]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
8f18cf13 | 6 | #include <linux/kernel.h> |
065631f6 | 7 | #include <linux/bio.h> |
39279cc3 | 8 | #include <linux/buffer_head.h> |
f2eb0a24 | 9 | #include <linux/file.h> |
39279cc3 CM |
10 | #include <linux/fs.h> |
11 | #include <linux/pagemap.h> | |
12 | #include <linux/highmem.h> | |
13 | #include <linux/time.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/string.h> | |
39279cc3 | 16 | #include <linux/backing-dev.h> |
39279cc3 | 17 | #include <linux/writeback.h> |
39279cc3 | 18 | #include <linux/compat.h> |
5103e947 | 19 | #include <linux/xattr.h> |
33268eaf | 20 | #include <linux/posix_acl.h> |
d899e052 | 21 | #include <linux/falloc.h> |
5a0e3ad6 | 22 | #include <linux/slab.h> |
7a36ddec | 23 | #include <linux/ratelimit.h> |
55e301fd | 24 | #include <linux/btrfs.h> |
53b381b3 | 25 | #include <linux/blkdev.h> |
f23b5a59 | 26 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 27 | #include <linux/uio.h> |
69fe2d75 | 28 | #include <linux/magic.h> |
ae5e165d | 29 | #include <linux/iversion.h> |
ed46ff3d | 30 | #include <linux/swap.h> |
b1c16ac9 | 31 | #include <linux/sched/mm.h> |
92d32170 | 32 | #include <asm/unaligned.h> |
602cbe91 | 33 | #include "misc.h" |
39279cc3 CM |
34 | #include "ctree.h" |
35 | #include "disk-io.h" | |
36 | #include "transaction.h" | |
37 | #include "btrfs_inode.h" | |
39279cc3 | 38 | #include "print-tree.h" |
e6dcd2dc | 39 | #include "ordered-data.h" |
95819c05 | 40 | #include "xattr.h" |
e02119d5 | 41 | #include "tree-log.h" |
4a54c8c1 | 42 | #include "volumes.h" |
c8b97818 | 43 | #include "compression.h" |
b4ce94de | 44 | #include "locking.h" |
dc89e982 | 45 | #include "free-space-cache.h" |
581bb050 | 46 | #include "inode-map.h" |
38c227d8 | 47 | #include "backref.h" |
63541927 | 48 | #include "props.h" |
31193213 | 49 | #include "qgroup.h" |
86736342 | 50 | #include "delalloc-space.h" |
aac0023c | 51 | #include "block-group.h" |
39279cc3 CM |
52 | |
53 | struct btrfs_iget_args { | |
90d3e592 | 54 | struct btrfs_key *location; |
39279cc3 CM |
55 | struct btrfs_root *root; |
56 | }; | |
57 | ||
f28a4928 | 58 | struct btrfs_dio_data { |
f28a4928 FM |
59 | u64 reserve; |
60 | u64 unsubmitted_oe_range_start; | |
61 | u64 unsubmitted_oe_range_end; | |
4aaedfb0 | 62 | int overwrite; |
f28a4928 FM |
63 | }; |
64 | ||
6e1d5dcc AD |
65 | static const struct inode_operations btrfs_dir_inode_operations; |
66 | static const struct inode_operations btrfs_symlink_inode_operations; | |
67 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
68 | static const struct inode_operations btrfs_special_inode_operations; | |
69 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b | 70 | static const struct address_space_operations btrfs_aops; |
828c0950 | 71 | static const struct file_operations btrfs_dir_file_operations; |
20e5506b | 72 | static const struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
73 | |
74 | static struct kmem_cache *btrfs_inode_cachep; | |
75 | struct kmem_cache *btrfs_trans_handle_cachep; | |
39279cc3 | 76 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 77 | struct kmem_cache *btrfs_free_space_cachep; |
3acd4850 | 78 | struct kmem_cache *btrfs_free_space_bitmap_cachep; |
39279cc3 | 79 | |
3972f260 | 80 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
213e8c55 | 81 | static int btrfs_truncate(struct inode *inode, bool skip_writeback); |
5fd02043 | 82 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
83 | static noinline int cow_file_range(struct inode *inode, |
84 | struct page *locked_page, | |
74e9194a | 85 | u64 start, u64 end, int *page_started, |
330a5827 | 86 | unsigned long *nr_written, int unlock); |
6f9994db LB |
87 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, |
88 | u64 orig_start, u64 block_start, | |
89 | u64 block_len, u64 orig_block_len, | |
90 | u64 ram_bytes, int compress_type, | |
91 | int type); | |
7b128766 | 92 | |
52427260 QW |
93 | static void __endio_write_update_ordered(struct inode *inode, |
94 | const u64 offset, const u64 bytes, | |
95 | const bool uptodate); | |
96 | ||
97 | /* | |
98 | * Cleanup all submitted ordered extents in specified range to handle errors | |
52042d8e | 99 | * from the btrfs_run_delalloc_range() callback. |
52427260 QW |
100 | * |
101 | * NOTE: caller must ensure that when an error happens, it can not call | |
102 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
103 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
104 | * to be released, which we want to happen only when finishing the ordered | |
d1051d6e | 105 | * extent (btrfs_finish_ordered_io()). |
52427260 QW |
106 | */ |
107 | static inline void btrfs_cleanup_ordered_extents(struct inode *inode, | |
d1051d6e NB |
108 | struct page *locked_page, |
109 | u64 offset, u64 bytes) | |
52427260 | 110 | { |
63d71450 NA |
111 | unsigned long index = offset >> PAGE_SHIFT; |
112 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
d1051d6e NB |
113 | u64 page_start = page_offset(locked_page); |
114 | u64 page_end = page_start + PAGE_SIZE - 1; | |
115 | ||
63d71450 NA |
116 | struct page *page; |
117 | ||
118 | while (index <= end_index) { | |
119 | page = find_get_page(inode->i_mapping, index); | |
120 | index++; | |
121 | if (!page) | |
122 | continue; | |
123 | ClearPagePrivate2(page); | |
124 | put_page(page); | |
125 | } | |
d1051d6e NB |
126 | |
127 | /* | |
128 | * In case this page belongs to the delalloc range being instantiated | |
129 | * then skip it, since the first page of a range is going to be | |
130 | * properly cleaned up by the caller of run_delalloc_range | |
131 | */ | |
132 | if (page_start >= offset && page_end <= (offset + bytes - 1)) { | |
133 | offset += PAGE_SIZE; | |
134 | bytes -= PAGE_SIZE; | |
135 | } | |
136 | ||
137 | return __endio_write_update_ordered(inode, offset, bytes, false); | |
52427260 QW |
138 | } |
139 | ||
48a3b636 | 140 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 141 | |
6a3891c5 JB |
142 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
143 | void btrfs_test_inode_set_ops(struct inode *inode) | |
144 | { | |
145 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
146 | } | |
147 | #endif | |
148 | ||
f34f57a3 | 149 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
150 | struct inode *inode, struct inode *dir, |
151 | const struct qstr *qstr) | |
0279b4cd JO |
152 | { |
153 | int err; | |
154 | ||
f34f57a3 | 155 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 156 | if (!err) |
2a7dba39 | 157 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
158 | return err; |
159 | } | |
160 | ||
c8b97818 CM |
161 | /* |
162 | * this does all the hard work for inserting an inline extent into | |
163 | * the btree. The caller should have done a btrfs_drop_extents so that | |
164 | * no overlapping inline items exist in the btree | |
165 | */ | |
40f76580 | 166 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 167 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
168 | struct btrfs_root *root, struct inode *inode, |
169 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 170 | int compress_type, |
c8b97818 CM |
171 | struct page **compressed_pages) |
172 | { | |
c8b97818 CM |
173 | struct extent_buffer *leaf; |
174 | struct page *page = NULL; | |
175 | char *kaddr; | |
176 | unsigned long ptr; | |
177 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
178 | int ret; |
179 | size_t cur_size = size; | |
c8b97818 | 180 | unsigned long offset; |
c8b97818 | 181 | |
982f1f5d JJB |
182 | ASSERT((compressed_size > 0 && compressed_pages) || |
183 | (compressed_size == 0 && !compressed_pages)); | |
184 | ||
fe3f566c | 185 | if (compressed_size && compressed_pages) |
c8b97818 | 186 | cur_size = compressed_size; |
c8b97818 | 187 | |
1acae57b | 188 | inode_add_bytes(inode, size); |
c8b97818 | 189 | |
1acae57b FDBM |
190 | if (!extent_inserted) { |
191 | struct btrfs_key key; | |
192 | size_t datasize; | |
c8b97818 | 193 | |
4a0cc7ca | 194 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b | 195 | key.offset = start; |
962a298f | 196 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 197 | |
1acae57b FDBM |
198 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
199 | path->leave_spinning = 1; | |
200 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
201 | datasize); | |
79b4f4c6 | 202 | if (ret) |
1acae57b | 203 | goto fail; |
c8b97818 CM |
204 | } |
205 | leaf = path->nodes[0]; | |
206 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
207 | struct btrfs_file_extent_item); | |
208 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
209 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
210 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
211 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
212 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
213 | ptr = btrfs_file_extent_inline_start(ei); | |
214 | ||
261507a0 | 215 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
216 | struct page *cpage; |
217 | int i = 0; | |
d397712b | 218 | while (compressed_size > 0) { |
c8b97818 | 219 | cpage = compressed_pages[i]; |
5b050f04 | 220 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 221 | PAGE_SIZE); |
c8b97818 | 222 | |
7ac687d9 | 223 | kaddr = kmap_atomic(cpage); |
c8b97818 | 224 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 225 | kunmap_atomic(kaddr); |
c8b97818 CM |
226 | |
227 | i++; | |
228 | ptr += cur_size; | |
229 | compressed_size -= cur_size; | |
230 | } | |
231 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 232 | compress_type); |
c8b97818 CM |
233 | } else { |
234 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 235 | start >> PAGE_SHIFT); |
c8b97818 | 236 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 237 | kaddr = kmap_atomic(page); |
7073017a | 238 | offset = offset_in_page(start); |
c8b97818 | 239 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 240 | kunmap_atomic(kaddr); |
09cbfeaf | 241 | put_page(page); |
c8b97818 CM |
242 | } |
243 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 244 | btrfs_release_path(path); |
c8b97818 | 245 | |
c2167754 YZ |
246 | /* |
247 | * we're an inline extent, so nobody can | |
248 | * extend the file past i_size without locking | |
249 | * a page we already have locked. | |
250 | * | |
251 | * We must do any isize and inode updates | |
252 | * before we unlock the pages. Otherwise we | |
253 | * could end up racing with unlink. | |
254 | */ | |
c8b97818 | 255 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 256 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 257 | |
c8b97818 | 258 | fail: |
79b4f4c6 | 259 | return ret; |
c8b97818 CM |
260 | } |
261 | ||
262 | ||
263 | /* | |
264 | * conditionally insert an inline extent into the file. This | |
265 | * does the checks required to make sure the data is small enough | |
266 | * to fit as an inline extent. | |
267 | */ | |
d02c0e20 | 268 | static noinline int cow_file_range_inline(struct inode *inode, u64 start, |
00361589 JB |
269 | u64 end, size_t compressed_size, |
270 | int compress_type, | |
271 | struct page **compressed_pages) | |
c8b97818 | 272 | { |
d02c0e20 | 273 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0b246afa | 274 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 275 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
276 | u64 isize = i_size_read(inode); |
277 | u64 actual_end = min(end + 1, isize); | |
278 | u64 inline_len = actual_end - start; | |
0b246afa | 279 | u64 aligned_end = ALIGN(end, fs_info->sectorsize); |
c8b97818 CM |
280 | u64 data_len = inline_len; |
281 | int ret; | |
1acae57b FDBM |
282 | struct btrfs_path *path; |
283 | int extent_inserted = 0; | |
284 | u32 extent_item_size; | |
c8b97818 CM |
285 | |
286 | if (compressed_size) | |
287 | data_len = compressed_size; | |
288 | ||
289 | if (start > 0 || | |
0b246afa JM |
290 | actual_end > fs_info->sectorsize || |
291 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || | |
c8b97818 | 292 | (!compressed_size && |
0b246afa | 293 | (actual_end & (fs_info->sectorsize - 1)) == 0) || |
c8b97818 | 294 | end + 1 < isize || |
0b246afa | 295 | data_len > fs_info->max_inline) { |
c8b97818 CM |
296 | return 1; |
297 | } | |
298 | ||
1acae57b FDBM |
299 | path = btrfs_alloc_path(); |
300 | if (!path) | |
301 | return -ENOMEM; | |
302 | ||
00361589 | 303 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
304 | if (IS_ERR(trans)) { |
305 | btrfs_free_path(path); | |
00361589 | 306 | return PTR_ERR(trans); |
1acae57b | 307 | } |
69fe2d75 | 308 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
00361589 | 309 | |
1acae57b FDBM |
310 | if (compressed_size && compressed_pages) |
311 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
312 | compressed_size); | |
313 | else | |
314 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
315 | inline_len); | |
316 | ||
317 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
318 | start, aligned_end, NULL, | |
319 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 | 320 | if (ret) { |
66642832 | 321 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
322 | goto out; |
323 | } | |
c8b97818 CM |
324 | |
325 | if (isize > actual_end) | |
326 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
327 | ret = insert_inline_extent(trans, path, extent_inserted, |
328 | root, inode, start, | |
c8b97818 | 329 | inline_len, compressed_size, |
fe3f566c | 330 | compress_type, compressed_pages); |
2adcac1a | 331 | if (ret && ret != -ENOSPC) { |
66642832 | 332 | btrfs_abort_transaction(trans, ret); |
00361589 | 333 | goto out; |
2adcac1a | 334 | } else if (ret == -ENOSPC) { |
00361589 JB |
335 | ret = 1; |
336 | goto out; | |
79787eaa | 337 | } |
2adcac1a | 338 | |
bdc20e67 | 339 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
dcdbc059 | 340 | btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0); |
00361589 | 341 | out: |
94ed938a QW |
342 | /* |
343 | * Don't forget to free the reserved space, as for inlined extent | |
344 | * it won't count as data extent, free them directly here. | |
345 | * And at reserve time, it's always aligned to page size, so | |
346 | * just free one page here. | |
347 | */ | |
bc42bda2 | 348 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 349 | btrfs_free_path(path); |
3a45bb20 | 350 | btrfs_end_transaction(trans); |
00361589 | 351 | return ret; |
c8b97818 CM |
352 | } |
353 | ||
771ed689 CM |
354 | struct async_extent { |
355 | u64 start; | |
356 | u64 ram_size; | |
357 | u64 compressed_size; | |
358 | struct page **pages; | |
359 | unsigned long nr_pages; | |
261507a0 | 360 | int compress_type; |
771ed689 CM |
361 | struct list_head list; |
362 | }; | |
363 | ||
97db1204 | 364 | struct async_chunk { |
771ed689 | 365 | struct inode *inode; |
771ed689 CM |
366 | struct page *locked_page; |
367 | u64 start; | |
368 | u64 end; | |
f82b7359 | 369 | unsigned int write_flags; |
771ed689 CM |
370 | struct list_head extents; |
371 | struct btrfs_work work; | |
97db1204 | 372 | atomic_t *pending; |
771ed689 CM |
373 | }; |
374 | ||
97db1204 NB |
375 | struct async_cow { |
376 | /* Number of chunks in flight; must be first in the structure */ | |
377 | atomic_t num_chunks; | |
378 | struct async_chunk chunks[]; | |
771ed689 CM |
379 | }; |
380 | ||
97db1204 | 381 | static noinline int add_async_extent(struct async_chunk *cow, |
771ed689 CM |
382 | u64 start, u64 ram_size, |
383 | u64 compressed_size, | |
384 | struct page **pages, | |
261507a0 LZ |
385 | unsigned long nr_pages, |
386 | int compress_type) | |
771ed689 CM |
387 | { |
388 | struct async_extent *async_extent; | |
389 | ||
390 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 391 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
392 | async_extent->start = start; |
393 | async_extent->ram_size = ram_size; | |
394 | async_extent->compressed_size = compressed_size; | |
395 | async_extent->pages = pages; | |
396 | async_extent->nr_pages = nr_pages; | |
261507a0 | 397 | async_extent->compress_type = compress_type; |
771ed689 CM |
398 | list_add_tail(&async_extent->list, &cow->extents); |
399 | return 0; | |
400 | } | |
401 | ||
42c16da6 QW |
402 | /* |
403 | * Check if the inode has flags compatible with compression | |
404 | */ | |
405 | static inline bool inode_can_compress(struct inode *inode) | |
406 | { | |
407 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW || | |
408 | BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
409 | return false; | |
410 | return true; | |
411 | } | |
412 | ||
413 | /* | |
414 | * Check if the inode needs to be submitted to compression, based on mount | |
415 | * options, defragmentation, properties or heuristics. | |
416 | */ | |
c2fcdcdf | 417 | static inline int inode_need_compress(struct inode *inode, u64 start, u64 end) |
f79707b0 | 418 | { |
0b246afa | 419 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
f79707b0 | 420 | |
42c16da6 QW |
421 | if (!inode_can_compress(inode)) { |
422 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), | |
423 | KERN_ERR "BTRFS: unexpected compression for ino %llu\n", | |
424 | btrfs_ino(BTRFS_I(inode))); | |
425 | return 0; | |
426 | } | |
f79707b0 | 427 | /* force compress */ |
0b246afa | 428 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 429 | return 1; |
eec63c65 DS |
430 | /* defrag ioctl */ |
431 | if (BTRFS_I(inode)->defrag_compress) | |
432 | return 1; | |
f79707b0 WS |
433 | /* bad compression ratios */ |
434 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
435 | return 0; | |
0b246afa | 436 | if (btrfs_test_opt(fs_info, COMPRESS) || |
f79707b0 | 437 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || |
b52aa8c9 | 438 | BTRFS_I(inode)->prop_compress) |
c2fcdcdf | 439 | return btrfs_compress_heuristic(inode, start, end); |
f79707b0 WS |
440 | return 0; |
441 | } | |
442 | ||
6158e1ce | 443 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
26d30f85 AJ |
444 | u64 start, u64 end, u64 num_bytes, u64 small_write) |
445 | { | |
446 | /* If this is a small write inside eof, kick off a defrag */ | |
447 | if (num_bytes < small_write && | |
6158e1ce | 448 | (start > 0 || end + 1 < inode->disk_i_size)) |
26d30f85 AJ |
449 | btrfs_add_inode_defrag(NULL, inode); |
450 | } | |
451 | ||
d352ac68 | 452 | /* |
771ed689 CM |
453 | * we create compressed extents in two phases. The first |
454 | * phase compresses a range of pages that have already been | |
455 | * locked (both pages and state bits are locked). | |
c8b97818 | 456 | * |
771ed689 CM |
457 | * This is done inside an ordered work queue, and the compression |
458 | * is spread across many cpus. The actual IO submission is step | |
459 | * two, and the ordered work queue takes care of making sure that | |
460 | * happens in the same order things were put onto the queue by | |
461 | * writepages and friends. | |
c8b97818 | 462 | * |
771ed689 CM |
463 | * If this code finds it can't get good compression, it puts an |
464 | * entry onto the work queue to write the uncompressed bytes. This | |
465 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
466 | * are written in the same order that the flusher thread sent them |
467 | * down. | |
d352ac68 | 468 | */ |
ac3e9933 | 469 | static noinline int compress_file_range(struct async_chunk *async_chunk) |
b888db2b | 470 | { |
1368c6da | 471 | struct inode *inode = async_chunk->inode; |
0b246afa | 472 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
0b246afa | 473 | u64 blocksize = fs_info->sectorsize; |
1368c6da NB |
474 | u64 start = async_chunk->start; |
475 | u64 end = async_chunk->end; | |
c8b97818 | 476 | u64 actual_end; |
e6dcd2dc | 477 | int ret = 0; |
c8b97818 CM |
478 | struct page **pages = NULL; |
479 | unsigned long nr_pages; | |
c8b97818 CM |
480 | unsigned long total_compressed = 0; |
481 | unsigned long total_in = 0; | |
c8b97818 CM |
482 | int i; |
483 | int will_compress; | |
0b246afa | 484 | int compress_type = fs_info->compress_type; |
ac3e9933 | 485 | int compressed_extents = 0; |
4adaa611 | 486 | int redirty = 0; |
b888db2b | 487 | |
6158e1ce NB |
488 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
489 | SZ_16K); | |
4cb5300b | 490 | |
62b37622 | 491 | actual_end = min_t(u64, i_size_read(inode), end + 1); |
c8b97818 CM |
492 | again: |
493 | will_compress = 0; | |
09cbfeaf | 494 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
495 | BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0); |
496 | nr_pages = min_t(unsigned long, nr_pages, | |
497 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 498 | |
f03d9301 CM |
499 | /* |
500 | * we don't want to send crud past the end of i_size through | |
501 | * compression, that's just a waste of CPU time. So, if the | |
502 | * end of the file is before the start of our current | |
503 | * requested range of bytes, we bail out to the uncompressed | |
504 | * cleanup code that can deal with all of this. | |
505 | * | |
506 | * It isn't really the fastest way to fix things, but this is a | |
507 | * very uncommon corner. | |
508 | */ | |
509 | if (actual_end <= start) | |
510 | goto cleanup_and_bail_uncompressed; | |
511 | ||
c8b97818 CM |
512 | total_compressed = actual_end - start; |
513 | ||
4bcbb332 SW |
514 | /* |
515 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 516 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
517 | */ |
518 | if (total_compressed <= blocksize && | |
519 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
520 | goto cleanup_and_bail_uncompressed; | |
521 | ||
069eac78 DS |
522 | total_compressed = min_t(unsigned long, total_compressed, |
523 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
524 | total_in = 0; |
525 | ret = 0; | |
db94535d | 526 | |
771ed689 CM |
527 | /* |
528 | * we do compression for mount -o compress and when the | |
529 | * inode has not been flagged as nocompress. This flag can | |
530 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 531 | */ |
c2fcdcdf | 532 | if (inode_need_compress(inode, start, end)) { |
c8b97818 | 533 | WARN_ON(pages); |
31e818fe | 534 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
535 | if (!pages) { |
536 | /* just bail out to the uncompressed code */ | |
3527a018 | 537 | nr_pages = 0; |
560f7d75 LZ |
538 | goto cont; |
539 | } | |
c8b97818 | 540 | |
eec63c65 DS |
541 | if (BTRFS_I(inode)->defrag_compress) |
542 | compress_type = BTRFS_I(inode)->defrag_compress; | |
543 | else if (BTRFS_I(inode)->prop_compress) | |
b52aa8c9 | 544 | compress_type = BTRFS_I(inode)->prop_compress; |
261507a0 | 545 | |
4adaa611 CM |
546 | /* |
547 | * we need to call clear_page_dirty_for_io on each | |
548 | * page in the range. Otherwise applications with the file | |
549 | * mmap'd can wander in and change the page contents while | |
550 | * we are compressing them. | |
551 | * | |
552 | * If the compression fails for any reason, we set the pages | |
553 | * dirty again later on. | |
e9679de3 TT |
554 | * |
555 | * Note that the remaining part is redirtied, the start pointer | |
556 | * has moved, the end is the original one. | |
4adaa611 | 557 | */ |
e9679de3 TT |
558 | if (!redirty) { |
559 | extent_range_clear_dirty_for_io(inode, start, end); | |
560 | redirty = 1; | |
561 | } | |
f51d2b59 DS |
562 | |
563 | /* Compression level is applied here and only here */ | |
564 | ret = btrfs_compress_pages( | |
565 | compress_type | (fs_info->compress_level << 4), | |
261507a0 | 566 | inode->i_mapping, start, |
38c31464 | 567 | pages, |
4d3a800e | 568 | &nr_pages, |
261507a0 | 569 | &total_in, |
e5d74902 | 570 | &total_compressed); |
c8b97818 CM |
571 | |
572 | if (!ret) { | |
7073017a | 573 | unsigned long offset = offset_in_page(total_compressed); |
4d3a800e | 574 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
575 | char *kaddr; |
576 | ||
577 | /* zero the tail end of the last page, we might be | |
578 | * sending it down to disk | |
579 | */ | |
580 | if (offset) { | |
7ac687d9 | 581 | kaddr = kmap_atomic(page); |
c8b97818 | 582 | memset(kaddr + offset, 0, |
09cbfeaf | 583 | PAGE_SIZE - offset); |
7ac687d9 | 584 | kunmap_atomic(kaddr); |
c8b97818 CM |
585 | } |
586 | will_compress = 1; | |
587 | } | |
588 | } | |
560f7d75 | 589 | cont: |
c8b97818 CM |
590 | if (start == 0) { |
591 | /* lets try to make an inline extent */ | |
6018ba0a | 592 | if (ret || total_in < actual_end) { |
c8b97818 | 593 | /* we didn't compress the entire range, try |
771ed689 | 594 | * to make an uncompressed inline extent. |
c8b97818 | 595 | */ |
d02c0e20 NB |
596 | ret = cow_file_range_inline(inode, start, end, 0, |
597 | BTRFS_COMPRESS_NONE, NULL); | |
c8b97818 | 598 | } else { |
771ed689 | 599 | /* try making a compressed inline extent */ |
d02c0e20 | 600 | ret = cow_file_range_inline(inode, start, end, |
fe3f566c LZ |
601 | total_compressed, |
602 | compress_type, pages); | |
c8b97818 | 603 | } |
79787eaa | 604 | if (ret <= 0) { |
151a41bc | 605 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
606 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
607 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 FM |
608 | unsigned long page_error_op; |
609 | ||
e6eb4314 | 610 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 611 | |
771ed689 | 612 | /* |
79787eaa JM |
613 | * inline extent creation worked or returned error, |
614 | * we don't need to create any more async work items. | |
615 | * Unlock and free up our temp pages. | |
8b62f87b JB |
616 | * |
617 | * We use DO_ACCOUNTING here because we need the | |
618 | * delalloc_release_metadata to be done _after_ we drop | |
619 | * our outstanding extent for clearing delalloc for this | |
620 | * range. | |
771ed689 | 621 | */ |
74e9194a NB |
622 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
623 | clear_flags, | |
ba8b04c1 | 624 | PAGE_UNLOCK | |
c2790a2e JB |
625 | PAGE_CLEAR_DIRTY | |
626 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 627 | page_error_op | |
c2790a2e | 628 | PAGE_END_WRITEBACK); |
cecc8d90 NB |
629 | |
630 | for (i = 0; i < nr_pages; i++) { | |
631 | WARN_ON(pages[i]->mapping); | |
632 | put_page(pages[i]); | |
633 | } | |
634 | kfree(pages); | |
635 | ||
636 | return 0; | |
c8b97818 CM |
637 | } |
638 | } | |
639 | ||
640 | if (will_compress) { | |
641 | /* | |
642 | * we aren't doing an inline extent round the compressed size | |
643 | * up to a block size boundary so the allocator does sane | |
644 | * things | |
645 | */ | |
fda2832f | 646 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
647 | |
648 | /* | |
649 | * one last check to make sure the compression is really a | |
170607eb TT |
650 | * win, compare the page count read with the blocks on disk, |
651 | * compression must free at least one sector size | |
c8b97818 | 652 | */ |
09cbfeaf | 653 | total_in = ALIGN(total_in, PAGE_SIZE); |
170607eb | 654 | if (total_compressed + blocksize <= total_in) { |
ac3e9933 | 655 | compressed_extents++; |
c8bb0c8b AS |
656 | |
657 | /* | |
658 | * The async work queues will take care of doing actual | |
659 | * allocation on disk for these compressed pages, and | |
660 | * will submit them to the elevator. | |
661 | */ | |
b5326271 | 662 | add_async_extent(async_chunk, start, total_in, |
4d3a800e | 663 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
664 | compress_type); |
665 | ||
1170862d TT |
666 | if (start + total_in < end) { |
667 | start += total_in; | |
c8bb0c8b AS |
668 | pages = NULL; |
669 | cond_resched(); | |
670 | goto again; | |
671 | } | |
ac3e9933 | 672 | return compressed_extents; |
c8b97818 CM |
673 | } |
674 | } | |
c8bb0c8b | 675 | if (pages) { |
c8b97818 CM |
676 | /* |
677 | * the compression code ran but failed to make things smaller, | |
678 | * free any pages it allocated and our page pointer array | |
679 | */ | |
4d3a800e | 680 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 681 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 682 | put_page(pages[i]); |
c8b97818 CM |
683 | } |
684 | kfree(pages); | |
685 | pages = NULL; | |
686 | total_compressed = 0; | |
4d3a800e | 687 | nr_pages = 0; |
c8b97818 CM |
688 | |
689 | /* flag the file so we don't compress in the future */ | |
0b246afa | 690 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
b52aa8c9 | 691 | !(BTRFS_I(inode)->prop_compress)) { |
a555f810 | 692 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 693 | } |
c8b97818 | 694 | } |
f03d9301 | 695 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
696 | /* |
697 | * No compression, but we still need to write the pages in the file | |
698 | * we've been given so far. redirty the locked page if it corresponds | |
699 | * to our extent and set things up for the async work queue to run | |
700 | * cow_file_range to do the normal delalloc dance. | |
701 | */ | |
1368c6da NB |
702 | if (page_offset(async_chunk->locked_page) >= start && |
703 | page_offset(async_chunk->locked_page) <= end) | |
704 | __set_page_dirty_nobuffers(async_chunk->locked_page); | |
c8bb0c8b AS |
705 | /* unlocked later on in the async handlers */ |
706 | ||
707 | if (redirty) | |
708 | extent_range_redirty_for_io(inode, start, end); | |
b5326271 | 709 | add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0, |
c8bb0c8b | 710 | BTRFS_COMPRESS_NONE); |
ac3e9933 | 711 | compressed_extents++; |
3b951516 | 712 | |
ac3e9933 | 713 | return compressed_extents; |
771ed689 | 714 | } |
771ed689 | 715 | |
40ae837b FM |
716 | static void free_async_extent_pages(struct async_extent *async_extent) |
717 | { | |
718 | int i; | |
719 | ||
720 | if (!async_extent->pages) | |
721 | return; | |
722 | ||
723 | for (i = 0; i < async_extent->nr_pages; i++) { | |
724 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 725 | put_page(async_extent->pages[i]); |
40ae837b FM |
726 | } |
727 | kfree(async_extent->pages); | |
728 | async_extent->nr_pages = 0; | |
729 | async_extent->pages = NULL; | |
771ed689 CM |
730 | } |
731 | ||
732 | /* | |
733 | * phase two of compressed writeback. This is the ordered portion | |
734 | * of the code, which only gets called in the order the work was | |
735 | * queued. We walk all the async extents created by compress_file_range | |
736 | * and send them down to the disk. | |
737 | */ | |
b5326271 | 738 | static noinline void submit_compressed_extents(struct async_chunk *async_chunk) |
771ed689 | 739 | { |
b5326271 | 740 | struct inode *inode = async_chunk->inode; |
0b246afa | 741 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
742 | struct async_extent *async_extent; |
743 | u64 alloc_hint = 0; | |
771ed689 CM |
744 | struct btrfs_key ins; |
745 | struct extent_map *em; | |
746 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4336650a | 747 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
f5a84ee3 | 748 | int ret = 0; |
771ed689 | 749 | |
3e04e7f1 | 750 | again: |
b5326271 NB |
751 | while (!list_empty(&async_chunk->extents)) { |
752 | async_extent = list_entry(async_chunk->extents.next, | |
771ed689 CM |
753 | struct async_extent, list); |
754 | list_del(&async_extent->list); | |
c8b97818 | 755 | |
f5a84ee3 | 756 | retry: |
7447555f NB |
757 | lock_extent(io_tree, async_extent->start, |
758 | async_extent->start + async_extent->ram_size - 1); | |
771ed689 CM |
759 | /* did the compression code fall back to uncompressed IO? */ |
760 | if (!async_extent->pages) { | |
761 | int page_started = 0; | |
762 | unsigned long nr_written = 0; | |
763 | ||
771ed689 | 764 | /* allocate blocks */ |
b5326271 | 765 | ret = cow_file_range(inode, async_chunk->locked_page, |
f5a84ee3 JB |
766 | async_extent->start, |
767 | async_extent->start + | |
768 | async_extent->ram_size - 1, | |
330a5827 | 769 | &page_started, &nr_written, 0); |
771ed689 | 770 | |
79787eaa JM |
771 | /* JDM XXX */ |
772 | ||
771ed689 CM |
773 | /* |
774 | * if page_started, cow_file_range inserted an | |
775 | * inline extent and took care of all the unlocking | |
776 | * and IO for us. Otherwise, we need to submit | |
777 | * all those pages down to the drive. | |
778 | */ | |
f5a84ee3 | 779 | if (!page_started && !ret) |
5e3ee236 NB |
780 | extent_write_locked_range(inode, |
781 | async_extent->start, | |
d397712b | 782 | async_extent->start + |
771ed689 | 783 | async_extent->ram_size - 1, |
771ed689 | 784 | WB_SYNC_ALL); |
3e04e7f1 | 785 | else if (ret) |
b5326271 | 786 | unlock_page(async_chunk->locked_page); |
771ed689 CM |
787 | kfree(async_extent); |
788 | cond_resched(); | |
789 | continue; | |
790 | } | |
791 | ||
18513091 | 792 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
771ed689 CM |
793 | async_extent->compressed_size, |
794 | async_extent->compressed_size, | |
e570fd27 | 795 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 796 | if (ret) { |
40ae837b | 797 | free_async_extent_pages(async_extent); |
3e04e7f1 | 798 | |
fdf8e2ea JB |
799 | if (ret == -ENOSPC) { |
800 | unlock_extent(io_tree, async_extent->start, | |
801 | async_extent->start + | |
802 | async_extent->ram_size - 1); | |
ce62003f LB |
803 | |
804 | /* | |
805 | * we need to redirty the pages if we decide to | |
806 | * fallback to uncompressed IO, otherwise we | |
807 | * will not submit these pages down to lower | |
808 | * layers. | |
809 | */ | |
810 | extent_range_redirty_for_io(inode, | |
811 | async_extent->start, | |
812 | async_extent->start + | |
813 | async_extent->ram_size - 1); | |
814 | ||
79787eaa | 815 | goto retry; |
fdf8e2ea | 816 | } |
3e04e7f1 | 817 | goto out_free; |
f5a84ee3 | 818 | } |
c2167754 YZ |
819 | /* |
820 | * here we're doing allocation and writeback of the | |
821 | * compressed pages | |
822 | */ | |
6f9994db LB |
823 | em = create_io_em(inode, async_extent->start, |
824 | async_extent->ram_size, /* len */ | |
825 | async_extent->start, /* orig_start */ | |
826 | ins.objectid, /* block_start */ | |
827 | ins.offset, /* block_len */ | |
828 | ins.offset, /* orig_block_len */ | |
829 | async_extent->ram_size, /* ram_bytes */ | |
830 | async_extent->compress_type, | |
831 | BTRFS_ORDERED_COMPRESSED); | |
832 | if (IS_ERR(em)) | |
833 | /* ret value is not necessary due to void function */ | |
3e04e7f1 | 834 | goto out_free_reserve; |
6f9994db | 835 | free_extent_map(em); |
3e04e7f1 | 836 | |
261507a0 LZ |
837 | ret = btrfs_add_ordered_extent_compress(inode, |
838 | async_extent->start, | |
839 | ins.objectid, | |
840 | async_extent->ram_size, | |
841 | ins.offset, | |
842 | BTRFS_ORDERED_COMPRESSED, | |
843 | async_extent->compress_type); | |
d9f85963 | 844 | if (ret) { |
dcdbc059 NB |
845 | btrfs_drop_extent_cache(BTRFS_I(inode), |
846 | async_extent->start, | |
d9f85963 FM |
847 | async_extent->start + |
848 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 849 | goto out_free_reserve; |
d9f85963 | 850 | } |
0b246afa | 851 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
771ed689 | 852 | |
771ed689 CM |
853 | /* |
854 | * clear dirty, set writeback and unlock the pages. | |
855 | */ | |
c2790a2e | 856 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
857 | async_extent->start + |
858 | async_extent->ram_size - 1, | |
151a41bc JB |
859 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
860 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 861 | PAGE_SET_WRITEBACK); |
4e4cbee9 | 862 | if (btrfs_submit_compressed_write(inode, |
d397712b CM |
863 | async_extent->start, |
864 | async_extent->ram_size, | |
865 | ins.objectid, | |
866 | ins.offset, async_extent->pages, | |
f82b7359 | 867 | async_extent->nr_pages, |
b5326271 | 868 | async_chunk->write_flags)) { |
fce2a4e6 FM |
869 | struct page *p = async_extent->pages[0]; |
870 | const u64 start = async_extent->start; | |
871 | const u64 end = start + async_extent->ram_size - 1; | |
872 | ||
873 | p->mapping = inode->i_mapping; | |
c629732d | 874 | btrfs_writepage_endio_finish_ordered(p, start, end, 0); |
7087a9d8 | 875 | |
fce2a4e6 | 876 | p->mapping = NULL; |
74e9194a | 877 | extent_clear_unlock_delalloc(inode, start, end, |
ba8b04c1 | 878 | NULL, 0, |
fce2a4e6 FM |
879 | PAGE_END_WRITEBACK | |
880 | PAGE_SET_ERROR); | |
40ae837b | 881 | free_async_extent_pages(async_extent); |
fce2a4e6 | 882 | } |
771ed689 CM |
883 | alloc_hint = ins.objectid + ins.offset; |
884 | kfree(async_extent); | |
885 | cond_resched(); | |
886 | } | |
dec8f175 | 887 | return; |
3e04e7f1 | 888 | out_free_reserve: |
0b246afa | 889 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 890 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 891 | out_free: |
c2790a2e | 892 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
893 | async_extent->start + |
894 | async_extent->ram_size - 1, | |
c2790a2e | 895 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 896 | EXTENT_DELALLOC_NEW | |
151a41bc JB |
897 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
898 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
899 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
900 | PAGE_SET_ERROR); | |
40ae837b | 901 | free_async_extent_pages(async_extent); |
79787eaa | 902 | kfree(async_extent); |
3e04e7f1 | 903 | goto again; |
771ed689 CM |
904 | } |
905 | ||
4b46fce2 JB |
906 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
907 | u64 num_bytes) | |
908 | { | |
909 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
910 | struct extent_map *em; | |
911 | u64 alloc_hint = 0; | |
912 | ||
913 | read_lock(&em_tree->lock); | |
914 | em = search_extent_mapping(em_tree, start, num_bytes); | |
915 | if (em) { | |
916 | /* | |
917 | * if block start isn't an actual block number then find the | |
918 | * first block in this inode and use that as a hint. If that | |
919 | * block is also bogus then just don't worry about it. | |
920 | */ | |
921 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
922 | free_extent_map(em); | |
923 | em = search_extent_mapping(em_tree, 0, 0); | |
924 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
925 | alloc_hint = em->block_start; | |
926 | if (em) | |
927 | free_extent_map(em); | |
928 | } else { | |
929 | alloc_hint = em->block_start; | |
930 | free_extent_map(em); | |
931 | } | |
932 | } | |
933 | read_unlock(&em_tree->lock); | |
934 | ||
935 | return alloc_hint; | |
936 | } | |
937 | ||
771ed689 CM |
938 | /* |
939 | * when extent_io.c finds a delayed allocation range in the file, | |
940 | * the call backs end up in this code. The basic idea is to | |
941 | * allocate extents on disk for the range, and create ordered data structs | |
942 | * in ram to track those extents. | |
943 | * | |
944 | * locked_page is the page that writepage had locked already. We use | |
945 | * it to make sure we don't do extra locks or unlocks. | |
946 | * | |
947 | * *page_started is set to one if we unlock locked_page and do everything | |
948 | * required to start IO on it. It may be clean and already done with | |
949 | * IO when we return. | |
950 | */ | |
00361589 JB |
951 | static noinline int cow_file_range(struct inode *inode, |
952 | struct page *locked_page, | |
74e9194a | 953 | u64 start, u64 end, int *page_started, |
330a5827 | 954 | unsigned long *nr_written, int unlock) |
771ed689 | 955 | { |
0b246afa | 956 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00361589 | 957 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
958 | u64 alloc_hint = 0; |
959 | u64 num_bytes; | |
960 | unsigned long ram_size; | |
a315e68f | 961 | u64 cur_alloc_size = 0; |
0b246afa | 962 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
963 | struct btrfs_key ins; |
964 | struct extent_map *em; | |
a315e68f FM |
965 | unsigned clear_bits; |
966 | unsigned long page_ops; | |
967 | bool extent_reserved = false; | |
771ed689 CM |
968 | int ret = 0; |
969 | ||
70ddc553 | 970 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
02ecd2c2 | 971 | WARN_ON_ONCE(1); |
29bce2f3 JB |
972 | ret = -EINVAL; |
973 | goto out_unlock; | |
02ecd2c2 | 974 | } |
771ed689 | 975 | |
fda2832f | 976 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 | 977 | num_bytes = max(blocksize, num_bytes); |
566b1760 | 978 | ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 979 | |
6158e1ce | 980 | inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K); |
4cb5300b | 981 | |
771ed689 CM |
982 | if (start == 0) { |
983 | /* lets try to make an inline extent */ | |
d02c0e20 NB |
984 | ret = cow_file_range_inline(inode, start, end, 0, |
985 | BTRFS_COMPRESS_NONE, NULL); | |
771ed689 | 986 | if (ret == 0) { |
8b62f87b JB |
987 | /* |
988 | * We use DO_ACCOUNTING here because we need the | |
989 | * delalloc_release_metadata to be run _after_ we drop | |
990 | * our outstanding extent for clearing delalloc for this | |
991 | * range. | |
992 | */ | |
74e9194a | 993 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
c2790a2e | 994 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
995 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
996 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
c2790a2e JB |
997 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
998 | PAGE_END_WRITEBACK); | |
771ed689 | 999 | *nr_written = *nr_written + |
09cbfeaf | 1000 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 1001 | *page_started = 1; |
771ed689 | 1002 | goto out; |
79787eaa | 1003 | } else if (ret < 0) { |
79787eaa | 1004 | goto out_unlock; |
771ed689 CM |
1005 | } |
1006 | } | |
1007 | ||
4b46fce2 | 1008 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
dcdbc059 NB |
1009 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
1010 | start + num_bytes - 1, 0); | |
771ed689 | 1011 | |
3752d22f AJ |
1012 | while (num_bytes > 0) { |
1013 | cur_alloc_size = num_bytes; | |
18513091 | 1014 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
0b246afa | 1015 | fs_info->sectorsize, 0, alloc_hint, |
e570fd27 | 1016 | &ins, 1, 1); |
00361589 | 1017 | if (ret < 0) |
79787eaa | 1018 | goto out_unlock; |
a315e68f FM |
1019 | cur_alloc_size = ins.offset; |
1020 | extent_reserved = true; | |
d397712b | 1021 | |
771ed689 | 1022 | ram_size = ins.offset; |
6f9994db LB |
1023 | em = create_io_em(inode, start, ins.offset, /* len */ |
1024 | start, /* orig_start */ | |
1025 | ins.objectid, /* block_start */ | |
1026 | ins.offset, /* block_len */ | |
1027 | ins.offset, /* orig_block_len */ | |
1028 | ram_size, /* ram_bytes */ | |
1029 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1030 | BTRFS_ORDERED_REGULAR /* type */); |
090a127a SY |
1031 | if (IS_ERR(em)) { |
1032 | ret = PTR_ERR(em); | |
ace68bac | 1033 | goto out_reserve; |
090a127a | 1034 | } |
6f9994db | 1035 | free_extent_map(em); |
e6dcd2dc | 1036 | |
e6dcd2dc | 1037 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1038 | ram_size, cur_alloc_size, 0); |
ace68bac | 1039 | if (ret) |
d9f85963 | 1040 | goto out_drop_extent_cache; |
c8b97818 | 1041 | |
17d217fe YZ |
1042 | if (root->root_key.objectid == |
1043 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1044 | ret = btrfs_reloc_clone_csums(inode, start, | |
1045 | cur_alloc_size); | |
4dbd80fb QW |
1046 | /* |
1047 | * Only drop cache here, and process as normal. | |
1048 | * | |
1049 | * We must not allow extent_clear_unlock_delalloc() | |
1050 | * at out_unlock label to free meta of this ordered | |
1051 | * extent, as its meta should be freed by | |
1052 | * btrfs_finish_ordered_io(). | |
1053 | * | |
1054 | * So we must continue until @start is increased to | |
1055 | * skip current ordered extent. | |
1056 | */ | |
00361589 | 1057 | if (ret) |
4dbd80fb QW |
1058 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
1059 | start + ram_size - 1, 0); | |
17d217fe YZ |
1060 | } |
1061 | ||
0b246afa | 1062 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1063 | |
c8b97818 CM |
1064 | /* we're not doing compressed IO, don't unlock the first |
1065 | * page (which the caller expects to stay locked), don't | |
1066 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1067 | * |
1068 | * Do set the Private2 bit so we know this page was properly | |
1069 | * setup for writepage | |
c8b97818 | 1070 | */ |
a315e68f FM |
1071 | page_ops = unlock ? PAGE_UNLOCK : 0; |
1072 | page_ops |= PAGE_SET_PRIVATE2; | |
a791e35e | 1073 | |
c2790a2e | 1074 | extent_clear_unlock_delalloc(inode, start, |
ba8b04c1 | 1075 | start + ram_size - 1, |
74e9194a | 1076 | locked_page, |
c2790a2e | 1077 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1078 | page_ops); |
3752d22f AJ |
1079 | if (num_bytes < cur_alloc_size) |
1080 | num_bytes = 0; | |
4dbd80fb | 1081 | else |
3752d22f | 1082 | num_bytes -= cur_alloc_size; |
c59f8951 CM |
1083 | alloc_hint = ins.objectid + ins.offset; |
1084 | start += cur_alloc_size; | |
a315e68f | 1085 | extent_reserved = false; |
4dbd80fb QW |
1086 | |
1087 | /* | |
1088 | * btrfs_reloc_clone_csums() error, since start is increased | |
1089 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1090 | * free metadata of current ordered extent, we're OK to exit. | |
1091 | */ | |
1092 | if (ret) | |
1093 | goto out_unlock; | |
b888db2b | 1094 | } |
79787eaa | 1095 | out: |
be20aa9d | 1096 | return ret; |
b7d5b0a8 | 1097 | |
d9f85963 | 1098 | out_drop_extent_cache: |
dcdbc059 | 1099 | btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0); |
ace68bac | 1100 | out_reserve: |
0b246afa | 1101 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1102 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1103 | out_unlock: |
a7e3b975 FM |
1104 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1105 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
a315e68f FM |
1106 | page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
1107 | PAGE_END_WRITEBACK; | |
1108 | /* | |
1109 | * If we reserved an extent for our delalloc range (or a subrange) and | |
1110 | * failed to create the respective ordered extent, then it means that | |
1111 | * when we reserved the extent we decremented the extent's size from | |
1112 | * the data space_info's bytes_may_use counter and incremented the | |
1113 | * space_info's bytes_reserved counter by the same amount. We must make | |
1114 | * sure extent_clear_unlock_delalloc() does not try to decrement again | |
1115 | * the data space_info's bytes_may_use counter, therefore we do not pass | |
1116 | * it the flag EXTENT_CLEAR_DATA_RESV. | |
1117 | */ | |
1118 | if (extent_reserved) { | |
1119 | extent_clear_unlock_delalloc(inode, start, | |
a315e68f FM |
1120 | start + cur_alloc_size, |
1121 | locked_page, | |
1122 | clear_bits, | |
1123 | page_ops); | |
1124 | start += cur_alloc_size; | |
1125 | if (start >= end) | |
1126 | goto out; | |
1127 | } | |
74e9194a | 1128 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
a315e68f FM |
1129 | clear_bits | EXTENT_CLEAR_DATA_RESV, |
1130 | page_ops); | |
79787eaa | 1131 | goto out; |
771ed689 | 1132 | } |
c8b97818 | 1133 | |
771ed689 CM |
1134 | /* |
1135 | * work queue call back to started compression on a file and pages | |
1136 | */ | |
1137 | static noinline void async_cow_start(struct btrfs_work *work) | |
1138 | { | |
b5326271 | 1139 | struct async_chunk *async_chunk; |
ac3e9933 | 1140 | int compressed_extents; |
771ed689 | 1141 | |
b5326271 | 1142 | async_chunk = container_of(work, struct async_chunk, work); |
771ed689 | 1143 | |
ac3e9933 NB |
1144 | compressed_extents = compress_file_range(async_chunk); |
1145 | if (compressed_extents == 0) { | |
b5326271 NB |
1146 | btrfs_add_delayed_iput(async_chunk->inode); |
1147 | async_chunk->inode = NULL; | |
8180ef88 | 1148 | } |
771ed689 CM |
1149 | } |
1150 | ||
1151 | /* | |
1152 | * work queue call back to submit previously compressed pages | |
1153 | */ | |
1154 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1155 | { | |
c5a68aec NB |
1156 | struct async_chunk *async_chunk = container_of(work, struct async_chunk, |
1157 | work); | |
1158 | struct btrfs_fs_info *fs_info = btrfs_work_owner(work); | |
771ed689 CM |
1159 | unsigned long nr_pages; |
1160 | ||
b5326271 | 1161 | nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >> |
09cbfeaf | 1162 | PAGE_SHIFT; |
771ed689 | 1163 | |
093258e6 | 1164 | /* atomic_sub_return implies a barrier */ |
0b246afa | 1165 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < |
093258e6 DS |
1166 | 5 * SZ_1M) |
1167 | cond_wake_up_nomb(&fs_info->async_submit_wait); | |
771ed689 | 1168 | |
4546d178 | 1169 | /* |
b5326271 | 1170 | * ->inode could be NULL if async_chunk_start has failed to compress, |
4546d178 NB |
1171 | * in which case we don't have anything to submit, yet we need to |
1172 | * always adjust ->async_delalloc_pages as its paired with the init | |
1173 | * happening in cow_file_range_async | |
1174 | */ | |
b5326271 NB |
1175 | if (async_chunk->inode) |
1176 | submit_compressed_extents(async_chunk); | |
771ed689 | 1177 | } |
c8b97818 | 1178 | |
771ed689 CM |
1179 | static noinline void async_cow_free(struct btrfs_work *work) |
1180 | { | |
b5326271 | 1181 | struct async_chunk *async_chunk; |
97db1204 | 1182 | |
b5326271 NB |
1183 | async_chunk = container_of(work, struct async_chunk, work); |
1184 | if (async_chunk->inode) | |
1185 | btrfs_add_delayed_iput(async_chunk->inode); | |
97db1204 NB |
1186 | /* |
1187 | * Since the pointer to 'pending' is at the beginning of the array of | |
b5326271 | 1188 | * async_chunk's, freeing it ensures the whole array has been freed. |
97db1204 | 1189 | */ |
b5326271 | 1190 | if (atomic_dec_and_test(async_chunk->pending)) |
b1c16ac9 | 1191 | kvfree(async_chunk->pending); |
771ed689 CM |
1192 | } |
1193 | ||
1194 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1195 | u64 start, u64 end, int *page_started, | |
f82b7359 LB |
1196 | unsigned long *nr_written, |
1197 | unsigned int write_flags) | |
771ed689 | 1198 | { |
0b246afa | 1199 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
97db1204 NB |
1200 | struct async_cow *ctx; |
1201 | struct async_chunk *async_chunk; | |
771ed689 CM |
1202 | unsigned long nr_pages; |
1203 | u64 cur_end; | |
97db1204 NB |
1204 | u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K); |
1205 | int i; | |
1206 | bool should_compress; | |
b1c16ac9 | 1207 | unsigned nofs_flag; |
771ed689 | 1208 | |
69684c5a | 1209 | unlock_extent(&BTRFS_I(inode)->io_tree, start, end); |
97db1204 NB |
1210 | |
1211 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && | |
1212 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) { | |
1213 | num_chunks = 1; | |
1214 | should_compress = false; | |
1215 | } else { | |
1216 | should_compress = true; | |
1217 | } | |
1218 | ||
b1c16ac9 NB |
1219 | nofs_flag = memalloc_nofs_save(); |
1220 | ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL); | |
1221 | memalloc_nofs_restore(nofs_flag); | |
1222 | ||
97db1204 NB |
1223 | if (!ctx) { |
1224 | unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | | |
1225 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | | |
1226 | EXTENT_DO_ACCOUNTING; | |
1227 | unsigned long page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1228 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | | |
1229 | PAGE_SET_ERROR; | |
1230 | ||
74e9194a | 1231 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
97db1204 NB |
1232 | clear_bits, page_ops); |
1233 | return -ENOMEM; | |
1234 | } | |
1235 | ||
1236 | async_chunk = ctx->chunks; | |
1237 | atomic_set(&ctx->num_chunks, num_chunks); | |
1238 | ||
1239 | for (i = 0; i < num_chunks; i++) { | |
1240 | if (should_compress) | |
1241 | cur_end = min(end, start + SZ_512K - 1); | |
1242 | else | |
1243 | cur_end = end; | |
771ed689 | 1244 | |
bd4691a0 NB |
1245 | /* |
1246 | * igrab is called higher up in the call chain, take only the | |
1247 | * lightweight reference for the callback lifetime | |
1248 | */ | |
1249 | ihold(inode); | |
97db1204 NB |
1250 | async_chunk[i].pending = &ctx->num_chunks; |
1251 | async_chunk[i].inode = inode; | |
1252 | async_chunk[i].start = start; | |
1253 | async_chunk[i].end = cur_end; | |
97db1204 NB |
1254 | async_chunk[i].locked_page = locked_page; |
1255 | async_chunk[i].write_flags = write_flags; | |
1256 | INIT_LIST_HEAD(&async_chunk[i].extents); | |
1257 | ||
1258 | btrfs_init_work(&async_chunk[i].work, | |
9e0af237 LB |
1259 | btrfs_delalloc_helper, |
1260 | async_cow_start, async_cow_submit, | |
1261 | async_cow_free); | |
771ed689 | 1262 | |
97db1204 | 1263 | nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE); |
0b246afa | 1264 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1265 | |
97db1204 | 1266 | btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work); |
771ed689 | 1267 | |
771ed689 CM |
1268 | *nr_written += nr_pages; |
1269 | start = cur_end + 1; | |
1270 | } | |
1271 | *page_started = 1; | |
1272 | return 0; | |
be20aa9d CM |
1273 | } |
1274 | ||
2ff7e61e | 1275 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1276 | u64 bytenr, u64 num_bytes) |
1277 | { | |
1278 | int ret; | |
1279 | struct btrfs_ordered_sum *sums; | |
1280 | LIST_HEAD(list); | |
1281 | ||
0b246afa | 1282 | ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr, |
a2de733c | 1283 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1284 | if (ret == 0 && list_empty(&list)) |
1285 | return 0; | |
1286 | ||
1287 | while (!list_empty(&list)) { | |
1288 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1289 | list_del(&sums->list); | |
1290 | kfree(sums); | |
1291 | } | |
58113753 LB |
1292 | if (ret < 0) |
1293 | return ret; | |
17d217fe YZ |
1294 | return 1; |
1295 | } | |
1296 | ||
d352ac68 CM |
1297 | /* |
1298 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1299 | * of the extents that exist in the file, and COWs the file as required. | |
1300 | * | |
1301 | * If no cow copies or snapshots exist, we write directly to the existing | |
1302 | * blocks on disk | |
1303 | */ | |
7f366cfe CM |
1304 | static noinline int run_delalloc_nocow(struct inode *inode, |
1305 | struct page *locked_page, | |
3e024846 NB |
1306 | const u64 start, const u64 end, |
1307 | int *page_started, int force, | |
1308 | unsigned long *nr_written) | |
be20aa9d | 1309 | { |
0b246afa | 1310 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
be20aa9d | 1311 | struct btrfs_root *root = BTRFS_I(inode)->root; |
be20aa9d | 1312 | struct btrfs_path *path; |
3e024846 NB |
1313 | u64 cow_start = (u64)-1; |
1314 | u64 cur_offset = start; | |
8ecebf4d | 1315 | int ret; |
3e024846 NB |
1316 | bool check_prev = true; |
1317 | const bool freespace_inode = btrfs_is_free_space_inode(BTRFS_I(inode)); | |
4a0cc7ca | 1318 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
762bf098 NB |
1319 | bool nocow = false; |
1320 | u64 disk_bytenr = 0; | |
be20aa9d CM |
1321 | |
1322 | path = btrfs_alloc_path(); | |
17ca04af | 1323 | if (!path) { |
74e9194a | 1324 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
c2790a2e | 1325 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1326 | EXTENT_DO_ACCOUNTING | |
1327 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1328 | PAGE_CLEAR_DIRTY | |
1329 | PAGE_SET_WRITEBACK | | |
