]>
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> |
92d32170 | 31 | #include <asm/unaligned.h> |
39279cc3 CM |
32 | #include "ctree.h" |
33 | #include "disk-io.h" | |
34 | #include "transaction.h" | |
35 | #include "btrfs_inode.h" | |
39279cc3 | 36 | #include "print-tree.h" |
e6dcd2dc | 37 | #include "ordered-data.h" |
95819c05 | 38 | #include "xattr.h" |
e02119d5 | 39 | #include "tree-log.h" |
4a54c8c1 | 40 | #include "volumes.h" |
c8b97818 | 41 | #include "compression.h" |
b4ce94de | 42 | #include "locking.h" |
dc89e982 | 43 | #include "free-space-cache.h" |
581bb050 | 44 | #include "inode-map.h" |
38c227d8 | 45 | #include "backref.h" |
63541927 | 46 | #include "props.h" |
31193213 | 47 | #include "qgroup.h" |
dda3245e | 48 | #include "dedupe.h" |
39279cc3 CM |
49 | |
50 | struct btrfs_iget_args { | |
90d3e592 | 51 | struct btrfs_key *location; |
39279cc3 CM |
52 | struct btrfs_root *root; |
53 | }; | |
54 | ||
f28a4928 | 55 | struct btrfs_dio_data { |
f28a4928 FM |
56 | u64 reserve; |
57 | u64 unsubmitted_oe_range_start; | |
58 | u64 unsubmitted_oe_range_end; | |
4aaedfb0 | 59 | int overwrite; |
f28a4928 FM |
60 | }; |
61 | ||
6e1d5dcc AD |
62 | static const struct inode_operations btrfs_dir_inode_operations; |
63 | static const struct inode_operations btrfs_symlink_inode_operations; | |
64 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
65 | static const struct inode_operations btrfs_special_inode_operations; | |
66 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b | 67 | static const struct address_space_operations btrfs_aops; |
828c0950 | 68 | static const struct file_operations btrfs_dir_file_operations; |
20e5506b | 69 | static const struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
70 | |
71 | static struct kmem_cache *btrfs_inode_cachep; | |
72 | struct kmem_cache *btrfs_trans_handle_cachep; | |
39279cc3 | 73 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 74 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
75 | |
76 | #define S_SHIFT 12 | |
4d4ab6d6 | 77 | static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { |
39279cc3 CM |
78 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, |
79 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
80 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
81 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
82 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
83 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
84 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
85 | }; | |
86 | ||
3972f260 | 87 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
213e8c55 | 88 | static int btrfs_truncate(struct inode *inode, bool skip_writeback); |
5fd02043 | 89 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
90 | static noinline int cow_file_range(struct inode *inode, |
91 | struct page *locked_page, | |
dda3245e WX |
92 | u64 start, u64 end, u64 delalloc_end, |
93 | int *page_started, unsigned long *nr_written, | |
94 | int unlock, struct btrfs_dedupe_hash *hash); | |
6f9994db LB |
95 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, |
96 | u64 orig_start, u64 block_start, | |
97 | u64 block_len, u64 orig_block_len, | |
98 | u64 ram_bytes, int compress_type, | |
99 | int type); | |
7b128766 | 100 | |
52427260 QW |
101 | static void __endio_write_update_ordered(struct inode *inode, |
102 | const u64 offset, const u64 bytes, | |
103 | const bool uptodate); | |
104 | ||
105 | /* | |
106 | * Cleanup all submitted ordered extents in specified range to handle errors | |
107 | * from the fill_dellaloc() callback. | |
108 | * | |
109 | * NOTE: caller must ensure that when an error happens, it can not call | |
110 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
111 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
112 | * to be released, which we want to happen only when finishing the ordered | |
d1051d6e | 113 | * extent (btrfs_finish_ordered_io()). |
52427260 QW |
114 | */ |
115 | static inline void btrfs_cleanup_ordered_extents(struct inode *inode, | |
d1051d6e NB |
116 | struct page *locked_page, |
117 | u64 offset, u64 bytes) | |
52427260 | 118 | { |
63d71450 NA |
119 | unsigned long index = offset >> PAGE_SHIFT; |
120 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
d1051d6e NB |
121 | u64 page_start = page_offset(locked_page); |
122 | u64 page_end = page_start + PAGE_SIZE - 1; | |
123 | ||
63d71450 NA |
124 | struct page *page; |
125 | ||
126 | while (index <= end_index) { | |
127 | page = find_get_page(inode->i_mapping, index); | |
128 | index++; | |
129 | if (!page) | |
130 | continue; | |
131 | ClearPagePrivate2(page); | |
132 | put_page(page); | |
133 | } | |
d1051d6e NB |
134 | |
135 | /* | |
136 | * In case this page belongs to the delalloc range being instantiated | |
137 | * then skip it, since the first page of a range is going to be | |
138 | * properly cleaned up by the caller of run_delalloc_range | |
139 | */ | |
140 | if (page_start >= offset && page_end <= (offset + bytes - 1)) { | |
141 | offset += PAGE_SIZE; | |
142 | bytes -= PAGE_SIZE; | |
143 | } | |
144 | ||
145 | return __endio_write_update_ordered(inode, offset, bytes, false); | |
52427260 QW |
146 | } |
147 | ||
48a3b636 | 148 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 149 | |
6a3891c5 JB |
150 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
151 | void btrfs_test_inode_set_ops(struct inode *inode) | |
152 | { | |
153 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
154 | } | |
155 | #endif | |
156 | ||
f34f57a3 | 157 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
158 | struct inode *inode, struct inode *dir, |
159 | const struct qstr *qstr) | |
0279b4cd JO |
160 | { |
161 | int err; | |
162 | ||
f34f57a3 | 163 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 164 | if (!err) |
2a7dba39 | 165 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
166 | return err; |
167 | } | |
168 | ||
c8b97818 CM |
169 | /* |
170 | * this does all the hard work for inserting an inline extent into | |
171 | * the btree. The caller should have done a btrfs_drop_extents so that | |
172 | * no overlapping inline items exist in the btree | |
173 | */ | |
40f76580 | 174 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 175 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
176 | struct btrfs_root *root, struct inode *inode, |
177 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 178 | int compress_type, |
c8b97818 CM |
179 | struct page **compressed_pages) |
180 | { | |
c8b97818 CM |
181 | struct extent_buffer *leaf; |
182 | struct page *page = NULL; | |
183 | char *kaddr; | |
184 | unsigned long ptr; | |
185 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
186 | int ret; |
187 | size_t cur_size = size; | |
c8b97818 | 188 | unsigned long offset; |
c8b97818 | 189 | |
fe3f566c | 190 | if (compressed_size && compressed_pages) |
c8b97818 | 191 | cur_size = compressed_size; |
c8b97818 | 192 | |
1acae57b | 193 | inode_add_bytes(inode, size); |
c8b97818 | 194 | |
1acae57b FDBM |
195 | if (!extent_inserted) { |
196 | struct btrfs_key key; | |
197 | size_t datasize; | |
c8b97818 | 198 | |
4a0cc7ca | 199 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b | 200 | key.offset = start; |
962a298f | 201 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 202 | |
1acae57b FDBM |
203 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
204 | path->leave_spinning = 1; | |
205 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
206 | datasize); | |
79b4f4c6 | 207 | if (ret) |
1acae57b | 208 | goto fail; |
c8b97818 CM |
209 | } |
210 | leaf = path->nodes[0]; | |
211 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
212 | struct btrfs_file_extent_item); | |
213 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
214 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
215 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
216 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
217 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
218 | ptr = btrfs_file_extent_inline_start(ei); | |
219 | ||
261507a0 | 220 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
221 | struct page *cpage; |
222 | int i = 0; | |
d397712b | 223 | while (compressed_size > 0) { |
c8b97818 | 224 | cpage = compressed_pages[i]; |
5b050f04 | 225 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 226 | PAGE_SIZE); |
c8b97818 | 227 | |
7ac687d9 | 228 | kaddr = kmap_atomic(cpage); |
c8b97818 | 229 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 230 | kunmap_atomic(kaddr); |
c8b97818 CM |
231 | |
232 | i++; | |
233 | ptr += cur_size; | |
234 | compressed_size -= cur_size; | |
235 | } | |
236 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 237 | compress_type); |
c8b97818 CM |
238 | } else { |
239 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 240 | start >> PAGE_SHIFT); |
c8b97818 | 241 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 242 | kaddr = kmap_atomic(page); |
7073017a | 243 | offset = offset_in_page(start); |
c8b97818 | 244 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 245 | kunmap_atomic(kaddr); |
09cbfeaf | 246 | put_page(page); |
c8b97818 CM |
247 | } |
248 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 249 | btrfs_release_path(path); |
c8b97818 | 250 | |
c2167754 YZ |
251 | /* |
252 | * we're an inline extent, so nobody can | |
253 | * extend the file past i_size without locking | |
254 | * a page we already have locked. | |
255 | * | |
256 | * We must do any isize and inode updates | |
257 | * before we unlock the pages. Otherwise we | |
258 | * could end up racing with unlink. | |
259 | */ | |
c8b97818 | 260 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 261 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 262 | |
c8b97818 | 263 | fail: |
79b4f4c6 | 264 | return ret; |
c8b97818 CM |
265 | } |
266 | ||
267 | ||
268 | /* | |
269 | * conditionally insert an inline extent into the file. This | |
270 | * does the checks required to make sure the data is small enough | |
271 | * to fit as an inline extent. | |
272 | */ | |
d02c0e20 | 273 | static noinline int cow_file_range_inline(struct inode *inode, u64 start, |
00361589 JB |
274 | u64 end, size_t compressed_size, |
275 | int compress_type, | |
276 | struct page **compressed_pages) | |
c8b97818 | 277 | { |
d02c0e20 | 278 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0b246afa | 279 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 280 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
281 | u64 isize = i_size_read(inode); |
282 | u64 actual_end = min(end + 1, isize); | |
283 | u64 inline_len = actual_end - start; | |
0b246afa | 284 | u64 aligned_end = ALIGN(end, fs_info->sectorsize); |
c8b97818 CM |
285 | u64 data_len = inline_len; |
286 | int ret; | |
1acae57b FDBM |
287 | struct btrfs_path *path; |
288 | int extent_inserted = 0; | |
289 | u32 extent_item_size; | |
c8b97818 CM |
290 | |
291 | if (compressed_size) | |
292 | data_len = compressed_size; | |
293 | ||
294 | if (start > 0 || | |
0b246afa JM |
295 | actual_end > fs_info->sectorsize || |
296 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || | |
c8b97818 | 297 | (!compressed_size && |
0b246afa | 298 | (actual_end & (fs_info->sectorsize - 1)) == 0) || |
c8b97818 | 299 | end + 1 < isize || |
0b246afa | 300 | data_len > fs_info->max_inline) { |
c8b97818 CM |
301 | return 1; |
302 | } | |
303 | ||
1acae57b FDBM |
304 | path = btrfs_alloc_path(); |
305 | if (!path) | |
306 | return -ENOMEM; | |
307 | ||
00361589 | 308 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
309 | if (IS_ERR(trans)) { |
310 | btrfs_free_path(path); | |
00361589 | 311 | return PTR_ERR(trans); |
1acae57b | 312 | } |
69fe2d75 | 313 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
00361589 | 314 | |
1acae57b FDBM |
315 | if (compressed_size && compressed_pages) |
316 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
317 | compressed_size); | |
318 | else | |
319 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
320 | inline_len); | |
321 | ||
322 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
323 | start, aligned_end, NULL, | |
324 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 | 325 | if (ret) { |
66642832 | 326 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
327 | goto out; |
328 | } | |
c8b97818 CM |
329 | |
330 | if (isize > actual_end) | |
331 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
332 | ret = insert_inline_extent(trans, path, extent_inserted, |
333 | root, inode, start, | |
c8b97818 | 334 | inline_len, compressed_size, |
fe3f566c | 335 | compress_type, compressed_pages); |
2adcac1a | 336 | if (ret && ret != -ENOSPC) { |
66642832 | 337 | btrfs_abort_transaction(trans, ret); |
00361589 | 338 | goto out; |
2adcac1a | 339 | } else if (ret == -ENOSPC) { |
00361589 JB |
340 | ret = 1; |
341 | goto out; | |
79787eaa | 342 | } |
2adcac1a | 343 | |
bdc20e67 | 344 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
dcdbc059 | 345 | btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0); |
00361589 | 346 | out: |
94ed938a QW |
347 | /* |
348 | * Don't forget to free the reserved space, as for inlined extent | |
349 | * it won't count as data extent, free them directly here. | |
350 | * And at reserve time, it's always aligned to page size, so | |
351 | * just free one page here. | |
352 | */ | |
bc42bda2 | 353 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 354 | btrfs_free_path(path); |
3a45bb20 | 355 | btrfs_end_transaction(trans); |
00361589 | 356 | return ret; |
c8b97818 CM |
357 | } |
358 | ||
771ed689 CM |
359 | struct async_extent { |
360 | u64 start; | |
361 | u64 ram_size; | |
362 | u64 compressed_size; | |
363 | struct page **pages; | |
364 | unsigned long nr_pages; | |
261507a0 | 365 | int compress_type; |
771ed689 CM |
366 | struct list_head list; |
367 | }; | |
368 | ||
369 | struct async_cow { | |
370 | struct inode *inode; | |
600b6cf4 | 371 | struct btrfs_fs_info *fs_info; |
771ed689 CM |
372 | struct page *locked_page; |
373 | u64 start; | |
374 | u64 end; | |
f82b7359 | 375 | unsigned int write_flags; |
771ed689 CM |
376 | struct list_head extents; |
377 | struct btrfs_work work; | |
378 | }; | |
379 | ||
380 | static noinline int add_async_extent(struct async_cow *cow, | |
381 | u64 start, u64 ram_size, | |
382 | u64 compressed_size, | |
383 | struct page **pages, | |
261507a0 LZ |
384 | unsigned long nr_pages, |
385 | int compress_type) | |
771ed689 CM |
386 | { |
387 | struct async_extent *async_extent; | |
388 | ||
389 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 390 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
391 | async_extent->start = start; |
392 | async_extent->ram_size = ram_size; | |
393 | async_extent->compressed_size = compressed_size; | |
394 | async_extent->pages = pages; | |
395 | async_extent->nr_pages = nr_pages; | |
261507a0 | 396 | async_extent->compress_type = compress_type; |
771ed689 CM |
397 | list_add_tail(&async_extent->list, &cow->extents); |
398 | return 0; | |
399 | } | |
400 | ||
c2fcdcdf | 401 | static inline int inode_need_compress(struct inode *inode, u64 start, u64 end) |
f79707b0 | 402 | { |
0b246afa | 403 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
f79707b0 WS |
404 | |
405 | /* force compress */ | |
0b246afa | 406 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 407 | return 1; |
eec63c65 DS |
408 | /* defrag ioctl */ |
409 | if (BTRFS_I(inode)->defrag_compress) | |
410 | return 1; | |
f79707b0 WS |
411 | /* bad compression ratios */ |
412 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
413 | return 0; | |
0b246afa | 414 | if (btrfs_test_opt(fs_info, COMPRESS) || |
f79707b0 | 415 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || |
b52aa8c9 | 416 | BTRFS_I(inode)->prop_compress) |
c2fcdcdf | 417 | return btrfs_compress_heuristic(inode, start, end); |
f79707b0 WS |
418 | return 0; |
419 | } | |
420 | ||
6158e1ce | 421 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
26d30f85 AJ |
422 | u64 start, u64 end, u64 num_bytes, u64 small_write) |
423 | { | |
424 | /* If this is a small write inside eof, kick off a defrag */ | |
425 | if (num_bytes < small_write && | |
6158e1ce | 426 | (start > 0 || end + 1 < inode->disk_i_size)) |
26d30f85 AJ |
427 | btrfs_add_inode_defrag(NULL, inode); |
428 | } | |
429 | ||
d352ac68 | 430 | /* |
771ed689 CM |
431 | * we create compressed extents in two phases. The first |
432 | * phase compresses a range of pages that have already been | |
433 | * locked (both pages and state bits are locked). | |
c8b97818 | 434 | * |
771ed689 CM |
435 | * This is done inside an ordered work queue, and the compression |
436 | * is spread across many cpus. The actual IO submission is step | |
437 | * two, and the ordered work queue takes care of making sure that | |
438 | * happens in the same order things were put onto the queue by | |
439 | * writepages and friends. | |
c8b97818 | 440 | * |
771ed689 CM |
441 | * If this code finds it can't get good compression, it puts an |
442 | * entry onto the work queue to write the uncompressed bytes. This | |
443 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
444 | * are written in the same order that the flusher thread sent them |
445 | * down. | |
d352ac68 | 446 | */ |
c44f649e | 447 | static noinline void compress_file_range(struct inode *inode, |
771ed689 CM |
448 | struct page *locked_page, |
449 | u64 start, u64 end, | |
450 | struct async_cow *async_cow, | |
451 | int *num_added) | |
b888db2b | 452 | { |
0b246afa | 453 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
0b246afa | 454 | u64 blocksize = fs_info->sectorsize; |
c8b97818 | 455 | u64 actual_end; |
42dc7bab | 456 | u64 isize = i_size_read(inode); |
e6dcd2dc | 457 | int ret = 0; |
c8b97818 CM |
458 | struct page **pages = NULL; |
459 | unsigned long nr_pages; | |
c8b97818 CM |
460 | unsigned long total_compressed = 0; |
461 | unsigned long total_in = 0; | |
c8b97818 CM |
462 | int i; |
463 | int will_compress; | |
0b246afa | 464 | int compress_type = fs_info->compress_type; |
4adaa611 | 465 | int redirty = 0; |
b888db2b | 466 | |
6158e1ce NB |
467 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
468 | SZ_16K); | |
4cb5300b | 469 | |
42dc7bab | 470 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
471 | again: |
472 | will_compress = 0; | |
09cbfeaf | 473 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
474 | BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0); |
475 | nr_pages = min_t(unsigned long, nr_pages, | |
476 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 477 | |
f03d9301 CM |
478 | /* |
479 | * we don't want to send crud past the end of i_size through | |
480 | * compression, that's just a waste of CPU time. So, if the | |
481 | * end of the file is before the start of our current | |
482 | * requested range of bytes, we bail out to the uncompressed | |
483 | * cleanup code that can deal with all of this. | |
484 | * | |
485 | * It isn't really the fastest way to fix things, but this is a | |
486 | * very uncommon corner. | |
487 | */ | |
488 | if (actual_end <= start) | |
489 | goto cleanup_and_bail_uncompressed; | |
490 | ||
c8b97818 CM |
491 | total_compressed = actual_end - start; |
492 | ||
4bcbb332 SW |
493 | /* |
494 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 495 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
496 | */ |
497 | if (total_compressed <= blocksize && | |
498 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
499 | goto cleanup_and_bail_uncompressed; | |
500 | ||
069eac78 DS |
501 | total_compressed = min_t(unsigned long, total_compressed, |
502 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
503 | total_in = 0; |
504 | ret = 0; | |
db94535d | 505 | |
771ed689 CM |
506 | /* |
507 | * we do compression for mount -o compress and when the | |
508 | * inode has not been flagged as nocompress. This flag can | |
509 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 510 | */ |
c2fcdcdf | 511 | if (inode_need_compress(inode, start, end)) { |
c8b97818 | 512 | WARN_ON(pages); |
31e818fe | 513 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
514 | if (!pages) { |
515 | /* just bail out to the uncompressed code */ | |
3527a018 | 516 | nr_pages = 0; |
560f7d75 LZ |
517 | goto cont; |
518 | } | |
c8b97818 | 519 | |
eec63c65 DS |
520 | if (BTRFS_I(inode)->defrag_compress) |
521 | compress_type = BTRFS_I(inode)->defrag_compress; | |
522 | else if (BTRFS_I(inode)->prop_compress) | |
b52aa8c9 | 523 | compress_type = BTRFS_I(inode)->prop_compress; |
261507a0 | 524 | |
4adaa611 CM |
525 | /* |
526 | * we need to call clear_page_dirty_for_io on each | |
527 | * page in the range. Otherwise applications with the file | |
528 | * mmap'd can wander in and change the page contents while | |
529 | * we are compressing them. | |
530 | * | |
531 | * If the compression fails for any reason, we set the pages | |
532 | * dirty again later on. | |
e9679de3 TT |
533 | * |
534 | * Note that the remaining part is redirtied, the start pointer | |
535 | * has moved, the end is the original one. | |
4adaa611 | 536 | */ |
e9679de3 TT |
537 | if (!redirty) { |
538 | extent_range_clear_dirty_for_io(inode, start, end); | |
539 | redirty = 1; | |
540 | } | |
f51d2b59 DS |
541 | |
542 | /* Compression level is applied here and only here */ | |
543 | ret = btrfs_compress_pages( | |
544 | compress_type | (fs_info->compress_level << 4), | |
261507a0 | 545 | inode->i_mapping, start, |
38c31464 | 546 | pages, |
4d3a800e | 547 | &nr_pages, |
261507a0 | 548 | &total_in, |
e5d74902 | 549 | &total_compressed); |
c8b97818 CM |
550 | |
551 | if (!ret) { | |
7073017a | 552 | unsigned long offset = offset_in_page(total_compressed); |
4d3a800e | 553 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
554 | char *kaddr; |
555 | ||
556 | /* zero the tail end of the last page, we might be | |
557 | * sending it down to disk | |
558 | */ | |
559 | if (offset) { | |
7ac687d9 | 560 | kaddr = kmap_atomic(page); |
c8b97818 | 561 | memset(kaddr + offset, 0, |
09cbfeaf | 562 | PAGE_SIZE - offset); |
7ac687d9 | 563 | kunmap_atomic(kaddr); |
c8b97818 CM |
564 | } |
565 | will_compress = 1; | |
566 | } | |
567 | } | |
560f7d75 | 568 | cont: |
c8b97818 CM |
569 | if (start == 0) { |
570 | /* lets try to make an inline extent */ | |
6018ba0a | 571 | if (ret || total_in < actual_end) { |
c8b97818 | 572 | /* we didn't compress the entire range, try |
771ed689 | 573 | * to make an uncompressed inline extent. |
c8b97818 | 574 | */ |
d02c0e20 NB |
575 | ret = cow_file_range_inline(inode, start, end, 0, |
576 | BTRFS_COMPRESS_NONE, NULL); | |
c8b97818 | 577 | } else { |
771ed689 | 578 | /* try making a compressed inline extent */ |
d02c0e20 | 579 | ret = cow_file_range_inline(inode, start, end, |
fe3f566c LZ |
580 | total_compressed, |
581 | compress_type, pages); | |
c8b97818 | 582 | } |
79787eaa | 583 | if (ret <= 0) { |
151a41bc | 584 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
585 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
586 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 FM |
587 | unsigned long page_error_op; |
588 | ||
e6eb4314 | 589 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 590 | |
771ed689 | 591 | /* |
79787eaa JM |
592 | * inline extent creation worked or returned error, |
593 | * we don't need to create any more async work items. | |
594 | * Unlock and free up our temp pages. | |
8b62f87b JB |
595 | * |
596 | * We use DO_ACCOUNTING here because we need the | |
597 | * delalloc_release_metadata to be done _after_ we drop | |
598 | * our outstanding extent for clearing delalloc for this | |
599 | * range. | |
771ed689 | 600 | */ |
ba8b04c1 QW |
601 | extent_clear_unlock_delalloc(inode, start, end, end, |
602 | NULL, clear_flags, | |
603 | PAGE_UNLOCK | | |
c2790a2e JB |
604 | PAGE_CLEAR_DIRTY | |
605 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 606 | page_error_op | |
c2790a2e | 607 | PAGE_END_WRITEBACK); |
c8b97818 CM |
608 | goto free_pages_out; |
609 | } | |
610 | } | |
611 | ||
612 | if (will_compress) { | |
613 | /* | |
614 | * we aren't doing an inline extent round the compressed size | |
615 | * up to a block size boundary so the allocator does sane | |
616 | * things | |
617 | */ | |
fda2832f | 618 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
619 | |
620 | /* | |
621 | * one last check to make sure the compression is really a | |
170607eb TT |
622 | * win, compare the page count read with the blocks on disk, |
623 | * compression must free at least one sector size | |
c8b97818 | 624 | */ |
09cbfeaf | 625 | total_in = ALIGN(total_in, PAGE_SIZE); |
170607eb | 626 | if (total_compressed + blocksize <= total_in) { |
c8bb0c8b AS |
627 | *num_added += 1; |
628 | ||
629 | /* | |
630 | * The async work queues will take care of doing actual | |
631 | * allocation on disk for these compressed pages, and | |
632 | * will submit them to the elevator. | |
633 | */ | |
1170862d | 634 | add_async_extent(async_cow, start, total_in, |
4d3a800e | 635 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
636 | compress_type); |
637 | ||
1170862d TT |
638 | if (start + total_in < end) { |
639 | start += total_in; | |
c8bb0c8b AS |
640 | pages = NULL; |
641 | cond_resched(); | |
642 | goto again; | |
643 | } | |
644 | return; | |
c8b97818 CM |
645 | } |
646 | } | |
c8bb0c8b | 647 | if (pages) { |
c8b97818 CM |
648 | /* |
649 | * the compression code ran but failed to make things smaller, | |
650 | * free any pages it allocated and our page pointer array | |
651 | */ | |
4d3a800e | 652 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 653 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 654 | put_page(pages[i]); |
c8b97818 CM |
655 | } |
656 | kfree(pages); | |
657 | pages = NULL; | |
658 | total_compressed = 0; | |
4d3a800e | 659 | nr_pages = 0; |
c8b97818 CM |
660 | |
661 | /* flag the file so we don't compress in the future */ | |
0b246afa | 662 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
b52aa8c9 | 663 | !(BTRFS_I(inode)->prop_compress)) { |
a555f810 | 664 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 665 | } |
c8b97818 | 666 | } |
f03d9301 | 667 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
668 | /* |
669 | * No compression, but we still need to write the pages in the file | |
670 | * we've been given so far. redirty the locked page if it corresponds | |
671 | * to our extent and set things up for the async work queue to run | |
672 | * cow_file_range to do the normal delalloc dance. | |
673 | */ | |
674 | if (page_offset(locked_page) >= start && | |
675 | page_offset(locked_page) <= end) | |
676 | __set_page_dirty_nobuffers(locked_page); | |
677 | /* unlocked later on in the async handlers */ | |
678 | ||
679 | if (redirty) | |
680 | extent_range_redirty_for_io(inode, start, end); | |
681 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0, | |
682 | BTRFS_COMPRESS_NONE); | |
683 | *num_added += 1; | |
3b951516 | 684 | |
c44f649e | 685 | return; |
771ed689 CM |
686 | |
687 | free_pages_out: | |
4d3a800e | 688 | for (i = 0; i < nr_pages; i++) { |
771ed689 | 689 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 690 | put_page(pages[i]); |
771ed689 | 691 | } |
d397712b | 692 | kfree(pages); |
771ed689 | 693 | } |
771ed689 | 694 | |
40ae837b FM |
695 | static void free_async_extent_pages(struct async_extent *async_extent) |
696 | { | |
697 | int i; | |
698 | ||
699 | if (!async_extent->pages) | |
700 | return; | |
701 | ||
702 | for (i = 0; i < async_extent->nr_pages; i++) { | |
703 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 704 | put_page(async_extent->pages[i]); |
40ae837b FM |
705 | } |
706 | kfree(async_extent->pages); | |
707 | async_extent->nr_pages = 0; | |
708 | async_extent->pages = NULL; | |
771ed689 CM |
709 | } |
710 | ||
711 | /* | |
712 | * phase two of compressed writeback. This is the ordered portion | |
713 | * of the code, which only gets called in the order the work was | |
714 | * queued. We walk all the async extents created by compress_file_range | |
715 | * and send them down to the disk. | |
716 | */ | |
dec8f175 | 717 | static noinline void submit_compressed_extents(struct inode *inode, |
771ed689 CM |
718 | struct async_cow *async_cow) |
719 | { | |
0b246afa | 720 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
721 | struct async_extent *async_extent; |
722 | u64 alloc_hint = 0; | |
771ed689 CM |
723 | struct btrfs_key ins; |
724 | struct extent_map *em; | |
725 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
771ed689 | 726 | struct extent_io_tree *io_tree; |
f5a84ee3 | 727 | int ret = 0; |
771ed689 | 728 | |
3e04e7f1 | 729 | again: |
d397712b | 730 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
731 | async_extent = list_entry(async_cow->extents.next, |
732 | struct async_extent, list); | |
733 | list_del(&async_extent->list); | |
c8b97818 | 734 | |
771ed689 CM |
735 | io_tree = &BTRFS_I(inode)->io_tree; |
736 | ||
f5a84ee3 | 737 | retry: |
771ed689 CM |
738 | /* did the compression code fall back to uncompressed IO? */ |
739 | if (!async_extent->pages) { | |
740 | int page_started = 0; | |
741 | unsigned long nr_written = 0; | |
742 | ||
743 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 744 | async_extent->start + |
d0082371 | 745 | async_extent->ram_size - 1); |
771ed689 CM |
746 | |
747 | /* allocate blocks */ | |
f5a84ee3 JB |
748 | ret = cow_file_range(inode, async_cow->locked_page, |
749 | async_extent->start, | |
750 | async_extent->start + | |
751 | async_extent->ram_size - 1, | |
dda3245e WX |
752 | async_extent->start + |
753 | async_extent->ram_size - 1, | |
754 | &page_started, &nr_written, 0, | |
755 | NULL); | |
771ed689 | 756 | |
79787eaa JM |
757 | /* JDM XXX */ |
758 | ||
771ed689 CM |
759 | /* |
760 | * if page_started, cow_file_range inserted an | |
761 | * inline extent and took care of all the unlocking | |
762 | * and IO for us. Otherwise, we need to submit | |
763 | * all those pages down to the drive. | |
764 | */ | |
f5a84ee3 | 765 | if (!page_started && !ret) |
5e3ee236 NB |
766 | extent_write_locked_range(inode, |
767 | async_extent->start, | |
d397712b | 768 | async_extent->start + |
771ed689 | 769 | async_extent->ram_size - 1, |
771ed689 | 770 | WB_SYNC_ALL); |
3e04e7f1 JB |
771 | else if (ret) |
772 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
773 | kfree(async_extent); |
774 | cond_resched(); | |
775 | continue; | |
776 | } | |
777 | ||
778 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 779 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 780 | |
18513091 | 781 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
771ed689 CM |
782 | async_extent->compressed_size, |
783 | async_extent->compressed_size, | |
e570fd27 | 784 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 785 | if (ret) { |
40ae837b | 786 | free_async_extent_pages(async_extent); |
3e04e7f1 | 787 | |
fdf8e2ea JB |
788 | if (ret == -ENOSPC) { |
789 | unlock_extent(io_tree, async_extent->start, | |
790 | async_extent->start + | |
791 | async_extent->ram_size - 1); | |
ce62003f LB |
792 | |
793 | /* | |
794 | * we need to redirty the pages if we decide to | |
795 | * fallback to uncompressed IO, otherwise we | |
796 | * will not submit these pages down to lower | |
797 | * layers. | |
798 | */ | |
799 | extent_range_redirty_for_io(inode, | |
800 | async_extent->start, | |
801 | async_extent->start + | |
802 | async_extent->ram_size - 1); | |
803 | ||
79787eaa | 804 | goto retry; |
fdf8e2ea | 805 | } |
3e04e7f1 | 806 | goto out_free; |
f5a84ee3 | 807 | } |
c2167754 YZ |
808 | /* |
809 | * here we're doing allocation and writeback of the | |
810 | * compressed pages | |
811 | */ | |
6f9994db LB |
812 | em = create_io_em(inode, async_extent->start, |
813 | async_extent->ram_size, /* len */ | |
814 | async_extent->start, /* orig_start */ | |
815 | ins.objectid, /* block_start */ | |
816 | ins.offset, /* block_len */ | |
817 | ins.offset, /* orig_block_len */ | |
818 | async_extent->ram_size, /* ram_bytes */ | |
819 | async_extent->compress_type, | |
820 | BTRFS_ORDERED_COMPRESSED); | |
821 | if (IS_ERR(em)) | |
822 | /* ret value is not necessary due to void function */ | |
3e04e7f1 | 823 | goto out_free_reserve; |
6f9994db | 824 | free_extent_map(em); |
3e04e7f1 | 825 | |
261507a0 LZ |
826 | ret = btrfs_add_ordered_extent_compress(inode, |
827 | async_extent->start, | |
828 | ins.objectid, | |
829 | async_extent->ram_size, | |
830 | ins.offset, | |
831 | BTRFS_ORDERED_COMPRESSED, | |
832 | async_extent->compress_type); | |
d9f85963 | 833 | if (ret) { |
dcdbc059 NB |
834 | btrfs_drop_extent_cache(BTRFS_I(inode), |
835 | async_extent->start, | |
d9f85963 FM |
836 | async_extent->start + |
837 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 838 | goto out_free_reserve; |
d9f85963 | 839 | } |
0b246afa | 840 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
771ed689 | 841 | |
771ed689 CM |
842 | /* |
843 | * clear dirty, set writeback and unlock the pages. | |
844 | */ | |
c2790a2e | 845 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
846 | async_extent->start + |
847 | async_extent->ram_size - 1, | |
a791e35e CM |
848 | async_extent->start + |
849 | async_extent->ram_size - 1, | |
151a41bc JB |
850 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
851 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 852 | PAGE_SET_WRITEBACK); |
4e4cbee9 | 853 | if (btrfs_submit_compressed_write(inode, |
d397712b CM |
854 | async_extent->start, |
855 | async_extent->ram_size, | |
856 | ins.objectid, | |
857 | ins.offset, async_extent->pages, | |
f82b7359 LB |
858 | async_extent->nr_pages, |
859 | async_cow->write_flags)) { | |
fce2a4e6 FM |
860 | struct page *p = async_extent->pages[0]; |
861 | const u64 start = async_extent->start; | |
862 | const u64 end = start + async_extent->ram_size - 1; | |
863 | ||
864 | p->mapping = inode->i_mapping; | |
c629732d | 865 | btrfs_writepage_endio_finish_ordered(p, start, end, 0); |
7087a9d8 | 866 | |
fce2a4e6 | 867 | p->mapping = NULL; |
ba8b04c1 QW |
868 | extent_clear_unlock_delalloc(inode, start, end, end, |
869 | NULL, 0, | |
fce2a4e6 FM |
870 | PAGE_END_WRITEBACK | |
871 | PAGE_SET_ERROR); | |
40ae837b | 872 | free_async_extent_pages(async_extent); |
fce2a4e6 | 873 | } |
771ed689 CM |
874 | alloc_hint = ins.objectid + ins.offset; |
875 | kfree(async_extent); | |
876 | cond_resched(); | |
877 | } | |
dec8f175 | 878 | return; |
3e04e7f1 | 879 | out_free_reserve: |
0b246afa | 880 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 881 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 882 | out_free: |
c2790a2e | 883 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
884 | async_extent->start + |
885 | async_extent->ram_size - 1, | |
3e04e7f1 JB |
886 | async_extent->start + |
887 | async_extent->ram_size - 1, | |
c2790a2e | 888 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 889 | EXTENT_DELALLOC_NEW | |
151a41bc JB |
890 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
891 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
892 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
893 | PAGE_SET_ERROR); | |
40ae837b | 894 | free_async_extent_pages(async_extent); |
79787eaa | 895 | kfree(async_extent); |
3e04e7f1 | 896 | goto again; |
771ed689 CM |
897 | } |
898 | ||
4b46fce2 JB |
899 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
900 | u64 num_bytes) | |
901 | { | |
902 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
903 | struct extent_map *em; | |
904 | u64 alloc_hint = 0; | |
905 | ||
906 | read_lock(&em_tree->lock); | |
907 | em = search_extent_mapping(em_tree, start, num_bytes); | |
908 | if (em) { | |
909 | /* | |
910 | * if block start isn't an actual block number then find the | |
911 | * first block in this inode and use that as a hint. If that | |
912 | * block is also bogus then just don't worry about it. | |
913 | */ | |
914 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
915 | free_extent_map(em); | |
916 | em = search_extent_mapping(em_tree, 0, 0); | |
917 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
918 | alloc_hint = em->block_start; | |
919 | if (em) | |
920 | free_extent_map(em); | |
921 | } else { | |
922 | alloc_hint = em->block_start; | |
923 | free_extent_map(em); | |
924 | } | |
925 | } | |
926 | read_unlock(&em_tree->lock); | |
927 | ||
928 | return alloc_hint; | |
929 | } | |
930 | ||
771ed689 CM |
931 | /* |
932 | * when extent_io.c finds a delayed allocation range in the file, | |
933 | * the call backs end up in this code. The basic idea is to | |
934 | * allocate extents on disk for the range, and create ordered data structs | |
935 | * in ram to track those extents. | |
936 | * | |
937 | * locked_page is the page that writepage had locked already. We use | |
938 | * it to make sure we don't do extra locks or unlocks. | |
939 | * | |
940 | * *page_started is set to one if we unlock locked_page and do everything | |
941 | * required to start IO on it. It may be clean and already done with | |
942 | * IO when we return. | |
943 | */ | |
00361589 JB |
944 | static noinline int cow_file_range(struct inode *inode, |
945 | struct page *locked_page, | |
dda3245e WX |
946 | u64 start, u64 end, u64 delalloc_end, |
947 | int *page_started, unsigned long *nr_written, | |
948 | int unlock, struct btrfs_dedupe_hash *hash) | |
771ed689 | 949 | { |
0b246afa | 950 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00361589 | 951 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
952 | u64 alloc_hint = 0; |
953 | u64 num_bytes; | |
954 | unsigned long ram_size; | |
a315e68f | 955 | u64 cur_alloc_size = 0; |
0b246afa | 956 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
957 | struct btrfs_key ins; |
958 | struct extent_map *em; | |
a315e68f FM |
959 | unsigned clear_bits; |
960 | unsigned long page_ops; | |
961 | bool extent_reserved = false; | |
771ed689 CM |
962 | int ret = 0; |
963 | ||
70ddc553 | 964 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
02ecd2c2 | 965 | WARN_ON_ONCE(1); |
29bce2f3 JB |
966 | ret = -EINVAL; |
967 | goto out_unlock; | |
02ecd2c2 | 968 | } |
771ed689 | 969 | |
fda2832f | 970 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 | 971 | num_bytes = max(blocksize, num_bytes); |
566b1760 | 972 | ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 973 | |
6158e1ce | 974 | inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K); |
4cb5300b | 975 | |
771ed689 CM |
976 | if (start == 0) { |
977 | /* lets try to make an inline extent */ | |
d02c0e20 NB |
978 | ret = cow_file_range_inline(inode, start, end, 0, |
979 | BTRFS_COMPRESS_NONE, NULL); | |
771ed689 | 980 | if (ret == 0) { |
8b62f87b JB |
981 | /* |
982 | * We use DO_ACCOUNTING here because we need the | |
983 | * delalloc_release_metadata to be run _after_ we drop | |
984 | * our outstanding extent for clearing delalloc for this | |
985 | * range. | |
986 | */ | |
ba8b04c1 QW |
987 | extent_clear_unlock_delalloc(inode, start, end, |
988 | delalloc_end, NULL, | |
c2790a2e | 989 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
990 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
991 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
c2790a2e JB |
992 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
993 | PAGE_END_WRITEBACK); | |
771ed689 | 994 | *nr_written = *nr_written + |
09cbfeaf | 995 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 996 | *page_started = 1; |
771ed689 | 997 | goto out; |
79787eaa | 998 | } else if (ret < 0) { |
79787eaa | 999 | goto out_unlock; |
771ed689 CM |
1000 | } |
1001 | } | |
1002 | ||
4b46fce2 | 1003 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
dcdbc059 NB |
1004 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
1005 | start + num_bytes - 1, 0); | |
771ed689 | 1006 | |
3752d22f AJ |
1007 | while (num_bytes > 0) { |
1008 | cur_alloc_size = num_bytes; | |
18513091 | 1009 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
0b246afa | 1010 | fs_info->sectorsize, 0, alloc_hint, |
e570fd27 | 1011 | &ins, 1, 1); |
00361589 | 1012 | if (ret < 0) |
79787eaa | 1013 | goto out_unlock; |
a315e68f FM |
1014 | cur_alloc_size = ins.offset; |
1015 | extent_reserved = true; | |
d397712b | 1016 | |
771ed689 | 1017 | ram_size = ins.offset; |
6f9994db LB |
1018 | em = create_io_em(inode, start, ins.offset, /* len */ |
1019 | start, /* orig_start */ | |
1020 | ins.objectid, /* block_start */ | |
1021 | ins.offset, /* block_len */ | |
1022 | ins.offset, /* orig_block_len */ | |
1023 | ram_size, /* ram_bytes */ | |
1024 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1025 | BTRFS_ORDERED_REGULAR /* type */); |
090a127a SY |
1026 | if (IS_ERR(em)) { |
1027 | ret = PTR_ERR(em); | |
ace68bac | 1028 | goto out_reserve; |
090a127a | 1029 | } |
6f9994db | 1030 | free_extent_map(em); |
e6dcd2dc | 1031 | |
e6dcd2dc | 1032 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1033 | ram_size, cur_alloc_size, 0); |
ace68bac | 1034 | if (ret) |
d9f85963 | 1035 | goto out_drop_extent_cache; |
c8b97818 | 1036 | |
17d217fe YZ |
1037 | if (root->root_key.objectid == |
1038 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1039 | ret = btrfs_reloc_clone_csums(inode, start, | |
1040 | cur_alloc_size); | |
4dbd80fb QW |
1041 | /* |
1042 | * Only drop cache here, and process as normal. | |
1043 | * | |
1044 | * We must not allow extent_clear_unlock_delalloc() | |
1045 | * at out_unlock label to free meta of this ordered | |
1046 | * extent, as its meta should be freed by | |
1047 | * btrfs_finish_ordered_io(). | |
1048 | * | |
1049 | * So we must continue until @start is increased to | |
1050 | * skip current ordered extent. | |
1051 | */ | |
00361589 | 1052 | if (ret) |
4dbd80fb QW |
1053 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
1054 | start + ram_size - 1, 0); | |
17d217fe YZ |
1055 | } |
1056 | ||
0b246afa | 1057 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1058 | |
c8b97818 CM |
1059 | /* we're not doing compressed IO, don't unlock the first |
1060 | * page (which the caller expects to stay locked), don't | |
1061 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1062 | * |
1063 | * Do set the Private2 bit so we know this page was properly | |
1064 | * setup for writepage | |
c8b97818 | 1065 | */ |
a315e68f FM |
1066 | page_ops = unlock ? PAGE_UNLOCK : 0; |
1067 | page_ops |= PAGE_SET_PRIVATE2; | |
a791e35e | 1068 | |
c2790a2e | 1069 | extent_clear_unlock_delalloc(inode, start, |
ba8b04c1 QW |
1070 | start + ram_size - 1, |
1071 | delalloc_end, locked_page, | |
c2790a2e | 1072 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1073 | page_ops); |
3752d22f AJ |
1074 | if (num_bytes < cur_alloc_size) |
1075 | num_bytes = 0; | |
4dbd80fb | 1076 | else |
3752d22f | 1077 | num_bytes -= cur_alloc_size; |
c59f8951 CM |
1078 | alloc_hint = ins.objectid + ins.offset; |
1079 | start += cur_alloc_size; | |
a315e68f | 1080 | extent_reserved = false; |
4dbd80fb QW |
1081 | |
1082 | /* | |
1083 | * btrfs_reloc_clone_csums() error, since start is increased | |
1084 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1085 | * free metadata of current ordered extent, we're OK to exit. | |
1086 | */ | |
1087 | if (ret) | |
1088 | goto out_unlock; | |
b888db2b | 1089 | } |
79787eaa | 1090 | out: |
be20aa9d | 1091 | return ret; |
b7d5b0a8 | 1092 | |
d9f85963 | 1093 | out_drop_extent_cache: |
dcdbc059 | 1094 | btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0); |
ace68bac | 1095 | out_reserve: |
0b246afa | 1096 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1097 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1098 | out_unlock: |
a7e3b975 FM |
1099 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1100 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
a315e68f FM |
1101 | page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
1102 | PAGE_END_WRITEBACK; | |
1103 | /* | |
1104 | * If we reserved an extent for our delalloc range (or a subrange) and | |
1105 | * failed to create the respective ordered extent, then it means that | |
1106 | * when we reserved the extent we decremented the extent's size from | |
1107 | * the data space_info's bytes_may_use counter and incremented the | |
1108 | * space_info's bytes_reserved counter by the same amount. We must make | |
1109 | * sure extent_clear_unlock_delalloc() does not try to decrement again | |
1110 | * the data space_info's bytes_may_use counter, therefore we do not pass | |
1111 | * it the flag EXTENT_CLEAR_DATA_RESV. | |
1112 | */ | |
1113 | if (extent_reserved) { | |
1114 | extent_clear_unlock_delalloc(inode, start, | |
1115 | start + cur_alloc_size, | |
1116 | start + cur_alloc_size, | |
1117 | locked_page, | |
1118 | clear_bits, | |
1119 | page_ops); | |
1120 | start += cur_alloc_size; | |
1121 | if (start >= end) | |
1122 | goto out; | |
1123 | } | |
ba8b04c1 QW |
1124 | extent_clear_unlock_delalloc(inode, start, end, delalloc_end, |
1125 | locked_page, | |
a315e68f FM |
1126 | clear_bits | EXTENT_CLEAR_DATA_RESV, |
1127 | page_ops); | |
79787eaa | 1128 | goto out; |
771ed689 | 1129 | } |
c8b97818 | 1130 | |
771ed689 CM |
1131 | /* |
1132 | * work queue call back to started compression on a file and pages | |
1133 | */ | |
1134 | static noinline void async_cow_start(struct btrfs_work *work) | |
1135 | { | |
1136 | struct async_cow *async_cow; | |
1137 | int num_added = 0; | |
1138 | async_cow = container_of(work, struct async_cow, work); | |
1139 | ||
1140 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1141 | async_cow->start, async_cow->end, async_cow, | |
1142 | &num_added); | |
8180ef88 | 1143 | if (num_added == 0) { |
cb77fcd8 | 1144 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1145 | async_cow->inode = NULL; |
8180ef88 | 1146 | } |
771ed689 CM |
1147 | } |
1148 | ||
1149 | /* | |
1150 | * work queue call back to submit previously compressed pages | |
1151 | */ | |
1152 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1153 | { | |
0b246afa | 1154 | struct btrfs_fs_info *fs_info; |
771ed689 | 1155 | struct async_cow *async_cow; |
771ed689 CM |
1156 | unsigned long nr_pages; |
1157 | ||
1158 | async_cow = container_of(work, struct async_cow, work); | |
1159 | ||
600b6cf4 | 1160 | fs_info = async_cow->fs_info; |
09cbfeaf KS |
1161 | nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >> |
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 | |
d397712b | 1169 | if (async_cow->inode) |
771ed689 | 1170 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1171 | } |
c8b97818 | 1172 | |
771ed689 CM |
1173 | static noinline void async_cow_free(struct btrfs_work *work) |
1174 | { | |
1175 | struct async_cow *async_cow; | |
1176 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1177 | if (async_cow->inode) |
cb77fcd8 | 1178 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1179 | kfree(async_cow); |
1180 | } | |
1181 | ||
1182 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1183 | u64 start, u64 end, int *page_started, | |
f82b7359 LB |
1184 | unsigned long *nr_written, |
1185 | unsigned int write_flags) | |
771ed689 | 1186 | { |
0b246afa | 1187 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 | 1188 | struct async_cow *async_cow; |
771ed689 CM |
1189 | unsigned long nr_pages; |
1190 | u64 cur_end; | |
771ed689 | 1191 | |
a3429ab7 | 1192 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
ae0f1625 | 1193 | 1, 0, NULL); |
d397712b | 1194 | while (start < end) { |
771ed689 | 1195 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1196 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1197 | async_cow->inode = igrab(inode); |
600b6cf4 | 1198 | async_cow->fs_info = fs_info; |
771ed689 CM |
1199 | async_cow->locked_page = locked_page; |
1200 | async_cow->start = start; | |
f82b7359 | 1201 | async_cow->write_flags = write_flags; |
771ed689 | 1202 | |
f79707b0 | 1203 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
0b246afa | 1204 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
771ed689 CM |
1205 | cur_end = end; |
1206 | else | |
ee22184b | 1207 | cur_end = min(end, start + SZ_512K - 1); |
771ed689 CM |
1208 | |
1209 | async_cow->end = cur_end; | |
1210 | INIT_LIST_HEAD(&async_cow->extents); | |
1211 | ||
9e0af237 LB |
1212 | btrfs_init_work(&async_cow->work, |
1213 | btrfs_delalloc_helper, | |
1214 | async_cow_start, async_cow_submit, | |
1215 | async_cow_free); | |
771ed689 | 1216 | |
09cbfeaf KS |
1217 | nr_pages = (cur_end - start + PAGE_SIZE) >> |
1218 | PAGE_SHIFT; | |
0b246afa | 1219 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1220 | |
0b246afa | 1221 | btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work); |
771ed689 | 1222 | |
771ed689 CM |
1223 | *nr_written += nr_pages; |
1224 | start = cur_end + 1; | |
1225 | } | |
1226 | *page_started = 1; | |
1227 | return 0; | |
be20aa9d CM |
1228 | } |
1229 | ||
2ff7e61e | 1230 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1231 | u64 bytenr, u64 num_bytes) |
1232 | { | |
1233 | int ret; | |
1234 | struct btrfs_ordered_sum *sums; | |
1235 | LIST_HEAD(list); | |
1236 | ||
0b246afa | 1237 | ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr, |
a2de733c | 1238 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1239 | if (ret == 0 && list_empty(&list)) |
1240 | return 0; | |
1241 | ||
1242 | while (!list_empty(&list)) { | |
1243 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1244 | list_del(&sums->list); | |
1245 | kfree(sums); | |
1246 | } | |
58113753 LB |
1247 | if (ret < 0) |
1248 | return ret; | |
17d217fe YZ |
1249 | return 1; |
1250 | } | |
1251 | ||
d352ac68 CM |
1252 | /* |
1253 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1254 | * of the extents that exist in the file, and COWs the file as required. | |
1255 | * | |
1256 | * If no cow copies or snapshots exist, we write directly to the existing | |
1257 | * blocks on disk | |
1258 | */ | |
7f366cfe CM |
1259 | static noinline int run_delalloc_nocow(struct inode *inode, |
1260 | struct page *locked_page, | |
771ed689 CM |
1261 | u64 start, u64 end, int *page_started, int force, |
1262 | unsigned long *nr_written) | |
be20aa9d | 1263 | { |
0b246afa | 1264 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
be20aa9d CM |
1265 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1266 | struct extent_buffer *leaf; | |
be20aa9d | 1267 | struct btrfs_path *path; |
80ff3856 | 1268 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1269 | struct btrfs_key found_key; |
6f9994db | 1270 | struct extent_map *em; |
80ff3856 YZ |
1271 | u64 cow_start; |
1272 | u64 cur_offset; | |
1273 | u64 extent_end; | |
5d4f98a2 | 1274 | u64 extent_offset; |
80ff3856 YZ |
1275 | u64 disk_bytenr; |
1276 | u64 num_bytes; | |
b4939680 | 1277 | u64 disk_num_bytes; |
cc95bef6 | 1278 | u64 ram_bytes; |
80ff3856 | 1279 | int extent_type; |
8ecebf4d | 1280 | int ret; |
d899e052 | 1281 | int type; |
80ff3856 YZ |
1282 | int nocow; |
1283 | int check_prev = 1; | |
82d5902d | 1284 | bool nolock; |
4a0cc7ca | 1285 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
be20aa9d CM |
1286 | |
1287 | path = btrfs_alloc_path(); | |
17ca04af | 1288 | if (!path) { |
ba8b04c1 QW |
1289 | extent_clear_unlock_delalloc(inode, start, end, end, |
1290 | locked_page, | |
c2790a2e | 1291 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1292 | EXTENT_DO_ACCOUNTING | |
1293 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1294 | PAGE_CLEAR_DIRTY | |
1295 | PAGE_SET_WRITEBACK | | |
1296 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1297 | return -ENOMEM; |
17ca04af | 1298 | } |
82d5902d | 1299 | |
70ddc553 | 1300 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
82d5902d | 1301 | |
80ff3856 YZ |
1302 | cow_start = (u64)-1; |
1303 | cur_offset = start; | |
1304 | while (1) { | |
e4c3b2dc | 1305 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1306 | cur_offset, 0); |
d788a349 | 1307 | if (ret < 0) |
79787eaa | 1308 | goto error; |
80ff3856 YZ |
1309 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1310 | leaf = path->nodes[0]; | |
1311 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1312 | path->slots[0] - 1); | |
33345d01 | 1313 | if (found_key.objectid == ino && |
80ff3856 YZ |
1314 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1315 | path->slots[0]--; | |
1316 | } | |
1317 | check_prev = 0; | |
1318 | next_slot: | |
1319 | leaf = path->nodes[0]; | |
1320 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1321 | ret = btrfs_next_leaf(root, path); | |
e8916699 LB |
1322 | if (ret < 0) { |
1323 | if (cow_start != (u64)-1) | |
1324 | cur_offset = cow_start; | |
79787eaa | 1325 | goto error; |
e8916699 | 1326 | } |
80ff3856 YZ |
1327 | if (ret > 0) |
1328 | break; | |
1329 | leaf = path->nodes[0]; | |
1330 | } | |
be20aa9d | 1331 | |
80ff3856 YZ |
1332 | nocow = 0; |
1333 | disk_bytenr = 0; | |
17d217fe | 1334 | num_bytes = 0; |
80ff3856 YZ |
1335 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1336 | ||
1d512cb7 FM |
1337 | if (found_key.objectid > ino) |
1338 | break; | |
1339 | if (WARN_ON_ONCE(found_key.objectid < ino) || | |
1340 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1341 | path->slots[0]++; | |
1342 | goto next_slot; | |
1343 | } | |
1344 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || | |
80ff3856 YZ |
1345 | found_key.offset > end) |
1346 | break; | |
1347 | ||
1348 | if (found_key.offset > cur_offset) { | |
1349 | extent_end = found_key.offset; | |
e9061e21 | 1350 | extent_type = 0; |
80ff3856 YZ |
1351 | goto out_check; |
1352 | } | |
1353 | ||
1354 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1355 | struct btrfs_file_extent_item); | |
1356 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1357 | ||
cc95bef6 | 1358 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1359 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1360 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1361 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1362 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1363 | extent_end = found_key.offset + |
1364 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1365 | disk_num_bytes = |
1366 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1367 | if (extent_end <= start) { |
1368 | path->slots[0]++; | |
1369 | goto next_slot; | |
1370 | } | |
17d217fe YZ |
1371 | if (disk_bytenr == 0) |
1372 | goto out_check; | |
80ff3856 YZ |
1373 | if (btrfs_file_extent_compression(leaf, fi) || |
1374 | btrfs_file_extent_encryption(leaf, fi) || | |
1375 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1376 | goto out_check; | |
78d4295b EL |
1377 | /* |
1378 | * Do the same check as in btrfs_cross_ref_exist but | |
1379 | * without the unnecessary search. | |
1380 | */ | |
27a7ff55 LF |
1381 | if (!nolock && |
1382 | btrfs_file_extent_generation(leaf, fi) <= | |
78d4295b EL |
1383 | btrfs_root_last_snapshot(&root->root_item)) |
1384 | goto out_check; | |
d899e052 YZ |
1385 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1386 | goto out_check; | |
2ff7e61e | 1387 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
80ff3856 | 1388 | goto out_check; |
58113753 LB |
1389 | ret = btrfs_cross_ref_exist(root, ino, |
1390 | found_key.offset - | |
1391 | extent_offset, disk_bytenr); | |
1392 | if (ret) { | |
1393 | /* | |
1394 | * ret could be -EIO if the above fails to read | |
1395 | * metadata. | |
1396 | */ | |
1397 | if (ret < 0) { | |
1398 | if (cow_start != (u64)-1) | |
1399 | cur_offset = cow_start; | |
1400 | goto error; | |
1401 | } | |
1402 | ||
1403 | WARN_ON_ONCE(nolock); | |
17d217fe | 1404 | goto out_check; |
58113753 | 1405 | } |
5d4f98a2 | 1406 | disk_bytenr += extent_offset; |
17d217fe YZ |
1407 | disk_bytenr += cur_offset - found_key.offset; |
1408 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1409 | /* |
1410 | * if there are pending snapshots for this root, | |
1411 | * we fall into common COW way. | |
1412 | */ | |
8ecebf4d RK |
1413 | if (!nolock && atomic_read(&root->snapshot_force_cow)) |
1414 | goto out_check; | |
17d217fe YZ |
1415 | /* |
1416 | * force cow if csum exists in the range. | |
1417 | * this ensure that csum for a given extent are | |
1418 | * either valid or do not exist. | |
1419 | */ | |
58113753 LB |
1420 | ret = csum_exist_in_range(fs_info, disk_bytenr, |
1421 | num_bytes); | |
1422 | if (ret) { | |
58113753 LB |
1423 | /* |
1424 | * ret could be -EIO if the above fails to read | |
1425 | * metadata. | |
1426 | */ | |
1427 | if (ret < 0) { | |
1428 | if (cow_start != (u64)-1) | |
1429 | cur_offset = cow_start; | |
1430 | goto error; | |
1431 | } | |
1432 | WARN_ON_ONCE(nolock); | |
17d217fe | 1433 | goto out_check; |
91e1f56a | 1434 | } |
8ecebf4d | 1435 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) |
f78c436c | 1436 | goto out_check; |
80ff3856 YZ |
1437 | nocow = 1; |
1438 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1439 | extent_end = found_key.offset + | |
e41ca589 | 1440 | btrfs_file_extent_ram_bytes(leaf, fi); |
da17066c | 1441 | extent_end = ALIGN(extent_end, |
0b246afa | 1442 | fs_info->sectorsize); |
80ff3856 YZ |
1443 | } else { |
1444 | BUG_ON(1); | |
1445 | } | |
1446 | out_check: | |
1447 | if (extent_end <= start) { | |
1448 | path->slots[0]++; | |
f78c436c | 1449 | if (nocow) |
0b246afa | 1450 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
80ff3856 YZ |
1451 | goto next_slot; |
1452 | } | |
1453 | if (!nocow) { | |
1454 | if (cow_start == (u64)-1) | |
1455 | cow_start = cur_offset; | |
1456 | cur_offset = extent_end; | |
1457 | if (cur_offset > end) | |
1458 | break; | |
1459 | path->slots[0]++; | |
1460 | goto next_slot; | |
7ea394f1 YZ |
1461 | } |
1462 | ||
b3b4aa74 | 1463 | btrfs_release_path(path); |
80ff3856 | 1464 | if (cow_start != (u64)-1) { |
00361589 JB |
1465 | ret = cow_file_range(inode, locked_page, |
1466 | cow_start, found_key.offset - 1, | |
dda3245e WX |
1467 | end, page_started, nr_written, 1, |
1468 | NULL); | |
e9894fd3 | 1469 | if (ret) { |
f78c436c | 1470 | if (nocow) |
0b246afa | 1471 | btrfs_dec_nocow_writers(fs_info, |
f78c436c | 1472 | disk_bytenr); |
79787eaa | 1473 | goto error; |
e9894fd3 | 1474 | } |
80ff3856 | 1475 | cow_start = (u64)-1; |
7ea394f1 | 1476 | } |
80ff3856 | 1477 | |
d899e052 | 1478 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
6f9994db LB |
1479 | u64 orig_start = found_key.offset - extent_offset; |
1480 | ||
1481 | em = create_io_em(inode, cur_offset, num_bytes, | |
1482 | orig_start, | |
1483 | disk_bytenr, /* block_start */ | |
1484 | num_bytes, /* block_len */ | |
1485 | disk_num_bytes, /* orig_block_len */ | |
1486 | ram_bytes, BTRFS_COMPRESS_NONE, | |
1487 | BTRFS_ORDERED_PREALLOC); | |
1488 | if (IS_ERR(em)) { | |
6f9994db LB |
1489 | if (nocow) |
1490 | btrfs_dec_nocow_writers(fs_info, | |
1491 | disk_bytenr); | |
1492 | ret = PTR_ERR(em); | |
1493 | goto error; | |
d899e052 | 1494 | } |
6f9994db LB |
1495 | free_extent_map(em); |
1496 | } | |
1497 | ||
1498 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
d899e052 YZ |
1499 | type = BTRFS_ORDERED_PREALLOC; |
1500 | } else { | |
1501 | type = BTRFS_ORDERED_NOCOW; | |
1502 | } | |
80ff3856 YZ |
1503 | |
1504 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1505 | num_bytes, num_bytes, type); |
f78c436c | 1506 | if (nocow) |
0b246afa | 1507 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
79787eaa | 1508 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1509 | |
efa56464 | 1510 | if (root->root_key.objectid == |
4dbd80fb QW |
1511 | BTRFS_DATA_RELOC_TREE_OBJECTID) |
1512 | /* | |
1513 | * Error handled later, as we must prevent | |
1514 | * extent_clear_unlock_delalloc() in error handler | |
1515 | * from freeing metadata of created ordered extent. | |
1516 | */ | |
efa56464 YZ |
1517 | ret = btrfs_reloc_clone_csums(inode, cur_offset, |
1518 | num_bytes); | |
efa56464 | 1519 | |
c2790a2e | 1520 | extent_clear_unlock_delalloc(inode, cur_offset, |
ba8b04c1 | 1521 | cur_offset + num_bytes - 1, end, |
c2790a2e | 1522 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1523 | EXTENT_DELALLOC | |
1524 | EXTENT_CLEAR_DATA_RESV, | |
1525 | PAGE_UNLOCK | PAGE_SET_PRIVATE2); | |
1526 | ||
80ff3856 | 1527 | cur_offset = extent_end; |
4dbd80fb QW |
1528 | |
1529 | /* | |
1530 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
1531 | * handler, as metadata for created ordered extent will only | |
1532 | * be freed by btrfs_finish_ordered_io(). | |
1533 | */ | |
1534 | if (ret) | |
1535 | goto error; | |
80ff3856 YZ |
1536 | if (cur_offset > end) |
1537 | break; | |
be20aa9d | 1538 | } |
b3b4aa74 | 1539 | btrfs_release_path(path); |
80ff3856 | 1540 | |
506481b2 | 1541 | if (cur_offset <= end && cow_start == (u64)-1) |
80ff3856 | 1542 | cow_start = cur_offset; |
17ca04af | 1543 | |
80ff3856 | 1544 | if (cow_start != (u64)-1) { |
506481b2 | 1545 | cur_offset = end; |
dda3245e WX |
1546 | ret = cow_file_range(inode, locked_page, cow_start, end, end, |
1547 | page_started, nr_written, 1, NULL); | |
d788a349 | 1548 | if (ret) |
79787eaa | 1549 | goto error; |
80ff3856 YZ |
1550 | } |
1551 | ||
79787eaa | 1552 | error: |
17ca04af | 1553 | if (ret && cur_offset < end) |
ba8b04c1 | 1554 | extent_clear_unlock_delalloc(inode, cur_offset, end, end, |
c2790a2e | 1555 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1556 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1557 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1558 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1559 | PAGE_SET_WRITEBACK | |
1560 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1561 | btrfs_free_path(path); |
79787eaa | 1562 | return ret; |
be20aa9d CM |
1563 | } |
1564 | ||
47059d93 WS |
1565 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1566 | { | |
1567 | ||
1568 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1569 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1570 | return 0; | |
1571 | ||
1572 | /* | |
1573 | * @defrag_bytes is a hint value, no spinlock held here, | |
1574 | * if is not zero, it means the file is defragging. | |
1575 | * Force cow if given extent needs to be defragged. | |
1576 | */ | |
1577 | if (BTRFS_I(inode)->defrag_bytes && | |
1578 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1579 | EXTENT_DEFRAG, 0, NULL)) | |
1580 | return 1; | |
1581 | ||
1582 | return 0; | |
1583 | } | |
1584 | ||
d352ac68 | 1585 | /* |
5eaad97a NB |
1586 | * Function to process delayed allocation (create CoW) for ranges which are |
1587 | * being touched for the first time. | |
d352ac68 | 1588 | */ |
5eaad97a NB |
1589 | int btrfs_run_delalloc_range(void *private_data, struct page *locked_page, |
1590 | u64 start, u64 end, int *page_started, unsigned long *nr_written, | |
1591 | struct writeback_control *wbc) | |
be20aa9d | 1592 | { |
c6100a4b | 1593 | struct inode *inode = private_data; |
be20aa9d | 1594 | int ret; |
47059d93 | 1595 | int force_cow = need_force_cow(inode, start, end); |
f82b7359 | 1596 | unsigned int write_flags = wbc_to_write_flags(wbc); |
a2135011 | 1597 | |
47059d93 | 1598 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1599 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1600 | page_started, 1, nr_written); |
47059d93 | 1601 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1602 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1603 | page_started, 0, nr_written); |
c2fcdcdf | 1604 | } else if (!inode_need_compress(inode, start, end)) { |
dda3245e WX |
1605 | ret = cow_file_range(inode, locked_page, start, end, end, |
1606 | page_started, nr_written, 1, NULL); | |
7ddf5a42 JB |
1607 | } else { |
1608 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1609 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1610 | ret = cow_file_range_async(inode, locked_page, start, end, |
f82b7359 LB |
1611 | page_started, nr_written, |
1612 | write_flags); | |
7ddf5a42 | 1613 | } |
52427260 | 1614 | if (ret) |
d1051d6e NB |
1615 | btrfs_cleanup_ordered_extents(inode, locked_page, start, |
1616 | end - start + 1); | |
b888db2b CM |
1617 | return ret; |
1618 | } | |
1619 | ||
abbb55f4 NB |
1620 | void btrfs_split_delalloc_extent(struct inode *inode, |
1621 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1622 | { |
dcab6a3b JB |
1623 | u64 size; |
1624 | ||
0ca1f7ce | 1625 | /* not delalloc, ignore it */ |
9ed74f2d | 1626 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1627 | return; |
9ed74f2d | 1628 | |
dcab6a3b JB |
1629 | size = orig->end - orig->start + 1; |
1630 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 1631 | u32 num_extents; |
dcab6a3b JB |
1632 | u64 new_size; |
1633 | ||
1634 | /* | |
5c848198 | 1635 | * See the explanation in btrfs_merge_delalloc_extent, the same |
ba117213 | 1636 | * applies here, just in reverse. |
dcab6a3b JB |
1637 | */ |
1638 | new_size = orig->end - split + 1; | |
823bb20a | 1639 | num_extents = count_max_extents(new_size); |
ba117213 | 1640 | new_size = split - orig->start; |
823bb20a DS |
1641 | num_extents += count_max_extents(new_size); |
1642 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
1643 | return; |
1644 | } | |
1645 | ||
9e0baf60 | 1646 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1647 | btrfs_mod_outstanding_extents(BTRFS_I(inode), 1); |
9e0baf60 | 1648 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1649 | } |
1650 | ||
1651 | /* | |
5c848198 NB |
1652 | * Handle merged delayed allocation extents so we can keep track of new extents |
1653 | * that are just merged onto old extents, such as when we are doing sequential | |
1654 | * writes, so we can properly account for the metadata space we'll need. | |
9ed74f2d | 1655 | */ |
5c848198 NB |
1656 | void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new, |
1657 | struct extent_state *other) | |
9ed74f2d | 1658 | { |
dcab6a3b | 1659 | u64 new_size, old_size; |
823bb20a | 1660 | u32 num_extents; |
dcab6a3b | 1661 | |
9ed74f2d JB |
1662 | /* not delalloc, ignore it */ |
1663 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1664 | return; |
9ed74f2d | 1665 | |
8461a3de JB |
1666 | if (new->start > other->start) |
1667 | new_size = new->end - other->start + 1; | |
1668 | else | |
1669 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1670 | |
1671 | /* we're not bigger than the max, unreserve the space and go */ | |
1672 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1673 | spin_lock(&BTRFS_I(inode)->lock); | |
8b62f87b | 1674 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
dcab6a3b JB |
1675 | spin_unlock(&BTRFS_I(inode)->lock); |
1676 | return; | |
1677 | } | |
1678 | ||
1679 | /* | |
ba117213 JB |
1680 | * We have to add up either side to figure out how many extents were |
1681 | * accounted for before we merged into one big extent. If the number of | |
1682 | * extents we accounted for is <= the amount we need for the new range | |
1683 | * then we can return, otherwise drop. Think of it like this | |
1684 | * | |
1685 | * [ 4k][MAX_SIZE] | |
1686 | * | |
1687 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1688 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1689 | * we have 1 so they are == and we can return. But in this case | |
1690 | * | |
1691 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1692 | * | |
1693 | * Each range on their own accounts for 2 extents, but merged together | |
1694 | * they are only 3 extents worth of accounting, so we need to drop in | |
1695 | * this case. | |
dcab6a3b | 1696 | */ |
ba117213 | 1697 | old_size = other->end - other->start + 1; |
823bb20a | 1698 | num_extents = count_max_extents(old_size); |
ba117213 | 1699 | old_size = new->end - new->start + 1; |
823bb20a DS |
1700 | num_extents += count_max_extents(old_size); |
1701 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
1702 | return; |
1703 | ||
9e0baf60 | 1704 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1705 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
9e0baf60 | 1706 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1707 | } |
1708 | ||
eb73c1b7 MX |
1709 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1710 | struct inode *inode) | |
1711 | { | |
0b246afa JM |
1712 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1713 | ||
eb73c1b7 MX |
1714 | spin_lock(&root->delalloc_lock); |
1715 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1716 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1717 | &root->delalloc_inodes); | |
1718 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1719 | &BTRFS_I(inode)->runtime_flags); | |
1720 | root->nr_delalloc_inodes++; | |
1721 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 1722 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1723 | BUG_ON(!list_empty(&root->delalloc_root)); |
1724 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
1725 | &fs_info->delalloc_roots); |
1726 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
1727 | } |
1728 | } | |
1729 | spin_unlock(&root->delalloc_lock); | |
1730 | } | |
1731 | ||
2b877331 NB |
1732 | |
1733 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1734 | struct btrfs_inode *inode) | |
eb73c1b7 | 1735 | { |
3ffbd68c | 1736 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b246afa | 1737 | |
9e3e97f4 NB |
1738 | if (!list_empty(&inode->delalloc_inodes)) { |
1739 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 1740 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1741 | &inode->runtime_flags); |
eb73c1b7 MX |
1742 | root->nr_delalloc_inodes--; |
1743 | if (!root->nr_delalloc_inodes) { | |
7c8a0d36 | 1744 | ASSERT(list_empty(&root->delalloc_inodes)); |
0b246afa | 1745 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1746 | BUG_ON(list_empty(&root->delalloc_root)); |
1747 | list_del_init(&root->delalloc_root); | |
0b246afa | 1748 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1749 | } |
1750 | } | |
2b877331 NB |
1751 | } |
1752 | ||
1753 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1754 | struct btrfs_inode *inode) | |
1755 | { | |
1756 | spin_lock(&root->delalloc_lock); | |
1757 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
1758 | spin_unlock(&root->delalloc_lock); |
1759 | } | |
1760 | ||
d352ac68 | 1761 | /* |
e06a1fc9 NB |
1762 | * Properly track delayed allocation bytes in the inode and to maintain the |
1763 | * list of inodes that have pending delalloc work to be done. | |
d352ac68 | 1764 | */ |
e06a1fc9 NB |
1765 | void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state, |
1766 | unsigned *bits) | |
291d673e | 1767 | { |
0b246afa JM |
1768 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1769 | ||
47059d93 WS |
1770 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1771 | WARN_ON(1); | |
75eff68e CM |
1772 | /* |
1773 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1774 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1775 | * bit, which is only set or cleared with irqs on |
1776 | */ | |
0ca1f7ce | 1777 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1778 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1779 | u64 len = state->end + 1 - state->start; |
8b62f87b | 1780 | u32 num_extents = count_max_extents(len); |
70ddc553 | 1781 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 1782 | |
8b62f87b JB |
1783 | spin_lock(&BTRFS_I(inode)->lock); |
1784 | btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents); | |
1785 | spin_unlock(&BTRFS_I(inode)->lock); | |
287a0ab9 | 1786 | |
6a3891c5 | 1787 | /* For sanity tests */ |
0b246afa | 1788 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1789 | return; |
1790 | ||
104b4e51 NB |
1791 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
1792 | fs_info->delalloc_batch); | |
df0af1a5 | 1793 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1794 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1795 | if (*bits & EXTENT_DEFRAG) |
1796 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1797 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1798 | &BTRFS_I(inode)->runtime_flags)) |
1799 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1800 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1801 | } |
a7e3b975 FM |
1802 | |
1803 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
1804 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1805 | spin_lock(&BTRFS_I(inode)->lock); | |
1806 | BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 - | |
1807 | state->start; | |
1808 | spin_unlock(&BTRFS_I(inode)->lock); | |
1809 | } | |
291d673e CM |
1810 | } |
1811 | ||
d352ac68 | 1812 | /* |
a36bb5f9 NB |
1813 | * Once a range is no longer delalloc this function ensures that proper |
1814 | * accounting happens. | |
d352ac68 | 1815 | */ |
a36bb5f9 NB |
1816 | void btrfs_clear_delalloc_extent(struct inode *vfs_inode, |
1817 | struct extent_state *state, unsigned *bits) | |
291d673e | 1818 | { |
a36bb5f9 NB |
1819 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
1820 | struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb); | |
47059d93 | 1821 | u64 len = state->end + 1 - state->start; |
823bb20a | 1822 | u32 num_extents = count_max_extents(len); |
47059d93 | 1823 | |
4a4b964f FM |
1824 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) { |
1825 | spin_lock(&inode->lock); | |
6fc0ef68 | 1826 | inode->defrag_bytes -= len; |
4a4b964f FM |
1827 | spin_unlock(&inode->lock); |
1828 | } | |
47059d93 | 1829 | |
75eff68e CM |
1830 | /* |
1831 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1832 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1833 | * bit, which is only set or cleared with irqs on |
1834 | */ | |
0ca1f7ce | 1835 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 1836 | struct btrfs_root *root = inode->root; |
83eea1f1 | 1837 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1838 | |
8b62f87b JB |
1839 | spin_lock(&inode->lock); |
1840 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
1841 | spin_unlock(&inode->lock); | |
0ca1f7ce | 1842 | |
b6d08f06 JB |
1843 | /* |
1844 | * We don't reserve metadata space for space cache inodes so we | |
1845 | * don't need to call dellalloc_release_metadata if there is an | |
1846 | * error. | |
1847 | */ | |
a315e68f | 1848 | if (*bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 1849 | root != fs_info->tree_root) |
43b18595 | 1850 | btrfs_delalloc_release_metadata(inode, len, false); |
0ca1f7ce | 1851 | |
6a3891c5 | 1852 | /* For sanity tests. */ |
0b246afa | 1853 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1854 | return; |
1855 | ||
a315e68f FM |
1856 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID && |
1857 | do_list && !(state->state & EXTENT_NORESERVE) && | |
1858 | (*bits & EXTENT_CLEAR_DATA_RESV)) | |
6fc0ef68 NB |
1859 | btrfs_free_reserved_data_space_noquota( |
1860 | &inode->vfs_inode, | |
51773bec | 1861 | state->start, len); |
9ed74f2d | 1862 | |
104b4e51 NB |
1863 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
1864 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
1865 | spin_lock(&inode->lock); |
1866 | inode->delalloc_bytes -= len; | |
1867 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 1868 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1869 | &inode->runtime_flags)) |
eb73c1b7 | 1870 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 1871 | spin_unlock(&inode->lock); |
291d673e | 1872 | } |
a7e3b975 FM |
1873 | |
1874 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
1875 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1876 | spin_lock(&inode->lock); | |
1877 | ASSERT(inode->new_delalloc_bytes >= len); | |
1878 | inode->new_delalloc_bytes -= len; | |
1879 | spin_unlock(&inode->lock); | |
1880 | } | |
291d673e CM |
1881 | } |
1882 | ||
d352ac68 | 1883 | /* |
da12fe54 NB |
1884 | * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit |
1885 | * in a chunk's stripe. This function ensures that bios do not span a | |
1886 | * stripe/chunk | |
6f034ece | 1887 | * |
da12fe54 NB |
1888 | * @page - The page we are about to add to the bio |
1889 | * @size - size we want to add to the bio | |
1890 | * @bio - bio we want to ensure is smaller than a stripe | |
1891 | * @bio_flags - flags of the bio | |
1892 | * | |
1893 | * return 1 if page cannot be added to the bio | |
1894 | * return 0 if page can be added to the bio | |
6f034ece | 1895 | * return error otherwise |
d352ac68 | 1896 | */ |
da12fe54 NB |
1897 | int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio, |
1898 | unsigned long bio_flags) | |
239b14b3 | 1899 | { |
0b246afa JM |
1900 | struct inode *inode = page->mapping->host; |
1901 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
4f024f37 | 1902 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1903 | u64 length = 0; |
1904 | u64 map_length; | |
239b14b3 CM |
1905 | int ret; |
1906 | ||
771ed689 CM |
1907 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1908 | return 0; | |
1909 | ||
4f024f37 | 1910 | length = bio->bi_iter.bi_size; |
239b14b3 | 1911 | map_length = length; |
0b246afa JM |
1912 | ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length, |
1913 | NULL, 0); | |
6f034ece LB |
1914 | if (ret < 0) |
1915 | return ret; | |
d397712b | 1916 | if (map_length < length + size) |
239b14b3 | 1917 | return 1; |
3444a972 | 1918 | return 0; |
239b14b3 CM |
1919 | } |
1920 | ||
d352ac68 CM |
1921 | /* |
1922 | * in order to insert checksums into the metadata in large chunks, | |
1923 | * we wait until bio submission time. All the pages in the bio are | |
1924 | * checksummed and sums are attached onto the ordered extent record. | |
1925 | * | |
1926 | * At IO completion time the cums attached on the ordered extent record | |
1927 | * are inserted into the btree | |
1928 | */ | |
d0ee3934 | 1929 | static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio, |
eaf25d93 | 1930 | u64 bio_offset) |
065631f6 | 1931 | { |
c6100a4b | 1932 | struct inode *inode = private_data; |
4e4cbee9 | 1933 | blk_status_t ret = 0; |
e015640f | 1934 | |
2ff7e61e | 1935 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
79787eaa | 1936 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1937 | return 0; |
1938 | } | |
e015640f | 1939 | |
d352ac68 | 1940 | /* |
cad321ad | 1941 | * extent_io.c submission hook. This does the right thing for csum calculation |
4c274bc6 LB |
1942 | * on write, or reading the csums from the tree before a read. |
1943 | * | |
1944 | * Rules about async/sync submit, | |
1945 | * a) read: sync submit | |
1946 | * | |
1947 | * b) write without checksum: sync submit | |
1948 | * | |
1949 | * c) write with checksum: | |
1950 | * c-1) if bio is issued by fsync: sync submit | |
1951 | * (sync_writers != 0) | |
1952 | * | |
1953 | * c-2) if root is reloc root: sync submit | |
1954 | * (only in case of buffered IO) | |
1955 | * | |
1956 | * c-3) otherwise: async submit | |
d352ac68 | 1957 | */ |
8c27cb35 | 1958 | static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio, |
c6100a4b JB |
1959 | int mirror_num, unsigned long bio_flags, |
1960 | u64 bio_offset) | |
44b8bd7e | 1961 | { |
c6100a4b | 1962 | struct inode *inode = private_data; |
0b246afa | 1963 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 1964 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 1965 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
4e4cbee9 | 1966 | blk_status_t ret = 0; |
19b9bdb0 | 1967 | int skip_sum; |
b812ce28 | 1968 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1969 | |
6cbff00f | 1970 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1971 | |
70ddc553 | 1972 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 1973 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 1974 | |
37226b21 | 1975 | if (bio_op(bio) != REQ_OP_WRITE) { |
0b246afa | 1976 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 1977 | if (ret) |
61891923 | 1978 | goto out; |
5fd02043 | 1979 | |
d20f7043 | 1980 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1981 | ret = btrfs_submit_compressed_read(inode, bio, |
1982 | mirror_num, | |
1983 | bio_flags); | |
1984 | goto out; | |
c2db1073 | 1985 | } else if (!skip_sum) { |
2ff7e61e | 1986 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 1987 | if (ret) |
61891923 | 1988 | goto out; |
c2db1073 | 1989 | } |
4d1b5fb4 | 1990 | goto mapit; |
b812ce28 | 1991 | } else if (async && !skip_sum) { |
17d217fe YZ |
1992 | /* csum items have already been cloned */ |
1993 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1994 | goto mapit; | |
19b9bdb0 | 1995 | /* we're doing a write, do the async checksumming */ |
c6100a4b JB |
1996 | ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags, |
1997 | bio_offset, inode, | |
e288c080 | 1998 | btrfs_submit_bio_start); |
61891923 | 1999 | goto out; |
b812ce28 | 2000 | } else if (!skip_sum) { |
2ff7e61e | 2001 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
b812ce28 JB |
2002 | if (ret) |
2003 | goto out; | |
19b9bdb0 CM |
2004 | } |
2005 | ||
0b86a832 | 2006 | mapit: |
2ff7e61e | 2007 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
61891923 SB |
2008 | |
2009 | out: | |
4e4cbee9 CH |
2010 | if (ret) { |
2011 | bio->bi_status = ret; | |
4246a0b6 CH |
2012 | bio_endio(bio); |
2013 | } | |
61891923 | 2014 | return ret; |
065631f6 | 2015 | } |
6885f308 | 2016 | |
d352ac68 CM |
2017 | /* |
2018 | * given a list of ordered sums record them in the inode. This happens | |
2019 | * at IO completion time based on sums calculated at bio submission time. | |
2020 | */ | |
ba1da2f4 | 2021 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
df9f628e | 2022 | struct inode *inode, struct list_head *list) |
e6dcd2dc | 2023 | { |
e6dcd2dc | 2024 | struct btrfs_ordered_sum *sum; |
ac01f26a | 2025 | int ret; |
e6dcd2dc | 2026 | |
c6e30871 | 2027 | list_for_each_entry(sum, list, list) { |
7c2871a2 | 2028 | trans->adding_csums = true; |
ac01f26a | 2029 | ret = btrfs_csum_file_blocks(trans, |
d20f7043 | 2030 | BTRFS_I(inode)->root->fs_info->csum_root, sum); |
7c2871a2 | 2031 | trans->adding_csums = false; |
ac01f26a NB |
2032 | if (ret) |
2033 | return ret; | |
e6dcd2dc CM |
2034 | } |
2035 | return 0; | |
2036 | } | |
2037 | ||
2ac55d41 | 2038 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
e3b8a485 | 2039 | unsigned int extra_bits, |
ba8b04c1 | 2040 | struct extent_state **cached_state, int dedupe) |
ea8c2819 | 2041 | { |
fdb1e121 | 2042 | WARN_ON(PAGE_ALIGNED(end)); |
ea8c2819 | 2043 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
e3b8a485 | 2044 | extra_bits, cached_state); |
ea8c2819 CM |
2045 | } |
2046 | ||
d352ac68 | 2047 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2048 | struct btrfs_writepage_fixup { |
2049 | struct page *page; | |
2050 | struct btrfs_work work; | |
2051 | }; | |
2052 | ||
b2950863 | 2053 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2054 | { |
2055 | struct btrfs_writepage_fixup *fixup; | |
2056 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2057 | struct extent_state *cached_state = NULL; |
364ecf36 | 2058 | struct extent_changeset *data_reserved = NULL; |
247e743c CM |
2059 | struct page *page; |
2060 | struct inode *inode; | |
2061 | u64 page_start; | |
2062 | u64 page_end; | |
87826df0 | 2063 | int ret; |
247e743c CM |
2064 | |
2065 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2066 | page = fixup->page; | |
4a096752 | 2067 | again: |
247e743c CM |
2068 | lock_page(page); |
2069 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
2070 | ClearPageChecked(page); | |
2071 | goto out_page; | |
2072 | } | |
2073 | ||
2074 | inode = page->mapping->host; | |
2075 | page_start = page_offset(page); | |
09cbfeaf | 2076 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 2077 | |
ff13db41 | 2078 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 2079 | &cached_state); |
4a096752 CM |
2080 | |
2081 | /* already ordered? We're done */ | |
8b62b72b | 2082 | if (PagePrivate2(page)) |
247e743c | 2083 | goto out; |
4a096752 | 2084 | |
a776c6fa | 2085 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
09cbfeaf | 2086 | PAGE_SIZE); |
4a096752 | 2087 | if (ordered) { |
2ac55d41 | 2088 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
e43bbe5e | 2089 | page_end, &cached_state); |
4a096752 CM |
2090 | unlock_page(page); |
2091 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 2092 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2093 | goto again; |
2094 | } | |
247e743c | 2095 | |
364ecf36 | 2096 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
09cbfeaf | 2097 | PAGE_SIZE); |
87826df0 JM |
2098 | if (ret) { |
2099 | mapping_set_error(page->mapping, ret); | |
2100 | end_extent_writepage(page, ret, page_start, page_end); | |
2101 | ClearPageChecked(page); | |
2102 | goto out; | |
2103 | } | |
2104 | ||
f3038ee3 NB |
2105 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, |
2106 | &cached_state, 0); | |
2107 | if (ret) { | |
2108 | mapping_set_error(page->mapping, ret); | |
2109 | end_extent_writepage(page, ret, page_start, page_end); | |
2110 | ClearPageChecked(page); | |
2111 | goto out; | |
2112 | } | |
2113 | ||
247e743c | 2114 | ClearPageChecked(page); |
87826df0 | 2115 | set_page_dirty(page); |
43b18595 | 2116 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false); |
247e743c | 2117 | out: |
2ac55d41 | 2118 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
e43bbe5e | 2119 | &cached_state); |
247e743c CM |
2120 | out_page: |
2121 | unlock_page(page); | |
09cbfeaf | 2122 | put_page(page); |
b897abec | 2123 | kfree(fixup); |
364ecf36 | 2124 | extent_changeset_free(data_reserved); |
247e743c CM |
2125 | } |
2126 | ||
2127 | /* | |
2128 | * There are a few paths in the higher layers of the kernel that directly | |
2129 | * set the page dirty bit without asking the filesystem if it is a | |
2130 | * good idea. This causes problems because we want to make sure COW | |
2131 | * properly happens and the data=ordered rules are followed. | |
2132 | * | |
c8b97818 | 2133 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2134 | * hasn't been properly setup for IO. We kick off an async process |
2135 | * to fix it up. The async helper will wait for ordered extents, set | |
2136 | * the delalloc bit and make it safe to write the page. | |
2137 | */ | |
d75855b4 | 2138 | int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end) |
247e743c CM |
2139 | { |
2140 | struct inode *inode = page->mapping->host; | |
0b246afa | 2141 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2142 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2143 | |
8b62b72b CM |
2144 | /* this page is properly in the ordered list */ |
2145 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2146 | return 0; |
2147 | ||
2148 | if (PageChecked(page)) | |
2149 | return -EAGAIN; | |
2150 | ||
2151 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2152 | if (!fixup) | |
2153 | return -EAGAIN; | |
f421950f | 2154 | |
247e743c | 2155 | SetPageChecked(page); |
09cbfeaf | 2156 | get_page(page); |
9e0af237 LB |
2157 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2158 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2159 | fixup->page = page; |
0b246afa | 2160 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
87826df0 | 2161 | return -EBUSY; |
247e743c CM |
2162 | } |
2163 | ||
d899e052 YZ |
2164 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2165 | struct inode *inode, u64 file_pos, | |
2166 | u64 disk_bytenr, u64 disk_num_bytes, | |
2167 | u64 num_bytes, u64 ram_bytes, | |
2168 | u8 compression, u8 encryption, | |
2169 | u16 other_encoding, int extent_type) | |
2170 | { | |
2171 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2172 | struct btrfs_file_extent_item *fi; | |
2173 | struct btrfs_path *path; | |
2174 | struct extent_buffer *leaf; | |
2175 | struct btrfs_key ins; | |
a12b877b | 2176 | u64 qg_released; |
1acae57b | 2177 | int extent_inserted = 0; |
d899e052 YZ |
2178 | int ret; |
2179 | ||
2180 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2181 | if (!path) |
2182 | return -ENOMEM; | |
d899e052 | 2183 | |
a1ed835e CM |
2184 | /* |
2185 | * we may be replacing one extent in the tree with another. | |
2186 | * The new extent is pinned in the extent map, and we don't want | |
2187 | * to drop it from the cache until it is completely in the btree. | |
2188 | * | |
2189 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2190 | * the caller is expected to unpin it and allow it to be merged | |
2191 | * with the others. | |
2192 | */ | |
1acae57b FDBM |
2193 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2194 | file_pos + num_bytes, NULL, 0, | |
2195 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2196 | if (ret) |
2197 | goto out; | |
d899e052 | 2198 | |
1acae57b | 2199 | if (!extent_inserted) { |
4a0cc7ca | 2200 | ins.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b FDBM |
2201 | ins.offset = file_pos; |
2202 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2203 | ||
2204 | path->leave_spinning = 1; | |
2205 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2206 | sizeof(*fi)); | |
2207 | if (ret) | |
2208 | goto out; | |
2209 | } | |
d899e052 YZ |
2210 | leaf = path->nodes[0]; |
2211 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2212 | struct btrfs_file_extent_item); | |
2213 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2214 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2215 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2216 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2217 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2218 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2219 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2220 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2221 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2222 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2223 | |
d899e052 | 2224 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2225 | btrfs_release_path(path); |
d899e052 YZ |
2226 | |
2227 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2228 | |
2229 | ins.objectid = disk_bytenr; | |
2230 | ins.offset = disk_num_bytes; | |
2231 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2232 | |
297d750b | 2233 | /* |
5846a3c2 QW |
2234 | * Release the reserved range from inode dirty range map, as it is |
2235 | * already moved into delayed_ref_head | |
297d750b | 2236 | */ |
a12b877b QW |
2237 | ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes); |
2238 | if (ret < 0) | |
2239 | goto out; | |
2240 | qg_released = ret; | |
84f7d8e6 JB |
2241 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2242 | btrfs_ino(BTRFS_I(inode)), | |
2243 | file_pos, qg_released, &ins); | |
79787eaa | 2244 | out: |
d899e052 | 2245 | btrfs_free_path(path); |
b9473439 | 2246 | |
79787eaa | 2247 | return ret; |
d899e052 YZ |
2248 | } |
2249 | ||
38c227d8 LB |
2250 | /* snapshot-aware defrag */ |
2251 | struct sa_defrag_extent_backref { | |
2252 | struct rb_node node; | |
2253 | struct old_sa_defrag_extent *old; | |
2254 | u64 root_id; | |
2255 | u64 inum; | |
2256 | u64 file_pos; | |
2257 | u64 extent_offset; | |
2258 | u64 num_bytes; | |
2259 | u64 generation; | |
2260 | }; | |
2261 | ||
2262 | struct old_sa_defrag_extent { | |
2263 | struct list_head list; | |
2264 | struct new_sa_defrag_extent *new; | |