1330 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1331 | return -ENOMEM; |
17ca04af | 1332 | } |
82d5902d | 1333 | |
80ff3856 | 1334 | while (1) { |
3e024846 NB |
1335 | struct btrfs_key found_key; |
1336 | struct btrfs_file_extent_item *fi; | |
1337 | struct extent_buffer *leaf; | |
1338 | u64 extent_end; | |
1339 | u64 extent_offset; | |
3e024846 NB |
1340 | u64 num_bytes = 0; |
1341 | u64 disk_num_bytes; | |
3e024846 NB |
1342 | u64 ram_bytes; |
1343 | int extent_type; | |
762bf098 NB |
1344 | |
1345 | nocow = false; | |
3e024846 | 1346 | |
e4c3b2dc | 1347 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1348 | cur_offset, 0); |
d788a349 | 1349 | if (ret < 0) |
79787eaa | 1350 | goto error; |
a6bd9cd1 NB |
1351 | |
1352 | /* | |
1353 | * If there is no extent for our range when doing the initial | |
1354 | * search, then go back to the previous slot as it will be the | |
1355 | * one containing the search offset | |
1356 | */ | |
80ff3856 YZ |
1357 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1358 | leaf = path->nodes[0]; | |
1359 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1360 | path->slots[0] - 1); | |
33345d01 | 1361 | if (found_key.objectid == ino && |
80ff3856 YZ |
1362 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1363 | path->slots[0]--; | |
1364 | } | |
3e024846 | 1365 | check_prev = false; |
80ff3856 | 1366 | next_slot: |
a6bd9cd1 | 1367 | /* Go to next leaf if we have exhausted the current one */ |
80ff3856 YZ |
1368 | leaf = path->nodes[0]; |
1369 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1370 | ret = btrfs_next_leaf(root, path); | |
e8916699 LB |
1371 | if (ret < 0) { |
1372 | if (cow_start != (u64)-1) | |
1373 | cur_offset = cow_start; | |
79787eaa | 1374 | goto error; |
e8916699 | 1375 | } |
80ff3856 YZ |
1376 | if (ret > 0) |
1377 | break; | |
1378 | leaf = path->nodes[0]; | |
1379 | } | |
be20aa9d | 1380 | |
80ff3856 YZ |
1381 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1382 | ||
a6bd9cd1 | 1383 | /* Didn't find anything for our INO */ |
1d512cb7 FM |
1384 | if (found_key.objectid > ino) |
1385 | break; | |
a6bd9cd1 NB |
1386 | /* |
1387 | * Keep searching until we find an EXTENT_ITEM or there are no | |
1388 | * more extents for this inode | |
1389 | */ | |
1d512cb7 FM |
1390 | if (WARN_ON_ONCE(found_key.objectid < ino) || |
1391 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1392 | path->slots[0]++; | |
1393 | goto next_slot; | |
1394 | } | |
a6bd9cd1 NB |
1395 | |
1396 | /* Found key is not EXTENT_DATA_KEY or starts after req range */ | |
1d512cb7 | 1397 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || |
80ff3856 YZ |
1398 | found_key.offset > end) |
1399 | break; | |
1400 | ||
a6bd9cd1 NB |
1401 | /* |
1402 | * If the found extent starts after requested offset, then | |
1403 | * adjust extent_end to be right before this extent begins | |
1404 | */ | |
80ff3856 YZ |
1405 | if (found_key.offset > cur_offset) { |
1406 | extent_end = found_key.offset; | |
e9061e21 | 1407 | extent_type = 0; |
80ff3856 YZ |
1408 | goto out_check; |
1409 | } | |
1410 | ||
a6bd9cd1 NB |
1411 | /* |
1412 | * Found extent which begins before our range and potentially | |
1413 | * intersect it | |
1414 | */ | |
80ff3856 YZ |
1415 | fi = btrfs_item_ptr(leaf, path->slots[0], |
1416 | struct btrfs_file_extent_item); | |
1417 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1418 | ||
cc95bef6 | 1419 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1420 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1421 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1422 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1423 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1424 | extent_end = found_key.offset + |
1425 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1426 | disk_num_bytes = |
1427 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
a6bd9cd1 NB |
1428 | /* |
1429 | * If extent we got ends before our range starts, skip | |
1430 | * to next extent | |
1431 | */ | |
80ff3856 YZ |
1432 | if (extent_end <= start) { |
1433 | path->slots[0]++; | |
1434 | goto next_slot; | |
1435 | } | |
a6bd9cd1 | 1436 | /* Skip holes */ |
17d217fe YZ |
1437 | if (disk_bytenr == 0) |
1438 | goto out_check; | |
a6bd9cd1 | 1439 | /* Skip compressed/encrypted/encoded extents */ |
80ff3856 YZ |
1440 | if (btrfs_file_extent_compression(leaf, fi) || |
1441 | btrfs_file_extent_encryption(leaf, fi) || | |
1442 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1443 | goto out_check; | |
78d4295b | 1444 | /* |
a6bd9cd1 NB |
1445 | * If extent is created before the last volume's snapshot |
1446 | * this implies the extent is shared, hence we can't do | |
1447 | * nocow. This is the same check as in | |
1448 | * btrfs_cross_ref_exist but without calling | |
1449 | * btrfs_search_slot. | |
78d4295b | 1450 | */ |
3e024846 | 1451 | if (!freespace_inode && |
27a7ff55 | 1452 | btrfs_file_extent_generation(leaf, fi) <= |
78d4295b EL |
1453 | btrfs_root_last_snapshot(&root->root_item)) |
1454 | goto out_check; | |
d899e052 YZ |
1455 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1456 | goto out_check; | |
a6bd9cd1 | 1457 | /* If extent is RO, we must COW it */ |
2ff7e61e | 1458 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
80ff3856 | 1459 | goto out_check; |
58113753 LB |
1460 | ret = btrfs_cross_ref_exist(root, ino, |
1461 | found_key.offset - | |
1462 | extent_offset, disk_bytenr); | |
1463 | if (ret) { | |
1464 | /* | |
1465 | * ret could be -EIO if the above fails to read | |
1466 | * metadata. | |
1467 | */ | |
1468 | if (ret < 0) { | |
1469 | if (cow_start != (u64)-1) | |
1470 | cur_offset = cow_start; | |
1471 | goto error; | |
1472 | } | |
1473 | ||
3e024846 | 1474 | WARN_ON_ONCE(freespace_inode); |
17d217fe | 1475 | goto out_check; |
58113753 | 1476 | } |
5d4f98a2 | 1477 | disk_bytenr += extent_offset; |
17d217fe YZ |
1478 | disk_bytenr += cur_offset - found_key.offset; |
1479 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 | 1480 | /* |
a6bd9cd1 NB |
1481 | * If there are pending snapshots for this root, we |
1482 | * fall into common COW way | |
e9894fd3 | 1483 | */ |
3e024846 | 1484 | if (!freespace_inode && atomic_read(&root->snapshot_force_cow)) |
8ecebf4d | 1485 | goto out_check; |
17d217fe YZ |
1486 | /* |
1487 | * force cow if csum exists in the range. | |
1488 | * this ensure that csum for a given extent are | |
1489 | * either valid or do not exist. | |
1490 | */ | |
58113753 LB |
1491 | ret = csum_exist_in_range(fs_info, disk_bytenr, |
1492 | num_bytes); | |
1493 | if (ret) { | |
58113753 LB |
1494 | /* |
1495 | * ret could be -EIO if the above fails to read | |
1496 | * metadata. | |
1497 | */ | |
1498 | if (ret < 0) { | |
1499 | if (cow_start != (u64)-1) | |
1500 | cur_offset = cow_start; | |
1501 | goto error; | |
1502 | } | |
3e024846 | 1503 | WARN_ON_ONCE(freespace_inode); |
17d217fe | 1504 | goto out_check; |
91e1f56a | 1505 | } |
8ecebf4d | 1506 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) |
f78c436c | 1507 | goto out_check; |
3e024846 | 1508 | nocow = true; |
80ff3856 | 1509 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
e8e21007 NB |
1510 | extent_end = found_key.offset + ram_bytes; |
1511 | extent_end = ALIGN(extent_end, fs_info->sectorsize); | |
922f0518 NB |
1512 | /* Skip extents outside of our requested range */ |
1513 | if (extent_end <= start) { | |
1514 | path->slots[0]++; | |
1515 | goto next_slot; | |
1516 | } | |
80ff3856 | 1517 | } else { |
e8e21007 | 1518 | /* If this triggers then we have a memory corruption */ |
290342f6 | 1519 | BUG(); |
80ff3856 YZ |
1520 | } |
1521 | out_check: | |
a6bd9cd1 NB |
1522 | /* |
1523 | * If nocow is false then record the beginning of the range | |
1524 | * that needs to be COWed | |
1525 | */ | |
80ff3856 YZ |
1526 | if (!nocow) { |
1527 | if (cow_start == (u64)-1) | |
1528 | cow_start = cur_offset; | |
1529 | cur_offset = extent_end; | |
1530 | if (cur_offset > end) | |
1531 | break; | |
1532 | path->slots[0]++; | |
1533 | goto next_slot; | |
7ea394f1 YZ |
1534 | } |
1535 | ||
b3b4aa74 | 1536 | btrfs_release_path(path); |
a6bd9cd1 NB |
1537 | |
1538 | /* | |
1539 | * COW range from cow_start to found_key.offset - 1. As the key | |
1540 | * will contain the beginning of the first extent that can be | |
1541 | * NOCOW, following one which needs to be COW'ed | |
1542 | */ | |
80ff3856 | 1543 | if (cow_start != (u64)-1) { |
00361589 JB |
1544 | ret = cow_file_range(inode, locked_page, |
1545 | cow_start, found_key.offset - 1, | |
330a5827 | 1546 | page_started, nr_written, 1); |
e9894fd3 | 1547 | if (ret) { |
f78c436c | 1548 | if (nocow) |
0b246afa | 1549 | btrfs_dec_nocow_writers(fs_info, |
f78c436c | 1550 | disk_bytenr); |
79787eaa | 1551 | goto error; |
e9894fd3 | 1552 | } |
80ff3856 | 1553 | cow_start = (u64)-1; |
7ea394f1 | 1554 | } |
80ff3856 | 1555 | |
d899e052 | 1556 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
6f9994db | 1557 | u64 orig_start = found_key.offset - extent_offset; |
3e024846 | 1558 | struct extent_map *em; |
6f9994db LB |
1559 | |
1560 | em = create_io_em(inode, cur_offset, num_bytes, | |
1561 | orig_start, | |
1562 | disk_bytenr, /* block_start */ | |
1563 | num_bytes, /* block_len */ | |
1564 | disk_num_bytes, /* orig_block_len */ | |
1565 | ram_bytes, BTRFS_COMPRESS_NONE, | |
1566 | BTRFS_ORDERED_PREALLOC); | |
1567 | if (IS_ERR(em)) { | |
6f9994db LB |
1568 | if (nocow) |
1569 | btrfs_dec_nocow_writers(fs_info, | |
1570 | disk_bytenr); | |
1571 | ret = PTR_ERR(em); | |
1572 | goto error; | |
d899e052 | 1573 | } |
6f9994db | 1574 | free_extent_map(em); |
bb55f626 NB |
1575 | ret = btrfs_add_ordered_extent(inode, cur_offset, |
1576 | disk_bytenr, num_bytes, | |
1577 | num_bytes, | |
1578 | BTRFS_ORDERED_PREALLOC); | |
762bf098 NB |
1579 | if (ret) { |
1580 | btrfs_drop_extent_cache(BTRFS_I(inode), | |
1581 | cur_offset, | |
1582 | cur_offset + num_bytes - 1, | |
1583 | 0); | |
1584 | goto error; | |
1585 | } | |
d899e052 | 1586 | } else { |
bb55f626 NB |
1587 | ret = btrfs_add_ordered_extent(inode, cur_offset, |
1588 | disk_bytenr, num_bytes, | |
1589 | num_bytes, | |
1590 | BTRFS_ORDERED_NOCOW); | |
762bf098 NB |
1591 | if (ret) |
1592 | goto error; | |
d899e052 | 1593 | } |
80ff3856 | 1594 | |
f78c436c | 1595 | if (nocow) |
0b246afa | 1596 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
762bf098 | 1597 | nocow = false; |
771ed689 | 1598 | |
efa56464 | 1599 | if (root->root_key.objectid == |
4dbd80fb QW |
1600 | BTRFS_DATA_RELOC_TREE_OBJECTID) |
1601 | /* | |
1602 | * Error handled later, as we must prevent | |
1603 | * extent_clear_unlock_delalloc() in error handler | |
1604 | * from freeing metadata of created ordered extent. | |
1605 | */ | |
efa56464 YZ |
1606 | ret = btrfs_reloc_clone_csums(inode, cur_offset, |
1607 | num_bytes); | |
efa56464 | 1608 | |
c2790a2e | 1609 | extent_clear_unlock_delalloc(inode, cur_offset, |
74e9194a | 1610 | cur_offset + num_bytes - 1, |
c2790a2e | 1611 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1612 | EXTENT_DELALLOC | |
1613 | EXTENT_CLEAR_DATA_RESV, | |
1614 | PAGE_UNLOCK | PAGE_SET_PRIVATE2); | |
1615 | ||
80ff3856 | 1616 | cur_offset = extent_end; |
4dbd80fb QW |
1617 | |
1618 | /* | |
1619 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
1620 | * handler, as metadata for created ordered extent will only | |
1621 | * be freed by btrfs_finish_ordered_io(). | |
1622 | */ | |
1623 | if (ret) | |
1624 | goto error; | |
80ff3856 YZ |
1625 | if (cur_offset > end) |
1626 | break; | |
be20aa9d | 1627 | } |
b3b4aa74 | 1628 | btrfs_release_path(path); |
80ff3856 | 1629 | |
506481b2 | 1630 | if (cur_offset <= end && cow_start == (u64)-1) |
80ff3856 | 1631 | cow_start = cur_offset; |
17ca04af | 1632 | |
80ff3856 | 1633 | if (cow_start != (u64)-1) { |
506481b2 | 1634 | cur_offset = end; |
74e9194a | 1635 | ret = cow_file_range(inode, locked_page, cow_start, end, |
330a5827 | 1636 | page_started, nr_written, 1); |
d788a349 | 1637 | if (ret) |
79787eaa | 1638 | goto error; |
80ff3856 YZ |
1639 | } |
1640 | ||
79787eaa | 1641 | error: |
762bf098 NB |
1642 | if (nocow) |
1643 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); | |
1644 | ||
17ca04af | 1645 | if (ret && cur_offset < end) |
74e9194a | 1646 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
c2790a2e | 1647 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1648 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1649 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1650 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1651 | PAGE_SET_WRITEBACK | |
1652 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1653 | btrfs_free_path(path); |
79787eaa | 1654 | return ret; |
be20aa9d CM |
1655 | } |
1656 | ||
47059d93 WS |
1657 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1658 | { | |
1659 | ||
1660 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1661 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1662 | return 0; | |
1663 | ||
1664 | /* | |
1665 | * @defrag_bytes is a hint value, no spinlock held here, | |
1666 | * if is not zero, it means the file is defragging. | |
1667 | * Force cow if given extent needs to be defragged. | |
1668 | */ | |
1669 | if (BTRFS_I(inode)->defrag_bytes && | |
1670 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1671 | EXTENT_DEFRAG, 0, NULL)) | |
1672 | return 1; | |
1673 | ||
1674 | return 0; | |
1675 | } | |
1676 | ||
d352ac68 | 1677 | /* |
5eaad97a NB |
1678 | * Function to process delayed allocation (create CoW) for ranges which are |
1679 | * being touched for the first time. | |
d352ac68 | 1680 | */ |
bc9a8bf7 | 1681 | int btrfs_run_delalloc_range(struct inode *inode, struct page *locked_page, |
5eaad97a NB |
1682 | u64 start, u64 end, int *page_started, unsigned long *nr_written, |
1683 | struct writeback_control *wbc) | |
be20aa9d | 1684 | { |
be20aa9d | 1685 | int ret; |
47059d93 | 1686 | int force_cow = need_force_cow(inode, start, end); |
f82b7359 | 1687 | unsigned int write_flags = wbc_to_write_flags(wbc); |
a2135011 | 1688 | |
47059d93 | 1689 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1690 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1691 | page_started, 1, nr_written); |
47059d93 | 1692 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1693 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1694 | page_started, 0, nr_written); |
42c16da6 QW |
1695 | } else if (!inode_can_compress(inode) || |
1696 | !inode_need_compress(inode, start, end)) { | |
74e9194a | 1697 | ret = cow_file_range(inode, locked_page, start, end, |
330a5827 | 1698 | page_started, nr_written, 1); |
7ddf5a42 JB |
1699 | } else { |
1700 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1701 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1702 | ret = cow_file_range_async(inode, locked_page, start, end, |
f82b7359 LB |
1703 | page_started, nr_written, |
1704 | write_flags); | |
7ddf5a42 | 1705 | } |
52427260 | 1706 | if (ret) |
d1051d6e NB |
1707 | btrfs_cleanup_ordered_extents(inode, locked_page, start, |
1708 | end - start + 1); | |
b888db2b CM |
1709 | return ret; |
1710 | } | |
1711 | ||
abbb55f4 NB |
1712 | void btrfs_split_delalloc_extent(struct inode *inode, |
1713 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1714 | { |
dcab6a3b JB |
1715 | u64 size; |
1716 | ||
0ca1f7ce | 1717 | /* not delalloc, ignore it */ |
9ed74f2d | 1718 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1719 | return; |
9ed74f2d | 1720 | |
dcab6a3b JB |
1721 | size = orig->end - orig->start + 1; |
1722 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 1723 | u32 num_extents; |
dcab6a3b JB |
1724 | u64 new_size; |
1725 | ||
1726 | /* | |
5c848198 | 1727 | * See the explanation in btrfs_merge_delalloc_extent, the same |
ba117213 | 1728 | * applies here, just in reverse. |
dcab6a3b JB |
1729 | */ |
1730 | new_size = orig->end - split + 1; | |
823bb20a | 1731 | num_extents = count_max_extents(new_size); |
ba117213 | 1732 | new_size = split - orig->start; |
823bb20a DS |
1733 | num_extents += count_max_extents(new_size); |
1734 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
1735 | return; |
1736 | } | |
1737 | ||
9e0baf60 | 1738 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1739 | btrfs_mod_outstanding_extents(BTRFS_I(inode), 1); |
9e0baf60 | 1740 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1741 | } |
1742 | ||
1743 | /* | |
5c848198 NB |
1744 | * Handle merged delayed allocation extents so we can keep track of new extents |
1745 | * that are just merged onto old extents, such as when we are doing sequential | |
1746 | * writes, so we can properly account for the metadata space we'll need. | |
9ed74f2d | 1747 | */ |
5c848198 NB |
1748 | void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new, |
1749 | struct extent_state *other) | |
9ed74f2d | 1750 | { |
dcab6a3b | 1751 | u64 new_size, old_size; |
823bb20a | 1752 | u32 num_extents; |
dcab6a3b | 1753 | |
9ed74f2d JB |
1754 | /* not delalloc, ignore it */ |
1755 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1756 | return; |
9ed74f2d | 1757 | |
8461a3de JB |
1758 | if (new->start > other->start) |
1759 | new_size = new->end - other->start + 1; | |
1760 | else | |
1761 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1762 | |
1763 | /* we're not bigger than the max, unreserve the space and go */ | |
1764 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1765 | spin_lock(&BTRFS_I(inode)->lock); | |
8b62f87b | 1766 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
dcab6a3b JB |
1767 | spin_unlock(&BTRFS_I(inode)->lock); |
1768 | return; | |
1769 | } | |
1770 | ||
1771 | /* | |
ba117213 JB |
1772 | * We have to add up either side to figure out how many extents were |
1773 | * accounted for before we merged into one big extent. If the number of | |
1774 | * extents we accounted for is <= the amount we need for the new range | |
1775 | * then we can return, otherwise drop. Think of it like this | |
1776 | * | |
1777 | * [ 4k][MAX_SIZE] | |
1778 | * | |
1779 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1780 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1781 | * we have 1 so they are == and we can return. But in this case | |
1782 | * | |
1783 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1784 | * | |
1785 | * Each range on their own accounts for 2 extents, but merged together | |
1786 | * they are only 3 extents worth of accounting, so we need to drop in | |
1787 | * this case. | |
dcab6a3b | 1788 | */ |
ba117213 | 1789 | old_size = other->end - other->start + 1; |
823bb20a | 1790 | num_extents = count_max_extents(old_size); |
ba117213 | 1791 | old_size = new->end - new->start + 1; |
823bb20a DS |
1792 | num_extents += count_max_extents(old_size); |
1793 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
1794 | return; |
1795 | ||
9e0baf60 | 1796 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1797 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
9e0baf60 | 1798 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1799 | } |
1800 | ||
eb73c1b7 MX |
1801 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1802 | struct inode *inode) | |
1803 | { | |
0b246afa JM |
1804 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1805 | ||
eb73c1b7 MX |
1806 | spin_lock(&root->delalloc_lock); |
1807 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1808 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1809 | &root->delalloc_inodes); | |
1810 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1811 | &BTRFS_I(inode)->runtime_flags); | |
1812 | root->nr_delalloc_inodes++; | |
1813 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 1814 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1815 | BUG_ON(!list_empty(&root->delalloc_root)); |
1816 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
1817 | &fs_info->delalloc_roots); |
1818 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
1819 | } |
1820 | } | |
1821 | spin_unlock(&root->delalloc_lock); | |
1822 | } | |
1823 | ||
2b877331 NB |
1824 | |
1825 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1826 | struct btrfs_inode *inode) | |
eb73c1b7 | 1827 | { |
3ffbd68c | 1828 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b246afa | 1829 | |
9e3e97f4 NB |
1830 | if (!list_empty(&inode->delalloc_inodes)) { |
1831 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 1832 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1833 | &inode->runtime_flags); |
eb73c1b7 MX |
1834 | root->nr_delalloc_inodes--; |
1835 | if (!root->nr_delalloc_inodes) { | |
7c8a0d36 | 1836 | ASSERT(list_empty(&root->delalloc_inodes)); |
0b246afa | 1837 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1838 | BUG_ON(list_empty(&root->delalloc_root)); |
1839 | list_del_init(&root->delalloc_root); | |
0b246afa | 1840 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1841 | } |
1842 | } | |
2b877331 NB |
1843 | } |
1844 | ||
1845 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1846 | struct btrfs_inode *inode) | |
1847 | { | |
1848 | spin_lock(&root->delalloc_lock); | |
1849 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
1850 | spin_unlock(&root->delalloc_lock); |
1851 | } | |
1852 | ||
d352ac68 | 1853 | /* |
e06a1fc9 NB |
1854 | * Properly track delayed allocation bytes in the inode and to maintain the |
1855 | * list of inodes that have pending delalloc work to be done. | |
d352ac68 | 1856 | */ |
e06a1fc9 NB |
1857 | void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state, |
1858 | unsigned *bits) | |
291d673e | 1859 | { |
0b246afa JM |
1860 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1861 | ||
47059d93 WS |
1862 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1863 | WARN_ON(1); | |
75eff68e CM |
1864 | /* |
1865 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1866 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1867 | * bit, which is only set or cleared with irqs on |
1868 | */ | |
0ca1f7ce | 1869 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1870 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1871 | u64 len = state->end + 1 - state->start; |
8b62f87b | 1872 | u32 num_extents = count_max_extents(len); |
70ddc553 | 1873 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 1874 | |
8b62f87b JB |
1875 | spin_lock(&BTRFS_I(inode)->lock); |
1876 | btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents); | |
1877 | spin_unlock(&BTRFS_I(inode)->lock); | |
287a0ab9 | 1878 | |
6a3891c5 | 1879 | /* For sanity tests */ |
0b246afa | 1880 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1881 | return; |
1882 | ||
104b4e51 NB |
1883 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
1884 | fs_info->delalloc_batch); | |
df0af1a5 | 1885 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1886 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1887 | if (*bits & EXTENT_DEFRAG) |
1888 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1889 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1890 | &BTRFS_I(inode)->runtime_flags)) |
1891 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1892 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1893 | } |
a7e3b975 FM |
1894 | |
1895 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
1896 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1897 | spin_lock(&BTRFS_I(inode)->lock); | |
1898 | BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 - | |
1899 | state->start; | |
1900 | spin_unlock(&BTRFS_I(inode)->lock); | |
1901 | } | |
291d673e CM |
1902 | } |
1903 | ||
d352ac68 | 1904 | /* |
a36bb5f9 NB |
1905 | * Once a range is no longer delalloc this function ensures that proper |
1906 | * accounting happens. | |
d352ac68 | 1907 | */ |
a36bb5f9 NB |
1908 | void btrfs_clear_delalloc_extent(struct inode *vfs_inode, |
1909 | struct extent_state *state, unsigned *bits) | |
291d673e | 1910 | { |
a36bb5f9 NB |
1911 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
1912 | struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb); | |
47059d93 | 1913 | u64 len = state->end + 1 - state->start; |
823bb20a | 1914 | u32 num_extents = count_max_extents(len); |
47059d93 | 1915 | |
4a4b964f FM |
1916 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) { |
1917 | spin_lock(&inode->lock); | |
6fc0ef68 | 1918 | inode->defrag_bytes -= len; |
4a4b964f FM |
1919 | spin_unlock(&inode->lock); |
1920 | } | |
47059d93 | 1921 | |
75eff68e CM |
1922 | /* |
1923 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1924 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1925 | * bit, which is only set or cleared with irqs on |
1926 | */ | |
0ca1f7ce | 1927 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 1928 | struct btrfs_root *root = inode->root; |
83eea1f1 | 1929 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1930 | |
8b62f87b JB |
1931 | spin_lock(&inode->lock); |
1932 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
1933 | spin_unlock(&inode->lock); | |
0ca1f7ce | 1934 | |
b6d08f06 JB |
1935 | /* |
1936 | * We don't reserve metadata space for space cache inodes so we | |
52042d8e | 1937 | * don't need to call delalloc_release_metadata if there is an |
b6d08f06 JB |
1938 | * error. |
1939 | */ | |
a315e68f | 1940 | if (*bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 1941 | root != fs_info->tree_root) |
43b18595 | 1942 | btrfs_delalloc_release_metadata(inode, len, false); |
0ca1f7ce | 1943 | |
6a3891c5 | 1944 | /* For sanity tests. */ |
0b246afa | 1945 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1946 | return; |
1947 | ||
a315e68f FM |
1948 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID && |
1949 | do_list && !(state->state & EXTENT_NORESERVE) && | |
1950 | (*bits & EXTENT_CLEAR_DATA_RESV)) | |
6fc0ef68 NB |
1951 | btrfs_free_reserved_data_space_noquota( |
1952 | &inode->vfs_inode, | |
51773bec | 1953 | state->start, len); |
9ed74f2d | 1954 | |
104b4e51 NB |
1955 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
1956 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
1957 | spin_lock(&inode->lock); |
1958 | inode->delalloc_bytes -= len; | |
1959 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 1960 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1961 | &inode->runtime_flags)) |
eb73c1b7 | 1962 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 1963 | spin_unlock(&inode->lock); |
291d673e | 1964 | } |
a7e3b975 FM |
1965 | |
1966 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
1967 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1968 | spin_lock(&inode->lock); | |
1969 | ASSERT(inode->new_delalloc_bytes >= len); | |
1970 | inode->new_delalloc_bytes -= len; | |
1971 | spin_unlock(&inode->lock); | |
1972 | } | |
291d673e CM |
1973 | } |
1974 | ||
d352ac68 | 1975 | /* |
da12fe54 NB |
1976 | * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit |
1977 | * in a chunk's stripe. This function ensures that bios do not span a | |
1978 | * stripe/chunk | |
6f034ece | 1979 | * |
da12fe54 NB |
1980 | * @page - The page we are about to add to the bio |
1981 | * @size - size we want to add to the bio | |
1982 | * @bio - bio we want to ensure is smaller than a stripe | |
1983 | * @bio_flags - flags of the bio | |
1984 | * | |
1985 | * return 1 if page cannot be added to the bio | |
1986 | * return 0 if page can be added to the bio | |
6f034ece | 1987 | * return error otherwise |
d352ac68 | 1988 | */ |
da12fe54 NB |
1989 | int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio, |
1990 | unsigned long bio_flags) | |
239b14b3 | 1991 | { |
0b246afa JM |
1992 | struct inode *inode = page->mapping->host; |
1993 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
4f024f37 | 1994 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1995 | u64 length = 0; |
1996 | u64 map_length; | |
239b14b3 | 1997 | int ret; |
89b798ad | 1998 | struct btrfs_io_geometry geom; |
239b14b3 | 1999 | |
771ed689 CM |
2000 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
2001 | return 0; | |
2002 | ||
4f024f37 | 2003 | length = bio->bi_iter.bi_size; |
239b14b3 | 2004 | map_length = length; |
89b798ad NB |
2005 | ret = btrfs_get_io_geometry(fs_info, btrfs_op(bio), logical, map_length, |
2006 | &geom); | |
6f034ece LB |
2007 | if (ret < 0) |
2008 | return ret; | |
89b798ad NB |
2009 | |
2010 | if (geom.len < length + size) | |
239b14b3 | 2011 | return 1; |
3444a972 | 2012 | return 0; |
239b14b3 CM |
2013 | } |
2014 | ||
d352ac68 CM |
2015 | /* |
2016 | * in order to insert checksums into the metadata in large chunks, | |
2017 | * we wait until bio submission time. All the pages in the bio are | |
2018 | * checksummed and sums are attached onto the ordered extent record. | |
2019 | * | |
2020 | * At IO completion time the cums attached on the ordered extent record | |
2021 | * are inserted into the btree | |
2022 | */ | |
d0ee3934 | 2023 | static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio, |
eaf25d93 | 2024 | u64 bio_offset) |
065631f6 | 2025 | { |
c6100a4b | 2026 | struct inode *inode = private_data; |
4e4cbee9 | 2027 | blk_status_t ret = 0; |
e015640f | 2028 | |
2ff7e61e | 2029 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
79787eaa | 2030 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
2031 | return 0; |
2032 | } | |
e015640f | 2033 | |
d352ac68 | 2034 | /* |
cad321ad | 2035 | * extent_io.c submission hook. This does the right thing for csum calculation |
4c274bc6 LB |
2036 | * on write, or reading the csums from the tree before a read. |
2037 | * | |
2038 | * Rules about async/sync submit, | |
2039 | * a) read: sync submit | |
2040 | * | |
2041 | * b) write without checksum: sync submit | |
2042 | * | |
2043 | * c) write with checksum: | |
2044 | * c-1) if bio is issued by fsync: sync submit | |
2045 | * (sync_writers != 0) | |
2046 | * | |
2047 | * c-2) if root is reloc root: sync submit | |
2048 | * (only in case of buffered IO) | |
2049 | * | |
2050 | * c-3) otherwise: async submit | |
d352ac68 | 2051 | */ |
a56b1c7b | 2052 | static blk_status_t btrfs_submit_bio_hook(struct inode *inode, struct bio *bio, |
50489a57 NB |
2053 | int mirror_num, |
2054 | unsigned long bio_flags) | |
2055 | ||
44b8bd7e | 2056 | { |
0b246afa | 2057 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 2058 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 2059 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
4e4cbee9 | 2060 | blk_status_t ret = 0; |
19b9bdb0 | 2061 | int skip_sum; |
b812ce28 | 2062 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 2063 | |
6cbff00f | 2064 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 2065 | |
70ddc553 | 2066 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 2067 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 2068 | |
37226b21 | 2069 | if (bio_op(bio) != REQ_OP_WRITE) { |
0b246afa | 2070 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 2071 | if (ret) |
61891923 | 2072 | goto out; |
5fd02043 | 2073 | |
d20f7043 | 2074 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
2075 | ret = btrfs_submit_compressed_read(inode, bio, |
2076 | mirror_num, | |
2077 | bio_flags); | |
2078 | goto out; | |
c2db1073 | 2079 | } else if (!skip_sum) { |
2ff7e61e | 2080 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 2081 | if (ret) |
61891923 | 2082 | goto out; |
c2db1073 | 2083 | } |
4d1b5fb4 | 2084 | goto mapit; |
b812ce28 | 2085 | } else if (async && !skip_sum) { |
17d217fe YZ |
2086 | /* csum items have already been cloned */ |
2087 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
2088 | goto mapit; | |
19b9bdb0 | 2089 | /* we're doing a write, do the async checksumming */ |
c6100a4b | 2090 | ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags, |
e7681167 | 2091 | 0, inode, btrfs_submit_bio_start); |
61891923 | 2092 | goto out; |
b812ce28 | 2093 | } else if (!skip_sum) { |
2ff7e61e | 2094 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
b812ce28 JB |
2095 | if (ret) |
2096 | goto out; | |
19b9bdb0 CM |
2097 | } |
2098 | ||
0b86a832 | 2099 | mapit: |
2ff7e61e | 2100 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
61891923 SB |
2101 | |
2102 | out: | |
4e4cbee9 CH |
2103 | if (ret) { |
2104 | bio->bi_status = ret; | |
4246a0b6 CH |
2105 | bio_endio(bio); |
2106 | } | |
61891923 | 2107 | return ret; |
065631f6 | 2108 | } |
6885f308 | 2109 | |
d352ac68 CM |
2110 | /* |
2111 | * given a list of ordered sums record them in the inode. This happens | |
2112 | * at IO completion time based on sums calculated at bio submission time. | |
2113 | */ | |
ba1da2f4 | 2114 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
df9f628e | 2115 | struct inode *inode, struct list_head *list) |
e6dcd2dc | 2116 | { |
e6dcd2dc | 2117 | struct btrfs_ordered_sum *sum; |
ac01f26a | 2118 | int ret; |
e6dcd2dc | 2119 | |
c6e30871 | 2120 | list_for_each_entry(sum, list, list) { |
7c2871a2 | 2121 | trans->adding_csums = true; |
ac01f26a | 2122 | ret = btrfs_csum_file_blocks(trans, |
d20f7043 | 2123 | BTRFS_I(inode)->root->fs_info->csum_root, sum); |
7c2871a2 | 2124 | trans->adding_csums = false; |
ac01f26a NB |
2125 | if (ret) |
2126 | return ret; | |
e6dcd2dc CM |
2127 | } |
2128 | return 0; | |
2129 | } | |
2130 | ||
2ac55d41 | 2131 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
e3b8a485 | 2132 | unsigned int extra_bits, |
330a5827 | 2133 | struct extent_state **cached_state) |
ea8c2819 | 2134 | { |
fdb1e121 | 2135 | WARN_ON(PAGE_ALIGNED(end)); |
ea8c2819 | 2136 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
e3b8a485 | 2137 | extra_bits, cached_state); |
ea8c2819 CM |
2138 | } |
2139 | ||
d352ac68 | 2140 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2141 | struct btrfs_writepage_fixup { |
2142 | struct page *page; | |
2143 | struct btrfs_work work; | |
2144 | }; | |
2145 | ||
b2950863 | 2146 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2147 | { |
2148 | struct btrfs_writepage_fixup *fixup; | |
2149 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2150 | struct extent_state *cached_state = NULL; |
364ecf36 | 2151 | struct extent_changeset *data_reserved = NULL; |
247e743c CM |
2152 | struct page *page; |
2153 | struct inode *inode; | |
2154 | u64 page_start; | |
2155 | u64 page_end; | |
87826df0 | 2156 | int ret; |
247e743c CM |
2157 | |
2158 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2159 | page = fixup->page; | |
4a096752 | 2160 | again: |
247e743c CM |
2161 | lock_page(page); |
2162 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
2163 | ClearPageChecked(page); | |
2164 | goto out_page; | |
2165 | } | |
2166 | ||
2167 | inode = page->mapping->host; | |
2168 | page_start = page_offset(page); | |
09cbfeaf | 2169 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 2170 | |
ff13db41 | 2171 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 2172 | &cached_state); |
4a096752 CM |
2173 | |
2174 | /* already ordered? We're done */ | |
8b62b72b | 2175 | if (PagePrivate2(page)) |
247e743c | 2176 | goto out; |
4a096752 | 2177 | |
a776c6fa | 2178 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
09cbfeaf | 2179 | PAGE_SIZE); |
4a096752 | 2180 | if (ordered) { |
2ac55d41 | 2181 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
e43bbe5e | 2182 | page_end, &cached_state); |
4a096752 CM |
2183 | unlock_page(page); |
2184 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 2185 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2186 | goto again; |
2187 | } | |
247e743c | 2188 | |
364ecf36 | 2189 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
09cbfeaf | 2190 | PAGE_SIZE); |
87826df0 JM |
2191 | if (ret) { |
2192 | mapping_set_error(page->mapping, ret); | |
2193 | end_extent_writepage(page, ret, page_start, page_end); | |
2194 | ClearPageChecked(page); | |
2195 | goto out; | |
2196 | } | |
2197 | ||
f3038ee3 | 2198 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, |
330a5827 | 2199 | &cached_state); |
f3038ee3 NB |
2200 | if (ret) { |
2201 | mapping_set_error(page->mapping, ret); | |
2202 | end_extent_writepage(page, ret, page_start, page_end); | |
2203 | ClearPageChecked(page); | |
2204 | goto out; | |
2205 | } | |
2206 | ||
247e743c | 2207 | ClearPageChecked(page); |
87826df0 | 2208 | set_page_dirty(page); |
43b18595 | 2209 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false); |
247e743c | 2210 | out: |
2ac55d41 | 2211 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
e43bbe5e | 2212 | &cached_state); |
247e743c CM |
2213 | out_page: |
2214 | unlock_page(page); | |
09cbfeaf | 2215 | put_page(page); |
b897abec | 2216 | kfree(fixup); |
364ecf36 | 2217 | extent_changeset_free(data_reserved); |
247e743c CM |
2218 | } |
2219 | ||
2220 | /* | |
2221 | * There are a few paths in the higher layers of the kernel that directly | |
2222 | * set the page dirty bit without asking the filesystem if it is a | |
2223 | * good idea. This causes problems because we want to make sure COW | |
2224 | * properly happens and the data=ordered rules are followed. | |
2225 | * | |
c8b97818 | 2226 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2227 | * hasn't been properly setup for IO. We kick off an async process |
2228 | * to fix it up. The async helper will wait for ordered extents, set | |
2229 | * the delalloc bit and make it safe to write the page. | |
2230 | */ | |
d75855b4 | 2231 | int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end) |
247e743c CM |
2232 | { |
2233 | struct inode *inode = page->mapping->host; | |
0b246afa | 2234 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2235 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2236 | |
8b62b72b CM |
2237 | /* this page is properly in the ordered list */ |
2238 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2239 | return 0; |
2240 | ||
2241 | if (PageChecked(page)) | |
2242 | return -EAGAIN; | |
2243 | ||
2244 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2245 | if (!fixup) | |
2246 | return -EAGAIN; | |
f421950f | 2247 | |
247e743c | 2248 | SetPageChecked(page); |
09cbfeaf | 2249 | get_page(page); |
9e0af237 LB |
2250 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2251 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2252 | fixup->page = page; |
0b246afa | 2253 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
87826df0 | 2254 | return -EBUSY; |
247e743c CM |
2255 | } |
2256 | ||
d899e052 YZ |
2257 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2258 | struct inode *inode, u64 file_pos, | |
2259 | u64 disk_bytenr, u64 disk_num_bytes, | |
2260 | u64 num_bytes, u64 ram_bytes, | |
2261 | u8 compression, u8 encryption, | |
2262 | u16 other_encoding, int extent_type) | |
2263 | { | |
2264 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2265 | struct btrfs_file_extent_item *fi; | |
2266 | struct btrfs_path *path; | |
2267 | struct extent_buffer *leaf; | |
2268 | struct btrfs_key ins; | |
a12b877b | 2269 | u64 qg_released; |
1acae57b | 2270 | int extent_inserted = 0; |
d899e052 YZ |
2271 | int ret; |
2272 | ||
2273 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2274 | if (!path) |
2275 | return -ENOMEM; | |
d899e052 | 2276 | |
a1ed835e CM |
2277 | /* |
2278 | * we may be replacing one extent in the tree with another. | |
2279 | * The new extent is pinned in the extent map, and we don't want | |
2280 | * to drop it from the cache until it is completely in the btree. | |
2281 | * | |
2282 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2283 | * the caller is expected to unpin it and allow it to be merged | |
2284 | * with the others. | |
2285 | */ | |
1acae57b FDBM |
2286 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2287 | file_pos + num_bytes, NULL, 0, | |
2288 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2289 | if (ret) |
2290 | goto out; | |
d899e052 | 2291 | |
1acae57b | 2292 | if (!extent_inserted) { |
4a0cc7ca | 2293 | ins.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b FDBM |
2294 | ins.offset = file_pos; |
2295 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2296 | ||
2297 | path->leave_spinning = 1; | |
2298 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2299 | sizeof(*fi)); | |
2300 | if (ret) | |
2301 | goto out; | |
2302 | } | |
d899e052 YZ |
2303 | leaf = path->nodes[0]; |
2304 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2305 | struct btrfs_file_extent_item); | |
2306 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2307 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2308 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2309 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2310 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2311 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2312 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2313 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2314 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2315 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2316 | |
d899e052 | 2317 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2318 | btrfs_release_path(path); |
d899e052 YZ |
2319 | |
2320 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2321 | |
2322 | ins.objectid = disk_bytenr; | |
2323 | ins.offset = disk_num_bytes; | |
2324 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2325 | |
297d750b | 2326 | /* |
5846a3c2 QW |
2327 | * Release the reserved range from inode dirty range map, as it is |
2328 | * already moved into delayed_ref_head | |
297d750b | 2329 | */ |
a12b877b QW |
2330 | ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes); |
2331 | if (ret < 0) | |
2332 | goto out; | |
2333 | qg_released = ret; | |
84f7d8e6 JB |
2334 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2335 | btrfs_ino(BTRFS_I(inode)), | |
2336 | file_pos, qg_released, &ins); | |
79787eaa | 2337 | out: |
d899e052 | 2338 | btrfs_free_path(path); |
b9473439 | 2339 | |
79787eaa | 2340 | return ret; |
d899e052 YZ |
2341 | } |
2342 | ||
38c227d8 LB |
2343 | /* snapshot-aware defrag */ |
2344 | struct sa_defrag_extent_backref { | |
2345 | struct rb_node node; | |
2346 | struct old_sa_defrag_extent *old; | |
2347 | u64 root_id; | |
2348 | u64 inum; | |
2349 | u64 file_pos; | |
2350 | u64 extent_offset; | |
2351 | u64 num_bytes; | |
2352 | u64 generation; | |
2353 | }; | |
2354 | ||
2355 | struct old_sa_defrag_extent { | |
2356 | struct list_head list; | |
2357 | struct new_sa_defrag_extent *new; | |
2358 | ||
2359 | u64 extent_offset; | |
2360 | u64 bytenr; | |
2361 | u64 offset; | |
2362 | u64 len; | |
2363 | int count; | |
2364 | }; | |
2365 | ||
2366 | struct new_sa_defrag_extent { | |
2367 | struct rb_root root; | |
2368 | struct list_head head; | |
2369 | struct btrfs_path *path; | |
2370 | struct inode *inode; | |
2371 | u64 file_pos; | |
2372 | u64 len; | |
2373 | u64 bytenr; | |
2374 | u64 disk_len; | |
2375 | u8 compress_type; | |
2376 | }; | |
2377 | ||
2378 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2379 | struct sa_defrag_extent_backref *b2) | |
2380 | { | |
2381 | if (b1->root_id < b2->root_id) | |
2382 | return -1; | |
2383 | else if (b1->root_id > b2->root_id) | |
2384 | return 1; | |
2385 | ||
2386 | if (b1->inum < b2->inum) | |
2387 | return -1; | |
2388 | else if (b1->inum > b2->inum) | |
2389 | return 1; | |
2390 | ||
2391 | if (b1->file_pos < b2->file_pos) | |
2392 | return -1; | |
2393 | else if (b1->file_pos > b2->file_pos) | |
2394 | return 1; | |
2395 | ||
2396 | /* | |
2397 | * [------------------------------] ===> (a range of space) | |
2398 | * |<--->| |<---->| =============> (fs/file tree A) | |
2399 | * |<---------------------------->| ===> (fs/file tree B) | |
2400 | * | |
2401 | * A range of space can refer to two file extents in one tree while | |
2402 | * refer to only one file extent in another tree. | |
2403 | * | |
2404 | * So we may process a disk offset more than one time(two extents in A) | |
2405 | * and locate at the same extent(one extent in B), then insert two same | |
2406 | * backrefs(both refer to the extent in B). | |
2407 | */ | |
2408 | return 0; | |
2409 | } | |
2410 | ||
2411 | static void backref_insert(struct rb_root *root, | |
2412 | struct sa_defrag_extent_backref *backref) | |
2413 | { | |
2414 | struct rb_node **p = &root->rb_node; | |
2415 | struct rb_node *parent = NULL; | |
2416 | struct sa_defrag_extent_backref *entry; | |
2417 | int ret; | |
2418 | ||
2419 | while (*p) { | |
2420 | parent = *p; | |
2421 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2422 | ||
2423 | ret = backref_comp(backref, entry); | |
2424 | if (ret < 0) | |
2425 | p = &(*p)->rb_left; | |
2426 | else | |
2427 | p = &(*p)->rb_right; | |
2428 | } | |
2429 | ||
2430 | rb_link_node(&backref->node, parent, p); | |
2431 | rb_insert_color(&backref->node, root); | |
2432 | } | |
2433 | ||
2434 | /* | |
2435 | * Note the backref might has changed, and in this case we just return 0. | |
2436 | */ | |
2437 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2438 | void *ctx) | |
2439 | { | |
2440 | struct btrfs_file_extent_item *extent; | |
38c227d8 LB |
2441 | struct old_sa_defrag_extent *old = ctx; |
2442 | struct new_sa_defrag_extent *new = old->new; | |
2443 | struct btrfs_path *path = new->path; | |
2444 | struct btrfs_key key; | |
2445 | struct btrfs_root *root; | |
2446 | struct sa_defrag_extent_backref *backref; | |
2447 | struct extent_buffer *leaf; | |
2448 | struct inode *inode = new->inode; | |
0b246afa | 2449 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2450 | int slot; |
2451 | int ret; | |
2452 | u64 extent_offset; | |
2453 | u64 num_bytes; | |
2454 | ||
2455 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
4a0cc7ca | 2456 | inum == btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2457 | return 0; |
2458 | ||
2459 | key.objectid = root_id; | |
2460 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2461 | key.offset = (u64)-1; | |
2462 | ||
38c227d8 LB |
2463 | root = btrfs_read_fs_root_no_name(fs_info, &key); |
2464 | if (IS_ERR(root)) { | |
2465 | if (PTR_ERR(root) == -ENOENT) | |
2466 | return 0; | |
2467 | WARN_ON(1); | |
ab8d0fc4 | 2468 | btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu", |
38c227d8 LB |
2469 | inum, offset, root_id); |
2470 | return PTR_ERR(root); | |
2471 | } | |
2472 | ||
2473 | key.objectid = inum; | |
2474 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2475 | if (offset > (u64)-1 << 32) | |
2476 | key.offset = 0; | |
2477 | else | |
2478 | key.offset = offset; | |
2479 | ||
2480 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2481 | if (WARN_ON(ret < 0)) |
38c227d8 | 2482 | return ret; |
50f1319c | 2483 | ret = 0; |
38c227d8 LB |
2484 | |
2485 | while (1) { | |
2486 | cond_resched(); | |
2487 | ||
2488 | leaf = path->nodes[0]; | |
2489 | slot = path->slots[0]; | |
2490 | ||
2491 | if (slot >= btrfs_header_nritems(leaf)) { | |
2492 | ret = btrfs_next_leaf(root, path); | |
2493 | if (ret < 0) { | |
2494 | goto out; | |
2495 | } else if (ret > 0) { | |
2496 | ret = 0; | |
2497 | goto out; | |
2498 | } | |
2499 | continue; | |
2500 | } | |
2501 | ||
2502 | path->slots[0]++; | |
2503 | ||
2504 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2505 | ||
2506 | if (key.objectid > inum) | |
2507 | goto out; | |
2508 | ||
2509 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2510 | continue; | |
2511 | ||
2512 | extent = btrfs_item_ptr(leaf, slot, | |
2513 | struct btrfs_file_extent_item); | |
2514 | ||
2515 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2516 | continue; | |
2517 | ||
e68afa49 LB |
2518 | /* |
2519 | * 'offset' refers to the exact key.offset, | |
2520 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2521 | * (key.offset - extent_offset). | |
2522 | */ | |
2523 | if (key.offset != offset) | |
38c227d8 LB |
2524 | continue; |
2525 | ||
e68afa49 | 2526 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2527 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2528 | |
38c227d8 LB |
2529 | if (extent_offset >= old->extent_offset + old->offset + |
2530 | old->len || extent_offset + num_bytes <= | |
2531 | old->extent_offset + old->offset) | |
2532 | continue; | |
38c227d8 LB |
2533 | break; |
2534 | } | |
2535 | ||
2536 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2537 | if (!backref) { | |
2538 | ret = -ENOENT; | |
2539 | goto out; | |
2540 | } | |
2541 | ||
2542 | backref->root_id = root_id; | |
2543 | backref->inum = inum; | |
e68afa49 | 2544 | backref->file_pos = offset; |
38c227d8 LB |
2545 | backref->num_bytes = num_bytes; |
2546 | backref->extent_offset = extent_offset; | |
2547 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2548 | backref->old = old; | |
2549 | backref_insert(&new->root, backref); | |
2550 | old->count++; | |
2551 | out: | |
2552 | btrfs_release_path(path); | |
2553 | WARN_ON(ret); | |
2554 | return ret; | |
2555 | } | |
2556 | ||
2557 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2558 | struct new_sa_defrag_extent *new) | |
2559 | { | |
0b246afa | 2560 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2561 | struct old_sa_defrag_extent *old, *tmp; |
2562 | int ret; | |
2563 | ||
2564 | new->path = path; | |
2565 | ||
2566 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2567 | ret = iterate_inodes_from_logical(old->bytenr + |
2568 | old->extent_offset, fs_info, | |
38c227d8 | 2569 | path, record_one_backref, |
c995ab3c | 2570 | old, false); |
4724b106 JB |
2571 | if (ret < 0 && ret != -ENOENT) |
2572 | return false; | |
38c227d8 LB |
2573 | |
2574 | /* no backref to be processed for this extent */ | |
2575 | if (!old->count) { | |
2576 | list_del(&old->list); | |
2577 | kfree(old); | |
2578 | } | |
2579 | } | |
2580 | ||
2581 | if (list_empty(&new->head)) | |
2582 | return false; | |
2583 | ||
2584 | return true; | |
2585 | } | |
2586 | ||
2587 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2588 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2589 | struct new_sa_defrag_extent *new) |
38c227d8 | 2590 | { |
116e0024 | 2591 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2592 | return 0; |
2593 | ||
2594 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2595 | return 0; | |
2596 | ||
116e0024 LB |
2597 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2598 | return 0; | |
2599 | ||
2600 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2601 | btrfs_file_extent_other_encoding(leaf, fi)) |
2602 | return 0; | |
2603 | ||
2604 | return 1; | |
2605 | } | |
2606 | ||
2607 | /* | |
2608 | * Note the backref might has changed, and in this case we just return 0. | |
2609 | */ | |
2610 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2611 | struct sa_defrag_extent_backref *prev, | |
2612 | struct sa_defrag_extent_backref *backref) | |
2613 | { | |
2614 | struct btrfs_file_extent_item *extent; | |
2615 | struct btrfs_file_extent_item *item; | |
2616 | struct btrfs_ordered_extent *ordered; | |
2617 | struct btrfs_trans_handle *trans; | |
82fa113f | 2618 | struct btrfs_ref ref = { 0 }; |
38c227d8 LB |
2619 | struct btrfs_root *root; |
2620 | struct btrfs_key key; | |
2621 | struct extent_buffer *leaf; | |
2622 | struct old_sa_defrag_extent *old = backref->old; | |
2623 | struct new_sa_defrag_extent *new = old->new; | |
0b246afa | 2624 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2625 | struct inode *inode; |
2626 | struct extent_state *cached = NULL; | |
2627 | int ret = 0; | |
2628 | u64 start; | |
2629 | u64 len; | |
2630 | u64 lock_start; | |
2631 | u64 lock_end; | |
2632 | bool merge = false; | |
2633 | int index; | |
2634 | ||
2635 | if (prev && prev->root_id == backref->root_id && | |
2636 | prev->inum == backref->inum && | |
2637 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2638 | merge = true; | |
2639 | ||
2640 | /* step 1: get root */ | |
2641 | key.objectid = backref->root_id; | |
2642 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2643 | key.offset = (u64)-1; | |
2644 | ||
38c227d8 LB |
2645 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2646 | ||
2647 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2648 | if (IS_ERR(root)) { | |
2649 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2650 | if (PTR_ERR(root) == -ENOENT) | |
2651 | return 0; | |
2652 | return PTR_ERR(root); | |
2653 | } | |
38c227d8 | 2654 | |
bcbba5e6 WS |
2655 | if (btrfs_root_readonly(root)) { |
2656 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2657 | return 0; | |
2658 | } | |
2659 | ||
38c227d8 LB |
2660 | /* step 2: get inode */ |
2661 | key.objectid = backref->inum; | |
2662 | key.type = BTRFS_INODE_ITEM_KEY; | |
2663 | key.offset = 0; | |
2664 | ||
2665 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2666 | if (IS_ERR(inode)) { | |
2667 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2668 | return 0; | |
2669 | } | |
2670 | ||
2671 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2672 | ||
2673 | /* step 3: relink backref */ | |
2674 | lock_start = backref->file_pos; | |
2675 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2676 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2677 | &cached); |
38c227d8 LB |
2678 | |
2679 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2680 | if (ordered) { | |
2681 | btrfs_put_ordered_extent(ordered); | |
2682 | goto out_unlock; | |
2683 | } | |
2684 | ||
2685 | trans = btrfs_join_transaction(root); | |
2686 | if (IS_ERR(trans)) { | |
2687 | ret = PTR_ERR(trans); | |
2688 | goto out_unlock; | |
2689 | } | |
2690 | ||
2691 | key.objectid = backref->inum; | |
2692 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2693 | key.offset = backref->file_pos; | |
2694 | ||
2695 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2696 | if (ret < 0) { | |
2697 | goto out_free_path; | |
2698 | } else if (ret > 0) { | |
2699 | ret = 0; | |
2700 | goto out_free_path; | |
2701 | } | |
2702 | ||
2703 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2704 | struct btrfs_file_extent_item); | |
2705 | ||
2706 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2707 | backref->generation) | |
2708 | goto out_free_path; | |
2709 | ||
2710 | btrfs_release_path(path); | |
2711 | ||
2712 | start = backref->file_pos; | |
2713 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2714 | start += old->extent_offset + old->offset - | |
2715 | backref->extent_offset; | |
2716 | ||
2717 | len = min(backref->extent_offset + backref->num_bytes, | |
2718 | old->extent_offset + old->offset + old->len); | |
2719 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2720 | ||
2721 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2722 | start + len, 1); | |
2723 | if (ret) | |
2724 | goto out_free_path; | |
2725 | again: | |
4a0cc7ca | 2726 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2727 | key.type = BTRFS_EXTENT_DATA_KEY; |
2728 | key.offset = start; | |
2729 | ||
a09a0a70 | 2730 | path->leave_spinning = 1; |
38c227d8 LB |
2731 | if (merge) { |
2732 | struct btrfs_file_extent_item *fi; | |
2733 | u64 extent_len; | |
2734 | struct btrfs_key found_key; | |
2735 | ||
3c9665df | 2736 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2737 | if (ret < 0) |
2738 | goto out_free_path; | |
2739 | ||
2740 | path->slots[0]--; | |
2741 | leaf = path->nodes[0]; | |
2742 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2743 | ||
2744 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2745 | struct btrfs_file_extent_item); | |
2746 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2747 | ||
116e0024 LB |
2748 | if (extent_len + found_key.offset == start && |
2749 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2750 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2751 | extent_len + len); | |
2752 | btrfs_mark_buffer_dirty(leaf); | |
2753 | inode_add_bytes(inode, len); | |
2754 | ||
2755 | ret = 1; | |
2756 | goto out_free_path; | |
2757 | } else { | |
2758 | merge = false; | |
2759 | btrfs_release_path(path); | |
2760 | goto again; | |
2761 | } | |
2762 | } | |
2763 | ||
2764 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2765 | sizeof(*extent)); | |
2766 | if (ret) { | |
66642832 | 2767 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2768 | goto out_free_path; |
2769 | } | |
2770 | ||
2771 | leaf = path->nodes[0]; | |
2772 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2773 | struct btrfs_file_extent_item); | |
2774 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2775 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2776 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2777 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2778 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2779 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2780 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2781 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2782 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2783 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2784 | ||
2785 | btrfs_mark_buffer_dirty(leaf); | |
2786 | inode_add_bytes(inode, len); | |
a09a0a70 | 2787 | btrfs_release_path(path); |
38c227d8 | 2788 | |
82fa113f QW |
2789 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new->bytenr, |
2790 | new->disk_len, 0); | |
2791 | btrfs_init_data_ref(&ref, backref->root_id, backref->inum, | |
2792 | new->file_pos); /* start - extent_offset */ | |
2793 | ret = btrfs_inc_extent_ref(trans, &ref); | |
38c227d8 | 2794 | if (ret) { |
66642832 | 2795 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2796 | goto out_free_path; |
2797 | } | |
2798 | ||
2799 | ret = 1; | |
2800 | out_free_path: | |
2801 | btrfs_release_path(path); | |
a09a0a70 | 2802 | path->leave_spinning = 0; |
3a45bb20 | 2803 | btrfs_end_transaction(trans); |
38c227d8 LB |
2804 | out_unlock: |
2805 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
e43bbe5e | 2806 | &cached); |
38c227d8 LB |
2807 | iput(inode); |
2808 | return ret; | |
2809 | } | |
2810 | ||
6f519564 LB |
2811 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2812 | { | |
2813 | struct old_sa_defrag_extent *old, *tmp; | |
2814 | ||
2815 | if (!new) | |
2816 | return; | |
2817 | ||
2818 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2819 | kfree(old); |
2820 | } | |
2821 | kfree(new); | |
2822 | } | |
2823 | ||
38c227d8 LB |
2824 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2825 | { | |
0b246afa | 2826 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 | 2827 | struct btrfs_path *path; |
38c227d8 LB |
2828 | struct sa_defrag_extent_backref *backref; |
2829 | struct sa_defrag_extent_backref *prev = NULL; | |
38c227d8 LB |
2830 | struct rb_node *node; |
2831 | int ret; | |
2832 | ||
38c227d8 LB |
2833 | path = btrfs_alloc_path(); |
2834 | if (!path) | |
2835 | return; | |
2836 | ||
2837 | if (!record_extent_backrefs(path, new)) { | |
2838 | btrfs_free_path(path); | |
2839 | goto out; | |
2840 | } | |
2841 | btrfs_release_path(path); | |
2842 | ||
2843 | while (1) { | |
2844 | node = rb_first(&new->root); | |
2845 | if (!node) | |
2846 | break; | |
2847 | rb_erase(node, &new->root); | |
2848 | ||
2849 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2850 | ||
2851 | ret = relink_extent_backref(path, prev, backref); | |
2852 | WARN_ON(ret < 0); | |
2853 | ||
2854 | kfree(prev); | |
2855 | ||
2856 | if (ret == 1) | |
2857 | prev = backref; | |
2858 | else | |
2859 | prev = NULL; | |
2860 | cond_resched(); | |
2861 | } | |
2862 | kfree(prev); | |
2863 | ||
2864 | btrfs_free_path(path); | |
38c227d8 | 2865 | out: |
6f519564 LB |
2866 | free_sa_defrag_extent(new); |
2867 | ||
0b246afa JM |
2868 | atomic_dec(&fs_info->defrag_running); |
2869 | wake_up(&fs_info->transaction_wait); | |
38c227d8 LB |
2870 | } |
2871 | ||
2872 | static struct new_sa_defrag_extent * | |
2873 | record_old_file_extents(struct inode *inode, | |
2874 | struct btrfs_ordered_extent *ordered) | |
2875 | { | |
0b246afa | 2876 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2877 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2878 | struct btrfs_path *path; | |
2879 | struct btrfs_key key; | |
6f519564 | 2880 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2881 | struct new_sa_defrag_extent *new; |
2882 | int ret; | |
2883 | ||
2884 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2885 | if (!new) | |
2886 | return NULL; | |
2887 | ||
2888 | new->inode = inode; | |
2889 | new->file_pos = ordered->file_offset; | |
2890 | new->len = ordered->len; | |
2891 | new->bytenr = ordered->start; | |
2892 | new->disk_len = ordered->disk_len; | |
2893 | new->compress_type = ordered->compress_type; | |
2894 | new->root = RB_ROOT; | |
2895 | INIT_LIST_HEAD(&new->head); | |
2896 | ||
2897 | path = btrfs_alloc_path(); | |
2898 | if (!path) | |
2899 | goto out_kfree; | |
2900 | ||
4a0cc7ca | 2901 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2902 | key.type = BTRFS_EXTENT_DATA_KEY; |
2903 | key.offset = new->file_pos; | |
2904 | ||
2905 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2906 | if (ret < 0) | |
2907 | goto out_free_path; | |
2908 | if (ret > 0 && path->slots[0] > 0) | |
2909 | path->slots[0]--; | |
2910 | ||
2911 | /* find out all the old extents for the file range */ | |
2912 | while (1) { | |
2913 | struct btrfs_file_extent_item *extent; | |
2914 | struct extent_buffer *l; | |
2915 | int slot; | |
2916 | u64 num_bytes; | |
2917 | u64 offset; | |
2918 | u64 end; | |
2919 | u64 disk_bytenr; | |
2920 | u64 extent_offset; | |
2921 | ||
2922 | l = path->nodes[0]; | |
2923 | slot = path->slots[0]; | |
2924 | ||
2925 | if (slot >= btrfs_header_nritems(l)) { | |
2926 | ret = btrfs_next_leaf(root, path); | |
2927 | if (ret < 0) | |
6f519564 | 2928 | goto out_free_path; |
38c227d8 LB |
2929 | else if (ret > 0) |
2930 | break; | |
2931 | continue; | |
2932 | } | |
2933 | ||
2934 | btrfs_item_key_to_cpu(l, &key, slot); | |
2935 | ||
4a0cc7ca | 2936 | if (key.objectid != btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2937 | break; |
2938 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2939 | break; | |
2940 | if (key.offset >= new->file_pos + new->len) | |
2941 | break; | |
2942 | ||
2943 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2944 | ||
2945 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2946 | if (key.offset + num_bytes < new->file_pos) | |
2947 | goto next; | |
2948 | ||
2949 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2950 | if (!disk_bytenr) | |
2951 | goto next; | |
2952 | ||
2953 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2954 | ||
2955 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2956 | if (!old) | |
6f519564 | 2957 | goto out_free_path; |
38c227d8 LB |
2958 | |
2959 | offset = max(new->file_pos, key.offset); | |
2960 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2961 | ||
2962 | old->bytenr = disk_bytenr; | |
2963 | old->extent_offset = extent_offset; | |
2964 | old->offset = offset - key.offset; | |
2965 | old->len = end - offset; | |
2966 | old->new = new; | |
2967 | old->count = 0; | |
2968 | list_add_tail(&old->list, &new->head); | |
2969 | next: | |
2970 | path->slots[0]++; | |
2971 | cond_resched(); | |
2972 | } | |
2973 | ||
2974 | btrfs_free_path(path); | |
0b246afa | 2975 | atomic_inc(&fs_info->defrag_running); |
38c227d8 LB |
2976 | |
2977 | return new; | |
2978 | ||
38c227d8 LB |
2979 | out_free_path: |
2980 | btrfs_free_path(path); | |
2981 | out_kfree: | |
6f519564 | 2982 | free_sa_defrag_extent(new); |
38c227d8 LB |
2983 | return NULL; |
2984 | } | |
2985 | ||
2ff7e61e | 2986 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2987 | u64 start, u64 len) |
2988 | { | |
2989 | struct btrfs_block_group_cache *cache; | |
2990 | ||
0b246afa | 2991 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2992 | ASSERT(cache); |
2993 | ||
2994 | spin_lock(&cache->lock); | |
2995 | cache->delalloc_bytes -= len; | |
2996 | spin_unlock(&cache->lock); | |
2997 | ||
2998 | btrfs_put_block_group(cache); | |
2999 | } | |
3000 | ||
d352ac68 CM |
3001 | /* as ordered data IO finishes, this gets called so we can finish |
3002 | * an ordered extent if the range of bytes in the file it covers are | |
3003 | * fully written. | |
3004 | */ | |
5fd02043 | 3005 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 3006 | { |
5fd02043 | 3007 | struct inode *inode = ordered_extent->inode; |
0b246afa | 3008 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 3009 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 3010 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 3011 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 3012 | struct extent_state *cached_state = NULL; |
38c227d8 | 3013 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 3014 | int compress_type = 0; |
77cef2ec JB |
3015 | int ret = 0; |
3016 | u64 logical_len = ordered_extent->len; | |
82d5902d | 3017 | bool nolock; |
77cef2ec | 3018 | bool truncated = false; |
a7e3b975 FM |
3019 | bool range_locked = false; |
3020 | bool clear_new_delalloc_bytes = false; | |
49940bdd | 3021 | bool clear_reserved_extent = true; |
a7e3b975 FM |
3022 | |
3023 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
3024 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
3025 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags)) | |
3026 | clear_new_delalloc_bytes = true; | |
e6dcd2dc | 3027 | |
70ddc553 | 3028 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
0cb59c99 | 3029 | |
5fd02043 JB |
3030 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
3031 | ret = -EIO; | |
3032 | goto out; | |
3033 | } | |
3034 | ||
7ab7956e NB |
3035 | btrfs_free_io_failure_record(BTRFS_I(inode), |
3036 | ordered_extent->file_offset, | |
3037 | ordered_extent->file_offset + | |
3038 | ordered_extent->len - 1); | |
f612496b | 3039 | |
77cef2ec JB |
3040 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
3041 | truncated = true; | |
3042 | logical_len = ordered_extent->truncated_len; | |
3043 | /* Truncated the entire extent, don't bother adding */ | |
3044 | if (!logical_len) | |
3045 | goto out; | |
3046 | } | |
3047 | ||
c2167754 | 3048 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 3049 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
3050 | |
3051 | /* | |
3052 | * For mwrite(mmap + memset to write) case, we still reserve | |
3053 | * space for NOCOW range. | |
3054 | * As NOCOW won't cause a new delayed ref, just free the space | |
3055 | */ | |
bc42bda2 | 3056 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
94ed938a | 3057 | ordered_extent->len); |
6c760c07 JB |
3058 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
3059 | if (nolock) | |
3060 | trans = btrfs_join_transaction_nolock(root); | |
3061 | else | |
3062 | trans = btrfs_join_transaction(root); | |
3063 | if (IS_ERR(trans)) { | |
3064 | ret = PTR_ERR(trans); | |
3065 | trans = NULL; | |
3066 | goto out; | |
c2167754 | 3067 | } |
69fe2d75 | 3068 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
6c760c07 JB |
3069 | ret = btrfs_update_inode_fallback(trans, root, inode); |
3070 | if (ret) /* -ENOMEM or corruption */ | |
66642832 | 3071 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
3072 | goto out; |
3073 | } | |
e6dcd2dc | 3074 | |
a7e3b975 | 3075 | range_locked = true; |
2ac55d41 JB |
3076 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
3077 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 3078 | &cached_state); |
e6dcd2dc | 3079 | |
38c227d8 LB |
3080 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
3081 | ordered_extent->file_offset + ordered_extent->len - 1, | |
452e62b7 | 3082 | EXTENT_DEFRAG, 0, cached_state); |
38c227d8 LB |
3083 | if (ret) { |
3084 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 3085 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
3086 | /* the inode is shared */ |
3087 | new = record_old_file_extents(inode, ordered_extent); | |
3088 | ||
3089 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
3090 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ae0f1625 | 3091 | EXTENT_DEFRAG, 0, 0, &cached_state); |
38c227d8 LB |
3092 | } |
3093 | ||
0cb59c99 | 3094 | if (nolock) |
7a7eaa40 | 3095 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 3096 | else |
7a7eaa40 | 3097 | trans = btrfs_join_transaction(root); |
79787eaa JM |
3098 | if (IS_ERR(trans)) { |
3099 | ret = PTR_ERR(trans); | |
3100 | trans = NULL; | |
a7e3b975 | 3101 | goto out; |
79787eaa | 3102 | } |
a79b7d4b | 3103 | |
69fe2d75 | 3104 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
c2167754 | 3105 | |
c8b97818 | 3106 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3107 | compress_type = ordered_extent->compress_type; |
d899e052 | 3108 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3109 | BUG_ON(compress_type); |
b430b775 JM |
3110 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
3111 | ordered_extent->len); | |
7a6d7067 | 3112 | ret = btrfs_mark_extent_written(trans, BTRFS_I(inode), |
d899e052 YZ |
3113 | ordered_extent->file_offset, |
3114 | ordered_extent->file_offset + | |
77cef2ec | 3115 | logical_len); |
d899e052 | 3116 | } else { |
0b246afa | 3117 | BUG_ON(root == fs_info->tree_root); |
d899e052 YZ |
3118 | ret = insert_reserved_file_extent(trans, inode, |
3119 | ordered_extent->file_offset, | |
3120 | ordered_extent->start, | |
3121 | ordered_extent->disk_len, | |
77cef2ec | 3122 | logical_len, logical_len, |
261507a0 | 3123 | compress_type, 0, 0, |
d899e052 | 3124 | BTRFS_FILE_EXTENT_REG); |
49940bdd JB |
3125 | if (!ret) { |
3126 | clear_reserved_extent = false; | |
2ff7e61e | 3127 | btrfs_release_delalloc_bytes(fs_info, |
e570fd27 MX |
3128 | ordered_extent->start, |
3129 | ordered_extent->disk_len); | |
49940bdd | 3130 | } |
d899e052 | 3131 | } |
5dc562c5 JB |
3132 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
3133 | ordered_extent->file_offset, ordered_extent->len, | |
3134 | trans->transid); | |
79787eaa | 3135 | if (ret < 0) { |
66642832 | 3136 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3137 | goto out; |
79787eaa | 3138 | } |
2ac55d41 | 3139 | |
ac01f26a NB |
3140 | ret = add_pending_csums(trans, inode, &ordered_extent->list); |
3141 | if (ret) { | |
3142 | btrfs_abort_transaction(trans, ret); | |
3143 | goto out; | |
3144 | } | |
e6dcd2dc | 3145 | |
6c760c07 JB |
3146 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
3147 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
3148 | if (ret) { /* -ENOMEM or corruption */ | |
66642832 | 3149 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3150 | goto out; |
1ef30be1 JB |
3151 | } |
3152 | ret = 0; | |
c2167754 | 3153 | out: |
a7e3b975 FM |
3154 | if (range_locked || clear_new_delalloc_bytes) { |
3155 | unsigned int clear_bits = 0; | |
3156 | ||
3157 | if (range_locked) | |
3158 | clear_bits |= EXTENT_LOCKED; | |
3159 | if (clear_new_delalloc_bytes) | |
3160 | clear_bits |= EXTENT_DELALLOC_NEW; | |
3161 | clear_extent_bit(&BTRFS_I(inode)->io_tree, | |
3162 | ordered_extent->file_offset, | |
3163 | ordered_extent->file_offset + | |
3164 | ordered_extent->len - 1, | |
3165 | clear_bits, | |
3166 | (clear_bits & EXTENT_LOCKED) ? 1 : 0, | |
ae0f1625 | 3167 | 0, &cached_state); |
a7e3b975 FM |
3168 | } |
3169 | ||
a698d075 | 3170 | if (trans) |
3a45bb20 | 3171 | btrfs_end_transaction(trans); |
0cb59c99 | 3172 | |
77cef2ec JB |
3173 | if (ret || truncated) { |
3174 | u64 start, end; | |
3175 | ||
3176 | if (truncated) | |
3177 | start = ordered_extent->file_offset + logical_len; | |
3178 | else | |
3179 | start = ordered_extent->file_offset; | |
3180 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
f08dc36f | 3181 | clear_extent_uptodate(io_tree, start, end, NULL); |
77cef2ec JB |
3182 | |
3183 | /* Drop the cache for the part of the extent we didn't write. */ | |
dcdbc059 | 3184 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); |
5fd02043 | 3185 | |
0bec9ef5 JB |
3186 | /* |
3187 | * If the ordered extent had an IOERR or something else went | |
3188 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3189 | * back to the allocator. We only free the extent in the |
3190 | * truncated case if we didn't write out the extent at all. | |
49940bdd JB |
3191 | * |
3192 | * If we made it past insert_reserved_file_extent before we | |
3193 | * errored out then we don't need to do this as the accounting | |
3194 | * has already been done. | |
0bec9ef5 | 3195 | */ |
77cef2ec | 3196 | if ((ret || !logical_len) && |
49940bdd | 3197 | clear_reserved_extent && |
77cef2ec | 3198 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
0bec9ef5 | 3199 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2ff7e61e JM |
3200 | btrfs_free_reserved_extent(fs_info, |
3201 | ordered_extent->start, | |
e570fd27 | 3202 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
3203 | } |
3204 | ||
3205 | ||
5fd02043 | 3206 | /* |
8bad3c02 LB |
3207 | * This needs to be done to make sure anybody waiting knows we are done |
3208 | * updating everything for this ordered extent. | |
5fd02043 JB |
3209 | */ |
3210 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
3211 | ||
38c227d8 | 3212 | /* for snapshot-aware defrag */ |
6f519564 LB |
3213 | if (new) { |
3214 | if (ret) { | |
3215 | free_sa_defrag_extent(new); | |
0b246afa | 3216 | atomic_dec(&fs_info->defrag_running); |
6f519564 LB |
3217 | } else { |
3218 | relink_file_extents(new); | |
3219 | } | |
3220 | } | |
38c227d8 | 3221 | |
e6dcd2dc CM |
3222 | /* once for us */ |
3223 | btrfs_put_ordered_extent(ordered_extent); | |
3224 | /* once for the tree */ | |
3225 | btrfs_put_ordered_extent(ordered_extent); | |
3226 | ||
5fd02043 JB |
3227 | return ret; |
3228 | } | |
3229 | ||
3230 | static void finish_ordered_fn(struct btrfs_work *work) | |
3231 | { | |
3232 | struct btrfs_ordered_extent *ordered_extent; | |
3233 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3234 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3235 | } |
3236 | ||
c629732d NB |
3237 | void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start, |
3238 | u64 end, int uptodate) | |
211f90e6 | 3239 | { |
5fd02043 | 3240 | struct inode *inode = page->mapping->host; |
0b246afa | 3241 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5fd02043 | 3242 | struct btrfs_ordered_extent *ordered_extent = NULL; |
9e0af237 LB |
3243 | struct btrfs_workqueue *wq; |
3244 | btrfs_work_func_t func; | |
5fd02043 | 3245 | |
1abe9b8a | 3246 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
3247 | ||
8b62b72b | 3248 | ClearPagePrivate2(page); |
5fd02043 JB |
3249 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
3250 | end - start + 1, uptodate)) | |
c3988d63 | 3251 | return; |
5fd02043 | 3252 | |
70ddc553 | 3253 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
0b246afa | 3254 | wq = fs_info->endio_freespace_worker; |
9e0af237 LB |
3255 | func = btrfs_freespace_write_helper; |
3256 | } else { | |
0b246afa | 3257 | wq = fs_info->endio_write_workers; |
9e0af237 LB |
3258 | func = btrfs_endio_write_helper; |
3259 | } | |
5fd02043 | 3260 | |
9e0af237 LB |
3261 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3262 | NULL); | |
3263 | btrfs_queue_work(wq, &ordered_extent->work); | |
211f90e6 CM |
3264 | } |
3265 | ||
dc380aea MX |
3266 | static int __readpage_endio_check(struct inode *inode, |
3267 | struct btrfs_io_bio *io_bio, | |
3268 | int icsum, struct page *page, | |
3269 | int pgoff, u64 start, size_t len) | |
3270 | { | |
d5178578 JT |
3271 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
3272 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
dc380aea | 3273 | char *kaddr; |
d5178578 JT |
3274 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); |
3275 | u8 *csum_expected; | |
3276 | u8 csum[BTRFS_CSUM_SIZE]; | |
dc380aea | 3277 | |
d5178578 | 3278 | csum_expected = ((u8 *)io_bio->csum) + icsum * csum_size; |
dc380aea MX |
3279 | |
3280 | kaddr = kmap_atomic(page); | |
d5178578 JT |
3281 | shash->tfm = fs_info->csum_shash; |
3282 | ||
3283 | crypto_shash_init(shash); | |
3284 | crypto_shash_update(shash, kaddr + pgoff, len); | |
3285 | crypto_shash_final(shash, csum); | |
3286 | ||
3287 | if (memcmp(csum, csum_expected, csum_size)) | |
dc380aea MX |
3288 | goto zeroit; |
3289 | ||
3290 | kunmap_atomic(kaddr); | |
3291 | return 0; | |
3292 | zeroit: | |
ea41d6b2 JT |
3293 | btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected, |
3294 | io_bio->mirror_num); | |
dc380aea MX |
3295 | memset(kaddr + pgoff, 1, len); |
3296 | flush_dcache_page(page); | |
3297 | kunmap_atomic(kaddr); | |
dc380aea MX |
3298 | return -EIO; |
3299 | } | |
3300 | ||
d352ac68 CM |
3301 | /* |
3302 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3303 | * if there's a match, we allow the bio to finish. If not, the code in |
3304 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3305 | */ |
facc8a22 MX |
3306 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3307 | u64 phy_offset, struct page *page, | |
3308 | u64 start, u64 end, int mirror) | |
07157aac | 3309 | { |
4eee4fa4 | 3310 | size_t offset = start - page_offset(page); |
07157aac | 3311 | struct inode *inode = page->mapping->host; |
d1310b2e | 3312 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3313 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3314 | |
d20f7043 CM |
3315 | if (PageChecked(page)) { |
3316 | ClearPageChecked(page); | |
dc380aea | 3317 | return 0; |
d20f7043 | 3318 | } |
6cbff00f CH |
3319 | |
3320 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3321 | return 0; |
17d217fe YZ |
3322 | |
3323 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3324 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3325 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3326 | return 0; |
17d217fe | 3327 | } |
d20f7043 | 3328 | |
facc8a22 | 3329 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3330 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3331 | start, (size_t)(end - start + 1)); | |
07157aac | 3332 | } |
b888db2b | 3333 | |
c1c3fac2 NB |
3334 | /* |
3335 | * btrfs_add_delayed_iput - perform a delayed iput on @inode | |
3336 | * | |
3337 | * @inode: The inode we want to perform iput on | |
3338 | * | |
3339 | * This function uses the generic vfs_inode::i_count to track whether we should | |
3340 | * just decrement it (in case it's > 1) or if this is the last iput then link | |
3341 | * the inode to the delayed iput machinery. Delayed iputs are processed at | |
3342 | * transaction commit time/superblock commit/cleaner kthread. | |
3343 | */ | |
24bbcf04 YZ |
3344 | void btrfs_add_delayed_iput(struct inode *inode) |
3345 | { | |
0b246afa | 3346 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3347 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3348 | |
3349 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3350 | return; | |
3351 | ||
034f784d | 3352 | atomic_inc(&fs_info->nr_delayed_iputs); |
24bbcf04 | 3353 | spin_lock(&fs_info->delayed_iput_lock); |
c1c3fac2 NB |
3354 | ASSERT(list_empty(&binode->delayed_iput)); |
3355 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
24bbcf04 | 3356 | spin_unlock(&fs_info->delayed_iput_lock); |
fd340d0f JB |
3357 | if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags)) |
3358 | wake_up_process(fs_info->cleaner_kthread); | |
24bbcf04 YZ |
3359 | } |
3360 | ||
63611e73 JB |
3361 | static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info, |
3362 | struct btrfs_inode *inode) | |
3363 | { | |
3364 | list_del_init(&inode->delayed_iput); | |
3365 | spin_unlock(&fs_info->delayed_iput_lock); | |
3366 | iput(&inode->vfs_inode); | |
3367 | if (atomic_dec_and_test(&fs_info->nr_delayed_iputs)) | |
3368 | wake_up(&fs_info->delayed_iputs_wait); | |
3369 | spin_lock(&fs_info->delayed_iput_lock); | |
3370 | } | |
3371 | ||
3372 | static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info, | |
3373 | struct btrfs_inode *inode) | |
3374 | { | |
3375 | if (!list_empty(&inode->delayed_iput)) { | |
3376 | spin_lock(&fs_info->delayed_iput_lock); | |
3377 | if (!list_empty(&inode->delayed_iput)) | |
3378 | run_delayed_iput_locked(fs_info, inode); | |
3379 | spin_unlock(&fs_info->delayed_iput_lock); | |
3380 | } | |
3381 | } | |
3382 | ||
2ff7e61e | 3383 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3384 | { |
24bbcf04 | 3385 | |
24bbcf04 | 3386 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3387 | while (!list_empty(&fs_info->delayed_iputs)) { |
3388 | struct btrfs_inode *inode; | |
3389 | ||
3390 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3391 | struct btrfs_inode, delayed_iput); | |
63611e73 | 3392 | run_delayed_iput_locked(fs_info, inode); |
24bbcf04 | 3393 | } |
8089fe62 | 3394 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3395 | } |
3396 | ||
034f784d JB |
3397 | /** |
3398 | * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running | |
3399 | * @fs_info - the fs_info for this fs | |
3400 | * @return - EINTR if we were killed, 0 if nothing's pending | |
3401 | * | |
3402 | * This will wait on any delayed iputs that are currently running with KILLABLE | |
3403 | * set. Once they are all done running we will return, unless we are killed in | |
3404 | * which case we return EINTR. This helps in user operations like fallocate etc | |
3405 | * that might get blocked on the iputs. | |
3406 | */ | |
3407 | int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info) | |
3408 | { | |
3409 | int ret = wait_event_killable(fs_info->delayed_iputs_wait, | |
3410 | atomic_read(&fs_info->nr_delayed_iputs) == 0); | |
3411 | if (ret) | |
3412 | return -EINTR; | |
3413 | return 0; | |
3414 | } | |
3415 | ||
7b128766 | 3416 | /* |
f7e9e8fc OS |
3417 | * This creates an orphan entry for the given inode in case something goes wrong |
3418 | * in the middle of an unlink. | |
7b128766 | 3419 | */ |
73f2e545 | 3420 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
27919067 | 3421 | struct btrfs_inode *inode) |
7b128766 | 3422 | { |
d68fc57b | 3423 | int ret; |
7b128766 | 3424 | |
27919067 OS |
3425 | ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode)); |
3426 | if (ret && ret != -EEXIST) { | |
3427 | btrfs_abort_transaction(trans, ret); | |
3428 | return ret; | |
d68fc57b YZ |
3429 | } |
3430 | ||
d68fc57b | 3431 | return 0; |
7b128766 JB |
3432 | } |
3433 | ||
3434 | /* | |
f7e9e8fc OS |
3435 | * We have done the delete so we can go ahead and remove the orphan item for |
3436 | * this particular inode. | |
7b128766 | 3437 | */ |
48a3b636 | 3438 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3439 | struct btrfs_inode *inode) |
7b128766 | 3440 | { |
27919067 | 3441 | return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode)); |
7b128766 JB |
3442 | } |
3443 | ||
3444 | /* | |
3445 | * this cleans up any orphans that may be left on the list from the last use | |
3446 | * of this root. | |
3447 | */ | |
66b4ffd1 | 3448 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3449 | { |
0b246afa | 3450 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3451 | struct btrfs_path *path; |
3452 | struct extent_buffer *leaf; | |
7b128766 JB |
3453 | struct btrfs_key key, found_key; |
3454 | struct btrfs_trans_handle *trans; | |
3455 | struct inode *inode; | |
8f6d7f4f | 3456 | u64 last_objectid = 0; |
f7e9e8fc | 3457 | int ret = 0, nr_unlink = 0; |
7b128766 | 3458 | |
d68fc57b | 3459 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3460 | return 0; |
c71bf099 YZ |
3461 | |
3462 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3463 | if (!path) { |
3464 | ret = -ENOMEM; | |
3465 | goto out; | |
3466 | } | |
e4058b54 | 3467 | path->reada = READA_BACK; |
7b128766 JB |
3468 | |
3469 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3470 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3471 | key.offset = (u64)-1; |
3472 | ||
7b128766 JB |
3473 | while (1) { |
3474 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3475 | if (ret < 0) |
3476 | goto out; | |
7b128766 JB |
3477 | |
3478 | /* | |
3479 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3480 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3481 | * find the key and see if we have stuff that matches |
3482 | */ | |
3483 | if (ret > 0) { | |
66b4ffd1 | 3484 | ret = 0; |
7b128766 JB |
3485 | if (path->slots[0] == 0) |
3486 | break; | |
3487 | path->slots[0]--; | |
3488 | } | |
3489 | ||
3490 | /* pull out the item */ | |
3491 | leaf = path->nodes[0]; | |
7b128766 JB |
3492 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3493 | ||
3494 | /* make sure the item matches what we want */ | |
3495 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3496 | break; | |
962a298f | 3497 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3498 | break; |
3499 | ||
3500 | /* release the path since we're done with it */ | |
b3b4aa74 | 3501 | btrfs_release_path(path); |
7b128766 JB |
3502 | |
3503 | /* | |
3504 | * this is where we are basically btrfs_lookup, without the | |
3505 | * crossing root thing. we store the inode number in the | |
3506 | * offset of the orphan item. | |
3507 | */ | |
8f6d7f4f JB |
3508 | |
3509 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3510 | btrfs_err(fs_info, |
3511 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3512 | ret = -EINVAL; |
3513 | goto out; | |
3514 | } | |
3515 | ||
3516 | last_objectid = found_key.offset; | |
3517 | ||
5d4f98a2 YZ |
3518 | found_key.objectid = found_key.offset; |
3519 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3520 | found_key.offset = 0; | |
0b246afa | 3521 | inode = btrfs_iget(fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3522 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3523 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3524 | goto out; |
7b128766 | 3525 | |
0b246afa | 3526 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 AJ |
3527 | struct btrfs_root *dead_root; |
3528 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3529 | int is_dead_root = 0; | |
3530 | ||
3531 | /* | |
3532 | * this is an orphan in the tree root. Currently these | |
3533 | * could come from 2 sources: | |
3534 | * a) a snapshot deletion in progress | |
3535 | * b) a free space cache inode | |
3536 | * We need to distinguish those two, as the snapshot | |
3537 | * orphan must not get deleted. | |
3538 | * find_dead_roots already ran before us, so if this | |
3539 | * is a snapshot deletion, we should find the root | |
3540 | * in the dead_roots list | |
3541 | */ | |
3542 | spin_lock(&fs_info->trans_lock); | |
3543 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3544 | root_list) { | |
3545 | if (dead_root->root_key.objectid == | |
3546 | found_key.objectid) { | |
3547 | is_dead_root = 1; | |
3548 | break; | |
3549 | } | |
3550 | } | |
3551 | spin_unlock(&fs_info->trans_lock); | |
3552 | if (is_dead_root) { | |
3553 | /* prevent this orphan from being found again */ | |
3554 | key.offset = found_key.objectid - 1; | |
3555 | continue; | |
3556 | } | |
f7e9e8fc | 3557 | |
f8e9e0b0 | 3558 | } |
f7e9e8fc | 3559 | |
7b128766 | 3560 | /* |
f7e9e8fc OS |
3561 | * If we have an inode with links, there are a couple of |
3562 | * possibilities. Old kernels (before v3.12) used to create an | |
3563 | * orphan item for truncate indicating that there were possibly | |
3564 | * extent items past i_size that needed to be deleted. In v3.12, | |
3565 | * truncate was changed to update i_size in sync with the extent | |
3566 | * items, but the (useless) orphan item was still created. Since | |
3567 | * v4.18, we don't create the orphan item for truncate at all. | |
3568 | * | |
3569 | * So, this item could mean that we need to do a truncate, but | |
3570 | * only if this filesystem was last used on a pre-v3.12 kernel | |
3571 | * and was not cleanly unmounted. The odds of that are quite | |
3572 | * slim, and it's a pain to do the truncate now, so just delete | |
3573 | * the orphan item. | |
3574 | * | |
3575 | * It's also possible that this orphan item was supposed to be | |
3576 | * deleted but wasn't. The inode number may have been reused, | |
3577 | * but either way, we can delete the orphan item. | |
7b128766 | 3578 | */ |
f7e9e8fc OS |
3579 | if (ret == -ENOENT || inode->i_nlink) { |
3580 | if (!ret) | |
3581 | iput(inode); | |
a8c9e576 | 3582 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3583 | if (IS_ERR(trans)) { |
3584 | ret = PTR_ERR(trans); | |
3585 | goto out; | |
3586 | } | |
0b246afa JM |
3587 | btrfs_debug(fs_info, "auto deleting %Lu", |
3588 | found_key.objectid); | |
a8c9e576 JB |
3589 | ret = btrfs_del_orphan_item(trans, root, |
3590 | found_key.objectid); | |
3a45bb20 | 3591 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3592 | if (ret) |
3593 | goto out; | |
7b128766 JB |
3594 | continue; |
3595 | } | |
3596 | ||
f7e9e8fc | 3597 | nr_unlink++; |
7b128766 JB |
3598 | |
3599 | /* this will do delete_inode and everything for us */ | |
3600 | iput(inode); | |
3601 | } | |
3254c876 MX |
3602 | /* release the path since we're done with it */ |
3603 | btrfs_release_path(path); | |
3604 | ||
d68fc57b YZ |
3605 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3606 | ||
a575ceeb | 3607 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { |
7a7eaa40 | 3608 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3609 | if (!IS_ERR(trans)) |
3a45bb20 | 3610 | btrfs_end_transaction(trans); |
d68fc57b | 3611 | } |
7b128766 JB |
3612 | |
3613 | if (nr_unlink) | |
0b246afa | 3614 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
66b4ffd1 JB |
3615 | |
3616 | out: | |
3617 | if (ret) | |
0b246afa | 3618 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3619 | btrfs_free_path(path); |
3620 | return ret; | |
7b128766 JB |
3621 | } |
3622 | ||
46a53cca CM |
3623 | /* |
3624 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3625 | * don't find any xattrs, we know there can't be any acls. | |
3626 | * | |
3627 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3628 | */ | |
3629 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3630 | int slot, u64 objectid, |
3631 | int *first_xattr_slot) | |
46a53cca CM |
3632 | { |
3633 | u32 nritems = btrfs_header_nritems(leaf); | |
3634 | struct btrfs_key found_key; | |
f23b5a59 JB |
3635 | static u64 xattr_access = 0; |
3636 | static u64 xattr_default = 0; | |
46a53cca CM |
3637 | int scanned = 0; |
3638 | ||
f23b5a59 | 3639 | if (!xattr_access) { |
97d79299 AG |
3640 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3641 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3642 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3643 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3644 | } |
3645 | ||
46a53cca | 3646 | slot++; |
63541927 | 3647 | *first_xattr_slot = -1; |
46a53cca CM |
3648 | while (slot < nritems) { |
3649 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3650 | ||
3651 | /* we found a different objectid, there must not be acls */ | |
3652 | if (found_key.objectid != objectid) | |
3653 | return 0; | |
3654 | ||
3655 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3656 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3657 | if (*first_xattr_slot == -1) |
3658 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3659 | if (found_key.offset == xattr_access || |
3660 | found_key.offset == xattr_default) | |
3661 | return 1; | |
3662 | } | |
46a53cca CM |
3663 | |
3664 | /* | |
3665 | * we found a key greater than an xattr key, there can't | |
3666 | * be any acls later on | |
3667 | */ | |
3668 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3669 | return 0; | |
3670 | ||
3671 | slot++; | |
3672 | scanned++; | |
3673 | ||
3674 | /* | |
3675 | * it goes inode, inode backrefs, xattrs, extents, | |
3676 | * so if there are a ton of hard links to an inode there can | |
3677 | * be a lot of backrefs. Don't waste time searching too hard, | |
3678 | * this is just an optimization | |
3679 | */ | |
3680 | if (scanned >= 8) | |
3681 | break; | |
3682 | } | |
3683 | /* we hit the end of the leaf before we found an xattr or | |
3684 | * something larger than an xattr. We have to assume the inode | |
3685 | * has acls | |
3686 | */ | |
63541927 FDBM |
3687 | if (*first_xattr_slot == -1) |
3688 | *first_xattr_slot = slot; | |
46a53cca CM |
3689 | return 1; |
3690 | } | |
3691 | ||
d352ac68 CM |
3692 | /* |
3693 | * read an inode from the btree into the in-memory inode | |
3694 | */ | |
4222ea71 FM |
3695 | static int btrfs_read_locked_inode(struct inode *inode, |
3696 | struct btrfs_path *in_path) | |
39279cc3 | 3697 | { |
0b246afa | 3698 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4222ea71 | 3699 | struct btrfs_path *path = in_path; |
5f39d397 | 3700 | struct extent_buffer *leaf; |
39279cc3 CM |
3701 | struct btrfs_inode_item *inode_item; |
3702 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3703 | struct btrfs_key location; | |
67de1176 | 3704 | unsigned long ptr; |
46a53cca | 3705 | int maybe_acls; |
618e21d5 | 3706 | u32 rdev; |
39279cc3 | 3707 | int ret; |
2f7e33d4 | 3708 | bool filled = false; |
63541927 | 3709 | int first_xattr_slot; |
2f7e33d4 MX |
3710 | |
3711 | ret = btrfs_fill_inode(inode, &rdev); | |
3712 | if (!ret) | |
3713 | filled = true; | |
39279cc3 | 3714 | |
4222ea71 FM |
3715 | if (!path) { |
3716 | path = btrfs_alloc_path(); | |
3717 | if (!path) | |
3718 | return -ENOMEM; | |
3719 | } | |
1748f843 | 3720 | |
39279cc3 | 3721 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3722 | |
39279cc3 | 3723 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 | 3724 | if (ret) { |
4222ea71 FM |
3725 | if (path != in_path) |
3726 | btrfs_free_path(path); | |
f5b3a417 | 3727 | return ret; |
67710892 | 3728 | } |
39279cc3 | 3729 | |
5f39d397 | 3730 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3731 | |
3732 | if (filled) | |
67de1176 | 3733 | goto cache_index; |
2f7e33d4 | 3734 | |
5f39d397 CM |
3735 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3736 | struct btrfs_inode_item); | |
5f39d397 | 3737 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3738 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3739 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3740 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3741 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3742 | |
a937b979 DS |
3743 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3744 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3745 | |
a937b979 DS |
3746 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3747 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3748 | |
a937b979 DS |
3749 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3750 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3751 | |
9cc97d64 | 3752 | BTRFS_I(inode)->i_otime.tv_sec = |
3753 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3754 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3755 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3756 | |
a76a3cd4 | 3757 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3758 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3759 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3760 | ||
c7f88c4e JL |
3761 | inode_set_iversion_queried(inode, |
3762 | btrfs_inode_sequence(leaf, inode_item)); | |
6e17d30b YD |
3763 | inode->i_generation = BTRFS_I(inode)->generation; |
3764 | inode->i_rdev = 0; | |
3765 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3766 | ||
3767 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3768 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3769 | ||
3770 | cache_index: | |
5dc562c5 JB |
3771 | /* |
3772 | * If we were modified in the current generation and evicted from memory | |
3773 | * and then re-read we need to do a full sync since we don't have any | |
3774 | * idea about which extents were modified before we were evicted from | |
3775 | * cache. | |
6e17d30b YD |
3776 | * |
3777 | * This is required for both inode re-read from disk and delayed inode | |
3778 | * in delayed_nodes_tree. | |
5dc562c5 | 3779 | */ |
0b246afa | 3780 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3781 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3782 | &BTRFS_I(inode)->runtime_flags); | |
3783 | ||
bde6c242 FM |
3784 | /* |
3785 | * We don't persist the id of the transaction where an unlink operation | |
3786 | * against the inode was last made. So here we assume the inode might | |
3787 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3788 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3789 | * between the inode and its parent if the inode is fsync'ed and the log | |
3790 | * replayed. For example, in the scenario: | |
3791 | * | |
3792 | * touch mydir/foo | |
3793 | * ln mydir/foo mydir/bar | |
3794 | * sync | |
3795 | * unlink mydir/bar | |
3796 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3797 | * xfs_io -c fsync mydir/foo | |
3798 | * <power failure> | |
3799 | * mount fs, triggers fsync log replay | |
3800 | * | |
3801 | * We must make sure that when we fsync our inode foo we also log its | |
3802 | * parent inode, otherwise after log replay the parent still has the | |
3803 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3804 | * and doesn't have an inode ref with the name "bar" anymore. | |
3805 | * | |
3806 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3807 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3808 | * transaction commits on fsync if our inode is a directory, or if our |
3809 | * inode is not a directory, logging its parent unnecessarily. | |
3810 | */ | |
3811 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3812 | ||
67de1176 MX |
3813 | path->slots[0]++; |
3814 | if (inode->i_nlink != 1 || | |
3815 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3816 | goto cache_acl; | |
3817 | ||
3818 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3819 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3820 | goto cache_acl; |
3821 | ||
3822 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3823 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3824 | struct btrfs_inode_ref *ref; | |
3825 | ||
3826 | ref = (struct btrfs_inode_ref *)ptr; | |
3827 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3828 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3829 | struct btrfs_inode_extref *extref; | |
3830 | ||
3831 | extref = (struct btrfs_inode_extref *)ptr; | |
3832 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3833 | extref); | |
3834 | } | |
2f7e33d4 | 3835 | cache_acl: |
46a53cca CM |
3836 | /* |
3837 | * try to precache a NULL acl entry for files that don't have | |
3838 | * any xattrs or acls | |
3839 | */ | |
33345d01 | 3840 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3841 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3842 | if (first_xattr_slot != -1) { |
3843 | path->slots[0] = first_xattr_slot; | |
3844 | ret = btrfs_load_inode_props(inode, path); | |
3845 | if (ret) | |
0b246afa | 3846 | btrfs_err(fs_info, |
351fd353 | 3847 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3848 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3849 | root->root_key.objectid, ret); |
3850 | } | |
4222ea71 FM |
3851 | if (path != in_path) |
3852 | btrfs_free_path(path); | |
63541927 | 3853 | |
72c04902 AV |
3854 | if (!maybe_acls) |
3855 | cache_no_acl(inode); | |
46a53cca | 3856 | |
39279cc3 | 3857 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3858 | case S_IFREG: |
3859 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3860 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3861 | inode->i_fop = &btrfs_file_operations; |
3862 | inode->i_op = &btrfs_file_inode_operations; | |
3863 | break; | |
3864 | case S_IFDIR: | |
3865 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3866 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3867 | break; |
3868 | case S_IFLNK: | |
3869 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3870 | inode_nohighmem(inode); |
4779cc04 | 3871 | inode->i_mapping->a_ops = &btrfs_aops; |
39279cc3 | 3872 | break; |
618e21d5 | 3873 | default: |
0279b4cd | 3874 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3875 | init_special_inode(inode, inode->i_mode, rdev); |
3876 | break; | |
39279cc3 | 3877 | } |
6cbff00f | 3878 | |
7b6a221e | 3879 | btrfs_sync_inode_flags_to_i_flags(inode); |
67710892 | 3880 | return 0; |
39279cc3 CM |
3881 | } |
3882 | ||
d352ac68 CM |
3883 | /* |
3884 | * given a leaf and an inode, copy the inode fields into the leaf | |
3885 | */ | |
e02119d5 CM |
3886 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3887 | struct extent_buffer *leaf, | |
5f39d397 | 3888 | struct btrfs_inode_item *item, |
39279cc3 CM |
3889 | struct inode *inode) |
3890 | { | |
51fab693 LB |
3891 | struct btrfs_map_token token; |
3892 | ||
3893 | btrfs_init_map_token(&token); | |
5f39d397 | 3894 | |
51fab693 LB |
3895 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3896 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3897 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3898 | &token); | |
3899 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3900 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3901 | |
a937b979 | 3902 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3903 | inode->i_atime.tv_sec, &token); |
a937b979 | 3904 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3905 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3906 | |
a937b979 | 3907 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3908 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3909 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3910 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3911 | |
a937b979 | 3912 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3913 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3914 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3915 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3916 | |
9cc97d64 | 3917 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3918 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3919 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3920 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3921 | ||
51fab693 LB |
3922 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3923 | &token); | |
3924 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3925 | &token); | |
c7f88c4e JL |
3926 | btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode), |
3927 | &token); | |
51fab693 LB |
3928 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); |
3929 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3930 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3931 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3932 | } |
3933 | ||
d352ac68 CM |
3934 | /* |
3935 | * copy everything in the in-memory inode into the btree. | |
3936 | */ | |
2115133f | 3937 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3938 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3939 | { |
3940 | struct btrfs_inode_item *inode_item; | |
3941 | struct btrfs_path *path; | |
5f39d397 | 3942 | struct extent_buffer *leaf; |
39279cc3 CM |
3943 | int ret; |
3944 | ||
3945 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3946 | if (!path) |
3947 | return -ENOMEM; | |
3948 | ||
b9473439 | 3949 | path->leave_spinning = 1; |
16cdcec7 MX |
3950 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3951 | 1); | |
39279cc3 CM |
3952 | if (ret) { |
3953 | if (ret > 0) | |
3954 | ret = -ENOENT; | |
3955 | goto failed; | |
3956 | } | |
3957 | ||
5f39d397 CM |
3958 | leaf = path->nodes[0]; |
3959 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3960 | struct btrfs_inode_item); |
39279cc3 | 3961 | |
e02119d5 | 3962 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3963 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3964 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3965 | ret = 0; |
3966 | failed: | |
39279cc3 CM |
3967 | btrfs_free_path(path); |
3968 | return ret; | |
3969 | } | |
3970 | ||
2115133f CM |
3971 | /* |
3972 | * copy everything in the in-memory inode into the btree. | |
3973 | */ | |
3974 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3975 | struct btrfs_root *root, struct inode *inode) | |
3976 | { | |
0b246afa | 3977 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
3978 | int ret; |
3979 | ||
3980 | /* | |
3981 | * If the inode is a free space inode, we can deadlock during commit | |
3982 | * if we put it into the delayed code. | |
3983 | * | |
3984 | * The data relocation inode should also be directly updated | |
3985 | * without delay | |
3986 | */ | |
70ddc553 | 3987 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) |
1d52c78a | 3988 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 3989 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
3990 | btrfs_update_root_times(trans, root); |
3991 | ||
2115133f CM |
3992 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3993 | if (!ret) | |
3994 | btrfs_set_inode_last_trans(trans, inode); | |
3995 | return ret; | |
3996 | } | |
3997 | ||
3998 | return btrfs_update_inode_item(trans, root, inode); | |
3999 | } | |
4000 | ||
be6aef60 JB |
4001 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
4002 | struct btrfs_root *root, | |
4003 | struct inode *inode) | |
2115133f CM |
4004 | { |
4005 | int ret; | |
4006 | ||
4007 | ret = btrfs_update_inode(trans, root, inode); | |
4008 | if (ret == -ENOSPC) | |
4009 | return btrfs_update_inode_item(trans, root, inode); | |
4010 | return ret; | |
4011 | } | |
4012 | ||
d352ac68 CM |
4013 | /* |
4014 | * unlink helper that gets used here in inode.c and in the tree logging | |
4015 | * recovery code. It remove a link in a directory with a given name, and | |
4016 | * also drops the back refs in the inode to the directory | |
4017 | */ | |
92986796 AV |
4018 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4019 | struct btrfs_root *root, | |
4ec5934e NB |
4020 | struct btrfs_inode *dir, |
4021 | struct btrfs_inode *inode, | |
92986796 | 4022 | const char *name, int name_len) |
39279cc3 | 4023 | { |
0b246afa | 4024 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4025 | struct btrfs_path *path; |
39279cc3 | 4026 | int ret = 0; |
39279cc3 | 4027 | struct btrfs_dir_item *di; |
aec7477b | 4028 | u64 index; |
33345d01 LZ |
4029 | u64 ino = btrfs_ino(inode); |
4030 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
4031 | |
4032 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
4033 | if (!path) { |
4034 | ret = -ENOMEM; | |
554233a6 | 4035 | goto out; |
54aa1f4d CM |
4036 | } |
4037 | ||
b9473439 | 4038 | path->leave_spinning = 1; |
33345d01 | 4039 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 | 4040 | name, name_len, -1); |
3cf5068f LB |
4041 | if (IS_ERR_OR_NULL(di)) { |
4042 | ret = di ? PTR_ERR(di) : -ENOENT; | |
39279cc3 CM |
4043 | goto err; |
4044 | } | |
39279cc3 | 4045 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
4046 | if (ret) |
4047 | goto err; | |
b3b4aa74 | 4048 | btrfs_release_path(path); |
39279cc3 | 4049 | |
67de1176 MX |
4050 | /* |
4051 | * If we don't have dir index, we have to get it by looking up | |
4052 | * the inode ref, since we get the inode ref, remove it directly, | |
4053 | * it is unnecessary to do delayed deletion. | |
4054 | * | |
4055 | * But if we have dir index, needn't search inode ref to get it. | |
4056 | * Since the inode ref is close to the inode item, it is better | |
4057 | * that we delay to delete it, and just do this deletion when | |
4058 | * we update the inode item. | |
4059 | */ | |
4ec5934e | 4060 | if (inode->dir_index) { |
67de1176 MX |
4061 | ret = btrfs_delayed_delete_inode_ref(inode); |
4062 | if (!ret) { | |
4ec5934e | 4063 | index = inode->dir_index; |
67de1176 MX |
4064 | goto skip_backref; |
4065 | } | |
4066 | } | |
4067 | ||
33345d01 LZ |
4068 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
4069 | dir_ino, &index); | |
aec7477b | 4070 | if (ret) { |
0b246afa | 4071 | btrfs_info(fs_info, |
c2cf52eb | 4072 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 4073 | name_len, name, ino, dir_ino); |
66642832 | 4074 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
4075 | goto err; |
4076 | } | |
67de1176 | 4077 | skip_backref: |
9add2945 | 4078 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 4079 | if (ret) { |
66642832 | 4080 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 4081 | goto err; |
79787eaa | 4082 | } |
39279cc3 | 4083 | |
4ec5934e NB |
4084 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
4085 | dir_ino); | |
79787eaa | 4086 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 4087 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4088 | goto err; |
4089 | } | |
e02119d5 | 4090 | |
4ec5934e NB |
4091 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
4092 | index); | |
6418c961 CM |
4093 | if (ret == -ENOENT) |
4094 | ret = 0; | |
d4e3991b | 4095 | else if (ret) |
66642832 | 4096 | btrfs_abort_transaction(trans, ret); |
63611e73 JB |
4097 | |
4098 | /* | |
4099 | * If we have a pending delayed iput we could end up with the final iput | |
4100 | * being run in btrfs-cleaner context. If we have enough of these built | |
4101 | * up we can end up burning a lot of time in btrfs-cleaner without any | |
4102 | * way to throttle the unlinks. Since we're currently holding a ref on | |
4103 | * the inode we can run the delayed iput here without any issues as the | |
4104 | * final iput won't be done until after we drop the ref we're currently | |
4105 | * holding. | |
4106 | */ | |
4107 | btrfs_run_delayed_iput(fs_info, inode); | |
39279cc3 CM |
4108 | err: |
4109 | btrfs_free_path(path); | |
e02119d5 CM |
4110 | if (ret) |
4111 | goto out; | |
4112 | ||
6ef06d27 | 4113 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
4114 | inode_inc_iversion(&inode->vfs_inode); |
4115 | inode_inc_iversion(&dir->vfs_inode); | |
4116 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