2265 | ||
2266 | u64 extent_offset; | |
2267 | u64 bytenr; | |
2268 | u64 offset; | |
2269 | u64 len; | |
2270 | int count; | |
2271 | }; | |
2272 | ||
2273 | struct new_sa_defrag_extent { | |
2274 | struct rb_root root; | |
2275 | struct list_head head; | |
2276 | struct btrfs_path *path; | |
2277 | struct inode *inode; | |
2278 | u64 file_pos; | |
2279 | u64 len; | |
2280 | u64 bytenr; | |
2281 | u64 disk_len; | |
2282 | u8 compress_type; | |
2283 | }; | |
2284 | ||
2285 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2286 | struct sa_defrag_extent_backref *b2) | |
2287 | { | |
2288 | if (b1->root_id < b2->root_id) | |
2289 | return -1; | |
2290 | else if (b1->root_id > b2->root_id) | |
2291 | return 1; | |
2292 | ||
2293 | if (b1->inum < b2->inum) | |
2294 | return -1; | |
2295 | else if (b1->inum > b2->inum) | |
2296 | return 1; | |
2297 | ||
2298 | if (b1->file_pos < b2->file_pos) | |
2299 | return -1; | |
2300 | else if (b1->file_pos > b2->file_pos) | |
2301 | return 1; | |
2302 | ||
2303 | /* | |
2304 | * [------------------------------] ===> (a range of space) | |
2305 | * |<--->| |<---->| =============> (fs/file tree A) | |
2306 | * |<---------------------------->| ===> (fs/file tree B) | |
2307 | * | |
2308 | * A range of space can refer to two file extents in one tree while | |
2309 | * refer to only one file extent in another tree. | |
2310 | * | |
2311 | * So we may process a disk offset more than one time(two extents in A) | |
2312 | * and locate at the same extent(one extent in B), then insert two same | |
2313 | * backrefs(both refer to the extent in B). | |
2314 | */ | |
2315 | return 0; | |
2316 | } | |
2317 | ||
2318 | static void backref_insert(struct rb_root *root, | |
2319 | struct sa_defrag_extent_backref *backref) | |
2320 | { | |
2321 | struct rb_node **p = &root->rb_node; | |
2322 | struct rb_node *parent = NULL; | |
2323 | struct sa_defrag_extent_backref *entry; | |
2324 | int ret; | |
2325 | ||
2326 | while (*p) { | |
2327 | parent = *p; | |
2328 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2329 | ||
2330 | ret = backref_comp(backref, entry); | |
2331 | if (ret < 0) | |
2332 | p = &(*p)->rb_left; | |
2333 | else | |
2334 | p = &(*p)->rb_right; | |
2335 | } | |
2336 | ||
2337 | rb_link_node(&backref->node, parent, p); | |
2338 | rb_insert_color(&backref->node, root); | |
2339 | } | |
2340 | ||
2341 | /* | |
2342 | * Note the backref might has changed, and in this case we just return 0. | |
2343 | */ | |
2344 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2345 | void *ctx) | |
2346 | { | |
2347 | struct btrfs_file_extent_item *extent; | |
38c227d8 LB |
2348 | struct old_sa_defrag_extent *old = ctx; |
2349 | struct new_sa_defrag_extent *new = old->new; | |
2350 | struct btrfs_path *path = new->path; | |
2351 | struct btrfs_key key; | |
2352 | struct btrfs_root *root; | |
2353 | struct sa_defrag_extent_backref *backref; | |
2354 | struct extent_buffer *leaf; | |
2355 | struct inode *inode = new->inode; | |
0b246afa | 2356 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2357 | int slot; |
2358 | int ret; | |
2359 | u64 extent_offset; | |
2360 | u64 num_bytes; | |
2361 | ||
2362 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
4a0cc7ca | 2363 | inum == btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2364 | return 0; |
2365 | ||
2366 | key.objectid = root_id; | |
2367 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2368 | key.offset = (u64)-1; | |
2369 | ||
38c227d8 LB |
2370 | root = btrfs_read_fs_root_no_name(fs_info, &key); |
2371 | if (IS_ERR(root)) { | |
2372 | if (PTR_ERR(root) == -ENOENT) | |
2373 | return 0; | |
2374 | WARN_ON(1); | |
ab8d0fc4 | 2375 | btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu", |
38c227d8 LB |
2376 | inum, offset, root_id); |
2377 | return PTR_ERR(root); | |
2378 | } | |
2379 | ||
2380 | key.objectid = inum; | |
2381 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2382 | if (offset > (u64)-1 << 32) | |
2383 | key.offset = 0; | |
2384 | else | |
2385 | key.offset = offset; | |
2386 | ||
2387 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2388 | if (WARN_ON(ret < 0)) |
38c227d8 | 2389 | return ret; |
50f1319c | 2390 | ret = 0; |
38c227d8 LB |
2391 | |
2392 | while (1) { | |
2393 | cond_resched(); | |
2394 | ||
2395 | leaf = path->nodes[0]; | |
2396 | slot = path->slots[0]; | |
2397 | ||
2398 | if (slot >= btrfs_header_nritems(leaf)) { | |
2399 | ret = btrfs_next_leaf(root, path); | |
2400 | if (ret < 0) { | |
2401 | goto out; | |
2402 | } else if (ret > 0) { | |
2403 | ret = 0; | |
2404 | goto out; | |
2405 | } | |
2406 | continue; | |
2407 | } | |
2408 | ||
2409 | path->slots[0]++; | |
2410 | ||
2411 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2412 | ||
2413 | if (key.objectid > inum) | |
2414 | goto out; | |
2415 | ||
2416 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2417 | continue; | |
2418 | ||
2419 | extent = btrfs_item_ptr(leaf, slot, | |
2420 | struct btrfs_file_extent_item); | |
2421 | ||
2422 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2423 | continue; | |
2424 | ||
e68afa49 LB |
2425 | /* |
2426 | * 'offset' refers to the exact key.offset, | |
2427 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2428 | * (key.offset - extent_offset). | |
2429 | */ | |
2430 | if (key.offset != offset) | |
38c227d8 LB |
2431 | continue; |
2432 | ||
e68afa49 | 2433 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2434 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2435 | |
38c227d8 LB |
2436 | if (extent_offset >= old->extent_offset + old->offset + |
2437 | old->len || extent_offset + num_bytes <= | |
2438 | old->extent_offset + old->offset) | |
2439 | continue; | |
38c227d8 LB |
2440 | break; |
2441 | } | |
2442 | ||
2443 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2444 | if (!backref) { | |
2445 | ret = -ENOENT; | |
2446 | goto out; | |
2447 | } | |
2448 | ||
2449 | backref->root_id = root_id; | |
2450 | backref->inum = inum; | |
e68afa49 | 2451 | backref->file_pos = offset; |
38c227d8 LB |
2452 | backref->num_bytes = num_bytes; |
2453 | backref->extent_offset = extent_offset; | |
2454 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2455 | backref->old = old; | |
2456 | backref_insert(&new->root, backref); | |
2457 | old->count++; | |
2458 | out: | |
2459 | btrfs_release_path(path); | |
2460 | WARN_ON(ret); | |
2461 | return ret; | |
2462 | } | |
2463 | ||
2464 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2465 | struct new_sa_defrag_extent *new) | |
2466 | { | |
0b246afa | 2467 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2468 | struct old_sa_defrag_extent *old, *tmp; |
2469 | int ret; | |
2470 | ||
2471 | new->path = path; | |
2472 | ||
2473 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2474 | ret = iterate_inodes_from_logical(old->bytenr + |
2475 | old->extent_offset, fs_info, | |
38c227d8 | 2476 | path, record_one_backref, |
c995ab3c | 2477 | old, false); |
4724b106 JB |
2478 | if (ret < 0 && ret != -ENOENT) |
2479 | return false; | |
38c227d8 LB |
2480 | |
2481 | /* no backref to be processed for this extent */ | |
2482 | if (!old->count) { | |
2483 | list_del(&old->list); | |
2484 | kfree(old); | |
2485 | } | |
2486 | } | |
2487 | ||
2488 | if (list_empty(&new->head)) | |
2489 | return false; | |
2490 | ||
2491 | return true; | |
2492 | } | |
2493 | ||
2494 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2495 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2496 | struct new_sa_defrag_extent *new) |
38c227d8 | 2497 | { |
116e0024 | 2498 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2499 | return 0; |
2500 | ||
2501 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2502 | return 0; | |
2503 | ||
116e0024 LB |
2504 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2505 | return 0; | |
2506 | ||
2507 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2508 | btrfs_file_extent_other_encoding(leaf, fi)) |
2509 | return 0; | |
2510 | ||
2511 | return 1; | |
2512 | } | |
2513 | ||
2514 | /* | |
2515 | * Note the backref might has changed, and in this case we just return 0. | |
2516 | */ | |
2517 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2518 | struct sa_defrag_extent_backref *prev, | |
2519 | struct sa_defrag_extent_backref *backref) | |
2520 | { | |
2521 | struct btrfs_file_extent_item *extent; | |
2522 | struct btrfs_file_extent_item *item; | |
2523 | struct btrfs_ordered_extent *ordered; | |
2524 | struct btrfs_trans_handle *trans; | |
38c227d8 LB |
2525 | struct btrfs_root *root; |
2526 | struct btrfs_key key; | |
2527 | struct extent_buffer *leaf; | |
2528 | struct old_sa_defrag_extent *old = backref->old; | |
2529 | struct new_sa_defrag_extent *new = old->new; | |
0b246afa | 2530 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2531 | struct inode *inode; |
2532 | struct extent_state *cached = NULL; | |
2533 | int ret = 0; | |
2534 | u64 start; | |
2535 | u64 len; | |
2536 | u64 lock_start; | |
2537 | u64 lock_end; | |
2538 | bool merge = false; | |
2539 | int index; | |
2540 | ||
2541 | if (prev && prev->root_id == backref->root_id && | |
2542 | prev->inum == backref->inum && | |
2543 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2544 | merge = true; | |
2545 | ||
2546 | /* step 1: get root */ | |
2547 | key.objectid = backref->root_id; | |
2548 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2549 | key.offset = (u64)-1; | |
2550 | ||
38c227d8 LB |
2551 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2552 | ||
2553 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2554 | if (IS_ERR(root)) { | |
2555 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2556 | if (PTR_ERR(root) == -ENOENT) | |
2557 | return 0; | |
2558 | return PTR_ERR(root); | |
2559 | } | |
38c227d8 | 2560 | |
bcbba5e6 WS |
2561 | if (btrfs_root_readonly(root)) { |
2562 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2563 | return 0; | |
2564 | } | |
2565 | ||
38c227d8 LB |
2566 | /* step 2: get inode */ |
2567 | key.objectid = backref->inum; | |
2568 | key.type = BTRFS_INODE_ITEM_KEY; | |
2569 | key.offset = 0; | |
2570 | ||
2571 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2572 | if (IS_ERR(inode)) { | |
2573 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2574 | return 0; | |
2575 | } | |
2576 | ||
2577 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2578 | ||
2579 | /* step 3: relink backref */ | |
2580 | lock_start = backref->file_pos; | |
2581 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2582 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2583 | &cached); |
38c227d8 LB |
2584 | |
2585 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2586 | if (ordered) { | |
2587 | btrfs_put_ordered_extent(ordered); | |
2588 | goto out_unlock; | |
2589 | } | |
2590 | ||
2591 | trans = btrfs_join_transaction(root); | |
2592 | if (IS_ERR(trans)) { | |
2593 | ret = PTR_ERR(trans); | |
2594 | goto out_unlock; | |
2595 | } | |
2596 | ||
2597 | key.objectid = backref->inum; | |
2598 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2599 | key.offset = backref->file_pos; | |
2600 | ||
2601 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2602 | if (ret < 0) { | |
2603 | goto out_free_path; | |
2604 | } else if (ret > 0) { | |
2605 | ret = 0; | |
2606 | goto out_free_path; | |
2607 | } | |
2608 | ||
2609 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2610 | struct btrfs_file_extent_item); | |
2611 | ||
2612 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2613 | backref->generation) | |
2614 | goto out_free_path; | |
2615 | ||
2616 | btrfs_release_path(path); | |
2617 | ||
2618 | start = backref->file_pos; | |
2619 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2620 | start += old->extent_offset + old->offset - | |
2621 | backref->extent_offset; | |
2622 | ||
2623 | len = min(backref->extent_offset + backref->num_bytes, | |
2624 | old->extent_offset + old->offset + old->len); | |
2625 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2626 | ||
2627 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2628 | start + len, 1); | |
2629 | if (ret) | |
2630 | goto out_free_path; | |
2631 | again: | |
4a0cc7ca | 2632 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2633 | key.type = BTRFS_EXTENT_DATA_KEY; |
2634 | key.offset = start; | |
2635 | ||
a09a0a70 | 2636 | path->leave_spinning = 1; |
38c227d8 LB |
2637 | if (merge) { |
2638 | struct btrfs_file_extent_item *fi; | |
2639 | u64 extent_len; | |
2640 | struct btrfs_key found_key; | |
2641 | ||
3c9665df | 2642 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2643 | if (ret < 0) |
2644 | goto out_free_path; | |
2645 | ||
2646 | path->slots[0]--; | |
2647 | leaf = path->nodes[0]; | |
2648 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2649 | ||
2650 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2651 | struct btrfs_file_extent_item); | |
2652 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2653 | ||
116e0024 LB |
2654 | if (extent_len + found_key.offset == start && |
2655 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2656 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2657 | extent_len + len); | |
2658 | btrfs_mark_buffer_dirty(leaf); | |
2659 | inode_add_bytes(inode, len); | |
2660 | ||
2661 | ret = 1; | |
2662 | goto out_free_path; | |
2663 | } else { | |
2664 | merge = false; | |
2665 | btrfs_release_path(path); | |
2666 | goto again; | |
2667 | } | |
2668 | } | |
2669 | ||
2670 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2671 | sizeof(*extent)); | |
2672 | if (ret) { | |
66642832 | 2673 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2674 | goto out_free_path; |
2675 | } | |
2676 | ||
2677 | leaf = path->nodes[0]; | |
2678 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2679 | struct btrfs_file_extent_item); | |
2680 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2681 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2682 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2683 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2684 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2685 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2686 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2687 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2688 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2689 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2690 | ||
2691 | btrfs_mark_buffer_dirty(leaf); | |
2692 | inode_add_bytes(inode, len); | |
a09a0a70 | 2693 | btrfs_release_path(path); |
38c227d8 | 2694 | |
84f7d8e6 | 2695 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, |
38c227d8 LB |
2696 | new->disk_len, 0, |
2697 | backref->root_id, backref->inum, | |
b06c4bf5 | 2698 | new->file_pos); /* start - extent_offset */ |
38c227d8 | 2699 | if (ret) { |
66642832 | 2700 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2701 | goto out_free_path; |
2702 | } | |
2703 | ||
2704 | ret = 1; | |
2705 | out_free_path: | |
2706 | btrfs_release_path(path); | |
a09a0a70 | 2707 | path->leave_spinning = 0; |
3a45bb20 | 2708 | btrfs_end_transaction(trans); |
38c227d8 LB |
2709 | out_unlock: |
2710 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
e43bbe5e | 2711 | &cached); |
38c227d8 LB |
2712 | iput(inode); |
2713 | return ret; | |
2714 | } | |
2715 | ||
6f519564 LB |
2716 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2717 | { | |
2718 | struct old_sa_defrag_extent *old, *tmp; | |
2719 | ||
2720 | if (!new) | |
2721 | return; | |
2722 | ||
2723 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2724 | kfree(old); |
2725 | } | |
2726 | kfree(new); | |
2727 | } | |
2728 | ||
38c227d8 LB |
2729 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2730 | { | |
0b246afa | 2731 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 | 2732 | struct btrfs_path *path; |
38c227d8 LB |
2733 | struct sa_defrag_extent_backref *backref; |
2734 | struct sa_defrag_extent_backref *prev = NULL; | |
38c227d8 LB |
2735 | struct rb_node *node; |
2736 | int ret; | |
2737 | ||
38c227d8 LB |
2738 | path = btrfs_alloc_path(); |
2739 | if (!path) | |
2740 | return; | |
2741 | ||
2742 | if (!record_extent_backrefs(path, new)) { | |
2743 | btrfs_free_path(path); | |
2744 | goto out; | |
2745 | } | |
2746 | btrfs_release_path(path); | |
2747 | ||
2748 | while (1) { | |
2749 | node = rb_first(&new->root); | |
2750 | if (!node) | |
2751 | break; | |
2752 | rb_erase(node, &new->root); | |
2753 | ||
2754 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2755 | ||
2756 | ret = relink_extent_backref(path, prev, backref); | |
2757 | WARN_ON(ret < 0); | |
2758 | ||
2759 | kfree(prev); | |
2760 | ||
2761 | if (ret == 1) | |
2762 | prev = backref; | |
2763 | else | |
2764 | prev = NULL; | |
2765 | cond_resched(); | |
2766 | } | |
2767 | kfree(prev); | |
2768 | ||
2769 | btrfs_free_path(path); | |
38c227d8 | 2770 | out: |
6f519564 LB |
2771 | free_sa_defrag_extent(new); |
2772 | ||
0b246afa JM |
2773 | atomic_dec(&fs_info->defrag_running); |
2774 | wake_up(&fs_info->transaction_wait); | |
38c227d8 LB |
2775 | } |
2776 | ||
2777 | static struct new_sa_defrag_extent * | |
2778 | record_old_file_extents(struct inode *inode, | |
2779 | struct btrfs_ordered_extent *ordered) | |
2780 | { | |
0b246afa | 2781 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2782 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2783 | struct btrfs_path *path; | |
2784 | struct btrfs_key key; | |
6f519564 | 2785 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2786 | struct new_sa_defrag_extent *new; |
2787 | int ret; | |
2788 | ||
2789 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2790 | if (!new) | |
2791 | return NULL; | |
2792 | ||
2793 | new->inode = inode; | |
2794 | new->file_pos = ordered->file_offset; | |
2795 | new->len = ordered->len; | |
2796 | new->bytenr = ordered->start; | |
2797 | new->disk_len = ordered->disk_len; | |
2798 | new->compress_type = ordered->compress_type; | |
2799 | new->root = RB_ROOT; | |
2800 | INIT_LIST_HEAD(&new->head); | |
2801 | ||
2802 | path = btrfs_alloc_path(); | |
2803 | if (!path) | |
2804 | goto out_kfree; | |
2805 | ||
4a0cc7ca | 2806 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2807 | key.type = BTRFS_EXTENT_DATA_KEY; |
2808 | key.offset = new->file_pos; | |
2809 | ||
2810 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2811 | if (ret < 0) | |
2812 | goto out_free_path; | |
2813 | if (ret > 0 && path->slots[0] > 0) | |
2814 | path->slots[0]--; | |
2815 | ||
2816 | /* find out all the old extents for the file range */ | |
2817 | while (1) { | |
2818 | struct btrfs_file_extent_item *extent; | |
2819 | struct extent_buffer *l; | |
2820 | int slot; | |
2821 | u64 num_bytes; | |
2822 | u64 offset; | |
2823 | u64 end; | |
2824 | u64 disk_bytenr; | |
2825 | u64 extent_offset; | |
2826 | ||
2827 | l = path->nodes[0]; | |
2828 | slot = path->slots[0]; | |
2829 | ||
2830 | if (slot >= btrfs_header_nritems(l)) { | |
2831 | ret = btrfs_next_leaf(root, path); | |
2832 | if (ret < 0) | |
6f519564 | 2833 | goto out_free_path; |
38c227d8 LB |
2834 | else if (ret > 0) |
2835 | break; | |
2836 | continue; | |
2837 | } | |
2838 | ||
2839 | btrfs_item_key_to_cpu(l, &key, slot); | |
2840 | ||
4a0cc7ca | 2841 | if (key.objectid != btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2842 | break; |
2843 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2844 | break; | |
2845 | if (key.offset >= new->file_pos + new->len) | |
2846 | break; | |
2847 | ||
2848 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2849 | ||
2850 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2851 | if (key.offset + num_bytes < new->file_pos) | |
2852 | goto next; | |
2853 | ||
2854 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2855 | if (!disk_bytenr) | |
2856 | goto next; | |
2857 | ||
2858 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2859 | ||
2860 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2861 | if (!old) | |
6f519564 | 2862 | goto out_free_path; |
38c227d8 LB |
2863 | |
2864 | offset = max(new->file_pos, key.offset); | |
2865 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2866 | ||
2867 | old->bytenr = disk_bytenr; | |
2868 | old->extent_offset = extent_offset; | |
2869 | old->offset = offset - key.offset; | |
2870 | old->len = end - offset; | |
2871 | old->new = new; | |
2872 | old->count = 0; | |
2873 | list_add_tail(&old->list, &new->head); | |
2874 | next: | |
2875 | path->slots[0]++; | |
2876 | cond_resched(); | |
2877 | } | |
2878 | ||
2879 | btrfs_free_path(path); | |
0b246afa | 2880 | atomic_inc(&fs_info->defrag_running); |
38c227d8 LB |
2881 | |
2882 | return new; | |
2883 | ||
38c227d8 LB |
2884 | out_free_path: |
2885 | btrfs_free_path(path); | |
2886 | out_kfree: | |
6f519564 | 2887 | free_sa_defrag_extent(new); |
38c227d8 LB |
2888 | return NULL; |
2889 | } | |
2890 | ||
2ff7e61e | 2891 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2892 | u64 start, u64 len) |
2893 | { | |
2894 | struct btrfs_block_group_cache *cache; | |
2895 | ||
0b246afa | 2896 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2897 | ASSERT(cache); |
2898 | ||
2899 | spin_lock(&cache->lock); | |
2900 | cache->delalloc_bytes -= len; | |
2901 | spin_unlock(&cache->lock); | |
2902 | ||
2903 | btrfs_put_block_group(cache); | |
2904 | } | |
2905 | ||
d352ac68 CM |
2906 | /* as ordered data IO finishes, this gets called so we can finish |
2907 | * an ordered extent if the range of bytes in the file it covers are | |
2908 | * fully written. | |
2909 | */ | |
5fd02043 | 2910 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2911 | { |
5fd02043 | 2912 | struct inode *inode = ordered_extent->inode; |
0b246afa | 2913 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 2914 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2915 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2916 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2917 | struct extent_state *cached_state = NULL; |
38c227d8 | 2918 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2919 | int compress_type = 0; |
77cef2ec JB |
2920 | int ret = 0; |
2921 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2922 | bool nolock; |
77cef2ec | 2923 | bool truncated = false; |
a7e3b975 FM |
2924 | bool range_locked = false; |
2925 | bool clear_new_delalloc_bytes = false; | |
49940bdd | 2926 | bool clear_reserved_extent = true; |
a7e3b975 FM |
2927 | |
2928 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
2929 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
2930 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags)) | |
2931 | clear_new_delalloc_bytes = true; | |
e6dcd2dc | 2932 | |
70ddc553 | 2933 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
0cb59c99 | 2934 | |
5fd02043 JB |
2935 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2936 | ret = -EIO; | |
2937 | goto out; | |
2938 | } | |
2939 | ||
7ab7956e NB |
2940 | btrfs_free_io_failure_record(BTRFS_I(inode), |
2941 | ordered_extent->file_offset, | |
2942 | ordered_extent->file_offset + | |
2943 | ordered_extent->len - 1); | |
f612496b | 2944 | |
77cef2ec JB |
2945 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2946 | truncated = true; | |
2947 | logical_len = ordered_extent->truncated_len; | |
2948 | /* Truncated the entire extent, don't bother adding */ | |
2949 | if (!logical_len) | |
2950 | goto out; | |
2951 | } | |
2952 | ||
c2167754 | 2953 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2954 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
2955 | |
2956 | /* | |
2957 | * For mwrite(mmap + memset to write) case, we still reserve | |
2958 | * space for NOCOW range. | |
2959 | * As NOCOW won't cause a new delayed ref, just free the space | |
2960 | */ | |
bc42bda2 | 2961 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
94ed938a | 2962 | ordered_extent->len); |
6c760c07 JB |
2963 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2964 | if (nolock) | |
2965 | trans = btrfs_join_transaction_nolock(root); | |
2966 | else | |
2967 | trans = btrfs_join_transaction(root); | |
2968 | if (IS_ERR(trans)) { | |
2969 | ret = PTR_ERR(trans); | |
2970 | trans = NULL; | |
2971 | goto out; | |
c2167754 | 2972 | } |
69fe2d75 | 2973 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
6c760c07 JB |
2974 | ret = btrfs_update_inode_fallback(trans, root, inode); |
2975 | if (ret) /* -ENOMEM or corruption */ | |
66642832 | 2976 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
2977 | goto out; |
2978 | } | |
e6dcd2dc | 2979 | |
a7e3b975 | 2980 | range_locked = true; |
2ac55d41 JB |
2981 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2982 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 2983 | &cached_state); |
e6dcd2dc | 2984 | |
38c227d8 LB |
2985 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2986 | ordered_extent->file_offset + ordered_extent->len - 1, | |
452e62b7 | 2987 | EXTENT_DEFRAG, 0, cached_state); |
38c227d8 LB |
2988 | if (ret) { |
2989 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2990 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2991 | /* the inode is shared */ |
2992 | new = record_old_file_extents(inode, ordered_extent); | |
2993 | ||
2994 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2995 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ae0f1625 | 2996 | EXTENT_DEFRAG, 0, 0, &cached_state); |
38c227d8 LB |
2997 | } |
2998 | ||
0cb59c99 | 2999 | if (nolock) |
7a7eaa40 | 3000 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 3001 | else |
7a7eaa40 | 3002 | trans = btrfs_join_transaction(root); |
79787eaa JM |
3003 | if (IS_ERR(trans)) { |
3004 | ret = PTR_ERR(trans); | |
3005 | trans = NULL; | |
a7e3b975 | 3006 | goto out; |
79787eaa | 3007 | } |
a79b7d4b | 3008 | |
69fe2d75 | 3009 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
c2167754 | 3010 | |
c8b97818 | 3011 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3012 | compress_type = ordered_extent->compress_type; |
d899e052 | 3013 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3014 | BUG_ON(compress_type); |
b430b775 JM |
3015 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
3016 | ordered_extent->len); | |
7a6d7067 | 3017 | ret = btrfs_mark_extent_written(trans, BTRFS_I(inode), |
d899e052 YZ |
3018 | ordered_extent->file_offset, |
3019 | ordered_extent->file_offset + | |
77cef2ec | 3020 | logical_len); |
d899e052 | 3021 | } else { |
0b246afa | 3022 | BUG_ON(root == fs_info->tree_root); |
d899e052 YZ |
3023 | ret = insert_reserved_file_extent(trans, inode, |
3024 | ordered_extent->file_offset, | |
3025 | ordered_extent->start, | |
3026 | ordered_extent->disk_len, | |
77cef2ec | 3027 | logical_len, logical_len, |
261507a0 | 3028 | compress_type, 0, 0, |
d899e052 | 3029 | BTRFS_FILE_EXTENT_REG); |
49940bdd JB |
3030 | if (!ret) { |
3031 | clear_reserved_extent = false; | |
2ff7e61e | 3032 | btrfs_release_delalloc_bytes(fs_info, |
e570fd27 MX |
3033 | ordered_extent->start, |
3034 | ordered_extent->disk_len); | |
49940bdd | 3035 | } |
d899e052 | 3036 | } |
5dc562c5 JB |
3037 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
3038 | ordered_extent->file_offset, ordered_extent->len, | |
3039 | trans->transid); | |
79787eaa | 3040 | if (ret < 0) { |
66642832 | 3041 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3042 | goto out; |
79787eaa | 3043 | } |
2ac55d41 | 3044 | |
ac01f26a NB |
3045 | ret = add_pending_csums(trans, inode, &ordered_extent->list); |
3046 | if (ret) { | |
3047 | btrfs_abort_transaction(trans, ret); | |
3048 | goto out; | |
3049 | } | |
e6dcd2dc | 3050 | |
6c760c07 JB |
3051 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
3052 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
3053 | if (ret) { /* -ENOMEM or corruption */ | |
66642832 | 3054 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3055 | goto out; |
1ef30be1 JB |
3056 | } |
3057 | ret = 0; | |
c2167754 | 3058 | out: |
a7e3b975 FM |
3059 | if (range_locked || clear_new_delalloc_bytes) { |
3060 | unsigned int clear_bits = 0; | |
3061 | ||
3062 | if (range_locked) | |
3063 | clear_bits |= EXTENT_LOCKED; | |
3064 | if (clear_new_delalloc_bytes) | |
3065 | clear_bits |= EXTENT_DELALLOC_NEW; | |
3066 | clear_extent_bit(&BTRFS_I(inode)->io_tree, | |
3067 | ordered_extent->file_offset, | |
3068 | ordered_extent->file_offset + | |
3069 | ordered_extent->len - 1, | |
3070 | clear_bits, | |
3071 | (clear_bits & EXTENT_LOCKED) ? 1 : 0, | |
ae0f1625 | 3072 | 0, &cached_state); |
a7e3b975 FM |
3073 | } |
3074 | ||
a698d075 | 3075 | if (trans) |
3a45bb20 | 3076 | btrfs_end_transaction(trans); |
0cb59c99 | 3077 | |
77cef2ec JB |
3078 | if (ret || truncated) { |
3079 | u64 start, end; | |
3080 | ||
3081 | if (truncated) | |
3082 | start = ordered_extent->file_offset + logical_len; | |
3083 | else | |
3084 | start = ordered_extent->file_offset; | |
3085 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
f08dc36f | 3086 | clear_extent_uptodate(io_tree, start, end, NULL); |
77cef2ec JB |
3087 | |
3088 | /* Drop the cache for the part of the extent we didn't write. */ | |
dcdbc059 | 3089 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); |
5fd02043 | 3090 | |
0bec9ef5 JB |
3091 | /* |
3092 | * If the ordered extent had an IOERR or something else went | |
3093 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3094 | * back to the allocator. We only free the extent in the |
3095 | * truncated case if we didn't write out the extent at all. | |
49940bdd JB |
3096 | * |
3097 | * If we made it past insert_reserved_file_extent before we | |
3098 | * errored out then we don't need to do this as the accounting | |
3099 | * has already been done. | |
0bec9ef5 | 3100 | */ |
77cef2ec | 3101 | if ((ret || !logical_len) && |
49940bdd | 3102 | clear_reserved_extent && |
77cef2ec | 3103 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
0bec9ef5 | 3104 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2ff7e61e JM |
3105 | btrfs_free_reserved_extent(fs_info, |
3106 | ordered_extent->start, | |
e570fd27 | 3107 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
3108 | } |
3109 | ||
3110 | ||
5fd02043 | 3111 | /* |
8bad3c02 LB |
3112 | * This needs to be done to make sure anybody waiting knows we are done |
3113 | * updating everything for this ordered extent. | |
5fd02043 JB |
3114 | */ |
3115 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
3116 | ||
38c227d8 | 3117 | /* for snapshot-aware defrag */ |
6f519564 LB |
3118 | if (new) { |
3119 | if (ret) { | |
3120 | free_sa_defrag_extent(new); | |
0b246afa | 3121 | atomic_dec(&fs_info->defrag_running); |
6f519564 LB |
3122 | } else { |
3123 | relink_file_extents(new); | |
3124 | } | |
3125 | } | |
38c227d8 | 3126 | |
e6dcd2dc CM |
3127 | /* once for us */ |
3128 | btrfs_put_ordered_extent(ordered_extent); | |
3129 | /* once for the tree */ | |
3130 | btrfs_put_ordered_extent(ordered_extent); | |
3131 | ||
e73e81b6 EL |
3132 | /* Try to release some metadata so we don't get an OOM but don't wait */ |
3133 | btrfs_btree_balance_dirty_nodelay(fs_info); | |
3134 | ||
5fd02043 JB |
3135 | return ret; |
3136 | } | |
3137 | ||
3138 | static void finish_ordered_fn(struct btrfs_work *work) | |
3139 | { | |
3140 | struct btrfs_ordered_extent *ordered_extent; | |
3141 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3142 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3143 | } |
3144 | ||
c629732d NB |
3145 | void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start, |
3146 | u64 end, int uptodate) | |
211f90e6 | 3147 | { |
5fd02043 | 3148 | struct inode *inode = page->mapping->host; |
0b246afa | 3149 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5fd02043 | 3150 | struct btrfs_ordered_extent *ordered_extent = NULL; |
9e0af237 LB |
3151 | struct btrfs_workqueue *wq; |
3152 | btrfs_work_func_t func; | |
5fd02043 | 3153 | |
1abe9b8a | 3154 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
3155 | ||
8b62b72b | 3156 | ClearPagePrivate2(page); |
5fd02043 JB |
3157 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
3158 | end - start + 1, uptodate)) | |
c3988d63 | 3159 | return; |
5fd02043 | 3160 | |
70ddc553 | 3161 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
0b246afa | 3162 | wq = fs_info->endio_freespace_worker; |
9e0af237 LB |
3163 | func = btrfs_freespace_write_helper; |
3164 | } else { | |
0b246afa | 3165 | wq = fs_info->endio_write_workers; |
9e0af237 LB |
3166 | func = btrfs_endio_write_helper; |
3167 | } | |
5fd02043 | 3168 | |
9e0af237 LB |
3169 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3170 | NULL); | |
3171 | btrfs_queue_work(wq, &ordered_extent->work); | |
211f90e6 CM |
3172 | } |
3173 | ||
dc380aea MX |
3174 | static int __readpage_endio_check(struct inode *inode, |
3175 | struct btrfs_io_bio *io_bio, | |
3176 | int icsum, struct page *page, | |
3177 | int pgoff, u64 start, size_t len) | |
3178 | { | |
3179 | char *kaddr; | |
3180 | u32 csum_expected; | |
3181 | u32 csum = ~(u32)0; | |
dc380aea MX |
3182 | |
3183 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3184 | ||
3185 | kaddr = kmap_atomic(page); | |
3186 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
0b5e3daf | 3187 | btrfs_csum_final(csum, (u8 *)&csum); |
dc380aea MX |
3188 | if (csum != csum_expected) |
3189 | goto zeroit; | |
3190 | ||
3191 | kunmap_atomic(kaddr); | |
3192 | return 0; | |
3193 | zeroit: | |
0970a22e | 3194 | btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected, |
6f6b643e | 3195 | io_bio->mirror_num); |
dc380aea MX |
3196 | memset(kaddr + pgoff, 1, len); |
3197 | flush_dcache_page(page); | |
3198 | kunmap_atomic(kaddr); | |
dc380aea MX |
3199 | return -EIO; |
3200 | } | |
3201 | ||
d352ac68 CM |
3202 | /* |
3203 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3204 | * if there's a match, we allow the bio to finish. If not, the code in |
3205 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3206 | */ |
facc8a22 MX |
3207 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3208 | u64 phy_offset, struct page *page, | |
3209 | u64 start, u64 end, int mirror) | |
07157aac | 3210 | { |
4eee4fa4 | 3211 | size_t offset = start - page_offset(page); |
07157aac | 3212 | struct inode *inode = page->mapping->host; |
d1310b2e | 3213 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3214 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3215 | |
d20f7043 CM |
3216 | if (PageChecked(page)) { |
3217 | ClearPageChecked(page); | |
dc380aea | 3218 | return 0; |
d20f7043 | 3219 | } |
6cbff00f CH |
3220 | |
3221 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3222 | return 0; |
17d217fe YZ |
3223 | |
3224 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3225 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3226 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3227 | return 0; |
17d217fe | 3228 | } |
d20f7043 | 3229 | |
facc8a22 | 3230 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3231 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3232 | start, (size_t)(end - start + 1)); | |
07157aac | 3233 | } |
b888db2b | 3234 | |
c1c3fac2 NB |
3235 | /* |
3236 | * btrfs_add_delayed_iput - perform a delayed iput on @inode | |
3237 | * | |
3238 | * @inode: The inode we want to perform iput on | |
3239 | * | |
3240 | * This function uses the generic vfs_inode::i_count to track whether we should | |
3241 | * just decrement it (in case it's > 1) or if this is the last iput then link | |
3242 | * the inode to the delayed iput machinery. Delayed iputs are processed at | |
3243 | * transaction commit time/superblock commit/cleaner kthread. | |
3244 | */ | |
24bbcf04 YZ |
3245 | void btrfs_add_delayed_iput(struct inode *inode) |
3246 | { | |
0b246afa | 3247 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3248 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3249 | |
3250 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3251 | return; | |
3252 | ||
24bbcf04 | 3253 | spin_lock(&fs_info->delayed_iput_lock); |
c1c3fac2 NB |
3254 | ASSERT(list_empty(&binode->delayed_iput)); |
3255 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
24bbcf04 YZ |
3256 | spin_unlock(&fs_info->delayed_iput_lock); |
3257 | } | |
3258 | ||
2ff7e61e | 3259 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3260 | { |
24bbcf04 | 3261 | |
24bbcf04 | 3262 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3263 | while (!list_empty(&fs_info->delayed_iputs)) { |
3264 | struct btrfs_inode *inode; | |
3265 | ||
3266 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3267 | struct btrfs_inode, delayed_iput); | |
c1c3fac2 | 3268 | list_del_init(&inode->delayed_iput); |
8089fe62 DS |
3269 | spin_unlock(&fs_info->delayed_iput_lock); |
3270 | iput(&inode->vfs_inode); | |
3271 | spin_lock(&fs_info->delayed_iput_lock); | |
24bbcf04 | 3272 | } |
8089fe62 | 3273 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3274 | } |
3275 | ||
7b128766 | 3276 | /* |
f7e9e8fc OS |
3277 | * This creates an orphan entry for the given inode in case something goes wrong |
3278 | * in the middle of an unlink. | |
7b128766 | 3279 | */ |
73f2e545 | 3280 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
27919067 | 3281 | struct btrfs_inode *inode) |
7b128766 | 3282 | { |
d68fc57b | 3283 | int ret; |
7b128766 | 3284 | |
27919067 OS |
3285 | ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode)); |
3286 | if (ret && ret != -EEXIST) { | |
3287 | btrfs_abort_transaction(trans, ret); | |
3288 | return ret; | |
d68fc57b YZ |
3289 | } |
3290 | ||
d68fc57b | 3291 | return 0; |
7b128766 JB |
3292 | } |
3293 | ||
3294 | /* | |
f7e9e8fc OS |
3295 | * We have done the delete so we can go ahead and remove the orphan item for |
3296 | * this particular inode. | |
7b128766 | 3297 | */ |
48a3b636 | 3298 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3299 | struct btrfs_inode *inode) |
7b128766 | 3300 | { |
27919067 | 3301 | return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode)); |
7b128766 JB |
3302 | } |
3303 | ||
3304 | /* | |
3305 | * this cleans up any orphans that may be left on the list from the last use | |
3306 | * of this root. | |
3307 | */ | |
66b4ffd1 | 3308 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3309 | { |
0b246afa | 3310 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3311 | struct btrfs_path *path; |
3312 | struct extent_buffer *leaf; | |
7b128766 JB |
3313 | struct btrfs_key key, found_key; |
3314 | struct btrfs_trans_handle *trans; | |
3315 | struct inode *inode; | |
8f6d7f4f | 3316 | u64 last_objectid = 0; |
f7e9e8fc | 3317 | int ret = 0, nr_unlink = 0; |
7b128766 | 3318 | |
d68fc57b | 3319 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3320 | return 0; |
c71bf099 YZ |
3321 | |
3322 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3323 | if (!path) { |
3324 | ret = -ENOMEM; | |
3325 | goto out; | |
3326 | } | |
e4058b54 | 3327 | path->reada = READA_BACK; |
7b128766 JB |
3328 | |
3329 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3330 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3331 | key.offset = (u64)-1; |
3332 | ||
7b128766 JB |
3333 | while (1) { |
3334 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3335 | if (ret < 0) |
3336 | goto out; | |
7b128766 JB |
3337 | |
3338 | /* | |
3339 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3340 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3341 | * find the key and see if we have stuff that matches |
3342 | */ | |
3343 | if (ret > 0) { | |
66b4ffd1 | 3344 | ret = 0; |
7b128766 JB |
3345 | if (path->slots[0] == 0) |
3346 | break; | |
3347 | path->slots[0]--; | |
3348 | } | |
3349 | ||
3350 | /* pull out the item */ | |
3351 | leaf = path->nodes[0]; | |
7b128766 JB |
3352 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3353 | ||
3354 | /* make sure the item matches what we want */ | |
3355 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3356 | break; | |
962a298f | 3357 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3358 | break; |
3359 | ||
3360 | /* release the path since we're done with it */ | |
b3b4aa74 | 3361 | btrfs_release_path(path); |
7b128766 JB |
3362 | |
3363 | /* | |
3364 | * this is where we are basically btrfs_lookup, without the | |
3365 | * crossing root thing. we store the inode number in the | |
3366 | * offset of the orphan item. | |
3367 | */ | |
8f6d7f4f JB |
3368 | |
3369 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3370 | btrfs_err(fs_info, |
3371 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3372 | ret = -EINVAL; |
3373 | goto out; | |
3374 | } | |
3375 | ||
3376 | last_objectid = found_key.offset; | |
3377 | ||
5d4f98a2 YZ |
3378 | found_key.objectid = found_key.offset; |
3379 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3380 | found_key.offset = 0; | |
0b246afa | 3381 | inode = btrfs_iget(fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3382 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3383 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3384 | goto out; |
7b128766 | 3385 | |
0b246afa | 3386 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 AJ |
3387 | struct btrfs_root *dead_root; |
3388 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3389 | int is_dead_root = 0; | |
3390 | ||
3391 | /* | |
3392 | * this is an orphan in the tree root. Currently these | |
3393 | * could come from 2 sources: | |
3394 | * a) a snapshot deletion in progress | |
3395 | * b) a free space cache inode | |
3396 | * We need to distinguish those two, as the snapshot | |
3397 | * orphan must not get deleted. | |
3398 | * find_dead_roots already ran before us, so if this | |
3399 | * is a snapshot deletion, we should find the root | |
3400 | * in the dead_roots list | |
3401 | */ | |
3402 | spin_lock(&fs_info->trans_lock); | |
3403 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3404 | root_list) { | |
3405 | if (dead_root->root_key.objectid == | |
3406 | found_key.objectid) { | |
3407 | is_dead_root = 1; | |
3408 | break; | |
3409 | } | |
3410 | } | |
3411 | spin_unlock(&fs_info->trans_lock); | |
3412 | if (is_dead_root) { | |
3413 | /* prevent this orphan from being found again */ | |
3414 | key.offset = found_key.objectid - 1; | |
3415 | continue; | |
3416 | } | |
f7e9e8fc | 3417 | |
f8e9e0b0 | 3418 | } |
f7e9e8fc | 3419 | |
7b128766 | 3420 | /* |
f7e9e8fc OS |
3421 | * If we have an inode with links, there are a couple of |
3422 | * possibilities. Old kernels (before v3.12) used to create an | |
3423 | * orphan item for truncate indicating that there were possibly | |
3424 | * extent items past i_size that needed to be deleted. In v3.12, | |
3425 | * truncate was changed to update i_size in sync with the extent | |
3426 | * items, but the (useless) orphan item was still created. Since | |
3427 | * v4.18, we don't create the orphan item for truncate at all. | |
3428 | * | |
3429 | * So, this item could mean that we need to do a truncate, but | |
3430 | * only if this filesystem was last used on a pre-v3.12 kernel | |
3431 | * and was not cleanly unmounted. The odds of that are quite | |
3432 | * slim, and it's a pain to do the truncate now, so just delete | |
3433 | * the orphan item. | |
3434 | * | |
3435 | * It's also possible that this orphan item was supposed to be | |
3436 | * deleted but wasn't. The inode number may have been reused, | |
3437 | * but either way, we can delete the orphan item. | |
7b128766 | 3438 | */ |
f7e9e8fc OS |
3439 | if (ret == -ENOENT || inode->i_nlink) { |
3440 | if (!ret) | |
3441 | iput(inode); | |
a8c9e576 | 3442 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3443 | if (IS_ERR(trans)) { |
3444 | ret = PTR_ERR(trans); | |
3445 | goto out; | |
3446 | } | |
0b246afa JM |
3447 | btrfs_debug(fs_info, "auto deleting %Lu", |
3448 | found_key.objectid); | |
a8c9e576 JB |
3449 | ret = btrfs_del_orphan_item(trans, root, |
3450 | found_key.objectid); | |
3a45bb20 | 3451 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3452 | if (ret) |
3453 | goto out; | |
7b128766 JB |
3454 | continue; |
3455 | } | |
3456 | ||
f7e9e8fc | 3457 | nr_unlink++; |
7b128766 JB |
3458 | |
3459 | /* this will do delete_inode and everything for us */ | |
3460 | iput(inode); | |
3461 | } | |
3254c876 MX |
3462 | /* release the path since we're done with it */ |
3463 | btrfs_release_path(path); | |
3464 | ||
d68fc57b YZ |
3465 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3466 | ||
a575ceeb | 3467 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { |
7a7eaa40 | 3468 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3469 | if (!IS_ERR(trans)) |
3a45bb20 | 3470 | btrfs_end_transaction(trans); |
d68fc57b | 3471 | } |
7b128766 JB |
3472 | |
3473 | if (nr_unlink) | |
0b246afa | 3474 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
66b4ffd1 JB |
3475 | |
3476 | out: | |
3477 | if (ret) | |
0b246afa | 3478 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3479 | btrfs_free_path(path); |
3480 | return ret; | |
7b128766 JB |
3481 | } |
3482 | ||
46a53cca CM |
3483 | /* |
3484 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3485 | * don't find any xattrs, we know there can't be any acls. | |
3486 | * | |
3487 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3488 | */ | |
3489 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3490 | int slot, u64 objectid, |
3491 | int *first_xattr_slot) | |
46a53cca CM |
3492 | { |
3493 | u32 nritems = btrfs_header_nritems(leaf); | |
3494 | struct btrfs_key found_key; | |
f23b5a59 JB |
3495 | static u64 xattr_access = 0; |
3496 | static u64 xattr_default = 0; | |
46a53cca CM |
3497 | int scanned = 0; |
3498 | ||
f23b5a59 | 3499 | if (!xattr_access) { |
97d79299 AG |
3500 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3501 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3502 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3503 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3504 | } |
3505 | ||
46a53cca | 3506 | slot++; |
63541927 | 3507 | *first_xattr_slot = -1; |
46a53cca CM |
3508 | while (slot < nritems) { |
3509 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3510 | ||
3511 | /* we found a different objectid, there must not be acls */ | |
3512 | if (found_key.objectid != objectid) | |
3513 | return 0; | |
3514 | ||
3515 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3516 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3517 | if (*first_xattr_slot == -1) |
3518 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3519 | if (found_key.offset == xattr_access || |
3520 | found_key.offset == xattr_default) | |
3521 | return 1; | |
3522 | } | |
46a53cca CM |
3523 | |
3524 | /* | |
3525 | * we found a key greater than an xattr key, there can't | |
3526 | * be any acls later on | |
3527 | */ | |
3528 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3529 | return 0; | |
3530 | ||
3531 | slot++; | |
3532 | scanned++; | |
3533 | ||
3534 | /* | |
3535 | * it goes inode, inode backrefs, xattrs, extents, | |
3536 | * so if there are a ton of hard links to an inode there can | |
3537 | * be a lot of backrefs. Don't waste time searching too hard, | |
3538 | * this is just an optimization | |
3539 | */ | |
3540 | if (scanned >= 8) | |
3541 | break; | |
3542 | } | |
3543 | /* we hit the end of the leaf before we found an xattr or | |
3544 | * something larger than an xattr. We have to assume the inode | |
3545 | * has acls | |
3546 | */ | |
63541927 FDBM |
3547 | if (*first_xattr_slot == -1) |
3548 | *first_xattr_slot = slot; | |
46a53cca CM |
3549 | return 1; |
3550 | } | |
3551 | ||
d352ac68 CM |
3552 | /* |
3553 | * read an inode from the btree into the in-memory inode | |
3554 | */ | |
4222ea71 FM |
3555 | static int btrfs_read_locked_inode(struct inode *inode, |
3556 | struct btrfs_path *in_path) | |
39279cc3 | 3557 | { |
0b246afa | 3558 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4222ea71 | 3559 | struct btrfs_path *path = in_path; |
5f39d397 | 3560 | struct extent_buffer *leaf; |
39279cc3 CM |
3561 | struct btrfs_inode_item *inode_item; |
3562 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3563 | struct btrfs_key location; | |
67de1176 | 3564 | unsigned long ptr; |
46a53cca | 3565 | int maybe_acls; |
618e21d5 | 3566 | u32 rdev; |
39279cc3 | 3567 | int ret; |
2f7e33d4 | 3568 | bool filled = false; |
63541927 | 3569 | int first_xattr_slot; |
2f7e33d4 MX |
3570 | |
3571 | ret = btrfs_fill_inode(inode, &rdev); | |
3572 | if (!ret) | |
3573 | filled = true; | |
39279cc3 | 3574 | |
4222ea71 FM |
3575 | if (!path) { |
3576 | path = btrfs_alloc_path(); | |
3577 | if (!path) | |
3578 | return -ENOMEM; | |
3579 | } | |
1748f843 | 3580 | |
39279cc3 | 3581 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3582 | |
39279cc3 | 3583 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 | 3584 | if (ret) { |
4222ea71 FM |
3585 | if (path != in_path) |
3586 | btrfs_free_path(path); | |
f5b3a417 | 3587 | return ret; |
67710892 | 3588 | } |
39279cc3 | 3589 | |
5f39d397 | 3590 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3591 | |
3592 | if (filled) | |
67de1176 | 3593 | goto cache_index; |
2f7e33d4 | 3594 | |
5f39d397 CM |
3595 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3596 | struct btrfs_inode_item); | |
5f39d397 | 3597 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3598 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3599 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3600 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3601 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3602 | |
a937b979 DS |
3603 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3604 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3605 | |
a937b979 DS |
3606 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3607 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3608 | |
a937b979 DS |
3609 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3610 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3611 | |
9cc97d64 | 3612 | BTRFS_I(inode)->i_otime.tv_sec = |
3613 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3614 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3615 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3616 | |
a76a3cd4 | 3617 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3618 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3619 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3620 | ||
c7f88c4e JL |
3621 | inode_set_iversion_queried(inode, |
3622 | btrfs_inode_sequence(leaf, inode_item)); | |
6e17d30b YD |
3623 | inode->i_generation = BTRFS_I(inode)->generation; |
3624 | inode->i_rdev = 0; | |
3625 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3626 | ||
3627 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3628 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3629 | ||
3630 | cache_index: | |
5dc562c5 JB |
3631 | /* |
3632 | * If we were modified in the current generation and evicted from memory | |
3633 | * and then re-read we need to do a full sync since we don't have any | |
3634 | * idea about which extents were modified before we were evicted from | |
3635 | * cache. | |
6e17d30b YD |
3636 | * |
3637 | * This is required for both inode re-read from disk and delayed inode | |
3638 | * in delayed_nodes_tree. | |
5dc562c5 | 3639 | */ |
0b246afa | 3640 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3641 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3642 | &BTRFS_I(inode)->runtime_flags); | |
3643 | ||
bde6c242 FM |
3644 | /* |
3645 | * We don't persist the id of the transaction where an unlink operation | |
3646 | * against the inode was last made. So here we assume the inode might | |
3647 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3648 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3649 | * between the inode and its parent if the inode is fsync'ed and the log | |
3650 | * replayed. For example, in the scenario: | |
3651 | * | |
3652 | * touch mydir/foo | |
3653 | * ln mydir/foo mydir/bar | |
3654 | * sync | |
3655 | * unlink mydir/bar | |
3656 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3657 | * xfs_io -c fsync mydir/foo | |
3658 | * <power failure> | |
3659 | * mount fs, triggers fsync log replay | |
3660 | * | |
3661 | * We must make sure that when we fsync our inode foo we also log its | |
3662 | * parent inode, otherwise after log replay the parent still has the | |
3663 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3664 | * and doesn't have an inode ref with the name "bar" anymore. | |
3665 | * | |
3666 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3667 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3668 | * transaction commits on fsync if our inode is a directory, or if our |
3669 | * inode is not a directory, logging its parent unnecessarily. | |
3670 | */ | |
3671 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
41bd6067 FM |
3672 | /* |
3673 | * Similar reasoning for last_link_trans, needs to be set otherwise | |
3674 | * for a case like the following: | |
3675 | * | |
3676 | * mkdir A | |
3677 | * touch foo | |
3678 | * ln foo A/bar | |
3679 | * echo 2 > /proc/sys/vm/drop_caches | |
3680 | * fsync foo | |
3681 | * <power failure> | |
3682 | * | |
3683 | * Would result in link bar and directory A not existing after the power | |
3684 | * failure. | |
3685 | */ | |
3686 | BTRFS_I(inode)->last_link_trans = BTRFS_I(inode)->last_trans; | |
bde6c242 | 3687 | |
67de1176 MX |
3688 | path->slots[0]++; |
3689 | if (inode->i_nlink != 1 || | |
3690 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3691 | goto cache_acl; | |
3692 | ||
3693 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3694 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3695 | goto cache_acl; |
3696 | ||
3697 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3698 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3699 | struct btrfs_inode_ref *ref; | |
3700 | ||
3701 | ref = (struct btrfs_inode_ref *)ptr; | |
3702 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3703 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3704 | struct btrfs_inode_extref *extref; | |
3705 | ||
3706 | extref = (struct btrfs_inode_extref *)ptr; | |
3707 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3708 | extref); | |
3709 | } | |
2f7e33d4 | 3710 | cache_acl: |
46a53cca CM |
3711 | /* |
3712 | * try to precache a NULL acl entry for files that don't have | |
3713 | * any xattrs or acls | |
3714 | */ | |
33345d01 | 3715 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3716 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3717 | if (first_xattr_slot != -1) { |
3718 | path->slots[0] = first_xattr_slot; | |
3719 | ret = btrfs_load_inode_props(inode, path); | |
3720 | if (ret) | |
0b246afa | 3721 | btrfs_err(fs_info, |
351fd353 | 3722 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3723 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3724 | root->root_key.objectid, ret); |
3725 | } | |
4222ea71 FM |
3726 | if (path != in_path) |
3727 | btrfs_free_path(path); | |
63541927 | 3728 | |
72c04902 AV |
3729 | if (!maybe_acls) |
3730 | cache_no_acl(inode); | |
46a53cca | 3731 | |
39279cc3 | 3732 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3733 | case S_IFREG: |
3734 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3735 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3736 | inode->i_fop = &btrfs_file_operations; |
3737 | inode->i_op = &btrfs_file_inode_operations; | |
3738 | break; | |
3739 | case S_IFDIR: | |
3740 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3741 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3742 | break; |
3743 | case S_IFLNK: | |
3744 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3745 | inode_nohighmem(inode); |
4779cc04 | 3746 | inode->i_mapping->a_ops = &btrfs_aops; |
39279cc3 | 3747 | break; |
618e21d5 | 3748 | default: |
0279b4cd | 3749 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3750 | init_special_inode(inode, inode->i_mode, rdev); |
3751 | break; | |
39279cc3 | 3752 | } |
6cbff00f | 3753 | |
7b6a221e | 3754 | btrfs_sync_inode_flags_to_i_flags(inode); |
67710892 | 3755 | return 0; |
39279cc3 CM |
3756 | } |
3757 | ||
d352ac68 CM |
3758 | /* |
3759 | * given a leaf and an inode, copy the inode fields into the leaf | |
3760 | */ | |
e02119d5 CM |
3761 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3762 | struct extent_buffer *leaf, | |
5f39d397 | 3763 | struct btrfs_inode_item *item, |
39279cc3 CM |
3764 | struct inode *inode) |
3765 | { | |
51fab693 LB |
3766 | struct btrfs_map_token token; |
3767 | ||
3768 | btrfs_init_map_token(&token); | |
5f39d397 | 3769 | |
51fab693 LB |
3770 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3771 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3772 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3773 | &token); | |
3774 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3775 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3776 | |
a937b979 | 3777 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3778 | inode->i_atime.tv_sec, &token); |
a937b979 | 3779 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3780 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3781 | |
a937b979 | 3782 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3783 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3784 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3785 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3786 | |
a937b979 | 3787 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3788 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3789 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3790 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3791 | |
9cc97d64 | 3792 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3793 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3794 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3795 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3796 | ||
51fab693 LB |
3797 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3798 | &token); | |
3799 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3800 | &token); | |
c7f88c4e JL |
3801 | btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode), |
3802 | &token); | |
51fab693 LB |
3803 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); |
3804 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3805 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3806 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3807 | } |
3808 | ||
d352ac68 CM |
3809 | /* |
3810 | * copy everything in the in-memory inode into the btree. | |
3811 | */ | |
2115133f | 3812 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3813 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3814 | { |
3815 | struct btrfs_inode_item *inode_item; | |
3816 | struct btrfs_path *path; | |
5f39d397 | 3817 | struct extent_buffer *leaf; |
39279cc3 CM |
3818 | int ret; |
3819 | ||
3820 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3821 | if (!path) |
3822 | return -ENOMEM; | |
3823 | ||
b9473439 | 3824 | path->leave_spinning = 1; |
16cdcec7 MX |
3825 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3826 | 1); | |
39279cc3 CM |
3827 | if (ret) { |
3828 | if (ret > 0) | |
3829 | ret = -ENOENT; | |
3830 | goto failed; | |
3831 | } | |
3832 | ||
5f39d397 CM |
3833 | leaf = path->nodes[0]; |
3834 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3835 | struct btrfs_inode_item); |
39279cc3 | 3836 | |
e02119d5 | 3837 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3838 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3839 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3840 | ret = 0; |
3841 | failed: | |
39279cc3 CM |
3842 | btrfs_free_path(path); |
3843 | return ret; | |
3844 | } | |
3845 | ||
2115133f CM |
3846 | /* |
3847 | * copy everything in the in-memory inode into the btree. | |
3848 | */ | |
3849 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3850 | struct btrfs_root *root, struct inode *inode) | |
3851 | { | |
0b246afa | 3852 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
3853 | int ret; |
3854 | ||
3855 | /* | |
3856 | * If the inode is a free space inode, we can deadlock during commit | |
3857 | * if we put it into the delayed code. | |
3858 | * | |
3859 | * The data relocation inode should also be directly updated | |
3860 | * without delay | |
3861 | */ | |
70ddc553 | 3862 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) |
1d52c78a | 3863 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 3864 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
3865 | btrfs_update_root_times(trans, root); |
3866 | ||
2115133f CM |
3867 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3868 | if (!ret) | |
3869 | btrfs_set_inode_last_trans(trans, inode); | |
3870 | return ret; | |
3871 | } | |
3872 | ||
3873 | return btrfs_update_inode_item(trans, root, inode); | |
3874 | } | |
3875 | ||
be6aef60 JB |
3876 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3877 | struct btrfs_root *root, | |
3878 | struct inode *inode) | |
2115133f CM |
3879 | { |
3880 | int ret; | |
3881 | ||
3882 | ret = btrfs_update_inode(trans, root, inode); | |
3883 | if (ret == -ENOSPC) | |
3884 | return btrfs_update_inode_item(trans, root, inode); | |
3885 | return ret; | |
3886 | } | |
3887 | ||
d352ac68 CM |
3888 | /* |
3889 | * unlink helper that gets used here in inode.c and in the tree logging | |
3890 | * recovery code. It remove a link in a directory with a given name, and | |
3891 | * also drops the back refs in the inode to the directory | |
3892 | */ | |
92986796 AV |
3893 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3894 | struct btrfs_root *root, | |
4ec5934e NB |
3895 | struct btrfs_inode *dir, |
3896 | struct btrfs_inode *inode, | |
92986796 | 3897 | const char *name, int name_len) |
39279cc3 | 3898 | { |
0b246afa | 3899 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 3900 | struct btrfs_path *path; |
39279cc3 | 3901 | int ret = 0; |
5f39d397 | 3902 | struct extent_buffer *leaf; |
39279cc3 | 3903 | struct btrfs_dir_item *di; |
5f39d397 | 3904 | struct btrfs_key key; |
aec7477b | 3905 | u64 index; |
33345d01 LZ |
3906 | u64 ino = btrfs_ino(inode); |
3907 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3908 | |
3909 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3910 | if (!path) { |
3911 | ret = -ENOMEM; | |
554233a6 | 3912 | goto out; |
54aa1f4d CM |
3913 | } |
3914 | ||
b9473439 | 3915 | path->leave_spinning = 1; |
33345d01 | 3916 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 | 3917 | name, name_len, -1); |
3cf5068f LB |
3918 | if (IS_ERR_OR_NULL(di)) { |
3919 | ret = di ? PTR_ERR(di) : -ENOENT; | |
39279cc3 CM |
3920 | goto err; |
3921 | } | |
5f39d397 CM |
3922 | leaf = path->nodes[0]; |
3923 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3924 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3925 | if (ret) |
3926 | goto err; | |
b3b4aa74 | 3927 | btrfs_release_path(path); |
39279cc3 | 3928 | |
67de1176 MX |
3929 | /* |
3930 | * If we don't have dir index, we have to get it by looking up | |
3931 | * the inode ref, since we get the inode ref, remove it directly, | |
3932 | * it is unnecessary to do delayed deletion. | |
3933 | * | |
3934 | * But if we have dir index, needn't search inode ref to get it. | |
3935 | * Since the inode ref is close to the inode item, it is better | |
3936 | * that we delay to delete it, and just do this deletion when | |
3937 | * we update the inode item. | |
3938 | */ | |
4ec5934e | 3939 | if (inode->dir_index) { |
67de1176 MX |
3940 | ret = btrfs_delayed_delete_inode_ref(inode); |
3941 | if (!ret) { | |
4ec5934e | 3942 | index = inode->dir_index; |
67de1176 MX |
3943 | goto skip_backref; |
3944 | } | |
3945 | } | |
3946 | ||
33345d01 LZ |
3947 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3948 | dir_ino, &index); | |
aec7477b | 3949 | if (ret) { |
0b246afa | 3950 | btrfs_info(fs_info, |
c2cf52eb | 3951 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 3952 | name_len, name, ino, dir_ino); |
66642832 | 3953 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
3954 | goto err; |
3955 | } | |
67de1176 | 3956 | skip_backref: |
9add2945 | 3957 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 3958 | if (ret) { |
66642832 | 3959 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 3960 | goto err; |
79787eaa | 3961 | } |
39279cc3 | 3962 | |
4ec5934e NB |
3963 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
3964 | dir_ino); | |
79787eaa | 3965 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 3966 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3967 | goto err; |
3968 | } | |
e02119d5 | 3969 | |
4ec5934e NB |
3970 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
3971 | index); | |
6418c961 CM |
3972 | if (ret == -ENOENT) |
3973 | ret = 0; | |
d4e3991b | 3974 | else if (ret) |
66642832 | 3975 | btrfs_abort_transaction(trans, ret); |
39279cc3 CM |
3976 | err: |
3977 | btrfs_free_path(path); | |
e02119d5 CM |
3978 | if (ret) |
3979 | goto out; | |
3980 | ||
6ef06d27 | 3981 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
3982 | inode_inc_iversion(&inode->vfs_inode); |
3983 | inode_inc_iversion(&dir->vfs_inode); | |
3984 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
3985 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
3986 | ret = btrfs_update_inode(trans, root, &dir->vfs_inode); | |
e02119d5 | 3987 | out: |
39279cc3 CM |
3988 | return ret; |
3989 | } | |
3990 | ||
92986796 AV |
3991 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3992 | struct btrfs_root *root, | |
4ec5934e | 3993 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
3994 | const char *name, int name_len) |
3995 | { | |
3996 | int ret; | |
3997 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
3998 | if (!ret) { | |
4ec5934e NB |
3999 | drop_nlink(&inode->vfs_inode); |
4000 | ret = btrfs_update_inode(trans, root, &inode->vfs_inode); | |
92986796 AV |
4001 | } |
4002 | return ret; | |
4003 | } | |
39279cc3 | 4004 | |
a22285a6 YZ |
4005 | /* |
4006 | * helper to start transaction for unlink and rmdir. | |
4007 | * | |
d52be818 JB |
4008 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4009 | * if we cannot make our reservations the normal way try and see if there is | |
4010 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4011 | * allow the unlink to occur. | |
a22285a6 | 4012 | */ |
d52be818 | 4013 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4014 | { |
a22285a6 | 4015 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4016 | |
e70bea5f JB |
4017 | /* |
4018 | * 1 for the possible orphan item | |
4019 | * 1 for the dir item | |
4020 | * 1 for the dir index | |
4021 | * 1 for the inode ref | |
e70bea5f JB |
4022 | * 1 for the inode |
4023 | */ | |
8eab77ff | 4024 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4025 | } |
4026 | ||
4027 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4028 | { | |
4029 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4030 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4031 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4032 | int ret; |
a22285a6 | 4033 | |
d52be818 | 4034 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4035 | if (IS_ERR(trans)) |
4036 | return PTR_ERR(trans); | |
5f39d397 | 4037 | |
4ec5934e NB |
4038 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4039 | 0); | |
12fcfd22 | 4040 | |
4ec5934e NB |
4041 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4042 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4043 | dentry->d_name.len); | |
b532402e TI |
4044 | if (ret) |
4045 | goto out; | |
7b128766 | 4046 | |
a22285a6 | 4047 | if (inode->i_nlink == 0) { |
73f2e545 | 4048 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
4049 | if (ret) |
4050 | goto out; | |
a22285a6 | 4051 | } |
7b128766 | 4052 | |
b532402e | 4053 | out: |
3a45bb20 | 4054 | btrfs_end_transaction(trans); |
2ff7e61e | 4055 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
4056 | return ret; |
4057 | } | |
4058 | ||
f60a2364 | 4059 | static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
401b3b19 LF |
4060 | struct inode *dir, u64 objectid, |
4061 | const char *name, int name_len) | |
4df27c4d | 4062 | { |
401b3b19 | 4063 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d YZ |
4064 | struct btrfs_path *path; |
4065 | struct extent_buffer *leaf; | |
4066 | struct btrfs_dir_item *di; | |
4067 | struct btrfs_key key; | |
4068 | u64 index; | |
4069 | int ret; | |
4a0cc7ca | 4070 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d YZ |
4071 | |
4072 | path = btrfs_alloc_path(); | |
4073 | if (!path) | |
4074 | return -ENOMEM; | |
4075 | ||
33345d01 | 4076 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4077 | name, name_len, -1); |
79787eaa | 4078 | if (IS_ERR_OR_NULL(di)) { |
3cf5068f | 4079 | ret = di ? PTR_ERR(di) : -ENOENT; |
79787eaa JM |
4080 | goto out; |
4081 | } | |
4df27c4d YZ |
4082 | |
4083 | leaf = path->nodes[0]; | |
4084 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4085 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4086 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4087 | if (ret) { |
66642832 | 4088 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4089 | goto out; |
4090 | } | |
b3b4aa74 | 4091 | btrfs_release_path(path); |
4df27c4d | 4092 | |
3ee1c553 LF |
4093 | ret = btrfs_del_root_ref(trans, objectid, root->root_key.objectid, |
4094 | dir_ino, &index, name, name_len); | |
4df27c4d | 4095 | if (ret < 0) { |
79787eaa | 4096 | if (ret != -ENOENT) { |
66642832 | 4097 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4098 | goto out; |
4099 | } | |
33345d01 | 4100 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4101 | name, name_len); |
79787eaa JM |
4102 | if (IS_ERR_OR_NULL(di)) { |
4103 | if (!di) | |
4104 | ret = -ENOENT; | |
4105 | else | |
4106 | ret = PTR_ERR(di); | |
66642832 | 4107 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4108 | goto out; |
4109 | } | |
4df27c4d YZ |
4110 | |
4111 | leaf = path->nodes[0]; | |
4112 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4df27c4d YZ |
4113 | index = key.offset; |
4114 | } | |
945d8962 | 4115 | btrfs_release_path(path); |
4df27c4d | 4116 | |
9add2945 | 4117 | ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index); |
79787eaa | 4118 | if (ret) { |
66642832 | 4119 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4120 | goto out; |
4121 | } | |
4df27c4d | 4122 | |
6ef06d27 | 4123 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 4124 | inode_inc_iversion(dir); |
c2050a45 | 4125 | dir->i_mtime = dir->i_ctime = current_time(dir); |
5a24e84c | 4126 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa | 4127 | if (ret) |
66642832 | 4128 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4129 | out: |
71d7aed0 | 4130 | btrfs_free_path(path); |
79787eaa | 4131 | return ret; |
4df27c4d YZ |
4132 | } |
4133 | ||
ec42f167 MT |
4134 | /* |
4135 | * Helper to check if the subvolume references other subvolumes or if it's | |
4136 | * default. | |
4137 | */ | |
f60a2364 | 4138 | static noinline int may_destroy_subvol(struct btrfs_root *root) |
ec42f167 MT |
4139 | { |
4140 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4141 | struct btrfs_path *path; | |
4142 | struct btrfs_dir_item *di; | |
4143 | struct btrfs_key key; | |
4144 | u64 dir_id; | |
4145 | int ret; | |
4146 | ||
4147 | path = btrfs_alloc_path(); | |
4148 | if (!path) | |
4149 | return -ENOMEM; | |
4150 | ||
4151 | /* Make sure this root isn't set as the default subvol */ | |
4152 | dir_id = btrfs_super_root_dir(fs_info->super_copy); | |
4153 | di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path, | |
4154 | dir_id, "default", 7, 0); | |
4155 | if (di && !IS_ERR(di)) { | |
4156 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
4157 | if (key.objectid == root->root_key.objectid) { | |
4158 | ret = -EPERM; | |
4159 | btrfs_err(fs_info, | |
4160 | "deleting default subvolume %llu is not allowed", | |
4161 | key.objectid); | |
4162 | goto out; | |
4163 | } | |
4164 | btrfs_release_path(path); | |
4165 | } | |
4166 | ||
4167 | key.objectid = root->root_key.objectid; | |
4168 | key.type = BTRFS_ROOT_REF_KEY; | |
4169 | key.offset = (u64)-1; | |
4170 | ||
4171 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
4172 | if (ret < 0) | |
4173 | goto out; | |
4174 | BUG_ON(ret == 0); | |
4175 | ||
4176 | ret = 0; | |
4177 | if (path->slots[0] > 0) { | |
4178 | path->slots[0]--; | |
4179 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
4180 | if (key.objectid == root->root_key.objectid && | |
4181 | key.type == BTRFS_ROOT_REF_KEY) | |
4182 | ret = -ENOTEMPTY; | |
4183 | } | |
4184 | out: | |
4185 | btrfs_free_path(path); | |
4186 | return ret; | |
4187 | } | |
4188 | ||
20a68004 NB |
4189 | /* Delete all dentries for inodes belonging to the root */ |
4190 | static void btrfs_prune_dentries(struct btrfs_root *root) | |
4191 | { | |
4192 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4193 | struct rb_node *node; | |
4194 | struct rb_node *prev; | |
4195 | struct btrfs_inode *entry; | |
4196 | struct inode *inode; | |
4197 | u64 objectid = 0; | |
4198 | ||
4199 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) | |
4200 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4201 | ||
4202 | spin_lock(&root->inode_lock); | |
4203 | again: | |
4204 | node = root->inode_tree.rb_node; | |
4205 | prev = NULL; | |
4206 | while (node) { | |
4207 | prev = node; | |
4208 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4209 | ||
37508515 | 4210 | if (objectid < btrfs_ino(entry)) |
20a68004 | 4211 | node = node->rb_left; |
37508515 | 4212 | else if (objectid > btrfs_ino(entry)) |
20a68004 NB |
4213 | node = node->rb_right; |
4214 | else | |
4215 | break; | |
4216 | } | |
4217 | if (!node) { | |
4218 | while (prev) { | |
4219 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
37508515 | 4220 | if (objectid <= btrfs_ino(entry)) { |
20a68004 NB |
4221 | node = prev; |
4222 | break; | |
4223 | } | |
4224 | prev = rb_next(prev); | |
4225 | } | |
4226 | } | |
4227 | while (node) { | |
4228 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
37508515 | 4229 | objectid = btrfs_ino(entry) + 1; |
20a68004 NB |
4230 | inode = igrab(&entry->vfs_inode); |
4231 | if (inode) { | |
4232 | spin_unlock(&root->inode_lock); | |
4233 | if (atomic_read(&inode->i_count) > 1) | |
4234 | d_prune_aliases(inode); | |
4235 | /* | |
4236 | * btrfs_drop_inode will have it removed from the inode | |
4237 | * cache when its usage count hits zero. | |
4238 | */ | |
4239 | iput(inode); | |
4240 | cond_resched(); | |
4241 | spin_lock(&root->inode_lock); | |
4242 | goto again; | |
4243 | } | |
4244 | ||
4245 | if (cond_resched_lock(&root->inode_lock)) | |
4246 | goto again; | |
4247 | ||
4248 | node = rb_next(node); | |
4249 | } | |
4250 | spin_unlock(&root->inode_lock); | |
4251 | } | |
4252 | ||
f60a2364 MT |
4253 | int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry) |
4254 | { | |
4255 | struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb); | |
4256 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4257 | struct inode *inode = d_inode(dentry); | |
4258 | struct btrfs_root *dest = BTRFS_I(inode)->root; | |
4259 | struct btrfs_trans_handle *trans; | |
4260 | struct btrfs_block_rsv block_rsv; | |
4261 | u64 root_flags; | |
f60a2364 MT |
4262 | int ret; |
4263 | int err; | |
4264 | ||
4265 | /* | |
4266 | * Don't allow to delete a subvolume with send in progress. This is | |
4267 | * inside the inode lock so the error handling that has to drop the bit | |
4268 | * again is not run concurrently. | |
4269 | */ | |
4270 | spin_lock(&dest->root_item_lock); | |
a7176f74 | 4271 | if (dest->send_in_progress) { |
f60a2364 MT |
4272 | spin_unlock(&dest->root_item_lock); |
4273 | btrfs_warn(fs_info, | |
4274 | "attempt to delete subvolume %llu during send", | |
4275 | dest->root_key.objectid); | |
4276 | return -EPERM; | |
4277 | } | |
a7176f74 LF |
4278 | root_flags = btrfs_root_flags(&dest->root_item); |
4279 | btrfs_set_root_flags(&dest->root_item, | |
4280 | root_flags | BTRFS_ROOT_SUBVOL_DEAD); | |
4281 | spin_unlock(&dest->root_item_lock); | |
f60a2364 MT |
4282 | |
4283 | down_write(&fs_info->subvol_sem); | |
4284 | ||
4285 | err = may_destroy_subvol(dest); | |
4286 | if (err) | |
4287 | goto out_up_write; | |
4288 | ||
4289 | btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); | |
4290 | /* | |
4291 | * One for dir inode, | |
4292 | * two for dir entries, | |
4293 | * two for root ref/backref. | |
4294 | */ | |
c4c129db | 4295 | err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true); |
f60a2364 MT |
4296 | if (err) |
4297 | goto out_up_write; | |
4298 | ||
4299 | trans = btrfs_start_transaction(root, 0); | |
4300 | if (IS_ERR(trans)) { | |
4301 | err = PTR_ERR(trans); | |
4302 | goto out_release; | |
4303 | } | |
4304 | trans->block_rsv = &block_rsv; | |
4305 | trans->bytes_reserved = block_rsv.size; | |
4306 | ||
4307 | btrfs_record_snapshot_destroy(trans, BTRFS_I(dir)); | |
4308 | ||
401b3b19 LF |
4309 | ret = btrfs_unlink_subvol(trans, dir, dest->root_key.objectid, |
4310 | dentry->d_name.name, dentry->d_name.len); | |
f60a2364 MT |
4311 | if (ret) { |
4312 | err = ret; | |
4313 | btrfs_abort_transaction(trans, ret); | |
4314 | goto out_end_trans; | |
4315 | } | |
4316 | ||
4317 | btrfs_record_root_in_trans(trans, dest); | |
4318 | ||
4319 | memset(&dest->root_item.drop_progress, 0, | |
4320 | sizeof(dest->root_item.drop_progress)); | |
4321 | dest->root_item.drop_level = 0; | |
4322 | btrfs_set_root_refs(&dest->root_item, 0); | |
4323 | ||
4324 | if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { | |
4325 | ret = btrfs_insert_orphan_item(trans, | |
4326 | fs_info->tree_root, | |
4327 | dest->root_key.objectid); | |
4328 | if (ret) { | |
4329 | btrfs_abort_transaction(trans, ret); | |
4330 | err = ret; | |
4331 | goto out_end_trans; | |
4332 | } | |
4333 | } | |
4334 | ||
d1957791 | 4335 | ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid, |
f60a2364 MT |
4336 | BTRFS_UUID_KEY_SUBVOL, |
4337 | dest->root_key.objectid); | |
4338 | if (ret && ret != -ENOENT) { | |
4339 | btrfs_abort_transaction(trans, ret); | |
4340 | err = ret; | |
4341 | goto out_end_trans; | |
4342 | } | |
4343 | if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { | |
d1957791 | 4344 | ret = btrfs_uuid_tree_remove(trans, |
f60a2364 MT |
4345 | dest->root_item.received_uuid, |
4346 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4347 | dest->root_key.objectid); | |
4348 | if (ret && ret != -ENOENT) { | |
4349 | btrfs_abort_transaction(trans, ret); | |
4350 | err = ret; | |
4351 | goto out_end_trans; | |
4352 | } | |
4353 | } | |
4354 | ||
4355 | out_end_trans: | |
4356 | trans->block_rsv = NULL; | |
4357 | trans->bytes_reserved = 0; | |
4358 | ret = btrfs_end_transaction(trans); | |
4359 | if (ret && !err) | |
4360 | err = ret; | |
4361 | inode->i_flags |= S_DEAD; | |
4362 | out_release: | |
4363 | btrfs_subvolume_release_metadata(fs_info, &block_rsv); | |
4364 | out_up_write: | |
4365 | up_write(&fs_info->subvol_sem); | |
4366 | if (err) { | |
4367 | spin_lock(&dest->root_item_lock); | |
4368 | root_flags = btrfs_root_flags(&dest->root_item); | |
4369 | btrfs_set_root_flags(&dest->root_item, | |
4370 | root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); | |
4371 | spin_unlock(&dest->root_item_lock); | |
4372 | } else { | |
4373 | d_invalidate(dentry); | |
20a68004 | 4374 | btrfs_prune_dentries(dest); |
f60a2364 MT |
4375 | ASSERT(dest->send_in_progress == 0); |
4376 | ||
4377 | /* the last ref */ | |
4378 | if (dest->ino_cache_inode) { | |
4379 | iput(dest->ino_cache_inode); | |
4380 | dest->ino_cache_inode = NULL; | |
4381 | } | |
4382 | } | |
4383 | ||
4384 | return err; | |
4385 | } | |
4386 | ||
39279cc3 CM |
4387 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4388 | { | |
2b0143b5 | 4389 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4390 | int err = 0; |
39279cc3 | 4391 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4392 | struct btrfs_trans_handle *trans; |
44f714da | 4393 | u64 last_unlink_trans; |
39279cc3 | 4394 | |
b3ae244e | 4395 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4396 | return -ENOTEMPTY; |
4a0cc7ca | 4397 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
a79a464d | 4398 | return btrfs_delete_subvolume(dir, dentry); |
134d4512 | 4399 | |
d52be818 | 4400 | trans = __unlink_start_trans(dir); |
a22285a6 | 4401 | if (IS_ERR(trans)) |
5df6a9f6 | 4402 | return PTR_ERR(trans); |
5df6a9f6 | 4403 | |
4a0cc7ca | 4404 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
401b3b19 | 4405 | err = btrfs_unlink_subvol(trans, dir, |
4df27c4d YZ |
4406 | BTRFS_I(inode)->location.objectid, |
4407 | dentry->d_name.name, | |
4408 | dentry->d_name.len); | |
4409 | goto out; | |
4410 | } | |
4411 | ||
73f2e545 | 4412 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4413 | if (err) |
4df27c4d | 4414 | goto out; |
7b128766 | 4415 | |
44f714da FM |
4416 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4417 | ||
39279cc3 | 4418 | /* now the directory is empty */ |
4ec5934e NB |
4419 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4420 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4421 | dentry->d_name.len); | |
44f714da | 4422 | if (!err) { |
6ef06d27 | 4423 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4424 | /* |
4425 | * Propagate the last_unlink_trans value of the deleted dir to | |
4426 | * its parent directory. This is to prevent an unrecoverable | |
4427 | * log tree in the case we do something like this: | |
4428 | * 1) create dir foo | |
4429 | * 2) create snapshot under dir foo | |
4430 | * 3) delete the snapshot | |
4431 | * 4) rmdir foo | |
4432 | * 5) mkdir foo | |
4433 | * 6) fsync foo or some file inside foo | |
4434 | */ | |
4435 | if (last_unlink_trans >= trans->transid) | |
4436 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4437 | } | |
4df27c4d | 4438 | out: |
3a45bb20 | 4439 | btrfs_end_transaction(trans); |
2ff7e61e | 4440 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4441 | |
39279cc3 CM |
4442 | return err; |
4443 | } | |
4444 | ||
28f75a0e CM |
4445 | static int truncate_space_check(struct btrfs_trans_handle *trans, |
4446 | struct btrfs_root *root, | |
4447 | u64 bytes_deleted) | |
4448 | { | |
0b246afa | 4449 | struct btrfs_fs_info *fs_info = root->fs_info; |
28f75a0e CM |
4450 | int ret; |
4451 | ||
dc95f7bf JB |
4452 | /* |
4453 | * This is only used to apply pressure to the enospc system, we don't | |
4454 | * intend to use this reservation at all. | |
4455 | */ | |
2ff7e61e | 4456 | bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted); |
0b246afa JM |
4457 | bytes_deleted *= fs_info->nodesize; |
4458 | ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv, | |
28f75a0e | 4459 | bytes_deleted, BTRFS_RESERVE_NO_FLUSH); |
dc95f7bf | 4460 | if (!ret) { |
0b246afa | 4461 | trace_btrfs_space_reservation(fs_info, "transaction", |
dc95f7bf JB |
4462 | trans->transid, |
4463 | bytes_deleted, 1); | |
28f75a0e | 4464 | trans->bytes_reserved += bytes_deleted; |
dc95f7bf | 4465 | } |
28f75a0e CM |
4466 | return ret; |
4467 | ||
4468 | } | |
4469 | ||
ddfae63c JB |
4470 | /* |
4471 | * Return this if we need to call truncate_block for the last bit of the | |
4472 | * truncate. | |
4473 | */ | |
4474 | #define NEED_TRUNCATE_BLOCK 1 | |
0305cd5f | 4475 | |
39279cc3 CM |
4476 | /* |
4477 | * this can truncate away extent items, csum items and directory items. | |
4478 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4479 | * any higher than new_size |
39279cc3 CM |
4480 | * |
4481 | * csum items that cross the new i_size are truncated to the new size | |
4482 | * as well. | |
7b128766 JB |
4483 | * |
4484 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4485 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4486 | */ |
8082510e YZ |
4487 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4488 | struct btrfs_root *root, | |
4489 | struct inode *inode, | |
4490 | u64 new_size, u32 min_type) | |
39279cc3 | 4491 | { |
0b246afa | 4492 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4493 | struct btrfs_path *path; |
5f39d397 | 4494 | struct extent_buffer *leaf; |
39279cc3 | 4495 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4496 | struct btrfs_key key; |
4497 | struct btrfs_key found_key; | |
39279cc3 | 4498 | u64 extent_start = 0; |
db94535d | 4499 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4500 | u64 extent_offset = 0; |
39279cc3 | 4501 | u64 item_end = 0; |
c1aa4575 | 4502 | u64 last_size = new_size; |
8082510e | 4503 | u32 found_type = (u8)-1; |
39279cc3 CM |
4504 | int found_extent; |
4505 | int del_item; | |
85e21bac CM |
4506 | int pending_del_nr = 0; |
4507 | int pending_del_slot = 0; | |
179e29e4 | 4508 | int extent_type = -1; |
8082510e | 4509 | int ret; |
4a0cc7ca | 4510 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
28ed1345 | 4511 | u64 bytes_deleted = 0; |
897ca819 TM |
4512 | bool be_nice = false; |
4513 | bool should_throttle = false; | |
4514 | bool should_end = false; | |
8082510e YZ |
4515 | |
4516 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4517 | |
28ed1345 CM |
4518 | /* |
4519 | * for non-free space inodes and ref cows, we want to back off from | |
4520 | * time to time | |
4521 | */ | |
70ddc553 | 4522 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) && |
28ed1345 | 4523 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
897ca819 | 4524 | be_nice = true; |
28ed1345 | 4525 | |
0eb0e19c MF |
4526 | path = btrfs_alloc_path(); |
4527 | if (!path) | |
4528 | return -ENOMEM; | |
e4058b54 | 4529 | path->reada = READA_BACK; |
0eb0e19c | 4530 | |
5dc562c5 JB |
4531 | /* |
4532 | * We want to drop from the next block forward in case this new size is | |
4533 | * not block aligned since we will be keeping the last block of the | |
4534 | * extent just the way it is. | |
4535 | */ | |
27cdeb70 | 4536 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 4537 | root == fs_info->tree_root) |
dcdbc059 | 4538 | btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size, |
0b246afa | 4539 | fs_info->sectorsize), |
da17066c | 4540 | (u64)-1, 0); |
8082510e | 4541 | |
16cdcec7 MX |
4542 | /* |
4543 | * This function is also used to drop the items in the log tree before | |
4544 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4545 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4546 | * items. | |
4547 | */ | |
4548 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4ccb5c72 | 4549 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); |
16cdcec7 | 4550 | |
33345d01 | 4551 | key.objectid = ino; |
39279cc3 | 4552 | key.offset = (u64)-1; |
5f39d397 CM |
4553 | key.type = (u8)-1; |
4554 | ||
85e21bac | 4555 | search_again: |
28ed1345 CM |
4556 | /* |
4557 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4558 | * up a huge file in a single leaf. Most of the time that | |
4559 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4560 | */ | |
fd86a3a3 OS |
4561 | if (be_nice && bytes_deleted > SZ_32M && |
4562 | btrfs_should_end_transaction(trans)) { | |
4563 | ret = -EAGAIN; | |
4564 | goto out; | |
28ed1345 CM |
4565 | } |
4566 | ||
b9473439 | 4567 | path->leave_spinning = 1; |
85e21bac | 4568 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
fd86a3a3 | 4569 | if (ret < 0) |
8082510e | 4570 | goto out; |
d397712b | 4571 | |
85e21bac | 4572 | if (ret > 0) { |
fd86a3a3 | 4573 | ret = 0; |
e02119d5 CM |
4574 | /* there are no items in the tree for us to truncate, we're |
4575 | * done | |
4576 | */ | |
8082510e YZ |
4577 | if (path->slots[0] == 0) |
4578 | goto out; | |
85e21bac CM |
4579 | path->slots[0]--; |
4580 | } | |
4581 | ||
d397712b | 4582 | while (1) { |
39279cc3 | 4583 | fi = NULL; |
5f39d397 CM |
4584 | leaf = path->nodes[0]; |
4585 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4586 | found_type = found_key.type; |
39279cc3 | 4587 | |
33345d01 | 4588 | if (found_key.objectid != ino) |
39279cc3 | 4589 | break; |
5f39d397 | 4590 | |
85e21bac | 4591 | if (found_type < min_type) |
39279cc3 CM |
4592 | break; |
4593 | ||
5f39d397 | 4594 | item_end = found_key.offset; |
39279cc3 | 4595 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4596 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4597 | struct btrfs_file_extent_item); |
179e29e4 CM |
4598 | extent_type = btrfs_file_extent_type(leaf, fi); |
4599 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4600 | item_end += |
db94535d | 4601 | btrfs_file_extent_num_bytes(leaf, fi); |
09ed2f16 LB |
4602 | |
4603 | trace_btrfs_truncate_show_fi_regular( | |
4604 | BTRFS_I(inode), leaf, fi, | |
4605 | found_key.offset); | |
179e29e4 | 4606 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
e41ca589 QW |
4607 | item_end += btrfs_file_extent_ram_bytes(leaf, |
4608 | fi); | |
09ed2f16 LB |
4609 | |
4610 | trace_btrfs_truncate_show_fi_inline( | |
4611 | BTRFS_I(inode), leaf, fi, path->slots[0], | |
4612 | found_key.offset); | |
39279cc3 | 4613 | } |
008630c1 | 4614 | item_end--; |
39279cc3 | 4615 | } |
8082510e YZ |
4616 | if (found_type > min_type) { |
4617 | del_item = 1; | |
4618 | } else { | |
76b42abb | 4619 | if (item_end < new_size) |
b888db2b | 4620 | break; |
8082510e YZ |
4621 | if (found_key.offset >= new_size) |
4622 | del_item = 1; | |
4623 | else | |
4624 | del_item = 0; | |
39279cc3 | 4625 | } |
39279cc3 | 4626 | found_extent = 0; |
39279cc3 | 4627 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4628 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4629 | goto delete; | |
4630 | ||
4631 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 4632 | u64 num_dec; |
db94535d | 4633 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4634 | if (!del_item) { |
db94535d CM |
4635 | u64 orig_num_bytes = |
4636 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4637 | extent_num_bytes = ALIGN(new_size - |
4638 | found_key.offset, | |
0b246afa | 4639 | fs_info->sectorsize); |
db94535d CM |
4640 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4641 | extent_num_bytes); | |
4642 | num_dec = (orig_num_bytes - | |
9069218d | 4643 | extent_num_bytes); |
27cdeb70 MX |
4644 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4645 | &root->state) && | |
4646 | extent_start != 0) | |
a76a3cd4 | 4647 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4648 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4649 | } else { |
db94535d CM |
4650 | extent_num_bytes = |
4651 | btrfs_file_extent_disk_num_bytes(leaf, | |
4652 | fi); | |
5d4f98a2 YZ |
4653 | extent_offset = found_key.offset - |
4654 | btrfs_file_extent_offset(leaf, fi); | |
4655 | ||
39279cc3 | 4656 | /* FIXME blocksize != 4096 */ |
9069218d | 4657 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4658 | if (extent_start != 0) { |
4659 | found_extent = 1; | |
27cdeb70 MX |
4660 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4661 | &root->state)) | |
a76a3cd4 | 4662 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4663 | } |
39279cc3 | 4664 | } |
9069218d | 4665 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4666 | /* |