4117 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
4118 | ret = btrfs_update_inode(trans, root, &dir->vfs_inode); | |
e02119d5 | 4119 | out: |
39279cc3 CM |
4120 | return ret; |
4121 | } | |
4122 | ||
92986796 AV |
4123 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4124 | struct btrfs_root *root, | |
4ec5934e | 4125 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
4126 | const char *name, int name_len) |
4127 | { | |
4128 | int ret; | |
4129 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4130 | if (!ret) { | |
4ec5934e NB |
4131 | drop_nlink(&inode->vfs_inode); |
4132 | ret = btrfs_update_inode(trans, root, &inode->vfs_inode); | |
92986796 AV |
4133 | } |
4134 | return ret; | |
4135 | } | |
39279cc3 | 4136 | |
a22285a6 YZ |
4137 | /* |
4138 | * helper to start transaction for unlink and rmdir. | |
4139 | * | |
d52be818 JB |
4140 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4141 | * if we cannot make our reservations the normal way try and see if there is | |
4142 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4143 | * allow the unlink to occur. | |
a22285a6 | 4144 | */ |
d52be818 | 4145 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4146 | { |
a22285a6 | 4147 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4148 | |
e70bea5f JB |
4149 | /* |
4150 | * 1 for the possible orphan item | |
4151 | * 1 for the dir item | |
4152 | * 1 for the dir index | |
4153 | * 1 for the inode ref | |
e70bea5f JB |
4154 | * 1 for the inode |
4155 | */ | |
8eab77ff | 4156 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4157 | } |
4158 | ||
4159 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4160 | { | |
4161 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4162 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4163 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4164 | int ret; |
a22285a6 | 4165 | |
d52be818 | 4166 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4167 | if (IS_ERR(trans)) |
4168 | return PTR_ERR(trans); | |
5f39d397 | 4169 | |
4ec5934e NB |
4170 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4171 | 0); | |
12fcfd22 | 4172 | |
4ec5934e NB |
4173 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4174 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4175 | dentry->d_name.len); | |
b532402e TI |
4176 | if (ret) |
4177 | goto out; | |
7b128766 | 4178 | |
a22285a6 | 4179 | if (inode->i_nlink == 0) { |
73f2e545 | 4180 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
4181 | if (ret) |
4182 | goto out; | |
a22285a6 | 4183 | } |
7b128766 | 4184 | |
b532402e | 4185 | out: |
3a45bb20 | 4186 | btrfs_end_transaction(trans); |
2ff7e61e | 4187 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
4188 | return ret; |
4189 | } | |
4190 | ||
f60a2364 | 4191 | static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
401b3b19 LF |
4192 | struct inode *dir, u64 objectid, |
4193 | const char *name, int name_len) | |
4df27c4d | 4194 | { |
401b3b19 | 4195 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d YZ |
4196 | struct btrfs_path *path; |
4197 | struct extent_buffer *leaf; | |
4198 | struct btrfs_dir_item *di; | |
4199 | struct btrfs_key key; | |
4200 | u64 index; | |
4201 | int ret; | |
4a0cc7ca | 4202 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d YZ |
4203 | |
4204 | path = btrfs_alloc_path(); | |
4205 | if (!path) | |
4206 | return -ENOMEM; | |
4207 | ||
33345d01 | 4208 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4209 | name, name_len, -1); |
79787eaa | 4210 | if (IS_ERR_OR_NULL(di)) { |
3cf5068f | 4211 | ret = di ? PTR_ERR(di) : -ENOENT; |
79787eaa JM |
4212 | goto out; |
4213 | } | |
4df27c4d YZ |
4214 | |
4215 | leaf = path->nodes[0]; | |
4216 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4217 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4218 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4219 | if (ret) { |
66642832 | 4220 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4221 | goto out; |
4222 | } | |
b3b4aa74 | 4223 | btrfs_release_path(path); |
4df27c4d | 4224 | |
3ee1c553 LF |
4225 | ret = btrfs_del_root_ref(trans, objectid, root->root_key.objectid, |
4226 | dir_ino, &index, name, name_len); | |
4df27c4d | 4227 | if (ret < 0) { |
79787eaa | 4228 | if (ret != -ENOENT) { |
66642832 | 4229 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4230 | goto out; |
4231 | } | |
33345d01 | 4232 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4233 | name, name_len); |
79787eaa JM |
4234 | if (IS_ERR_OR_NULL(di)) { |
4235 | if (!di) | |
4236 | ret = -ENOENT; | |
4237 | else | |
4238 | ret = PTR_ERR(di); | |
66642832 | 4239 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4240 | goto out; |
4241 | } | |
4df27c4d YZ |
4242 | |
4243 | leaf = path->nodes[0]; | |
4244 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4df27c4d YZ |
4245 | index = key.offset; |
4246 | } | |
945d8962 | 4247 | btrfs_release_path(path); |
4df27c4d | 4248 | |
9add2945 | 4249 | ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index); |
79787eaa | 4250 | if (ret) { |
66642832 | 4251 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4252 | goto out; |
4253 | } | |
4df27c4d | 4254 | |
6ef06d27 | 4255 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 4256 | inode_inc_iversion(dir); |
c2050a45 | 4257 | dir->i_mtime = dir->i_ctime = current_time(dir); |
5a24e84c | 4258 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa | 4259 | if (ret) |
66642832 | 4260 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4261 | out: |
71d7aed0 | 4262 | btrfs_free_path(path); |
79787eaa | 4263 | return ret; |
4df27c4d YZ |
4264 | } |
4265 | ||
ec42f167 MT |
4266 | /* |
4267 | * Helper to check if the subvolume references other subvolumes or if it's | |
4268 | * default. | |
4269 | */ | |
f60a2364 | 4270 | static noinline int may_destroy_subvol(struct btrfs_root *root) |
ec42f167 MT |
4271 | { |
4272 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4273 | struct btrfs_path *path; | |
4274 | struct btrfs_dir_item *di; | |
4275 | struct btrfs_key key; | |
4276 | u64 dir_id; | |
4277 | int ret; | |
4278 | ||
4279 | path = btrfs_alloc_path(); | |
4280 | if (!path) | |
4281 | return -ENOMEM; | |
4282 | ||
4283 | /* Make sure this root isn't set as the default subvol */ | |
4284 | dir_id = btrfs_super_root_dir(fs_info->super_copy); | |
4285 | di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path, | |
4286 | dir_id, "default", 7, 0); | |
4287 | if (di && !IS_ERR(di)) { | |
4288 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
4289 | if (key.objectid == root->root_key.objectid) { | |
4290 | ret = -EPERM; | |
4291 | btrfs_err(fs_info, | |
4292 | "deleting default subvolume %llu is not allowed", | |
4293 | key.objectid); | |
4294 | goto out; | |
4295 | } | |
4296 | btrfs_release_path(path); | |
4297 | } | |
4298 | ||
4299 | key.objectid = root->root_key.objectid; | |
4300 | key.type = BTRFS_ROOT_REF_KEY; | |
4301 | key.offset = (u64)-1; | |
4302 | ||
4303 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
4304 | if (ret < 0) | |
4305 | goto out; | |
4306 | BUG_ON(ret == 0); | |
4307 | ||
4308 | ret = 0; | |
4309 | if (path->slots[0] > 0) { | |
4310 | path->slots[0]--; | |
4311 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
4312 | if (key.objectid == root->root_key.objectid && | |
4313 | key.type == BTRFS_ROOT_REF_KEY) | |
4314 | ret = -ENOTEMPTY; | |
4315 | } | |
4316 | out: | |
4317 | btrfs_free_path(path); | |
4318 | return ret; | |
4319 | } | |
4320 | ||
20a68004 NB |
4321 | /* Delete all dentries for inodes belonging to the root */ |
4322 | static void btrfs_prune_dentries(struct btrfs_root *root) | |
4323 | { | |
4324 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4325 | struct rb_node *node; | |
4326 | struct rb_node *prev; | |
4327 | struct btrfs_inode *entry; | |
4328 | struct inode *inode; | |
4329 | u64 objectid = 0; | |
4330 | ||
4331 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) | |
4332 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4333 | ||
4334 | spin_lock(&root->inode_lock); | |
4335 | again: | |
4336 | node = root->inode_tree.rb_node; | |
4337 | prev = NULL; | |
4338 | while (node) { | |
4339 | prev = node; | |
4340 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4341 | ||
37508515 | 4342 | if (objectid < btrfs_ino(entry)) |
20a68004 | 4343 | node = node->rb_left; |
37508515 | 4344 | else if (objectid > btrfs_ino(entry)) |
20a68004 NB |
4345 | node = node->rb_right; |
4346 | else | |
4347 | break; | |
4348 | } | |
4349 | if (!node) { | |
4350 | while (prev) { | |
4351 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
37508515 | 4352 | if (objectid <= btrfs_ino(entry)) { |
20a68004 NB |
4353 | node = prev; |
4354 | break; | |
4355 | } | |
4356 | prev = rb_next(prev); | |
4357 | } | |
4358 | } | |
4359 | while (node) { | |
4360 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
37508515 | 4361 | objectid = btrfs_ino(entry) + 1; |
20a68004 NB |
4362 | inode = igrab(&entry->vfs_inode); |
4363 | if (inode) { | |
4364 | spin_unlock(&root->inode_lock); | |
4365 | if (atomic_read(&inode->i_count) > 1) | |
4366 | d_prune_aliases(inode); | |
4367 | /* | |
4368 | * btrfs_drop_inode will have it removed from the inode | |
4369 | * cache when its usage count hits zero. | |
4370 | */ | |
4371 | iput(inode); | |
4372 | cond_resched(); | |
4373 | spin_lock(&root->inode_lock); | |
4374 | goto again; | |
4375 | } | |
4376 | ||
4377 | if (cond_resched_lock(&root->inode_lock)) | |
4378 | goto again; | |
4379 | ||
4380 | node = rb_next(node); | |
4381 | } | |
4382 | spin_unlock(&root->inode_lock); | |
4383 | } | |
4384 | ||
f60a2364 MT |
4385 | int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry) |
4386 | { | |
4387 | struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb); | |
4388 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4389 | struct inode *inode = d_inode(dentry); | |
4390 | struct btrfs_root *dest = BTRFS_I(inode)->root; | |
4391 | struct btrfs_trans_handle *trans; | |
4392 | struct btrfs_block_rsv block_rsv; | |
4393 | u64 root_flags; | |
f60a2364 MT |
4394 | int ret; |
4395 | int err; | |
4396 | ||
4397 | /* | |
4398 | * Don't allow to delete a subvolume with send in progress. This is | |
4399 | * inside the inode lock so the error handling that has to drop the bit | |
4400 | * again is not run concurrently. | |
4401 | */ | |
4402 | spin_lock(&dest->root_item_lock); | |
a7176f74 | 4403 | if (dest->send_in_progress) { |
f60a2364 MT |
4404 | spin_unlock(&dest->root_item_lock); |
4405 | btrfs_warn(fs_info, | |
4406 | "attempt to delete subvolume %llu during send", | |
4407 | dest->root_key.objectid); | |
4408 | return -EPERM; | |
4409 | } | |
a7176f74 LF |
4410 | root_flags = btrfs_root_flags(&dest->root_item); |
4411 | btrfs_set_root_flags(&dest->root_item, | |
4412 | root_flags | BTRFS_ROOT_SUBVOL_DEAD); | |
4413 | spin_unlock(&dest->root_item_lock); | |
f60a2364 MT |
4414 | |
4415 | down_write(&fs_info->subvol_sem); | |
4416 | ||
4417 | err = may_destroy_subvol(dest); | |
4418 | if (err) | |
4419 | goto out_up_write; | |
4420 | ||
4421 | btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); | |
4422 | /* | |
4423 | * One for dir inode, | |
4424 | * two for dir entries, | |
4425 | * two for root ref/backref. | |
4426 | */ | |
c4c129db | 4427 | err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true); |
f60a2364 MT |
4428 | if (err) |
4429 | goto out_up_write; | |
4430 | ||
4431 | trans = btrfs_start_transaction(root, 0); | |
4432 | if (IS_ERR(trans)) { | |
4433 | err = PTR_ERR(trans); | |
4434 | goto out_release; | |
4435 | } | |
4436 | trans->block_rsv = &block_rsv; | |
4437 | trans->bytes_reserved = block_rsv.size; | |
4438 | ||
4439 | btrfs_record_snapshot_destroy(trans, BTRFS_I(dir)); | |
4440 | ||
401b3b19 LF |
4441 | ret = btrfs_unlink_subvol(trans, dir, dest->root_key.objectid, |
4442 | dentry->d_name.name, dentry->d_name.len); | |
f60a2364 MT |
4443 | if (ret) { |
4444 | err = ret; | |
4445 | btrfs_abort_transaction(trans, ret); | |
4446 | goto out_end_trans; | |
4447 | } | |
4448 | ||
4449 | btrfs_record_root_in_trans(trans, dest); | |
4450 | ||
4451 | memset(&dest->root_item.drop_progress, 0, | |
4452 | sizeof(dest->root_item.drop_progress)); | |
4453 | dest->root_item.drop_level = 0; | |
4454 | btrfs_set_root_refs(&dest->root_item, 0); | |
4455 | ||
4456 | if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { | |
4457 | ret = btrfs_insert_orphan_item(trans, | |
4458 | fs_info->tree_root, | |
4459 | dest->root_key.objectid); | |
4460 | if (ret) { | |
4461 | btrfs_abort_transaction(trans, ret); | |
4462 | err = ret; | |
4463 | goto out_end_trans; | |
4464 | } | |
4465 | } | |
4466 | ||
d1957791 | 4467 | ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid, |
f60a2364 MT |
4468 | BTRFS_UUID_KEY_SUBVOL, |
4469 | dest->root_key.objectid); | |
4470 | if (ret && ret != -ENOENT) { | |
4471 | btrfs_abort_transaction(trans, ret); | |
4472 | err = ret; | |
4473 | goto out_end_trans; | |
4474 | } | |
4475 | if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { | |
d1957791 | 4476 | ret = btrfs_uuid_tree_remove(trans, |
f60a2364 MT |
4477 | dest->root_item.received_uuid, |
4478 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4479 | dest->root_key.objectid); | |
4480 | if (ret && ret != -ENOENT) { | |
4481 | btrfs_abort_transaction(trans, ret); | |
4482 | err = ret; | |
4483 | goto out_end_trans; | |
4484 | } | |
4485 | } | |
4486 | ||
4487 | out_end_trans: | |
4488 | trans->block_rsv = NULL; | |
4489 | trans->bytes_reserved = 0; | |
4490 | ret = btrfs_end_transaction(trans); | |
4491 | if (ret && !err) | |
4492 | err = ret; | |
4493 | inode->i_flags |= S_DEAD; | |
4494 | out_release: | |
4495 | btrfs_subvolume_release_metadata(fs_info, &block_rsv); | |
4496 | out_up_write: | |
4497 | up_write(&fs_info->subvol_sem); | |
4498 | if (err) { | |
4499 | spin_lock(&dest->root_item_lock); | |
4500 | root_flags = btrfs_root_flags(&dest->root_item); | |
4501 | btrfs_set_root_flags(&dest->root_item, | |
4502 | root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); | |
4503 | spin_unlock(&dest->root_item_lock); | |
4504 | } else { | |
4505 | d_invalidate(dentry); | |
20a68004 | 4506 | btrfs_prune_dentries(dest); |
f60a2364 MT |
4507 | ASSERT(dest->send_in_progress == 0); |
4508 | ||
4509 | /* the last ref */ | |
4510 | if (dest->ino_cache_inode) { | |
4511 | iput(dest->ino_cache_inode); | |
4512 | dest->ino_cache_inode = NULL; | |
4513 | } | |
4514 | } | |
4515 | ||
4516 | return err; | |
4517 | } | |
4518 | ||
39279cc3 CM |
4519 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4520 | { | |
2b0143b5 | 4521 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4522 | int err = 0; |
39279cc3 | 4523 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4524 | struct btrfs_trans_handle *trans; |
44f714da | 4525 | u64 last_unlink_trans; |
39279cc3 | 4526 | |
b3ae244e | 4527 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4528 | return -ENOTEMPTY; |
4a0cc7ca | 4529 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
a79a464d | 4530 | return btrfs_delete_subvolume(dir, dentry); |
134d4512 | 4531 | |
d52be818 | 4532 | trans = __unlink_start_trans(dir); |
a22285a6 | 4533 | if (IS_ERR(trans)) |
5df6a9f6 | 4534 | return PTR_ERR(trans); |
5df6a9f6 | 4535 | |
4a0cc7ca | 4536 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
401b3b19 | 4537 | err = btrfs_unlink_subvol(trans, dir, |
4df27c4d YZ |
4538 | BTRFS_I(inode)->location.objectid, |
4539 | dentry->d_name.name, | |
4540 | dentry->d_name.len); | |
4541 | goto out; | |
4542 | } | |
4543 | ||
73f2e545 | 4544 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4545 | if (err) |
4df27c4d | 4546 | goto out; |
7b128766 | 4547 | |
44f714da FM |
4548 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4549 | ||
39279cc3 | 4550 | /* now the directory is empty */ |
4ec5934e NB |
4551 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4552 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4553 | dentry->d_name.len); | |
44f714da | 4554 | if (!err) { |
6ef06d27 | 4555 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4556 | /* |
4557 | * Propagate the last_unlink_trans value of the deleted dir to | |
4558 | * its parent directory. This is to prevent an unrecoverable | |
4559 | * log tree in the case we do something like this: | |
4560 | * 1) create dir foo | |
4561 | * 2) create snapshot under dir foo | |
4562 | * 3) delete the snapshot | |
4563 | * 4) rmdir foo | |
4564 | * 5) mkdir foo | |
4565 | * 6) fsync foo or some file inside foo | |
4566 | */ | |
4567 | if (last_unlink_trans >= trans->transid) | |
4568 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4569 | } | |
4df27c4d | 4570 | out: |
3a45bb20 | 4571 | btrfs_end_transaction(trans); |
2ff7e61e | 4572 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4573 | |
39279cc3 CM |
4574 | return err; |
4575 | } | |
4576 | ||
ddfae63c JB |
4577 | /* |
4578 | * Return this if we need to call truncate_block for the last bit of the | |
4579 | * truncate. | |
4580 | */ | |
4581 | #define NEED_TRUNCATE_BLOCK 1 | |
0305cd5f | 4582 | |
39279cc3 CM |
4583 | /* |
4584 | * this can truncate away extent items, csum items and directory items. | |
4585 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4586 | * any higher than new_size |
39279cc3 CM |
4587 | * |
4588 | * csum items that cross the new i_size are truncated to the new size | |
4589 | * as well. | |
7b128766 JB |
4590 | * |
4591 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4592 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4593 | */ |
8082510e YZ |
4594 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4595 | struct btrfs_root *root, | |
4596 | struct inode *inode, | |
4597 | u64 new_size, u32 min_type) | |
39279cc3 | 4598 | { |
0b246afa | 4599 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4600 | struct btrfs_path *path; |
5f39d397 | 4601 | struct extent_buffer *leaf; |
39279cc3 | 4602 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4603 | struct btrfs_key key; |
4604 | struct btrfs_key found_key; | |
39279cc3 | 4605 | u64 extent_start = 0; |
db94535d | 4606 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4607 | u64 extent_offset = 0; |
39279cc3 | 4608 | u64 item_end = 0; |
c1aa4575 | 4609 | u64 last_size = new_size; |
8082510e | 4610 | u32 found_type = (u8)-1; |
39279cc3 CM |
4611 | int found_extent; |
4612 | int del_item; | |
85e21bac CM |
4613 | int pending_del_nr = 0; |
4614 | int pending_del_slot = 0; | |
179e29e4 | 4615 | int extent_type = -1; |
8082510e | 4616 | int ret; |
4a0cc7ca | 4617 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
28ed1345 | 4618 | u64 bytes_deleted = 0; |
897ca819 TM |
4619 | bool be_nice = false; |
4620 | bool should_throttle = false; | |
8082510e YZ |
4621 | |
4622 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4623 | |
28ed1345 CM |
4624 | /* |
4625 | * for non-free space inodes and ref cows, we want to back off from | |
4626 | * time to time | |
4627 | */ | |
70ddc553 | 4628 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) && |
28ed1345 | 4629 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
897ca819 | 4630 | be_nice = true; |
28ed1345 | 4631 | |
0eb0e19c MF |
4632 | path = btrfs_alloc_path(); |
4633 | if (!path) | |
4634 | return -ENOMEM; | |
e4058b54 | 4635 | path->reada = READA_BACK; |
0eb0e19c | 4636 | |
5dc562c5 JB |
4637 | /* |
4638 | * We want to drop from the next block forward in case this new size is | |
4639 | * not block aligned since we will be keeping the last block of the | |
4640 | * extent just the way it is. | |
4641 | */ | |
27cdeb70 | 4642 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 4643 | root == fs_info->tree_root) |
dcdbc059 | 4644 | btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size, |
0b246afa | 4645 | fs_info->sectorsize), |
da17066c | 4646 | (u64)-1, 0); |
8082510e | 4647 | |
16cdcec7 MX |
4648 | /* |
4649 | * This function is also used to drop the items in the log tree before | |
4650 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
52042d8e | 4651 | * it is used to drop the logged items. So we shouldn't kill the delayed |
16cdcec7 MX |
4652 | * items. |
4653 | */ | |
4654 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4ccb5c72 | 4655 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); |
16cdcec7 | 4656 | |
33345d01 | 4657 | key.objectid = ino; |
39279cc3 | 4658 | key.offset = (u64)-1; |
5f39d397 CM |
4659 | key.type = (u8)-1; |
4660 | ||
85e21bac | 4661 | search_again: |
28ed1345 CM |
4662 | /* |
4663 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4664 | * up a huge file in a single leaf. Most of the time that | |
4665 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4666 | */ | |
fd86a3a3 OS |
4667 | if (be_nice && bytes_deleted > SZ_32M && |
4668 | btrfs_should_end_transaction(trans)) { | |
4669 | ret = -EAGAIN; | |
4670 | goto out; | |
28ed1345 CM |
4671 | } |
4672 | ||
b9473439 | 4673 | path->leave_spinning = 1; |
85e21bac | 4674 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
fd86a3a3 | 4675 | if (ret < 0) |
8082510e | 4676 | goto out; |
d397712b | 4677 | |
85e21bac | 4678 | if (ret > 0) { |
fd86a3a3 | 4679 | ret = 0; |
e02119d5 CM |
4680 | /* there are no items in the tree for us to truncate, we're |
4681 | * done | |
4682 | */ | |
8082510e YZ |
4683 | if (path->slots[0] == 0) |
4684 | goto out; | |
85e21bac CM |
4685 | path->slots[0]--; |
4686 | } | |
4687 | ||
d397712b | 4688 | while (1) { |
39279cc3 | 4689 | fi = NULL; |
5f39d397 CM |
4690 | leaf = path->nodes[0]; |
4691 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4692 | found_type = found_key.type; |
39279cc3 | 4693 | |
33345d01 | 4694 | if (found_key.objectid != ino) |
39279cc3 | 4695 | break; |
5f39d397 | 4696 | |
85e21bac | 4697 | if (found_type < min_type) |
39279cc3 CM |
4698 | break; |
4699 | ||
5f39d397 | 4700 | item_end = found_key.offset; |
39279cc3 | 4701 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4702 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4703 | struct btrfs_file_extent_item); |
179e29e4 CM |
4704 | extent_type = btrfs_file_extent_type(leaf, fi); |
4705 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4706 | item_end += |
db94535d | 4707 | btrfs_file_extent_num_bytes(leaf, fi); |
09ed2f16 LB |
4708 | |
4709 | trace_btrfs_truncate_show_fi_regular( | |
4710 | BTRFS_I(inode), leaf, fi, | |
4711 | found_key.offset); | |
179e29e4 | 4712 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
e41ca589 QW |
4713 | item_end += btrfs_file_extent_ram_bytes(leaf, |
4714 | fi); | |
09ed2f16 LB |
4715 | |
4716 | trace_btrfs_truncate_show_fi_inline( | |
4717 | BTRFS_I(inode), leaf, fi, path->slots[0], | |
4718 | found_key.offset); | |
39279cc3 | 4719 | } |
008630c1 | 4720 | item_end--; |
39279cc3 | 4721 | } |
8082510e YZ |
4722 | if (found_type > min_type) { |
4723 | del_item = 1; | |
4724 | } else { | |
76b42abb | 4725 | if (item_end < new_size) |
b888db2b | 4726 | break; |
8082510e YZ |
4727 | if (found_key.offset >= new_size) |
4728 | del_item = 1; | |
4729 | else | |
4730 | del_item = 0; | |
39279cc3 | 4731 | } |
39279cc3 | 4732 | found_extent = 0; |
39279cc3 | 4733 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4734 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4735 | goto delete; | |
4736 | ||
4737 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 4738 | u64 num_dec; |
db94535d | 4739 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4740 | if (!del_item) { |
db94535d CM |
4741 | u64 orig_num_bytes = |
4742 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4743 | extent_num_bytes = ALIGN(new_size - |
4744 | found_key.offset, | |
0b246afa | 4745 | fs_info->sectorsize); |
db94535d CM |
4746 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4747 | extent_num_bytes); | |
4748 | num_dec = (orig_num_bytes - | |
9069218d | 4749 | extent_num_bytes); |
27cdeb70 MX |
4750 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4751 | &root->state) && | |
4752 | extent_start != 0) | |
a76a3cd4 | 4753 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4754 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4755 | } else { |
db94535d CM |
4756 | extent_num_bytes = |
4757 | btrfs_file_extent_disk_num_bytes(leaf, | |
4758 | fi); | |
5d4f98a2 YZ |
4759 | extent_offset = found_key.offset - |
4760 | btrfs_file_extent_offset(leaf, fi); | |
4761 | ||
39279cc3 | 4762 | /* FIXME blocksize != 4096 */ |
9069218d | 4763 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4764 | if (extent_start != 0) { |
4765 | found_extent = 1; | |
27cdeb70 MX |
4766 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4767 | &root->state)) | |
a76a3cd4 | 4768 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4769 | } |
39279cc3 | 4770 | } |
9069218d | 4771 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4772 | /* |
4773 | * we can't truncate inline items that have had | |
4774 | * special encodings | |
4775 | */ | |
4776 | if (!del_item && | |
c8b97818 | 4777 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
ddfae63c JB |
4778 | btrfs_file_extent_other_encoding(leaf, fi) == 0 && |
4779 | btrfs_file_extent_compression(leaf, fi) == 0) { | |
4780 | u32 size = (u32)(new_size - found_key.offset); | |
4781 | ||
4782 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4783 | size = btrfs_file_extent_calc_inline_size(size); | |
78ac4f9e | 4784 | btrfs_truncate_item(path, size, 1); |
ddfae63c | 4785 | } else if (!del_item) { |
514ac8ad | 4786 | /* |
ddfae63c JB |
4787 | * We have to bail so the last_size is set to |
4788 | * just before this extent. | |
514ac8ad | 4789 | */ |
fd86a3a3 | 4790 | ret = NEED_TRUNCATE_BLOCK; |
ddfae63c JB |
4791 | break; |
4792 | } | |
0305cd5f | 4793 | |
ddfae63c | 4794 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
0305cd5f | 4795 | inode_sub_bytes(inode, item_end + 1 - new_size); |
39279cc3 | 4796 | } |
179e29e4 | 4797 | delete: |
ddfae63c JB |
4798 | if (del_item) |
4799 | last_size = found_key.offset; | |
4800 | else | |
4801 | last_size = new_size; | |
39279cc3 | 4802 | if (del_item) { |
85e21bac CM |
4803 | if (!pending_del_nr) { |
4804 | /* no pending yet, add ourselves */ | |
4805 | pending_del_slot = path->slots[0]; | |
4806 | pending_del_nr = 1; | |
4807 | } else if (pending_del_nr && | |
4808 | path->slots[0] + 1 == pending_del_slot) { | |
4809 | /* hop on the pending chunk */ | |
4810 | pending_del_nr++; | |
4811 | pending_del_slot = path->slots[0]; | |
4812 | } else { | |
d397712b | 4813 | BUG(); |
85e21bac | 4814 | } |
39279cc3 CM |
4815 | } else { |
4816 | break; | |
4817 | } | |
897ca819 | 4818 | should_throttle = false; |
28f75a0e | 4819 | |
27cdeb70 MX |
4820 | if (found_extent && |
4821 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
0b246afa | 4822 | root == fs_info->tree_root)) { |
ffd4bb2a QW |
4823 | struct btrfs_ref ref = { 0 }; |
4824 | ||
b9473439 | 4825 | btrfs_set_path_blocking(path); |
28ed1345 | 4826 | bytes_deleted += extent_num_bytes; |
ffd4bb2a QW |
4827 | |
4828 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, | |
4829 | extent_start, extent_num_bytes, 0); | |
4830 | ref.real_root = root->root_key.objectid; | |
4831 | btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), | |
4832 | ino, extent_offset); | |
4833 | ret = btrfs_free_extent(trans, &ref); | |
05522109 OS |
4834 | if (ret) { |
4835 | btrfs_abort_transaction(trans, ret); | |
4836 | break; | |
4837 | } | |
28f75a0e | 4838 | if (be_nice) { |
7c861627 | 4839 | if (btrfs_should_throttle_delayed_refs(trans)) |
897ca819 | 4840 | should_throttle = true; |
28f75a0e | 4841 | } |
39279cc3 | 4842 | } |
85e21bac | 4843 | |
8082510e YZ |
4844 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4845 | break; | |
4846 | ||
4847 | if (path->slots[0] == 0 || | |
1262133b | 4848 | path->slots[0] != pending_del_slot || |
28bad212 | 4849 | should_throttle) { |
8082510e YZ |
4850 | if (pending_del_nr) { |
4851 | ret = btrfs_del_items(trans, root, path, | |
4852 | pending_del_slot, | |
4853 | pending_del_nr); | |
79787eaa | 4854 | if (ret) { |
66642832 | 4855 | btrfs_abort_transaction(trans, ret); |
fd86a3a3 | 4856 | break; |
79787eaa | 4857 | } |
8082510e YZ |
4858 | pending_del_nr = 0; |
4859 | } | |
b3b4aa74 | 4860 | btrfs_release_path(path); |
28bad212 | 4861 | |
28f75a0e | 4862 | /* |
28bad212 JB |
4863 | * We can generate a lot of delayed refs, so we need to |
4864 | * throttle every once and a while and make sure we're | |
4865 | * adding enough space to keep up with the work we are | |
4866 | * generating. Since we hold a transaction here we | |
4867 | * can't flush, and we don't want to FLUSH_LIMIT because | |
4868 | * we could have generated too many delayed refs to | |
4869 | * actually allocate, so just bail if we're short and | |
4870 | * let the normal reservation dance happen higher up. | |
28f75a0e | 4871 | */ |
28bad212 JB |
4872 | if (should_throttle) { |
4873 | ret = btrfs_delayed_refs_rsv_refill(fs_info, | |
4874 | BTRFS_RESERVE_NO_FLUSH); | |
4875 | if (ret) { | |
4876 | ret = -EAGAIN; | |
4877 | break; | |
4878 | } | |
28f75a0e | 4879 | } |
85e21bac | 4880 | goto search_again; |
8082510e YZ |
4881 | } else { |
4882 | path->slots[0]--; | |
85e21bac | 4883 | } |
39279cc3 | 4884 | } |
8082510e | 4885 | out: |
fd86a3a3 OS |
4886 | if (ret >= 0 && pending_del_nr) { |
4887 | int err; | |
4888 | ||
4889 | err = btrfs_del_items(trans, root, path, pending_del_slot, | |
85e21bac | 4890 | pending_del_nr); |
fd86a3a3 OS |
4891 | if (err) { |
4892 | btrfs_abort_transaction(trans, err); | |
4893 | ret = err; | |
4894 | } | |
85e21bac | 4895 | } |
76b42abb FM |
4896 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
4897 | ASSERT(last_size >= new_size); | |
fd86a3a3 | 4898 | if (!ret && last_size > new_size) |
76b42abb | 4899 | last_size = new_size; |
7f4f6e0a | 4900 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
76b42abb | 4901 | } |
28ed1345 | 4902 | |
39279cc3 | 4903 | btrfs_free_path(path); |
fd86a3a3 | 4904 | return ret; |
39279cc3 CM |
4905 | } |
4906 | ||
4907 | /* | |
9703fefe | 4908 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4909 | * @inode - inode that we're zeroing |
4910 | * @from - the offset to start zeroing | |
4911 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4912 | * offset | |
4913 | * @front - zero up to the offset instead of from the offset on | |
4914 | * | |
9703fefe | 4915 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4916 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4917 | */ |
9703fefe | 4918 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4919 | int front) |
39279cc3 | 4920 | { |
0b246afa | 4921 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 | 4922 | struct address_space *mapping = inode->i_mapping; |
e6dcd2dc CM |
4923 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4924 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4925 | struct extent_state *cached_state = NULL; |
364ecf36 | 4926 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 4927 | char *kaddr; |
0b246afa | 4928 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4929 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4930 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4931 | struct page *page; |
3b16a4e3 | 4932 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4933 | int ret = 0; |
9703fefe CR |
4934 | u64 block_start; |
4935 | u64 block_end; | |
39279cc3 | 4936 | |
b03ebd99 NB |
4937 | if (IS_ALIGNED(offset, blocksize) && |
4938 | (!len || IS_ALIGNED(len, blocksize))) | |
39279cc3 | 4939 | goto out; |
9703fefe | 4940 | |
8b62f87b JB |
4941 | block_start = round_down(from, blocksize); |
4942 | block_end = block_start + blocksize - 1; | |
4943 | ||
364ecf36 | 4944 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8b62f87b | 4945 | block_start, blocksize); |
5d5e103a JB |
4946 | if (ret) |
4947 | goto out; | |
39279cc3 | 4948 | |
211c17f5 | 4949 | again: |
3b16a4e3 | 4950 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4951 | if (!page) { |
bc42bda2 | 4952 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
4953 | block_start, blocksize, true); |
4954 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true); | |
ac6a2b36 | 4955 | ret = -ENOMEM; |
39279cc3 | 4956 | goto out; |
5d5e103a | 4957 | } |
e6dcd2dc | 4958 | |
39279cc3 | 4959 | if (!PageUptodate(page)) { |
9ebefb18 | 4960 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4961 | lock_page(page); |
211c17f5 CM |
4962 | if (page->mapping != mapping) { |
4963 | unlock_page(page); | |
09cbfeaf | 4964 | put_page(page); |
211c17f5 CM |
4965 | goto again; |
4966 | } | |
39279cc3 CM |
4967 | if (!PageUptodate(page)) { |
4968 | ret = -EIO; | |
89642229 | 4969 | goto out_unlock; |
39279cc3 CM |
4970 | } |
4971 | } | |
211c17f5 | 4972 | wait_on_page_writeback(page); |
e6dcd2dc | 4973 | |
9703fefe | 4974 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4975 | set_page_extent_mapped(page); |
4976 | ||
9703fefe | 4977 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4978 | if (ordered) { |
9703fefe | 4979 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4980 | &cached_state); |
e6dcd2dc | 4981 | unlock_page(page); |
09cbfeaf | 4982 | put_page(page); |
eb84ae03 | 4983 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4984 | btrfs_put_ordered_extent(ordered); |
4985 | goto again; | |
4986 | } | |
4987 | ||
9703fefe | 4988 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
4989 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4990 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 4991 | 0, 0, &cached_state); |
5d5e103a | 4992 | |
e3b8a485 | 4993 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
330a5827 | 4994 | &cached_state); |
9ed74f2d | 4995 | if (ret) { |
9703fefe | 4996 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4997 | &cached_state); |
9ed74f2d JB |
4998 | goto out_unlock; |
4999 | } | |
5000 | ||
9703fefe | 5001 | if (offset != blocksize) { |
2aaa6655 | 5002 | if (!len) |
9703fefe | 5003 | len = blocksize - offset; |
e6dcd2dc | 5004 | kaddr = kmap(page); |
2aaa6655 | 5005 | if (front) |
9703fefe CR |
5006 | memset(kaddr + (block_start - page_offset(page)), |
5007 | 0, offset); | |
2aaa6655 | 5008 | else |
9703fefe CR |
5009 | memset(kaddr + (block_start - page_offset(page)) + offset, |
5010 | 0, len); | |
e6dcd2dc CM |
5011 | flush_dcache_page(page); |
5012 | kunmap(page); | |
5013 | } | |
247e743c | 5014 | ClearPageChecked(page); |
e6dcd2dc | 5015 | set_page_dirty(page); |
e43bbe5e | 5016 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state); |
39279cc3 | 5017 | |
89642229 | 5018 | out_unlock: |
5d5e103a | 5019 | if (ret) |
bc42bda2 | 5020 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
43b18595 QW |
5021 | blocksize, true); |
5022 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0)); | |
39279cc3 | 5023 | unlock_page(page); |
09cbfeaf | 5024 | put_page(page); |
39279cc3 | 5025 | out: |
364ecf36 | 5026 | extent_changeset_free(data_reserved); |
39279cc3 CM |
5027 | return ret; |
5028 | } | |
5029 | ||
16e7549f JB |
5030 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
5031 | u64 offset, u64 len) | |
5032 | { | |
0b246afa | 5033 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
16e7549f JB |
5034 | struct btrfs_trans_handle *trans; |
5035 | int ret; | |
5036 | ||
5037 | /* | |
5038 | * Still need to make sure the inode looks like it's been updated so | |
5039 | * that any holes get logged if we fsync. | |
5040 | */ | |
0b246afa JM |
5041 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) { |
5042 | BTRFS_I(inode)->last_trans = fs_info->generation; | |
16e7549f JB |
5043 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
5044 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
5045 | return 0; | |
5046 | } | |
5047 | ||
5048 | /* | |
5049 | * 1 - for the one we're dropping | |
5050 | * 1 - for the one we're adding | |
5051 | * 1 - for updating the inode. | |
5052 | */ | |
5053 | trans = btrfs_start_transaction(root, 3); | |
5054 | if (IS_ERR(trans)) | |
5055 | return PTR_ERR(trans); | |
5056 | ||
5057 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
5058 | if (ret) { | |
66642832 | 5059 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 5060 | btrfs_end_transaction(trans); |
16e7549f JB |
5061 | return ret; |
5062 | } | |
5063 | ||
f85b7379 DS |
5064 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)), |
5065 | offset, 0, 0, len, 0, len, 0, 0, 0); | |
16e7549f | 5066 | if (ret) |
66642832 | 5067 | btrfs_abort_transaction(trans, ret); |
16e7549f JB |
5068 | else |
5069 | btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 5070 | btrfs_end_transaction(trans); |
16e7549f JB |
5071 | return ret; |
5072 | } | |
5073 | ||
695a0d0d JB |
5074 | /* |
5075 | * This function puts in dummy file extents for the area we're creating a hole | |
5076 | * for. So if we are truncating this file to a larger size we need to insert | |
5077 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
5078 | * the range between oldsize and size | |
5079 | */ | |
a41ad394 | 5080 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 5081 | { |
0b246afa | 5082 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9036c102 YZ |
5083 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5084 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 5085 | struct extent_map *em = NULL; |
2ac55d41 | 5086 | struct extent_state *cached_state = NULL; |
5dc562c5 | 5087 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
0b246afa JM |
5088 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
5089 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
5090 | u64 last_byte; |
5091 | u64 cur_offset; | |
5092 | u64 hole_size; | |
9ed74f2d | 5093 | int err = 0; |
39279cc3 | 5094 | |
a71754fc | 5095 | /* |
9703fefe CR |
5096 | * If our size started in the middle of a block we need to zero out the |
5097 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
5098 | * expose stale data. |
5099 | */ | |
9703fefe | 5100 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
5101 | if (err) |
5102 | return err; | |
5103 | ||
9036c102 YZ |
5104 | if (size <= hole_start) |
5105 | return 0; | |
5106 | ||
23d31bd4 NB |
5107 | btrfs_lock_and_flush_ordered_range(io_tree, BTRFS_I(inode), hole_start, |
5108 | block_end - 1, &cached_state); | |
9036c102 YZ |
5109 | cur_offset = hole_start; |
5110 | while (1) { | |
fc4f21b1 | 5111 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
9036c102 | 5112 | block_end - cur_offset, 0); |
79787eaa JM |
5113 | if (IS_ERR(em)) { |
5114 | err = PTR_ERR(em); | |
f2767956 | 5115 | em = NULL; |
79787eaa JM |
5116 | break; |
5117 | } | |
9036c102 | 5118 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 5119 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
8082510e | 5120 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 5121 | struct extent_map *hole_em; |
9036c102 | 5122 | hole_size = last_byte - cur_offset; |
9ed74f2d | 5123 | |
16e7549f JB |
5124 | err = maybe_insert_hole(root, inode, cur_offset, |
5125 | hole_size); | |
5126 | if (err) | |
3893e33b | 5127 | break; |
dcdbc059 | 5128 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
5129 | cur_offset + hole_size - 1, 0); |
5130 | hole_em = alloc_extent_map(); | |
5131 | if (!hole_em) { | |
5132 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
5133 | &BTRFS_I(inode)->runtime_flags); | |
5134 | goto next; | |
5135 | } | |
5136 | hole_em->start = cur_offset; | |
5137 | hole_em->len = hole_size; | |
5138 | hole_em->orig_start = cur_offset; | |
8082510e | 5139 | |
5dc562c5 JB |
5140 | hole_em->block_start = EXTENT_MAP_HOLE; |
5141 | hole_em->block_len = 0; | |
b4939680 | 5142 | hole_em->orig_block_len = 0; |
cc95bef6 | 5143 | hole_em->ram_bytes = hole_size; |
0b246afa | 5144 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 | 5145 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 5146 | hole_em->generation = fs_info->generation; |
8082510e | 5147 | |
5dc562c5 JB |
5148 | while (1) { |
5149 | write_lock(&em_tree->lock); | |
09a2a8f9 | 5150 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
5151 | write_unlock(&em_tree->lock); |
5152 | if (err != -EEXIST) | |
5153 | break; | |
dcdbc059 NB |
5154 | btrfs_drop_extent_cache(BTRFS_I(inode), |
5155 | cur_offset, | |
5dc562c5 JB |
5156 | cur_offset + |
5157 | hole_size - 1, 0); | |
5158 | } | |
5159 | free_extent_map(hole_em); | |
9036c102 | 5160 | } |
16e7549f | 5161 | next: |
9036c102 | 5162 | free_extent_map(em); |
a22285a6 | 5163 | em = NULL; |
9036c102 | 5164 | cur_offset = last_byte; |
8082510e | 5165 | if (cur_offset >= block_end) |
9036c102 YZ |
5166 | break; |
5167 | } | |
a22285a6 | 5168 | free_extent_map(em); |
e43bbe5e | 5169 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state); |
9036c102 YZ |
5170 | return err; |
5171 | } | |
39279cc3 | 5172 | |
3972f260 | 5173 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 5174 | { |
f4a2f4c5 MX |
5175 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5176 | struct btrfs_trans_handle *trans; | |
a41ad394 | 5177 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
5178 | loff_t newsize = attr->ia_size; |
5179 | int mask = attr->ia_valid; | |
8082510e YZ |
5180 | int ret; |
5181 | ||
3972f260 ES |
5182 | /* |
5183 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
5184 | * special case where we need to update the times despite not having | |
5185 | * these flags set. For all other operations the VFS set these flags | |
5186 | * explicitly if it wants a timestamp update. | |
5187 | */ | |
dff6efc3 CH |
5188 | if (newsize != oldsize) { |
5189 | inode_inc_iversion(inode); | |
5190 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
5191 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 5192 | current_time(inode); |
dff6efc3 | 5193 | } |
3972f260 | 5194 | |
a41ad394 | 5195 | if (newsize > oldsize) { |
9ea24bbe | 5196 | /* |
ea14b57f | 5197 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
5198 | * This is to ensure the snapshot captures a fully consistent |
5199 | * state of this file - if the snapshot captures this expanding | |
5200 | * truncation, it must capture all writes that happened before | |
5201 | * this truncation. | |
5202 | */ | |
0bc19f90 | 5203 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 5204 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe | 5205 | if (ret) { |
ea14b57f | 5206 | btrfs_end_write_no_snapshotting(root); |
8082510e | 5207 | return ret; |
9ea24bbe | 5208 | } |
8082510e | 5209 | |
f4a2f4c5 | 5210 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 5211 | if (IS_ERR(trans)) { |
ea14b57f | 5212 | btrfs_end_write_no_snapshotting(root); |
f4a2f4c5 | 5213 | return PTR_ERR(trans); |
9ea24bbe | 5214 | } |
f4a2f4c5 MX |
5215 | |
5216 | i_size_write(inode, newsize); | |
5217 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 5218 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 5219 | ret = btrfs_update_inode(trans, root, inode); |
ea14b57f | 5220 | btrfs_end_write_no_snapshotting(root); |
3a45bb20 | 5221 | btrfs_end_transaction(trans); |
a41ad394 | 5222 | } else { |
8082510e | 5223 | |
a41ad394 JB |
5224 | /* |
5225 | * We're truncating a file that used to have good data down to | |
5226 | * zero. Make sure it gets into the ordered flush list so that | |
5227 | * any new writes get down to disk quickly. | |
5228 | */ | |
5229 | if (newsize == 0) | |
72ac3c0d JB |
5230 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
5231 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 5232 | |
a41ad394 | 5233 | truncate_setsize(inode, newsize); |
2e60a51e | 5234 | |
52042d8e | 5235 | /* Disable nonlocked read DIO to avoid the endless truncate */ |
abcefb1e | 5236 | btrfs_inode_block_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5237 | inode_dio_wait(inode); |
0b581701 | 5238 | btrfs_inode_resume_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5239 | |
213e8c55 | 5240 | ret = btrfs_truncate(inode, newsize == oldsize); |
7f4f6e0a JB |
5241 | if (ret && inode->i_nlink) { |
5242 | int err; | |
5243 | ||
5244 | /* | |
f7e9e8fc OS |
5245 | * Truncate failed, so fix up the in-memory size. We |
5246 | * adjusted disk_i_size down as we removed extents, so | |
5247 | * wait for disk_i_size to be stable and then update the | |
5248 | * in-memory size to match. | |
7f4f6e0a | 5249 | */ |
f7e9e8fc | 5250 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
7f4f6e0a | 5251 | if (err) |
f7e9e8fc OS |
5252 | return err; |
5253 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
7f4f6e0a | 5254 | } |
8082510e YZ |
5255 | } |
5256 | ||
a41ad394 | 5257 | return ret; |
8082510e YZ |
5258 | } |
5259 | ||
9036c102 YZ |
5260 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5261 | { | |
2b0143b5 | 5262 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5263 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5264 | int err; |
39279cc3 | 5265 | |
b83cc969 LZ |
5266 | if (btrfs_root_readonly(root)) |
5267 | return -EROFS; | |
5268 | ||
31051c85 | 5269 | err = setattr_prepare(dentry, attr); |
9036c102 YZ |
5270 | if (err) |
5271 | return err; | |
2bf5a725 | 5272 | |
5a3f23d5 | 5273 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5274 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5275 | if (err) |
5276 | return err; | |
39279cc3 | 5277 | } |
9036c102 | 5278 | |
1025774c CH |
5279 | if (attr->ia_valid) { |
5280 | setattr_copy(inode, attr); | |
0c4d2d95 | 5281 | inode_inc_iversion(inode); |
22c44fe6 | 5282 | err = btrfs_dirty_inode(inode); |
1025774c | 5283 | |
22c44fe6 | 5284 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5285 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5286 | } |
33268eaf | 5287 | |
39279cc3 CM |
5288 | return err; |
5289 | } | |
61295eb8 | 5290 | |
131e404a FDBM |
5291 | /* |
5292 | * While truncating the inode pages during eviction, we get the VFS calling | |
5293 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5294 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5295 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5296 | * extent_state structures over and over, wasting lots of time. | |
5297 | * | |
5298 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5299 | * those expensive operations on a per page basis and do only the ordered io | |
5300 | * finishing, while we release here the extent_map and extent_state structures, | |
5301 | * without the excessive merging and splitting. | |
5302 | */ | |
5303 | static void evict_inode_truncate_pages(struct inode *inode) | |
5304 | { | |
5305 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5306 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5307 | struct rb_node *node; | |
5308 | ||
5309 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5310 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5311 | |
5312 | write_lock(&map_tree->lock); | |
07e1ce09 | 5313 | while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) { |
131e404a FDBM |
5314 | struct extent_map *em; |
5315 | ||
07e1ce09 | 5316 | node = rb_first_cached(&map_tree->map); |
131e404a | 5317 | em = rb_entry(node, struct extent_map, rb_node); |
180589ef WS |
5318 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5319 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5320 | remove_extent_mapping(map_tree, em); |
5321 | free_extent_map(em); | |
7064dd5c FM |
5322 | if (need_resched()) { |
5323 | write_unlock(&map_tree->lock); | |
5324 | cond_resched(); | |
5325 | write_lock(&map_tree->lock); | |
5326 | } | |
131e404a FDBM |
5327 | } |
5328 | write_unlock(&map_tree->lock); | |
5329 | ||
6ca07097 FM |
5330 | /* |
5331 | * Keep looping until we have no more ranges in the io tree. | |
5332 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5333 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5334 | * still in progress (unlocked the pages in the bio but did not yet | |
5335 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5336 | * ranges can still be locked and eviction started because before |
5337 | * submitting those bios, which are executed by a separate task (work | |
5338 | * queue kthread), inode references (inode->i_count) were not taken | |
5339 | * (which would be dropped in the end io callback of each bio). | |
5340 | * Therefore here we effectively end up waiting for those bios and | |
5341 | * anyone else holding locked ranges without having bumped the inode's | |
5342 | * reference count - if we don't do it, when they access the inode's | |
5343 | * io_tree to unlock a range it may be too late, leading to an | |
5344 | * use-after-free issue. | |
5345 | */ | |
131e404a FDBM |
5346 | spin_lock(&io_tree->lock); |
5347 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5348 | struct extent_state *state; | |
5349 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5350 | u64 start; |
5351 | u64 end; | |
421f0922 | 5352 | unsigned state_flags; |
131e404a FDBM |
5353 | |
5354 | node = rb_first(&io_tree->state); | |
5355 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5356 | start = state->start; |
5357 | end = state->end; | |
421f0922 | 5358 | state_flags = state->state; |
131e404a FDBM |
5359 | spin_unlock(&io_tree->lock); |
5360 | ||
ff13db41 | 5361 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5362 | |
5363 | /* | |
5364 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5365 | * and its reserved space won't be freed by delayed_ref. | |
5366 | * So we need to free its reserved space here. | |
5367 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5368 | * | |
5369 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5370 | */ | |
421f0922 | 5371 | if (state_flags & EXTENT_DELALLOC) |
bc42bda2 | 5372 | btrfs_qgroup_free_data(inode, NULL, start, end - start + 1); |
b9d0b389 | 5373 | |
6ca07097 | 5374 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5375 | EXTENT_LOCKED | EXTENT_DIRTY | |
5376 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
ae0f1625 | 5377 | EXTENT_DEFRAG, 1, 1, &cached_state); |
131e404a | 5378 | |
7064dd5c | 5379 | cond_resched(); |
131e404a FDBM |
5380 | spin_lock(&io_tree->lock); |
5381 | } | |
5382 | spin_unlock(&io_tree->lock); | |
5383 | } | |
5384 | ||
4b9d7b59 | 5385 | static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root, |
ad80cf50 | 5386 | struct btrfs_block_rsv *rsv) |
4b9d7b59 OS |
5387 | { |
5388 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5389 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d3984c90 | 5390 | struct btrfs_trans_handle *trans; |
2bd36e7b | 5391 | u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1); |
d3984c90 | 5392 | int ret; |
4b9d7b59 | 5393 | |
d3984c90 JB |
5394 | /* |
5395 | * Eviction should be taking place at some place safe because of our | |
5396 | * delayed iputs. However the normal flushing code will run delayed | |
5397 | * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock. | |
5398 | * | |
5399 | * We reserve the delayed_refs_extra here again because we can't use | |
5400 | * btrfs_start_transaction(root, 0) for the same deadlocky reason as | |
5401 | * above. We reserve our extra bit here because we generate a ton of | |
5402 | * delayed refs activity by truncating. | |
5403 | * | |
5404 | * If we cannot make our reservation we'll attempt to steal from the | |
5405 | * global reserve, because we really want to be able to free up space. | |
5406 | */ | |
5407 | ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra, | |
5408 | BTRFS_RESERVE_FLUSH_EVICT); | |
5409 | if (ret) { | |
4b9d7b59 OS |
5410 | /* |
5411 | * Try to steal from the global reserve if there is space for | |
5412 | * it. | |
5413 | */ | |
d3984c90 JB |
5414 | if (btrfs_check_space_for_delayed_refs(fs_info) || |
5415 | btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) { | |
5416 | btrfs_warn(fs_info, | |
5417 | "could not allocate space for delete; will truncate on mount"); | |
5418 | return ERR_PTR(-ENOSPC); | |
5419 | } | |
5420 | delayed_refs_extra = 0; | |
5421 | } | |
4b9d7b59 | 5422 | |
d3984c90 JB |
5423 | trans = btrfs_join_transaction(root); |
5424 | if (IS_ERR(trans)) | |
5425 | return trans; | |
5426 | ||
5427 | if (delayed_refs_extra) { | |
5428 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5429 | trans->bytes_reserved = delayed_refs_extra; | |
5430 | btrfs_block_rsv_migrate(rsv, trans->block_rsv, | |
5431 | delayed_refs_extra, 1); | |
4b9d7b59 | 5432 | } |
d3984c90 | 5433 | return trans; |
4b9d7b59 OS |
5434 | } |
5435 | ||
bd555975 | 5436 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5437 | { |
0b246afa | 5438 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5439 | struct btrfs_trans_handle *trans; |
5440 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b9d7b59 | 5441 | struct btrfs_block_rsv *rsv; |
39279cc3 CM |
5442 | int ret; |
5443 | ||
1abe9b8a | 5444 | trace_btrfs_inode_evict(inode); |
5445 | ||
3d48d981 | 5446 | if (!root) { |
e8f1bc14 | 5447 | clear_inode(inode); |
3d48d981 NB |
5448 | return; |
5449 | } | |
5450 | ||
131e404a FDBM |
5451 | evict_inode_truncate_pages(inode); |
5452 | ||
69e9c6c6 SB |
5453 | if (inode->i_nlink && |
5454 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5455 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5456 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5457 | goto no_delete; |
5458 | ||
27919067 | 5459 | if (is_bad_inode(inode)) |
39279cc3 | 5460 | goto no_delete; |
5f39d397 | 5461 | |
7ab7956e | 5462 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5463 | |
7b40b695 | 5464 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
c71bf099 | 5465 | goto no_delete; |
c71bf099 | 5466 | |
76dda93c | 5467 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5468 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5469 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5470 | goto no_delete; |
5471 | } | |
5472 | ||
aa79021f | 5473 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
27919067 | 5474 | if (ret) |
0e8c36a9 | 5475 | goto no_delete; |
0e8c36a9 | 5476 | |
2ff7e61e | 5477 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
27919067 | 5478 | if (!rsv) |
4289a667 | 5479 | goto no_delete; |
2bd36e7b | 5480 | rsv->size = btrfs_calc_metadata_size(fs_info, 1); |
ca7e70f5 | 5481 | rsv->failfast = 1; |
4289a667 | 5482 | |
6ef06d27 | 5483 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5484 | |
8082510e | 5485 | while (1) { |
ad80cf50 | 5486 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5487 | if (IS_ERR(trans)) |
5488 | goto free_rsv; | |
7b128766 | 5489 | |
4289a667 JB |
5490 | trans->block_rsv = rsv; |
5491 | ||
d68fc57b | 5492 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
27919067 OS |
5493 | trans->block_rsv = &fs_info->trans_block_rsv; |
5494 | btrfs_end_transaction(trans); | |
5495 | btrfs_btree_balance_dirty(fs_info); | |