4667 | * we can't truncate inline items that have had | |
4668 | * special encodings | |
4669 | */ | |
4670 | if (!del_item && | |
c8b97818 | 4671 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
ddfae63c JB |
4672 | btrfs_file_extent_other_encoding(leaf, fi) == 0 && |
4673 | btrfs_file_extent_compression(leaf, fi) == 0) { | |
4674 | u32 size = (u32)(new_size - found_key.offset); | |
4675 | ||
4676 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4677 | size = btrfs_file_extent_calc_inline_size(size); | |
4678 | btrfs_truncate_item(root->fs_info, path, size, 1); | |
4679 | } else if (!del_item) { | |
514ac8ad | 4680 | /* |
ddfae63c JB |
4681 | * We have to bail so the last_size is set to |
4682 | * just before this extent. | |
514ac8ad | 4683 | */ |
fd86a3a3 | 4684 | ret = NEED_TRUNCATE_BLOCK; |
ddfae63c JB |
4685 | break; |
4686 | } | |
0305cd5f | 4687 | |
ddfae63c | 4688 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
0305cd5f | 4689 | inode_sub_bytes(inode, item_end + 1 - new_size); |
39279cc3 | 4690 | } |
179e29e4 | 4691 | delete: |
ddfae63c JB |
4692 | if (del_item) |
4693 | last_size = found_key.offset; | |
4694 | else | |
4695 | last_size = new_size; | |
39279cc3 | 4696 | if (del_item) { |
85e21bac CM |
4697 | if (!pending_del_nr) { |
4698 | /* no pending yet, add ourselves */ | |
4699 | pending_del_slot = path->slots[0]; | |
4700 | pending_del_nr = 1; | |
4701 | } else if (pending_del_nr && | |
4702 | path->slots[0] + 1 == pending_del_slot) { | |
4703 | /* hop on the pending chunk */ | |
4704 | pending_del_nr++; | |
4705 | pending_del_slot = path->slots[0]; | |
4706 | } else { | |
d397712b | 4707 | BUG(); |
85e21bac | 4708 | } |
39279cc3 CM |
4709 | } else { |
4710 | break; | |
4711 | } | |
897ca819 | 4712 | should_throttle = false; |
28f75a0e | 4713 | |
27cdeb70 MX |
4714 | if (found_extent && |
4715 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
0b246afa | 4716 | root == fs_info->tree_root)) { |
b9473439 | 4717 | btrfs_set_path_blocking(path); |
28ed1345 | 4718 | bytes_deleted += extent_num_bytes; |
84f7d8e6 | 4719 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4720 | extent_num_bytes, 0, |
4721 | btrfs_header_owner(leaf), | |
b06c4bf5 | 4722 | ino, extent_offset); |
05522109 OS |
4723 | if (ret) { |
4724 | btrfs_abort_transaction(trans, ret); | |
4725 | break; | |
4726 | } | |
7c861627 | 4727 | if (btrfs_should_throttle_delayed_refs(trans)) |
2ff7e61e | 4728 | btrfs_async_run_delayed_refs(fs_info, |
dd4b857a WX |
4729 | trans->delayed_ref_updates * 2, |
4730 | trans->transid, 0); | |
28f75a0e CM |
4731 | if (be_nice) { |
4732 | if (truncate_space_check(trans, root, | |
4733 | extent_num_bytes)) { | |
897ca819 | 4734 | should_end = true; |
28f75a0e | 4735 | } |
7c861627 | 4736 | if (btrfs_should_throttle_delayed_refs(trans)) |
897ca819 | 4737 | should_throttle = true; |
28f75a0e | 4738 | } |
39279cc3 | 4739 | } |
85e21bac | 4740 | |
8082510e YZ |
4741 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4742 | break; | |
4743 | ||
4744 | if (path->slots[0] == 0 || | |
1262133b | 4745 | path->slots[0] != pending_del_slot || |
28f75a0e | 4746 | should_throttle || should_end) { |
8082510e YZ |
4747 | if (pending_del_nr) { |
4748 | ret = btrfs_del_items(trans, root, path, | |
4749 | pending_del_slot, | |
4750 | pending_del_nr); | |
79787eaa | 4751 | if (ret) { |
66642832 | 4752 | btrfs_abort_transaction(trans, ret); |
fd86a3a3 | 4753 | break; |
79787eaa | 4754 | } |
8082510e YZ |
4755 | pending_del_nr = 0; |
4756 | } | |
b3b4aa74 | 4757 | btrfs_release_path(path); |
28f75a0e | 4758 | if (should_throttle) { |
1262133b JB |
4759 | unsigned long updates = trans->delayed_ref_updates; |
4760 | if (updates) { | |
4761 | trans->delayed_ref_updates = 0; | |
2ff7e61e | 4762 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 4763 | updates * 2); |
fd86a3a3 OS |
4764 | if (ret) |
4765 | break; | |
1262133b JB |
4766 | } |
4767 | } | |
28f75a0e CM |
4768 | /* |
4769 | * if we failed to refill our space rsv, bail out | |
4770 | * and let the transaction restart | |
4771 | */ | |
4772 | if (should_end) { | |
fd86a3a3 OS |
4773 | ret = -EAGAIN; |
4774 | break; | |
28f75a0e | 4775 | } |
85e21bac | 4776 | goto search_again; |
8082510e YZ |
4777 | } else { |
4778 | path->slots[0]--; | |
85e21bac | 4779 | } |
39279cc3 | 4780 | } |
8082510e | 4781 | out: |
fd86a3a3 OS |
4782 | if (ret >= 0 && pending_del_nr) { |
4783 | int err; | |
4784 | ||
4785 | err = btrfs_del_items(trans, root, path, pending_del_slot, | |
85e21bac | 4786 | pending_del_nr); |
fd86a3a3 OS |
4787 | if (err) { |
4788 | btrfs_abort_transaction(trans, err); | |
4789 | ret = err; | |
4790 | } | |
85e21bac | 4791 | } |
76b42abb FM |
4792 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
4793 | ASSERT(last_size >= new_size); | |
fd86a3a3 | 4794 | if (!ret && last_size > new_size) |
76b42abb | 4795 | last_size = new_size; |
7f4f6e0a | 4796 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
76b42abb | 4797 | } |
28ed1345 | 4798 | |
39279cc3 | 4799 | btrfs_free_path(path); |
28ed1345 | 4800 | |
fd86a3a3 | 4801 | if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) { |
28ed1345 | 4802 | unsigned long updates = trans->delayed_ref_updates; |
fd86a3a3 OS |
4803 | int err; |
4804 | ||
28ed1345 CM |
4805 | if (updates) { |
4806 | trans->delayed_ref_updates = 0; | |
fd86a3a3 OS |
4807 | err = btrfs_run_delayed_refs(trans, updates * 2); |
4808 | if (err) | |
4809 | ret = err; | |
28ed1345 CM |
4810 | } |
4811 | } | |
fd86a3a3 | 4812 | return ret; |
39279cc3 CM |
4813 | } |
4814 | ||
4815 | /* | |
9703fefe | 4816 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4817 | * @inode - inode that we're zeroing |
4818 | * @from - the offset to start zeroing | |
4819 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4820 | * offset | |
4821 | * @front - zero up to the offset instead of from the offset on | |
4822 | * | |
9703fefe | 4823 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4824 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4825 | */ |
9703fefe | 4826 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4827 | int front) |
39279cc3 | 4828 | { |
0b246afa | 4829 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 | 4830 | struct address_space *mapping = inode->i_mapping; |
e6dcd2dc CM |
4831 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4832 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4833 | struct extent_state *cached_state = NULL; |
364ecf36 | 4834 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 4835 | char *kaddr; |
0b246afa | 4836 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4837 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4838 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4839 | struct page *page; |
3b16a4e3 | 4840 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4841 | int ret = 0; |
9703fefe CR |
4842 | u64 block_start; |
4843 | u64 block_end; | |
39279cc3 | 4844 | |
b03ebd99 NB |
4845 | if (IS_ALIGNED(offset, blocksize) && |
4846 | (!len || IS_ALIGNED(len, blocksize))) | |
39279cc3 | 4847 | goto out; |
9703fefe | 4848 | |
8b62f87b JB |
4849 | block_start = round_down(from, blocksize); |
4850 | block_end = block_start + blocksize - 1; | |
4851 | ||
364ecf36 | 4852 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8b62f87b | 4853 | block_start, blocksize); |
5d5e103a JB |
4854 | if (ret) |
4855 | goto out; | |
39279cc3 | 4856 | |
211c17f5 | 4857 | again: |
3b16a4e3 | 4858 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4859 | if (!page) { |
bc42bda2 | 4860 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
4861 | block_start, blocksize, true); |
4862 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true); | |
ac6a2b36 | 4863 | ret = -ENOMEM; |
39279cc3 | 4864 | goto out; |
5d5e103a | 4865 | } |
e6dcd2dc | 4866 | |
39279cc3 | 4867 | if (!PageUptodate(page)) { |
9ebefb18 | 4868 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4869 | lock_page(page); |
211c17f5 CM |
4870 | if (page->mapping != mapping) { |
4871 | unlock_page(page); | |
09cbfeaf | 4872 | put_page(page); |
211c17f5 CM |
4873 | goto again; |
4874 | } | |
39279cc3 CM |
4875 | if (!PageUptodate(page)) { |
4876 | ret = -EIO; | |
89642229 | 4877 | goto out_unlock; |
39279cc3 CM |
4878 | } |
4879 | } | |
211c17f5 | 4880 | wait_on_page_writeback(page); |
e6dcd2dc | 4881 | |
9703fefe | 4882 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4883 | set_page_extent_mapped(page); |
4884 | ||
9703fefe | 4885 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4886 | if (ordered) { |
9703fefe | 4887 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4888 | &cached_state); |
e6dcd2dc | 4889 | unlock_page(page); |
09cbfeaf | 4890 | put_page(page); |
eb84ae03 | 4891 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4892 | btrfs_put_ordered_extent(ordered); |
4893 | goto again; | |
4894 | } | |
4895 | ||
9703fefe | 4896 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
4897 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4898 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 4899 | 0, 0, &cached_state); |
5d5e103a | 4900 | |
e3b8a485 | 4901 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
ba8b04c1 | 4902 | &cached_state, 0); |
9ed74f2d | 4903 | if (ret) { |
9703fefe | 4904 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4905 | &cached_state); |
9ed74f2d JB |
4906 | goto out_unlock; |
4907 | } | |
4908 | ||
9703fefe | 4909 | if (offset != blocksize) { |
2aaa6655 | 4910 | if (!len) |
9703fefe | 4911 | len = blocksize - offset; |
e6dcd2dc | 4912 | kaddr = kmap(page); |
2aaa6655 | 4913 | if (front) |
9703fefe CR |
4914 | memset(kaddr + (block_start - page_offset(page)), |
4915 | 0, offset); | |
2aaa6655 | 4916 | else |
9703fefe CR |
4917 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4918 | 0, len); | |
e6dcd2dc CM |
4919 | flush_dcache_page(page); |
4920 | kunmap(page); | |
4921 | } | |
247e743c | 4922 | ClearPageChecked(page); |
e6dcd2dc | 4923 | set_page_dirty(page); |
e43bbe5e | 4924 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state); |
39279cc3 | 4925 | |
89642229 | 4926 | out_unlock: |
5d5e103a | 4927 | if (ret) |
bc42bda2 | 4928 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
43b18595 QW |
4929 | blocksize, true); |
4930 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0)); | |
39279cc3 | 4931 | unlock_page(page); |
09cbfeaf | 4932 | put_page(page); |
39279cc3 | 4933 | out: |
364ecf36 | 4934 | extent_changeset_free(data_reserved); |
39279cc3 CM |
4935 | return ret; |
4936 | } | |
4937 | ||
16e7549f JB |
4938 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4939 | u64 offset, u64 len) | |
4940 | { | |
0b246afa | 4941 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
16e7549f JB |
4942 | struct btrfs_trans_handle *trans; |
4943 | int ret; | |
4944 | ||
4945 | /* | |
4946 | * Still need to make sure the inode looks like it's been updated so | |
4947 | * that any holes get logged if we fsync. | |
4948 | */ | |
0b246afa JM |
4949 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) { |
4950 | BTRFS_I(inode)->last_trans = fs_info->generation; | |
16e7549f JB |
4951 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
4952 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4953 | return 0; | |
4954 | } | |
4955 | ||
4956 | /* | |
4957 | * 1 - for the one we're dropping | |
4958 | * 1 - for the one we're adding | |
4959 | * 1 - for updating the inode. | |
4960 | */ | |
4961 | trans = btrfs_start_transaction(root, 3); | |
4962 | if (IS_ERR(trans)) | |
4963 | return PTR_ERR(trans); | |
4964 | ||
4965 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4966 | if (ret) { | |
66642832 | 4967 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4968 | btrfs_end_transaction(trans); |
16e7549f JB |
4969 | return ret; |
4970 | } | |
4971 | ||
f85b7379 DS |
4972 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)), |
4973 | offset, 0, 0, len, 0, len, 0, 0, 0); | |
16e7549f | 4974 | if (ret) |
66642832 | 4975 | btrfs_abort_transaction(trans, ret); |
16e7549f JB |
4976 | else |
4977 | btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 4978 | btrfs_end_transaction(trans); |
16e7549f JB |
4979 | return ret; |
4980 | } | |
4981 | ||
695a0d0d JB |
4982 | /* |
4983 | * This function puts in dummy file extents for the area we're creating a hole | |
4984 | * for. So if we are truncating this file to a larger size we need to insert | |
4985 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4986 | * the range between oldsize and size | |
4987 | */ | |
a41ad394 | 4988 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4989 | { |
0b246afa | 4990 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9036c102 YZ |
4991 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4992 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4993 | struct extent_map *em = NULL; |
2ac55d41 | 4994 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4995 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
0b246afa JM |
4996 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
4997 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
4998 | u64 last_byte; |
4999 | u64 cur_offset; | |
5000 | u64 hole_size; | |
9ed74f2d | 5001 | int err = 0; |
39279cc3 | 5002 | |
a71754fc | 5003 | /* |
9703fefe CR |
5004 | * If our size started in the middle of a block we need to zero out the |
5005 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
5006 | * expose stale data. |
5007 | */ | |
9703fefe | 5008 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
5009 | if (err) |
5010 | return err; | |
5011 | ||
9036c102 YZ |
5012 | if (size <= hole_start) |
5013 | return 0; | |
5014 | ||
9036c102 YZ |
5015 | while (1) { |
5016 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 5017 | |
ff13db41 | 5018 | lock_extent_bits(io_tree, hole_start, block_end - 1, |
d0082371 | 5019 | &cached_state); |
a776c6fa | 5020 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start, |
fa7c1494 | 5021 | block_end - hole_start); |
9036c102 YZ |
5022 | if (!ordered) |
5023 | break; | |
2ac55d41 | 5024 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
e43bbe5e | 5025 | &cached_state); |
fa7c1494 | 5026 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
5027 | btrfs_put_ordered_extent(ordered); |
5028 | } | |
39279cc3 | 5029 | |
9036c102 YZ |
5030 | cur_offset = hole_start; |
5031 | while (1) { | |
fc4f21b1 | 5032 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
9036c102 | 5033 | block_end - cur_offset, 0); |
79787eaa JM |
5034 | if (IS_ERR(em)) { |
5035 | err = PTR_ERR(em); | |
f2767956 | 5036 | em = NULL; |
79787eaa JM |
5037 | break; |
5038 | } | |
9036c102 | 5039 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 5040 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
8082510e | 5041 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 5042 | struct extent_map *hole_em; |
9036c102 | 5043 | hole_size = last_byte - cur_offset; |
9ed74f2d | 5044 | |
16e7549f JB |
5045 | err = maybe_insert_hole(root, inode, cur_offset, |
5046 | hole_size); | |
5047 | if (err) | |
3893e33b | 5048 | break; |
dcdbc059 | 5049 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
5050 | cur_offset + hole_size - 1, 0); |
5051 | hole_em = alloc_extent_map(); | |
5052 | if (!hole_em) { | |
5053 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
5054 | &BTRFS_I(inode)->runtime_flags); | |
5055 | goto next; | |
5056 | } | |
5057 | hole_em->start = cur_offset; | |
5058 | hole_em->len = hole_size; | |
5059 | hole_em->orig_start = cur_offset; | |
8082510e | 5060 | |
5dc562c5 JB |
5061 | hole_em->block_start = EXTENT_MAP_HOLE; |
5062 | hole_em->block_len = 0; | |
b4939680 | 5063 | hole_em->orig_block_len = 0; |
cc95bef6 | 5064 | hole_em->ram_bytes = hole_size; |
0b246afa | 5065 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 | 5066 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 5067 | hole_em->generation = fs_info->generation; |
8082510e | 5068 | |
5dc562c5 JB |
5069 | while (1) { |
5070 | write_lock(&em_tree->lock); | |
09a2a8f9 | 5071 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
5072 | write_unlock(&em_tree->lock); |
5073 | if (err != -EEXIST) | |
5074 | break; | |
dcdbc059 NB |
5075 | btrfs_drop_extent_cache(BTRFS_I(inode), |
5076 | cur_offset, | |
5dc562c5 JB |
5077 | cur_offset + |
5078 | hole_size - 1, 0); | |
5079 | } | |
5080 | free_extent_map(hole_em); | |
9036c102 | 5081 | } |
16e7549f | 5082 | next: |
9036c102 | 5083 | free_extent_map(em); |
a22285a6 | 5084 | em = NULL; |
9036c102 | 5085 | cur_offset = last_byte; |
8082510e | 5086 | if (cur_offset >= block_end) |
9036c102 YZ |
5087 | break; |
5088 | } | |
a22285a6 | 5089 | free_extent_map(em); |
e43bbe5e | 5090 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state); |
9036c102 YZ |
5091 | return err; |
5092 | } | |
39279cc3 | 5093 | |
3972f260 | 5094 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 5095 | { |
f4a2f4c5 MX |
5096 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5097 | struct btrfs_trans_handle *trans; | |
a41ad394 | 5098 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
5099 | loff_t newsize = attr->ia_size; |
5100 | int mask = attr->ia_valid; | |
8082510e YZ |
5101 | int ret; |
5102 | ||
3972f260 ES |
5103 | /* |
5104 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
5105 | * special case where we need to update the times despite not having | |
5106 | * these flags set. For all other operations the VFS set these flags | |
5107 | * explicitly if it wants a timestamp update. | |
5108 | */ | |
dff6efc3 CH |
5109 | if (newsize != oldsize) { |
5110 | inode_inc_iversion(inode); | |
5111 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
5112 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 5113 | current_time(inode); |
dff6efc3 | 5114 | } |
3972f260 | 5115 | |
a41ad394 | 5116 | if (newsize > oldsize) { |
9ea24bbe | 5117 | /* |
ea14b57f | 5118 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
5119 | * This is to ensure the snapshot captures a fully consistent |
5120 | * state of this file - if the snapshot captures this expanding | |
5121 | * truncation, it must capture all writes that happened before | |
5122 | * this truncation. | |
5123 | */ | |
0bc19f90 | 5124 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 5125 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe | 5126 | if (ret) { |
ea14b57f | 5127 | btrfs_end_write_no_snapshotting(root); |
8082510e | 5128 | return ret; |
9ea24bbe | 5129 | } |
8082510e | 5130 | |
f4a2f4c5 | 5131 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 5132 | if (IS_ERR(trans)) { |
ea14b57f | 5133 | btrfs_end_write_no_snapshotting(root); |
f4a2f4c5 | 5134 | return PTR_ERR(trans); |
9ea24bbe | 5135 | } |
f4a2f4c5 MX |
5136 | |
5137 | i_size_write(inode, newsize); | |
5138 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 5139 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 5140 | ret = btrfs_update_inode(trans, root, inode); |
ea14b57f | 5141 | btrfs_end_write_no_snapshotting(root); |
3a45bb20 | 5142 | btrfs_end_transaction(trans); |
a41ad394 | 5143 | } else { |
8082510e | 5144 | |
a41ad394 JB |
5145 | /* |
5146 | * We're truncating a file that used to have good data down to | |
5147 | * zero. Make sure it gets into the ordered flush list so that | |
5148 | * any new writes get down to disk quickly. | |
5149 | */ | |
5150 | if (newsize == 0) | |
72ac3c0d JB |
5151 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
5152 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 5153 | |
a41ad394 | 5154 | truncate_setsize(inode, newsize); |
2e60a51e MX |
5155 | |
5156 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
abcefb1e | 5157 | btrfs_inode_block_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5158 | inode_dio_wait(inode); |
0b581701 | 5159 | btrfs_inode_resume_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5160 | |
213e8c55 | 5161 | ret = btrfs_truncate(inode, newsize == oldsize); |
7f4f6e0a JB |
5162 | if (ret && inode->i_nlink) { |
5163 | int err; | |
5164 | ||
5165 | /* | |
f7e9e8fc OS |
5166 | * Truncate failed, so fix up the in-memory size. We |
5167 | * adjusted disk_i_size down as we removed extents, so | |
5168 | * wait for disk_i_size to be stable and then update the | |
5169 | * in-memory size to match. | |
7f4f6e0a | 5170 | */ |
f7e9e8fc | 5171 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
7f4f6e0a | 5172 | if (err) |
f7e9e8fc OS |
5173 | return err; |
5174 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
7f4f6e0a | 5175 | } |
8082510e YZ |
5176 | } |
5177 | ||
a41ad394 | 5178 | return ret; |
8082510e YZ |
5179 | } |
5180 | ||
9036c102 YZ |
5181 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5182 | { | |
2b0143b5 | 5183 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5184 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5185 | int err; |
39279cc3 | 5186 | |
b83cc969 LZ |
5187 | if (btrfs_root_readonly(root)) |
5188 | return -EROFS; | |
5189 | ||
31051c85 | 5190 | err = setattr_prepare(dentry, attr); |
9036c102 YZ |
5191 | if (err) |
5192 | return err; | |
2bf5a725 | 5193 | |
5a3f23d5 | 5194 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5195 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5196 | if (err) |
5197 | return err; | |
39279cc3 | 5198 | } |
9036c102 | 5199 | |
1025774c CH |
5200 | if (attr->ia_valid) { |
5201 | setattr_copy(inode, attr); | |
0c4d2d95 | 5202 | inode_inc_iversion(inode); |
22c44fe6 | 5203 | err = btrfs_dirty_inode(inode); |
1025774c | 5204 | |
22c44fe6 | 5205 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5206 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5207 | } |
33268eaf | 5208 | |
39279cc3 CM |
5209 | return err; |
5210 | } | |
61295eb8 | 5211 | |
131e404a FDBM |
5212 | /* |
5213 | * While truncating the inode pages during eviction, we get the VFS calling | |
5214 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5215 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5216 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5217 | * extent_state structures over and over, wasting lots of time. | |
5218 | * | |
5219 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5220 | * those expensive operations on a per page basis and do only the ordered io | |
5221 | * finishing, while we release here the extent_map and extent_state structures, | |
5222 | * without the excessive merging and splitting. | |
5223 | */ | |
5224 | static void evict_inode_truncate_pages(struct inode *inode) | |
5225 | { | |
5226 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5227 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5228 | struct rb_node *node; | |
5229 | ||
5230 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5231 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5232 | |
5233 | write_lock(&map_tree->lock); | |
07e1ce09 | 5234 | while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) { |
131e404a FDBM |
5235 | struct extent_map *em; |
5236 | ||
07e1ce09 | 5237 | node = rb_first_cached(&map_tree->map); |
131e404a | 5238 | em = rb_entry(node, struct extent_map, rb_node); |
180589ef WS |
5239 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5240 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5241 | remove_extent_mapping(map_tree, em); |
5242 | free_extent_map(em); | |
7064dd5c FM |
5243 | if (need_resched()) { |
5244 | write_unlock(&map_tree->lock); | |
5245 | cond_resched(); | |
5246 | write_lock(&map_tree->lock); | |
5247 | } | |
131e404a FDBM |
5248 | } |
5249 | write_unlock(&map_tree->lock); | |
5250 | ||
6ca07097 FM |
5251 | /* |
5252 | * Keep looping until we have no more ranges in the io tree. | |
5253 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5254 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5255 | * still in progress (unlocked the pages in the bio but did not yet | |
5256 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5257 | * ranges can still be locked and eviction started because before |
5258 | * submitting those bios, which are executed by a separate task (work | |
5259 | * queue kthread), inode references (inode->i_count) were not taken | |
5260 | * (which would be dropped in the end io callback of each bio). | |
5261 | * Therefore here we effectively end up waiting for those bios and | |
5262 | * anyone else holding locked ranges without having bumped the inode's | |
5263 | * reference count - if we don't do it, when they access the inode's | |
5264 | * io_tree to unlock a range it may be too late, leading to an | |
5265 | * use-after-free issue. | |
5266 | */ | |
131e404a FDBM |
5267 | spin_lock(&io_tree->lock); |
5268 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5269 | struct extent_state *state; | |
5270 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5271 | u64 start; |
5272 | u64 end; | |
421f0922 | 5273 | unsigned state_flags; |
131e404a FDBM |
5274 | |
5275 | node = rb_first(&io_tree->state); | |
5276 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5277 | start = state->start; |
5278 | end = state->end; | |
421f0922 | 5279 | state_flags = state->state; |
131e404a FDBM |
5280 | spin_unlock(&io_tree->lock); |
5281 | ||
ff13db41 | 5282 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5283 | |
5284 | /* | |
5285 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5286 | * and its reserved space won't be freed by delayed_ref. | |
5287 | * So we need to free its reserved space here. | |
5288 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5289 | * | |
5290 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5291 | */ | |
421f0922 | 5292 | if (state_flags & EXTENT_DELALLOC) |
bc42bda2 | 5293 | btrfs_qgroup_free_data(inode, NULL, start, end - start + 1); |
b9d0b389 | 5294 | |
6ca07097 | 5295 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5296 | EXTENT_LOCKED | EXTENT_DIRTY | |
5297 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
ae0f1625 | 5298 | EXTENT_DEFRAG, 1, 1, &cached_state); |
131e404a | 5299 | |
7064dd5c | 5300 | cond_resched(); |
131e404a FDBM |
5301 | spin_lock(&io_tree->lock); |
5302 | } | |
5303 | spin_unlock(&io_tree->lock); | |
5304 | } | |
5305 | ||
4b9d7b59 | 5306 | static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root, |
ad80cf50 | 5307 | struct btrfs_block_rsv *rsv) |
4b9d7b59 OS |
5308 | { |
5309 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5310 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5311 | int failures = 0; | |
5312 | ||
5313 | for (;;) { | |
5314 | struct btrfs_trans_handle *trans; | |
5315 | int ret; | |
5316 | ||
ad80cf50 | 5317 | ret = btrfs_block_rsv_refill(root, rsv, rsv->size, |
4b9d7b59 OS |
5318 | BTRFS_RESERVE_FLUSH_LIMIT); |
5319 | ||
5320 | if (ret && ++failures > 2) { | |
5321 | btrfs_warn(fs_info, | |
5322 | "could not allocate space for a delete; will truncate on mount"); | |
5323 | return ERR_PTR(-ENOSPC); | |
5324 | } | |
5325 | ||
5326 | trans = btrfs_join_transaction(root); | |
5327 | if (IS_ERR(trans) || !ret) | |
5328 | return trans; | |
5329 | ||
5330 | /* | |
5331 | * Try to steal from the global reserve if there is space for | |
5332 | * it. | |
5333 | */ | |
64403612 JB |
5334 | if (!btrfs_check_space_for_delayed_refs(fs_info) && |
5335 | !btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) | |
4b9d7b59 OS |
5336 | return trans; |
5337 | ||
5338 | /* If not, commit and try again. */ | |
5339 | ret = btrfs_commit_transaction(trans); | |
5340 | if (ret) | |
5341 | return ERR_PTR(ret); | |
5342 | } | |
5343 | } | |
5344 | ||
bd555975 | 5345 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5346 | { |
0b246afa | 5347 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5348 | struct btrfs_trans_handle *trans; |
5349 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b9d7b59 | 5350 | struct btrfs_block_rsv *rsv; |
39279cc3 CM |
5351 | int ret; |
5352 | ||
1abe9b8a | 5353 | trace_btrfs_inode_evict(inode); |
5354 | ||
3d48d981 | 5355 | if (!root) { |
e8f1bc14 | 5356 | clear_inode(inode); |
3d48d981 NB |
5357 | return; |
5358 | } | |
5359 | ||
131e404a FDBM |
5360 | evict_inode_truncate_pages(inode); |
5361 | ||
69e9c6c6 SB |
5362 | if (inode->i_nlink && |
5363 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5364 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5365 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5366 | goto no_delete; |
5367 | ||
27919067 | 5368 | if (is_bad_inode(inode)) |
39279cc3 | 5369 | goto no_delete; |
5f39d397 | 5370 | |
7ab7956e | 5371 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5372 | |
7b40b695 | 5373 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
c71bf099 | 5374 | goto no_delete; |
c71bf099 | 5375 | |
76dda93c | 5376 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5377 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5378 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5379 | goto no_delete; |
5380 | } | |
5381 | ||
aa79021f | 5382 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
27919067 | 5383 | if (ret) |
0e8c36a9 | 5384 | goto no_delete; |
0e8c36a9 | 5385 | |
2ff7e61e | 5386 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
27919067 | 5387 | if (!rsv) |
4289a667 | 5388 | goto no_delete; |
ad80cf50 | 5389 | rsv->size = btrfs_calc_trunc_metadata_size(fs_info, 1); |
ca7e70f5 | 5390 | rsv->failfast = 1; |
4289a667 | 5391 | |
6ef06d27 | 5392 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5393 | |
8082510e | 5394 | while (1) { |
ad80cf50 | 5395 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5396 | if (IS_ERR(trans)) |
5397 | goto free_rsv; | |
7b128766 | 5398 | |
4289a667 JB |
5399 | trans->block_rsv = rsv; |
5400 | ||
d68fc57b | 5401 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
27919067 OS |
5402 | trans->block_rsv = &fs_info->trans_block_rsv; |
5403 | btrfs_end_transaction(trans); | |
5404 | btrfs_btree_balance_dirty(fs_info); | |
5405 | if (ret && ret != -ENOSPC && ret != -EAGAIN) | |
5406 | goto free_rsv; | |
5407 | else if (!ret) | |
8082510e | 5408 | break; |
8082510e | 5409 | } |
5f39d397 | 5410 | |
4ef31a45 | 5411 | /* |
27919067 OS |
5412 | * Errors here aren't a big deal, it just means we leave orphan items in |
5413 | * the tree. They will be cleaned up on the next mount. If the inode | |
5414 | * number gets reused, cleanup deletes the orphan item without doing | |
5415 | * anything, and unlink reuses the existing orphan item. | |
5416 | * | |
5417 | * If it turns out that we are dropping too many of these, we might want | |
5418 | * to add a mechanism for retrying these after a commit. | |
4ef31a45 | 5419 | */ |
ad80cf50 | 5420 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5421 | if (!IS_ERR(trans)) { |
5422 | trans->block_rsv = rsv; | |
5423 | btrfs_orphan_del(trans, BTRFS_I(inode)); | |
5424 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5425 | btrfs_end_transaction(trans); | |
5426 | } | |
54aa1f4d | 5427 | |
0b246afa | 5428 | if (!(root == fs_info->tree_root || |
581bb050 | 5429 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) |
4a0cc7ca | 5430 | btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode))); |
581bb050 | 5431 | |
27919067 OS |
5432 | free_rsv: |
5433 | btrfs_free_block_rsv(fs_info, rsv); | |
39279cc3 | 5434 | no_delete: |
27919067 OS |
5435 | /* |
5436 | * If we didn't successfully delete, the orphan item will still be in | |
5437 | * the tree and we'll retry on the next mount. Again, we might also want | |
5438 | * to retry these periodically in the future. | |
5439 | */ | |
f48d1cf5 | 5440 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5441 | clear_inode(inode); |
39279cc3 CM |
5442 | } |
5443 | ||
5444 | /* | |
5445 | * this returns the key found in the dir entry in the location pointer. | |
005d6712 SY |
5446 | * If no dir entries were found, returns -ENOENT. |
5447 | * If found a corrupted location in dir entry, returns -EUCLEAN. | |
39279cc3 CM |
5448 | */ |
5449 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5450 | struct btrfs_key *location) | |
5451 | { | |
5452 | const char *name = dentry->d_name.name; | |
5453 | int namelen = dentry->d_name.len; | |
5454 | struct btrfs_dir_item *di; | |
5455 | struct btrfs_path *path; | |
5456 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5457 | int ret = 0; |
39279cc3 CM |
5458 | |
5459 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5460 | if (!path) |
5461 | return -ENOMEM; | |
3954401f | 5462 | |
f85b7379 DS |
5463 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5464 | name, namelen, 0); | |
3cf5068f LB |
5465 | if (IS_ERR_OR_NULL(di)) { |
5466 | ret = di ? PTR_ERR(di) : -ENOENT; | |
005d6712 SY |
5467 | goto out; |
5468 | } | |
d397712b | 5469 | |
5f39d397 | 5470 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5471 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5472 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
005d6712 | 5473 | ret = -EUCLEAN; |
56a0e706 LB |
5474 | btrfs_warn(root->fs_info, |
5475 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5476 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5477 | location->objectid, location->type, location->offset); | |
56a0e706 | 5478 | } |
39279cc3 | 5479 | out: |
39279cc3 CM |
5480 | btrfs_free_path(path); |
5481 | return ret; | |
5482 | } | |
5483 | ||
5484 | /* | |
5485 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5486 | * needs to be changed to reflect the root directory of the tree root. This | |
5487 | * is kind of like crossing a mount point. | |
5488 | */ | |
2ff7e61e | 5489 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5490 | struct inode *dir, |
5491 | struct dentry *dentry, | |
5492 | struct btrfs_key *location, | |
5493 | struct btrfs_root **sub_root) | |
39279cc3 | 5494 | { |
4df27c4d YZ |
5495 | struct btrfs_path *path; |
5496 | struct btrfs_root *new_root; | |
5497 | struct btrfs_root_ref *ref; | |
5498 | struct extent_buffer *leaf; | |
1d4c08e0 | 5499 | struct btrfs_key key; |
4df27c4d YZ |
5500 | int ret; |
5501 | int err = 0; | |
39279cc3 | 5502 | |
4df27c4d YZ |
5503 | path = btrfs_alloc_path(); |
5504 | if (!path) { | |
5505 | err = -ENOMEM; | |
5506 | goto out; | |
5507 | } | |
39279cc3 | 5508 | |
4df27c4d | 5509 | err = -ENOENT; |
1d4c08e0 DS |
5510 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5511 | key.type = BTRFS_ROOT_REF_KEY; | |
5512 | key.offset = location->objectid; | |
5513 | ||
0b246afa | 5514 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5515 | if (ret) { |
5516 | if (ret < 0) | |
5517 | err = ret; | |
5518 | goto out; | |
5519 | } | |
39279cc3 | 5520 | |
4df27c4d YZ |
5521 | leaf = path->nodes[0]; |
5522 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5523 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5524 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5525 | goto out; | |
39279cc3 | 5526 | |
4df27c4d YZ |
5527 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5528 | (unsigned long)(ref + 1), | |
5529 | dentry->d_name.len); | |
5530 | if (ret) | |
5531 | goto out; | |
5532 | ||
b3b4aa74 | 5533 | btrfs_release_path(path); |
4df27c4d | 5534 | |
0b246afa | 5535 | new_root = btrfs_read_fs_root_no_name(fs_info, location); |
4df27c4d YZ |
5536 | if (IS_ERR(new_root)) { |
5537 | err = PTR_ERR(new_root); | |
5538 | goto out; | |
5539 | } | |
5540 | ||
4df27c4d YZ |
5541 | *sub_root = new_root; |
5542 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5543 | location->type = BTRFS_INODE_ITEM_KEY; | |
5544 | location->offset = 0; | |
5545 | err = 0; | |
5546 | out: | |
5547 | btrfs_free_path(path); | |
5548 | return err; | |
39279cc3 CM |
5549 | } |
5550 | ||
5d4f98a2 YZ |
5551 | static void inode_tree_add(struct inode *inode) |
5552 | { | |
5553 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5554 | struct btrfs_inode *entry; | |
03e860bd FNP |
5555 | struct rb_node **p; |
5556 | struct rb_node *parent; | |
cef21937 | 5557 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5558 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5559 | |
1d3382cb | 5560 | if (inode_unhashed(inode)) |
76dda93c | 5561 | return; |
e1409cef | 5562 | parent = NULL; |
5d4f98a2 | 5563 | spin_lock(&root->inode_lock); |
e1409cef | 5564 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5565 | while (*p) { |
5566 | parent = *p; | |
5567 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5568 | ||
37508515 | 5569 | if (ino < btrfs_ino(entry)) |
03e860bd | 5570 | p = &parent->rb_left; |
37508515 | 5571 | else if (ino > btrfs_ino(entry)) |
03e860bd | 5572 | p = &parent->rb_right; |
5d4f98a2 YZ |
5573 | else { |
5574 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5575 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5576 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5577 | RB_CLEAR_NODE(parent); |
5578 | spin_unlock(&root->inode_lock); | |
cef21937 | 5579 | return; |
5d4f98a2 YZ |
5580 | } |
5581 | } | |
cef21937 FDBM |
5582 | rb_link_node(new, parent, p); |
5583 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5584 | spin_unlock(&root->inode_lock); |
5585 | } | |
5586 | ||
5587 | static void inode_tree_del(struct inode *inode) | |
5588 | { | |
0b246afa | 5589 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5d4f98a2 | 5590 | struct btrfs_root *root = BTRFS_I(inode)->root; |
76dda93c | 5591 | int empty = 0; |
5d4f98a2 | 5592 | |
03e860bd | 5593 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5594 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5595 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5596 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5597 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5598 | } |
03e860bd | 5599 | spin_unlock(&root->inode_lock); |
76dda93c | 5600 | |
69e9c6c6 | 5601 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
0b246afa | 5602 | synchronize_srcu(&fs_info->subvol_srcu); |
76dda93c YZ |
5603 | spin_lock(&root->inode_lock); |
5604 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5605 | spin_unlock(&root->inode_lock); | |
5606 | if (empty) | |
5607 | btrfs_add_dead_root(root); | |
5608 | } | |
5609 | } | |
5610 | ||
5d4f98a2 | 5611 | |
e02119d5 CM |
5612 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5613 | { | |
5614 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5615 | inode->i_ino = args->location->objectid; |
5616 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5617 | sizeof(*args->location)); | |
e02119d5 | 5618 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5619 | return 0; |
5620 | } | |
5621 | ||
5622 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5623 | { | |
5624 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5625 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5626 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5627 | } |
5628 | ||
5d4f98a2 | 5629 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5630 | struct btrfs_key *location, |
5d4f98a2 | 5631 | struct btrfs_root *root) |
39279cc3 CM |
5632 | { |
5633 | struct inode *inode; | |
5634 | struct btrfs_iget_args args; | |
90d3e592 | 5635 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5636 | |
90d3e592 | 5637 | args.location = location; |
39279cc3 CM |
5638 | args.root = root; |
5639 | ||
778ba82b | 5640 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5641 | btrfs_init_locked_inode, |
5642 | (void *)&args); | |
5643 | return inode; | |
5644 | } | |
5645 | ||
1a54ef8c BR |
5646 | /* Get an inode object given its location and corresponding root. |
5647 | * Returns in *is_new if the inode was read from disk | |
5648 | */ | |
4222ea71 FM |
5649 | struct inode *btrfs_iget_path(struct super_block *s, struct btrfs_key *location, |
5650 | struct btrfs_root *root, int *new, | |
5651 | struct btrfs_path *path) | |
1a54ef8c BR |
5652 | { |
5653 | struct inode *inode; | |
5654 | ||
90d3e592 | 5655 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5656 | if (!inode) |
5d4f98a2 | 5657 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5658 | |
5659 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5660 | int ret; |
5661 | ||
4222ea71 | 5662 | ret = btrfs_read_locked_inode(inode, path); |
9bc2ceff | 5663 | if (!ret) { |
1748f843 MF |
5664 | inode_tree_add(inode); |
5665 | unlock_new_inode(inode); | |
5666 | if (new) | |
5667 | *new = 1; | |
5668 | } else { | |
f5b3a417 AV |
5669 | iget_failed(inode); |
5670 | /* | |
5671 | * ret > 0 can come from btrfs_search_slot called by | |
5672 | * btrfs_read_locked_inode, this means the inode item | |
5673 | * was not found. | |
5674 | */ | |
5675 | if (ret > 0) | |
5676 | ret = -ENOENT; | |
5677 | inode = ERR_PTR(ret); | |
1748f843 MF |
5678 | } |
5679 | } | |
5680 | ||
1a54ef8c BR |
5681 | return inode; |
5682 | } | |
5683 | ||
4222ea71 FM |
5684 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, |
5685 | struct btrfs_root *root, int *new) | |
5686 | { | |
5687 | return btrfs_iget_path(s, location, root, new, NULL); | |
5688 | } | |
5689 | ||
4df27c4d YZ |
5690 | static struct inode *new_simple_dir(struct super_block *s, |
5691 | struct btrfs_key *key, | |
5692 | struct btrfs_root *root) | |
5693 | { | |
5694 | struct inode *inode = new_inode(s); | |
5695 | ||
5696 | if (!inode) | |
5697 | return ERR_PTR(-ENOMEM); | |
5698 | ||
4df27c4d YZ |
5699 | BTRFS_I(inode)->root = root; |
5700 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5701 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5702 | |
5703 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5704 | inode->i_op = &btrfs_dir_ro_inode_operations; |
1fdf4194 | 5705 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5706 | inode->i_fop = &simple_dir_operations; |
5707 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5708 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5709 | inode->i_atime = inode->i_mtime; |
5710 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 5711 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
4df27c4d YZ |
5712 | |
5713 | return inode; | |
5714 | } | |
5715 | ||
3de4586c | 5716 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5717 | { |
0b246afa | 5718 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5719 | struct inode *inode; |
4df27c4d | 5720 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5721 | struct btrfs_root *sub_root = root; |
5722 | struct btrfs_key location; | |
76dda93c | 5723 | int index; |
b4aff1f8 | 5724 | int ret = 0; |
39279cc3 CM |
5725 | |
5726 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5727 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5728 | |
39e3c955 | 5729 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5730 | if (ret < 0) |
5731 | return ERR_PTR(ret); | |
5f39d397 | 5732 | |
4df27c4d | 5733 | if (location.type == BTRFS_INODE_ITEM_KEY) { |
73f73415 | 5734 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5735 | return inode; |
5736 | } | |
5737 | ||
0b246afa | 5738 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2ff7e61e | 5739 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5740 | &location, &sub_root); |
5741 | if (ret < 0) { | |
5742 | if (ret != -ENOENT) | |
5743 | inode = ERR_PTR(ret); | |
5744 | else | |
5745 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5746 | } else { | |
73f73415 | 5747 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5748 | } |
0b246afa | 5749 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
76dda93c | 5750 | |
34d19bad | 5751 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5752 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5753 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5754 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5755 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5756 | if (ret) { |
5757 | iput(inode); | |
66b4ffd1 | 5758 | inode = ERR_PTR(ret); |
01cd3367 | 5759 | } |
c71bf099 YZ |
5760 | } |
5761 | ||
3de4586c CM |
5762 | return inode; |
5763 | } | |
5764 | ||
fe15ce44 | 5765 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5766 | { |
5767 | struct btrfs_root *root; | |
2b0143b5 | 5768 | struct inode *inode = d_inode(dentry); |
76dda93c | 5769 | |
848cce0d | 5770 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5771 | inode = d_inode(dentry->d_parent); |
76dda93c | 5772 | |
848cce0d LZ |
5773 | if (inode) { |
5774 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5775 | if (btrfs_root_refs(&root->root_item) == 0) |
5776 | return 1; | |
848cce0d | 5777 | |
4a0cc7ca | 5778 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5779 | return 1; |
efefb143 | 5780 | } |
76dda93c YZ |
5781 | return 0; |
5782 | } | |
5783 | ||
3de4586c | 5784 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5785 | unsigned int flags) |
3de4586c | 5786 | { |
3837d208 | 5787 | struct inode *inode = btrfs_lookup_dentry(dir, dentry); |
5662344b | 5788 | |
3837d208 AV |
5789 | if (inode == ERR_PTR(-ENOENT)) |
5790 | inode = NULL; | |
41d28bca | 5791 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5792 | } |
5793 | ||
16cdcec7 | 5794 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5795 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5796 | }; | |
5797 | ||
23b5ec74 JB |
5798 | /* |
5799 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5800 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5801 | * our information into that, and then dir_emit from the buffer. This is | |
5802 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5803 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5804 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5805 | * tree lock. | |
5806 | */ | |
5807 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5808 | { | |
5809 | struct btrfs_file_private *private; | |