5496 | if (ret && ret != -ENOSPC && ret != -EAGAIN) | |
5497 | goto free_rsv; | |
5498 | else if (!ret) | |
8082510e | 5499 | break; |
8082510e | 5500 | } |
5f39d397 | 5501 | |
4ef31a45 | 5502 | /* |
27919067 OS |
5503 | * Errors here aren't a big deal, it just means we leave orphan items in |
5504 | * the tree. They will be cleaned up on the next mount. If the inode | |
5505 | * number gets reused, cleanup deletes the orphan item without doing | |
5506 | * anything, and unlink reuses the existing orphan item. | |
5507 | * | |
5508 | * If it turns out that we are dropping too many of these, we might want | |
5509 | * to add a mechanism for retrying these after a commit. | |
4ef31a45 | 5510 | */ |
ad80cf50 | 5511 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5512 | if (!IS_ERR(trans)) { |
5513 | trans->block_rsv = rsv; | |
5514 | btrfs_orphan_del(trans, BTRFS_I(inode)); | |
5515 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5516 | btrfs_end_transaction(trans); | |
5517 | } | |
54aa1f4d | 5518 | |
0b246afa | 5519 | if (!(root == fs_info->tree_root || |
581bb050 | 5520 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) |
4a0cc7ca | 5521 | btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode))); |
581bb050 | 5522 | |
27919067 OS |
5523 | free_rsv: |
5524 | btrfs_free_block_rsv(fs_info, rsv); | |
39279cc3 | 5525 | no_delete: |
27919067 OS |
5526 | /* |
5527 | * If we didn't successfully delete, the orphan item will still be in | |
5528 | * the tree and we'll retry on the next mount. Again, we might also want | |
5529 | * to retry these periodically in the future. | |
5530 | */ | |
f48d1cf5 | 5531 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5532 | clear_inode(inode); |
39279cc3 CM |
5533 | } |
5534 | ||
5535 | /* | |
6bf9e4bd QW |
5536 | * Return the key found in the dir entry in the location pointer, fill @type |
5537 | * with BTRFS_FT_*, and return 0. | |
5538 | * | |
005d6712 SY |
5539 | * If no dir entries were found, returns -ENOENT. |
5540 | * If found a corrupted location in dir entry, returns -EUCLEAN. | |
39279cc3 CM |
5541 | */ |
5542 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
6bf9e4bd | 5543 | struct btrfs_key *location, u8 *type) |
39279cc3 CM |
5544 | { |
5545 | const char *name = dentry->d_name.name; | |
5546 | int namelen = dentry->d_name.len; | |
5547 | struct btrfs_dir_item *di; | |
5548 | struct btrfs_path *path; | |
5549 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5550 | int ret = 0; |
39279cc3 CM |
5551 | |
5552 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5553 | if (!path) |
5554 | return -ENOMEM; | |
3954401f | 5555 | |
f85b7379 DS |
5556 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5557 | name, namelen, 0); | |
3cf5068f LB |
5558 | if (IS_ERR_OR_NULL(di)) { |
5559 | ret = di ? PTR_ERR(di) : -ENOENT; | |
005d6712 SY |
5560 | goto out; |
5561 | } | |
d397712b | 5562 | |
5f39d397 | 5563 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5564 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5565 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
005d6712 | 5566 | ret = -EUCLEAN; |
56a0e706 LB |
5567 | btrfs_warn(root->fs_info, |
5568 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5569 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5570 | location->objectid, location->type, location->offset); | |
56a0e706 | 5571 | } |
6bf9e4bd QW |
5572 | if (!ret) |
5573 | *type = btrfs_dir_type(path->nodes[0], di); | |
39279cc3 | 5574 | out: |
39279cc3 CM |
5575 | btrfs_free_path(path); |
5576 | return ret; | |
5577 | } | |
5578 | ||
5579 | /* | |
5580 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5581 | * needs to be changed to reflect the root directory of the tree root. This | |
5582 | * is kind of like crossing a mount point. | |
5583 | */ | |
2ff7e61e | 5584 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5585 | struct inode *dir, |
5586 | struct dentry *dentry, | |
5587 | struct btrfs_key *location, | |
5588 | struct btrfs_root **sub_root) | |
39279cc3 | 5589 | { |
4df27c4d YZ |
5590 | struct btrfs_path *path; |
5591 | struct btrfs_root *new_root; | |
5592 | struct btrfs_root_ref *ref; | |
5593 | struct extent_buffer *leaf; | |
1d4c08e0 | 5594 | struct btrfs_key key; |
4df27c4d YZ |
5595 | int ret; |
5596 | int err = 0; | |
39279cc3 | 5597 | |
4df27c4d YZ |
5598 | path = btrfs_alloc_path(); |
5599 | if (!path) { | |
5600 | err = -ENOMEM; | |
5601 | goto out; | |
5602 | } | |
39279cc3 | 5603 | |
4df27c4d | 5604 | err = -ENOENT; |
1d4c08e0 DS |
5605 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5606 | key.type = BTRFS_ROOT_REF_KEY; | |
5607 | key.offset = location->objectid; | |
5608 | ||
0b246afa | 5609 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5610 | if (ret) { |
5611 | if (ret < 0) | |
5612 | err = ret; | |
5613 | goto out; | |
5614 | } | |
39279cc3 | 5615 | |
4df27c4d YZ |
5616 | leaf = path->nodes[0]; |
5617 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5618 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5619 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5620 | goto out; | |
39279cc3 | 5621 | |
4df27c4d YZ |
5622 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5623 | (unsigned long)(ref + 1), | |
5624 | dentry->d_name.len); | |
5625 | if (ret) | |
5626 | goto out; | |
5627 | ||
b3b4aa74 | 5628 | btrfs_release_path(path); |
4df27c4d | 5629 | |
0b246afa | 5630 | new_root = btrfs_read_fs_root_no_name(fs_info, location); |
4df27c4d YZ |
5631 | if (IS_ERR(new_root)) { |
5632 | err = PTR_ERR(new_root); | |
5633 | goto out; | |
5634 | } | |
5635 | ||
4df27c4d YZ |
5636 | *sub_root = new_root; |
5637 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5638 | location->type = BTRFS_INODE_ITEM_KEY; | |
5639 | location->offset = 0; | |
5640 | err = 0; | |
5641 | out: | |
5642 | btrfs_free_path(path); | |
5643 | return err; | |
39279cc3 CM |
5644 | } |
5645 | ||
5d4f98a2 YZ |
5646 | static void inode_tree_add(struct inode *inode) |
5647 | { | |
5648 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5649 | struct btrfs_inode *entry; | |
03e860bd FNP |
5650 | struct rb_node **p; |
5651 | struct rb_node *parent; | |
cef21937 | 5652 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5653 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5654 | |
1d3382cb | 5655 | if (inode_unhashed(inode)) |
76dda93c | 5656 | return; |
e1409cef | 5657 | parent = NULL; |
5d4f98a2 | 5658 | spin_lock(&root->inode_lock); |
e1409cef | 5659 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5660 | while (*p) { |
5661 | parent = *p; | |
5662 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5663 | ||
37508515 | 5664 | if (ino < btrfs_ino(entry)) |
03e860bd | 5665 | p = &parent->rb_left; |
37508515 | 5666 | else if (ino > btrfs_ino(entry)) |
03e860bd | 5667 | p = &parent->rb_right; |
5d4f98a2 YZ |
5668 | else { |
5669 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5670 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5671 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5672 | RB_CLEAR_NODE(parent); |
5673 | spin_unlock(&root->inode_lock); | |
cef21937 | 5674 | return; |
5d4f98a2 YZ |
5675 | } |
5676 | } | |
cef21937 FDBM |
5677 | rb_link_node(new, parent, p); |
5678 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5679 | spin_unlock(&root->inode_lock); |
5680 | } | |
5681 | ||
5682 | static void inode_tree_del(struct inode *inode) | |
5683 | { | |
0b246afa | 5684 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5d4f98a2 | 5685 | struct btrfs_root *root = BTRFS_I(inode)->root; |
76dda93c | 5686 | int empty = 0; |
5d4f98a2 | 5687 | |
03e860bd | 5688 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5689 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5690 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5691 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5692 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5693 | } |
03e860bd | 5694 | spin_unlock(&root->inode_lock); |
76dda93c | 5695 | |
69e9c6c6 | 5696 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
0b246afa | 5697 | synchronize_srcu(&fs_info->subvol_srcu); |
76dda93c YZ |
5698 | spin_lock(&root->inode_lock); |
5699 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5700 | spin_unlock(&root->inode_lock); | |
5701 | if (empty) | |
5702 | btrfs_add_dead_root(root); | |
5703 | } | |
5704 | } | |
5705 | ||
5d4f98a2 | 5706 | |
e02119d5 CM |
5707 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5708 | { | |
5709 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5710 | inode->i_ino = args->location->objectid; |
5711 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5712 | sizeof(*args->location)); | |
e02119d5 | 5713 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5714 | return 0; |
5715 | } | |
5716 | ||
5717 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5718 | { | |
5719 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5720 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5721 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5722 | } |
5723 | ||
5d4f98a2 | 5724 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5725 | struct btrfs_key *location, |
5d4f98a2 | 5726 | struct btrfs_root *root) |
39279cc3 CM |
5727 | { |
5728 | struct inode *inode; | |
5729 | struct btrfs_iget_args args; | |
90d3e592 | 5730 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5731 | |
90d3e592 | 5732 | args.location = location; |
39279cc3 CM |
5733 | args.root = root; |
5734 | ||
778ba82b | 5735 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5736 | btrfs_init_locked_inode, |
5737 | (void *)&args); | |
5738 | return inode; | |
5739 | } | |
5740 | ||
1a54ef8c BR |
5741 | /* Get an inode object given its location and corresponding root. |
5742 | * Returns in *is_new if the inode was read from disk | |
5743 | */ | |
4222ea71 FM |
5744 | struct inode *btrfs_iget_path(struct super_block *s, struct btrfs_key *location, |
5745 | struct btrfs_root *root, int *new, | |
5746 | struct btrfs_path *path) | |
1a54ef8c BR |
5747 | { |
5748 | struct inode *inode; | |
5749 | ||
90d3e592 | 5750 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5751 | if (!inode) |
5d4f98a2 | 5752 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5753 | |
5754 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5755 | int ret; |
5756 | ||
4222ea71 | 5757 | ret = btrfs_read_locked_inode(inode, path); |
9bc2ceff | 5758 | if (!ret) { |
1748f843 MF |
5759 | inode_tree_add(inode); |
5760 | unlock_new_inode(inode); | |
5761 | if (new) | |
5762 | *new = 1; | |
5763 | } else { | |
f5b3a417 AV |
5764 | iget_failed(inode); |
5765 | /* | |
5766 | * ret > 0 can come from btrfs_search_slot called by | |
5767 | * btrfs_read_locked_inode, this means the inode item | |
5768 | * was not found. | |
5769 | */ | |
5770 | if (ret > 0) | |
5771 | ret = -ENOENT; | |
5772 | inode = ERR_PTR(ret); | |
1748f843 MF |
5773 | } |
5774 | } | |
5775 | ||
1a54ef8c BR |
5776 | return inode; |
5777 | } | |
5778 | ||
4222ea71 FM |
5779 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, |
5780 | struct btrfs_root *root, int *new) | |
5781 | { | |
5782 | return btrfs_iget_path(s, location, root, new, NULL); | |
5783 | } | |
5784 | ||
4df27c4d YZ |
5785 | static struct inode *new_simple_dir(struct super_block *s, |
5786 | struct btrfs_key *key, | |
5787 | struct btrfs_root *root) | |
5788 | { | |
5789 | struct inode *inode = new_inode(s); | |
5790 | ||
5791 | if (!inode) | |
5792 | return ERR_PTR(-ENOMEM); | |
5793 | ||
4df27c4d YZ |
5794 | BTRFS_I(inode)->root = root; |
5795 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5796 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5797 | |
5798 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5799 | inode->i_op = &btrfs_dir_ro_inode_operations; |
1fdf4194 | 5800 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5801 | inode->i_fop = &simple_dir_operations; |
5802 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5803 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5804 | inode->i_atime = inode->i_mtime; |
5805 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 5806 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
4df27c4d YZ |
5807 | |
5808 | return inode; | |
5809 | } | |
5810 | ||
6bf9e4bd QW |
5811 | static inline u8 btrfs_inode_type(struct inode *inode) |
5812 | { | |
5813 | /* | |
5814 | * Compile-time asserts that generic FT_* types still match | |
5815 | * BTRFS_FT_* types | |
5816 | */ | |
5817 | BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN); | |
5818 | BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE); | |
5819 | BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR); | |
5820 | BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV); | |
5821 | BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV); | |
5822 | BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO); | |
5823 | BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK); | |
5824 | BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK); | |
5825 | ||
5826 | return fs_umode_to_ftype(inode->i_mode); | |
5827 | } | |
5828 | ||
3de4586c | 5829 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5830 | { |
0b246afa | 5831 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5832 | struct inode *inode; |
4df27c4d | 5833 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5834 | struct btrfs_root *sub_root = root; |
5835 | struct btrfs_key location; | |
6bf9e4bd | 5836 | u8 di_type = 0; |
76dda93c | 5837 | int index; |
b4aff1f8 | 5838 | int ret = 0; |
39279cc3 CM |
5839 | |
5840 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5841 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5842 | |
6bf9e4bd | 5843 | ret = btrfs_inode_by_name(dir, dentry, &location, &di_type); |
39279cc3 CM |
5844 | if (ret < 0) |
5845 | return ERR_PTR(ret); | |
5f39d397 | 5846 | |
4df27c4d | 5847 | if (location.type == BTRFS_INODE_ITEM_KEY) { |
73f73415 | 5848 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
6bf9e4bd QW |
5849 | if (IS_ERR(inode)) |
5850 | return inode; | |
5851 | ||
5852 | /* Do extra check against inode mode with di_type */ | |
5853 | if (btrfs_inode_type(inode) != di_type) { | |
5854 | btrfs_crit(fs_info, | |
5855 | "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u", | |
5856 | inode->i_mode, btrfs_inode_type(inode), | |
5857 | di_type); | |
5858 | iput(inode); | |
5859 | return ERR_PTR(-EUCLEAN); | |
5860 | } | |
4df27c4d YZ |
5861 | return inode; |
5862 | } | |
5863 | ||
0b246afa | 5864 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2ff7e61e | 5865 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5866 | &location, &sub_root); |
5867 | if (ret < 0) { | |
5868 | if (ret != -ENOENT) | |
5869 | inode = ERR_PTR(ret); | |
5870 | else | |
5871 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5872 | } else { | |
73f73415 | 5873 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5874 | } |
0b246afa | 5875 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
76dda93c | 5876 | |
34d19bad | 5877 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5878 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5879 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5880 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5881 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5882 | if (ret) { |
5883 | iput(inode); | |
66b4ffd1 | 5884 | inode = ERR_PTR(ret); |
01cd3367 | 5885 | } |
c71bf099 YZ |
5886 | } |
5887 | ||
3de4586c CM |
5888 | return inode; |
5889 | } | |
5890 | ||
fe15ce44 | 5891 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5892 | { |
5893 | struct btrfs_root *root; | |
2b0143b5 | 5894 | struct inode *inode = d_inode(dentry); |
76dda93c | 5895 | |
848cce0d | 5896 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5897 | inode = d_inode(dentry->d_parent); |
76dda93c | 5898 | |
848cce0d LZ |
5899 | if (inode) { |
5900 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5901 | if (btrfs_root_refs(&root->root_item) == 0) |
5902 | return 1; | |
848cce0d | 5903 | |
4a0cc7ca | 5904 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5905 | return 1; |
efefb143 | 5906 | } |
76dda93c YZ |
5907 | return 0; |
5908 | } | |
5909 | ||
3de4586c | 5910 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5911 | unsigned int flags) |
3de4586c | 5912 | { |
3837d208 | 5913 | struct inode *inode = btrfs_lookup_dentry(dir, dentry); |
5662344b | 5914 | |
3837d208 AV |
5915 | if (inode == ERR_PTR(-ENOENT)) |
5916 | inode = NULL; | |
41d28bca | 5917 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5918 | } |
5919 | ||
23b5ec74 JB |
5920 | /* |
5921 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5922 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5923 | * our information into that, and then dir_emit from the buffer. This is | |
5924 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5925 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5926 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5927 | * tree lock. | |
5928 | */ | |
5929 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5930 | { | |
5931 | struct btrfs_file_private *private; | |
5932 | ||
5933 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5934 | if (!private) | |
5935 | return -ENOMEM; | |
5936 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
5937 | if (!private->filldir_buf) { | |
5938 | kfree(private); | |
5939 | return -ENOMEM; | |
5940 | } | |
5941 | file->private_data = private; | |
5942 | return 0; | |
5943 | } | |
5944 | ||
5945 | struct dir_entry { | |
5946 | u64 ino; | |
5947 | u64 offset; | |
5948 | unsigned type; | |
5949 | int name_len; | |
5950 | }; | |
5951 | ||
5952 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5953 | { | |
5954 | while (entries--) { | |
5955 | struct dir_entry *entry = addr; | |
5956 | char *name = (char *)(entry + 1); | |
5957 | ||
92d32170 DS |
5958 | ctx->pos = get_unaligned(&entry->offset); |
5959 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
5960 | get_unaligned(&entry->ino), | |
5961 | get_unaligned(&entry->type))) | |
23b5ec74 | 5962 | return 1; |
92d32170 DS |
5963 | addr += sizeof(struct dir_entry) + |
5964 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
5965 | ctx->pos++; |
5966 | } | |
5967 | return 0; | |
5968 | } | |
5969 | ||
9cdda8d3 | 5970 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5971 | { |
9cdda8d3 | 5972 | struct inode *inode = file_inode(file); |
39279cc3 | 5973 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 5974 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
5975 | struct btrfs_dir_item *di; |
5976 | struct btrfs_key key; | |
5f39d397 | 5977 | struct btrfs_key found_key; |
39279cc3 | 5978 | struct btrfs_path *path; |
23b5ec74 | 5979 | void *addr; |
16cdcec7 MX |
5980 | struct list_head ins_list; |
5981 | struct list_head del_list; | |
39279cc3 | 5982 | int ret; |
5f39d397 | 5983 | struct extent_buffer *leaf; |
39279cc3 | 5984 | int slot; |
5f39d397 CM |
5985 | char *name_ptr; |
5986 | int name_len; | |
23b5ec74 JB |
5987 | int entries = 0; |
5988 | int total_len = 0; | |
02dbfc99 | 5989 | bool put = false; |
c2951f32 | 5990 | struct btrfs_key location; |
5f39d397 | 5991 | |
9cdda8d3 AV |
5992 | if (!dir_emit_dots(file, ctx)) |
5993 | return 0; | |
5994 | ||
49593bfa | 5995 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5996 | if (!path) |
5997 | return -ENOMEM; | |
ff5714cc | 5998 | |
23b5ec74 | 5999 | addr = private->filldir_buf; |
e4058b54 | 6000 | path->reada = READA_FORWARD; |
49593bfa | 6001 | |
c2951f32 JM |
6002 | INIT_LIST_HEAD(&ins_list); |
6003 | INIT_LIST_HEAD(&del_list); | |
6004 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 6005 | |
23b5ec74 | 6006 | again: |
c2951f32 | 6007 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 6008 | key.offset = ctx->pos; |
4a0cc7ca | 6009 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 6010 | |
39279cc3 CM |
6011 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6012 | if (ret < 0) | |
6013 | goto err; | |
49593bfa DW |
6014 | |
6015 | while (1) { | |
23b5ec74 JB |
6016 | struct dir_entry *entry; |
6017 | ||
5f39d397 | 6018 | leaf = path->nodes[0]; |
39279cc3 | 6019 | slot = path->slots[0]; |
b9e03af0 LZ |
6020 | if (slot >= btrfs_header_nritems(leaf)) { |
6021 | ret = btrfs_next_leaf(root, path); | |
6022 | if (ret < 0) | |
6023 | goto err; | |
6024 | else if (ret > 0) | |
6025 | break; | |
6026 | continue; | |
39279cc3 | 6027 | } |
3de4586c | 6028 | |
5f39d397 CM |
6029 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
6030 | ||
6031 | if (found_key.objectid != key.objectid) | |
39279cc3 | 6032 | break; |
c2951f32 | 6033 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 6034 | break; |
9cdda8d3 | 6035 | if (found_key.offset < ctx->pos) |
b9e03af0 | 6036 | goto next; |
c2951f32 | 6037 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 6038 | goto next; |
39279cc3 | 6039 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
c2951f32 | 6040 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
6041 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
6042 | PAGE_SIZE) { | |
6043 | btrfs_release_path(path); | |
6044 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6045 | if (ret) | |
6046 | goto nopos; | |
6047 | addr = private->filldir_buf; | |
6048 | entries = 0; | |
6049 | total_len = 0; | |
6050 | goto again; | |
c2951f32 | 6051 | } |
23b5ec74 JB |
6052 | |
6053 | entry = addr; | |
92d32170 | 6054 | put_unaligned(name_len, &entry->name_len); |
23b5ec74 | 6055 | name_ptr = (char *)(entry + 1); |
c2951f32 JM |
6056 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
6057 | name_len); | |
7d157c3d | 6058 | put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)), |
92d32170 | 6059 | &entry->type); |
c2951f32 | 6060 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
92d32170 DS |
6061 | put_unaligned(location.objectid, &entry->ino); |
6062 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
6063 | entries++; |
6064 | addr += sizeof(struct dir_entry) + name_len; | |
6065 | total_len += sizeof(struct dir_entry) + name_len; | |
b9e03af0 LZ |
6066 | next: |
6067 | path->slots[0]++; | |
39279cc3 | 6068 | } |
23b5ec74 JB |
6069 | btrfs_release_path(path); |
6070 | ||
6071 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6072 | if (ret) | |
6073 | goto nopos; | |
49593bfa | 6074 | |
d2fbb2b5 | 6075 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 6076 | if (ret) |
bc4ef759 DS |
6077 | goto nopos; |
6078 | ||
db62efbb ZB |
6079 | /* |
6080 | * Stop new entries from being returned after we return the last | |
6081 | * entry. | |
6082 | * | |
6083 | * New directory entries are assigned a strictly increasing | |
6084 | * offset. This means that new entries created during readdir | |
6085 | * are *guaranteed* to be seen in the future by that readdir. | |
6086 | * This has broken buggy programs which operate on names as | |
6087 | * they're returned by readdir. Until we re-use freed offsets | |
6088 | * we have this hack to stop new entries from being returned | |
6089 | * under the assumption that they'll never reach this huge | |
6090 | * offset. | |
6091 | * | |
6092 | * This is being careful not to overflow 32bit loff_t unless the | |
6093 | * last entry requires it because doing so has broken 32bit apps | |
6094 | * in the past. | |
6095 | */ | |
c2951f32 JM |
6096 | if (ctx->pos >= INT_MAX) |
6097 | ctx->pos = LLONG_MAX; | |
6098 | else | |
6099 | ctx->pos = INT_MAX; | |
39279cc3 CM |
6100 | nopos: |
6101 | ret = 0; | |
6102 | err: | |
02dbfc99 OS |
6103 | if (put) |
6104 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 6105 | btrfs_free_path(path); |
39279cc3 CM |
6106 | return ret; |
6107 | } | |
6108 | ||
39279cc3 | 6109 | /* |
54aa1f4d | 6110 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
6111 | * inode changes. But, it is most likely to find the inode in cache. |
6112 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
6113 | * to keep or drop this code. | |
6114 | */ | |
48a3b636 | 6115 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 6116 | { |
2ff7e61e | 6117 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
6118 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6119 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
6120 | int ret; |
6121 | ||
72ac3c0d | 6122 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 6123 | return 0; |
39279cc3 | 6124 | |
7a7eaa40 | 6125 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
6126 | if (IS_ERR(trans)) |
6127 | return PTR_ERR(trans); | |
8929ecfa YZ |
6128 | |
6129 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
6130 | if (ret && ret == -ENOSPC) { |
6131 | /* whoops, lets try again with the full transaction */ | |
3a45bb20 | 6132 | btrfs_end_transaction(trans); |
94b60442 | 6133 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
6134 | if (IS_ERR(trans)) |
6135 | return PTR_ERR(trans); | |
8929ecfa | 6136 | |
94b60442 | 6137 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 6138 | } |
3a45bb20 | 6139 | btrfs_end_transaction(trans); |
16cdcec7 | 6140 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 6141 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6142 | |
6143 | return ret; | |
6144 | } | |
6145 | ||
6146 | /* | |
6147 | * This is a copy of file_update_time. We need this so we can return error on | |
6148 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6149 | */ | |
95582b00 | 6150 | static int btrfs_update_time(struct inode *inode, struct timespec64 *now, |
e41f941a | 6151 | int flags) |
22c44fe6 | 6152 | { |
2bc55652 | 6153 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3a8c7231 | 6154 | bool dirty = flags & ~S_VERSION; |
2bc55652 AB |
6155 | |
6156 | if (btrfs_root_readonly(root)) | |
6157 | return -EROFS; | |
6158 | ||
e41f941a | 6159 | if (flags & S_VERSION) |
3a8c7231 | 6160 | dirty |= inode_maybe_inc_iversion(inode, dirty); |
e41f941a JB |
6161 | if (flags & S_CTIME) |
6162 | inode->i_ctime = *now; | |
6163 | if (flags & S_MTIME) | |
6164 | inode->i_mtime = *now; | |
6165 | if (flags & S_ATIME) | |
6166 | inode->i_atime = *now; | |
3a8c7231 | 6167 | return dirty ? btrfs_dirty_inode(inode) : 0; |
39279cc3 CM |
6168 | } |
6169 | ||
d352ac68 CM |
6170 | /* |
6171 | * find the highest existing sequence number in a directory | |
6172 | * and then set the in-memory index_cnt variable to reflect | |
6173 | * free sequence numbers | |
6174 | */ | |
4c570655 | 6175 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 6176 | { |
4c570655 | 6177 | struct btrfs_root *root = inode->root; |
aec7477b JB |
6178 | struct btrfs_key key, found_key; |
6179 | struct btrfs_path *path; | |
6180 | struct extent_buffer *leaf; | |
6181 | int ret; | |
6182 | ||
4c570655 | 6183 | key.objectid = btrfs_ino(inode); |
962a298f | 6184 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6185 | key.offset = (u64)-1; |
6186 | ||
6187 | path = btrfs_alloc_path(); | |
6188 | if (!path) | |
6189 | return -ENOMEM; | |
6190 | ||
6191 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6192 | if (ret < 0) | |
6193 | goto out; | |
6194 | /* FIXME: we should be able to handle this */ | |
6195 | if (ret == 0) | |
6196 | goto out; | |
6197 | ret = 0; | |
6198 | ||
6199 | /* | |
6200 | * MAGIC NUMBER EXPLANATION: | |
6201 | * since we search a directory based on f_pos we have to start at 2 | |
6202 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6203 | * else has to start at 2 | |
6204 | */ | |
6205 | if (path->slots[0] == 0) { | |
4c570655 | 6206 | inode->index_cnt = 2; |
aec7477b JB |
6207 | goto out; |
6208 | } | |
6209 | ||
6210 | path->slots[0]--; | |
6211 | ||
6212 | leaf = path->nodes[0]; | |
6213 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6214 | ||
4c570655 | 6215 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6216 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
4c570655 | 6217 | inode->index_cnt = 2; |
aec7477b JB |
6218 | goto out; |
6219 | } | |
6220 | ||
4c570655 | 6221 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6222 | out: |
6223 | btrfs_free_path(path); | |
6224 | return ret; | |
6225 | } | |
6226 | ||
d352ac68 CM |
6227 | /* |
6228 | * helper to find a free sequence number in a given directory. This current | |
6229 | * code is very simple, later versions will do smarter things in the btree | |
6230 | */ | |
877574e2 | 6231 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6232 | { |
6233 | int ret = 0; | |
6234 | ||
877574e2 NB |
6235 | if (dir->index_cnt == (u64)-1) { |
6236 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6237 | if (ret) { |
6238 | ret = btrfs_set_inode_index_count(dir); | |
6239 | if (ret) | |
6240 | return ret; | |
6241 | } | |
aec7477b JB |
6242 | } |
6243 | ||
877574e2 NB |
6244 | *index = dir->index_cnt; |
6245 | dir->index_cnt++; | |
aec7477b JB |
6246 | |
6247 | return ret; | |
6248 | } | |
6249 | ||
b0d5d10f CM |
6250 | static int btrfs_insert_inode_locked(struct inode *inode) |
6251 | { | |
6252 | struct btrfs_iget_args args; | |
6253 | args.location = &BTRFS_I(inode)->location; | |
6254 | args.root = BTRFS_I(inode)->root; | |
6255 | ||
6256 | return insert_inode_locked4(inode, | |
6257 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6258 | btrfs_find_actor, &args); | |
6259 | } | |
6260 | ||
19aee8de AJ |
6261 | /* |
6262 | * Inherit flags from the parent inode. | |
6263 | * | |
6264 | * Currently only the compression flags and the cow flags are inherited. | |
6265 | */ | |
6266 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
6267 | { | |
6268 | unsigned int flags; | |
6269 | ||
6270 | if (!dir) | |
6271 | return; | |
6272 | ||
6273 | flags = BTRFS_I(dir)->flags; | |
6274 | ||
6275 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
6276 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
6277 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
6278 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
6279 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
6280 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
6281 | } | |
6282 | ||
6283 | if (flags & BTRFS_INODE_NODATACOW) { | |
6284 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
6285 | if (S_ISREG(inode->i_mode)) | |
6286 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6287 | } | |
6288 | ||
7b6a221e | 6289 | btrfs_sync_inode_flags_to_i_flags(inode); |
19aee8de AJ |
6290 | } |
6291 | ||
39279cc3 CM |
6292 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6293 | struct btrfs_root *root, | |
aec7477b | 6294 | struct inode *dir, |
9c58309d | 6295 | const char *name, int name_len, |
175a4eb7 AV |
6296 | u64 ref_objectid, u64 objectid, |
6297 | umode_t mode, u64 *index) | |
39279cc3 | 6298 | { |
0b246afa | 6299 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6300 | struct inode *inode; |
5f39d397 | 6301 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6302 | struct btrfs_key *location; |
5f39d397 | 6303 | struct btrfs_path *path; |
9c58309d CM |
6304 | struct btrfs_inode_ref *ref; |
6305 | struct btrfs_key key[2]; | |
6306 | u32 sizes[2]; | |
ef3b9af5 | 6307 | int nitems = name ? 2 : 1; |
9c58309d | 6308 | unsigned long ptr; |
39279cc3 | 6309 | int ret; |
39279cc3 | 6310 | |
5f39d397 | 6311 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6312 | if (!path) |
6313 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6314 | |
0b246afa | 6315 | inode = new_inode(fs_info->sb); |
8fb27640 YS |
6316 | if (!inode) { |
6317 | btrfs_free_path(path); | |
39279cc3 | 6318 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6319 | } |
39279cc3 | 6320 | |
5762b5c9 FM |
6321 | /* |
6322 | * O_TMPFILE, set link count to 0, so that after this point, | |
6323 | * we fill in an inode item with the correct link count. | |
6324 | */ | |
6325 | if (!name) | |
6326 | set_nlink(inode, 0); | |
6327 | ||
581bb050 LZ |
6328 | /* |
6329 | * we have to initialize this early, so we can reclaim the inode | |
6330 | * number if we fail afterwards in this function. | |
6331 | */ | |
6332 | inode->i_ino = objectid; | |
6333 | ||
ef3b9af5 | 6334 | if (dir && name) { |
1abe9b8a | 6335 | trace_btrfs_inode_request(dir); |
6336 | ||
877574e2 | 6337 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 6338 | if (ret) { |
8fb27640 | 6339 | btrfs_free_path(path); |
09771430 | 6340 | iput(inode); |
aec7477b | 6341 | return ERR_PTR(ret); |
09771430 | 6342 | } |
ef3b9af5 FM |
6343 | } else if (dir) { |
6344 | *index = 0; | |
aec7477b JB |
6345 | } |
6346 | /* | |
6347 | * index_cnt is ignored for everything but a dir, | |
df6703e1 | 6348 | * btrfs_set_inode_index_count has an explanation for the magic |
aec7477b JB |
6349 | * number |
6350 | */ | |
6351 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6352 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6353 | BTRFS_I(inode)->root = root; |
e02119d5 | 6354 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6355 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6356 | |
5dc562c5 JB |
6357 | /* |
6358 | * We could have gotten an inode number from somebody who was fsynced | |
6359 | * and then removed in this same transaction, so let's just set full | |
6360 | * sync since it will be a full sync anyway and this will blow away the | |
6361 | * old info in the log. | |
6362 | */ | |
6363 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6364 | ||
9c58309d | 6365 | key[0].objectid = objectid; |
962a298f | 6366 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6367 | key[0].offset = 0; |
6368 | ||
9c58309d | 6369 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6370 | |
6371 | if (name) { | |
6372 | /* | |
6373 | * Start new inodes with an inode_ref. This is slightly more | |
6374 | * efficient for small numbers of hard links since they will | |
6375 | * be packed into one item. Extended refs will kick in if we | |
6376 | * add more hard links than can fit in the ref item. | |
6377 | */ | |
6378 | key[1].objectid = objectid; | |
962a298f | 6379 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6380 | key[1].offset = ref_objectid; |
6381 | ||
6382 | sizes[1] = name_len + sizeof(*ref); | |
6383 | } | |
9c58309d | 6384 | |
b0d5d10f CM |
6385 | location = &BTRFS_I(inode)->location; |
6386 | location->objectid = objectid; | |
6387 | location->offset = 0; | |
962a298f | 6388 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6389 | |
6390 | ret = btrfs_insert_inode_locked(inode); | |
32955c54 AV |
6391 | if (ret < 0) { |
6392 | iput(inode); | |
b0d5d10f | 6393 | goto fail; |
32955c54 | 6394 | } |
b0d5d10f | 6395 | |
b9473439 | 6396 | path->leave_spinning = 1; |
ef3b9af5 | 6397 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6398 | if (ret != 0) |
b0d5d10f | 6399 | goto fail_unlock; |
5f39d397 | 6400 | |
ecc11fab | 6401 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6402 | inode_set_bytes(inode, 0); |
9cc97d64 | 6403 | |
c2050a45 | 6404 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6405 | inode->i_atime = inode->i_mtime; |
6406 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 6407 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
9cc97d64 | 6408 | |
5f39d397 CM |
6409 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6410 | struct btrfs_inode_item); | |
b159fa28 | 6411 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6412 | sizeof(*inode_item)); |
e02119d5 | 6413 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6414 | |
ef3b9af5 FM |
6415 | if (name) { |
6416 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6417 | struct btrfs_inode_ref); | |
6418 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6419 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6420 | ptr = (unsigned long)(ref + 1); | |
6421 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6422 | } | |
9c58309d | 6423 | |
5f39d397 CM |
6424 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6425 | btrfs_free_path(path); | |
6426 | ||
6cbff00f CH |
6427 | btrfs_inherit_iflags(inode, dir); |
6428 | ||
569254b0 | 6429 | if (S_ISREG(mode)) { |
0b246afa | 6430 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6431 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6432 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6433 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6434 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6435 | } |
6436 | ||
5d4f98a2 | 6437 | inode_tree_add(inode); |
1abe9b8a | 6438 | |
6439 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6440 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6441 | |
8ea05e3a AB |
6442 | btrfs_update_root_times(trans, root); |
6443 | ||
63541927 FDBM |
6444 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6445 | if (ret) | |
0b246afa | 6446 | btrfs_err(fs_info, |
63541927 | 6447 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6448 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6449 | |
39279cc3 | 6450 | return inode; |
b0d5d10f CM |
6451 | |
6452 | fail_unlock: | |
32955c54 | 6453 | discard_new_inode(inode); |
5f39d397 | 6454 | fail: |
ef3b9af5 | 6455 | if (dir && name) |
aec7477b | 6456 | BTRFS_I(dir)->index_cnt--; |
5f39d397 CM |
6457 | btrfs_free_path(path); |
6458 | return ERR_PTR(ret); | |
39279cc3 CM |
6459 | } |
6460 | ||
d352ac68 CM |
6461 | /* |
6462 | * utility function to add 'inode' into 'parent_inode' with | |
6463 | * a give name and a given sequence number. | |
6464 | * if 'add_backref' is true, also insert a backref from the | |
6465 | * inode to the parent directory. | |
6466 | */ | |
e02119d5 | 6467 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6468 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6469 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6470 | { |
4df27c4d | 6471 | int ret = 0; |
39279cc3 | 6472 | struct btrfs_key key; |
db0a669f NB |
6473 | struct btrfs_root *root = parent_inode->root; |
6474 | u64 ino = btrfs_ino(inode); | |
6475 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6476 | |
33345d01 | 6477 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6478 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6479 | } else { |
33345d01 | 6480 | key.objectid = ino; |
962a298f | 6481 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6482 | key.offset = 0; |
6483 | } | |
6484 | ||
33345d01 | 6485 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
6025c19f | 6486 | ret = btrfs_add_root_ref(trans, key.objectid, |
0b246afa JM |
6487 | root->root_key.objectid, parent_ino, |
6488 | index, name, name_len); | |
4df27c4d | 6489 | } else if (add_backref) { |
33345d01 LZ |
6490 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6491 | parent_ino, index); | |
4df27c4d | 6492 | } |
39279cc3 | 6493 | |
79787eaa JM |
6494 | /* Nothing to clean up yet */ |
6495 | if (ret) | |
6496 | return ret; | |
4df27c4d | 6497 | |
684572df | 6498 | ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key, |
db0a669f | 6499 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6500 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6501 | goto fail_dir_item; |
6502 | else if (ret) { | |
66642832 | 6503 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6504 | return ret; |
39279cc3 | 6505 | } |
79787eaa | 6506 | |
db0a669f | 6507 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6508 | name_len * 2); |
db0a669f | 6509 | inode_inc_iversion(&parent_inode->vfs_inode); |
5338e43a FM |
6510 | /* |
6511 | * If we are replaying a log tree, we do not want to update the mtime | |
6512 | * and ctime of the parent directory with the current time, since the | |
6513 | * log replay procedure is responsible for setting them to their correct | |
6514 | * values (the ones it had when the fsync was done). | |
6515 | */ | |
6516 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) { | |
6517 | struct timespec64 now = current_time(&parent_inode->vfs_inode); | |
6518 | ||
6519 | parent_inode->vfs_inode.i_mtime = now; | |
6520 | parent_inode->vfs_inode.i_ctime = now; | |
6521 | } | |
db0a669f | 6522 | ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode); |
79787eaa | 6523 | if (ret) |
66642832 | 6524 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6525 | return ret; |
fe66a05a CM |
6526 | |
6527 | fail_dir_item: | |
6528 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6529 | u64 local_index; | |
6530 | int err; | |
3ee1c553 | 6531 | err = btrfs_del_root_ref(trans, key.objectid, |
0b246afa JM |
6532 | root->root_key.objectid, parent_ino, |
6533 | &local_index, name, name_len); | |
1690dd41 JT |
6534 | if (err) |
6535 | btrfs_abort_transaction(trans, err); | |
fe66a05a CM |
6536 | } else if (add_backref) { |
6537 | u64 local_index; | |
6538 | int err; | |
6539 | ||
6540 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6541 | ino, parent_ino, &local_index); | |
1690dd41 JT |
6542 | if (err) |
6543 | btrfs_abort_transaction(trans, err); | |
fe66a05a | 6544 | } |
1690dd41 JT |
6545 | |
6546 | /* Return the original error code */ | |
fe66a05a | 6547 | return ret; |
39279cc3 CM |
6548 | } |
6549 | ||
6550 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
cef415af NB |
6551 | struct btrfs_inode *dir, struct dentry *dentry, |
6552 | struct btrfs_inode *inode, int backref, u64 index) | |
39279cc3 | 6553 | { |
a1b075d2 JB |
6554 | int err = btrfs_add_link(trans, dir, inode, |
6555 | dentry->d_name.name, dentry->d_name.len, | |
6556 | backref, index); | |
39279cc3 CM |
6557 | if (err > 0) |
6558 | err = -EEXIST; | |
6559 | return err; | |
6560 | } | |
6561 | ||
618e21d5 | 6562 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6563 | umode_t mode, dev_t rdev) |
618e21d5 | 6564 | { |
2ff7e61e | 6565 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6566 | struct btrfs_trans_handle *trans; |
6567 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6568 | struct inode *inode = NULL; |
618e21d5 | 6569 | int err; |
618e21d5 | 6570 | u64 objectid; |
00e4e6b3 | 6571 | u64 index = 0; |
618e21d5 | 6572 | |
9ed74f2d JB |
6573 | /* |
6574 | * 2 for inode item and ref | |
6575 | * 2 for dir items | |
6576 | * 1 for xattr if selinux is on | |
6577 | */ | |
a22285a6 YZ |
6578 | trans = btrfs_start_transaction(root, 5); |
6579 | if (IS_ERR(trans)) | |
6580 | return PTR_ERR(trans); | |
1832a6d5 | 6581 | |
581bb050 LZ |
6582 | err = btrfs_find_free_ino(root, &objectid); |
6583 | if (err) | |
6584 | goto out_unlock; | |
6585 | ||
aec7477b | 6586 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6587 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6588 | mode, &index); | |
7cf96da3 TI |
6589 | if (IS_ERR(inode)) { |
6590 | err = PTR_ERR(inode); | |
32955c54 | 6591 | inode = NULL; |
618e21d5 | 6592 | goto out_unlock; |
7cf96da3 | 6593 | } |
618e21d5 | 6594 | |
ad19db71 CS |
6595 | /* |
6596 | * If the active LSM wants to access the inode during | |
6597 | * d_instantiate it needs these. Smack checks to see | |
6598 | * if the filesystem supports xattrs by looking at the | |
6599 | * ops vector. | |
6600 | */ | |
ad19db71 | 6601 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6602 | init_special_inode(inode, inode->i_mode, rdev); |
6603 | ||
6604 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6605 | if (err) |
32955c54 | 6606 | goto out_unlock; |
b0d5d10f | 6607 | |
cef415af NB |
6608 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6609 | 0, index); | |
32955c54 AV |
6610 | if (err) |
6611 | goto out_unlock; | |
6612 | ||
6613 | btrfs_update_inode(trans, root, inode); | |
6614 | d_instantiate_new(dentry, inode); | |
b0d5d10f | 6615 | |
618e21d5 | 6616 | out_unlock: |
3a45bb20 | 6617 | btrfs_end_transaction(trans); |
2ff7e61e | 6618 | btrfs_btree_balance_dirty(fs_info); |
32955c54 | 6619 | if (err && inode) { |
618e21d5 | 6620 | inode_dec_link_count(inode); |
32955c54 | 6621 | discard_new_inode(inode); |
618e21d5 | 6622 | } |
618e21d5 JB |
6623 | return err; |
6624 | } | |
6625 | ||
39279cc3 | 6626 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6627 | umode_t mode, bool excl) |
39279cc3 | 6628 | { |
2ff7e61e | 6629 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6630 | struct btrfs_trans_handle *trans; |
6631 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6632 | struct inode *inode = NULL; |
a22285a6 | 6633 | int err; |
39279cc3 | 6634 | u64 objectid; |
00e4e6b3 | 6635 | u64 index = 0; |
39279cc3 | 6636 | |
9ed74f2d JB |
6637 | /* |
6638 | * 2 for inode item and ref | |
6639 | * 2 for dir items | |
6640 | * 1 for xattr if selinux is on | |
6641 | */ | |
a22285a6 YZ |
6642 | trans = btrfs_start_transaction(root, 5); |
6643 | if (IS_ERR(trans)) | |
6644 | return PTR_ERR(trans); | |
9ed74f2d | 6645 | |
581bb050 LZ |
6646 | err = btrfs_find_free_ino(root, &objectid); |
6647 | if (err) | |
6648 | goto out_unlock; | |
6649 | ||
aec7477b | 6650 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6651 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6652 | mode, &index); | |
7cf96da3 TI |
6653 | if (IS_ERR(inode)) { |
6654 | err = PTR_ERR(inode); | |
32955c54 | 6655 | inode = NULL; |
39279cc3 | 6656 | goto out_unlock; |
7cf96da3 | 6657 | } |
ad19db71 CS |
6658 | /* |
6659 | * If the active LSM wants to access the inode during | |
6660 | * d_instantiate it needs these. Smack checks to see | |
6661 | * if the filesystem supports xattrs by looking at the | |
6662 | * ops vector. | |
6663 | */ | |
6664 | inode->i_fop = &btrfs_file_operations; | |
6665 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6666 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6667 | |
6668 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6669 | if (err) | |
32955c54 | 6670 | goto out_unlock; |
b0d5d10f CM |
6671 | |
6672 | err = btrfs_update_inode(trans, root, inode); | |
6673 | if (err) | |
32955c54 | 6674 | goto out_unlock; |
ad19db71 | 6675 | |
cef415af NB |
6676 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6677 | 0, index); | |
39279cc3 | 6678 | if (err) |
32955c54 | 6679 | goto out_unlock; |
43baa579 | 6680 | |
43baa579 | 6681 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
1e2e547a | 6682 | d_instantiate_new(dentry, inode); |
43baa579 | 6683 | |
39279cc3 | 6684 | out_unlock: |
3a45bb20 | 6685 | btrfs_end_transaction(trans); |
32955c54 | 6686 | if (err && inode) { |
39279cc3 | 6687 | inode_dec_link_count(inode); |
32955c54 | 6688 | discard_new_inode(inode); |
39279cc3 | 6689 | } |
2ff7e61e | 6690 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6691 | return err; |
6692 | } | |
6693 | ||
6694 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6695 | struct dentry *dentry) | |
6696 | { | |
271dba45 | 6697 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6698 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6699 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6700 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6701 | u64 index; |
39279cc3 CM |
6702 | int err; |
6703 | int drop_inode = 0; | |
6704 | ||
4a8be425 | 6705 | /* do not allow sys_link's with other subvols of the same device */ |
4fd786e6 | 6706 | if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid) |
3ab3564f | 6707 | return -EXDEV; |
4a8be425 | 6708 | |
f186373f | 6709 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6710 | return -EMLINK; |
4a8be425 | 6711 | |
877574e2 | 6712 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6713 | if (err) |
6714 | goto fail; | |
6715 | ||
a22285a6 | 6716 | /* |
7e6b6465 | 6717 | * 2 items for inode and inode ref |
a22285a6 | 6718 | * 2 items for dir items |
7e6b6465 | 6719 | * 1 item for parent inode |
399b0bbf | 6720 | * 1 item for orphan item deletion if O_TMPFILE |
a22285a6 | 6721 | */ |
399b0bbf | 6722 | trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6); |
a22285a6 YZ |
6723 | if (IS_ERR(trans)) { |
6724 | err = PTR_ERR(trans); | |
271dba45 | 6725 | trans = NULL; |
a22285a6 YZ |
6726 | goto fail; |
6727 | } | |
5f39d397 | 6728 | |
67de1176 MX |
6729 | /* There are several dir indexes for this inode, clear the cache. */ |
6730 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6731 | inc_nlink(inode); |
0c4d2d95 | 6732 | inode_inc_iversion(inode); |
c2050a45 | 6733 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6734 | ihold(inode); |
e9976151 | 6735 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6736 | |
cef415af NB |
6737 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6738 | 1, index); | |
5f39d397 | 6739 | |
a5719521 | 6740 | if (err) { |
54aa1f4d | 6741 | drop_inode = 1; |
a5719521 | 6742 | } else { |
10d9f309 | 6743 | struct dentry *parent = dentry->d_parent; |
d4682ba0 FM |
6744 | int ret; |
6745 | ||
a5719521 | 6746 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6747 | if (err) |
6748 | goto fail; | |
ef3b9af5 FM |
6749 | if (inode->i_nlink == 1) { |
6750 | /* | |
6751 | * If new hard link count is 1, it's a file created | |
6752 | * with open(2) O_TMPFILE flag. | |
6753 | */ | |
3d6ae7bb | 6754 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6755 | if (err) |
6756 | goto fail; | |
6757 | } | |
08c422c2 | 6758 | d_instantiate(dentry, inode); |
d4682ba0 FM |
6759 | ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent, |
6760 | true, NULL); | |
6761 | if (ret == BTRFS_NEED_TRANS_COMMIT) { | |
6762 | err = btrfs_commit_transaction(trans); | |
6763 | trans = NULL; | |
6764 | } | |
a5719521 | 6765 | } |
39279cc3 | 6766 | |
1832a6d5 | 6767 | fail: |
271dba45 | 6768 | if (trans) |
3a45bb20 | 6769 | btrfs_end_transaction(trans); |
39279cc3 CM |
6770 | if (drop_inode) { |
6771 | inode_dec_link_count(inode); | |
6772 | iput(inode); | |
6773 | } | |
2ff7e61e | 6774 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6775 | return err; |
6776 | } | |
6777 | ||
18bb1db3 | 6778 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6779 | { |
2ff7e61e | 6780 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6781 | struct inode *inode = NULL; |
39279cc3 CM |
6782 | struct btrfs_trans_handle *trans; |
6783 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6784 | int err = 0; | |
b9d86667 | 6785 | u64 objectid = 0; |
00e4e6b3 | 6786 | u64 index = 0; |
39279cc3 | 6787 | |
9ed74f2d JB |
6788 | /* |
6789 | * 2 items for inode and ref | |
6790 | * 2 items for dir items | |
6791 | * 1 for xattr if selinux is on | |
6792 | */ | |
a22285a6 YZ |
6793 | trans = btrfs_start_transaction(root, 5); |
6794 | if (IS_ERR(trans)) | |
6795 | return PTR_ERR(trans); | |
39279cc3 | 6796 | |
581bb050 LZ |
6797 | err = btrfs_find_free_ino(root, &objectid); |
6798 | if (err) | |
6799 | goto out_fail; | |
6800 | ||
aec7477b | 6801 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6802 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6803 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6804 | if (IS_ERR(inode)) { |
6805 | err = PTR_ERR(inode); | |
32955c54 | 6806 | inode = NULL; |
39279cc3 CM |
6807 | goto out_fail; |
6808 | } | |
5f39d397 | 6809 | |
b0d5d10f CM |
6810 | /* these must be set before we unlock the inode */ |
6811 | inode->i_op = &btrfs_dir_inode_operations; | |
6812 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6813 | |
2a7dba39 | 6814 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6815 | if (err) |
32955c54 | 6816 | goto out_fail; |
39279cc3 | 6817 | |
6ef06d27 | 6818 | btrfs_i_size_write(BTRFS_I(inode), 0); |
39279cc3 CM |
6819 | err = btrfs_update_inode(trans, root, inode); |
6820 | if (err) | |
32955c54 | 6821 | goto out_fail; |
5f39d397 | 6822 | |
db0a669f NB |
6823 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6824 | dentry->d_name.name, | |
6825 | dentry->d_name.len, 0, index); | |
39279cc3 | 6826 | if (err) |
32955c54 | 6827 | goto out_fail; |
5f39d397 | 6828 | |
1e2e547a | 6829 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
6830 | |
6831 | out_fail: | |
3a45bb20 | 6832 | btrfs_end_transaction(trans); |
32955c54 | 6833 | if (err && inode) { |
c7cfb8a5 | 6834 | inode_dec_link_count(inode); |
32955c54 | 6835 | discard_new_inode(inode); |
c7cfb8a5 | 6836 | } |
2ff7e61e | 6837 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6838 | return err; |
6839 | } | |
6840 | ||
c8b97818 | 6841 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6842 | struct page *page, |
c8b97818 CM |
6843 | size_t pg_offset, u64 extent_offset, |
6844 | struct btrfs_file_extent_item *item) | |
6845 | { | |
6846 | int ret; | |
6847 | struct extent_buffer *leaf = path->nodes[0]; | |
6848 | char *tmp; | |
6849 | size_t max_size; | |
6850 | unsigned long inline_size; | |
6851 | unsigned long ptr; | |
261507a0 | 6852 | int compress_type; |
c8b97818 CM |
6853 | |
6854 | WARN_ON(pg_offset != 0); | |
261507a0 | 6855 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6856 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6857 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6858 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6859 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6860 | if (!tmp) |
6861 | return -ENOMEM; | |
c8b97818 CM |
6862 | ptr = btrfs_file_extent_inline_start(item); |
6863 | ||
6864 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6865 | ||
09cbfeaf | 6866 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6867 | ret = btrfs_decompress(compress_type, tmp, page, |
6868 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6869 | |
6870 | /* | |
6871 | * decompression code contains a memset to fill in any space between the end | |
6872 | * of the uncompressed data and the end of max_size in case the decompressed | |
6873 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6874 | * the end of an inline extent and the beginning of the next block, so we | |
6875 | * cover that region here. | |
6876 | */ | |
6877 | ||
6878 | if (max_size + pg_offset < PAGE_SIZE) { | |
6879 | char *map = kmap(page); | |
6880 | memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset); | |
6881 | kunmap(page); | |
6882 | } | |
c8b97818 | 6883 | kfree(tmp); |
166ae5a4 | 6884 | return ret; |
c8b97818 CM |
6885 | } |
6886 | ||
d352ac68 CM |
6887 | /* |
6888 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6889 | * the ugly parts come from merging extents from the disk with the in-ram |
6890 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6891 | * where the in-ram extents might be locked pending data=ordered completion. |
6892 | * | |
6893 | * This also copies inline extents directly into the page. | |
6894 | */ | |
fc4f21b1 | 6895 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
de2c6615 LB |
6896 | struct page *page, |
6897 | size_t pg_offset, u64 start, u64 len, | |
6898 | int create) | |
a52d9a80 | 6899 | { |
3ffbd68c | 6900 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
a52d9a80 CM |
6901 | int ret; |
6902 | int err = 0; | |
a52d9a80 CM |
6903 | u64 extent_start = 0; |
6904 | u64 extent_end = 0; | |
fc4f21b1 | 6905 | u64 objectid = btrfs_ino(inode); |
7e74e235 | 6906 | int extent_type = -1; |
f421950f | 6907 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6908 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6909 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6910 | struct extent_buffer *leaf; |
6911 | struct btrfs_key found_key; | |
a52d9a80 | 6912 | struct extent_map *em = NULL; |
fc4f21b1 NB |
6913 | struct extent_map_tree *em_tree = &inode->extent_tree; |
6914 | struct extent_io_tree *io_tree = &inode->io_tree; | |
7ffbb598 | 6915 | const bool new_inline = !page || create; |
a52d9a80 | 6916 | |
890871be | 6917 | read_lock(&em_tree->lock); |
d1310b2e | 6918 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d | 6919 | if (em) |
0b246afa | 6920 | em->bdev = fs_info->fs_devices->latest_bdev; |
890871be | 6921 | read_unlock(&em_tree->lock); |
d1310b2e | 6922 | |
a52d9a80 | 6923 | if (em) { |
e1c4b745 CM |
6924 | if (em->start > start || em->start + em->len <= start) |
6925 | free_extent_map(em); | |
6926 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6927 | free_extent_map(em); |
6928 | else | |
6929 | goto out; | |
a52d9a80 | 6930 | } |
172ddd60 | 6931 | em = alloc_extent_map(); |
a52d9a80 | 6932 | if (!em) { |
d1310b2e CM |
6933 | err = -ENOMEM; |
6934 | goto out; | |
a52d9a80 | 6935 | } |
0b246afa | 6936 | em->bdev = fs_info->fs_devices->latest_bdev; |
d1310b2e | 6937 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6938 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6939 | em->len = (u64)-1; |
c8b97818 | 6940 | em->block_len = (u64)-1; |
f421950f | 6941 | |
bee6ec82 | 6942 | path = btrfs_alloc_path(); |
f421950f | 6943 | if (!path) { |
bee6ec82 LB |
6944 | err = -ENOMEM; |
6945 | goto out; | |
f421950f CM |
6946 | } |
6947 | ||
bee6ec82 LB |
6948 | /* Chances are we'll be called again, so go ahead and do readahead */ |
6949 | path->reada = READA_FORWARD; | |
6950 | ||
e49aabd9 LB |
6951 | /* |
6952 | * Unless we're going to uncompress the inline extent, no sleep would | |
6953 | * happen. | |
6954 | */ | |
6955 | path->leave_spinning = 1; | |
6956 | ||
5c9a702e | 6957 | ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0); |
a52d9a80 CM |
6958 | if (ret < 0) { |
6959 | err = ret; | |
6960 | goto out; | |
b8eeab7f | 6961 | } else if (ret > 0) { |
a52d9a80 CM |
6962 | if (path->slots[0] == 0) |
6963 | goto not_found; | |
6964 | path->slots[0]--; | |
6965 | } | |
6966 | ||
5f39d397 CM |
6967 | leaf = path->nodes[0]; |
6968 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6969 | struct btrfs_file_extent_item); |
5f39d397 | 6970 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5f39d397 | 6971 | if (found_key.objectid != objectid || |
694c12ed | 6972 | found_key.type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6973 | /* |
6974 | * If we backup past the first extent we want to move forward | |
6975 | * and see if there is an extent in front of us, otherwise we'll | |
6976 | * say there is a hole for our whole search range which can | |
6977 | * cause problems. | |
6978 | */ | |
6979 | extent_end = start; | |
6980 | goto next; | |
a52d9a80 CM |
6981 | } |
6982 | ||
694c12ed | 6983 | extent_type = btrfs_file_extent_type(leaf, item); |
5f39d397 | 6984 | extent_start = found_key.offset; |
694c12ed NB |
6985 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6986 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6bf9e4bd QW |
6987 | /* Only regular file could have regular/prealloc extent */ |
6988 | if (!S_ISREG(inode->vfs_inode.i_mode)) { | |
6989 | ret = -EUCLEAN; | |
6990 | btrfs_crit(fs_info, | |
6991 | "regular/prealloc extent found for non-regular inode %llu", | |
6992 | btrfs_ino(inode)); | |
6993 | goto out; | |
6994 | } | |
a52d9a80 | 6995 | extent_end = extent_start + |
db94535d | 6996 | btrfs_file_extent_num_bytes(leaf, item); |
09ed2f16 LB |
6997 | |
6998 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, | |
6999 | extent_start); | |
694c12ed | 7000 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
9036c102 | 7001 | size_t size; |
e41ca589 QW |
7002 | |
7003 | size = btrfs_file_extent_ram_bytes(leaf, item); | |
da17066c | 7004 | extent_end = ALIGN(extent_start + size, |
0b246afa | 7005 | fs_info->sectorsize); |
09ed2f16 LB |
7006 | |
7007 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, | |
7008 | path->slots[0], | |
7009 | extent_start); | |
9036c102 | 7010 | } |
25a50341 | 7011 | next: |
9036c102 YZ |
7012 | if (start >= extent_end) { |
7013 | path->slots[0]++; | |
7014 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
7015 | ret = btrfs_next_leaf(root, path); | |
7016 | if (ret < 0) { | |
7017 | err = ret; | |
7018 | goto out; | |
b8eeab7f | 7019 | } else if (ret > 0) { |
9036c102 | 7020 | goto not_found; |
b8eeab7f | 7021 | } |
9036c102 | 7022 | leaf = path->nodes[0]; |
a52d9a80 | 7023 | } |
9036c102 YZ |
7024 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
7025 | if (found_key.objectid != objectid || | |
7026 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
7027 | goto not_found; | |
7028 | if (start + len <= found_key.offset) | |
7029 | goto not_found; | |
e2eca69d WS |
7030 | if (start > found_key.offset) |
7031 | goto next; | |
02a033df NB |
7032 | |
7033 | /* New extent overlaps with existing one */ | |
9036c102 | 7034 | em->start = start; |
70c8a91c | 7035 | em->orig_start = start; |
9036c102 | 7036 | em->len = found_key.offset - start; |
02a033df NB |
7037 | em->block_start = EXTENT_MAP_HOLE; |
7038 | goto insert; | |
9036c102 YZ |
7039 | } |
7040 | ||
fc4f21b1 | 7041 | btrfs_extent_item_to_extent_map(inode, path, item, |
9cdc5124 | 7042 | new_inline, em); |
7ffbb598 | 7043 | |
694c12ed NB |
7044 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
7045 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 7046 | goto insert; |
694c12ed | 7047 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 7048 | unsigned long ptr; |
a52d9a80 | 7049 | char *map; |
3326d1b0 CM |
7050 | size_t size; |
7051 | size_t extent_offset; | |
7052 | size_t copy_size; | |
a52d9a80 | 7053 | |
7ffbb598 | 7054 | if (new_inline) |
689f9346 | 7055 | goto out; |
5f39d397 | 7056 | |
e41ca589 | 7057 | size = btrfs_file_extent_ram_bytes(leaf, item); |
9036c102 | 7058 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
7059 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
7060 | size - extent_offset); | |
3326d1b0 | 7061 | em->start = extent_start + extent_offset; |
0b246afa | 7062 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 7063 | em->orig_block_len = em->len; |
70c8a91c | 7064 | em->orig_start = em->start; |
689f9346 | 7065 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
e49aabd9 LB |
7066 | |
7067 | btrfs_set_path_blocking(path); | |
bf46f52d | 7068 | if (!PageUptodate(page)) { |
261507a0 LZ |
7069 | if (btrfs_file_extent_compression(leaf, item) != |
7070 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 7071 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 7072 | extent_offset, item); |
166ae5a4 ZB |
7073 | if (ret) { |
7074 | err = ret; | |
7075 | goto out; | |
7076 | } | |
c8b97818 CM |
7077 | } else { |
7078 | map = kmap(page); | |
7079 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
7080 | copy_size); | |
09cbfeaf | 7081 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 7082 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 7083 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
7084 | copy_size); |
7085 | } | |
c8b97818 CM |
7086 | kunmap(page); |
7087 | } | |
179e29e4 | 7088 | flush_dcache_page(page); |
a52d9a80 | 7089 | } |
d1310b2e | 7090 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 7091 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 7092 | goto insert; |
a52d9a80 CM |
7093 | } |
7094 | not_found: | |
7095 | em->start = start; | |
70c8a91c | 7096 | em->orig_start = start; |
d1310b2e | 7097 | em->len = len; |
5f39d397 | 7098 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 | 7099 | insert: |
b3b4aa74 | 7100 | btrfs_release_path(path); |
d1310b2e | 7101 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 7102 | btrfs_err(fs_info, |
5d163e0e JM |
7103 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
7104 | em->start, em->len, start, len); | |
a52d9a80 CM |
7105 | err = -EIO; |
7106 | goto out; | |
7107 | } | |
d1310b2e CM |
7108 | |
7109 | err = 0; | |
890871be | 7110 | write_lock(&em_tree->lock); |
f46b24c9 | 7111 | err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len); |
890871be | 7112 | write_unlock(&em_tree->lock); |
a52d9a80 | 7113 | out: |
c6414280 | 7114 | btrfs_free_path(path); |
1abe9b8a | 7115 | |
fc4f21b1 | 7116 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 7117 | |
a52d9a80 CM |
7118 | if (err) { |
7119 | free_extent_map(em); | |
a52d9a80 CM |
7120 | return ERR_PTR(err); |
7121 | } | |
79787eaa | 7122 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
7123 | return em; |
7124 | } | |
7125 | ||
fc4f21b1 | 7126 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
4ab47a8d | 7127 | u64 start, u64 len) |
ec29ed5b CM |
7128 | { |
7129 | struct extent_map *em; | |
7130 | struct extent_map *hole_em = NULL; | |
f3714ef4 | 7131 | u64 delalloc_start = start; |
ec29ed5b | 7132 | u64 end; |
f3714ef4 NB |
7133 | u64 delalloc_len; |
7134 | u64 delalloc_end; | |
ec29ed5b CM |
7135 | int err = 0; |
7136 | ||
4ab47a8d | 7137 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
ec29ed5b CM |
7138 | if (IS_ERR(em)) |
7139 | return em; | |
9986277e DC |
7140 | /* |
7141 | * If our em maps to: | |
7142 | * - a hole or | |
7143 | * - a pre-alloc extent, | |
7144 | * there might actually be delalloc bytes behind it. | |
7145 | */ | |
7146 | if (em->block_start != EXTENT_MAP_HOLE && | |
7147 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7148 | return em; | |
7149 | else | |
7150 | hole_em = em; | |
ec29ed5b CM |
7151 | |
7152 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7153 | end = start + len; | |
7154 | if (end < start) | |
7155 | end = (u64)-1; | |
7156 | else | |
7157 | end -= 1; | |
7158 | ||
7159 | em = NULL; | |
7160 | ||
7161 | /* ok, we didn't find anything, lets look for delalloc */ | |
f3714ef4 | 7162 | delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start, |
ec29ed5b | 7163 | end, len, EXTENT_DELALLOC, 1); |
f3714ef4 NB |
7164 | delalloc_end = delalloc_start + delalloc_len; |
7165 | if (delalloc_end < delalloc_start) | |
7166 | delalloc_end = (u64)-1; | |
ec29ed5b CM |
7167 | |
7168 | /* | |
f3714ef4 NB |
7169 | * We didn't find anything useful, return the original results from |
7170 | * get_extent() | |
ec29ed5b | 7171 | */ |
f3714ef4 | 7172 | if (delalloc_start > end || delalloc_end <= start) { |
ec29ed5b CM |
7173 | em = hole_em; |
7174 | hole_em = NULL; | |
7175 | goto out; | |
7176 | } | |
7177 | ||
f3714ef4 NB |
7178 | /* |
7179 | * Adjust the delalloc_start to make sure it doesn't go backwards from | |
7180 | * the start they passed in | |
ec29ed5b | 7181 | */ |
f3714ef4 NB |
7182 | delalloc_start = max(start, delalloc_start); |
7183 | delalloc_len = delalloc_end - delalloc_start; | |
ec29ed5b | 7184 | |
f3714ef4 NB |
7185 | if (delalloc_len > 0) { |
7186 | u64 hole_start; | |
02950af4 | 7187 | u64 hole_len; |
f3714ef4 | 7188 | const u64 hole_end = extent_map_end(hole_em); |
ec29ed5b | 7189 | |
172ddd60 | 7190 | em = alloc_extent_map(); |
ec29ed5b CM |
7191 | if (!em) { |
7192 | err = -ENOMEM; | |
7193 | goto out; | |
7194 | } | |
f3714ef4 NB |
7195 | em->bdev = NULL; |
7196 | ||
7197 | ASSERT(hole_em); | |
ec29ed5b | 7198 | /* |
f3714ef4 NB |
7199 | * When btrfs_get_extent can't find anything it returns one |
7200 | * huge hole | |
ec29ed5b | 7201 | * |
f3714ef4 NB |
7202 | * Make sure what it found really fits our range, and adjust to |
7203 | * make sure it is based on the start from the caller | |
ec29ed5b | 7204 | */ |
f3714ef4 NB |
7205 | if (hole_end <= start || hole_em->start > end) { |
7206 | free_extent_map(hole_em); | |
7207 | hole_em = NULL; | |
7208 | } else { | |
7209 | hole_start = max(hole_em->start, start); | |
7210 | hole_len = hole_end - hole_start; | |
ec29ed5b | 7211 | } |
f3714ef4 NB |
7212 | |
7213 | if (hole_em && delalloc_start > hole_start) { | |
7214 | /* | |
7215 | * Our hole starts before our delalloc, so we have to | |
7216 | * return just the parts of the hole that go until the | |
7217 | * delalloc starts | |
ec29ed5b | 7218 | */ |
f3714ef4 | 7219 | em->len = min(hole_len, delalloc_start - hole_start); |
ec29ed5b CM |
7220 | em->start = hole_start; |
7221 | em->orig_start = hole_start; | |
7222 | /* | |
f3714ef4 NB |
7223 | * Don't adjust block start at all, it is fixed at |
7224 | * EXTENT_MAP_HOLE | |
ec29ed5b CM |
7225 | */ |
7226 | em->block_start = hole_em->block_start; | |
7227 | em->block_len = hole_len; | |
f9e4fb53 LB |
7228 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7229 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b | 7230 | } else { |
f3714ef4 NB |
7231 | /* |
7232 | * Hole is out of passed range or it starts after | |
7233 | * delalloc range | |
7234 | */ | |
7235 | em->start = delalloc_start; | |
7236 | em->len = delalloc_len; | |
7237 | em->orig_start = delalloc_start; | |
ec29ed5b | 7238 | em->block_start = EXTENT_MAP_DELALLOC; |
f3714ef4 | 7239 | em->block_len = delalloc_len; |
ec29ed5b | 7240 | } |
bf8d32b9 | 7241 | } else { |
ec29ed5b CM |
7242 | return hole_em; |
7243 | } | |
7244 | out: | |
7245 | ||
7246 | free_extent_map(hole_em); | |
7247 | if (err) { | |
7248 | free_extent_map(em); | |
7249 | return ERR_PTR(err); | |
7250 | } | |
7251 | return em; | |
7252 | } | |
7253 | ||
5f9a8a51 FM |
7254 | static struct extent_map *btrfs_create_dio_extent(struct inode *inode, |
7255 | const u64 start, | |
7256 | const u64 len, | |
7257 | const u64 orig_start, | |
7258 | const u64 block_start, | |
7259 | const u64 block_len, | |
7260 | const u64 orig_block_len, | |
7261 | const u64 ram_bytes, | |
7262 | const int type) | |
7263 | { | |
7264 | struct extent_map *em = NULL; | |
7265 | int ret; | |
7266 | ||
5f9a8a51 | 7267 | if (type != BTRFS_ORDERED_NOCOW) { |
6f9994db LB |
7268 | em = create_io_em(inode, start, len, orig_start, |
7269 | block_start, block_len, orig_block_len, | |
7270 | ram_bytes, | |
7271 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
7272 | type); | |
5f9a8a51 FM |
7273 | if (IS_ERR(em)) |
7274 | goto out; | |
7275 | } | |
7276 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, | |
7277 | len, block_len, type); | |
7278 | if (ret) { | |
7279 | if (em) { | |
7280 | free_extent_map(em); | |
dcdbc059 | 7281 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
5f9a8a51 FM |
7282 | start + len - 1, 0); |
7283 | } | |
7284 | em = ERR_PTR(ret); | |
7285 | } | |
7286 | out: | |
5f9a8a51 FM |
7287 | |
7288 | return em; | |
7289 | } | |
7290 | ||
4b46fce2 JB |
7291 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7292 | u64 start, u64 len) | |
7293 | { | |
0b246afa | 7294 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7295 | struct btrfs_root *root = BTRFS_I(inode)->root; |
70c8a91c | 7296 | struct extent_map *em; |
4b46fce2 JB |
7297 | struct btrfs_key ins; |
7298 | u64 alloc_hint; | |
7299 | int ret; | |
4b46fce2 | 7300 | |
4b46fce2 | 7301 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7302 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7303 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7304 | if (ret) |
7305 | return ERR_PTR(ret); | |
4b46fce2 | 7306 | |
5f9a8a51 FM |
7307 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
7308 | ins.objectid, ins.offset, ins.offset, | |
6288d6ea | 7309 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7310 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7311 | if (IS_ERR(em)) |
2ff7e61e JM |
7312 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
7313 | ins.offset, 1); | |
de0ee0ed | 7314 | |
4b46fce2 JB |
7315 | return em; |
7316 | } | |
7317 | ||
46bfbb5c CM |
7318 | /* |
7319 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7320 | * block must be cow'd | |
7321 | */ | |
00361589 | 7322 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7323 | u64 *orig_start, u64 *orig_block_len, |
7324 | u64 *ram_bytes) | |
46bfbb5c | 7325 | { |
2ff7e61e | 7326 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
7327 | struct btrfs_path *path; |
7328 | int ret; | |
7329 | struct extent_buffer *leaf; | |
7330 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7331 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7332 | struct btrfs_file_extent_item *fi; |
7333 | struct btrfs_key key; | |
7334 | u64 disk_bytenr; | |
7335 | u64 backref_offset; | |
7336 | u64 extent_end; | |
7337 | u64 num_bytes; | |
7338 | int slot; | |
7339 | int found_type; | |
7ee9e440 | 7340 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7341 | |
46bfbb5c CM |
7342 | path = btrfs_alloc_path(); |
7343 | if (!path) | |
7344 | return -ENOMEM; | |
7345 | ||
f85b7379 DS |
7346 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7347 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7348 | if (ret < 0) |
7349 | goto out; | |
7350 | ||
7351 | slot = path->slots[0]; | |
7352 | if (ret == 1) { | |
7353 | if (slot == 0) { | |
7354 | /* can't find the item, must cow */ | |
7355 | ret = 0; | |
7356 | goto out; | |
7357 | } | |
7358 | slot--; | |
7359 | } | |
7360 | ret = 0; | |
7361 | leaf = path->nodes[0]; | |
7362 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7363 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7364 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7365 | /* not our file or wrong item type, must cow */ | |
7366 | goto out; | |
7367 | } | |
7368 | ||
7369 | if (key.offset > offset) { | |
7370 | /* Wrong offset, must cow */ | |
7371 | goto out; | |
7372 | } | |
7373 | ||
7374 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7375 | found_type = btrfs_file_extent_type(leaf, fi); | |
7376 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7377 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7378 | /* not a regular extent, must cow */ | |
7379 | goto out; | |
7380 | } | |
7ee9e440 JB |
7381 | |
7382 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7383 | goto out; | |
7384 | ||
e77751aa MX |
7385 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7386 | if (extent_end <= offset) | |
7387 | goto out; | |
7388 | ||
46bfbb5c | 7389 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7390 | if (disk_bytenr == 0) |
7391 | goto out; | |
7392 | ||
7393 | if (btrfs_file_extent_compression(leaf, fi) || | |
7394 | btrfs_file_extent_encryption(leaf, fi) || | |
7395 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7396 | goto out; | |
7397 | ||
78d4295b EL |
7398 | /* |
7399 | * Do the same check as in btrfs_cross_ref_exist but without the | |
7400 | * unnecessary search. | |
7401 | */ | |
7402 | if (btrfs_file_extent_generation(leaf, fi) <= | |
7403 | btrfs_root_last_snapshot(&root->root_item)) | |
7404 | goto out; | |
7405 | ||
46bfbb5c CM |
7406 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7407 | ||
7ee9e440 JB |
7408 | if (orig_start) { |
7409 | *orig_start = key.offset - backref_offset; | |
7410 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7411 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7412 | } | |
eb384b55 | 7413 | |
2ff7e61e | 7414 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7415 | goto out; |
7b2b7085 MX |
7416 | |
7417 | num_bytes = min(offset + *len, extent_end) - offset; | |
7418 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7419 | u64 range_end; | |
7420 | ||
da17066c JM |
7421 | range_end = round_up(offset + num_bytes, |
7422 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7423 | ret = test_range_bit(io_tree, offset, range_end, |
7424 | EXTENT_DELALLOC, 0, NULL); | |
7425 | if (ret) { | |
7426 | ret = -EAGAIN; | |
7427 | goto out; | |
7428 | } | |
7429 | } | |
7430 | ||
1bda19eb | 7431 | btrfs_release_path(path); |
46bfbb5c CM |
7432 | |
7433 | /* | |
7434 | * look for other files referencing this extent, if we | |
7435 | * find any we must cow | |
7436 | */ | |
00361589 | 7437 | |
e4c3b2dc | 7438 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
00361589 | 7439 | key.offset - backref_offset, disk_bytenr); |
00361589 JB |
7440 | if (ret) { |
7441 | ret = 0; | |
7442 | goto out; | |
7443 | } | |
46bfbb5c CM |
7444 | |
7445 | /* | |
7446 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7447 | * in this extent we are about to write. If there | |
7448 | * are any csums in that range we have to cow in order | |
7449 | * to keep the csums correct | |
7450 | */ | |
7451 | disk_bytenr += backref_offset; | |
7452 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7453 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7454 | goto out; | |
46bfbb5c CM |
7455 | /* |
7456 | * all of the above have passed, it is safe to overwrite this extent | |
7457 | * without cow | |
7458 | */ | |
eb384b55 | 7459 | *len = num_bytes; |
46bfbb5c CM |
7460 | ret = 1; |
7461 | out: | |
7462 | btrfs_free_path(path); | |
7463 | return ret; | |
7464 | } | |
7465 | ||
eb838e73 JB |
7466 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7467 | struct extent_state **cached_state, int writing) | |
7468 | { | |
7469 | struct btrfs_ordered_extent *ordered; | |
7470 | int ret = 0; | |
7471 | ||
7472 | while (1) { | |
7473 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7474 | cached_state); |
eb838e73 JB |
7475 | /* |
7476 | * We're concerned with the entire range that we're going to be | |
01327610 | 7477 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7478 | * extents in this range. |
7479 | */ | |
a776c6fa | 7480 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7481 | lockend - lockstart + 1); |
7482 | ||
7483 | /* | |
7484 | * We need to make sure there are no buffered pages in this | |
7485 | * range either, we could have raced between the invalidate in | |
7486 | * generic_file_direct_write and locking the extent. The | |
7487 | * invalidate needs to happen so that reads after a write do not | |
7488 | * get stale data. | |
7489 | */ | |
fc4adbff | 7490 | if (!ordered && |
051c98eb DS |
7491 | (!writing || !filemap_range_has_page(inode->i_mapping, |
7492 | lockstart, lockend))) | |
eb838e73 JB |
7493 | break; |
7494 | ||
7495 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
e43bbe5e | 7496 | cached_state); |
eb838e73 JB |
7497 | |
7498 | if (ordered) { | |
ade77029 FM |
7499 | /* |
7500 | * If we are doing a DIO read and the ordered extent we | |
7501 | * found is for a buffered write, we can not wait for it | |
7502 | * to complete and retry, because if we do so we can | |
7503 | * deadlock with concurrent buffered writes on page | |
7504 | * locks. This happens only if our DIO read covers more | |
7505 | * than one extent map, if at this point has already | |
7506 | * created an ordered extent for a previous extent map | |
7507 | * and locked its range in the inode's io tree, and a | |
7508 | * concurrent write against that previous extent map's | |
7509 | * range and this range started (we unlock the ranges | |
7510 | * in the io tree only when the bios complete and | |
7511 | * buffered writes always lock pages before attempting | |
7512 | * to lock range in the io tree). | |
7513 | */ | |
7514 | if (writing || | |
7515 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7516 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7517 | else | |
7518 | ret = -ENOTBLK; | |
eb838e73 JB |
7519 | btrfs_put_ordered_extent(ordered); |
7520 | } else { | |
eb838e73 | 7521 | /* |
b850ae14 FM |
7522 | * We could trigger writeback for this range (and wait |
7523 | * for it to complete) and then invalidate the pages for | |
7524 | * this range (through invalidate_inode_pages2_range()), | |
7525 | * but that can lead us to a deadlock with a concurrent | |
7526 | * call to readpages() (a buffered read or a defrag call | |
7527 | * triggered a readahead) on a page lock due to an | |
7528 | * ordered dio extent we created before but did not have | |
7529 | * yet a corresponding bio submitted (whence it can not | |
7530 | * complete), which makes readpages() wait for that | |
7531 | * ordered extent to complete while holding a lock on | |
7532 | * that page. | |
eb838e73 | 7533 | */ |
b850ae14 | 7534 | ret = -ENOTBLK; |
eb838e73 JB |
7535 | } |
7536 | ||
ade77029 FM |
7537 | if (ret) |
7538 | break; | |
7539 | ||
eb838e73 JB |
7540 | cond_resched(); |
7541 | } | |
7542 | ||
7543 | return ret; | |
7544 | } | |
7545 | ||
6f9994db LB |
7546 | /* The callers of this must take lock_extent() */ |
7547 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, | |
7548 | u64 orig_start, u64 block_start, | |
7549 | u64 block_len, u64 orig_block_len, | |
7550 | u64 ram_bytes, int compress_type, | |
7551 | int type) | |
69ffb543 JB |
7552 | { |
7553 | struct extent_map_tree *em_tree; | |
7554 | struct extent_map *em; | |
7555 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7556 | int ret; | |
7557 | ||
6f9994db LB |
7558 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7559 | type == BTRFS_ORDERED_COMPRESSED || | |
7560 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7561 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7562 | |
69ffb543 JB |
7563 | em_tree = &BTRFS_I(inode)->extent_tree; |
7564 | em = alloc_extent_map(); | |
7565 | if (!em) | |
7566 | return ERR_PTR(-ENOMEM); | |
7567 | ||
7568 | em->start = start; | |
7569 | em->orig_start = orig_start; | |
7570 | em->len = len; | |
7571 | em->block_len = block_len; | |
7572 | em->block_start = block_start; | |
7573 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7574 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7575 | em->ram_bytes = ram_bytes; |
70c8a91c | 7576 | em->generation = -1; |
69ffb543 | 7577 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7578 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7579 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7580 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7581 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7582 | em->compress_type = compress_type; | |
7583 | } | |
69ffb543 JB |
7584 | |
7585 | do { | |
dcdbc059 | 7586 | btrfs_drop_extent_cache(BTRFS_I(inode), em->start, |
69ffb543 JB |
7587 | em->start + em->len - 1, 0); |
7588 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7589 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7590 | write_unlock(&em_tree->lock); |
6f9994db LB |
7591 | /* |
7592 | * The caller has taken lock_extent(), who could race with us | |
7593 | * to add em? | |
7594 | */ | |
69ffb543 JB |
7595 | } while (ret == -EEXIST); |
7596 | ||
7597 | if (ret) { | |
7598 | free_extent_map(em); | |
7599 | return ERR_PTR(ret); | |
7600 | } | |
7601 | ||
6f9994db | 7602 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7603 | return em; |
7604 | } | |
7605 | ||
1c8d0175 NB |
7606 | |
7607 | static int btrfs_get_blocks_direct_read(struct extent_map *em, | |
7608 | struct buffer_head *bh_result, | |
7609 | struct inode *inode, | |
7610 | u64 start, u64 len) | |
7611 | { | |
7612 | if (em->block_start == EXTENT_MAP_HOLE || | |
7613 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7614 | return -ENOENT; | |
7615 | ||
7616 | len = min(len, em->len - (start - em->start)); | |
7617 | ||
7618 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
7619 | inode->i_blkbits; | |
7620 | bh_result->b_size = len; | |
7621 | bh_result->b_bdev = em->bdev; | |
7622 | set_buffer_mapped(bh_result); | |
7623 | ||
7624 | return 0; | |
7625 | } | |
7626 | ||
c5794e51 NB |
7627 | static int btrfs_get_blocks_direct_write(struct extent_map **map, |
7628 | struct buffer_head *bh_result, | |
7629 | struct inode *inode, | |
7630 | struct btrfs_dio_data *dio_data, | |
7631 | u64 start, u64 len) | |
7632 | { | |
7633 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
7634 | struct extent_map *em = *map; | |
7635 | int ret = 0; | |
7636 | ||
7637 | /* | |
7638 | * We don't allocate a new extent in the following cases | |
7639 | * | |
7640 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7641 | * existing extent. | |
7642 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7643 | * just use the extent. | |
7644 | * | |
7645 | */ | |
7646 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7647 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7648 | em->block_start != EXTENT_MAP_HOLE)) { | |
7649 | int type; | |
7650 | u64 block_start, orig_start, orig_block_len, ram_bytes; | |
7651 | ||
7652 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7653 | type = BTRFS_ORDERED_PREALLOC; | |
7654 | else | |
7655 | type = BTRFS_ORDERED_NOCOW; | |
7656 | len = min(len, em->len - (start - em->start)); | |
7657 | block_start = em->block_start + (start - em->start); | |
7658 | ||
7659 | if (can_nocow_extent(inode, start, &len, &orig_start, | |
7660 | &orig_block_len, &ram_bytes) == 1 && | |
7661 | btrfs_inc_nocow_writers(fs_info, block_start)) { | |
7662 | struct extent_map *em2; | |
7663 | ||
7664 | em2 = btrfs_create_dio_extent(inode, start, len, | |
7665 | orig_start, block_start, | |
7666 | len, orig_block_len, | |
7667 | ram_bytes, type); | |
7668 | btrfs_dec_nocow_writers(fs_info, block_start); | |
7669 | if (type == BTRFS_ORDERED_PREALLOC) { | |
7670 | free_extent_map(em); | |
7671 | *map = em = em2; | |
7672 | } | |
7673 | ||
7674 | if (em2 && IS_ERR(em2)) { | |
7675 | ret = PTR_ERR(em2); | |
7676 | goto out; | |
7677 | } | |
7678 | /* | |
7679 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7680 | * use the existing or preallocated extent, so does not | |
7681 | * need to adjust btrfs_space_info's bytes_may_use. | |
7682 | */ | |
7683 | btrfs_free_reserved_data_space_noquota(inode, start, | |
7684 | len); | |
7685 | goto skip_cow; | |
7686 | } | |
7687 | } | |
7688 | ||
7689 | /* this will cow the extent */ | |
7690 | len = bh_result->b_size; | |
7691 | free_extent_map(em); | |
7692 | *map = em = btrfs_new_extent_direct(inode, start, len); | |
7693 | if (IS_ERR(em)) { | |
7694 | ret = PTR_ERR(em); | |
7695 | goto out; | |
7696 | } | |
7697 | ||
7698 | len = min(len, em->len - (start - em->start)); | |
7699 | ||
7700 | skip_cow: | |
7701 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
7702 | inode->i_blkbits; | |
7703 | bh_result->b_size = len; | |
7704 | bh_result->b_bdev = em->bdev; | |
7705 | set_buffer_mapped(bh_result); | |
7706 | ||
7707 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7708 | set_buffer_new(bh_result); | |
7709 | ||
7710 | /* | |
7711 | * Need to update the i_size under the extent lock so buffered | |
7712 | * readers will get the updated i_size when we unlock. | |
7713 | */ | |
7714 | if (!dio_data->overwrite && start + len > i_size_read(inode)) | |
7715 | i_size_write(inode, start + len); | |
7716 | ||
7717 | WARN_ON(dio_data->reserve < len); | |
7718 | dio_data->reserve -= len; | |
7719 | dio_data->unsubmitted_oe_range_end = start + len; | |
7720 | current->journal_info = dio_data; | |
7721 | out: | |
7722 | return ret; | |
7723 | } | |
7724 | ||
4b46fce2 JB |
7725 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7726 | struct buffer_head *bh_result, int create) | |
7727 | { | |
0b246afa | 7728 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7729 | struct extent_map *em; |
eb838e73 | 7730 | struct extent_state *cached_state = NULL; |
50745b0a | 7731 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7732 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7733 | u64 lockstart, lockend; |
4b46fce2 | 7734 | u64 len = bh_result->b_size; |
eb838e73 | 7735 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7736 | int ret = 0; |
eb838e73 | 7737 | |
172a5049 | 7738 | if (create) |
3266789f | 7739 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7740 | else |
0b246afa | 7741 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7742 | |
c329861d JB |
7743 | lockstart = start; |
7744 | lockend = start + len - 1; | |
7745 | ||
e1cbbfa5 JB |
7746 | if (current->journal_info) { |
7747 | /* | |
7748 | * Need to pull our outstanding extents and set journal_info to NULL so | |
01327610 | 7749 | * that anything that needs to check if there's a transaction doesn't get |
e1cbbfa5 JB |
7750 | * confused. |
7751 | */ | |
50745b0a | 7752 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7753 | current->journal_info = NULL; |
7754 | } | |
7755 | ||
eb838e73 JB |
7756 | /* |
7757 | * If this errors out it's because we couldn't invalidate pagecache for | |
7758 | * this range and we need to fallback to buffered. | |
7759 | */ | |
9c9464cc FM |
7760 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7761 | create)) { | |
7762 | ret = -ENOTBLK; | |
7763 | goto err; | |
7764 | } | |
eb838e73 | 7765 | |
fc4f21b1 | 7766 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); |
eb838e73 JB |
7767 | if (IS_ERR(em)) { |
7768 | ret = PTR_ERR(em); | |
7769 | goto unlock_err; | |
7770 | } | |
4b46fce2 JB |
7771 | |
7772 | /* | |
7773 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7774 | * io. INLINE is special, and we could probably kludge it in here, but | |
7775 | * it's still buffered so for safety lets just fall back to the generic | |
7776 | * buffered path. | |
7777 | * | |
7778 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7779 | * decompress it, so there will be buffering required no matter what we | |
7780 | * do, so go ahead and fallback to buffered. | |
7781 | * | |
01327610 | 7782 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7783 | * to buffered IO. Don't blame me, this is the price we pay for using |
7784 | * the generic code. | |
7785 | */ | |
7786 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7787 | em->block_start == EXTENT_MAP_INLINE) { | |
7788 | free_extent_map(em); | |
eb838e73 JB |
7789 | ret = -ENOTBLK; |
7790 | goto unlock_err; | |
4b46fce2 JB |
7791 | } |
7792 | ||
c5794e51 NB |
7793 | if (create) { |
7794 | ret = btrfs_get_blocks_direct_write(&em, bh_result, inode, | |
7795 | dio_data, start, len); | |
7796 | if (ret < 0) | |
7797 | goto unlock_err; | |
7798 | ||
7799 | /* clear and unlock the entire range */ | |
7800 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7801 | unlock_bits, 1, 0, &cached_state); | |
7802 | } else { | |
1c8d0175 NB |
7803 | ret = btrfs_get_blocks_direct_read(em, bh_result, inode, |
7804 | start, len); | |
7805 | /* Can be negative only if we read from a hole */ | |
7806 | if (ret < 0) { | |
7807 | ret = 0; | |
7808 | free_extent_map(em); | |
7809 | goto unlock_err; | |
7810 | } | |
7811 | /* | |
7812 | * We need to unlock only the end area that we aren't using. | |
7813 | * The rest is going to be unlocked by the endio routine. | |
7814 | */ | |
7815 | lockstart = start + bh_result->b_size; | |
7816 | if (lockstart < lockend) { | |
7817 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
7818 | lockend, unlock_bits, 1, 0, | |
7819 | &cached_state); | |
7820 | } else { | |
7821 | free_extent_state(cached_state); | |
7822 | } | |
4b46fce2 JB |
7823 | } |
7824 | ||
4b46fce2 JB |
7825 | free_extent_map(em); |
7826 | ||
7827 | return 0; | |
eb838e73 JB |
7828 | |
7829 | unlock_err: | |
eb838e73 | 7830 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
ae0f1625 | 7831 | unlock_bits, 1, 0, &cached_state); |
9c9464cc | 7832 | err: |
50745b0a | 7833 | if (dio_data) |
7834 | current->journal_info = dio_data; | |
eb838e73 | 7835 | return ret; |
4b46fce2 JB |
7836 | } |
7837 | ||
58efbc9f OS |
7838 | static inline blk_status_t submit_dio_repair_bio(struct inode *inode, |
7839 | struct bio *bio, | |
7840 | int mirror_num) | |
8b110e39 | 7841 | { |
2ff7e61e | 7842 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
58efbc9f | 7843 | blk_status_t ret; |
8b110e39 | 7844 | |
37226b21 | 7845 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 | 7846 | |
2ff7e61e | 7847 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR); |
8b110e39 | 7848 | if (ret) |
ea057f6d | 7849 | return ret; |
8b110e39 | 7850 | |
2ff7e61e | 7851 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
ea057f6d | 7852 | |
8b110e39 MX |
7853 | return ret; |
7854 | } | |
7855 | ||
7856 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7857 | struct bio *failed_bio, | |
7858 | struct io_failure_record *failrec, | |
7859 | int failed_mirror) | |
7860 | { | |
ab8d0fc4 | 7861 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 MX |
7862 | int num_copies; |
7863 | ||
ab8d0fc4 | 7864 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
8b110e39 MX |
7865 | if (num_copies == 1) { |
7866 | /* | |
7867 | * we only have a single copy of the data, so don't bother with | |
7868 | * all the retry and error correction code that follows. no | |
7869 | * matter what the error is, it is very likely to persist. | |
7870 | */ | |
ab8d0fc4 JM |
7871 | btrfs_debug(fs_info, |
7872 | "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
7873 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7874 | return 0; |
7875 | } | |
7876 | ||
7877 | failrec->failed_mirror = failed_mirror; | |
7878 | failrec->this_mirror++; | |
7879 | if (failrec->this_mirror == failed_mirror) | |
7880 | failrec->this_mirror++; | |
7881 | ||
7882 | if (failrec->this_mirror > num_copies) { | |
ab8d0fc4 JM |
7883 | btrfs_debug(fs_info, |
7884 | "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
7885 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7886 | return 0; |
7887 | } | |
7888 | ||
7889 | return 1; | |
7890 | } | |
7891 | ||
58efbc9f OS |
7892 | static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio, |
7893 | struct page *page, unsigned int pgoff, | |
7894 | u64 start, u64 end, int failed_mirror, | |
7895 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
7896 | { |
7897 | struct io_failure_record *failrec; | |
7870d082 JB |
7898 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
7899 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8b110e39 MX |
7900 | struct bio *bio; |
7901 | int isector; | |
f1c77c55 | 7902 | unsigned int read_mode = 0; |
17347cec | 7903 | int segs; |
8b110e39 | 7904 | int ret; |
58efbc9f | 7905 | blk_status_t status; |
c16a8ac3 | 7906 | struct bio_vec bvec; |
8b110e39 | 7907 | |
37226b21 | 7908 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
8b110e39 MX |
7909 | |
7910 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7911 | if (ret) | |
58efbc9f | 7912 | return errno_to_blk_status(ret); |
8b110e39 MX |
7913 | |
7914 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7915 | failed_mirror); | |
7916 | if (!ret) { | |
7870d082 | 7917 | free_io_failure(failure_tree, io_tree, failrec); |
58efbc9f | 7918 | return BLK_STS_IOERR; |
8b110e39 MX |
7919 | } |
7920 | ||
17347cec | 7921 | segs = bio_segments(failed_bio); |
c16a8ac3 | 7922 | bio_get_first_bvec(failed_bio, &bvec); |
17347cec | 7923 | if (segs > 1 || |
c16a8ac3 | 7924 | (bvec.bv_len > btrfs_inode_sectorsize(inode))) |
70fd7614 | 7925 | read_mode |= REQ_FAILFAST_DEV; |
8b110e39 MX |
7926 | |
7927 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7928 | isector >>= inode->i_sb->s_blocksize_bits; | |
7929 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 7930 | pgoff, isector, repair_endio, repair_arg); |
ebcc3263 | 7931 | bio->bi_opf = REQ_OP_READ | read_mode; |
8b110e39 MX |
7932 | |
7933 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
913e1535 | 7934 | "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d", |
8b110e39 MX |
7935 | read_mode, failrec->this_mirror, failrec->in_validation); |
7936 | ||
58efbc9f OS |
7937 | status = submit_dio_repair_bio(inode, bio, failrec->this_mirror); |
7938 | if (status) { | |
7870d082 | 7939 | free_io_failure(failure_tree, io_tree, failrec); |
8b110e39 MX |
7940 | bio_put(bio); |
7941 | } | |
7942 | ||
58efbc9f | 7943 | return status; |
8b110e39 MX |
7944 | } |
7945 | ||
7946 | struct btrfs_retry_complete { | |
7947 | struct completion done; | |
7948 | struct inode *inode; | |
7949 | u64 start; | |
7950 | int uptodate; | |
7951 | }; | |
7952 | ||
4246a0b6 | 7953 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
7954 | { |
7955 | struct btrfs_retry_complete *done = bio->bi_private; | |
7870d082 | 7956 | struct inode *inode = done->inode; |
8b110e39 | 7957 | struct bio_vec *bvec; |
7870d082 | 7958 | struct extent_io_tree *io_tree, *failure_tree; |
6dc4f100 | 7959 | struct bvec_iter_all iter_all; |
8b110e39 | 7960 | |
4e4cbee9 | 7961 | if (bio->bi_status) |
8b110e39 MX |
7962 | goto end; |
7963 | ||
2dabb324 | 7964 | ASSERT(bio->bi_vcnt == 1); |
7870d082 JB |
7965 | io_tree = &BTRFS_I(inode)->io_tree; |
7966 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
263663cd | 7967 | ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode)); |
2dabb324 | 7968 | |
8b110e39 | 7969 | done->uptodate = 1; |
c09abff8 | 7970 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 7971 | bio_for_each_segment_all(bvec, bio, iter_all) |
7870d082 JB |
7972 | clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree, |
7973 | io_tree, done->start, bvec->bv_page, | |
7974 | btrfs_ino(BTRFS_I(inode)), 0); | |
8b110e39 MX |
7975 | end: |
7976 | complete(&done->done); | |
7977 | bio_put(bio); | |
7978 | } | |
7979 | ||
58efbc9f OS |
7980 | static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode, |
7981 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 7982 | { |
2dabb324 | 7983 | struct btrfs_fs_info *fs_info; |
17347cec LB |
7984 | struct bio_vec bvec; |
7985 | struct bvec_iter iter; | |
8b110e39 | 7986 | struct btrfs_retry_complete done; |
4b46fce2 | 7987 | u64 start; |
2dabb324 CR |
7988 | unsigned int pgoff; |
7989 | u32 sectorsize; | |
7990 | int nr_sectors; | |
58efbc9f OS |
7991 | blk_status_t ret; |
7992 | blk_status_t err = BLK_STS_OK; | |
4b46fce2 | 7993 | |
2dabb324 | 7994 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 7995 | sectorsize = fs_info->sectorsize; |
2dabb324 | 7996 | |
8b110e39 MX |
7997 | start = io_bio->logical; |
7998 | done.inode = inode; | |
17347cec | 7999 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 8000 | |
17347cec LB |
8001 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
8002 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
8003 | pgoff = bvec.bv_offset; | |
2dabb324 CR |
8004 | |
8005 | next_block_or_try_again: | |
8b110e39 MX |
8006 | done.uptodate = 0; |
8007 | done.start = start; | |
8008 | init_completion(&done.done); | |
8009 | ||
17347cec | 8010 | ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
2dabb324 CR |
8011 | pgoff, start, start + sectorsize - 1, |
8012 | io_bio->mirror_num, | |
8013 | btrfs_retry_endio_nocsum, &done); | |
629ebf4f LB |
8014 | if (ret) { |
8015 | err = ret; | |
8016 | goto next; | |
8017 | } | |
8b110e39 | 8018 | |
9c17f6cd | 8019 | wait_for_completion_io(&done.done); |
8b110e39 MX |
8020 | |
8021 | if (!done.uptodate) { | |
8022 | /* We might have another mirror, so try again */ | |
2dabb324 | 8023 | goto next_block_or_try_again; |
8b110e39 MX |
8024 | } |
8025 | ||
629ebf4f | 8026 | next: |
2dabb324 CR |
8027 | start += sectorsize; |
8028 | ||
97bf5a55 LB |
8029 | nr_sectors--; |
8030 | if (nr_sectors) { | |
2dabb324 | 8031 | pgoff += sectorsize; |
97bf5a55 | 8032 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8033 | goto next_block_or_try_again; |
8034 | } | |
8b110e39 MX |
8035 | } |
8036 | ||
629ebf4f | 8037 | return err; |
8b110e39 MX |
8038 | } |
8039 | ||
4246a0b6 | 8040 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
8041 | { |
8042 | struct btrfs_retry_complete *done = bio->bi_private; | |
8043 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7870d082 JB |
8044 | struct extent_io_tree *io_tree, *failure_tree; |
8045 | struct inode *inode = done->inode; | |
8b110e39 MX |
8046 | struct bio_vec *bvec; |
8047 | int uptodate; | |
8048 | int ret; | |
2b070cfe | 8049 | int i = 0; |
6dc4f100 | 8050 | struct bvec_iter_all iter_all; |
8b110e39 | 8051 | |
4e4cbee9 | 8052 | if (bio->bi_status) |
8b110e39 MX |
8053 | goto end; |
8054 | ||
8055 | uptodate = 1; | |
2dabb324 | 8056 | |
2dabb324 | 8057 | ASSERT(bio->bi_vcnt == 1); |
263663cd | 8058 | ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode)); |
2dabb324 | 8059 | |
7870d082 JB |
8060 | io_tree = &BTRFS_I(inode)->io_tree; |
8061 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8062 | ||
c09abff8 | 8063 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 8064 | bio_for_each_segment_all(bvec, bio, iter_all) { |
7870d082 JB |
8065 | ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page, |
8066 | bvec->bv_offset, done->start, | |
8067 | bvec->bv_len); | |
8b110e39 | 8068 | if (!ret) |
7870d082 JB |
8069 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
8070 | failure_tree, io_tree, done->start, | |
8071 | bvec->bv_page, | |
8072 | btrfs_ino(BTRFS_I(inode)), | |
8073 | bvec->bv_offset); | |
8b110e39 MX |
8074 | else |
8075 | uptodate = 0; | |
2b070cfe | 8076 | i++; |
8b110e39 MX |
8077 | } |
8078 | ||
8079 | done->uptodate = uptodate; | |
8080 | end: | |
8081 | complete(&done->done); | |
8082 | bio_put(bio); | |
8083 | } | |
8084 | ||
4e4cbee9 CH |
8085 | static blk_status_t __btrfs_subio_endio_read(struct inode *inode, |
8086 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 | 8087 | { |
2dabb324 | 8088 | struct btrfs_fs_info *fs_info; |
17347cec LB |
8089 | struct bio_vec bvec; |
8090 | struct bvec_iter iter; | |
8b110e39 MX |
8091 | struct btrfs_retry_complete done; |
8092 | u64 start; | |
8093 | u64 offset = 0; | |
2dabb324 CR |
8094 | u32 sectorsize; |
8095 | int nr_sectors; | |
8096 | unsigned int pgoff; | |
8097 | int csum_pos; | |
ef7cdac1 | 8098 | bool uptodate = (err == 0); |
8b110e39 | 8099 | int ret; |
58efbc9f | 8100 | blk_status_t status; |
dc380aea | 8101 | |
2dabb324 | 8102 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8103 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8104 | |
58efbc9f | 8105 | err = BLK_STS_OK; |
c1dc0896 | 8106 | start = io_bio->logical; |
8b110e39 | 8107 | done.inode = inode; |
17347cec | 8108 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 8109 | |
17347cec LB |
8110 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
8111 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
2dabb324 | 8112 | |
17347cec | 8113 | pgoff = bvec.bv_offset; |
2dabb324 | 8114 | next_block: |
ef7cdac1 LB |
8115 | if (uptodate) { |
8116 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
8117 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
8118 | bvec.bv_page, pgoff, start, sectorsize); | |
8119 | if (likely(!ret)) | |
8120 | goto next; | |
8121 | } | |
8b110e39 MX |
8122 | try_again: |
8123 | done.uptodate = 0; | |
8124 | done.start = start; | |
8125 | init_completion(&done.done); | |
8126 | ||
58efbc9f OS |
8127 | status = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
8128 | pgoff, start, start + sectorsize - 1, | |
8129 | io_bio->mirror_num, btrfs_retry_endio, | |
8130 | &done); | |
8131 | if (status) { | |
8132 | err = status; | |
8b110e39 MX |
8133 | goto next; |
8134 | } | |
8135 | ||
9c17f6cd | 8136 | wait_for_completion_io(&done.done); |
8b110e39 MX |
8137 | |
8138 | if (!done.uptodate) { | |
8139 | /* We might have another mirror, so try again */ | |
8140 | goto try_again; | |
8141 | } | |
8142 | next: | |
2dabb324 CR |
8143 | offset += sectorsize; |
8144 | start += sectorsize; | |
8145 | ||
8146 | ASSERT(nr_sectors); | |
8147 | ||
97bf5a55 LB |
8148 | nr_sectors--; |
8149 | if (nr_sectors) { | |
2dabb324 | 8150 | pgoff += sectorsize; |
97bf5a55 | 8151 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8152 | goto next_block; |
8153 | } | |
2c30c71b | 8154 | } |
c1dc0896 MX |
8155 | |
8156 | return err; | |
8157 | } | |
8158 | ||
4e4cbee9 CH |
8159 | static blk_status_t btrfs_subio_endio_read(struct inode *inode, |
8160 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 MX |
8161 | { |
8162 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8163 | ||
8164 | if (skip_csum) { | |
8165 | if (unlikely(err)) | |
8166 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8167 | else | |
58efbc9f | 8168 | return BLK_STS_OK; |
8b110e39 MX |
8169 | } else { |
8170 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8171 | } | |
8172 | } | |
8173 | ||
4246a0b6 | 8174 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8175 | { |
8176 | struct btrfs_dio_private *dip = bio->bi_private; | |
8177 | struct inode *inode = dip->inode; | |
8178 | struct bio *dio_bio; | |
8179 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4e4cbee9 | 8180 | blk_status_t err = bio->bi_status; |
c1dc0896 | 8181 | |
99c4e3b9 | 8182 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8b110e39 | 8183 | err = btrfs_subio_endio_read(inode, io_bio, err); |
c1dc0896 | 8184 | |
4b46fce2 | 8185 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8186 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8187 | dio_bio = dip->dio_bio; |
4b46fce2 | 8188 | |
4b46fce2 | 8189 | kfree(dip); |
c0da7aa1 | 8190 | |
99c4e3b9 | 8191 | dio_bio->bi_status = err; |
4055351c | 8192 | dio_end_io(dio_bio); |
b3a0dd50 | 8193 | btrfs_io_bio_free_csum(io_bio); |
9be3395b | 8194 | bio_put(bio); |
4b46fce2 JB |
8195 | } |
8196 | ||
52427260 QW |
8197 | static void __endio_write_update_ordered(struct inode *inode, |
8198 | const u64 offset, const u64 bytes, | |
8199 | const bool uptodate) | |
4b46fce2 | 8200 | { |
0b246afa | 8201 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 8202 | struct btrfs_ordered_extent *ordered = NULL; |
52427260 QW |
8203 | struct btrfs_workqueue *wq; |
8204 | btrfs_work_func_t func; | |
14543774 FM |
8205 | u64 ordered_offset = offset; |
8206 | u64 ordered_bytes = bytes; | |
67c003f9 | 8207 | u64 last_offset; |
4b46fce2 | 8208 | |
52427260 QW |
8209 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
8210 | wq = fs_info->endio_freespace_worker; | |
8211 | func = btrfs_freespace_write_helper; | |
8212 | } else { | |
8213 | wq = fs_info->endio_write_workers; | |
8214 | func = btrfs_endio_write_helper; | |
8215 | } | |
8216 | ||
b25f0d00 NB |
8217 | while (ordered_offset < offset + bytes) { |
8218 | last_offset = ordered_offset; | |
8219 | if (btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
8220 | &ordered_offset, | |
8221 | ordered_bytes, | |
8222 | uptodate)) { | |
8223 | btrfs_init_work(&ordered->work, func, | |
8224 | finish_ordered_fn, | |
8225 | NULL, NULL); | |
8226 | btrfs_queue_work(wq, &ordered->work); | |
8227 | } | |
8228 | /* | |
8229 | * If btrfs_dec_test_ordered_pending does not find any ordered | |
8230 | * extent in the range, we can exit. | |
8231 | */ | |
8232 | if (ordered_offset == last_offset) | |
8233 | return; | |
8234 | /* | |
8235 | * Our bio might span multiple ordered extents. In this case | |
52042d8e | 8236 | * we keep going until we have accounted the whole dio. |
b25f0d00 NB |
8237 | */ |
8238 | if (ordered_offset < offset + bytes) { | |
8239 | ordered_bytes = offset + bytes - ordered_offset; | |
8240 | ordered = NULL; | |
8241 | } | |
163cf09c | 8242 | } |
14543774 FM |
8243 | } |
8244 | ||
8245 | static void btrfs_endio_direct_write(struct bio *bio) | |
8246 | { | |
8247 | struct btrfs_dio_private *dip = bio->bi_private; | |
8248 | struct bio *dio_bio = dip->dio_bio; | |
8249 | ||
52427260 | 8250 | __endio_write_update_ordered(dip->inode, dip->logical_offset, |
4e4cbee9 | 8251 | dip->bytes, !bio->bi_status); |
4b46fce2 | 8252 | |
4b46fce2 | 8253 | kfree(dip); |
c0da7aa1 | 8254 | |
4e4cbee9 | 8255 | dio_bio->bi_status = bio->bi_status; |
4055351c | 8256 | dio_end_io(dio_bio); |
9be3395b | 8257 | bio_put(bio); |
4b46fce2 JB |
8258 | } |
8259 | ||
d0ee3934 | 8260 | static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data, |
d0779291 | 8261 | struct bio *bio, u64 offset) |
eaf25d93 | 8262 | { |
c6100a4b | 8263 | struct inode *inode = private_data; |
4e4cbee9 | 8264 | blk_status_t ret; |
2ff7e61e | 8265 | ret = btrfs_csum_one_bio(inode, bio, offset, 1); |
79787eaa | 8266 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8267 | return 0; |
8268 | } | |
8269 | ||
4246a0b6 | 8270 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8271 | { |
8272 | struct btrfs_dio_private *dip = bio->bi_private; | |
4e4cbee9 | 8273 | blk_status_t err = bio->bi_status; |
e65e1535 | 8274 | |
8b110e39 MX |
8275 | if (err) |
8276 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 8277 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 DS |
8278 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
8279 | bio->bi_opf, | |
8b110e39 MX |
8280 | (unsigned long long)bio->bi_iter.bi_sector, |
8281 | bio->bi_iter.bi_size, err); | |
8282 | ||
8283 | if (dip->subio_endio) | |
8284 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8285 | |
8286 | if (err) { | |
e65e1535 | 8287 | /* |
de224b7c NB |
8288 | * We want to perceive the errors flag being set before |
8289 | * decrementing the reference count. We don't need a barrier | |
8290 | * since atomic operations with a return value are fully | |
8291 | * ordered as per atomic_t.txt | |
e65e1535 | 8292 | */ |
de224b7c | 8293 | dip->errors = 1; |
e65e1535 MX |
8294 | } |
8295 | ||
8296 | /* if there are more bios still pending for this dio, just exit */ | |
8297 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8298 | goto out; | |
8299 | ||
9be3395b | 8300 | if (dip->errors) { |
e65e1535 | 8301 | bio_io_error(dip->orig_bio); |
9be3395b | 8302 | } else { |
2dbe0c77 | 8303 | dip->dio_bio->bi_status = BLK_STS_OK; |
4246a0b6 | 8304 | bio_endio(dip->orig_bio); |
e65e1535 MX |
8305 | } |
8306 | out: | |
8307 | bio_put(bio); | |
8308 | } | |
8309 | ||
4e4cbee9 | 8310 | static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode, |
c1dc0896 MX |
8311 | struct btrfs_dio_private *dip, |
8312 | struct bio *bio, | |
8313 | u64 file_offset) | |
8314 | { | |
8315 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8316 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
4e4cbee9 | 8317 | blk_status_t ret; |
c1dc0896 MX |
8318 | |
8319 | /* | |
8320 | * We load all the csum data we need when we submit | |
8321 | * the first bio to reduce the csum tree search and | |
8322 | * contention. | |
8323 | */ | |
8324 | if (dip->logical_offset == file_offset) { | |
2ff7e61e | 8325 | ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio, |
c1dc0896 MX |
8326 | file_offset); |
8327 | if (ret) | |
8328 | return ret; | |
8329 | } | |
8330 | ||
8331 | if (bio == dip->orig_bio) | |
8332 | return 0; | |
8333 | ||
8334 | file_offset -= dip->logical_offset; | |
8335 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8336 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8337 | ||
8338 | return 0; | |
8339 | } | |
8340 | ||
d0ee3934 DS |
8341 | static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio, |
8342 | struct inode *inode, u64 file_offset, int async_submit) | |
e65e1535 | 8343 | { |
0b246afa | 8344 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 8345 | struct btrfs_dio_private *dip = bio->bi_private; |