5810 | ||
5811 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5812 | if (!private) | |
5813 | return -ENOMEM; | |
5814 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
5815 | if (!private->filldir_buf) { | |
5816 | kfree(private); | |
5817 | return -ENOMEM; | |
5818 | } | |
5819 | file->private_data = private; | |
5820 | return 0; | |
5821 | } | |
5822 | ||
5823 | struct dir_entry { | |
5824 | u64 ino; | |
5825 | u64 offset; | |
5826 | unsigned type; | |
5827 | int name_len; | |
5828 | }; | |
5829 | ||
5830 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5831 | { | |
5832 | while (entries--) { | |
5833 | struct dir_entry *entry = addr; | |
5834 | char *name = (char *)(entry + 1); | |
5835 | ||
92d32170 DS |
5836 | ctx->pos = get_unaligned(&entry->offset); |
5837 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
5838 | get_unaligned(&entry->ino), | |
5839 | get_unaligned(&entry->type))) | |
23b5ec74 | 5840 | return 1; |
92d32170 DS |
5841 | addr += sizeof(struct dir_entry) + |
5842 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
5843 | ctx->pos++; |
5844 | } | |
5845 | return 0; | |
5846 | } | |
5847 | ||
9cdda8d3 | 5848 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5849 | { |
9cdda8d3 | 5850 | struct inode *inode = file_inode(file); |
39279cc3 | 5851 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 5852 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
5853 | struct btrfs_dir_item *di; |
5854 | struct btrfs_key key; | |
5f39d397 | 5855 | struct btrfs_key found_key; |
39279cc3 | 5856 | struct btrfs_path *path; |
23b5ec74 | 5857 | void *addr; |
16cdcec7 MX |
5858 | struct list_head ins_list; |
5859 | struct list_head del_list; | |
39279cc3 | 5860 | int ret; |
5f39d397 | 5861 | struct extent_buffer *leaf; |
39279cc3 | 5862 | int slot; |
5f39d397 CM |
5863 | char *name_ptr; |
5864 | int name_len; | |
23b5ec74 JB |
5865 | int entries = 0; |
5866 | int total_len = 0; | |
02dbfc99 | 5867 | bool put = false; |
c2951f32 | 5868 | struct btrfs_key location; |
5f39d397 | 5869 | |
9cdda8d3 AV |
5870 | if (!dir_emit_dots(file, ctx)) |
5871 | return 0; | |
5872 | ||
49593bfa | 5873 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5874 | if (!path) |
5875 | return -ENOMEM; | |
ff5714cc | 5876 | |
23b5ec74 | 5877 | addr = private->filldir_buf; |
e4058b54 | 5878 | path->reada = READA_FORWARD; |
49593bfa | 5879 | |
c2951f32 JM |
5880 | INIT_LIST_HEAD(&ins_list); |
5881 | INIT_LIST_HEAD(&del_list); | |
5882 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 5883 | |
23b5ec74 | 5884 | again: |
c2951f32 | 5885 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 5886 | key.offset = ctx->pos; |
4a0cc7ca | 5887 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 5888 | |
39279cc3 CM |
5889 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5890 | if (ret < 0) | |
5891 | goto err; | |
49593bfa DW |
5892 | |
5893 | while (1) { | |
23b5ec74 JB |
5894 | struct dir_entry *entry; |
5895 | ||
5f39d397 | 5896 | leaf = path->nodes[0]; |
39279cc3 | 5897 | slot = path->slots[0]; |
b9e03af0 LZ |
5898 | if (slot >= btrfs_header_nritems(leaf)) { |
5899 | ret = btrfs_next_leaf(root, path); | |
5900 | if (ret < 0) | |
5901 | goto err; | |
5902 | else if (ret > 0) | |
5903 | break; | |
5904 | continue; | |
39279cc3 | 5905 | } |
3de4586c | 5906 | |
5f39d397 CM |
5907 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5908 | ||
5909 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5910 | break; |
c2951f32 | 5911 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 5912 | break; |
9cdda8d3 | 5913 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5914 | goto next; |
c2951f32 | 5915 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 5916 | goto next; |
39279cc3 | 5917 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
c2951f32 | 5918 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
5919 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
5920 | PAGE_SIZE) { | |
5921 | btrfs_release_path(path); | |
5922 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5923 | if (ret) | |
5924 | goto nopos; | |
5925 | addr = private->filldir_buf; | |
5926 | entries = 0; | |
5927 | total_len = 0; | |
5928 | goto again; | |
c2951f32 | 5929 | } |
23b5ec74 JB |
5930 | |
5931 | entry = addr; | |
92d32170 | 5932 | put_unaligned(name_len, &entry->name_len); |
23b5ec74 | 5933 | name_ptr = (char *)(entry + 1); |
c2951f32 JM |
5934 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
5935 | name_len); | |
92d32170 DS |
5936 | put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)], |
5937 | &entry->type); | |
c2951f32 | 5938 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
92d32170 DS |
5939 | put_unaligned(location.objectid, &entry->ino); |
5940 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
5941 | entries++; |
5942 | addr += sizeof(struct dir_entry) + name_len; | |
5943 | total_len += sizeof(struct dir_entry) + name_len; | |
b9e03af0 LZ |
5944 | next: |
5945 | path->slots[0]++; | |
39279cc3 | 5946 | } |
23b5ec74 JB |
5947 | btrfs_release_path(path); |
5948 | ||
5949 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5950 | if (ret) | |
5951 | goto nopos; | |
49593bfa | 5952 | |
d2fbb2b5 | 5953 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 5954 | if (ret) |
bc4ef759 DS |
5955 | goto nopos; |
5956 | ||
db62efbb ZB |
5957 | /* |
5958 | * Stop new entries from being returned after we return the last | |
5959 | * entry. | |
5960 | * | |
5961 | * New directory entries are assigned a strictly increasing | |
5962 | * offset. This means that new entries created during readdir | |
5963 | * are *guaranteed* to be seen in the future by that readdir. | |
5964 | * This has broken buggy programs which operate on names as | |
5965 | * they're returned by readdir. Until we re-use freed offsets | |
5966 | * we have this hack to stop new entries from being returned | |
5967 | * under the assumption that they'll never reach this huge | |
5968 | * offset. | |
5969 | * | |
5970 | * This is being careful not to overflow 32bit loff_t unless the | |
5971 | * last entry requires it because doing so has broken 32bit apps | |
5972 | * in the past. | |
5973 | */ | |
c2951f32 JM |
5974 | if (ctx->pos >= INT_MAX) |
5975 | ctx->pos = LLONG_MAX; | |
5976 | else | |
5977 | ctx->pos = INT_MAX; | |
39279cc3 CM |
5978 | nopos: |
5979 | ret = 0; | |
5980 | err: | |
02dbfc99 OS |
5981 | if (put) |
5982 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 5983 | btrfs_free_path(path); |
39279cc3 CM |
5984 | return ret; |
5985 | } | |
5986 | ||
39279cc3 | 5987 | /* |
54aa1f4d | 5988 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5989 | * inode changes. But, it is most likely to find the inode in cache. |
5990 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5991 | * to keep or drop this code. | |
5992 | */ | |
48a3b636 | 5993 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 5994 | { |
2ff7e61e | 5995 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5996 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5997 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5998 | int ret; |
5999 | ||
72ac3c0d | 6000 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 6001 | return 0; |
39279cc3 | 6002 | |
7a7eaa40 | 6003 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
6004 | if (IS_ERR(trans)) |
6005 | return PTR_ERR(trans); | |
8929ecfa YZ |
6006 | |
6007 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
6008 | if (ret && ret == -ENOSPC) { |
6009 | /* whoops, lets try again with the full transaction */ | |
3a45bb20 | 6010 | btrfs_end_transaction(trans); |
94b60442 | 6011 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
6012 | if (IS_ERR(trans)) |
6013 | return PTR_ERR(trans); | |
8929ecfa | 6014 | |
94b60442 | 6015 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 6016 | } |
3a45bb20 | 6017 | btrfs_end_transaction(trans); |
16cdcec7 | 6018 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 6019 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6020 | |
6021 | return ret; | |
6022 | } | |
6023 | ||
6024 | /* | |
6025 | * This is a copy of file_update_time. We need this so we can return error on | |
6026 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6027 | */ | |
95582b00 | 6028 | static int btrfs_update_time(struct inode *inode, struct timespec64 *now, |
e41f941a | 6029 | int flags) |
22c44fe6 | 6030 | { |
2bc55652 | 6031 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3a8c7231 | 6032 | bool dirty = flags & ~S_VERSION; |
2bc55652 AB |
6033 | |
6034 | if (btrfs_root_readonly(root)) | |
6035 | return -EROFS; | |
6036 | ||
e41f941a | 6037 | if (flags & S_VERSION) |
3a8c7231 | 6038 | dirty |= inode_maybe_inc_iversion(inode, dirty); |
e41f941a JB |
6039 | if (flags & S_CTIME) |
6040 | inode->i_ctime = *now; | |
6041 | if (flags & S_MTIME) | |
6042 | inode->i_mtime = *now; | |
6043 | if (flags & S_ATIME) | |
6044 | inode->i_atime = *now; | |
3a8c7231 | 6045 | return dirty ? btrfs_dirty_inode(inode) : 0; |
39279cc3 CM |
6046 | } |
6047 | ||
d352ac68 CM |
6048 | /* |
6049 | * find the highest existing sequence number in a directory | |
6050 | * and then set the in-memory index_cnt variable to reflect | |
6051 | * free sequence numbers | |
6052 | */ | |
4c570655 | 6053 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 6054 | { |
4c570655 | 6055 | struct btrfs_root *root = inode->root; |
aec7477b JB |
6056 | struct btrfs_key key, found_key; |
6057 | struct btrfs_path *path; | |
6058 | struct extent_buffer *leaf; | |
6059 | int ret; | |
6060 | ||
4c570655 | 6061 | key.objectid = btrfs_ino(inode); |
962a298f | 6062 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6063 | key.offset = (u64)-1; |
6064 | ||
6065 | path = btrfs_alloc_path(); | |
6066 | if (!path) | |
6067 | return -ENOMEM; | |
6068 | ||
6069 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6070 | if (ret < 0) | |
6071 | goto out; | |
6072 | /* FIXME: we should be able to handle this */ | |
6073 | if (ret == 0) | |
6074 | goto out; | |
6075 | ret = 0; | |
6076 | ||
6077 | /* | |
6078 | * MAGIC NUMBER EXPLANATION: | |
6079 | * since we search a directory based on f_pos we have to start at 2 | |
6080 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6081 | * else has to start at 2 | |
6082 | */ | |
6083 | if (path->slots[0] == 0) { | |
4c570655 | 6084 | inode->index_cnt = 2; |
aec7477b JB |
6085 | goto out; |
6086 | } | |
6087 | ||
6088 | path->slots[0]--; | |
6089 | ||
6090 | leaf = path->nodes[0]; | |
6091 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6092 | ||
4c570655 | 6093 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6094 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
4c570655 | 6095 | inode->index_cnt = 2; |
aec7477b JB |
6096 | goto out; |
6097 | } | |
6098 | ||
4c570655 | 6099 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6100 | out: |
6101 | btrfs_free_path(path); | |
6102 | return ret; | |
6103 | } | |
6104 | ||
d352ac68 CM |
6105 | /* |
6106 | * helper to find a free sequence number in a given directory. This current | |
6107 | * code is very simple, later versions will do smarter things in the btree | |
6108 | */ | |
877574e2 | 6109 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6110 | { |
6111 | int ret = 0; | |
6112 | ||
877574e2 NB |
6113 | if (dir->index_cnt == (u64)-1) { |
6114 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6115 | if (ret) { |
6116 | ret = btrfs_set_inode_index_count(dir); | |
6117 | if (ret) | |
6118 | return ret; | |
6119 | } | |
aec7477b JB |
6120 | } |
6121 | ||
877574e2 NB |
6122 | *index = dir->index_cnt; |
6123 | dir->index_cnt++; | |
aec7477b JB |
6124 | |
6125 | return ret; | |
6126 | } | |
6127 | ||
b0d5d10f CM |
6128 | static int btrfs_insert_inode_locked(struct inode *inode) |
6129 | { | |
6130 | struct btrfs_iget_args args; | |
6131 | args.location = &BTRFS_I(inode)->location; | |
6132 | args.root = BTRFS_I(inode)->root; | |
6133 | ||
6134 | return insert_inode_locked4(inode, | |
6135 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6136 | btrfs_find_actor, &args); | |
6137 | } | |
6138 | ||
19aee8de AJ |
6139 | /* |
6140 | * Inherit flags from the parent inode. | |
6141 | * | |
6142 | * Currently only the compression flags and the cow flags are inherited. | |
6143 | */ | |
6144 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
6145 | { | |
6146 | unsigned int flags; | |
6147 | ||
6148 | if (!dir) | |
6149 | return; | |
6150 | ||
6151 | flags = BTRFS_I(dir)->flags; | |
6152 | ||
6153 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
6154 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
6155 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
6156 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
6157 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
6158 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
6159 | } | |
6160 | ||
6161 | if (flags & BTRFS_INODE_NODATACOW) { | |
6162 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
6163 | if (S_ISREG(inode->i_mode)) | |
6164 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6165 | } | |
6166 | ||
7b6a221e | 6167 | btrfs_sync_inode_flags_to_i_flags(inode); |
19aee8de AJ |
6168 | } |
6169 | ||
39279cc3 CM |
6170 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6171 | struct btrfs_root *root, | |
aec7477b | 6172 | struct inode *dir, |
9c58309d | 6173 | const char *name, int name_len, |
175a4eb7 AV |
6174 | u64 ref_objectid, u64 objectid, |
6175 | umode_t mode, u64 *index) | |
39279cc3 | 6176 | { |
0b246afa | 6177 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6178 | struct inode *inode; |
5f39d397 | 6179 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6180 | struct btrfs_key *location; |
5f39d397 | 6181 | struct btrfs_path *path; |
9c58309d CM |
6182 | struct btrfs_inode_ref *ref; |
6183 | struct btrfs_key key[2]; | |
6184 | u32 sizes[2]; | |
ef3b9af5 | 6185 | int nitems = name ? 2 : 1; |
9c58309d | 6186 | unsigned long ptr; |
39279cc3 | 6187 | int ret; |
39279cc3 | 6188 | |
5f39d397 | 6189 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6190 | if (!path) |
6191 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6192 | |
0b246afa | 6193 | inode = new_inode(fs_info->sb); |
8fb27640 YS |
6194 | if (!inode) { |
6195 | btrfs_free_path(path); | |
39279cc3 | 6196 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6197 | } |
39279cc3 | 6198 | |
5762b5c9 FM |
6199 | /* |
6200 | * O_TMPFILE, set link count to 0, so that after this point, | |
6201 | * we fill in an inode item with the correct link count. | |
6202 | */ | |
6203 | if (!name) | |
6204 | set_nlink(inode, 0); | |
6205 | ||
581bb050 LZ |
6206 | /* |
6207 | * we have to initialize this early, so we can reclaim the inode | |
6208 | * number if we fail afterwards in this function. | |
6209 | */ | |
6210 | inode->i_ino = objectid; | |
6211 | ||
ef3b9af5 | 6212 | if (dir && name) { |
1abe9b8a | 6213 | trace_btrfs_inode_request(dir); |
6214 | ||
877574e2 | 6215 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 6216 | if (ret) { |
8fb27640 | 6217 | btrfs_free_path(path); |
09771430 | 6218 | iput(inode); |
aec7477b | 6219 | return ERR_PTR(ret); |
09771430 | 6220 | } |
ef3b9af5 FM |
6221 | } else if (dir) { |
6222 | *index = 0; | |
aec7477b JB |
6223 | } |
6224 | /* | |
6225 | * index_cnt is ignored for everything but a dir, | |
df6703e1 | 6226 | * btrfs_set_inode_index_count has an explanation for the magic |
aec7477b JB |
6227 | * number |
6228 | */ | |
6229 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6230 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6231 | BTRFS_I(inode)->root = root; |
e02119d5 | 6232 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6233 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6234 | |
5dc562c5 JB |
6235 | /* |
6236 | * We could have gotten an inode number from somebody who was fsynced | |
6237 | * and then removed in this same transaction, so let's just set full | |
6238 | * sync since it will be a full sync anyway and this will blow away the | |
6239 | * old info in the log. | |
6240 | */ | |
6241 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6242 | ||
9c58309d | 6243 | key[0].objectid = objectid; |
962a298f | 6244 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6245 | key[0].offset = 0; |
6246 | ||
9c58309d | 6247 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6248 | |
6249 | if (name) { | |
6250 | /* | |
6251 | * Start new inodes with an inode_ref. This is slightly more | |
6252 | * efficient for small numbers of hard links since they will | |
6253 | * be packed into one item. Extended refs will kick in if we | |
6254 | * add more hard links than can fit in the ref item. | |
6255 | */ | |
6256 | key[1].objectid = objectid; | |
962a298f | 6257 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6258 | key[1].offset = ref_objectid; |
6259 | ||
6260 | sizes[1] = name_len + sizeof(*ref); | |
6261 | } | |
9c58309d | 6262 | |
b0d5d10f CM |
6263 | location = &BTRFS_I(inode)->location; |
6264 | location->objectid = objectid; | |
6265 | location->offset = 0; | |
962a298f | 6266 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6267 | |
6268 | ret = btrfs_insert_inode_locked(inode); | |
32955c54 AV |
6269 | if (ret < 0) { |
6270 | iput(inode); | |
b0d5d10f | 6271 | goto fail; |
32955c54 | 6272 | } |
b0d5d10f | 6273 | |
b9473439 | 6274 | path->leave_spinning = 1; |
ef3b9af5 | 6275 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6276 | if (ret != 0) |
b0d5d10f | 6277 | goto fail_unlock; |
5f39d397 | 6278 | |
ecc11fab | 6279 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6280 | inode_set_bytes(inode, 0); |
9cc97d64 | 6281 | |
c2050a45 | 6282 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6283 | inode->i_atime = inode->i_mtime; |
6284 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 6285 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
9cc97d64 | 6286 | |
5f39d397 CM |
6287 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6288 | struct btrfs_inode_item); | |
b159fa28 | 6289 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6290 | sizeof(*inode_item)); |
e02119d5 | 6291 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6292 | |
ef3b9af5 FM |
6293 | if (name) { |
6294 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6295 | struct btrfs_inode_ref); | |
6296 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6297 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6298 | ptr = (unsigned long)(ref + 1); | |
6299 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6300 | } | |
9c58309d | 6301 | |
5f39d397 CM |
6302 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6303 | btrfs_free_path(path); | |
6304 | ||
6cbff00f CH |
6305 | btrfs_inherit_iflags(inode, dir); |
6306 | ||
569254b0 | 6307 | if (S_ISREG(mode)) { |
0b246afa | 6308 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6309 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6310 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6311 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6312 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6313 | } |
6314 | ||
5d4f98a2 | 6315 | inode_tree_add(inode); |
1abe9b8a | 6316 | |
6317 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6318 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6319 | |
8ea05e3a AB |
6320 | btrfs_update_root_times(trans, root); |
6321 | ||
63541927 FDBM |
6322 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6323 | if (ret) | |
0b246afa | 6324 | btrfs_err(fs_info, |
63541927 | 6325 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6326 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6327 | |
39279cc3 | 6328 | return inode; |
b0d5d10f CM |
6329 | |
6330 | fail_unlock: | |
32955c54 | 6331 | discard_new_inode(inode); |
5f39d397 | 6332 | fail: |
ef3b9af5 | 6333 | if (dir && name) |
aec7477b | 6334 | BTRFS_I(dir)->index_cnt--; |
5f39d397 CM |
6335 | btrfs_free_path(path); |
6336 | return ERR_PTR(ret); | |
39279cc3 CM |
6337 | } |
6338 | ||
6339 | static inline u8 btrfs_inode_type(struct inode *inode) | |
6340 | { | |
6341 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
6342 | } | |
6343 | ||
d352ac68 CM |
6344 | /* |
6345 | * utility function to add 'inode' into 'parent_inode' with | |
6346 | * a give name and a given sequence number. | |
6347 | * if 'add_backref' is true, also insert a backref from the | |
6348 | * inode to the parent directory. | |
6349 | */ | |
e02119d5 | 6350 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6351 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6352 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6353 | { |
4df27c4d | 6354 | int ret = 0; |
39279cc3 | 6355 | struct btrfs_key key; |
db0a669f NB |
6356 | struct btrfs_root *root = parent_inode->root; |
6357 | u64 ino = btrfs_ino(inode); | |
6358 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6359 | |
33345d01 | 6360 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6361 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6362 | } else { |
33345d01 | 6363 | key.objectid = ino; |
962a298f | 6364 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6365 | key.offset = 0; |
6366 | } | |
6367 | ||
33345d01 | 6368 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
6025c19f | 6369 | ret = btrfs_add_root_ref(trans, key.objectid, |
0b246afa JM |
6370 | root->root_key.objectid, parent_ino, |
6371 | index, name, name_len); | |
4df27c4d | 6372 | } else if (add_backref) { |
33345d01 LZ |
6373 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6374 | parent_ino, index); | |
4df27c4d | 6375 | } |
39279cc3 | 6376 | |
79787eaa JM |
6377 | /* Nothing to clean up yet */ |
6378 | if (ret) | |
6379 | return ret; | |
4df27c4d | 6380 | |
684572df | 6381 | ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key, |
db0a669f | 6382 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6383 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6384 | goto fail_dir_item; |
6385 | else if (ret) { | |
66642832 | 6386 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6387 | return ret; |
39279cc3 | 6388 | } |
79787eaa | 6389 | |
db0a669f | 6390 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6391 | name_len * 2); |
db0a669f NB |
6392 | inode_inc_iversion(&parent_inode->vfs_inode); |
6393 | parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime = | |
6394 | current_time(&parent_inode->vfs_inode); | |
6395 | ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode); | |
79787eaa | 6396 | if (ret) |
66642832 | 6397 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6398 | return ret; |
fe66a05a CM |
6399 | |
6400 | fail_dir_item: | |
6401 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6402 | u64 local_index; | |
6403 | int err; | |
3ee1c553 | 6404 | err = btrfs_del_root_ref(trans, key.objectid, |
0b246afa JM |
6405 | root->root_key.objectid, parent_ino, |
6406 | &local_index, name, name_len); | |
fe66a05a CM |
6407 | |
6408 | } else if (add_backref) { | |
6409 | u64 local_index; | |
6410 | int err; | |
6411 | ||
6412 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6413 | ino, parent_ino, &local_index); | |
6414 | } | |
6415 | return ret; | |
39279cc3 CM |
6416 | } |
6417 | ||
6418 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
cef415af NB |
6419 | struct btrfs_inode *dir, struct dentry *dentry, |
6420 | struct btrfs_inode *inode, int backref, u64 index) | |
39279cc3 | 6421 | { |
a1b075d2 JB |
6422 | int err = btrfs_add_link(trans, dir, inode, |
6423 | dentry->d_name.name, dentry->d_name.len, | |
6424 | backref, index); | |
39279cc3 CM |
6425 | if (err > 0) |
6426 | err = -EEXIST; | |
6427 | return err; | |
6428 | } | |
6429 | ||
618e21d5 | 6430 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6431 | umode_t mode, dev_t rdev) |
618e21d5 | 6432 | { |
2ff7e61e | 6433 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6434 | struct btrfs_trans_handle *trans; |
6435 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6436 | struct inode *inode = NULL; |
618e21d5 | 6437 | int err; |
618e21d5 | 6438 | u64 objectid; |
00e4e6b3 | 6439 | u64 index = 0; |
618e21d5 | 6440 | |
9ed74f2d JB |
6441 | /* |
6442 | * 2 for inode item and ref | |
6443 | * 2 for dir items | |
6444 | * 1 for xattr if selinux is on | |
6445 | */ | |
a22285a6 YZ |
6446 | trans = btrfs_start_transaction(root, 5); |
6447 | if (IS_ERR(trans)) | |
6448 | return PTR_ERR(trans); | |
1832a6d5 | 6449 | |
581bb050 LZ |
6450 | err = btrfs_find_free_ino(root, &objectid); |
6451 | if (err) | |
6452 | goto out_unlock; | |
6453 | ||
aec7477b | 6454 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6455 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6456 | mode, &index); | |
7cf96da3 TI |
6457 | if (IS_ERR(inode)) { |
6458 | err = PTR_ERR(inode); | |
32955c54 | 6459 | inode = NULL; |
618e21d5 | 6460 | goto out_unlock; |
7cf96da3 | 6461 | } |
618e21d5 | 6462 | |
ad19db71 CS |
6463 | /* |
6464 | * If the active LSM wants to access the inode during | |
6465 | * d_instantiate it needs these. Smack checks to see | |
6466 | * if the filesystem supports xattrs by looking at the | |
6467 | * ops vector. | |
6468 | */ | |
ad19db71 | 6469 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6470 | init_special_inode(inode, inode->i_mode, rdev); |
6471 | ||
6472 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6473 | if (err) |
32955c54 | 6474 | goto out_unlock; |
b0d5d10f | 6475 | |
cef415af NB |
6476 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6477 | 0, index); | |
32955c54 AV |
6478 | if (err) |
6479 | goto out_unlock; | |
6480 | ||
6481 | btrfs_update_inode(trans, root, inode); | |
6482 | d_instantiate_new(dentry, inode); | |
b0d5d10f | 6483 | |
618e21d5 | 6484 | out_unlock: |
3a45bb20 | 6485 | btrfs_end_transaction(trans); |
2ff7e61e | 6486 | btrfs_btree_balance_dirty(fs_info); |
32955c54 | 6487 | if (err && inode) { |
618e21d5 | 6488 | inode_dec_link_count(inode); |
32955c54 | 6489 | discard_new_inode(inode); |
618e21d5 | 6490 | } |
618e21d5 JB |
6491 | return err; |
6492 | } | |
6493 | ||
39279cc3 | 6494 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6495 | umode_t mode, bool excl) |
39279cc3 | 6496 | { |
2ff7e61e | 6497 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6498 | struct btrfs_trans_handle *trans; |
6499 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6500 | struct inode *inode = NULL; |
a22285a6 | 6501 | int err; |
39279cc3 | 6502 | u64 objectid; |
00e4e6b3 | 6503 | u64 index = 0; |
39279cc3 | 6504 | |
9ed74f2d JB |
6505 | /* |
6506 | * 2 for inode item and ref | |
6507 | * 2 for dir items | |
6508 | * 1 for xattr if selinux is on | |
6509 | */ | |
a22285a6 YZ |
6510 | trans = btrfs_start_transaction(root, 5); |
6511 | if (IS_ERR(trans)) | |
6512 | return PTR_ERR(trans); | |
9ed74f2d | 6513 | |
581bb050 LZ |
6514 | err = btrfs_find_free_ino(root, &objectid); |
6515 | if (err) | |
6516 | goto out_unlock; | |
6517 | ||
aec7477b | 6518 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6519 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6520 | mode, &index); | |
7cf96da3 TI |
6521 | if (IS_ERR(inode)) { |
6522 | err = PTR_ERR(inode); | |
32955c54 | 6523 | inode = NULL; |
39279cc3 | 6524 | goto out_unlock; |
7cf96da3 | 6525 | } |
ad19db71 CS |
6526 | /* |
6527 | * If the active LSM wants to access the inode during | |
6528 | * d_instantiate it needs these. Smack checks to see | |
6529 | * if the filesystem supports xattrs by looking at the | |
6530 | * ops vector. | |
6531 | */ | |
6532 | inode->i_fop = &btrfs_file_operations; | |
6533 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6534 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6535 | |
6536 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6537 | if (err) | |
32955c54 | 6538 | goto out_unlock; |
b0d5d10f CM |
6539 | |
6540 | err = btrfs_update_inode(trans, root, inode); | |
6541 | if (err) | |
32955c54 | 6542 | goto out_unlock; |
ad19db71 | 6543 | |
cef415af NB |
6544 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6545 | 0, index); | |
39279cc3 | 6546 | if (err) |
32955c54 | 6547 | goto out_unlock; |
43baa579 | 6548 | |
43baa579 | 6549 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
1e2e547a | 6550 | d_instantiate_new(dentry, inode); |
43baa579 | 6551 | |
39279cc3 | 6552 | out_unlock: |
3a45bb20 | 6553 | btrfs_end_transaction(trans); |
32955c54 | 6554 | if (err && inode) { |
39279cc3 | 6555 | inode_dec_link_count(inode); |
32955c54 | 6556 | discard_new_inode(inode); |
39279cc3 | 6557 | } |
2ff7e61e | 6558 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6559 | return err; |
6560 | } | |
6561 | ||
6562 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6563 | struct dentry *dentry) | |
6564 | { | |
271dba45 | 6565 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6566 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6567 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6568 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6569 | u64 index; |
39279cc3 CM |
6570 | int err; |
6571 | int drop_inode = 0; | |
6572 | ||
4a8be425 | 6573 | /* do not allow sys_link's with other subvols of the same device */ |
4fd786e6 | 6574 | if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid) |
3ab3564f | 6575 | return -EXDEV; |
4a8be425 | 6576 | |
f186373f | 6577 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6578 | return -EMLINK; |
4a8be425 | 6579 | |
877574e2 | 6580 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6581 | if (err) |
6582 | goto fail; | |
6583 | ||
a22285a6 | 6584 | /* |
7e6b6465 | 6585 | * 2 items for inode and inode ref |
a22285a6 | 6586 | * 2 items for dir items |
7e6b6465 | 6587 | * 1 item for parent inode |
399b0bbf | 6588 | * 1 item for orphan item deletion if O_TMPFILE |
a22285a6 | 6589 | */ |
399b0bbf | 6590 | trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6); |
a22285a6 YZ |
6591 | if (IS_ERR(trans)) { |
6592 | err = PTR_ERR(trans); | |
271dba45 | 6593 | trans = NULL; |
a22285a6 YZ |
6594 | goto fail; |
6595 | } | |
5f39d397 | 6596 | |
67de1176 MX |
6597 | /* There are several dir indexes for this inode, clear the cache. */ |
6598 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6599 | inc_nlink(inode); |
0c4d2d95 | 6600 | inode_inc_iversion(inode); |
c2050a45 | 6601 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6602 | ihold(inode); |
e9976151 | 6603 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6604 | |
cef415af NB |
6605 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6606 | 1, index); | |
5f39d397 | 6607 | |
a5719521 | 6608 | if (err) { |
54aa1f4d | 6609 | drop_inode = 1; |
a5719521 | 6610 | } else { |
10d9f309 | 6611 | struct dentry *parent = dentry->d_parent; |
d4682ba0 FM |
6612 | int ret; |
6613 | ||
a5719521 | 6614 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6615 | if (err) |
6616 | goto fail; | |
ef3b9af5 FM |
6617 | if (inode->i_nlink == 1) { |
6618 | /* | |
6619 | * If new hard link count is 1, it's a file created | |
6620 | * with open(2) O_TMPFILE flag. | |
6621 | */ | |
3d6ae7bb | 6622 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6623 | if (err) |
6624 | goto fail; | |
6625 | } | |
41bd6067 | 6626 | BTRFS_I(inode)->last_link_trans = trans->transid; |
08c422c2 | 6627 | d_instantiate(dentry, inode); |
d4682ba0 FM |
6628 | ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent, |
6629 | true, NULL); | |
6630 | if (ret == BTRFS_NEED_TRANS_COMMIT) { | |
6631 | err = btrfs_commit_transaction(trans); | |
6632 | trans = NULL; | |
6633 | } | |
a5719521 | 6634 | } |
39279cc3 | 6635 | |
1832a6d5 | 6636 | fail: |
271dba45 | 6637 | if (trans) |
3a45bb20 | 6638 | btrfs_end_transaction(trans); |
39279cc3 CM |
6639 | if (drop_inode) { |
6640 | inode_dec_link_count(inode); | |
6641 | iput(inode); | |
6642 | } | |
2ff7e61e | 6643 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6644 | return err; |
6645 | } | |
6646 | ||
18bb1db3 | 6647 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6648 | { |
2ff7e61e | 6649 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6650 | struct inode *inode = NULL; |
39279cc3 CM |
6651 | struct btrfs_trans_handle *trans; |
6652 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6653 | int err = 0; | |
b9d86667 | 6654 | u64 objectid = 0; |
00e4e6b3 | 6655 | u64 index = 0; |
39279cc3 | 6656 | |
9ed74f2d JB |
6657 | /* |
6658 | * 2 items for inode and ref | |
6659 | * 2 items for dir items | |
6660 | * 1 for xattr if selinux is on | |
6661 | */ | |
a22285a6 YZ |
6662 | trans = btrfs_start_transaction(root, 5); |
6663 | if (IS_ERR(trans)) | |
6664 | return PTR_ERR(trans); | |
39279cc3 | 6665 | |
581bb050 LZ |
6666 | err = btrfs_find_free_ino(root, &objectid); |
6667 | if (err) | |
6668 | goto out_fail; | |
6669 | ||
aec7477b | 6670 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6671 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6672 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6673 | if (IS_ERR(inode)) { |
6674 | err = PTR_ERR(inode); | |
32955c54 | 6675 | inode = NULL; |
39279cc3 CM |
6676 | goto out_fail; |
6677 | } | |
5f39d397 | 6678 | |
b0d5d10f CM |
6679 | /* these must be set before we unlock the inode */ |
6680 | inode->i_op = &btrfs_dir_inode_operations; | |
6681 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6682 | |
2a7dba39 | 6683 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6684 | if (err) |
32955c54 | 6685 | goto out_fail; |
39279cc3 | 6686 | |
6ef06d27 | 6687 | btrfs_i_size_write(BTRFS_I(inode), 0); |
39279cc3 CM |
6688 | err = btrfs_update_inode(trans, root, inode); |
6689 | if (err) | |
32955c54 | 6690 | goto out_fail; |
5f39d397 | 6691 | |
db0a669f NB |
6692 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6693 | dentry->d_name.name, | |
6694 | dentry->d_name.len, 0, index); | |
39279cc3 | 6695 | if (err) |
32955c54 | 6696 | goto out_fail; |
5f39d397 | 6697 | |
1e2e547a | 6698 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
6699 | |
6700 | out_fail: | |
3a45bb20 | 6701 | btrfs_end_transaction(trans); |
32955c54 | 6702 | if (err && inode) { |
c7cfb8a5 | 6703 | inode_dec_link_count(inode); |
32955c54 | 6704 | discard_new_inode(inode); |
c7cfb8a5 | 6705 | } |
2ff7e61e | 6706 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6707 | return err; |
6708 | } | |
6709 | ||
c8b97818 | 6710 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6711 | struct page *page, |
c8b97818 CM |
6712 | size_t pg_offset, u64 extent_offset, |
6713 | struct btrfs_file_extent_item *item) | |
6714 | { | |
6715 | int ret; | |
6716 | struct extent_buffer *leaf = path->nodes[0]; | |
6717 | char *tmp; | |
6718 | size_t max_size; | |
6719 | unsigned long inline_size; | |
6720 | unsigned long ptr; | |
261507a0 | 6721 | int compress_type; |
c8b97818 CM |
6722 | |
6723 | WARN_ON(pg_offset != 0); | |
261507a0 | 6724 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6725 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6726 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6727 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6728 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6729 | if (!tmp) |
6730 | return -ENOMEM; | |
c8b97818 CM |
6731 | ptr = btrfs_file_extent_inline_start(item); |
6732 | ||
6733 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6734 | ||
09cbfeaf | 6735 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6736 | ret = btrfs_decompress(compress_type, tmp, page, |
6737 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6738 | |
6739 | /* | |
6740 | * decompression code contains a memset to fill in any space between the end | |
6741 | * of the uncompressed data and the end of max_size in case the decompressed | |
6742 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6743 | * the end of an inline extent and the beginning of the next block, so we | |
6744 | * cover that region here. | |
6745 | */ | |
6746 | ||
6747 | if (max_size + pg_offset < PAGE_SIZE) { | |
6748 | char *map = kmap(page); | |
6749 | memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset); | |
6750 | kunmap(page); | |
6751 | } | |
c8b97818 | 6752 | kfree(tmp); |
166ae5a4 | 6753 | return ret; |
c8b97818 CM |
6754 | } |
6755 | ||
d352ac68 CM |
6756 | /* |
6757 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6758 | * the ugly parts come from merging extents from the disk with the in-ram |
6759 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6760 | * where the in-ram extents might be locked pending data=ordered completion. |
6761 | * | |
6762 | * This also copies inline extents directly into the page. | |
6763 | */ | |
fc4f21b1 | 6764 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
de2c6615 LB |
6765 | struct page *page, |
6766 | size_t pg_offset, u64 start, u64 len, | |
6767 | int create) | |
a52d9a80 | 6768 | { |
3ffbd68c | 6769 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
a52d9a80 CM |
6770 | int ret; |
6771 | int err = 0; | |
a52d9a80 CM |
6772 | u64 extent_start = 0; |
6773 | u64 extent_end = 0; | |
fc4f21b1 | 6774 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6775 | u32 found_type; |
f421950f | 6776 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6777 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6778 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6779 | struct extent_buffer *leaf; |
6780 | struct btrfs_key found_key; | |
a52d9a80 | 6781 | struct extent_map *em = NULL; |
fc4f21b1 NB |
6782 | struct extent_map_tree *em_tree = &inode->extent_tree; |
6783 | struct extent_io_tree *io_tree = &inode->io_tree; | |
7ffbb598 | 6784 | const bool new_inline = !page || create; |
a52d9a80 | 6785 | |
890871be | 6786 | read_lock(&em_tree->lock); |
d1310b2e | 6787 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d | 6788 | if (em) |
0b246afa | 6789 | em->bdev = fs_info->fs_devices->latest_bdev; |
890871be | 6790 | read_unlock(&em_tree->lock); |
d1310b2e | 6791 | |
a52d9a80 | 6792 | if (em) { |
e1c4b745 CM |
6793 | if (em->start > start || em->start + em->len <= start) |
6794 | free_extent_map(em); | |
6795 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6796 | free_extent_map(em); |
6797 | else | |
6798 | goto out; | |
a52d9a80 | 6799 | } |
172ddd60 | 6800 | em = alloc_extent_map(); |
a52d9a80 | 6801 | if (!em) { |
d1310b2e CM |
6802 | err = -ENOMEM; |
6803 | goto out; | |
a52d9a80 | 6804 | } |
0b246afa | 6805 | em->bdev = fs_info->fs_devices->latest_bdev; |
d1310b2e | 6806 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6807 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6808 | em->len = (u64)-1; |
c8b97818 | 6809 | em->block_len = (u64)-1; |
f421950f | 6810 | |
bee6ec82 | 6811 | path = btrfs_alloc_path(); |
f421950f | 6812 | if (!path) { |
bee6ec82 LB |
6813 | err = -ENOMEM; |
6814 | goto out; | |
f421950f CM |
6815 | } |
6816 | ||
bee6ec82 LB |
6817 | /* Chances are we'll be called again, so go ahead and do readahead */ |
6818 | path->reada = READA_FORWARD; | |
6819 | ||
e49aabd9 LB |
6820 | /* |
6821 | * Unless we're going to uncompress the inline extent, no sleep would | |
6822 | * happen. | |
6823 | */ | |
6824 | path->leave_spinning = 1; | |
6825 | ||
5c9a702e | 6826 | ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0); |
a52d9a80 CM |
6827 | if (ret < 0) { |
6828 | err = ret; | |
6829 | goto out; | |
6830 | } | |
6831 | ||
6832 | if (ret != 0) { | |
6833 | if (path->slots[0] == 0) | |
6834 | goto not_found; | |
6835 | path->slots[0]--; | |
6836 | } | |
6837 | ||
5f39d397 CM |
6838 | leaf = path->nodes[0]; |
6839 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6840 | struct btrfs_file_extent_item); |
a52d9a80 | 6841 | /* are we inside the extent that was found? */ |
5f39d397 | 6842 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 6843 | found_type = found_key.type; |
5f39d397 | 6844 | if (found_key.objectid != objectid || |
a52d9a80 | 6845 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6846 | /* |
6847 | * If we backup past the first extent we want to move forward | |
6848 | * and see if there is an extent in front of us, otherwise we'll | |
6849 | * say there is a hole for our whole search range which can | |
6850 | * cause problems. | |
6851 | */ | |
6852 | extent_end = start; | |
6853 | goto next; | |
a52d9a80 CM |
6854 | } |
6855 | ||
5f39d397 CM |
6856 | found_type = btrfs_file_extent_type(leaf, item); |
6857 | extent_start = found_key.offset; | |
d899e052 YZ |
6858 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6859 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6860 | extent_end = extent_start + |
db94535d | 6861 | btrfs_file_extent_num_bytes(leaf, item); |
09ed2f16 LB |
6862 | |
6863 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, | |
6864 | extent_start); | |
9036c102 YZ |
6865 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6866 | size_t size; | |
e41ca589 QW |
6867 | |
6868 | size = btrfs_file_extent_ram_bytes(leaf, item); | |
da17066c | 6869 | extent_end = ALIGN(extent_start + size, |
0b246afa | 6870 | fs_info->sectorsize); |
09ed2f16 LB |
6871 | |
6872 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, | |
6873 | path->slots[0], | |
6874 | extent_start); | |
9036c102 | 6875 | } |
25a50341 | 6876 | next: |
9036c102 YZ |
6877 | if (start >= extent_end) { |
6878 | path->slots[0]++; | |
6879 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6880 | ret = btrfs_next_leaf(root, path); | |
6881 | if (ret < 0) { | |
6882 | err = ret; | |
6883 | goto out; | |
a52d9a80 | 6884 | } |
9036c102 YZ |
6885 | if (ret > 0) |
6886 | goto not_found; | |
6887 | leaf = path->nodes[0]; | |
a52d9a80 | 6888 | } |
9036c102 YZ |
6889 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6890 | if (found_key.objectid != objectid || | |
6891 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6892 | goto not_found; | |
6893 | if (start + len <= found_key.offset) | |
6894 | goto not_found; | |
e2eca69d WS |
6895 | if (start > found_key.offset) |
6896 | goto next; | |
9036c102 | 6897 | em->start = start; |
70c8a91c | 6898 | em->orig_start = start; |
9036c102 YZ |
6899 | em->len = found_key.offset - start; |
6900 | goto not_found_em; | |
6901 | } | |
6902 | ||
fc4f21b1 | 6903 | btrfs_extent_item_to_extent_map(inode, path, item, |
9cdc5124 | 6904 | new_inline, em); |
7ffbb598 | 6905 | |
d899e052 YZ |
6906 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6907 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
6908 | goto insert; |
6909 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6910 | unsigned long ptr; |
a52d9a80 | 6911 | char *map; |
3326d1b0 CM |
6912 | size_t size; |
6913 | size_t extent_offset; | |
6914 | size_t copy_size; | |
a52d9a80 | 6915 | |
7ffbb598 | 6916 | if (new_inline) |
689f9346 | 6917 | goto out; |
5f39d397 | 6918 | |
e41ca589 | 6919 | size = btrfs_file_extent_ram_bytes(leaf, item); |
9036c102 | 6920 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
6921 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
6922 | size - extent_offset); | |
3326d1b0 | 6923 | em->start = extent_start + extent_offset; |
0b246afa | 6924 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 6925 | em->orig_block_len = em->len; |
70c8a91c | 6926 | em->orig_start = em->start; |
689f9346 | 6927 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
e49aabd9 LB |
6928 | |
6929 | btrfs_set_path_blocking(path); | |
bf46f52d | 6930 | if (!PageUptodate(page)) { |
261507a0 LZ |
6931 | if (btrfs_file_extent_compression(leaf, item) != |
6932 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 6933 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 6934 | extent_offset, item); |
166ae5a4 ZB |
6935 | if (ret) { |
6936 | err = ret; | |
6937 | goto out; | |
6938 | } | |
c8b97818 CM |
6939 | } else { |
6940 | map = kmap(page); | |
6941 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6942 | copy_size); | |
09cbfeaf | 6943 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 6944 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 6945 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
6946 | copy_size); |
6947 | } | |
c8b97818 CM |
6948 | kunmap(page); |
6949 | } | |
179e29e4 | 6950 | flush_dcache_page(page); |
a52d9a80 | 6951 | } |
d1310b2e | 6952 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6953 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6954 | goto insert; |
a52d9a80 CM |
6955 | } |
6956 | not_found: | |
6957 | em->start = start; | |
70c8a91c | 6958 | em->orig_start = start; |
d1310b2e | 6959 | em->len = len; |
a52d9a80 | 6960 | not_found_em: |
5f39d397 | 6961 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 | 6962 | insert: |
b3b4aa74 | 6963 | btrfs_release_path(path); |
d1310b2e | 6964 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 6965 | btrfs_err(fs_info, |
5d163e0e JM |
6966 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
6967 | em->start, em->len, start, len); | |
a52d9a80 CM |
6968 | err = -EIO; |
6969 | goto out; | |
6970 | } | |
d1310b2e CM |
6971 | |
6972 | err = 0; | |
890871be | 6973 | write_lock(&em_tree->lock); |
f46b24c9 | 6974 | err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len); |
890871be | 6975 | write_unlock(&em_tree->lock); |
a52d9a80 | 6976 | out: |
c6414280 | 6977 | btrfs_free_path(path); |
1abe9b8a | 6978 | |
fc4f21b1 | 6979 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 6980 | |
a52d9a80 CM |
6981 | if (err) { |
6982 | free_extent_map(em); | |
a52d9a80 CM |
6983 | return ERR_PTR(err); |
6984 | } | |
79787eaa | 6985 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
6986 | return em; |
6987 | } | |
6988 | ||
fc4f21b1 NB |