37226b21 | 8346 | bool write = bio_op(bio) == REQ_OP_WRITE; |
4e4cbee9 | 8347 | blk_status_t ret; |
e65e1535 | 8348 | |
4c274bc6 | 8349 | /* Check btrfs_submit_bio_hook() for rules about async submit. */ |
b812ce28 JB |
8350 | if (async_submit) |
8351 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8352 | ||
5fd02043 | 8353 | if (!write) { |
0b246afa | 8354 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
8355 | if (ret) |
8356 | goto err; | |
8357 | } | |
e65e1535 | 8358 | |
e6961cac | 8359 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
8360 | goto map; |
8361 | ||
8362 | if (write && async_submit) { | |
c6100a4b JB |
8363 | ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0, |
8364 | file_offset, inode, | |
e288c080 | 8365 | btrfs_submit_bio_start_direct_io); |
e65e1535 | 8366 | goto err; |
1ae39938 JB |
8367 | } else if (write) { |
8368 | /* | |
8369 | * If we aren't doing async submit, calculate the csum of the | |
8370 | * bio now. | |
8371 | */ | |
2ff7e61e | 8372 | ret = btrfs_csum_one_bio(inode, bio, file_offset, 1); |
1ae39938 JB |
8373 | if (ret) |
8374 | goto err; | |
23ea8e5a | 8375 | } else { |
2ff7e61e | 8376 | ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio, |
c1dc0896 | 8377 | file_offset); |
c2db1073 TI |
8378 | if (ret) |
8379 | goto err; | |
8380 | } | |
1ae39938 | 8381 | map: |
9b4a9b28 | 8382 | ret = btrfs_map_bio(fs_info, bio, 0, 0); |
e65e1535 | 8383 | err: |
e65e1535 MX |
8384 | return ret; |
8385 | } | |
8386 | ||
e6961cac | 8387 | static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip) |
e65e1535 MX |
8388 | { |
8389 | struct inode *inode = dip->inode; | |
0b246afa | 8390 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e65e1535 MX |
8391 | struct bio *bio; |
8392 | struct bio *orig_bio = dip->orig_bio; | |
4f024f37 | 8393 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 | 8394 | u64 file_offset = dip->logical_offset; |
1ae39938 | 8395 | int async_submit = 0; |
725130ba LB |
8396 | u64 submit_len; |
8397 | int clone_offset = 0; | |
8398 | int clone_len; | |
5f4dc8fc | 8399 | int ret; |
58efbc9f | 8400 | blk_status_t status; |
89b798ad | 8401 | struct btrfs_io_geometry geom; |
e65e1535 | 8402 | |
89b798ad NB |
8403 | submit_len = orig_bio->bi_iter.bi_size; |
8404 | ret = btrfs_get_io_geometry(fs_info, btrfs_op(orig_bio), | |
8405 | start_sector << 9, submit_len, &geom); | |
7a5c3c9b | 8406 | if (ret) |
e65e1535 | 8407 | return -EIO; |
facc8a22 | 8408 | |
89b798ad | 8409 | if (geom.len >= submit_len) { |
02f57c7a | 8410 | bio = orig_bio; |
c1dc0896 | 8411 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8412 | goto submit; |
8413 | } | |
8414 | ||
53b381b3 | 8415 | /* async crcs make it difficult to collect full stripe writes. */ |
1b86826d | 8416 | if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8417 | async_submit = 0; |
8418 | else | |
8419 | async_submit = 1; | |
8420 | ||
725130ba | 8421 | /* bio split */ |
89b798ad | 8422 | ASSERT(geom.len <= INT_MAX); |
02f57c7a | 8423 | atomic_inc(&dip->pending_bios); |
3c91ee69 | 8424 | do { |
89b798ad | 8425 | clone_len = min_t(int, submit_len, geom.len); |
02f57c7a | 8426 | |
725130ba LB |
8427 | /* |
8428 | * This will never fail as it's passing GPF_NOFS and | |
8429 | * the allocation is backed by btrfs_bioset. | |
8430 | */ | |
e477094f | 8431 | bio = btrfs_bio_clone_partial(orig_bio, clone_offset, |
725130ba LB |
8432 | clone_len); |
8433 | bio->bi_private = dip; | |
8434 | bio->bi_end_io = btrfs_end_dio_bio; | |
8435 | btrfs_io_bio(bio)->logical = file_offset; | |
8436 | ||
8437 | ASSERT(submit_len >= clone_len); | |
8438 | submit_len -= clone_len; | |
8439 | if (submit_len == 0) | |
8440 | break; | |
e65e1535 | 8441 | |
725130ba LB |
8442 | /* |
8443 | * Increase the count before we submit the bio so we know | |
8444 | * the end IO handler won't happen before we increase the | |
8445 | * count. Otherwise, the dip might get freed before we're | |
8446 | * done setting it up. | |
8447 | */ | |
8448 | atomic_inc(&dip->pending_bios); | |
e65e1535 | 8449 | |
d0ee3934 | 8450 | status = btrfs_submit_dio_bio(bio, inode, file_offset, |
58efbc9f OS |
8451 | async_submit); |
8452 | if (status) { | |
725130ba LB |
8453 | bio_put(bio); |
8454 | atomic_dec(&dip->pending_bios); | |
8455 | goto out_err; | |
8456 | } | |
e65e1535 | 8457 | |
725130ba LB |
8458 | clone_offset += clone_len; |
8459 | start_sector += clone_len >> 9; | |
8460 | file_offset += clone_len; | |
5f4dc8fc | 8461 | |
89b798ad NB |
8462 | ret = btrfs_get_io_geometry(fs_info, btrfs_op(orig_bio), |
8463 | start_sector << 9, submit_len, &geom); | |
725130ba LB |
8464 | if (ret) |
8465 | goto out_err; | |
3c91ee69 | 8466 | } while (submit_len > 0); |
e65e1535 | 8467 | |
02f57c7a | 8468 | submit: |
d0ee3934 | 8469 | status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit); |
58efbc9f | 8470 | if (!status) |
e65e1535 MX |
8471 | return 0; |
8472 | ||
8473 | bio_put(bio); | |
8474 | out_err: | |
8475 | dip->errors = 1; | |
8476 | /* | |
de224b7c NB |
8477 | * Before atomic variable goto zero, we must make sure dip->errors is |
8478 | * perceived to be set. This ordering is ensured by the fact that an | |
8479 | * atomic operations with a return value are fully ordered as per | |
8480 | * atomic_t.txt | |
e65e1535 | 8481 | */ |
e65e1535 MX |
8482 | if (atomic_dec_and_test(&dip->pending_bios)) |
8483 | bio_io_error(dip->orig_bio); | |
8484 | ||
8485 | /* bio_end_io() will handle error, so we needn't return it */ | |
8486 | return 0; | |
8487 | } | |
8488 | ||
8a4c1e42 MC |
8489 | static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode, |
8490 | loff_t file_offset) | |
4b46fce2 | 8491 | { |
61de718f | 8492 | struct btrfs_dio_private *dip = NULL; |
3892ac90 LB |
8493 | struct bio *bio = NULL; |
8494 | struct btrfs_io_bio *io_bio; | |
8a4c1e42 | 8495 | bool write = (bio_op(dio_bio) == REQ_OP_WRITE); |
4b46fce2 JB |
8496 | int ret = 0; |
8497 | ||
8b6c1d56 | 8498 | bio = btrfs_bio_clone(dio_bio); |
9be3395b | 8499 | |
c1dc0896 | 8500 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8501 | if (!dip) { |
8502 | ret = -ENOMEM; | |
61de718f | 8503 | goto free_ordered; |
4b46fce2 | 8504 | } |
4b46fce2 | 8505 | |
9be3395b | 8506 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8507 | dip->inode = inode; |
8508 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8509 | dip->bytes = dio_bio->bi_iter.bi_size; |
8510 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
3892ac90 LB |
8511 | bio->bi_private = dip; |
8512 | dip->orig_bio = bio; | |
9be3395b | 8513 | dip->dio_bio = dio_bio; |
e65e1535 | 8514 | atomic_set(&dip->pending_bios, 0); |
3892ac90 LB |
8515 | io_bio = btrfs_io_bio(bio); |
8516 | io_bio->logical = file_offset; | |
4b46fce2 | 8517 | |
c1dc0896 | 8518 | if (write) { |
3892ac90 | 8519 | bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8520 | } else { |
3892ac90 | 8521 | bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8522 | dip->subio_endio = btrfs_subio_endio_read; |
8523 | } | |
4b46fce2 | 8524 | |
f28a4928 FM |
8525 | /* |
8526 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8527 | * even if we fail to submit a bio, because in such case we do the | |
8528 | * corresponding error handling below and it must not be done a second | |
8529 | * time by btrfs_direct_IO(). | |
8530 | */ | |
8531 | if (write) { | |
8532 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8533 | ||
8534 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8535 | dip->bytes; | |
8536 | dio_data->unsubmitted_oe_range_start = | |
8537 | dio_data->unsubmitted_oe_range_end; | |
8538 | } | |
8539 | ||
e6961cac | 8540 | ret = btrfs_submit_direct_hook(dip); |
e65e1535 | 8541 | if (!ret) |
eaf25d93 | 8542 | return; |
9be3395b | 8543 | |
b3a0dd50 | 8544 | btrfs_io_bio_free_csum(io_bio); |
9be3395b | 8545 | |
4b46fce2 JB |
8546 | free_ordered: |
8547 | /* | |
61de718f FM |
8548 | * If we arrived here it means either we failed to submit the dip |
8549 | * or we either failed to clone the dio_bio or failed to allocate the | |
8550 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8551 | * call bio_endio against our io_bio so that we get proper resource | |
8552 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8553 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8554 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8555 | */ |
3892ac90 | 8556 | if (bio && dip) { |
054ec2f6 | 8557 | bio_io_error(bio); |
61de718f | 8558 | /* |
3892ac90 | 8559 | * The end io callbacks free our dip, do the final put on bio |
61de718f FM |
8560 | * and all the cleanup and final put for dio_bio (through |
8561 | * dio_end_io()). | |
8562 | */ | |
8563 | dip = NULL; | |
3892ac90 | 8564 | bio = NULL; |
61de718f | 8565 | } else { |
14543774 | 8566 | if (write) |
52427260 | 8567 | __endio_write_update_ordered(inode, |
14543774 FM |
8568 | file_offset, |
8569 | dio_bio->bi_iter.bi_size, | |
52427260 | 8570 | false); |
14543774 | 8571 | else |
61de718f FM |
8572 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8573 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8574 | |
4e4cbee9 | 8575 | dio_bio->bi_status = BLK_STS_IOERR; |
61de718f FM |
8576 | /* |
8577 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8578 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8579 | */ | |
4055351c | 8580 | dio_end_io(dio_bio); |
4b46fce2 | 8581 | } |
3892ac90 LB |
8582 | if (bio) |
8583 | bio_put(bio); | |
61de718f | 8584 | kfree(dip); |
4b46fce2 JB |
8585 | } |
8586 | ||
2ff7e61e | 8587 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
2ff7e61e | 8588 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
8589 | { |
8590 | int seg; | |
a1b75f7d | 8591 | int i; |
0b246afa | 8592 | unsigned int blocksize_mask = fs_info->sectorsize - 1; |
5a5f79b5 | 8593 | ssize_t retval = -EINVAL; |
5a5f79b5 CM |
8594 | |
8595 | if (offset & blocksize_mask) | |
8596 | goto out; | |
8597 | ||
28060d5d AV |
8598 | if (iov_iter_alignment(iter) & blocksize_mask) |
8599 | goto out; | |
a1b75f7d | 8600 | |
28060d5d | 8601 | /* If this is a write we don't need to check anymore */ |
cd27e455 | 8602 | if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter)) |
28060d5d AV |
8603 | return 0; |
8604 | /* | |
8605 | * Check to make sure we don't have duplicate iov_base's in this | |
8606 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8607 | * when reading back. | |
8608 | */ | |
8609 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8610 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8611 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8612 | goto out; |
8613 | } | |
5a5f79b5 CM |
8614 | } |
8615 | retval = 0; | |
8616 | out: | |
8617 | return retval; | |
8618 | } | |
eb838e73 | 8619 | |
c8b8e32d | 8620 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8621 | { |
4b46fce2 JB |
8622 | struct file *file = iocb->ki_filp; |
8623 | struct inode *inode = file->f_mapping->host; | |
0b246afa | 8624 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
50745b0a | 8625 | struct btrfs_dio_data dio_data = { 0 }; |
364ecf36 | 8626 | struct extent_changeset *data_reserved = NULL; |
c8b8e32d | 8627 | loff_t offset = iocb->ki_pos; |
0934856d | 8628 | size_t count = 0; |
2e60a51e | 8629 | int flags = 0; |
38851cc1 MX |
8630 | bool wakeup = true; |
8631 | bool relock = false; | |
0934856d | 8632 | ssize_t ret; |
4b46fce2 | 8633 | |
8c70c9f8 | 8634 | if (check_direct_IO(fs_info, iter, offset)) |
5a5f79b5 | 8635 | return 0; |
3f7c579c | 8636 | |
fe0f07d0 | 8637 | inode_dio_begin(inode); |
38851cc1 | 8638 | |
0e267c44 | 8639 | /* |
41bd9ca4 MX |
8640 | * The generic stuff only does filemap_write_and_wait_range, which |
8641 | * isn't enough if we've written compressed pages to this area, so | |
8642 | * we need to flush the dirty pages again to make absolutely sure | |
8643 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8644 | */ |
a6cbcd4a | 8645 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8646 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8647 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8648 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8649 | offset + count - 1); | |
0e267c44 | 8650 | |
6f673763 | 8651 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8652 | /* |
8653 | * If the write DIO is beyond the EOF, we need update | |
8654 | * the isize, but it is protected by i_mutex. So we can | |
8655 | * not unlock the i_mutex at this case. | |
8656 | */ | |
8657 | if (offset + count <= inode->i_size) { | |
4aaedfb0 | 8658 | dio_data.overwrite = 1; |
5955102c | 8659 | inode_unlock(inode); |
38851cc1 | 8660 | relock = true; |
edf064e7 GR |
8661 | } else if (iocb->ki_flags & IOCB_NOWAIT) { |
8662 | ret = -EAGAIN; | |
8663 | goto out; | |
38851cc1 | 8664 | } |
364ecf36 QW |
8665 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8666 | offset, count); | |
0934856d | 8667 | if (ret) |
38851cc1 | 8668 | goto out; |
e1cbbfa5 JB |
8669 | |
8670 | /* | |
8671 | * We need to know how many extents we reserved so that we can | |
8672 | * do the accounting properly if we go over the number we | |
8673 | * originally calculated. Abuse current->journal_info for this. | |
8674 | */ | |
da17066c | 8675 | dio_data.reserve = round_up(count, |
0b246afa | 8676 | fs_info->sectorsize); |
f28a4928 FM |
8677 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8678 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8679 | current->journal_info = &dio_data; |
97dcdea0 | 8680 | down_read(&BTRFS_I(inode)->dio_sem); |
ee39b432 DS |
8681 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8682 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8683 | inode_dio_end(inode); |
38851cc1 MX |
8684 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8685 | wakeup = false; | |
0934856d MX |
8686 | } |
8687 | ||
17f8c842 | 8688 | ret = __blockdev_direct_IO(iocb, inode, |
0b246afa | 8689 | fs_info->fs_devices->latest_bdev, |
c8b8e32d | 8690 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8691 | btrfs_submit_direct, flags); |
6f673763 | 8692 | if (iov_iter_rw(iter) == WRITE) { |
97dcdea0 | 8693 | up_read(&BTRFS_I(inode)->dio_sem); |
e1cbbfa5 | 8694 | current->journal_info = NULL; |
ddba1bfc | 8695 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8696 | if (dio_data.reserve) |
bc42bda2 | 8697 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 | 8698 | offset, dio_data.reserve, true); |
f28a4928 FM |
8699 | /* |
8700 | * On error we might have left some ordered extents | |
8701 | * without submitting corresponding bios for them, so | |
8702 | * cleanup them up to avoid other tasks getting them | |
8703 | * and waiting for them to complete forever. | |
8704 | */ | |
8705 | if (dio_data.unsubmitted_oe_range_start < | |
8706 | dio_data.unsubmitted_oe_range_end) | |
52427260 | 8707 | __endio_write_update_ordered(inode, |
f28a4928 FM |
8708 | dio_data.unsubmitted_oe_range_start, |
8709 | dio_data.unsubmitted_oe_range_end - | |
8710 | dio_data.unsubmitted_oe_range_start, | |
52427260 | 8711 | false); |
ddba1bfc | 8712 | } else if (ret >= 0 && (size_t)ret < count) |
bc42bda2 | 8713 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
8714 | offset, count - (size_t)ret, true); |
8715 | btrfs_delalloc_release_extents(BTRFS_I(inode), count, false); | |
0934856d | 8716 | } |
38851cc1 | 8717 | out: |
2e60a51e | 8718 | if (wakeup) |
fe0f07d0 | 8719 | inode_dio_end(inode); |
38851cc1 | 8720 | if (relock) |
5955102c | 8721 | inode_lock(inode); |
0934856d | 8722 | |
364ecf36 | 8723 | extent_changeset_free(data_reserved); |
0934856d | 8724 | return ret; |
16432985 CM |
8725 | } |
8726 | ||
05dadc09 TI |
8727 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8728 | ||
1506fcc8 YS |
8729 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8730 | __u64 start, __u64 len) | |
8731 | { | |
05dadc09 TI |
8732 | int ret; |
8733 | ||
8734 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8735 | if (ret) | |
8736 | return ret; | |
8737 | ||
2135fb9b | 8738 | return extent_fiemap(inode, fieinfo, start, len); |
1506fcc8 YS |
8739 | } |
8740 | ||
a52d9a80 | 8741 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8742 | { |
d1310b2e CM |
8743 | struct extent_io_tree *tree; |
8744 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8745 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8746 | } |
1832a6d5 | 8747 | |
a52d9a80 | 8748 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8749 | { |
be7bd730 JB |
8750 | struct inode *inode = page->mapping->host; |
8751 | int ret; | |
b888db2b CM |
8752 | |
8753 | if (current->flags & PF_MEMALLOC) { | |
8754 | redirty_page_for_writepage(wbc, page); | |
8755 | unlock_page(page); | |
8756 | return 0; | |
8757 | } | |
be7bd730 JB |
8758 | |
8759 | /* | |
8760 | * If we are under memory pressure we will call this directly from the | |
8761 | * VM, we need to make sure we have the inode referenced for the ordered | |
8762 | * extent. If not just return like we didn't do anything. | |
8763 | */ | |
8764 | if (!igrab(inode)) { | |
8765 | redirty_page_for_writepage(wbc, page); | |
8766 | return AOP_WRITEPAGE_ACTIVATE; | |
8767 | } | |
0a9b0e53 | 8768 | ret = extent_write_full_page(page, wbc); |
be7bd730 JB |
8769 | btrfs_add_delayed_iput(inode); |
8770 | return ret; | |
9ebefb18 CM |
8771 | } |
8772 | ||
48a3b636 ES |
8773 | static int btrfs_writepages(struct address_space *mapping, |
8774 | struct writeback_control *wbc) | |
b293f02e | 8775 | { |
8ae225a8 | 8776 | return extent_writepages(mapping, wbc); |
b293f02e CM |
8777 | } |
8778 | ||
3ab2fb5a CM |
8779 | static int |
8780 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8781 | struct list_head *pages, unsigned nr_pages) | |
8782 | { | |
2a3ff0ad | 8783 | return extent_readpages(mapping, pages, nr_pages); |
3ab2fb5a | 8784 | } |
2a3ff0ad | 8785 | |
e6dcd2dc | 8786 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8787 | { |
477a30ba | 8788 | int ret = try_release_extent_mapping(page, gfp_flags); |
a52d9a80 CM |
8789 | if (ret == 1) { |
8790 | ClearPagePrivate(page); | |
8791 | set_page_private(page, 0); | |
09cbfeaf | 8792 | put_page(page); |
39279cc3 | 8793 | } |
a52d9a80 | 8794 | return ret; |
39279cc3 CM |
8795 | } |
8796 | ||
e6dcd2dc CM |
8797 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8798 | { | |
98509cfc CM |
8799 | if (PageWriteback(page) || PageDirty(page)) |
8800 | return 0; | |
3ba7ab22 | 8801 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8802 | } |
8803 | ||
d47992f8 LC |
8804 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8805 | unsigned int length) | |
39279cc3 | 8806 | { |
5fd02043 | 8807 | struct inode *inode = page->mapping->host; |
d1310b2e | 8808 | struct extent_io_tree *tree; |
e6dcd2dc | 8809 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8810 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8811 | u64 page_start = page_offset(page); |
09cbfeaf | 8812 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8813 | u64 start; |
8814 | u64 end; | |
131e404a | 8815 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8816 | |
8b62b72b CM |
8817 | /* |
8818 | * we have the page locked, so new writeback can't start, | |
8819 | * and the dirty bit won't be cleared while we are here. | |
8820 | * | |
8821 | * Wait for IO on this page so that we can safely clear | |
8822 | * the PagePrivate2 bit and do ordered accounting | |
8823 | */ | |
e6dcd2dc | 8824 | wait_on_page_writeback(page); |
8b62b72b | 8825 | |
5fd02043 | 8826 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8827 | if (offset) { |
8828 | btrfs_releasepage(page, GFP_NOFS); | |
8829 | return; | |
8830 | } | |
131e404a FDBM |
8831 | |
8832 | if (!inode_evicting) | |
ff13db41 | 8833 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8834 | again: |
8835 | start = page_start; | |
a776c6fa | 8836 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
dbfdb6d1 | 8837 | page_end - start + 1); |
e6dcd2dc | 8838 | if (ordered) { |
dbfdb6d1 | 8839 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
8840 | /* |
8841 | * IO on this page will never be started, so we need | |
8842 | * to account for any ordered extents now | |
8843 | */ | |
131e404a | 8844 | if (!inode_evicting) |
dbfdb6d1 | 8845 | clear_extent_bit(tree, start, end, |
131e404a | 8846 | EXTENT_DIRTY | EXTENT_DELALLOC | |
a7e3b975 | 8847 | EXTENT_DELALLOC_NEW | |
131e404a | 8848 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
ae0f1625 | 8849 | EXTENT_DEFRAG, 1, 0, &cached_state); |
8b62b72b CM |
8850 | /* |
8851 | * whoever cleared the private bit is responsible | |
8852 | * for the finish_ordered_io | |
8853 | */ | |
77cef2ec JB |
8854 | if (TestClearPagePrivate2(page)) { |
8855 | struct btrfs_ordered_inode_tree *tree; | |
8856 | u64 new_len; | |
8857 | ||
8858 | tree = &BTRFS_I(inode)->ordered_tree; | |
8859 | ||
8860 | spin_lock_irq(&tree->lock); | |
8861 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8862 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8863 | if (new_len < ordered->truncated_len) |
8864 | ordered->truncated_len = new_len; | |
8865 | spin_unlock_irq(&tree->lock); | |
8866 | ||
8867 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
8868 | start, |
8869 | end - start + 1, 1)) | |
77cef2ec | 8870 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 8871 | } |
e6dcd2dc | 8872 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8873 | if (!inode_evicting) { |
8874 | cached_state = NULL; | |
dbfdb6d1 | 8875 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
8876 | &cached_state); |
8877 | } | |
dbfdb6d1 CR |
8878 | |
8879 | start = end + 1; | |
8880 | if (start < page_end) | |
8881 | goto again; | |
131e404a FDBM |
8882 | } |
8883 | ||
b9d0b389 QW |
8884 | /* |
8885 | * Qgroup reserved space handler | |
8886 | * Page here will be either | |
8887 | * 1) Already written to disk | |
8888 | * In this case, its reserved space is released from data rsv map | |
8889 | * and will be freed by delayed_ref handler finally. | |
8890 | * So even we call qgroup_free_data(), it won't decrease reserved | |
8891 | * space. | |
8892 | * 2) Not written to disk | |
0b34c261 GR |
8893 | * This means the reserved space should be freed here. However, |
8894 | * if a truncate invalidates the page (by clearing PageDirty) | |
8895 | * and the page is accounted for while allocating extent | |
8896 | * in btrfs_check_data_free_space() we let delayed_ref to | |
8897 | * free the entire extent. | |
b9d0b389 | 8898 | */ |
0b34c261 | 8899 | if (PageDirty(page)) |
bc42bda2 | 8900 | btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE); |
131e404a FDBM |
8901 | if (!inode_evicting) { |
8902 | clear_extent_bit(tree, page_start, page_end, | |
8903 | EXTENT_LOCKED | EXTENT_DIRTY | | |
a7e3b975 FM |
8904 | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
8905 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
ae0f1625 | 8906 | &cached_state); |
131e404a FDBM |
8907 | |
8908 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8909 | } |
e6dcd2dc | 8910 | |
4a096752 | 8911 | ClearPageChecked(page); |
9ad6b7bc | 8912 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8913 | ClearPagePrivate(page); |
8914 | set_page_private(page, 0); | |
09cbfeaf | 8915 | put_page(page); |
9ad6b7bc | 8916 | } |
39279cc3 CM |
8917 | } |
8918 | ||
9ebefb18 CM |
8919 | /* |
8920 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8921 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8922 | * be careful to check for EOF conditions here. We set the page up correctly | |
8923 | * for a written page which means we get ENOSPC checking when writing into | |
8924 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8925 | * support these features. | |
8926 | * | |
8927 | * We are not allowed to take the i_mutex here so we have to play games to | |
8928 | * protect against truncate races as the page could now be beyond EOF. Because | |
d1342aad OS |
8929 | * truncate_setsize() writes the inode size before removing pages, once we have |
8930 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
9ebefb18 CM |
8931 | * beyond EOF, then the page is guaranteed safe against truncation until we |
8932 | * unlock the page. | |
8933 | */ | |
a528a241 | 8934 | vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 8935 | { |
c2ec175c | 8936 | struct page *page = vmf->page; |
11bac800 | 8937 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 8938 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8939 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8940 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8941 | struct extent_state *cached_state = NULL; |
364ecf36 | 8942 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc CM |
8943 | char *kaddr; |
8944 | unsigned long zero_start; | |
9ebefb18 | 8945 | loff_t size; |
a528a241 SJ |
8946 | vm_fault_t ret; |
8947 | int ret2; | |
9998eb70 | 8948 | int reserved = 0; |
d0b7da88 | 8949 | u64 reserved_space; |
a52d9a80 | 8950 | u64 page_start; |
e6dcd2dc | 8951 | u64 page_end; |
d0b7da88 CR |
8952 | u64 end; |
8953 | ||
09cbfeaf | 8954 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8955 | |
b2b5ef5c | 8956 | sb_start_pagefault(inode->i_sb); |
df480633 | 8957 | page_start = page_offset(page); |
09cbfeaf | 8958 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8959 | end = page_end; |
df480633 | 8960 | |
d0b7da88 CR |
8961 | /* |
8962 | * Reserving delalloc space after obtaining the page lock can lead to | |
8963 | * deadlock. For example, if a dirty page is locked by this function | |
8964 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8965 | * dirty page write out, then the btrfs_writepage() function could | |
8966 | * end up waiting indefinitely to get a lock on the page currently | |
8967 | * being processed by btrfs_page_mkwrite() function. | |
8968 | */ | |
a528a241 | 8969 | ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
d0b7da88 | 8970 | reserved_space); |
a528a241 SJ |
8971 | if (!ret2) { |
8972 | ret2 = file_update_time(vmf->vma->vm_file); | |
9998eb70 CM |
8973 | reserved = 1; |
8974 | } | |
a528a241 SJ |
8975 | if (ret2) { |
8976 | ret = vmf_error(ret2); | |
9998eb70 CM |
8977 | if (reserved) |
8978 | goto out; | |
8979 | goto out_noreserve; | |
56a76f82 | 8980 | } |
1832a6d5 | 8981 | |
56a76f82 | 8982 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8983 | again: |
9ebefb18 | 8984 | lock_page(page); |
9ebefb18 | 8985 | size = i_size_read(inode); |
a52d9a80 | 8986 | |
9ebefb18 | 8987 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8988 | (page_start >= size)) { |
9ebefb18 CM |
8989 | /* page got truncated out from underneath us */ |
8990 | goto out_unlock; | |
8991 | } | |
e6dcd2dc CM |
8992 | wait_on_page_writeback(page); |
8993 | ||
ff13db41 | 8994 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
8995 | set_page_extent_mapped(page); |
8996 | ||
eb84ae03 CM |
8997 | /* |
8998 | * we can't set the delalloc bits if there are pending ordered | |
8999 | * extents. Drop our locks and wait for them to finish | |
9000 | */ | |
a776c6fa NB |
9001 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
9002 | PAGE_SIZE); | |
e6dcd2dc | 9003 | if (ordered) { |
2ac55d41 | 9004 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 9005 | &cached_state); |
e6dcd2dc | 9006 | unlock_page(page); |
eb84ae03 | 9007 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
9008 | btrfs_put_ordered_extent(ordered); |
9009 | goto again; | |
9010 | } | |
9011 | ||
09cbfeaf | 9012 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 9013 | reserved_space = round_up(size - page_start, |
0b246afa | 9014 | fs_info->sectorsize); |
09cbfeaf | 9015 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 9016 | end = page_start + reserved_space - 1; |
bc42bda2 | 9017 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
9018 | page_start, PAGE_SIZE - reserved_space, |
9019 | true); | |
d0b7da88 CR |
9020 | } |
9021 | } | |
9022 | ||
fbf19087 | 9023 | /* |
5416034f LB |
9024 | * page_mkwrite gets called when the page is firstly dirtied after it's |
9025 | * faulted in, but write(2) could also dirty a page and set delalloc | |
9026 | * bits, thus in this case for space account reason, we still need to | |
9027 | * clear any delalloc bits within this page range since we have to | |
9028 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 9029 | */ |
d0b7da88 | 9030 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
9031 | EXTENT_DIRTY | EXTENT_DELALLOC | |
9032 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 9033 | 0, 0, &cached_state); |
fbf19087 | 9034 | |
a528a241 | 9035 | ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0, |
330a5827 | 9036 | &cached_state); |
a528a241 | 9037 | if (ret2) { |
2ac55d41 | 9038 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 9039 | &cached_state); |
9ed74f2d JB |
9040 | ret = VM_FAULT_SIGBUS; |
9041 | goto out_unlock; | |
9042 | } | |
a528a241 | 9043 | ret2 = 0; |
9ebefb18 CM |
9044 | |
9045 | /* page is wholly or partially inside EOF */ | |
09cbfeaf | 9046 | if (page_start + PAGE_SIZE > size) |
7073017a | 9047 | zero_start = offset_in_page(size); |
9ebefb18 | 9048 | else |
09cbfeaf | 9049 | zero_start = PAGE_SIZE; |
9ebefb18 | 9050 | |
09cbfeaf | 9051 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 9052 | kaddr = kmap(page); |
09cbfeaf | 9053 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
9054 | flush_dcache_page(page); |
9055 | kunmap(page); | |
9056 | } | |
247e743c | 9057 | ClearPageChecked(page); |
e6dcd2dc | 9058 | set_page_dirty(page); |
50a9b214 | 9059 | SetPageUptodate(page); |
5a3f23d5 | 9060 | |
0b246afa | 9061 | BTRFS_I(inode)->last_trans = fs_info->generation; |
257c62e1 | 9062 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; |
46d8bc34 | 9063 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 9064 | |
e43bbe5e | 9065 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state); |
9ebefb18 | 9066 | |
a528a241 | 9067 | if (!ret2) { |
43b18595 | 9068 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true); |
b2b5ef5c | 9069 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 9070 | extent_changeset_free(data_reserved); |
50a9b214 | 9071 | return VM_FAULT_LOCKED; |
b2b5ef5c | 9072 | } |
717beb96 CM |
9073 | |
9074 | out_unlock: | |
9ebefb18 | 9075 | unlock_page(page); |
1832a6d5 | 9076 | out: |
43b18595 | 9077 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0)); |
bc42bda2 | 9078 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
43b18595 | 9079 | reserved_space, (ret != 0)); |
9998eb70 | 9080 | out_noreserve: |
b2b5ef5c | 9081 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 9082 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
9083 | return ret; |
9084 | } | |
9085 | ||
213e8c55 | 9086 | static int btrfs_truncate(struct inode *inode, bool skip_writeback) |
39279cc3 | 9087 | { |
0b246afa | 9088 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 9089 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 9090 | struct btrfs_block_rsv *rsv; |
ad7e1a74 | 9091 | int ret; |
39279cc3 | 9092 | struct btrfs_trans_handle *trans; |
0b246afa | 9093 | u64 mask = fs_info->sectorsize - 1; |
2bd36e7b | 9094 | u64 min_size = btrfs_calc_metadata_size(fs_info, 1); |
39279cc3 | 9095 | |
213e8c55 FM |
9096 | if (!skip_writeback) { |
9097 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), | |
9098 | (u64)-1); | |
9099 | if (ret) | |
9100 | return ret; | |
9101 | } | |
39279cc3 | 9102 | |
fcb80c2a | 9103 | /* |
f7e9e8fc OS |
9104 | * Yes ladies and gentlemen, this is indeed ugly. We have a couple of |
9105 | * things going on here: | |
fcb80c2a | 9106 | * |
f7e9e8fc | 9107 | * 1) We need to reserve space to update our inode. |
fcb80c2a | 9108 | * |
f7e9e8fc | 9109 | * 2) We need to have something to cache all the space that is going to |
fcb80c2a JB |
9110 | * be free'd up by the truncate operation, but also have some slack |
9111 | * space reserved in case it uses space during the truncate (thank you | |
9112 | * very much snapshotting). | |
9113 | * | |
f7e9e8fc | 9114 | * And we need these to be separate. The fact is we can use a lot of |
fcb80c2a | 9115 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 9116 | * we will use, so we need the truncate reservation to be separate so it |
f7e9e8fc OS |
9117 | * doesn't end up using space reserved for updating the inode. We also |
9118 | * need to be able to stop the transaction and start a new one, which | |
9119 | * means we need to be able to update the inode several times, and we | |
9120 | * have no idea of knowing how many times that will be, so we can't just | |
9121 | * reserve 1 item for the entirety of the operation, so that has to be | |
9122 | * done separately as well. | |
fcb80c2a JB |
9123 | * |
9124 | * So that leaves us with | |
9125 | * | |
f7e9e8fc | 9126 | * 1) rsv - for the truncate reservation, which we will steal from the |
fcb80c2a | 9127 | * transaction reservation. |
f7e9e8fc | 9128 | * 2) fs_info->trans_block_rsv - this will have 1 items worth left for |
fcb80c2a JB |
9129 | * updating the inode. |
9130 | */ | |
2ff7e61e | 9131 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
9132 | if (!rsv) |
9133 | return -ENOMEM; | |
4a338542 | 9134 | rsv->size = min_size; |
ca7e70f5 | 9135 | rsv->failfast = 1; |
f0cd846e | 9136 | |
907cbceb | 9137 | /* |
07127184 | 9138 | * 1 for the truncate slack space |
907cbceb JB |
9139 | * 1 for updating the inode. |
9140 | */ | |
f3fe820c | 9141 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a | 9142 | if (IS_ERR(trans)) { |
ad7e1a74 | 9143 | ret = PTR_ERR(trans); |
fcb80c2a JB |
9144 | goto out; |
9145 | } | |
f0cd846e | 9146 | |
907cbceb | 9147 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 9148 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
3a584174 | 9149 | min_size, false); |
fcb80c2a | 9150 | BUG_ON(ret); |
f0cd846e | 9151 | |
5dc562c5 JB |
9152 | /* |
9153 | * So if we truncate and then write and fsync we normally would just | |
9154 | * write the extents that changed, which is a problem if we need to | |
9155 | * first truncate that entire inode. So set this flag so we write out | |
9156 | * all of the extents in the inode to the sync log so we're completely | |
9157 | * safe. | |
9158 | */ | |
9159 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9160 | trans->block_rsv = rsv; |
907cbceb | 9161 | |
8082510e YZ |
9162 | while (1) { |
9163 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9164 | inode->i_size, | |
9165 | BTRFS_EXTENT_DATA_KEY); | |
ddfae63c | 9166 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 | 9167 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 9168 | break; |
39279cc3 | 9169 | |
8082510e | 9170 | ret = btrfs_update_inode(trans, root, inode); |
ad7e1a74 | 9171 | if (ret) |
3893e33b | 9172 | break; |
ca7e70f5 | 9173 | |
3a45bb20 | 9174 | btrfs_end_transaction(trans); |
2ff7e61e | 9175 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
9176 | |
9177 | trans = btrfs_start_transaction(root, 2); | |
9178 | if (IS_ERR(trans)) { | |
ad7e1a74 | 9179 | ret = PTR_ERR(trans); |
ca7e70f5 JB |
9180 | trans = NULL; |
9181 | break; | |
9182 | } | |
9183 | ||
47b5d646 | 9184 | btrfs_block_rsv_release(fs_info, rsv, -1); |
0b246afa | 9185 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
3a584174 | 9186 | rsv, min_size, false); |
ca7e70f5 JB |
9187 | BUG_ON(ret); /* shouldn't happen */ |
9188 | trans->block_rsv = rsv; | |
8082510e YZ |
9189 | } |
9190 | ||
ddfae63c JB |
9191 | /* |
9192 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
9193 | * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know | |
9194 | * we've truncated everything except the last little bit, and can do | |
9195 | * btrfs_truncate_block and then update the disk_i_size. | |
9196 | */ | |
9197 | if (ret == NEED_TRUNCATE_BLOCK) { | |
9198 | btrfs_end_transaction(trans); | |
9199 | btrfs_btree_balance_dirty(fs_info); | |
9200 | ||
9201 | ret = btrfs_truncate_block(inode, inode->i_size, 0, 0); | |
9202 | if (ret) | |
9203 | goto out; | |
9204 | trans = btrfs_start_transaction(root, 1); | |
9205 | if (IS_ERR(trans)) { | |
9206 | ret = PTR_ERR(trans); | |
9207 | goto out; | |
9208 | } | |
9209 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
9210 | } | |
9211 | ||
917c16b2 | 9212 | if (trans) { |
ad7e1a74 OS |
9213 | int ret2; |
9214 | ||
0b246afa | 9215 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 OS |
9216 | ret2 = btrfs_update_inode(trans, root, inode); |
9217 | if (ret2 && !ret) | |
9218 | ret = ret2; | |
7b128766 | 9219 | |
ad7e1a74 OS |
9220 | ret2 = btrfs_end_transaction(trans); |
9221 | if (ret2 && !ret) | |
9222 | ret = ret2; | |
2ff7e61e | 9223 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 9224 | } |
fcb80c2a | 9225 | out: |
2ff7e61e | 9226 | btrfs_free_block_rsv(fs_info, rsv); |
fcb80c2a | 9227 | |
ad7e1a74 | 9228 | return ret; |
39279cc3 CM |
9229 | } |
9230 | ||
d352ac68 CM |
9231 | /* |
9232 | * create a new subvolume directory/inode (helper for the ioctl). | |
9233 | */ | |
d2fb3437 | 9234 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9235 | struct btrfs_root *new_root, |
9236 | struct btrfs_root *parent_root, | |
9237 | u64 new_dirid) | |
39279cc3 | 9238 | { |
39279cc3 | 9239 | struct inode *inode; |
76dda93c | 9240 | int err; |
00e4e6b3 | 9241 | u64 index = 0; |
39279cc3 | 9242 | |
12fc9d09 FA |
9243 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9244 | new_dirid, new_dirid, | |
9245 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9246 | &index); | |
54aa1f4d | 9247 | if (IS_ERR(inode)) |
f46b5a66 | 9248 | return PTR_ERR(inode); |
39279cc3 CM |
9249 | inode->i_op = &btrfs_dir_inode_operations; |
9250 | inode->i_fop = &btrfs_dir_file_operations; | |
9251 | ||
bfe86848 | 9252 | set_nlink(inode, 1); |
6ef06d27 | 9253 | btrfs_i_size_write(BTRFS_I(inode), 0); |
b0d5d10f | 9254 | unlock_new_inode(inode); |
3b96362c | 9255 | |
63541927 FDBM |
9256 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9257 | if (err) | |
9258 | btrfs_err(new_root->fs_info, | |
351fd353 | 9259 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9260 | new_root->root_key.objectid, err); |
9261 | ||
76dda93c | 9262 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9263 | |
76dda93c | 9264 | iput(inode); |
ce598979 | 9265 | return err; |
39279cc3 CM |
9266 | } |
9267 | ||
39279cc3 CM |
9268 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9269 | { | |
69fe2d75 | 9270 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 9271 | struct btrfs_inode *ei; |
2ead6ae7 | 9272 | struct inode *inode; |
39279cc3 | 9273 | |
712e36c5 | 9274 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL); |
39279cc3 CM |
9275 | if (!ei) |
9276 | return NULL; | |
2ead6ae7 YZ |
9277 | |
9278 | ei->root = NULL; | |
2ead6ae7 | 9279 | ei->generation = 0; |
15ee9bc7 | 9280 | ei->last_trans = 0; |
257c62e1 | 9281 | ei->last_sub_trans = 0; |
e02119d5 | 9282 | ei->logged_trans = 0; |
2ead6ae7 | 9283 | ei->delalloc_bytes = 0; |
a7e3b975 | 9284 | ei->new_delalloc_bytes = 0; |
47059d93 | 9285 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9286 | ei->disk_i_size = 0; |
9287 | ei->flags = 0; | |
7709cde3 | 9288 | ei->csum_bytes = 0; |
2ead6ae7 | 9289 | ei->index_cnt = (u64)-1; |
67de1176 | 9290 | ei->dir_index = 0; |
2ead6ae7 | 9291 | ei->last_unlink_trans = 0; |
46d8bc34 | 9292 | ei->last_log_commit = 0; |
2ead6ae7 | 9293 | |
9e0baf60 JB |
9294 | spin_lock_init(&ei->lock); |
9295 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
9296 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
9297 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
9298 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 9299 | ei->runtime_flags = 0; |
b52aa8c9 | 9300 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 9301 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9302 | |
16cdcec7 MX |
9303 | ei->delayed_node = NULL; |
9304 | ||
9cc97d64 | 9305 | ei->i_otime.tv_sec = 0; |
9306 | ei->i_otime.tv_nsec = 0; | |
9307 | ||
2ead6ae7 | 9308 | inode = &ei->vfs_inode; |
a8067e02 | 9309 | extent_map_tree_init(&ei->extent_tree); |
43eb5f29 QW |
9310 | extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode); |
9311 | extent_io_tree_init(fs_info, &ei->io_failure_tree, | |
9312 | IO_TREE_INODE_IO_FAILURE, inode); | |
7b439738 DS |
9313 | ei->io_tree.track_uptodate = true; |
9314 | ei->io_failure_tree.track_uptodate = true; | |
b812ce28 | 9315 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9316 | mutex_init(&ei->log_mutex); |
f248679e | 9317 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9318 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9319 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9320 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9321 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 9322 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
9323 | |
9324 | return inode; | |
39279cc3 CM |
9325 | } |
9326 | ||
aaedb55b JB |
9327 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9328 | void btrfs_test_destroy_inode(struct inode *inode) | |
9329 | { | |
dcdbc059 | 9330 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
9331 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9332 | } | |
9333 | #endif | |
9334 | ||
26602cab | 9335 | void btrfs_free_inode(struct inode *inode) |
fa0d7e3d | 9336 | { |
fa0d7e3d NP |
9337 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9338 | } | |
9339 | ||
39279cc3 CM |
9340 | void btrfs_destroy_inode(struct inode *inode) |
9341 | { | |
0b246afa | 9342 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 9343 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9344 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9345 | ||
b3d9b7a3 | 9346 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9347 | WARN_ON(inode->i_data.nrpages); |
69fe2d75 JB |
9348 | WARN_ON(BTRFS_I(inode)->block_rsv.reserved); |
9349 | WARN_ON(BTRFS_I(inode)->block_rsv.size); | |
9e0baf60 | 9350 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7709cde3 | 9351 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
a7e3b975 | 9352 | WARN_ON(BTRFS_I(inode)->new_delalloc_bytes); |
7709cde3 | 9353 | WARN_ON(BTRFS_I(inode)->csum_bytes); |
47059d93 | 9354 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9355 | |
a6dbd429 JB |
9356 | /* |
9357 | * This can happen where we create an inode, but somebody else also | |
9358 | * created the same inode and we need to destroy the one we already | |
9359 | * created. | |
9360 | */ | |
9361 | if (!root) | |
26602cab | 9362 | return; |
a6dbd429 | 9363 | |
d397712b | 9364 | while (1) { |
e6dcd2dc CM |
9365 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9366 | if (!ordered) | |
9367 | break; | |
9368 | else { | |
0b246afa | 9369 | btrfs_err(fs_info, |
5d163e0e JM |
9370 | "found ordered extent %llu %llu on inode cleanup", |
9371 | ordered->file_offset, ordered->len); | |
e6dcd2dc CM |
9372 | btrfs_remove_ordered_extent(inode, ordered); |
9373 | btrfs_put_ordered_extent(ordered); | |
9374 | btrfs_put_ordered_extent(ordered); | |
9375 | } | |
9376 | } | |
56fa9d07 | 9377 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9378 | inode_tree_del(inode); |
dcdbc059 | 9379 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
39279cc3 CM |
9380 | } |
9381 | ||
45321ac5 | 9382 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9383 | { |
9384 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9385 | |
6379ef9f NA |
9386 | if (root == NULL) |
9387 | return 1; | |
9388 | ||
fa6ac876 | 9389 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9390 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9391 | return 1; |
76dda93c | 9392 | else |
45321ac5 | 9393 | return generic_drop_inode(inode); |
76dda93c YZ |
9394 | } |
9395 | ||
0ee0fda0 | 9396 | static void init_once(void *foo) |
39279cc3 CM |
9397 | { |
9398 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9399 | ||
9400 | inode_init_once(&ei->vfs_inode); | |
9401 | } | |
9402 | ||
e67c718b | 9403 | void __cold btrfs_destroy_cachep(void) |
39279cc3 | 9404 | { |
8c0a8537 KS |
9405 | /* |
9406 | * Make sure all delayed rcu free inodes are flushed before we | |
9407 | * destroy cache. | |
9408 | */ | |
9409 | rcu_barrier(); | |
5598e900 KM |
9410 | kmem_cache_destroy(btrfs_inode_cachep); |
9411 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
9412 | kmem_cache_destroy(btrfs_path_cachep); |
9413 | kmem_cache_destroy(btrfs_free_space_cachep); | |
3acd4850 | 9414 | kmem_cache_destroy(btrfs_free_space_bitmap_cachep); |
39279cc3 CM |
9415 | } |
9416 | ||
f5c29bd9 | 9417 | int __init btrfs_init_cachep(void) |
39279cc3 | 9418 | { |
837e1972 | 9419 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9420 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9421 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9422 | init_once); | |
39279cc3 CM |
9423 | if (!btrfs_inode_cachep) |
9424 | goto fail; | |
9601e3f6 | 9425 | |
837e1972 | 9426 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 9427 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 9428 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9429 | if (!btrfs_trans_handle_cachep) |
9430 | goto fail; | |
9601e3f6 | 9431 | |
837e1972 | 9432 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 9433 | sizeof(struct btrfs_path), 0, |
fba4b697 | 9434 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9435 | if (!btrfs_path_cachep) |
9436 | goto fail; | |
9601e3f6 | 9437 | |
837e1972 | 9438 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 9439 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 9440 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
9441 | if (!btrfs_free_space_cachep) |
9442 | goto fail; | |
9443 | ||
3acd4850 CL |
9444 | btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap", |
9445 | PAGE_SIZE, PAGE_SIZE, | |
9446 | SLAB_RED_ZONE, NULL); | |
9447 | if (!btrfs_free_space_bitmap_cachep) | |
9448 | goto fail; | |
9449 | ||
39279cc3 CM |
9450 | return 0; |
9451 | fail: | |
9452 | btrfs_destroy_cachep(); | |
9453 | return -ENOMEM; | |
9454 | } | |
9455 | ||
a528d35e DH |
9456 | static int btrfs_getattr(const struct path *path, struct kstat *stat, |
9457 | u32 request_mask, unsigned int flags) | |
39279cc3 | 9458 | { |
df0af1a5 | 9459 | u64 delalloc_bytes; |
a528d35e | 9460 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 9461 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 YS |
9462 | u32 bi_flags = BTRFS_I(inode)->flags; |
9463 | ||
9464 | stat->result_mask |= STATX_BTIME; | |
9465 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
9466 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
9467 | if (bi_flags & BTRFS_INODE_APPEND) | |
9468 | stat->attributes |= STATX_ATTR_APPEND; | |
9469 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
9470 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
9471 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
9472 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
9473 | if (bi_flags & BTRFS_INODE_NODUMP) | |
9474 | stat->attributes |= STATX_ATTR_NODUMP; | |
9475 | ||
9476 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
9477 | STATX_ATTR_COMPRESSED | | |
9478 | STATX_ATTR_IMMUTABLE | | |
9479 | STATX_ATTR_NODUMP); | |
fadc0d8b | 9480 | |
39279cc3 | 9481 | generic_fillattr(inode, stat); |
0ee5dc67 | 9482 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9483 | |
9484 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 9485 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
df0af1a5 | 9486 | spin_unlock(&BTRFS_I(inode)->lock); |
fadc0d8b | 9487 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9488 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9489 | return 0; |
9490 | } | |
9491 | ||
cdd1fedf DF |
9492 | static int btrfs_rename_exchange(struct inode *old_dir, |
9493 | struct dentry *old_dentry, | |
9494 | struct inode *new_dir, | |
9495 | struct dentry *new_dentry) | |
9496 | { | |
0b246afa | 9497 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
9498 | struct btrfs_trans_handle *trans; |
9499 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9500 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9501 | struct inode *new_inode = new_dentry->d_inode; | |
9502 | struct inode *old_inode = old_dentry->d_inode; | |
95582b00 | 9503 | struct timespec64 ctime = current_time(old_inode); |
cdd1fedf | 9504 | struct dentry *parent; |
4a0cc7ca NB |
9505 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9506 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9507 | u64 old_idx = 0; |
9508 | u64 new_idx = 0; | |
9509 | u64 root_objectid; | |
9510 | int ret; | |
86e8aa0e FM |
9511 | bool root_log_pinned = false; |
9512 | bool dest_log_pinned = false; | |
d4682ba0 FM |
9513 | struct btrfs_log_ctx ctx_root; |
9514 | struct btrfs_log_ctx ctx_dest; | |
9515 | bool sync_log_root = false; | |
9516 | bool sync_log_dest = false; | |
9517 | bool commit_transaction = false; | |
cdd1fedf DF |
9518 | |
9519 | /* we only allow rename subvolume link between subvolumes */ | |
9520 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
9521 | return -EXDEV; | |
9522 | ||
d4682ba0 FM |
9523 | btrfs_init_log_ctx(&ctx_root, old_inode); |
9524 | btrfs_init_log_ctx(&ctx_dest, new_inode); | |
9525 | ||
cdd1fedf DF |
9526 | /* close the race window with snapshot create/destroy ioctl */ |
9527 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9528 | down_read(&fs_info->subvol_sem); |
cdd1fedf | 9529 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9530 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9531 | |
9532 | /* | |
9533 | * We want to reserve the absolute worst case amount of items. So if | |
9534 | * both inodes are subvols and we need to unlink them then that would | |
9535 | * require 4 item modifications, but if they are both normal inodes it | |
9536 | * would require 5 item modifications, so we'll assume their normal | |
9537 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9538 | * should cover the worst case number of items we'll modify. | |
9539 | */ | |
9540 | trans = btrfs_start_transaction(root, 12); | |
9541 | if (IS_ERR(trans)) { | |
9542 | ret = PTR_ERR(trans); | |
9543 | goto out_notrans; | |
9544 | } | |
9545 | ||
9546 | /* | |
9547 | * We need to find a free sequence number both in the source and | |
9548 | * in the destination directory for the exchange. | |
9549 | */ | |
877574e2 | 9550 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
9551 | if (ret) |
9552 | goto out_fail; | |
877574e2 | 9553 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
9554 | if (ret) |
9555 | goto out_fail; | |
9556 | ||
9557 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9558 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9559 | ||
9560 | /* Reference for the source. */ | |
9561 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9562 | /* force full log commit if subvolume involved. */ | |
90787766 | 9563 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 9564 | } else { |
376e5a57 FM |
9565 | btrfs_pin_log_trans(root); |
9566 | root_log_pinned = true; | |
cdd1fedf DF |
9567 | ret = btrfs_insert_inode_ref(trans, dest, |
9568 | new_dentry->d_name.name, | |
9569 | new_dentry->d_name.len, | |
9570 | old_ino, | |
f85b7379 DS |
9571 | btrfs_ino(BTRFS_I(new_dir)), |
9572 | old_idx); | |
cdd1fedf DF |
9573 | if (ret) |
9574 | goto out_fail; | |
cdd1fedf DF |
9575 | } |
9576 | ||
9577 | /* And now for the dest. */ | |
9578 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9579 | /* force full log commit if subvolume involved. */ | |
90787766 | 9580 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 9581 | } else { |
376e5a57 FM |
9582 | btrfs_pin_log_trans(dest); |
9583 | dest_log_pinned = true; | |
cdd1fedf DF |
9584 | ret = btrfs_insert_inode_ref(trans, root, |
9585 | old_dentry->d_name.name, | |
9586 | old_dentry->d_name.len, | |
9587 | new_ino, | |
f85b7379 DS |
9588 | btrfs_ino(BTRFS_I(old_dir)), |
9589 | new_idx); | |
cdd1fedf DF |
9590 | if (ret) |
9591 | goto out_fail; | |
cdd1fedf DF |
9592 | } |
9593 | ||
9594 | /* Update inode version and ctime/mtime. */ | |
9595 | inode_inc_iversion(old_dir); | |
9596 | inode_inc_iversion(new_dir); | |
9597 | inode_inc_iversion(old_inode); | |
9598 | inode_inc_iversion(new_inode); | |
9599 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9600 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9601 | old_inode->i_ctime = ctime; | |
9602 | new_inode->i_ctime = ctime; | |
9603 | ||
9604 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9605 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9606 | BTRFS_I(old_inode), 1); | |
9607 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9608 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9609 | } |
9610 | ||
9611 | /* src is a subvolume */ | |
9612 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9613 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
401b3b19 | 9614 | ret = btrfs_unlink_subvol(trans, old_dir, root_objectid, |
cdd1fedf DF |
9615 | old_dentry->d_name.name, |
9616 | old_dentry->d_name.len); | |
9617 | } else { /* src is an inode */ | |
4ec5934e NB |
9618 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9619 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
9620 | old_dentry->d_name.name, |
9621 | old_dentry->d_name.len); | |
9622 | if (!ret) | |
9623 | ret = btrfs_update_inode(trans, root, old_inode); | |
9624 | } | |
9625 | if (ret) { | |
66642832 | 9626 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9627 | goto out_fail; |
9628 | } | |
9629 | ||
9630 | /* dest is a subvolume */ | |
9631 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9632 | root_objectid = BTRFS_I(new_inode)->root->root_key.objectid; | |
401b3b19 | 9633 | ret = btrfs_unlink_subvol(trans, new_dir, root_objectid, |
cdd1fedf DF |
9634 | new_dentry->d_name.name, |
9635 | new_dentry->d_name.len); | |
9636 | } else { /* dest is an inode */ | |
4ec5934e NB |
9637 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9638 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
9639 | new_dentry->d_name.name, |
9640 | new_dentry->d_name.len); | |
9641 | if (!ret) | |
9642 | ret = btrfs_update_inode(trans, dest, new_inode); | |
9643 | } | |
9644 | if (ret) { | |
66642832 | 9645 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9646 | goto out_fail; |
9647 | } | |
9648 | ||
db0a669f | 9649 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9650 | new_dentry->d_name.name, |
9651 | new_dentry->d_name.len, 0, old_idx); | |
9652 | if (ret) { | |
66642832 | 9653 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9654 | goto out_fail; |
9655 | } | |
9656 | ||
db0a669f | 9657 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9658 | old_dentry->d_name.name, |
9659 | old_dentry->d_name.len, 0, new_idx); | |
9660 | if (ret) { | |
66642832 | 9661 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9662 | goto out_fail; |
9663 | } | |
9664 | ||
9665 | if (old_inode->i_nlink == 1) | |
9666 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9667 | if (new_inode->i_nlink == 1) | |
9668 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9669 | ||
86e8aa0e | 9670 | if (root_log_pinned) { |
cdd1fedf | 9671 | parent = new_dentry->d_parent; |
d4682ba0 FM |
9672 | ret = btrfs_log_new_name(trans, BTRFS_I(old_inode), |
9673 | BTRFS_I(old_dir), parent, | |
9674 | false, &ctx_root); | |
9675 | if (ret == BTRFS_NEED_LOG_SYNC) | |
9676 | sync_log_root = true; | |
9677 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
9678 | commit_transaction = true; | |
9679 | ret = 0; | |
cdd1fedf | 9680 | btrfs_end_log_trans(root); |
86e8aa0e | 9681 | root_log_pinned = false; |
cdd1fedf | 9682 | } |
86e8aa0e | 9683 | if (dest_log_pinned) { |
d4682ba0 FM |
9684 | if (!commit_transaction) { |
9685 | parent = old_dentry->d_parent; | |
9686 | ret = btrfs_log_new_name(trans, BTRFS_I(new_inode), | |
9687 | BTRFS_I(new_dir), parent, | |
9688 | false, &ctx_dest); | |
9689 | if (ret == BTRFS_NEED_LOG_SYNC) | |
9690 | sync_log_dest = true; | |
9691 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
9692 | commit_transaction = true; | |
9693 | ret = 0; | |
9694 | } | |
cdd1fedf | 9695 | btrfs_end_log_trans(dest); |
86e8aa0e | 9696 | dest_log_pinned = false; |
cdd1fedf DF |
9697 | } |
9698 | out_fail: | |
86e8aa0e FM |
9699 | /* |
9700 | * If we have pinned a log and an error happened, we unpin tasks | |
9701 | * trying to sync the log and force them to fallback to a transaction | |
9702 | * commit if the log currently contains any of the inodes involved in | |
9703 | * this rename operation (to ensure we do not persist a log with an | |
9704 | * inconsistent state for any of these inodes or leading to any | |
9705 | * inconsistencies when replayed). If the transaction was aborted, the | |
9706 | * abortion reason is propagated to userspace when attempting to commit | |
9707 | * the transaction. If the log does not contain any of these inodes, we | |
9708 | * allow the tasks to sync it. | |
9709 | */ | |
9710 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
9711 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9712 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9713 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 9714 | (new_inode && |
0f8939b8 | 9715 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
90787766 | 9716 | btrfs_set_log_full_commit(trans); |
86e8aa0e FM |
9717 | |
9718 | if (root_log_pinned) { | |
9719 | btrfs_end_log_trans(root); | |
9720 | root_log_pinned = false; | |
9721 | } | |
9722 | if (dest_log_pinned) { | |
9723 | btrfs_end_log_trans(dest); | |
9724 | dest_log_pinned = false; | |
9725 | } | |
9726 | } | |
d4682ba0 FM |
9727 | if (!ret && sync_log_root && !commit_transaction) { |
9728 | ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, | |
9729 | &ctx_root); | |
9730 | if (ret) | |
9731 | commit_transaction = true; | |
9732 | } | |
9733 | if (!ret && sync_log_dest && !commit_transaction) { | |
9734 | ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root, | |
9735 | &ctx_dest); | |
9736 | if (ret) | |
9737 | commit_transaction = true; | |
9738 | } | |
9739 | if (commit_transaction) { | |
9740 | ret = btrfs_commit_transaction(trans); | |
9741 | } else { | |
9742 | int ret2; | |
9743 | ||
9744 | ret2 = btrfs_end_transaction(trans); | |
9745 | ret = ret ? ret : ret2; | |
9746 | } | |
cdd1fedf DF |
9747 | out_notrans: |
9748 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9749 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 9750 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9751 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9752 | |
9753 | return ret; | |
9754 | } | |
9755 | ||
9756 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9757 | struct btrfs_root *root, | |
9758 | struct inode *dir, | |
9759 | struct dentry *dentry) | |
9760 | { | |
9761 | int ret; | |
9762 | struct inode *inode; | |
9763 | u64 objectid; | |
9764 | u64 index; | |
9765 | ||
9766 | ret = btrfs_find_free_ino(root, &objectid); | |
9767 | if (ret) | |
9768 | return ret; | |
9769 | ||
9770 | inode = btrfs_new_inode(trans, root, dir, | |
9771 | dentry->d_name.name, | |
9772 | dentry->d_name.len, | |
4a0cc7ca | 9773 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9774 | objectid, |
9775 | S_IFCHR | WHITEOUT_MODE, | |
9776 | &index); | |
9777 | ||
9778 | if (IS_ERR(inode)) { | |
9779 | ret = PTR_ERR(inode); | |
9780 | return ret; | |
9781 | } | |
9782 | ||
9783 | inode->i_op = &btrfs_special_inode_operations; | |
9784 | init_special_inode(inode, inode->i_mode, | |
9785 | WHITEOUT_DEV); | |
9786 | ||
9787 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9788 | &dentry->d_name); | |
9789 | if (ret) | |
c9901618 | 9790 | goto out; |
cdd1fedf | 9791 | |
cef415af NB |
9792 | ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9793 | BTRFS_I(inode), 0, index); | |
cdd1fedf | 9794 | if (ret) |
c9901618 | 9795 | goto out; |
cdd1fedf DF |
9796 | |
9797 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 9798 | out: |
cdd1fedf | 9799 | unlock_new_inode(inode); |
c9901618 FM |
9800 | if (ret) |
9801 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9802 | iput(inode); |
9803 | ||
c9901618 | 9804 | return ret; |
cdd1fedf DF |
9805 | } |
9806 | ||
d397712b | 9807 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9808 | struct inode *new_dir, struct dentry *new_dentry, |
9809 | unsigned int flags) | |
39279cc3 | 9810 | { |
0b246afa | 9811 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9812 | struct btrfs_trans_handle *trans; |
5062af35 | 9813 | unsigned int trans_num_items; |
39279cc3 | 9814 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9815 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9816 | struct inode *new_inode = d_inode(new_dentry); |
9817 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9818 | u64 index = 0; |
4df27c4d | 9819 | u64 root_objectid; |
39279cc3 | 9820 | int ret; |
4a0cc7ca | 9821 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 9822 | bool log_pinned = false; |
d4682ba0 FM |
9823 | struct btrfs_log_ctx ctx; |
9824 | bool sync_log = false; | |
9825 | bool commit_transaction = false; | |
39279cc3 | 9826 | |
4a0cc7ca | 9827 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9828 | return -EPERM; |
9829 | ||
4df27c4d | 9830 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9831 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9832 | return -EXDEV; |
9833 | ||
33345d01 | 9834 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9835 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9836 | return -ENOTEMPTY; |
5f39d397 | 9837 | |
4df27c4d YZ |
9838 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9839 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9840 | return -ENOTEMPTY; | |
9c52057c CM |
9841 | |
9842 | ||
9843 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9844 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9845 | new_dentry->d_name.name, |
9846 | new_dentry->d_name.len); | |
9847 | ||
9848 | if (ret) { | |
9849 | if (ret == -EEXIST) { | |
9850 | /* we shouldn't get | |
9851 | * eexist without a new_inode */ | |
fae7f21c | 9852 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9853 | return ret; |
9854 | } | |
9855 | } else { | |
9856 | /* maybe -EOVERFLOW */ | |
9857 | return ret; | |
9858 | } | |
9859 | } | |
9860 | ret = 0; | |
9861 | ||
5a3f23d5 | 9862 | /* |
8d875f95 CM |
9863 | * we're using rename to replace one file with another. Start IO on it |
9864 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9865 | */ |
8d875f95 | 9866 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9867 | filemap_flush(old_inode->i_mapping); |
9868 | ||
76dda93c | 9869 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9870 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9871 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
9872 | /* |
9873 | * We want to reserve the absolute worst case amount of items. So if | |
9874 | * both inodes are subvols and we need to unlink them then that would | |
9875 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9876 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9877 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9878 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9879 | * If our rename has the whiteout flag, we need more 5 units for the |
9880 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9881 | * when selinux is enabled). | |
a22285a6 | 9882 | */ |
5062af35 FM |
9883 | trans_num_items = 11; |
9884 | if (flags & RENAME_WHITEOUT) | |
9885 | trans_num_items += 5; | |
9886 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9887 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9888 | ret = PTR_ERR(trans); |
9889 | goto out_notrans; | |
9890 | } | |
76dda93c | 9891 | |
4df27c4d YZ |
9892 | if (dest != root) |
9893 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9894 | |
877574e2 | 9895 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9896 | if (ret) |
9897 | goto out_fail; | |
5a3f23d5 | 9898 | |
67de1176 | 9899 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9900 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9901 | /* force full log commit if subvolume involved. */ |
90787766 | 9902 | btrfs_set_log_full_commit(trans); |
4df27c4d | 9903 | } else { |
c4aba954 FM |
9904 | btrfs_pin_log_trans(root); |
9905 | log_pinned = true; | |
a5719521 YZ |
9906 | ret = btrfs_insert_inode_ref(trans, dest, |
9907 | new_dentry->d_name.name, | |
9908 | new_dentry->d_name.len, | |
33345d01 | 9909 | old_ino, |
4a0cc7ca | 9910 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
9911 | if (ret) |
9912 | goto out_fail; | |
4df27c4d | 9913 | } |
5a3f23d5 | 9914 | |
0c4d2d95 JB |
9915 | inode_inc_iversion(old_dir); |
9916 | inode_inc_iversion(new_dir); | |
9917 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9918 | old_dir->i_ctime = old_dir->i_mtime = |
9919 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 9920 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 9921 | |
12fcfd22 | 9922 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
9923 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9924 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 9925 | |
33345d01 | 9926 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9927 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
401b3b19 | 9928 | ret = btrfs_unlink_subvol(trans, old_dir, root_objectid, |
4df27c4d YZ |
9929 | old_dentry->d_name.name, |
9930 | old_dentry->d_name.len); | |
9931 | } else { | |
4ec5934e NB |
9932 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9933 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
9934 | old_dentry->d_name.name, |
9935 | old_dentry->d_name.len); | |
9936 | if (!ret) | |
9937 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 9938 | } |
79787eaa | 9939 | if (ret) { |
66642832 | 9940 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9941 | goto out_fail; |
9942 | } | |
39279cc3 CM |
9943 | |
9944 | if (new_inode) { | |
0c4d2d95 | 9945 | inode_inc_iversion(new_inode); |
c2050a45 | 9946 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9947 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d YZ |
9948 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
9949 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
401b3b19 | 9950 | ret = btrfs_unlink_subvol(trans, new_dir, root_objectid, |
4df27c4d YZ |
9951 | new_dentry->d_name.name, |
9952 | new_dentry->d_name.len); | |
9953 | BUG_ON(new_inode->i_nlink == 0); | |
9954 | } else { | |
4ec5934e NB |
9955 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9956 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
9957 | new_dentry->d_name.name, |
9958 | new_dentry->d_name.len); | |
9959 | } | |
4ef31a45 | 9960 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
9961 | ret = btrfs_orphan_add(trans, |
9962 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 9963 | if (ret) { |
66642832 | 9964 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9965 | goto out_fail; |
9966 | } | |
39279cc3 | 9967 | } |
aec7477b | 9968 | |
db0a669f | 9969 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 9970 | new_dentry->d_name.name, |
a5719521 | 9971 | new_dentry->d_name.len, 0, index); |
79787eaa | 9972 | if (ret) { |
66642832 | 9973 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9974 | goto out_fail; |
9975 | } | |
39279cc3 | 9976 | |
67de1176 MX |
9977 | if (old_inode->i_nlink == 1) |
9978 | BTRFS_I(old_inode)->dir_index = index; | |
9979 | ||
3dc9e8f7 | 9980 | if (log_pinned) { |
10d9f309 | 9981 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 9982 | |
d4682ba0 FM |
9983 | btrfs_init_log_ctx(&ctx, old_inode); |
9984 | ret = btrfs_log_new_name(trans, BTRFS_I(old_inode), | |
9985 | BTRFS_I(old_dir), parent, | |
9986 | false, &ctx); | |
9987 | if (ret == BTRFS_NEED_LOG_SYNC) | |
9988 | sync_log = true; | |
9989 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
9990 | commit_transaction = true; | |
9991 | ret = 0; | |
4df27c4d | 9992 | btrfs_end_log_trans(root); |
3dc9e8f7 | 9993 | log_pinned = false; |
4df27c4d | 9994 | } |
cdd1fedf DF |
9995 | |
9996 | if (flags & RENAME_WHITEOUT) { | |
9997 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
9998 | old_dentry); | |
9999 | ||
10000 | if (ret) { | |
66642832 | 10001 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
10002 | goto out_fail; |
10003 | } | |
4df27c4d | 10004 | } |
39279cc3 | 10005 | out_fail: |
3dc9e8f7 FM |
10006 | /* |
10007 | * If we have pinned the log and an error happened, we unpin tasks | |
10008 | * trying to sync the log and force them to fallback to a transaction | |
10009 | * commit if the log currently contains any of the inodes involved in | |
10010 | * this rename operation (to ensure we do not persist a log with an | |
10011 | * inconsistent state for any of these inodes or leading to any | |
10012 | * inconsistencies when replayed). If the transaction was aborted, the | |
10013 | * abortion reason is propagated to userspace when attempting to commit | |
10014 | * the transaction. If the log does not contain any of these inodes, we | |
10015 | * allow the tasks to sync it. | |
10016 | */ | |
10017 | if (ret && log_pinned) { | |
0f8939b8 NB |
10018 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
10019 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
10020 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 10021 | (new_inode && |
0f8939b8 | 10022 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
90787766 | 10023 | btrfs_set_log_full_commit(trans); |
3dc9e8f7 FM |
10024 | |
10025 | btrfs_end_log_trans(root); | |
10026 | log_pinned = false; | |
10027 | } | |
d4682ba0 FM |
10028 | if (!ret && sync_log) { |
10029 | ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx); | |
10030 | if (ret) | |
10031 | commit_transaction = true; | |
10032 | } | |
10033 | if (commit_transaction) { | |
10034 | ret = btrfs_commit_transaction(trans); | |
10035 | } else { | |
10036 | int ret2; | |
10037 | ||
10038 | ret2 = btrfs_end_transaction(trans); | |
10039 | ret = ret ? ret : ret2; | |
10040 | } | |
b44c59a8 | 10041 | out_notrans: |
33345d01 | 10042 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 10043 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 10044 | |
39279cc3 CM |
10045 | return ret; |
10046 | } | |
10047 | ||
80ace85c MS |
10048 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
10049 | struct inode *new_dir, struct dentry *new_dentry, | |
10050 | unsigned int flags) | |
10051 | { | |
cdd1fedf | 10052 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
10053 | return -EINVAL; |
10054 | ||
cdd1fedf DF |
10055 | if (flags & RENAME_EXCHANGE) |
10056 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
10057 | new_dentry); | |
10058 | ||
10059 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
10060 | } |
10061 | ||
3a2f8c07 NB |
10062 | struct btrfs_delalloc_work { |
10063 | struct inode *inode; | |
10064 | struct completion completion; | |
10065 | struct list_head list; | |
10066 | struct btrfs_work work; | |
10067 | }; | |
10068 | ||
8ccf6f19 MX |
10069 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
10070 | { | |
10071 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 10072 | struct inode *inode; |
8ccf6f19 MX |
10073 | |
10074 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
10075 | work); | |
9f23e289 | 10076 | inode = delalloc_work->inode; |
30424601 DS |
10077 | filemap_flush(inode->i_mapping); |
10078 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
10079 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 10080 | filemap_flush(inode->i_mapping); |
8ccf6f19 | 10081 | |
076da91c | 10082 | iput(inode); |
8ccf6f19 MX |
10083 | complete(&delalloc_work->completion); |
10084 | } | |
10085 | ||
3a2f8c07 | 10086 | static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode) |
8ccf6f19 MX |
10087 | { |
10088 | struct btrfs_delalloc_work *work; | |
10089 | ||
100d5702 | 10090 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
10091 | if (!work) |
10092 | return NULL; | |
10093 | ||
10094 | init_completion(&work->completion); | |
10095 | INIT_LIST_HEAD(&work->list); | |
10096 | work->inode = inode; | |
9e0af237 LB |
10097 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, |
10098 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
10099 | |
10100 | return work; | |
10101 | } | |
10102 | ||
d352ac68 CM |
10103 | /* |
10104 | * some fairly slow code that needs optimization. This walks the list | |
10105 | * of all the inodes with pending delalloc and forces them to disk. | |
10106 | */ | |
3cd24c69 | 10107 | static int start_delalloc_inodes(struct btrfs_root *root, int nr, bool snapshot) |
ea8c2819 | 10108 | { |
ea8c2819 | 10109 | struct btrfs_inode *binode; |
5b21f2ed | 10110 | struct inode *inode; |
8ccf6f19 MX |
10111 | struct btrfs_delalloc_work *work, *next; |
10112 | struct list_head works; | |
1eafa6c7 | 10113 | struct list_head splice; |
8ccf6f19 | 10114 | int ret = 0; |
ea8c2819 | 10115 | |
8ccf6f19 | 10116 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 10117 | INIT_LIST_HEAD(&splice); |
63607cc8 | 10118 | |
573bfb72 | 10119 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
10120 | spin_lock(&root->delalloc_lock); |
10121 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
10122 | while (!list_empty(&splice)) { |
10123 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 10124 | delalloc_inodes); |
1eafa6c7 | 10125 | |
eb73c1b7 MX |
10126 | list_move_tail(&binode->delalloc_inodes, |
10127 | &root->delalloc_inodes); | |
5b21f2ed | 10128 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 10129 | if (!inode) { |
eb73c1b7 | 10130 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 10131 | continue; |
df0af1a5 | 10132 | } |
eb73c1b7 | 10133 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 10134 | |
3cd24c69 EL |
10135 | if (snapshot) |
10136 | set_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
10137 | &binode->runtime_flags); | |
076da91c | 10138 | work = btrfs_alloc_delalloc_work(inode); |
5d99a998 | 10139 | if (!work) { |
4fbb5147 | 10140 | iput(inode); |
1eafa6c7 | 10141 | ret = -ENOMEM; |
a1ecaabb | 10142 | goto out; |
5b21f2ed | 10143 | } |
1eafa6c7 | 10144 | list_add_tail(&work->list, &works); |
a44903ab QW |
10145 | btrfs_queue_work(root->fs_info->flush_workers, |
10146 | &work->work); | |
6c255e67 MX |
10147 | ret++; |
10148 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 10149 | goto out; |
5b21f2ed | 10150 | cond_resched(); |
eb73c1b7 | 10151 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 10152 | } |
eb73c1b7 | 10153 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 10154 | |
a1ecaabb | 10155 | out: |
eb73c1b7 MX |
10156 | list_for_each_entry_safe(work, next, &works, list) { |
10157 | list_del_init(&work->list); | |
40012f96 NB |
10158 | wait_for_completion(&work->completion); |
10159 | kfree(work); | |
eb73c1b7 MX |
10160 | } |
10161 | ||
81f1d390 | 10162 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
10163 | spin_lock(&root->delalloc_lock); |
10164 | list_splice_tail(&splice, &root->delalloc_inodes); | |
10165 | spin_unlock(&root->delalloc_lock); | |
10166 | } | |
573bfb72 | 10167 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
10168 | return ret; |
10169 | } | |
1eafa6c7 | 10170 | |
3cd24c69 | 10171 | int btrfs_start_delalloc_snapshot(struct btrfs_root *root) |
eb73c1b7 | 10172 | { |
0b246afa | 10173 | struct btrfs_fs_info *fs_info = root->fs_info; |
eb73c1b7 | 10174 | int ret; |
1eafa6c7 | 10175 | |
0b246afa | 10176 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10177 | return -EROFS; |
10178 | ||
3cd24c69 | 10179 | ret = start_delalloc_inodes(root, -1, true); |
6c255e67 MX |
10180 | if (ret > 0) |
10181 | ret = 0; | |
eb73c1b7 MX |
10182 | return ret; |
10183 | } | |
10184 | ||
82b3e53b | 10185 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr) |
eb73c1b7 MX |
10186 | { |
10187 | struct btrfs_root *root; | |
10188 | struct list_head splice; | |
10189 | int ret; | |
10190 | ||
2c21b4d7 | 10191 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10192 | return -EROFS; |
10193 | ||
10194 | INIT_LIST_HEAD(&splice); | |
10195 | ||
573bfb72 | 10196 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
10197 | spin_lock(&fs_info->delalloc_root_lock); |
10198 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 10199 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
10200 | root = list_first_entry(&splice, struct btrfs_root, |
10201 | delalloc_root); | |
10202 | root = btrfs_grab_fs_root(root); | |
10203 | BUG_ON(!root); | |
10204 | list_move_tail(&root->delalloc_root, | |
10205 | &fs_info->delalloc_roots); | |
10206 | spin_unlock(&fs_info->delalloc_root_lock); | |
10207 | ||
3cd24c69 | 10208 | ret = start_delalloc_inodes(root, nr, false); |
eb73c1b7 | 10209 | btrfs_put_fs_root(root); |
6c255e67 | 10210 | if (ret < 0) |
eb73c1b7 MX |
10211 | goto out; |
10212 | ||
6c255e67 MX |
10213 | if (nr != -1) { |
10214 | nr -= ret; | |
10215 | WARN_ON(nr < 0); | |
10216 | } | |
eb73c1b7 | 10217 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 10218 | } |
eb73c1b7 | 10219 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 10220 | |
6c255e67 | 10221 | ret = 0; |
eb73c1b7 | 10222 | out: |
81f1d390 | 10223 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
10224 | spin_lock(&fs_info->delalloc_root_lock); |
10225 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
10226 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 10227 | } |
573bfb72 | 10228 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 10229 | return ret; |
ea8c2819 CM |
10230 | } |
10231 | ||
39279cc3 CM |
10232 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
10233 | const char *symname) | |
10234 | { | |
0b246afa | 10235 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
10236 | struct btrfs_trans_handle *trans; |
10237 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10238 | struct btrfs_path *path; | |
10239 | struct btrfs_key key; | |
1832a6d5 | 10240 | struct inode *inode = NULL; |
39279cc3 | 10241 | int err; |
39279cc3 | 10242 | u64 objectid; |
67871254 | 10243 | u64 index = 0; |
39279cc3 CM |
10244 | int name_len; |
10245 | int datasize; | |
5f39d397 | 10246 | unsigned long ptr; |
39279cc3 | 10247 | struct btrfs_file_extent_item *ei; |
5f39d397 | 10248 | struct extent_buffer *leaf; |
39279cc3 | 10249 | |
f06becc4 | 10250 | name_len = strlen(symname); |
0b246afa | 10251 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 10252 | return -ENAMETOOLONG; |
1832a6d5 | 10253 | |
9ed74f2d JB |
10254 | /* |
10255 | * 2 items for inode item and ref | |
10256 | * 2 items for dir items | |
9269d12b FM |
10257 | * 1 item for updating parent inode item |
10258 | * 1 item for the inline extent item | |
9ed74f2d JB |
10259 | * 1 item for xattr if selinux is on |
10260 | */ | |
9269d12b | 10261 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10262 | if (IS_ERR(trans)) |
10263 | return PTR_ERR(trans); | |
1832a6d5 | 10264 | |
581bb050 LZ |
10265 | err = btrfs_find_free_ino(root, &objectid); |
10266 | if (err) | |
10267 | goto out_unlock; | |
10268 | ||
aec7477b | 10269 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
10270 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
10271 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
10272 | if (IS_ERR(inode)) { |
10273 | err = PTR_ERR(inode); | |
32955c54 | 10274 | inode = NULL; |
39279cc3 | 10275 | goto out_unlock; |
7cf96da3 | 10276 | } |
39279cc3 | 10277 | |
ad19db71 CS |
10278 | /* |
10279 | * If the active LSM wants to access the inode during | |
10280 | * d_instantiate it needs these. Smack checks to see | |
10281 | * if the filesystem supports xattrs by looking at the | |
10282 | * ops vector. | |
10283 | */ | |
10284 | inode->i_fop = &btrfs_file_operations; | |
10285 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10286 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10287 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10288 | ||
10289 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10290 | if (err) | |
32955c54 | 10291 | goto out_unlock; |
ad19db71 | 10292 | |
39279cc3 | 10293 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10294 | if (!path) { |
10295 | err = -ENOMEM; | |
32955c54 | 10296 | goto out_unlock; |
d8926bb3 | 10297 | } |
4a0cc7ca | 10298 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 10299 | key.offset = 0; |
962a298f | 10300 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10301 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10302 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10303 | datasize); | |
54aa1f4d | 10304 | if (err) { |
b0839166 | 10305 | btrfs_free_path(path); |
32955c54 | 10306 | goto out_unlock; |
54aa1f4d | 10307 | } |
5f39d397 CM |
10308 | leaf = path->nodes[0]; |
10309 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10310 | struct btrfs_file_extent_item); | |
10311 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10312 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10313 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10314 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10315 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10316 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10317 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10318 | ||
39279cc3 | 10319 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10320 | write_extent_buffer(leaf, symname, ptr, name_len); |
10321 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10322 | btrfs_free_path(path); |
5f39d397 | 10323 | |
39279cc3 | 10324 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10325 | inode_nohighmem(inode); |
d899e052 | 10326 | inode_set_bytes(inode, name_len); |
6ef06d27 | 10327 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
54aa1f4d | 10328 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
10329 | /* |
10330 | * Last step, add directory indexes for our symlink inode. This is the | |
10331 | * last step to avoid extra cleanup of these indexes if an error happens | |
10332 | * elsewhere above. | |
10333 | */ | |
10334 | if (!err) | |
cef415af NB |
10335 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
10336 | BTRFS_I(inode), 0, index); | |
32955c54 AV |
10337 | if (err) |
10338 | goto out_unlock; | |
b0d5d10f | 10339 | |
1e2e547a | 10340 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
10341 | |
10342 | out_unlock: | |
3a45bb20 | 10343 | btrfs_end_transaction(trans); |
32955c54 | 10344 | if (err && inode) { |
39279cc3 | 10345 | inode_dec_link_count(inode); |
32955c54 | 10346 | discard_new_inode(inode); |
39279cc3 | 10347 | } |
2ff7e61e | 10348 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
10349 | return err; |
10350 | } | |
16432985 | 10351 | |
0af3d00b JB |
10352 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10353 | u64 start, u64 num_bytes, u64 min_size, | |
10354 | loff_t actual_len, u64 *alloc_hint, | |
10355 | struct btrfs_trans_handle *trans) | |
d899e052 | 10356 | { |
0b246afa | 10357 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
10358 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10359 | struct extent_map *em; | |
d899e052 YZ |
10360 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10361 | struct btrfs_key ins; | |
d899e052 | 10362 | u64 cur_offset = start; |
55a61d1d | 10363 | u64 i_size; |
154ea289 | 10364 | u64 cur_bytes; |
0b670dc4 | 10365 | u64 last_alloc = (u64)-1; |
d899e052 | 10366 | int ret = 0; |
0af3d00b | 10367 | bool own_trans = true; |
18513091 | 10368 | u64 end = start + num_bytes - 1; |
d899e052 | 10369 | |
0af3d00b JB |
10370 | if (trans) |
10371 | own_trans = false; | |
d899e052 | 10372 | while (num_bytes > 0) { |
0af3d00b JB |
10373 | if (own_trans) { |
10374 | trans = btrfs_start_transaction(root, 3); | |
10375 | if (IS_ERR(trans)) { | |
10376 | ret = PTR_ERR(trans); | |
10377 | break; | |
10378 | } | |
5a303d5d YZ |
10379 | } |
10380 | ||
ee22184b | 10381 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10382 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10383 | /* |
10384 | * If we are severely fragmented we could end up with really | |
10385 | * small allocations, so if the allocator is returning small | |
10386 | * chunks lets make its job easier by only searching for those | |
10387 | * sized chunks. | |
10388 | */ | |
10389 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
10390 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
10391 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
5a303d5d | 10392 | if (ret) { |
0af3d00b | 10393 | if (own_trans) |
3a45bb20 | 10394 | btrfs_end_transaction(trans); |
a22285a6 | 10395 | break; |
d899e052 | 10396 | } |
0b246afa | 10397 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5a303d5d | 10398 | |
0b670dc4 | 10399 | last_alloc = ins.offset; |
d899e052 YZ |
10400 | ret = insert_reserved_file_extent(trans, inode, |
10401 | cur_offset, ins.objectid, | |
10402 | ins.offset, ins.offset, | |
920bbbfb | 10403 | ins.offset, 0, 0, 0, |
d899e052 | 10404 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 10405 | if (ret) { |
2ff7e61e | 10406 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 10407 | ins.offset, 0); |
66642832 | 10408 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10409 | if (own_trans) |
3a45bb20 | 10410 | btrfs_end_transaction(trans); |
79787eaa JM |
10411 | break; |
10412 | } | |
31193213 | 10413 | |
dcdbc059 | 10414 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 10415 | cur_offset + ins.offset -1, 0); |
5a303d5d | 10416 | |
5dc562c5 JB |
10417 | em = alloc_extent_map(); |
10418 | if (!em) { | |
10419 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10420 | &BTRFS_I(inode)->runtime_flags); | |
10421 | goto next; | |
10422 | } | |
10423 | ||
10424 | em->start = cur_offset; | |
10425 | em->orig_start = cur_offset; | |
10426 | em->len = ins.offset; | |
10427 | em->block_start = ins.objectid; | |
10428 | em->block_len = ins.offset; | |
b4939680 | 10429 | em->orig_block_len = ins.offset; |
cc95bef6 | 10430 | em->ram_bytes = ins.offset; |
0b246afa | 10431 | em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 JB |
10432 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
10433 | em->generation = trans->transid; | |
10434 | ||
10435 | while (1) { | |
10436 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10437 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10438 | write_unlock(&em_tree->lock); |
10439 | if (ret != -EEXIST) | |
10440 | break; | |
dcdbc059 | 10441 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
10442 | cur_offset + ins.offset - 1, |
10443 | 0); | |
10444 | } | |
10445 | free_extent_map(em); | |
10446 | next: | |
d899e052 YZ |
10447 | num_bytes -= ins.offset; |
10448 | cur_offset += ins.offset; | |
efa56464 | 10449 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10450 | |
0c4d2d95 | 10451 | inode_inc_iversion(inode); |
c2050a45 | 10452 | inode->i_ctime = current_time(inode); |
6cbff00f | 10453 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10454 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10455 | (actual_len > inode->i_size) && |
10456 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10457 | if (cur_offset > actual_len) |
55a61d1d | 10458 | i_size = actual_len; |
d1ea6a61 | 10459 | else |
55a61d1d JB |
10460 | i_size = cur_offset; |
10461 | i_size_write(inode, i_size); | |
10462 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10463 | } |
10464 | ||
d899e052 | 10465 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10466 | |
10467 | if (ret) { | |
66642832 | 10468 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10469 | if (own_trans) |
3a45bb20 | 10470 | btrfs_end_transaction(trans); |
79787eaa JM |
10471 | break; |
10472 | } | |
d899e052 | 10473 | |
0af3d00b | 10474 | if (own_trans) |
3a45bb20 | 10475 | btrfs_end_transaction(trans); |
5a303d5d | 10476 | } |
18513091 | 10477 | if (cur_offset < end) |
bc42bda2 | 10478 | btrfs_free_reserved_data_space(inode, NULL, cur_offset, |
18513091 | 10479 | end - cur_offset + 1); |
d899e052 YZ |
10480 | return ret; |
10481 | } | |
10482 | ||
0af3d00b JB |
10483 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10484 | u64 start, u64 num_bytes, u64 min_size, | |
10485 | loff_t actual_len, u64 *alloc_hint) | |
10486 | { | |
10487 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10488 | min_size, actual_len, alloc_hint, | |
10489 | NULL); | |
10490 | } | |
10491 | ||
10492 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10493 | struct btrfs_trans_handle *trans, int mode, | |
10494 | u64 start, u64 num_bytes, u64 min_size, | |
10495 | loff_t actual_len, u64 *alloc_hint) | |
10496 | { | |
10497 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10498 | min_size, actual_len, alloc_hint, trans); | |
10499 | } | |
10500 | ||
e6dcd2dc CM |
10501 | static int btrfs_set_page_dirty(struct page *page) |
10502 | { | |
e6dcd2dc CM |
10503 | return __set_page_dirty_nobuffers(page); |
10504 | } | |
10505 | ||
10556cb2 | 10506 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10507 | { |
b83cc969 | 10508 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10509 | umode_t mode = inode->i_mode; |
b83cc969 | 10510 | |
cb6db4e5 JM |
10511 | if (mask & MAY_WRITE && |
10512 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10513 | if (btrfs_root_readonly(root)) | |
10514 | return -EROFS; | |
10515 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10516 | return -EACCES; | |
10517 | } | |
2830ba7f | 10518 | return generic_permission(inode, mask); |
fdebe2bd | 10519 | } |
39279cc3 | 10520 | |
ef3b9af5 FM |
10521 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10522 | { | |
2ff7e61e | 10523 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10524 | struct btrfs_trans_handle *trans; |
10525 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10526 | struct inode *inode = NULL; | |
10527 | u64 objectid; | |
10528 | u64 index; | |
10529 | int ret = 0; | |
10530 | ||
10531 | /* | |
10532 | * 5 units required for adding orphan entry | |
10533 | */ | |
10534 | trans = btrfs_start_transaction(root, 5); | |
10535 | if (IS_ERR(trans)) | |
10536 | return PTR_ERR(trans); | |
10537 | ||
10538 | ret = btrfs_find_free_ino(root, &objectid); | |
10539 | if (ret) | |
10540 | goto out; | |
10541 | ||
10542 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 10543 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
10544 | if (IS_ERR(inode)) { |
10545 | ret = PTR_ERR(inode); | |
10546 | inode = NULL; | |
10547 | goto out; | |
10548 | } | |
10549 | ||
ef3b9af5 FM |
10550 | inode->i_fop = &btrfs_file_operations; |
10551 | inode->i_op = &btrfs_file_inode_operations; | |
10552 | ||
10553 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10554 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10555 | ||
b0d5d10f CM |
10556 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10557 | if (ret) | |
32955c54 | 10558 | goto out; |
b0d5d10f CM |
10559 | |
10560 | ret = btrfs_update_inode(trans, root, inode); | |
10561 | if (ret) | |
32955c54 | 10562 | goto out; |
73f2e545 | 10563 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 10564 | if (ret) |
32955c54 | 10565 | goto out; |
ef3b9af5 | 10566 | |
5762b5c9 FM |
10567 | /* |
10568 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10569 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10570 | * through: | |
10571 | * | |
10572 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10573 | */ | |
10574 | set_nlink(inode, 1); | |
ef3b9af5 | 10575 | d_tmpfile(dentry, inode); |
32955c54 | 10576 | unlock_new_inode(inode); |
ef3b9af5 | 10577 | mark_inode_dirty(inode); |
ef3b9af5 | 10578 | out: |
3a45bb20 | 10579 | btrfs_end_transaction(trans); |
32955c54 AV |
10580 | if (ret && inode) |
10581 | discard_new_inode(inode); | |
2ff7e61e | 10582 | btrfs_btree_balance_dirty(fs_info); |
ef3b9af5 FM |
10583 | return ret; |
10584 | } | |
10585 | ||
5cdc84bf | 10586 | void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
c6100a4b | 10587 | { |
5cdc84bf | 10588 | struct inode *inode = tree->private_data; |
c6100a4b JB |
10589 | unsigned long index = start >> PAGE_SHIFT; |
10590 | unsigned long end_index = end >> PAGE_SHIFT; | |
10591 | struct page *page; | |
10592 | ||
10593 | while (index <= end_index) { | |
10594 | page = find_get_page(inode->i_mapping, index); | |
10595 | ASSERT(page); /* Pages should be in the extent_io_tree */ | |
10596 | set_page_writeback(page); | |
10597 | put_page(page); | |
10598 | index++; | |
10599 | } | |
10600 | } | |
10601 | ||
ed46ff3d OS |
10602 | #ifdef CONFIG_SWAP |
10603 | /* | |
10604 | * Add an entry indicating a block group or device which is pinned by a | |
10605 | * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a | |
10606 | * negative errno on failure. | |
10607 | */ | |
10608 | static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr, | |
10609 | bool is_block_group) | |
10610 | { | |
10611 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10612 | struct btrfs_swapfile_pin *sp, *entry; | |
10613 | struct rb_node **p; | |
10614 | struct rb_node *parent = NULL; | |
10615 | ||
10616 | sp = kmalloc(sizeof(*sp), GFP_NOFS); | |
10617 | if (!sp) | |
10618 | return -ENOMEM; | |
10619 | sp->ptr = ptr; | |
10620 | sp->inode = inode; | |
10621 | sp->is_block_group = is_block_group; | |
10622 | ||
10623 | spin_lock(&fs_info->swapfile_pins_lock); | |
10624 | p = &fs_info->swapfile_pins.rb_node; | |
10625 | while (*p) { | |
10626 | parent = *p; | |
10627 | entry = rb_entry(parent, struct btrfs_swapfile_pin, node); | |
10628 | if (sp->ptr < entry->ptr || | |
10629 | (sp->ptr == entry->ptr && sp->inode < entry->inode)) { | |
10630 | p = &(*p)->rb_left; | |
10631 | } else if (sp->ptr > entry->ptr || | |
10632 | (sp->ptr == entry->ptr && sp->inode > entry->inode)) { | |
10633 | p = &(*p)->rb_right; | |
10634 | } else { | |
10635 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10636 | kfree(sp); | |
10637 | return 1; | |
10638 | } | |
10639 | } | |
10640 | rb_link_node(&sp->node, parent, p); | |
10641 | rb_insert_color(&sp->node, &fs_info->swapfile_pins); | |
10642 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10643 | return 0; | |
10644 | } | |
10645 | ||
10646 | /* Free all of the entries pinned by this swapfile. */ | |
10647 | static void btrfs_free_swapfile_pins(struct inode *inode) | |
10648 | { | |
10649 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10650 | struct btrfs_swapfile_pin *sp; | |
10651 | struct rb_node *node, *next; | |
10652 | ||
10653 | spin_lock(&fs_info->swapfile_pins_lock); | |
10654 | node = rb_first(&fs_info->swapfile_pins); | |
10655 | while (node) { | |
10656 | next = rb_next(node); | |
10657 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
10658 | if (sp->inode == inode) { | |
10659 | rb_erase(&sp->node, &fs_info->swapfile_pins); | |
10660 | if (sp->is_block_group) | |
10661 | btrfs_put_block_group(sp->ptr); | |
10662 | kfree(sp); | |
10663 | } | |
10664 | node = next; | |
10665 | } | |
10666 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10667 | } | |
10668 | ||
10669 | struct btrfs_swap_info { | |
10670 | u64 start; | |
10671 | u64 block_start; | |
10672 | u64 block_len; | |
10673 | u64 lowest_ppage; | |
10674 | u64 highest_ppage; | |
10675 | unsigned long nr_pages; | |
10676 | int nr_extents; | |
10677 | }; | |
10678 | ||
10679 | static int btrfs_add_swap_extent(struct swap_info_struct *sis, | |
10680 | struct btrfs_swap_info *bsi) | |
10681 | { | |
10682 | unsigned long nr_pages; | |
10683 | u64 first_ppage, first_ppage_reported, next_ppage; | |
10684 | int ret; | |
10685 | ||
10686 | first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT; | |
10687 | next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len, | |
10688 | PAGE_SIZE) >> PAGE_SHIFT; | |
10689 | ||
10690 | if (first_ppage >= next_ppage) | |
10691 | return 0; | |
10692 | nr_pages = next_ppage - first_ppage; | |
10693 | ||
10694 | first_ppage_reported = first_ppage; | |
10695 | if (bsi->start == 0) | |
10696 | first_ppage_reported++; | |
10697 | if (bsi->lowest_ppage > first_ppage_reported) | |
10698 | bsi->lowest_ppage = first_ppage_reported; | |
10699 | if (bsi->highest_ppage < (next_ppage - 1)) | |
10700 | bsi->highest_ppage = next_ppage - 1; | |
10701 | ||
10702 | ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage); | |
10703 | if (ret < 0) | |
10704 | return ret; | |
10705 | bsi->nr_extents += ret; | |
10706 | bsi->nr_pages += nr_pages; | |
10707 | return 0; | |
10708 | } | |
10709 | ||
10710 | static void btrfs_swap_deactivate(struct file *file) | |
10711 | { | |
10712 | struct inode *inode = file_inode(file); | |
10713 | ||
10714 | btrfs_free_swapfile_pins(inode); | |
10715 | atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles); | |
10716 | } | |
10717 | ||
10718 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10719 | sector_t *span) | |
10720 | { | |
10721 | struct inode *inode = file_inode(file); | |
10722 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10723 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
10724 | struct extent_state *cached_state = NULL; | |
10725 | struct extent_map *em = NULL; | |
10726 | struct btrfs_device *device = NULL; | |
10727 | struct btrfs_swap_info bsi = { | |
10728 | .lowest_ppage = (sector_t)-1ULL, | |
10729 | }; | |
10730 | int ret = 0; | |
10731 | u64 isize; | |
10732 | u64 start; | |
10733 | ||
10734 | /* | |
10735 | * If the swap file was just created, make sure delalloc is done. If the | |
10736 | * file changes again after this, the user is doing something stupid and | |
10737 | * we don't really care. | |
10738 | */ | |
10739 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
10740 | if (ret) | |
10741 | return ret; | |
10742 | ||
10743 | /* | |
10744 | * The inode is locked, so these flags won't change after we check them. | |
10745 | */ | |
10746 | if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) { | |
10747 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10748 | return -EINVAL; | |
10749 | } | |
10750 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) { | |
10751 | btrfs_warn(fs_info, "swapfile must not be copy-on-write"); | |
10752 | return -EINVAL; | |
10753 | } | |
10754 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
10755 | btrfs_warn(fs_info, "swapfile must not be checksummed"); | |
10756 | return -EINVAL; | |
10757 | } | |
10758 | ||
10759 | /* | |
10760 | * Balance or device remove/replace/resize can move stuff around from | |
10761 | * under us. The EXCL_OP flag makes sure they aren't running/won't run | |
10762 | * concurrently while we are mapping the swap extents, and | |
10763 | * fs_info->swapfile_pins prevents them from running while the swap file | |
10764 | * is active and moving the extents. Note that this also prevents a | |
10765 | * concurrent device add which isn't actually necessary, but it's not | |
10766 | * really worth the trouble to allow it. | |
10767 | */ | |
10768 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) { | |
10769 | btrfs_warn(fs_info, | |
10770 | "cannot activate swapfile while exclusive operation is running"); | |
10771 | return -EBUSY; | |
10772 | } | |
10773 | /* | |
10774 | * Snapshots can create extents which require COW even if NODATACOW is | |
10775 | * set. We use this counter to prevent snapshots. We must increment it | |
10776 | * before walking the extents because we don't want a concurrent | |
10777 | * snapshot to run after we've already checked the extents. | |
10778 | */ | |
10779 | atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles); | |
10780 | ||
10781 | isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize); | |
10782 | ||
10783 | lock_extent_bits(io_tree, 0, isize - 1, &cached_state); | |
10784 | start = 0; | |
10785 | while (start < isize) { | |
10786 | u64 logical_block_start, physical_block_start; | |
10787 | struct btrfs_block_group_cache *bg; | |
10788 | u64 len = isize - start; | |
10789 | ||
10790 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); | |
10791 | if (IS_ERR(em)) { | |
10792 | ret = PTR_ERR(em); | |
10793 | goto out; | |
10794 | } | |
10795 | ||
10796 | if (em->block_start == EXTENT_MAP_HOLE) { | |
10797 | btrfs_warn(fs_info, "swapfile must not have holes"); | |
10798 | ret = -EINVAL; | |
10799 | goto out; | |
10800 | } | |
10801 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10802 | /* | |
10803 | * It's unlikely we'll ever actually find ourselves | |
10804 | * here, as a file small enough to fit inline won't be | |
10805 | * big enough to store more than the swap header, but in | |
10806 | * case something changes in the future, let's catch it | |
10807 | * here rather than later. | |
10808 | */ | |
10809 | btrfs_warn(fs_info, "swapfile must not be inline"); | |
10810 | ret = -EINVAL; | |
10811 | goto out; | |
10812 | } | |
10813 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10814 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10815 | ret = -EINVAL; | |
10816 | goto out; | |
10817 | } | |
10818 | ||
10819 | logical_block_start = em->block_start + (start - em->start); | |
10820 | len = min(len, em->len - (start - em->start)); | |
10821 | free_extent_map(em); | |
10822 | em = NULL; | |
10823 | ||
10824 | ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL); | |
10825 | if (ret < 0) { | |
10826 | goto out; | |
10827 | } else if (ret) { | |
10828 | ret = 0; | |
10829 | } else { | |
10830 | btrfs_warn(fs_info, | |
10831 | "swapfile must not be copy-on-write"); | |
10832 | ret = -EINVAL; | |
10833 | goto out; | |
10834 | } | |
10835 | ||
10836 | em = btrfs_get_chunk_map(fs_info, logical_block_start, len); | |
10837 | if (IS_ERR(em)) { | |
10838 | ret = PTR_ERR(em); | |
10839 | goto out; | |
10840 | } | |
10841 | ||
10842 | if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
10843 | btrfs_warn(fs_info, | |
10844 | "swapfile must have single data profile"); | |
10845 | ret = -EINVAL; | |
10846 | goto out; | |
10847 | } | |
10848 | ||
10849 | if (device == NULL) { | |
10850 | device = em->map_lookup->stripes[0].dev; | |
10851 | ret = btrfs_add_swapfile_pin(inode, device, false); | |
10852 | if (ret == 1) | |
10853 | ret = 0; | |
10854 | else if (ret) | |
10855 | goto out; | |
10856 | } else if (device != em->map_lookup->stripes[0].dev) { | |
10857 | btrfs_warn(fs_info, "swapfile must be on one device"); | |
10858 | ret = -EINVAL; | |
10859 | goto out; | |
10860 | } | |
10861 | ||
10862 | physical_block_start = (em->map_lookup->stripes[0].physical + | |
10863 | (logical_block_start - em->start)); | |
10864 | len = min(len, em->len - (logical_block_start - em->start)); | |
10865 | free_extent_map(em); | |
10866 | em = NULL; | |
10867 | ||
10868 | bg = btrfs_lookup_block_group(fs_info, logical_block_start); | |
10869 | if (!bg) { | |
10870 | btrfs_warn(fs_info, | |
10871 | "could not find block group containing swapfile"); | |
10872 | ret = -EINVAL; | |
10873 | goto out; | |
10874 | } | |
10875 | ||
10876 | ret = btrfs_add_swapfile_pin(inode, bg, true); | |
10877 | if (ret) { | |
10878 | btrfs_put_block_group(bg); | |
10879 | if (ret == 1) | |
10880 | ret = 0; | |
10881 | else | |
10882 | goto out; | |
10883 | } | |
10884 | ||
10885 | if (bsi.block_len && | |
10886 | bsi.block_start + bsi.block_len == physical_block_start) { | |
10887 | bsi.block_len += len; | |
10888 | } else { | |
10889 | if (bsi.block_len) { | |
10890 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10891 | if (ret) | |
10892 | goto out; | |
10893 | } | |
10894 | bsi.start = start; | |
10895 | bsi.block_start = physical_block_start; | |
10896 | bsi.block_len = len; | |
10897 | } | |
10898 | ||
10899 | start += len; | |
10900 | } | |
10901 | ||
10902 | if (bsi.block_len) | |
10903 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10904 | ||
10905 | out: | |
10906 | if (!IS_ERR_OR_NULL(em)) | |
10907 | free_extent_map(em); | |
10908 | ||
10909 | unlock_extent_cached(io_tree, 0, isize - 1, &cached_state); | |
10910 | ||
10911 | if (ret) | |
10912 | btrfs_swap_deactivate(file); | |
10913 | ||
10914 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); | |
10915 | ||
10916 | if (ret) | |
10917 | return ret; | |
10918 | ||
10919 | if (device) | |
10920 | sis->bdev = device->bdev; | |
10921 | *span = bsi.highest_ppage - bsi.lowest_ppage + 1; | |
10922 | sis->max = bsi.nr_pages; | |
10923 | sis->pages = bsi.nr_pages - 1; | |
10924 | sis->highest_bit = bsi.nr_pages - 1; | |
10925 | return bsi.nr_extents; | |
10926 | } | |
10927 | #else | |
10928 | static void btrfs_swap_deactivate(struct file *file) | |
10929 | { | |
10930 | } | |
10931 | ||
10932 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10933 | sector_t *span) | |
10934 | { | |
10935 | return -EOPNOTSUPP; | |
10936 | } | |
10937 | #endif | |
10938 | ||
6e1d5dcc | 10939 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10940 | .getattr = btrfs_getattr, |
39279cc3 CM |
10941 | .lookup = btrfs_lookup, |
10942 | .create = btrfs_create, | |
10943 | .unlink = btrfs_unlink, | |
10944 | .link = btrfs_link, | |
10945 | .mkdir = btrfs_mkdir, | |
10946 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10947 | .rename = btrfs_rename2, |
39279cc3 CM |
10948 | .symlink = btrfs_symlink, |
10949 | .setattr = btrfs_setattr, | |
618e21d5 | 10950 | .mknod = btrfs_mknod, |
5103e947 | 10951 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10952 | .permission = btrfs_permission, |
4e34e719 | 10953 | .get_acl = btrfs_get_acl, |
996a710d | 10954 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10955 | .update_time = btrfs_update_time, |
ef3b9af5 | 10956 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10957 | }; |
6e1d5dcc | 10958 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10959 | .lookup = btrfs_lookup, |
fdebe2bd | 10960 | .permission = btrfs_permission, |
93fd63c2 | 10961 | .update_time = btrfs_update_time, |
39279cc3 | 10962 | }; |
76dda93c | 10963 | |
828c0950 | 10964 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10965 | .llseek = generic_file_llseek, |
10966 | .read = generic_read_dir, | |
02dbfc99 | 10967 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 10968 | .open = btrfs_opendir, |
34287aa3 | 10969 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10970 | #ifdef CONFIG_COMPAT |
4c63c245 | 10971 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10972 | #endif |
6bf13c0c | 10973 | .release = btrfs_release_file, |
e02119d5 | 10974 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10975 | }; |
10976 | ||
20e5506b | 10977 | static const struct extent_io_ops btrfs_extent_io_ops = { |
4d53dddb | 10978 | /* mandatory callbacks */ |
065631f6 | 10979 | .submit_bio_hook = btrfs_submit_bio_hook, |
07157aac CM |
10980 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
10981 | }; | |
10982 | ||
35054394 CM |
10983 | /* |
10984 | * btrfs doesn't support the bmap operation because swapfiles | |
10985 | * use bmap to make a mapping of extents in the file. They assume | |
10986 | * these extents won't change over the life of the file and they | |
10987 | * use the bmap result to do IO directly to the drive. | |
10988 | * | |
10989 | * the btrfs bmap call would return logical addresses that aren't | |
10990 | * suitable for IO and they also will change frequently as COW | |
10991 | * operations happen. So, swapfile + btrfs == corruption. | |
10992 | * | |
10993 | * For now we're avoiding this by dropping bmap. | |
10994 | */ | |
7f09410b | 10995 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10996 | .readpage = btrfs_readpage, |
10997 | .writepage = btrfs_writepage, | |
b293f02e | 10998 | .writepages = btrfs_writepages, |
3ab2fb5a | 10999 | .readpages = btrfs_readpages, |
16432985 | 11000 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
11001 | .invalidatepage = btrfs_invalidatepage, |
11002 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 11003 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 11004 | .error_remove_page = generic_error_remove_page, |
ed46ff3d OS |
11005 | .swap_activate = btrfs_swap_activate, |
11006 | .swap_deactivate = btrfs_swap_deactivate, | |
39279cc3 CM |
11007 | }; |
11008 | ||
6e1d5dcc | 11009 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
11010 | .getattr = btrfs_getattr, |
11011 | .setattr = btrfs_setattr, | |
5103e947 | 11012 | .listxattr = btrfs_listxattr, |
fdebe2bd | 11013 | .permission = btrfs_permission, |
1506fcc8 | 11014 | .fiemap = btrfs_fiemap, |
4e34e719 | 11015 | .get_acl = btrfs_get_acl, |
996a710d | 11016 | .set_acl = btrfs_set_acl, |
e41f941a | 11017 | .update_time = btrfs_update_time, |
39279cc3 | 11018 | }; |
6e1d5dcc | 11019 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
11020 | .getattr = btrfs_getattr, |
11021 | .setattr = btrfs_setattr, | |
fdebe2bd | 11022 | .permission = btrfs_permission, |
33268eaf | 11023 | .listxattr = btrfs_listxattr, |
4e34e719 | 11024 | .get_acl = btrfs_get_acl, |
996a710d | 11025 | .set_acl = btrfs_set_acl, |
e41f941a | 11026 | .update_time = btrfs_update_time, |
618e21d5 | 11027 | }; |
6e1d5dcc | 11028 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 11029 | .get_link = page_get_link, |
f209561a | 11030 | .getattr = btrfs_getattr, |
22c44fe6 | 11031 | .setattr = btrfs_setattr, |
fdebe2bd | 11032 | .permission = btrfs_permission, |
0279b4cd | 11033 | .listxattr = btrfs_listxattr, |
e41f941a | 11034 | .update_time = btrfs_update_time, |
39279cc3 | 11035 | }; |
76dda93c | 11036 | |
82d339d9 | 11037 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
11038 | .d_delete = btrfs_dentry_delete, |
11039 | }; |