6989 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
6990 | struct page *page, | |
6991 | size_t pg_offset, u64 start, u64 len, | |
6992 | int create) | |
ec29ed5b CM |
6993 | { |
6994 | struct extent_map *em; | |
6995 | struct extent_map *hole_em = NULL; | |
6996 | u64 range_start = start; | |
6997 | u64 end; | |
6998 | u64 found; | |
6999 | u64 found_end; | |
7000 | int err = 0; | |
7001 | ||
7002 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
7003 | if (IS_ERR(em)) | |
7004 | return em; | |
9986277e DC |
7005 | /* |
7006 | * If our em maps to: | |
7007 | * - a hole or | |
7008 | * - a pre-alloc extent, | |
7009 | * there might actually be delalloc bytes behind it. | |
7010 | */ | |
7011 | if (em->block_start != EXTENT_MAP_HOLE && | |
7012 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7013 | return em; | |
7014 | else | |
7015 | hole_em = em; | |
ec29ed5b CM |
7016 | |
7017 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7018 | end = start + len; | |
7019 | if (end < start) | |
7020 | end = (u64)-1; | |
7021 | else | |
7022 | end -= 1; | |
7023 | ||
7024 | em = NULL; | |
7025 | ||
7026 | /* ok, we didn't find anything, lets look for delalloc */ | |
fc4f21b1 | 7027 | found = count_range_bits(&inode->io_tree, &range_start, |
ec29ed5b CM |
7028 | end, len, EXTENT_DELALLOC, 1); |
7029 | found_end = range_start + found; | |
7030 | if (found_end < range_start) | |
7031 | found_end = (u64)-1; | |
7032 | ||
7033 | /* | |
7034 | * we didn't find anything useful, return | |
7035 | * the original results from get_extent() | |
7036 | */ | |
7037 | if (range_start > end || found_end <= start) { | |
7038 | em = hole_em; | |
7039 | hole_em = NULL; | |
7040 | goto out; | |
7041 | } | |
7042 | ||
7043 | /* adjust the range_start to make sure it doesn't | |
7044 | * go backwards from the start they passed in | |
7045 | */ | |
67871254 | 7046 | range_start = max(start, range_start); |
ec29ed5b CM |
7047 | found = found_end - range_start; |
7048 | ||
7049 | if (found > 0) { | |
7050 | u64 hole_start = start; | |
7051 | u64 hole_len = len; | |
7052 | ||
172ddd60 | 7053 | em = alloc_extent_map(); |
ec29ed5b CM |
7054 | if (!em) { |
7055 | err = -ENOMEM; | |
7056 | goto out; | |
7057 | } | |
7058 | /* | |
7059 | * when btrfs_get_extent can't find anything it | |
7060 | * returns one huge hole | |
7061 | * | |
7062 | * make sure what it found really fits our range, and | |
7063 | * adjust to make sure it is based on the start from | |
7064 | * the caller | |
7065 | */ | |
7066 | if (hole_em) { | |
7067 | u64 calc_end = extent_map_end(hole_em); | |
7068 | ||
7069 | if (calc_end <= start || (hole_em->start > end)) { | |
7070 | free_extent_map(hole_em); | |
7071 | hole_em = NULL; | |
7072 | } else { | |
7073 | hole_start = max(hole_em->start, start); | |
7074 | hole_len = calc_end - hole_start; | |
7075 | } | |
7076 | } | |
7077 | em->bdev = NULL; | |
7078 | if (hole_em && range_start > hole_start) { | |
7079 | /* our hole starts before our delalloc, so we | |
7080 | * have to return just the parts of the hole | |
7081 | * that go until the delalloc starts | |
7082 | */ | |
7083 | em->len = min(hole_len, | |
7084 | range_start - hole_start); | |
7085 | em->start = hole_start; | |
7086 | em->orig_start = hole_start; | |
7087 | /* | |
7088 | * don't adjust block start at all, | |
7089 | * it is fixed at EXTENT_MAP_HOLE | |
7090 | */ | |
7091 | em->block_start = hole_em->block_start; | |
7092 | em->block_len = hole_len; | |
f9e4fb53 LB |
7093 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7094 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
7095 | } else { |
7096 | em->start = range_start; | |
7097 | em->len = found; | |
7098 | em->orig_start = range_start; | |
7099 | em->block_start = EXTENT_MAP_DELALLOC; | |
7100 | em->block_len = found; | |
7101 | } | |
bf8d32b9 | 7102 | } else { |
ec29ed5b CM |
7103 | return hole_em; |
7104 | } | |
7105 | out: | |
7106 | ||
7107 | free_extent_map(hole_em); | |
7108 | if (err) { | |
7109 | free_extent_map(em); | |
7110 | return ERR_PTR(err); | |
7111 | } | |
7112 | return em; | |
7113 | } | |
7114 | ||
5f9a8a51 FM |
7115 | static struct extent_map *btrfs_create_dio_extent(struct inode *inode, |
7116 | const u64 start, | |
7117 | const u64 len, | |
7118 | const u64 orig_start, | |
7119 | const u64 block_start, | |
7120 | const u64 block_len, | |
7121 | const u64 orig_block_len, | |
7122 | const u64 ram_bytes, | |
7123 | const int type) | |
7124 | { | |
7125 | struct extent_map *em = NULL; | |
7126 | int ret; | |
7127 | ||
5f9a8a51 | 7128 | if (type != BTRFS_ORDERED_NOCOW) { |
6f9994db LB |
7129 | em = create_io_em(inode, start, len, orig_start, |
7130 | block_start, block_len, orig_block_len, | |
7131 | ram_bytes, | |
7132 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
7133 | type); | |
5f9a8a51 FM |
7134 | if (IS_ERR(em)) |
7135 | goto out; | |
7136 | } | |
7137 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, | |
7138 | len, block_len, type); | |
7139 | if (ret) { | |
7140 | if (em) { | |
7141 | free_extent_map(em); | |
dcdbc059 | 7142 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
5f9a8a51 FM |
7143 | start + len - 1, 0); |
7144 | } | |
7145 | em = ERR_PTR(ret); | |
7146 | } | |
7147 | out: | |
5f9a8a51 FM |
7148 | |
7149 | return em; | |
7150 | } | |
7151 | ||
4b46fce2 JB |
7152 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7153 | u64 start, u64 len) | |
7154 | { | |
0b246afa | 7155 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7156 | struct btrfs_root *root = BTRFS_I(inode)->root; |
70c8a91c | 7157 | struct extent_map *em; |
4b46fce2 JB |
7158 | struct btrfs_key ins; |
7159 | u64 alloc_hint; | |
7160 | int ret; | |
4b46fce2 | 7161 | |
4b46fce2 | 7162 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7163 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7164 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7165 | if (ret) |
7166 | return ERR_PTR(ret); | |
4b46fce2 | 7167 | |
5f9a8a51 FM |
7168 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
7169 | ins.objectid, ins.offset, ins.offset, | |
6288d6ea | 7170 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7171 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7172 | if (IS_ERR(em)) |
2ff7e61e JM |
7173 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
7174 | ins.offset, 1); | |
de0ee0ed | 7175 | |
4b46fce2 JB |
7176 | return em; |
7177 | } | |
7178 | ||
46bfbb5c CM |
7179 | /* |
7180 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7181 | * block must be cow'd | |
7182 | */ | |
00361589 | 7183 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7184 | u64 *orig_start, u64 *orig_block_len, |
7185 | u64 *ram_bytes) | |
46bfbb5c | 7186 | { |
2ff7e61e | 7187 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
7188 | struct btrfs_path *path; |
7189 | int ret; | |
7190 | struct extent_buffer *leaf; | |
7191 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7192 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7193 | struct btrfs_file_extent_item *fi; |
7194 | struct btrfs_key key; | |
7195 | u64 disk_bytenr; | |
7196 | u64 backref_offset; | |
7197 | u64 extent_end; | |
7198 | u64 num_bytes; | |
7199 | int slot; | |
7200 | int found_type; | |
7ee9e440 | 7201 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7202 | |
46bfbb5c CM |
7203 | path = btrfs_alloc_path(); |
7204 | if (!path) | |
7205 | return -ENOMEM; | |
7206 | ||
f85b7379 DS |
7207 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7208 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7209 | if (ret < 0) |
7210 | goto out; | |
7211 | ||
7212 | slot = path->slots[0]; | |
7213 | if (ret == 1) { | |
7214 | if (slot == 0) { | |
7215 | /* can't find the item, must cow */ | |
7216 | ret = 0; | |
7217 | goto out; | |
7218 | } | |
7219 | slot--; | |
7220 | } | |
7221 | ret = 0; | |
7222 | leaf = path->nodes[0]; | |
7223 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7224 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7225 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7226 | /* not our file or wrong item type, must cow */ | |
7227 | goto out; | |
7228 | } | |
7229 | ||
7230 | if (key.offset > offset) { | |
7231 | /* Wrong offset, must cow */ | |
7232 | goto out; | |
7233 | } | |
7234 | ||
7235 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7236 | found_type = btrfs_file_extent_type(leaf, fi); | |
7237 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7238 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7239 | /* not a regular extent, must cow */ | |
7240 | goto out; | |
7241 | } | |
7ee9e440 JB |
7242 | |
7243 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7244 | goto out; | |
7245 | ||
e77751aa MX |
7246 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7247 | if (extent_end <= offset) | |
7248 | goto out; | |
7249 | ||
46bfbb5c | 7250 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7251 | if (disk_bytenr == 0) |
7252 | goto out; | |
7253 | ||
7254 | if (btrfs_file_extent_compression(leaf, fi) || | |
7255 | btrfs_file_extent_encryption(leaf, fi) || | |
7256 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7257 | goto out; | |
7258 | ||
78d4295b EL |
7259 | /* |
7260 | * Do the same check as in btrfs_cross_ref_exist but without the | |
7261 | * unnecessary search. | |
7262 | */ | |
7263 | if (btrfs_file_extent_generation(leaf, fi) <= | |
7264 | btrfs_root_last_snapshot(&root->root_item)) | |
7265 | goto out; | |
7266 | ||
46bfbb5c CM |
7267 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7268 | ||
7ee9e440 JB |
7269 | if (orig_start) { |
7270 | *orig_start = key.offset - backref_offset; | |
7271 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7272 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7273 | } | |
eb384b55 | 7274 | |
2ff7e61e | 7275 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7276 | goto out; |
7b2b7085 MX |
7277 | |
7278 | num_bytes = min(offset + *len, extent_end) - offset; | |
7279 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7280 | u64 range_end; | |
7281 | ||
da17066c JM |
7282 | range_end = round_up(offset + num_bytes, |
7283 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7284 | ret = test_range_bit(io_tree, offset, range_end, |
7285 | EXTENT_DELALLOC, 0, NULL); | |
7286 | if (ret) { | |
7287 | ret = -EAGAIN; | |
7288 | goto out; | |
7289 | } | |
7290 | } | |
7291 | ||
1bda19eb | 7292 | btrfs_release_path(path); |
46bfbb5c CM |
7293 | |
7294 | /* | |
7295 | * look for other files referencing this extent, if we | |
7296 | * find any we must cow | |
7297 | */ | |
00361589 | 7298 | |
e4c3b2dc | 7299 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
00361589 | 7300 | key.offset - backref_offset, disk_bytenr); |
00361589 JB |
7301 | if (ret) { |
7302 | ret = 0; | |
7303 | goto out; | |
7304 | } | |
46bfbb5c CM |
7305 | |
7306 | /* | |
7307 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7308 | * in this extent we are about to write. If there | |
7309 | * are any csums in that range we have to cow in order | |
7310 | * to keep the csums correct | |
7311 | */ | |
7312 | disk_bytenr += backref_offset; | |
7313 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7314 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7315 | goto out; | |
46bfbb5c CM |
7316 | /* |
7317 | * all of the above have passed, it is safe to overwrite this extent | |
7318 | * without cow | |
7319 | */ | |
eb384b55 | 7320 | *len = num_bytes; |
46bfbb5c CM |
7321 | ret = 1; |
7322 | out: | |
7323 | btrfs_free_path(path); | |
7324 | return ret; | |
7325 | } | |
7326 | ||
eb838e73 JB |
7327 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7328 | struct extent_state **cached_state, int writing) | |
7329 | { | |
7330 | struct btrfs_ordered_extent *ordered; | |
7331 | int ret = 0; | |
7332 | ||
7333 | while (1) { | |
7334 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7335 | cached_state); |
eb838e73 JB |
7336 | /* |
7337 | * We're concerned with the entire range that we're going to be | |
01327610 | 7338 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7339 | * extents in this range. |
7340 | */ | |
a776c6fa | 7341 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7342 | lockend - lockstart + 1); |
7343 | ||
7344 | /* | |
7345 | * We need to make sure there are no buffered pages in this | |
7346 | * range either, we could have raced between the invalidate in | |
7347 | * generic_file_direct_write and locking the extent. The | |
7348 | * invalidate needs to happen so that reads after a write do not | |
7349 | * get stale data. | |
7350 | */ | |
fc4adbff | 7351 | if (!ordered && |
051c98eb DS |
7352 | (!writing || !filemap_range_has_page(inode->i_mapping, |
7353 | lockstart, lockend))) | |
eb838e73 JB |
7354 | break; |
7355 | ||
7356 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
e43bbe5e | 7357 | cached_state); |
eb838e73 JB |
7358 | |
7359 | if (ordered) { | |
ade77029 FM |
7360 | /* |
7361 | * If we are doing a DIO read and the ordered extent we | |
7362 | * found is for a buffered write, we can not wait for it | |
7363 | * to complete and retry, because if we do so we can | |
7364 | * deadlock with concurrent buffered writes on page | |
7365 | * locks. This happens only if our DIO read covers more | |
7366 | * than one extent map, if at this point has already | |
7367 | * created an ordered extent for a previous extent map | |
7368 | * and locked its range in the inode's io tree, and a | |
7369 | * concurrent write against that previous extent map's | |
7370 | * range and this range started (we unlock the ranges | |
7371 | * in the io tree only when the bios complete and | |
7372 | * buffered writes always lock pages before attempting | |
7373 | * to lock range in the io tree). | |
7374 | */ | |
7375 | if (writing || | |
7376 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7377 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7378 | else | |
7379 | ret = -ENOTBLK; | |
eb838e73 JB |
7380 | btrfs_put_ordered_extent(ordered); |
7381 | } else { | |
eb838e73 | 7382 | /* |
b850ae14 FM |
7383 | * We could trigger writeback for this range (and wait |
7384 | * for it to complete) and then invalidate the pages for | |
7385 | * this range (through invalidate_inode_pages2_range()), | |
7386 | * but that can lead us to a deadlock with a concurrent | |
7387 | * call to readpages() (a buffered read or a defrag call | |
7388 | * triggered a readahead) on a page lock due to an | |
7389 | * ordered dio extent we created before but did not have | |
7390 | * yet a corresponding bio submitted (whence it can not | |
7391 | * complete), which makes readpages() wait for that | |
7392 | * ordered extent to complete while holding a lock on | |
7393 | * that page. | |
eb838e73 | 7394 | */ |
b850ae14 | 7395 | ret = -ENOTBLK; |
eb838e73 JB |
7396 | } |
7397 | ||
ade77029 FM |
7398 | if (ret) |
7399 | break; | |
7400 | ||
eb838e73 JB |
7401 | cond_resched(); |
7402 | } | |
7403 | ||
7404 | return ret; | |
7405 | } | |
7406 | ||
6f9994db LB |
7407 | /* The callers of this must take lock_extent() */ |
7408 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, | |
7409 | u64 orig_start, u64 block_start, | |
7410 | u64 block_len, u64 orig_block_len, | |
7411 | u64 ram_bytes, int compress_type, | |
7412 | int type) | |
69ffb543 JB |
7413 | { |
7414 | struct extent_map_tree *em_tree; | |
7415 | struct extent_map *em; | |
7416 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7417 | int ret; | |
7418 | ||
6f9994db LB |
7419 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7420 | type == BTRFS_ORDERED_COMPRESSED || | |
7421 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7422 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7423 | |
69ffb543 JB |
7424 | em_tree = &BTRFS_I(inode)->extent_tree; |
7425 | em = alloc_extent_map(); | |
7426 | if (!em) | |
7427 | return ERR_PTR(-ENOMEM); | |
7428 | ||
7429 | em->start = start; | |
7430 | em->orig_start = orig_start; | |
7431 | em->len = len; | |
7432 | em->block_len = block_len; | |
7433 | em->block_start = block_start; | |
7434 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7435 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7436 | em->ram_bytes = ram_bytes; |
70c8a91c | 7437 | em->generation = -1; |
69ffb543 | 7438 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7439 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7440 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7441 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7442 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7443 | em->compress_type = compress_type; | |
7444 | } | |
69ffb543 JB |
7445 | |
7446 | do { | |
dcdbc059 | 7447 | btrfs_drop_extent_cache(BTRFS_I(inode), em->start, |
69ffb543 JB |
7448 | em->start + em->len - 1, 0); |
7449 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7450 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7451 | write_unlock(&em_tree->lock); |
6f9994db LB |
7452 | /* |
7453 | * The caller has taken lock_extent(), who could race with us | |
7454 | * to add em? | |
7455 | */ | |
69ffb543 JB |
7456 | } while (ret == -EEXIST); |
7457 | ||
7458 | if (ret) { | |
7459 | free_extent_map(em); | |
7460 | return ERR_PTR(ret); | |
7461 | } | |
7462 | ||
6f9994db | 7463 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7464 | return em; |
7465 | } | |
7466 | ||
1c8d0175 NB |
7467 | |
7468 | static int btrfs_get_blocks_direct_read(struct extent_map *em, | |
7469 | struct buffer_head *bh_result, | |
7470 | struct inode *inode, | |
7471 | u64 start, u64 len) | |
7472 | { | |
7473 | if (em->block_start == EXTENT_MAP_HOLE || | |
7474 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7475 | return -ENOENT; | |
7476 | ||
7477 | len = min(len, em->len - (start - em->start)); | |
7478 | ||
7479 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
7480 | inode->i_blkbits; | |
7481 | bh_result->b_size = len; | |
7482 | bh_result->b_bdev = em->bdev; | |
7483 | set_buffer_mapped(bh_result); | |
7484 | ||
7485 | return 0; | |
7486 | } | |
7487 | ||
c5794e51 NB |
7488 | static int btrfs_get_blocks_direct_write(struct extent_map **map, |
7489 | struct buffer_head *bh_result, | |
7490 | struct inode *inode, | |
7491 | struct btrfs_dio_data *dio_data, | |
7492 | u64 start, u64 len) | |
7493 | { | |
7494 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
7495 | struct extent_map *em = *map; | |
7496 | int ret = 0; | |
7497 | ||
7498 | /* | |
7499 | * We don't allocate a new extent in the following cases | |
7500 | * | |
7501 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7502 | * existing extent. | |
7503 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7504 | * just use the extent. | |
7505 | * | |
7506 | */ | |
7507 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7508 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7509 | em->block_start != EXTENT_MAP_HOLE)) { | |
7510 | int type; | |
7511 | u64 block_start, orig_start, orig_block_len, ram_bytes; | |
7512 | ||
7513 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7514 | type = BTRFS_ORDERED_PREALLOC; | |
7515 | else | |
7516 | type = BTRFS_ORDERED_NOCOW; | |
7517 | len = min(len, em->len - (start - em->start)); | |
7518 | block_start = em->block_start + (start - em->start); | |
7519 | ||
7520 | if (can_nocow_extent(inode, start, &len, &orig_start, | |
7521 | &orig_block_len, &ram_bytes) == 1 && | |
7522 | btrfs_inc_nocow_writers(fs_info, block_start)) { | |
7523 | struct extent_map *em2; | |
7524 | ||
7525 | em2 = btrfs_create_dio_extent(inode, start, len, | |
7526 | orig_start, block_start, | |
7527 | len, orig_block_len, | |
7528 | ram_bytes, type); | |
7529 | btrfs_dec_nocow_writers(fs_info, block_start); | |
7530 | if (type == BTRFS_ORDERED_PREALLOC) { | |
7531 | free_extent_map(em); | |
7532 | *map = em = em2; | |
7533 | } | |
7534 | ||
7535 | if (em2 && IS_ERR(em2)) { | |
7536 | ret = PTR_ERR(em2); | |
7537 | goto out; | |
7538 | } | |
7539 | /* | |
7540 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7541 | * use the existing or preallocated extent, so does not | |
7542 | * need to adjust btrfs_space_info's bytes_may_use. | |
7543 | */ | |
7544 | btrfs_free_reserved_data_space_noquota(inode, start, | |
7545 | len); | |
7546 | goto skip_cow; | |
7547 | } | |
7548 | } | |
7549 | ||
7550 | /* this will cow the extent */ | |
7551 | len = bh_result->b_size; | |
7552 | free_extent_map(em); | |
7553 | *map = em = btrfs_new_extent_direct(inode, start, len); | |
7554 | if (IS_ERR(em)) { | |
7555 | ret = PTR_ERR(em); | |
7556 | goto out; | |
7557 | } | |
7558 | ||
7559 | len = min(len, em->len - (start - em->start)); | |
7560 | ||
7561 | skip_cow: | |
7562 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
7563 | inode->i_blkbits; | |
7564 | bh_result->b_size = len; | |
7565 | bh_result->b_bdev = em->bdev; | |
7566 | set_buffer_mapped(bh_result); | |
7567 | ||
7568 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7569 | set_buffer_new(bh_result); | |
7570 | ||
7571 | /* | |
7572 | * Need to update the i_size under the extent lock so buffered | |
7573 | * readers will get the updated i_size when we unlock. | |
7574 | */ | |
7575 | if (!dio_data->overwrite && start + len > i_size_read(inode)) | |
7576 | i_size_write(inode, start + len); | |
7577 | ||
7578 | WARN_ON(dio_data->reserve < len); | |
7579 | dio_data->reserve -= len; | |
7580 | dio_data->unsubmitted_oe_range_end = start + len; | |
7581 | current->journal_info = dio_data; | |
7582 | out: | |
7583 | return ret; | |
7584 | } | |
7585 | ||
4b46fce2 JB |
7586 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7587 | struct buffer_head *bh_result, int create) | |
7588 | { | |
0b246afa | 7589 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7590 | struct extent_map *em; |
eb838e73 | 7591 | struct extent_state *cached_state = NULL; |
50745b0a | 7592 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7593 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7594 | u64 lockstart, lockend; |
4b46fce2 | 7595 | u64 len = bh_result->b_size; |
eb838e73 | 7596 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7597 | int ret = 0; |
eb838e73 | 7598 | |
172a5049 | 7599 | if (create) |
3266789f | 7600 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7601 | else |
0b246afa | 7602 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7603 | |
c329861d JB |
7604 | lockstart = start; |
7605 | lockend = start + len - 1; | |
7606 | ||
e1cbbfa5 JB |
7607 | if (current->journal_info) { |
7608 | /* | |
7609 | * Need to pull our outstanding extents and set journal_info to NULL so | |
01327610 | 7610 | * that anything that needs to check if there's a transaction doesn't get |
e1cbbfa5 JB |
7611 | * confused. |
7612 | */ | |
50745b0a | 7613 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7614 | current->journal_info = NULL; |
7615 | } | |
7616 | ||
eb838e73 JB |
7617 | /* |
7618 | * If this errors out it's because we couldn't invalidate pagecache for | |
7619 | * this range and we need to fallback to buffered. | |
7620 | */ | |
9c9464cc FM |
7621 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7622 | create)) { | |
7623 | ret = -ENOTBLK; | |
7624 | goto err; | |
7625 | } | |
eb838e73 | 7626 | |
fc4f21b1 | 7627 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); |
eb838e73 JB |
7628 | if (IS_ERR(em)) { |
7629 | ret = PTR_ERR(em); | |
7630 | goto unlock_err; | |
7631 | } | |
4b46fce2 JB |
7632 | |
7633 | /* | |
7634 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7635 | * io. INLINE is special, and we could probably kludge it in here, but | |
7636 | * it's still buffered so for safety lets just fall back to the generic | |
7637 | * buffered path. | |
7638 | * | |
7639 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7640 | * decompress it, so there will be buffering required no matter what we | |
7641 | * do, so go ahead and fallback to buffered. | |
7642 | * | |
01327610 | 7643 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7644 | * to buffered IO. Don't blame me, this is the price we pay for using |
7645 | * the generic code. | |
7646 | */ | |
7647 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7648 | em->block_start == EXTENT_MAP_INLINE) { | |
7649 | free_extent_map(em); | |
eb838e73 JB |
7650 | ret = -ENOTBLK; |
7651 | goto unlock_err; | |
4b46fce2 JB |
7652 | } |
7653 | ||
c5794e51 NB |
7654 | if (create) { |
7655 | ret = btrfs_get_blocks_direct_write(&em, bh_result, inode, | |
7656 | dio_data, start, len); | |
7657 | if (ret < 0) | |
7658 | goto unlock_err; | |
7659 | ||
7660 | /* clear and unlock the entire range */ | |
7661 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7662 | unlock_bits, 1, 0, &cached_state); | |
7663 | } else { | |
1c8d0175 NB |
7664 | ret = btrfs_get_blocks_direct_read(em, bh_result, inode, |
7665 | start, len); | |
7666 | /* Can be negative only if we read from a hole */ | |
7667 | if (ret < 0) { | |
7668 | ret = 0; | |
7669 | free_extent_map(em); | |
7670 | goto unlock_err; | |
7671 | } | |
7672 | /* | |
7673 | * We need to unlock only the end area that we aren't using. | |
7674 | * The rest is going to be unlocked by the endio routine. | |
7675 | */ | |
7676 | lockstart = start + bh_result->b_size; | |
7677 | if (lockstart < lockend) { | |
7678 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
7679 | lockend, unlock_bits, 1, 0, | |
7680 | &cached_state); | |
7681 | } else { | |
7682 | free_extent_state(cached_state); | |
7683 | } | |
4b46fce2 JB |
7684 | } |
7685 | ||
4b46fce2 JB |
7686 | free_extent_map(em); |
7687 | ||
7688 | return 0; | |
eb838e73 JB |
7689 | |
7690 | unlock_err: | |
eb838e73 | 7691 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
ae0f1625 | 7692 | unlock_bits, 1, 0, &cached_state); |
9c9464cc | 7693 | err: |
50745b0a | 7694 | if (dio_data) |
7695 | current->journal_info = dio_data; | |
eb838e73 | 7696 | return ret; |
4b46fce2 JB |
7697 | } |
7698 | ||
58efbc9f OS |
7699 | static inline blk_status_t submit_dio_repair_bio(struct inode *inode, |
7700 | struct bio *bio, | |
7701 | int mirror_num) | |
8b110e39 | 7702 | { |
2ff7e61e | 7703 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
58efbc9f | 7704 | blk_status_t ret; |
8b110e39 | 7705 | |
37226b21 | 7706 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 | 7707 | |
2ff7e61e | 7708 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR); |
8b110e39 | 7709 | if (ret) |
ea057f6d | 7710 | return ret; |
8b110e39 | 7711 | |
2ff7e61e | 7712 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
ea057f6d | 7713 | |
8b110e39 MX |
7714 | return ret; |
7715 | } | |
7716 | ||
7717 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7718 | struct bio *failed_bio, | |
7719 | struct io_failure_record *failrec, | |
7720 | int failed_mirror) | |
7721 | { | |
ab8d0fc4 | 7722 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 MX |
7723 | int num_copies; |
7724 | ||
ab8d0fc4 | 7725 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
8b110e39 MX |
7726 | if (num_copies == 1) { |
7727 | /* | |
7728 | * we only have a single copy of the data, so don't bother with | |
7729 | * all the retry and error correction code that follows. no | |
7730 | * matter what the error is, it is very likely to persist. | |
7731 | */ | |
ab8d0fc4 JM |
7732 | btrfs_debug(fs_info, |
7733 | "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
7734 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7735 | return 0; |
7736 | } | |
7737 | ||
7738 | failrec->failed_mirror = failed_mirror; | |
7739 | failrec->this_mirror++; | |
7740 | if (failrec->this_mirror == failed_mirror) | |
7741 | failrec->this_mirror++; | |
7742 | ||
7743 | if (failrec->this_mirror > num_copies) { | |
ab8d0fc4 JM |
7744 | btrfs_debug(fs_info, |
7745 | "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
7746 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7747 | return 0; |
7748 | } | |
7749 | ||
7750 | return 1; | |
7751 | } | |
7752 | ||
58efbc9f OS |
7753 | static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio, |
7754 | struct page *page, unsigned int pgoff, | |
7755 | u64 start, u64 end, int failed_mirror, | |
7756 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
7757 | { |
7758 | struct io_failure_record *failrec; | |
7870d082 JB |
7759 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
7760 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8b110e39 MX |
7761 | struct bio *bio; |
7762 | int isector; | |
f1c77c55 | 7763 | unsigned int read_mode = 0; |
17347cec | 7764 | int segs; |
8b110e39 | 7765 | int ret; |
58efbc9f | 7766 | blk_status_t status; |
c16a8ac3 | 7767 | struct bio_vec bvec; |
8b110e39 | 7768 | |
37226b21 | 7769 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
8b110e39 MX |
7770 | |
7771 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7772 | if (ret) | |
58efbc9f | 7773 | return errno_to_blk_status(ret); |
8b110e39 MX |
7774 | |
7775 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7776 | failed_mirror); | |
7777 | if (!ret) { | |
7870d082 | 7778 | free_io_failure(failure_tree, io_tree, failrec); |
58efbc9f | 7779 | return BLK_STS_IOERR; |
8b110e39 MX |
7780 | } |
7781 | ||
17347cec | 7782 | segs = bio_segments(failed_bio); |
c16a8ac3 | 7783 | bio_get_first_bvec(failed_bio, &bvec); |
17347cec | 7784 | if (segs > 1 || |
c16a8ac3 | 7785 | (bvec.bv_len > btrfs_inode_sectorsize(inode))) |
70fd7614 | 7786 | read_mode |= REQ_FAILFAST_DEV; |
8b110e39 MX |
7787 | |
7788 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7789 | isector >>= inode->i_sb->s_blocksize_bits; | |
7790 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 7791 | pgoff, isector, repair_endio, repair_arg); |
ebcc3263 | 7792 | bio->bi_opf = REQ_OP_READ | read_mode; |
8b110e39 MX |
7793 | |
7794 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
913e1535 | 7795 | "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d", |
8b110e39 MX |
7796 | read_mode, failrec->this_mirror, failrec->in_validation); |
7797 | ||
58efbc9f OS |
7798 | status = submit_dio_repair_bio(inode, bio, failrec->this_mirror); |
7799 | if (status) { | |
7870d082 | 7800 | free_io_failure(failure_tree, io_tree, failrec); |
8b110e39 MX |
7801 | bio_put(bio); |
7802 | } | |
7803 | ||
58efbc9f | 7804 | return status; |
8b110e39 MX |
7805 | } |
7806 | ||
7807 | struct btrfs_retry_complete { | |
7808 | struct completion done; | |
7809 | struct inode *inode; | |
7810 | u64 start; | |
7811 | int uptodate; | |
7812 | }; | |
7813 | ||
4246a0b6 | 7814 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
7815 | { |
7816 | struct btrfs_retry_complete *done = bio->bi_private; | |
7870d082 | 7817 | struct inode *inode = done->inode; |
8b110e39 | 7818 | struct bio_vec *bvec; |
7870d082 | 7819 | struct extent_io_tree *io_tree, *failure_tree; |
8b110e39 MX |
7820 | int i; |
7821 | ||
4e4cbee9 | 7822 | if (bio->bi_status) |
8b110e39 MX |
7823 | goto end; |
7824 | ||
2dabb324 | 7825 | ASSERT(bio->bi_vcnt == 1); |
7870d082 JB |
7826 | io_tree = &BTRFS_I(inode)->io_tree; |
7827 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
263663cd | 7828 | ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode)); |
2dabb324 | 7829 | |
8b110e39 | 7830 | done->uptodate = 1; |
c09abff8 | 7831 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
8b110e39 | 7832 | bio_for_each_segment_all(bvec, bio, i) |
7870d082 JB |
7833 | clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree, |
7834 | io_tree, done->start, bvec->bv_page, | |
7835 | btrfs_ino(BTRFS_I(inode)), 0); | |
8b110e39 MX |
7836 | end: |
7837 | complete(&done->done); | |
7838 | bio_put(bio); | |
7839 | } | |
7840 | ||
58efbc9f OS |
7841 | static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode, |
7842 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 7843 | { |
2dabb324 | 7844 | struct btrfs_fs_info *fs_info; |
17347cec LB |
7845 | struct bio_vec bvec; |
7846 | struct bvec_iter iter; | |
8b110e39 | 7847 | struct btrfs_retry_complete done; |
4b46fce2 | 7848 | u64 start; |
2dabb324 CR |
7849 | unsigned int pgoff; |
7850 | u32 sectorsize; | |
7851 | int nr_sectors; | |
58efbc9f OS |
7852 | blk_status_t ret; |
7853 | blk_status_t err = BLK_STS_OK; | |
4b46fce2 | 7854 | |
2dabb324 | 7855 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 7856 | sectorsize = fs_info->sectorsize; |
2dabb324 | 7857 | |
8b110e39 MX |
7858 | start = io_bio->logical; |
7859 | done.inode = inode; | |
17347cec | 7860 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 7861 | |
17347cec LB |
7862 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
7863 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
7864 | pgoff = bvec.bv_offset; | |
2dabb324 CR |
7865 | |
7866 | next_block_or_try_again: | |
8b110e39 MX |
7867 | done.uptodate = 0; |
7868 | done.start = start; | |
7869 | init_completion(&done.done); | |
7870 | ||
17347cec | 7871 | ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
2dabb324 CR |
7872 | pgoff, start, start + sectorsize - 1, |
7873 | io_bio->mirror_num, | |
7874 | btrfs_retry_endio_nocsum, &done); | |
629ebf4f LB |
7875 | if (ret) { |
7876 | err = ret; | |
7877 | goto next; | |
7878 | } | |
8b110e39 | 7879 | |
9c17f6cd | 7880 | wait_for_completion_io(&done.done); |
8b110e39 MX |
7881 | |
7882 | if (!done.uptodate) { | |
7883 | /* We might have another mirror, so try again */ | |
2dabb324 | 7884 | goto next_block_or_try_again; |
8b110e39 MX |
7885 | } |
7886 | ||
629ebf4f | 7887 | next: |
2dabb324 CR |
7888 | start += sectorsize; |
7889 | ||
97bf5a55 LB |
7890 | nr_sectors--; |
7891 | if (nr_sectors) { | |
2dabb324 | 7892 | pgoff += sectorsize; |
97bf5a55 | 7893 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
7894 | goto next_block_or_try_again; |
7895 | } | |
8b110e39 MX |
7896 | } |
7897 | ||
629ebf4f | 7898 | return err; |
8b110e39 MX |
7899 | } |
7900 | ||
4246a0b6 | 7901 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
7902 | { |
7903 | struct btrfs_retry_complete *done = bio->bi_private; | |
7904 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7870d082 JB |
7905 | struct extent_io_tree *io_tree, *failure_tree; |
7906 | struct inode *inode = done->inode; | |
8b110e39 MX |
7907 | struct bio_vec *bvec; |
7908 | int uptodate; | |
7909 | int ret; | |
7910 | int i; | |
7911 | ||
4e4cbee9 | 7912 | if (bio->bi_status) |
8b110e39 MX |
7913 | goto end; |
7914 | ||
7915 | uptodate = 1; | |
2dabb324 | 7916 | |
2dabb324 | 7917 | ASSERT(bio->bi_vcnt == 1); |
263663cd | 7918 | ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode)); |
2dabb324 | 7919 | |
7870d082 JB |
7920 | io_tree = &BTRFS_I(inode)->io_tree; |
7921 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
7922 | ||
c09abff8 | 7923 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
8b110e39 | 7924 | bio_for_each_segment_all(bvec, bio, i) { |
7870d082 JB |
7925 | ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page, |
7926 | bvec->bv_offset, done->start, | |
7927 | bvec->bv_len); | |
8b110e39 | 7928 | if (!ret) |
7870d082 JB |
7929 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
7930 | failure_tree, io_tree, done->start, | |
7931 | bvec->bv_page, | |
7932 | btrfs_ino(BTRFS_I(inode)), | |
7933 | bvec->bv_offset); | |
8b110e39 MX |
7934 | else |
7935 | uptodate = 0; | |
7936 | } | |
7937 | ||
7938 | done->uptodate = uptodate; | |
7939 | end: | |
7940 | complete(&done->done); | |
7941 | bio_put(bio); | |
7942 | } | |
7943 | ||
4e4cbee9 CH |
7944 | static blk_status_t __btrfs_subio_endio_read(struct inode *inode, |
7945 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 | 7946 | { |
2dabb324 | 7947 | struct btrfs_fs_info *fs_info; |
17347cec LB |
7948 | struct bio_vec bvec; |
7949 | struct bvec_iter iter; | |
8b110e39 MX |
7950 | struct btrfs_retry_complete done; |
7951 | u64 start; | |
7952 | u64 offset = 0; | |
2dabb324 CR |
7953 | u32 sectorsize; |
7954 | int nr_sectors; | |
7955 | unsigned int pgoff; | |
7956 | int csum_pos; | |
ef7cdac1 | 7957 | bool uptodate = (err == 0); |
8b110e39 | 7958 | int ret; |
58efbc9f | 7959 | blk_status_t status; |
dc380aea | 7960 | |
2dabb324 | 7961 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 7962 | sectorsize = fs_info->sectorsize; |
2dabb324 | 7963 | |
58efbc9f | 7964 | err = BLK_STS_OK; |
c1dc0896 | 7965 | start = io_bio->logical; |
8b110e39 | 7966 | done.inode = inode; |
17347cec | 7967 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 7968 | |
17347cec LB |
7969 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
7970 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
2dabb324 | 7971 | |
17347cec | 7972 | pgoff = bvec.bv_offset; |
2dabb324 | 7973 | next_block: |
ef7cdac1 LB |
7974 | if (uptodate) { |
7975 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
7976 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
7977 | bvec.bv_page, pgoff, start, sectorsize); | |
7978 | if (likely(!ret)) | |
7979 | goto next; | |
7980 | } | |
8b110e39 MX |
7981 | try_again: |
7982 | done.uptodate = 0; | |
7983 | done.start = start; | |
7984 | init_completion(&done.done); | |
7985 | ||
58efbc9f OS |
7986 | status = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
7987 | pgoff, start, start + sectorsize - 1, | |
7988 | io_bio->mirror_num, btrfs_retry_endio, | |
7989 | &done); | |
7990 | if (status) { | |
7991 | err = status; | |
8b110e39 MX |
7992 | goto next; |
7993 | } | |
7994 | ||
9c17f6cd | 7995 | wait_for_completion_io(&done.done); |
8b110e39 MX |
7996 | |
7997 | if (!done.uptodate) { | |
7998 | /* We might have another mirror, so try again */ | |
7999 | goto try_again; | |
8000 | } | |
8001 | next: | |
2dabb324 CR |
8002 | offset += sectorsize; |
8003 | start += sectorsize; | |
8004 | ||
8005 | ASSERT(nr_sectors); | |
8006 | ||
97bf5a55 LB |
8007 | nr_sectors--; |
8008 | if (nr_sectors) { | |
2dabb324 | 8009 | pgoff += sectorsize; |
97bf5a55 | 8010 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8011 | goto next_block; |
8012 | } | |
2c30c71b | 8013 | } |
c1dc0896 MX |
8014 | |
8015 | return err; | |
8016 | } | |
8017 | ||
4e4cbee9 CH |
8018 | static blk_status_t btrfs_subio_endio_read(struct inode *inode, |
8019 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 MX |
8020 | { |
8021 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8022 | ||
8023 | if (skip_csum) { | |
8024 | if (unlikely(err)) | |
8025 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8026 | else | |
58efbc9f | 8027 | return BLK_STS_OK; |
8b110e39 MX |
8028 | } else { |
8029 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8030 | } | |
8031 | } | |
8032 | ||
4246a0b6 | 8033 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8034 | { |
8035 | struct btrfs_dio_private *dip = bio->bi_private; | |
8036 | struct inode *inode = dip->inode; | |
8037 | struct bio *dio_bio; | |
8038 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4e4cbee9 | 8039 | blk_status_t err = bio->bi_status; |
c1dc0896 | 8040 | |
99c4e3b9 | 8041 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8b110e39 | 8042 | err = btrfs_subio_endio_read(inode, io_bio, err); |
c1dc0896 | 8043 | |
4b46fce2 | 8044 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8045 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8046 | dio_bio = dip->dio_bio; |
4b46fce2 | 8047 | |
4b46fce2 | 8048 | kfree(dip); |
c0da7aa1 | 8049 | |
99c4e3b9 | 8050 | dio_bio->bi_status = err; |
4055351c | 8051 | dio_end_io(dio_bio); |
b3a0dd50 | 8052 | btrfs_io_bio_free_csum(io_bio); |
9be3395b | 8053 | bio_put(bio); |
4b46fce2 JB |
8054 | } |
8055 | ||
52427260 QW |
8056 | static void __endio_write_update_ordered(struct inode *inode, |
8057 | const u64 offset, const u64 bytes, | |
8058 | const bool uptodate) | |
4b46fce2 | 8059 | { |
0b246afa | 8060 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 8061 | struct btrfs_ordered_extent *ordered = NULL; |
52427260 QW |
8062 | struct btrfs_workqueue *wq; |
8063 | btrfs_work_func_t func; | |
14543774 FM |
8064 | u64 ordered_offset = offset; |
8065 | u64 ordered_bytes = bytes; | |
67c003f9 | 8066 | u64 last_offset; |
4b46fce2 | 8067 | |
52427260 QW |
8068 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
8069 | wq = fs_info->endio_freespace_worker; | |
8070 | func = btrfs_freespace_write_helper; | |
8071 | } else { | |
8072 | wq = fs_info->endio_write_workers; | |
8073 | func = btrfs_endio_write_helper; | |
8074 | } | |
8075 | ||
b25f0d00 NB |
8076 | while (ordered_offset < offset + bytes) { |
8077 | last_offset = ordered_offset; | |
8078 | if (btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
8079 | &ordered_offset, | |
8080 | ordered_bytes, | |
8081 | uptodate)) { | |
8082 | btrfs_init_work(&ordered->work, func, | |
8083 | finish_ordered_fn, | |
8084 | NULL, NULL); | |
8085 | btrfs_queue_work(wq, &ordered->work); | |
8086 | } | |
8087 | /* | |
8088 | * If btrfs_dec_test_ordered_pending does not find any ordered | |
8089 | * extent in the range, we can exit. | |
8090 | */ | |
8091 | if (ordered_offset == last_offset) | |
8092 | return; | |
8093 | /* | |
8094 | * Our bio might span multiple ordered extents. In this case | |
8095 | * we keep goin until we have accounted the whole dio. | |
8096 | */ | |
8097 | if (ordered_offset < offset + bytes) { | |
8098 | ordered_bytes = offset + bytes - ordered_offset; | |
8099 | ordered = NULL; | |
8100 | } | |
163cf09c | 8101 | } |
14543774 FM |
8102 | } |
8103 | ||
8104 | static void btrfs_endio_direct_write(struct bio *bio) | |
8105 | { | |
8106 | struct btrfs_dio_private *dip = bio->bi_private; | |
8107 | struct bio *dio_bio = dip->dio_bio; | |
8108 | ||
52427260 | 8109 | __endio_write_update_ordered(dip->inode, dip->logical_offset, |
4e4cbee9 | 8110 | dip->bytes, !bio->bi_status); |
4b46fce2 | 8111 | |
4b46fce2 | 8112 | kfree(dip); |
c0da7aa1 | 8113 | |
4e4cbee9 | 8114 | dio_bio->bi_status = bio->bi_status; |
4055351c | 8115 | dio_end_io(dio_bio); |
9be3395b | 8116 | bio_put(bio); |
4b46fce2 JB |
8117 | } |
8118 | ||
d0ee3934 | 8119 | static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data, |
d0779291 | 8120 | struct bio *bio, u64 offset) |
eaf25d93 | 8121 | { |
c6100a4b | 8122 | struct inode *inode = private_data; |
4e4cbee9 | 8123 | blk_status_t ret; |
2ff7e61e | 8124 | ret = btrfs_csum_one_bio(inode, bio, offset, 1); |
79787eaa | 8125 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8126 | return 0; |
8127 | } | |
8128 | ||
4246a0b6 | 8129 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8130 | { |
8131 | struct btrfs_dio_private *dip = bio->bi_private; | |
4e4cbee9 | 8132 | blk_status_t err = bio->bi_status; |
e65e1535 | 8133 | |
8b110e39 MX |
8134 | if (err) |
8135 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 8136 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 DS |
8137 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
8138 | bio->bi_opf, | |
8b110e39 MX |
8139 | (unsigned long long)bio->bi_iter.bi_sector, |
8140 | bio->bi_iter.bi_size, err); | |
8141 | ||
8142 | if (dip->subio_endio) | |
8143 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8144 | |
8145 | if (err) { | |
e65e1535 | 8146 | /* |
de224b7c NB |
8147 | * We want to perceive the errors flag being set before |
8148 | * decrementing the reference count. We don't need a barrier | |
8149 | * since atomic operations with a return value are fully | |
8150 | * ordered as per atomic_t.txt | |
e65e1535 | 8151 | */ |
de224b7c | 8152 | dip->errors = 1; |
e65e1535 MX |
8153 | } |
8154 | ||
8155 | /* if there are more bios still pending for this dio, just exit */ | |
8156 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8157 | goto out; | |
8158 | ||
9be3395b | 8159 | if (dip->errors) { |
e65e1535 | 8160 | bio_io_error(dip->orig_bio); |
9be3395b | 8161 | } else { |
2dbe0c77 | 8162 | dip->dio_bio->bi_status = BLK_STS_OK; |
4246a0b6 | 8163 | bio_endio(dip->orig_bio); |
e65e1535 MX |
8164 | } |
8165 | out: | |
8166 | bio_put(bio); | |
8167 | } | |
8168 | ||
4e4cbee9 | 8169 | static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode, |
c1dc0896 MX |
8170 | struct btrfs_dio_private *dip, |
8171 | struct bio *bio, | |
8172 | u64 file_offset) | |
8173 | { | |
8174 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8175 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
4e4cbee9 | 8176 | blk_status_t ret; |
c1dc0896 MX |
8177 | |
8178 | /* | |
8179 | * We load all the csum data we need when we submit | |
8180 | * the first bio to reduce the csum tree search and | |
8181 | * contention. | |
8182 | */ | |
8183 | if (dip->logical_offset == file_offset) { | |
2ff7e61e | 8184 | ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio, |
c1dc0896 MX |
8185 | file_offset); |
8186 | if (ret) | |
8187 | return ret; | |
8188 | } | |
8189 | ||
8190 | if (bio == dip->orig_bio) | |
8191 | return 0; | |
8192 | ||
8193 | file_offset -= dip->logical_offset; | |
8194 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8195 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8196 | ||
8197 | return 0; | |
8198 | } | |
8199 | ||
d0ee3934 DS |
8200 | static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio, |
8201 | struct inode *inode, u64 file_offset, int async_submit) | |
e65e1535 | 8202 | { |
0b246afa | 8203 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 8204 | struct btrfs_dio_private *dip = bio->bi_private; |
37226b21 | 8205 | bool write = bio_op(bio) == REQ_OP_WRITE; |
4e4cbee9 | 8206 | blk_status_t ret; |
e65e1535 | 8207 | |
4c274bc6 | 8208 | /* Check btrfs_submit_bio_hook() for rules about async submit. */ |
b812ce28 JB |
8209 | if (async_submit) |
8210 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8211 | ||
5fd02043 | 8212 | if (!write) { |
0b246afa | 8213 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
8214 | if (ret) |
8215 | goto err; | |
8216 | } | |
e65e1535 | 8217 | |
e6961cac | 8218 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
8219 | goto map; |
8220 | ||
8221 | if (write && async_submit) { | |
c6100a4b JB |
8222 | ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0, |
8223 | file_offset, inode, | |
e288c080 | 8224 | btrfs_submit_bio_start_direct_io); |
e65e1535 | 8225 | goto err; |
1ae39938 JB |
8226 | } else if (write) { |
8227 | /* | |
8228 | * If we aren't doing async submit, calculate the csum of the | |
8229 | * bio now. | |
8230 | */ | |
2ff7e61e | 8231 | ret = btrfs_csum_one_bio(inode, bio, file_offset, 1); |
1ae39938 JB |
8232 | if (ret) |
8233 | goto err; | |
23ea8e5a | 8234 | } else { |
2ff7e61e | 8235 | ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio, |
c1dc0896 | 8236 | file_offset); |
c2db1073 TI |
8237 | if (ret) |
8238 | goto err; | |
8239 | } | |
1ae39938 | 8240 | map: |
9b4a9b28 | 8241 | ret = btrfs_map_bio(fs_info, bio, 0, 0); |
e65e1535 | 8242 | err: |
e65e1535 MX |
8243 | return ret; |
8244 | } | |
8245 | ||
e6961cac | 8246 | static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip) |
e65e1535 MX |
8247 | { |
8248 | struct inode *inode = dip->inode; | |
0b246afa | 8249 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e65e1535 MX |
8250 | struct bio *bio; |
8251 | struct bio *orig_bio = dip->orig_bio; | |
4f024f37 | 8252 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 | 8253 | u64 file_offset = dip->logical_offset; |
e65e1535 | 8254 | u64 map_length; |
1ae39938 | 8255 | int async_submit = 0; |
725130ba LB |
8256 | u64 submit_len; |
8257 | int clone_offset = 0; | |
8258 | int clone_len; | |
5f4dc8fc | 8259 | int ret; |
58efbc9f | 8260 | blk_status_t status; |
e65e1535 | 8261 | |
4f024f37 | 8262 | map_length = orig_bio->bi_iter.bi_size; |
725130ba | 8263 | submit_len = map_length; |
0b246afa JM |
8264 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9, |
8265 | &map_length, NULL, 0); | |
7a5c3c9b | 8266 | if (ret) |
e65e1535 | 8267 | return -EIO; |
facc8a22 | 8268 | |
725130ba | 8269 | if (map_length >= submit_len) { |
02f57c7a | 8270 | bio = orig_bio; |
c1dc0896 | 8271 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8272 | goto submit; |
8273 | } | |
8274 | ||
53b381b3 | 8275 | /* async crcs make it difficult to collect full stripe writes. */ |
1b86826d | 8276 | if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8277 | async_submit = 0; |
8278 | else | |
8279 | async_submit = 1; | |
8280 | ||
725130ba LB |
8281 | /* bio split */ |
8282 | ASSERT(map_length <= INT_MAX); | |
02f57c7a | 8283 | atomic_inc(&dip->pending_bios); |
3c91ee69 | 8284 | do { |
725130ba | 8285 | clone_len = min_t(int, submit_len, map_length); |
02f57c7a | 8286 | |
725130ba LB |
8287 | /* |
8288 | * This will never fail as it's passing GPF_NOFS and | |
8289 | * the allocation is backed by btrfs_bioset. | |
8290 | */ | |
e477094f | 8291 | bio = btrfs_bio_clone_partial(orig_bio, clone_offset, |
725130ba LB |
8292 | clone_len); |
8293 | bio->bi_private = dip; | |
8294 | bio->bi_end_io = btrfs_end_dio_bio; | |
8295 | btrfs_io_bio(bio)->logical = file_offset; | |
8296 | ||
8297 | ASSERT(submit_len >= clone_len); | |
8298 | submit_len -= clone_len; | |
8299 | if (submit_len == 0) | |
8300 | break; | |
e65e1535 | 8301 | |
725130ba LB |
8302 | /* |
8303 | * Increase the count before we submit the bio so we know | |
8304 | * the end IO handler won't happen before we increase the | |
8305 | * count. Otherwise, the dip might get freed before we're | |
8306 | * done setting it up. | |
8307 | */ | |
8308 | atomic_inc(&dip->pending_bios); | |
e65e1535 | 8309 | |
d0ee3934 | 8310 | status = btrfs_submit_dio_bio(bio, inode, file_offset, |
58efbc9f OS |
8311 | async_submit); |
8312 | if (status) { | |
725130ba LB |
8313 | bio_put(bio); |
8314 | atomic_dec(&dip->pending_bios); | |
8315 | goto out_err; | |
8316 | } | |
e65e1535 | 8317 | |
725130ba LB |
8318 | clone_offset += clone_len; |
8319 | start_sector += clone_len >> 9; | |
8320 | file_offset += clone_len; | |
5f4dc8fc | 8321 | |
725130ba LB |
8322 | map_length = submit_len; |
8323 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), | |
8324 | start_sector << 9, &map_length, NULL, 0); | |
8325 | if (ret) | |
8326 | goto out_err; | |
3c91ee69 | 8327 | } while (submit_len > 0); |
e65e1535 | 8328 | |
02f57c7a | 8329 | submit: |
d0ee3934 | 8330 | status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit); |
58efbc9f | 8331 | if (!status) |
e65e1535 MX |
8332 | return 0; |
8333 | ||
8334 | bio_put(bio); | |
8335 | out_err: | |
8336 | dip->errors = 1; | |
8337 | /* | |
de224b7c NB |
8338 | * Before atomic variable goto zero, we must make sure dip->errors is |
8339 | * perceived to be set. This ordering is ensured by the fact that an | |
8340 | * atomic operations with a return value are fully ordered as per | |
8341 | * atomic_t.txt | |
e65e1535 | 8342 | */ |
e65e1535 MX |
8343 | if (atomic_dec_and_test(&dip->pending_bios)) |
8344 | bio_io_error(dip->orig_bio); | |
8345 | ||
8346 | /* bio_end_io() will handle error, so we needn't return it */ | |
8347 | return 0; | |
8348 | } | |
8349 | ||
8a4c1e42 MC |
8350 | static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode, |
8351 | loff_t file_offset) | |
4b46fce2 | 8352 | { |
61de718f | 8353 | struct btrfs_dio_private *dip = NULL; |
3892ac90 LB |
8354 | struct bio *bio = NULL; |
8355 | struct btrfs_io_bio *io_bio; | |
8a4c1e42 | 8356 | bool write = (bio_op(dio_bio) == REQ_OP_WRITE); |
4b46fce2 JB |
8357 | int ret = 0; |
8358 | ||
8b6c1d56 | 8359 | bio = btrfs_bio_clone(dio_bio); |
9be3395b | 8360 | |
c1dc0896 | 8361 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8362 | if (!dip) { |
8363 | ret = -ENOMEM; | |
61de718f | 8364 | goto free_ordered; |
4b46fce2 | 8365 | } |
4b46fce2 | 8366 | |
9be3395b | 8367 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8368 | dip->inode = inode; |
8369 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8370 | dip->bytes = dio_bio->bi_iter.bi_size; |
8371 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
3892ac90 LB |
8372 | bio->bi_private = dip; |
8373 | dip->orig_bio = bio; | |
9be3395b | 8374 | dip->dio_bio = dio_bio; |
e65e1535 | 8375 | atomic_set(&dip->pending_bios, 0); |
3892ac90 LB |
8376 | io_bio = btrfs_io_bio(bio); |
8377 | io_bio->logical = file_offset; | |
4b46fce2 | 8378 | |
c1dc0896 | 8379 | if (write) { |
3892ac90 | 8380 | bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8381 | } else { |
3892ac90 | 8382 | bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8383 | dip->subio_endio = btrfs_subio_endio_read; |
8384 | } | |
4b46fce2 | 8385 | |
f28a4928 FM |
8386 | /* |
8387 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8388 | * even if we fail to submit a bio, because in such case we do the | |
8389 | * corresponding error handling below and it must not be done a second | |
8390 | * time by btrfs_direct_IO(). | |
8391 | */ | |
8392 | if (write) { | |
8393 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8394 | ||
8395 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8396 | dip->bytes; | |
8397 | dio_data->unsubmitted_oe_range_start = | |
8398 | dio_data->unsubmitted_oe_range_end; | |
8399 | } | |
8400 | ||
e6961cac | 8401 | ret = btrfs_submit_direct_hook(dip); |
e65e1535 | 8402 | if (!ret) |
eaf25d93 | 8403 | return; |
9be3395b | 8404 | |
b3a0dd50 | 8405 | btrfs_io_bio_free_csum(io_bio); |
9be3395b | 8406 | |
4b46fce2 JB |
8407 | free_ordered: |
8408 | /* | |
61de718f FM |
8409 | * If we arrived here it means either we failed to submit the dip |
8410 | * or we either failed to clone the dio_bio or failed to allocate the | |
8411 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8412 | * call bio_endio against our io_bio so that we get proper resource | |
8413 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8414 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8415 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8416 | */ |
3892ac90 | 8417 | if (bio && dip) { |
054ec2f6 | 8418 | bio_io_error(bio); |
61de718f | 8419 | /* |
3892ac90 | 8420 | * The end io callbacks free our dip, do the final put on bio |
61de718f FM |
8421 | * and all the cleanup and final put for dio_bio (through |
8422 | * dio_end_io()). | |
8423 | */ | |
8424 | dip = NULL; | |
3892ac90 | 8425 | bio = NULL; |
61de718f | 8426 | } else { |
14543774 | 8427 | if (write) |
52427260 | 8428 | __endio_write_update_ordered(inode, |
14543774 FM |
8429 | file_offset, |
8430 | dio_bio->bi_iter.bi_size, | |
52427260 | 8431 | false); |
14543774 | 8432 | else |
61de718f FM |
8433 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8434 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8435 | |
4e4cbee9 | 8436 | dio_bio->bi_status = BLK_STS_IOERR; |
61de718f FM |
8437 | /* |
8438 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8439 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8440 | */ | |
4055351c | 8441 | dio_end_io(dio_bio); |
4b46fce2 | 8442 | } |
3892ac90 LB |
8443 | if (bio) |
8444 | bio_put(bio); | |
61de718f | 8445 | kfree(dip); |
4b46fce2 JB |
8446 | } |
8447 | ||
2ff7e61e | 8448 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
2ff7e61e | 8449 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
8450 | { |
8451 | int seg; | |
a1b75f7d | 8452 | int i; |
0b246afa | 8453 | unsigned int blocksize_mask = fs_info->sectorsize - 1; |
5a5f79b5 | 8454 | ssize_t retval = -EINVAL; |
5a5f79b5 CM |
8455 | |
8456 | if (offset & blocksize_mask) | |
8457 | goto out; | |
8458 | ||
28060d5d AV |
8459 | if (iov_iter_alignment(iter) & blocksize_mask) |
8460 | goto out; | |
a1b75f7d | 8461 | |
28060d5d | 8462 | /* If this is a write we don't need to check anymore */ |
cd27e455 | 8463 | if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter)) |
28060d5d AV |
8464 | return 0; |
8465 | /* | |
8466 | * Check to make sure we don't have duplicate iov_base's in this | |
8467 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8468 | * when reading back. | |
8469 | */ | |
8470 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8471 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8472 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8473 | goto out; |
8474 | } | |
5a5f79b5 CM |
8475 | } |
8476 | retval = 0; | |
8477 | out: | |
8478 | return retval; | |
8479 | } | |
eb838e73 | 8480 | |
c8b8e32d | 8481 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8482 | { |
4b46fce2 JB |
8483 | struct file *file = iocb->ki_filp; |
8484 | struct inode *inode = file->f_mapping->host; | |
0b246afa | 8485 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
50745b0a | 8486 | struct btrfs_dio_data dio_data = { 0 }; |
364ecf36 | 8487 | struct extent_changeset *data_reserved = NULL; |
c8b8e32d | 8488 | loff_t offset = iocb->ki_pos; |
0934856d | 8489 | size_t count = 0; |
2e60a51e | 8490 | int flags = 0; |
38851cc1 MX |
8491 | bool wakeup = true; |
8492 | bool relock = false; | |
0934856d | 8493 | ssize_t ret; |
4b46fce2 | 8494 | |
8c70c9f8 | 8495 | if (check_direct_IO(fs_info, iter, offset)) |
5a5f79b5 | 8496 | return 0; |
3f7c579c | 8497 | |
fe0f07d0 | 8498 | inode_dio_begin(inode); |
38851cc1 | 8499 | |
0e267c44 | 8500 | /* |
41bd9ca4 MX |
8501 | * The generic stuff only does filemap_write_and_wait_range, which |
8502 | * isn't enough if we've written compressed pages to this area, so | |
8503 | * we need to flush the dirty pages again to make absolutely sure | |
8504 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8505 | */ |
a6cbcd4a | 8506 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8507 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8508 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8509 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8510 | offset + count - 1); | |
0e267c44 | 8511 | |
6f673763 | 8512 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8513 | /* |
8514 | * If the write DIO is beyond the EOF, we need update | |
8515 | * the isize, but it is protected by i_mutex. So we can | |
8516 | * not unlock the i_mutex at this case. | |
8517 | */ | |
8518 | if (offset + count <= inode->i_size) { | |
4aaedfb0 | 8519 | dio_data.overwrite = 1; |
5955102c | 8520 | inode_unlock(inode); |
38851cc1 | 8521 | relock = true; |
edf064e7 GR |
8522 | } else if (iocb->ki_flags & IOCB_NOWAIT) { |
8523 | ret = -EAGAIN; | |
8524 | goto out; | |
38851cc1 | 8525 | } |
364ecf36 QW |
8526 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8527 | offset, count); | |
0934856d | 8528 | if (ret) |
38851cc1 | 8529 | goto out; |
e1cbbfa5 JB |
8530 | |
8531 | /* | |
8532 | * We need to know how many extents we reserved so that we can | |
8533 | * do the accounting properly if we go over the number we | |
8534 | * originally calculated. Abuse current->journal_info for this. | |
8535 | */ | |
da17066c | 8536 | dio_data.reserve = round_up(count, |
0b246afa | 8537 | fs_info->sectorsize); |
f28a4928 FM |
8538 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8539 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8540 | current->journal_info = &dio_data; |
97dcdea0 | 8541 | down_read(&BTRFS_I(inode)->dio_sem); |
ee39b432 DS |
8542 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8543 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8544 | inode_dio_end(inode); |
38851cc1 MX |
8545 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8546 | wakeup = false; | |
0934856d MX |
8547 | } |
8548 | ||
17f8c842 | 8549 | ret = __blockdev_direct_IO(iocb, inode, |
0b246afa | 8550 | fs_info->fs_devices->latest_bdev, |
c8b8e32d | 8551 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8552 | btrfs_submit_direct, flags); |
6f673763 | 8553 | if (iov_iter_rw(iter) == WRITE) { |
97dcdea0 | 8554 | up_read(&BTRFS_I(inode)->dio_sem); |
e1cbbfa5 | 8555 | current->journal_info = NULL; |
ddba1bfc | 8556 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8557 | if (dio_data.reserve) |
bc42bda2 | 8558 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 | 8559 | offset, dio_data.reserve, true); |
f28a4928 FM |
8560 | /* |
8561 | * On error we might have left some ordered extents | |
8562 | * without submitting corresponding bios for them, so | |
8563 | * cleanup them up to avoid other tasks getting them | |
8564 | * and waiting for them to complete forever. | |
8565 | */ | |
8566 | if (dio_data.unsubmitted_oe_range_start < | |
8567 | dio_data.unsubmitted_oe_range_end) | |
52427260 | 8568 | __endio_write_update_ordered(inode, |
f28a4928 FM |
8569 | dio_data.unsubmitted_oe_range_start, |
8570 | dio_data.unsubmitted_oe_range_end - | |
8571 | dio_data.unsubmitted_oe_range_start, | |
52427260 | 8572 | false); |
ddba1bfc | 8573 | } else if (ret >= 0 && (size_t)ret < count) |
bc42bda2 | 8574 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
8575 | offset, count - (size_t)ret, true); |
8576 | btrfs_delalloc_release_extents(BTRFS_I(inode), count, false); | |
0934856d | 8577 | } |
38851cc1 | 8578 | out: |
2e60a51e | 8579 | if (wakeup) |
fe0f07d0 | 8580 | inode_dio_end(inode); |
38851cc1 | 8581 | if (relock) |
5955102c | 8582 | inode_lock(inode); |
0934856d | 8583 | |
364ecf36 | 8584 | extent_changeset_free(data_reserved); |
0934856d | 8585 | return ret; |
16432985 CM |
8586 | } |
8587 | ||
05dadc09 TI |
8588 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8589 | ||
1506fcc8 YS |
8590 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8591 | __u64 start, __u64 len) | |
8592 | { | |
05dadc09 TI |
8593 | int ret; |
8594 | ||
8595 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8596 | if (ret) | |
8597 | return ret; | |
8598 | ||
2135fb9b | 8599 | return extent_fiemap(inode, fieinfo, start, len); |
1506fcc8 YS |
8600 | } |
8601 | ||
a52d9a80 | 8602 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8603 | { |
d1310b2e CM |
8604 | struct extent_io_tree *tree; |
8605 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8606 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8607 | } |
1832a6d5 | 8608 | |
a52d9a80 | 8609 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8610 | { |
be7bd730 JB |
8611 | struct inode *inode = page->mapping->host; |
8612 | int ret; | |
b888db2b CM |
8613 | |
8614 | if (current->flags & PF_MEMALLOC) { | |
8615 | redirty_page_for_writepage(wbc, page); | |
8616 | unlock_page(page); | |
8617 | return 0; | |
8618 | } | |
be7bd730 JB |
8619 | |
8620 | /* | |
8621 | * If we are under memory pressure we will call this directly from the | |
8622 | * VM, we need to make sure we have the inode referenced for the ordered | |
8623 | * extent. If not just return like we didn't do anything. | |
8624 | */ | |
8625 | if (!igrab(inode)) { | |
8626 | redirty_page_for_writepage(wbc, page); | |
8627 | return AOP_WRITEPAGE_ACTIVATE; | |
8628 | } | |
0a9b0e53 | 8629 | ret = extent_write_full_page(page, wbc); |
be7bd730 JB |
8630 | btrfs_add_delayed_iput(inode); |
8631 | return ret; | |
9ebefb18 CM |
8632 | } |
8633 | ||
48a3b636 ES |
8634 | static int btrfs_writepages(struct address_space *mapping, |
8635 | struct writeback_control *wbc) | |
b293f02e | 8636 | { |
8ae225a8 | 8637 | return extent_writepages(mapping, wbc); |
b293f02e CM |
8638 | } |
8639 | ||
3ab2fb5a CM |
8640 | static int |
8641 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8642 | struct list_head *pages, unsigned nr_pages) | |
8643 | { | |
2a3ff0ad | 8644 | return extent_readpages(mapping, pages, nr_pages); |
3ab2fb5a | 8645 | } |
2a3ff0ad | 8646 | |
e6dcd2dc | 8647 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8648 | { |
477a30ba | 8649 | int ret = try_release_extent_mapping(page, gfp_flags); |
a52d9a80 CM |
8650 | if (ret == 1) { |
8651 | ClearPagePrivate(page); | |
8652 | set_page_private(page, 0); | |
09cbfeaf | 8653 | put_page(page); |
39279cc3 | 8654 | } |
a52d9a80 | 8655 | return ret; |
39279cc3 CM |
8656 | } |
8657 | ||
e6dcd2dc CM |
8658 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8659 | { | |
98509cfc CM |
8660 | if (PageWriteback(page) || PageDirty(page)) |
8661 | return 0; | |
3ba7ab22 | 8662 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8663 | } |
8664 | ||
d47992f8 LC |
8665 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8666 | unsigned int length) | |
39279cc3 | 8667 | { |
5fd02043 | 8668 | struct inode *inode = page->mapping->host; |
d1310b2e | 8669 | struct extent_io_tree *tree; |
e6dcd2dc | 8670 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8671 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8672 | u64 page_start = page_offset(page); |
09cbfeaf | 8673 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8674 | u64 start; |
8675 | u64 end; | |
131e404a | 8676 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8677 | |
8b62b72b CM |
8678 | /* |
8679 | * we have the page locked, so new writeback can't start, | |
8680 | * and the dirty bit won't be cleared while we are here. | |
8681 | * | |
8682 | * Wait for IO on this page so that we can safely clear | |
8683 | * the PagePrivate2 bit and do ordered accounting | |
8684 | */ | |
e6dcd2dc | 8685 | wait_on_page_writeback(page); |
8b62b72b | 8686 | |
5fd02043 | 8687 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8688 | if (offset) { |
8689 | btrfs_releasepage(page, GFP_NOFS); | |
8690 | return; | |
8691 | } | |
131e404a FDBM |
8692 | |
8693 | if (!inode_evicting) | |
ff13db41 | 8694 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8695 | again: |
8696 | start = page_start; | |
a776c6fa | 8697 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
dbfdb6d1 | 8698 | page_end - start + 1); |
e6dcd2dc | 8699 | if (ordered) { |
dbfdb6d1 | 8700 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
8701 | /* |
8702 | * IO on this page will never be started, so we need | |
8703 | * to account for any ordered extents now | |
8704 | */ | |
131e404a | 8705 | if (!inode_evicting) |
dbfdb6d1 | 8706 | clear_extent_bit(tree, start, end, |
131e404a | 8707 | EXTENT_DIRTY | EXTENT_DELALLOC | |
a7e3b975 | 8708 | EXTENT_DELALLOC_NEW | |
131e404a | 8709 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
ae0f1625 | 8710 | EXTENT_DEFRAG, 1, 0, &cached_state); |
8b62b72b CM |
8711 | /* |
8712 | * whoever cleared the private bit is responsible | |
8713 | * for the finish_ordered_io | |
8714 | */ | |
77cef2ec JB |
8715 | if (TestClearPagePrivate2(page)) { |
8716 | struct btrfs_ordered_inode_tree *tree; | |
8717 | u64 new_len; | |
8718 | ||
8719 | tree = &BTRFS_I(inode)->ordered_tree; | |
8720 | ||
8721 | spin_lock_irq(&tree->lock); | |
8722 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8723 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8724 | if (new_len < ordered->truncated_len) |
8725 | ordered->truncated_len = new_len; | |
8726 | spin_unlock_irq(&tree->lock); | |
8727 | ||
8728 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
8729 | start, |
8730 | end - start + 1, 1)) | |
77cef2ec | 8731 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 8732 | } |
e6dcd2dc | 8733 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8734 | if (!inode_evicting) { |
8735 | cached_state = NULL; | |
dbfdb6d1 | 8736 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
8737 | &cached_state); |
8738 | } | |
dbfdb6d1 CR |
8739 | |
8740 | start = end + 1; | |
8741 | if (start < page_end) | |
8742 | goto again; | |
131e404a FDBM |
8743 | } |
8744 | ||
b9d0b389 QW |
8745 | /* |
8746 | * Qgroup reserved space handler | |
8747 | * Page here will be either | |
8748 | * 1) Already written to disk | |
8749 | * In this case, its reserved space is released from data rsv map | |
8750 | * and will be freed by delayed_ref handler finally. | |
8751 | * So even we call qgroup_free_data(), it won't decrease reserved | |
8752 | * space. | |
8753 | * 2) Not written to disk | |
0b34c261 GR |
8754 | * This means the reserved space should be freed here. However, |
8755 | * if a truncate invalidates the page (by clearing PageDirty) | |
8756 | * and the page is accounted for while allocating extent | |
8757 | * in btrfs_check_data_free_space() we let delayed_ref to | |
8758 | * free the entire extent. | |
b9d0b389 | 8759 | */ |
0b34c261 | 8760 | if (PageDirty(page)) |
bc42bda2 | 8761 | btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE); |
131e404a FDBM |
8762 | if (!inode_evicting) { |
8763 | clear_extent_bit(tree, page_start, page_end, | |
8764 | EXTENT_LOCKED | EXTENT_DIRTY | | |
a7e3b975 FM |
8765 | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
8766 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
ae0f1625 | 8767 | &cached_state); |
131e404a FDBM |
8768 | |
8769 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8770 | } |
e6dcd2dc | 8771 | |
4a096752 | 8772 | ClearPageChecked(page); |
9ad6b7bc | 8773 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8774 | ClearPagePrivate(page); |
8775 | set_page_private(page, 0); | |
09cbfeaf | 8776 | put_page(page); |
9ad6b7bc | 8777 | } |
39279cc3 CM |
8778 | } |
8779 | ||
9ebefb18 CM |
8780 | /* |
8781 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8782 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8783 | * be careful to check for EOF conditions here. We set the page up correctly | |
8784 | * for a written page which means we get ENOSPC checking when writing into | |
8785 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8786 | * support these features. | |
8787 | * | |
8788 | * We are not allowed to take the i_mutex here so we have to play games to | |
8789 | * protect against truncate races as the page could now be beyond EOF. Because | |
d1342aad OS |
8790 | * truncate_setsize() writes the inode size before removing pages, once we have |
8791 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
9ebefb18 CM |
8792 | * beyond EOF, then the page is guaranteed safe against truncation until we |
8793 | * unlock the page. | |
8794 | */ | |
a528a241 | 8795 | vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 8796 | { |
c2ec175c | 8797 | struct page *page = vmf->page; |
11bac800 | 8798 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 8799 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8800 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8801 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8802 | struct extent_state *cached_state = NULL; |
364ecf36 | 8803 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc CM |
8804 | char *kaddr; |
8805 | unsigned long zero_start; | |
9ebefb18 | 8806 | loff_t size; |
a528a241 SJ |
8807 | vm_fault_t ret; |
8808 | int ret2; | |
9998eb70 | 8809 | int reserved = 0; |
d0b7da88 | 8810 | u64 reserved_space; |
a52d9a80 | 8811 | u64 page_start; |
e6dcd2dc | 8812 | u64 page_end; |
d0b7da88 CR |
8813 | u64 end; |
8814 | ||
09cbfeaf | 8815 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8816 | |
b2b5ef5c | 8817 | sb_start_pagefault(inode->i_sb); |
df480633 | 8818 | page_start = page_offset(page); |
09cbfeaf | 8819 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8820 | end = page_end; |
df480633 | 8821 | |
d0b7da88 CR |
8822 | /* |
8823 | * Reserving delalloc space after obtaining the page lock can lead to | |
8824 | * deadlock. For example, if a dirty page is locked by this function | |
8825 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8826 | * dirty page write out, then the btrfs_writepage() function could | |
8827 | * end up waiting indefinitely to get a lock on the page currently | |
8828 | * being processed by btrfs_page_mkwrite() function. | |
8829 | */ | |
a528a241 | 8830 | ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
d0b7da88 | 8831 | reserved_space); |
a528a241 SJ |
8832 | if (!ret2) { |
8833 | ret2 = file_update_time(vmf->vma->vm_file); | |
9998eb70 CM |
8834 | reserved = 1; |
8835 | } | |
a528a241 SJ |
8836 | if (ret2) { |
8837 | ret = vmf_error(ret2); | |
9998eb70 CM |
8838 | if (reserved) |
8839 | goto out; | |
8840 | goto out_noreserve; | |
56a76f82 | 8841 | } |
1832a6d5 | 8842 | |
56a76f82 | 8843 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8844 | again: |
9ebefb18 | 8845 | lock_page(page); |
9ebefb18 | 8846 | size = i_size_read(inode); |
a52d9a80 | 8847 | |
9ebefb18 | 8848 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8849 | (page_start >= size)) { |
9ebefb18 CM |
8850 | /* page got truncated out from underneath us */ |
8851 | goto out_unlock; | |
8852 | } | |
e6dcd2dc CM |
8853 | wait_on_page_writeback(page); |
8854 | ||
ff13db41 | 8855 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
8856 | set_page_extent_mapped(page); |
8857 | ||
eb84ae03 CM |
8858 | /* |
8859 | * we can't set the delalloc bits if there are pending ordered | |
8860 | * extents. Drop our locks and wait for them to finish | |
8861 | */ | |
a776c6fa NB |
8862 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
8863 | PAGE_SIZE); | |
e6dcd2dc | 8864 | if (ordered) { |
2ac55d41 | 8865 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8866 | &cached_state); |
e6dcd2dc | 8867 | unlock_page(page); |
eb84ae03 | 8868 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
8869 | btrfs_put_ordered_extent(ordered); |
8870 | goto again; | |
8871 | } | |
8872 | ||
09cbfeaf | 8873 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 8874 | reserved_space = round_up(size - page_start, |
0b246afa | 8875 | fs_info->sectorsize); |
09cbfeaf | 8876 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 8877 | end = page_start + reserved_space - 1; |
bc42bda2 | 8878 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
8879 | page_start, PAGE_SIZE - reserved_space, |
8880 | true); | |
d0b7da88 CR |
8881 | } |
8882 | } | |
8883 | ||
fbf19087 | 8884 | /* |
5416034f LB |
8885 | * page_mkwrite gets called when the page is firstly dirtied after it's |
8886 | * faulted in, but write(2) could also dirty a page and set delalloc | |
8887 | * bits, thus in this case for space account reason, we still need to | |
8888 | * clear any delalloc bits within this page range since we have to | |
8889 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 8890 | */ |
d0b7da88 | 8891 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
8892 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8893 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 8894 | 0, 0, &cached_state); |
fbf19087 | 8895 | |
a528a241 | 8896 | ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0, |
ba8b04c1 | 8897 | &cached_state, 0); |
a528a241 | 8898 | if (ret2) { |
2ac55d41 | 8899 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8900 | &cached_state); |
9ed74f2d JB |
8901 | ret = VM_FAULT_SIGBUS; |
8902 | goto out_unlock; | |
8903 | } | |
a528a241 | 8904 | ret2 = 0; |
9ebefb18 CM |
8905 | |
8906 | /* page is wholly or partially inside EOF */ | |
09cbfeaf | 8907 | if (page_start + PAGE_SIZE > size) |
7073017a | 8908 | zero_start = offset_in_page(size); |
9ebefb18 | 8909 | else |
09cbfeaf | 8910 | zero_start = PAGE_SIZE; |
9ebefb18 | 8911 | |
09cbfeaf | 8912 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 8913 | kaddr = kmap(page); |
09cbfeaf | 8914 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
8915 | flush_dcache_page(page); |
8916 | kunmap(page); | |
8917 | } | |
247e743c | 8918 | ClearPageChecked(page); |
e6dcd2dc | 8919 | set_page_dirty(page); |
50a9b214 | 8920 | SetPageUptodate(page); |
5a3f23d5 | 8921 | |
0b246afa | 8922 | BTRFS_I(inode)->last_trans = fs_info->generation; |
257c62e1 | 8923 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; |
46d8bc34 | 8924 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 8925 | |
e43bbe5e | 8926 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state); |
9ebefb18 | 8927 | |
a528a241 | 8928 | if (!ret2) { |
43b18595 | 8929 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true); |
b2b5ef5c | 8930 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 8931 | extent_changeset_free(data_reserved); |
50a9b214 | 8932 | return VM_FAULT_LOCKED; |
b2b5ef5c | 8933 | } |
717beb96 CM |
8934 | |
8935 | out_unlock: | |
9ebefb18 | 8936 | unlock_page(page); |
1832a6d5 | 8937 | out: |
43b18595 | 8938 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0)); |
bc42bda2 | 8939 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
43b18595 | 8940 | reserved_space, (ret != 0)); |
9998eb70 | 8941 | out_noreserve: |
b2b5ef5c | 8942 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 8943 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
8944 | return ret; |
8945 | } | |
8946 | ||
213e8c55 | 8947 | static int btrfs_truncate(struct inode *inode, bool skip_writeback) |
39279cc3 | 8948 | { |
0b246afa | 8949 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 8950 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 8951 | struct btrfs_block_rsv *rsv; |
ad7e1a74 | 8952 | int ret; |
39279cc3 | 8953 | struct btrfs_trans_handle *trans; |
0b246afa JM |
8954 | u64 mask = fs_info->sectorsize - 1; |
8955 | u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); | |
39279cc3 | 8956 | |
213e8c55 FM |
8957 | if (!skip_writeback) { |
8958 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), | |
8959 | (u64)-1); | |
8960 | if (ret) | |
8961 | return ret; | |
8962 | } | |
39279cc3 | 8963 | |
fcb80c2a | 8964 | /* |
f7e9e8fc OS |
8965 | * Yes ladies and gentlemen, this is indeed ugly. We have a couple of |
8966 | * things going on here: | |
fcb80c2a | 8967 | * |
f7e9e8fc | 8968 | * 1) We need to reserve space to update our inode. |
fcb80c2a | 8969 | * |
f7e9e8fc | 8970 | * 2) We need to have something to cache all the space that is going to |
fcb80c2a JB |
8971 | * be free'd up by the truncate operation, but also have some slack |
8972 | * space reserved in case it uses space during the truncate (thank you | |
8973 | * very much snapshotting). | |
8974 | * | |
f7e9e8fc | 8975 | * And we need these to be separate. The fact is we can use a lot of |
fcb80c2a | 8976 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 8977 | * we will use, so we need the truncate reservation to be separate so it |
f7e9e8fc OS |
8978 | * doesn't end up using space reserved for updating the inode. We also |
8979 | * need to be able to stop the transaction and start a new one, which | |
8980 | * means we need to be able to update the inode several times, and we | |
8981 | * have no idea of knowing how many times that will be, so we can't just | |
8982 | * reserve 1 item for the entirety of the operation, so that has to be | |
8983 | * done separately as well. | |
fcb80c2a JB |
8984 | * |
8985 | * So that leaves us with | |
8986 | * | |
f7e9e8fc | 8987 | * 1) rsv - for the truncate reservation, which we will steal from the |
fcb80c2a | 8988 | * transaction reservation. |
f7e9e8fc | 8989 | * 2) fs_info->trans_block_rsv - this will have 1 items worth left for |
fcb80c2a JB |
8990 | * updating the inode. |
8991 | */ | |
2ff7e61e | 8992 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8993 | if (!rsv) |
8994 | return -ENOMEM; | |
4a338542 | 8995 | rsv->size = min_size; |
ca7e70f5 | 8996 | rsv->failfast = 1; |
f0cd846e | 8997 | |
907cbceb | 8998 | /* |
07127184 | 8999 | * 1 for the truncate slack space |
907cbceb JB |
9000 | * 1 for updating the inode. |
9001 | */ | |
f3fe820c | 9002 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a | 9003 | if (IS_ERR(trans)) { |
ad7e1a74 | 9004 | ret = PTR_ERR(trans); |
fcb80c2a JB |
9005 | goto out; |
9006 | } | |
f0cd846e | 9007 | |
907cbceb | 9008 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 9009 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
3a584174 | 9010 | min_size, false); |
fcb80c2a | 9011 | BUG_ON(ret); |
f0cd846e | 9012 | |
5dc562c5 JB |
9013 | /* |
9014 | * So if we truncate and then write and fsync we normally would just | |
9015 | * write the extents that changed, which is a problem if we need to | |
9016 | * first truncate that entire inode. So set this flag so we write out | |
9017 | * all of the extents in the inode to the sync log so we're completely | |
9018 | * safe. | |
9019 | */ | |
9020 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9021 | trans->block_rsv = rsv; |
907cbceb | 9022 | |
8082510e YZ |
9023 | while (1) { |
9024 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9025 | inode->i_size, | |
9026 | BTRFS_EXTENT_DATA_KEY); | |
ddfae63c | 9027 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 | 9028 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 9029 | break; |
39279cc3 | 9030 | |
8082510e | 9031 | ret = btrfs_update_inode(trans, root, inode); |
ad7e1a74 | 9032 | if (ret) |
3893e33b | 9033 | break; |
ca7e70f5 | 9034 | |
3a45bb20 | 9035 | btrfs_end_transaction(trans); |
2ff7e61e | 9036 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
9037 | |
9038 | trans = btrfs_start_transaction(root, 2); | |
9039 | if (IS_ERR(trans)) { | |
ad7e1a74 | 9040 | ret = PTR_ERR(trans); |
ca7e70f5 JB |
9041 | trans = NULL; |
9042 | break; | |
9043 | } | |
9044 | ||
47b5d646 | 9045 | btrfs_block_rsv_release(fs_info, rsv, -1); |
0b246afa | 9046 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
3a584174 | 9047 | rsv, min_size, false); |
ca7e70f5 JB |
9048 | BUG_ON(ret); /* shouldn't happen */ |
9049 | trans->block_rsv = rsv; | |
8082510e YZ |
9050 | } |
9051 | ||
ddfae63c JB |
9052 | /* |
9053 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
9054 | * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know | |
9055 | * we've truncated everything except the last little bit, and can do | |
9056 | * btrfs_truncate_block and then update the disk_i_size. | |
9057 | */ | |
9058 | if (ret == NEED_TRUNCATE_BLOCK) { | |
9059 | btrfs_end_transaction(trans); | |
9060 | btrfs_btree_balance_dirty(fs_info); | |
9061 | ||
9062 | ret = btrfs_truncate_block(inode, inode->i_size, 0, 0); | |
9063 | if (ret) | |
9064 | goto out; | |
9065 | trans = btrfs_start_transaction(root, 1); | |
9066 | if (IS_ERR(trans)) { | |
9067 | ret = PTR_ERR(trans); | |
9068 | goto out; | |
9069 | } | |
9070 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
9071 | } | |
9072 | ||
917c16b2 | 9073 | if (trans) { |
ad7e1a74 OS |
9074 | int ret2; |
9075 | ||
0b246afa | 9076 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 OS |
9077 | ret2 = btrfs_update_inode(trans, root, inode); |
9078 | if (ret2 && !ret) | |
9079 | ret = ret2; | |
7b128766 | 9080 | |
ad7e1a74 OS |
9081 | ret2 = btrfs_end_transaction(trans); |
9082 | if (ret2 && !ret) | |
9083 | ret = ret2; | |
2ff7e61e | 9084 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 9085 | } |
fcb80c2a | 9086 | out: |
2ff7e61e | 9087 | btrfs_free_block_rsv(fs_info, rsv); |
fcb80c2a | 9088 | |
ad7e1a74 | 9089 | return ret; |
39279cc3 CM |
9090 | } |
9091 | ||
d352ac68 CM |
9092 | /* |
9093 | * create a new subvolume directory/inode (helper for the ioctl). | |
9094 | */ | |
d2fb3437 | 9095 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9096 | struct btrfs_root *new_root, |
9097 | struct btrfs_root *parent_root, | |
9098 | u64 new_dirid) | |
39279cc3 | 9099 | { |
39279cc3 | 9100 | struct inode *inode; |
76dda93c | 9101 | int err; |
00e4e6b3 | 9102 | u64 index = 0; |
39279cc3 | 9103 | |
12fc9d09 FA |
9104 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9105 | new_dirid, new_dirid, | |
9106 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9107 | &index); | |
54aa1f4d | 9108 | if (IS_ERR(inode)) |
f46b5a66 | 9109 | return PTR_ERR(inode); |
39279cc3 CM |
9110 | inode->i_op = &btrfs_dir_inode_operations; |
9111 | inode->i_fop = &btrfs_dir_file_operations; | |
9112 | ||
bfe86848 | 9113 | set_nlink(inode, 1); |
6ef06d27 | 9114 | btrfs_i_size_write(BTRFS_I(inode), 0); |
b0d5d10f | 9115 | unlock_new_inode(inode); |
3b96362c | 9116 | |
63541927 FDBM |
9117 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9118 | if (err) | |
9119 | btrfs_err(new_root->fs_info, | |
351fd353 | 9120 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9121 | new_root->root_key.objectid, err); |
9122 | ||
76dda93c | 9123 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9124 | |
76dda93c | 9125 | iput(inode); |
ce598979 | 9126 | return err; |
39279cc3 CM |
9127 | } |
9128 | ||
39279cc3 CM |
9129 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9130 | { | |
69fe2d75 | 9131 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 9132 | struct btrfs_inode *ei; |
2ead6ae7 | 9133 | struct inode *inode; |
39279cc3 | 9134 | |
712e36c5 | 9135 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL); |
39279cc3 CM |
9136 | if (!ei) |
9137 | return NULL; | |
2ead6ae7 YZ |
9138 | |
9139 | ei->root = NULL; | |
2ead6ae7 | 9140 | ei->generation = 0; |
15ee9bc7 | 9141 | ei->last_trans = 0; |
257c62e1 | 9142 | ei->last_sub_trans = 0; |
e02119d5 | 9143 | ei->logged_trans = 0; |
2ead6ae7 | 9144 | ei->delalloc_bytes = 0; |
a7e3b975 | 9145 | ei->new_delalloc_bytes = 0; |
47059d93 | 9146 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9147 | ei->disk_i_size = 0; |
9148 | ei->flags = 0; | |
7709cde3 | 9149 | ei->csum_bytes = 0; |
2ead6ae7 | 9150 | ei->index_cnt = (u64)-1; |
67de1176 | 9151 | ei->dir_index = 0; |
2ead6ae7 | 9152 | ei->last_unlink_trans = 0; |
41bd6067 | 9153 | ei->last_link_trans = 0; |
46d8bc34 | 9154 | ei->last_log_commit = 0; |
2ead6ae7 | 9155 | |
9e0baf60 JB |
9156 | spin_lock_init(&ei->lock); |
9157 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
9158 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
9159 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
9160 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 9161 | ei->runtime_flags = 0; |
b52aa8c9 | 9162 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 9163 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9164 | |
16cdcec7 MX |
9165 | ei->delayed_node = NULL; |
9166 | ||
9cc97d64 | 9167 | ei->i_otime.tv_sec = 0; |
9168 | ei->i_otime.tv_nsec = 0; | |
9169 | ||
2ead6ae7 | 9170 | inode = &ei->vfs_inode; |
a8067e02 | 9171 | extent_map_tree_init(&ei->extent_tree); |
c6100a4b JB |
9172 | extent_io_tree_init(&ei->io_tree, inode); |
9173 | extent_io_tree_init(&ei->io_failure_tree, inode); | |
0b32f4bb JB |
9174 | ei->io_tree.track_uptodate = 1; |
9175 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 9176 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9177 | mutex_init(&ei->log_mutex); |
f248679e | 9178 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9179 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9180 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9181 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9182 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 9183 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
9184 | |
9185 | return inode; | |
39279cc3 CM |
9186 | } |
9187 | ||
aaedb55b JB |
9188 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9189 | void btrfs_test_destroy_inode(struct inode *inode) | |
9190 | { | |
dcdbc059 | 9191 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
9192 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9193 | } | |
9194 | #endif | |
9195 | ||
fa0d7e3d NP |
9196 | static void btrfs_i_callback(struct rcu_head *head) |
9197 | { | |
9198 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
9199 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9200 | } | |
9201 | ||
39279cc3 CM |
9202 | void btrfs_destroy_inode(struct inode *inode) |
9203 | { | |
0b246afa | 9204 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 9205 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9206 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9207 | ||
b3d9b7a3 | 9208 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9209 | WARN_ON(inode->i_data.nrpages); |
69fe2d75 JB |
9210 | WARN_ON(BTRFS_I(inode)->block_rsv.reserved); |
9211 | WARN_ON(BTRFS_I(inode)->block_rsv.size); | |
9e0baf60 | 9212 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7709cde3 | 9213 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
a7e3b975 | 9214 | WARN_ON(BTRFS_I(inode)->new_delalloc_bytes); |
7709cde3 | 9215 | WARN_ON(BTRFS_I(inode)->csum_bytes); |
47059d93 | 9216 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9217 | |
a6dbd429 JB |
9218 | /* |
9219 | * This can happen where we create an inode, but somebody else also | |
9220 | * created the same inode and we need to destroy the one we already | |
9221 | * created. | |
9222 | */ | |
9223 | if (!root) | |
9224 | goto free; | |
9225 | ||
d397712b | 9226 | while (1) { |
e6dcd2dc CM |
9227 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9228 | if (!ordered) | |
9229 | break; | |
9230 | else { | |
0b246afa | 9231 | btrfs_err(fs_info, |
5d163e0e JM |
9232 | "found ordered extent %llu %llu on inode cleanup", |
9233 | ordered->file_offset, ordered->len); | |
e6dcd2dc CM |
9234 | btrfs_remove_ordered_extent(inode, ordered); |
9235 | btrfs_put_ordered_extent(ordered); | |
9236 | btrfs_put_ordered_extent(ordered); | |
9237 | } | |
9238 | } | |
56fa9d07 | 9239 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9240 | inode_tree_del(inode); |
dcdbc059 | 9241 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
a6dbd429 | 9242 | free: |
fa0d7e3d | 9243 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
9244 | } |
9245 | ||
45321ac5 | 9246 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9247 | { |
9248 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9249 | |
6379ef9f NA |
9250 | if (root == NULL) |
9251 | return 1; | |
9252 | ||
fa6ac876 | 9253 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9254 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9255 | return 1; |
76dda93c | 9256 | else |
45321ac5 | 9257 | return generic_drop_inode(inode); |
76dda93c YZ |
9258 | } |
9259 | ||
0ee0fda0 | 9260 | static void init_once(void *foo) |
39279cc3 CM |
9261 | { |
9262 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9263 | ||
9264 | inode_init_once(&ei->vfs_inode); | |
9265 | } | |
9266 | ||
e67c718b | 9267 | void __cold btrfs_destroy_cachep(void) |
39279cc3 | 9268 | { |
8c0a8537 KS |
9269 | /* |
9270 | * Make sure all delayed rcu free inodes are flushed before we | |
9271 | * destroy cache. | |
9272 | */ | |
9273 | rcu_barrier(); | |
5598e900 KM |
9274 | kmem_cache_destroy(btrfs_inode_cachep); |
9275 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
9276 | kmem_cache_destroy(btrfs_path_cachep); |
9277 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
9278 | } |
9279 | ||
f5c29bd9 | 9280 | int __init btrfs_init_cachep(void) |
39279cc3 | 9281 | { |
837e1972 | 9282 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9283 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9284 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9285 | init_once); | |
39279cc3 CM |
9286 | if (!btrfs_inode_cachep) |
9287 | goto fail; | |
9601e3f6 | 9288 | |
837e1972 | 9289 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 9290 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 9291 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9292 | if (!btrfs_trans_handle_cachep) |
9293 | goto fail; | |
9601e3f6 | 9294 | |
837e1972 | 9295 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 9296 | sizeof(struct btrfs_path), 0, |
fba4b697 | 9297 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9298 | if (!btrfs_path_cachep) |
9299 | goto fail; | |
9601e3f6 | 9300 | |
837e1972 | 9301 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 9302 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 9303 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
9304 | if (!btrfs_free_space_cachep) |
9305 | goto fail; | |
9306 | ||
39279cc3 CM |
9307 | return 0; |
9308 | fail: | |
9309 | btrfs_destroy_cachep(); | |
9310 | return -ENOMEM; | |
9311 | } | |
9312 | ||
a528d35e DH |
9313 | static int btrfs_getattr(const struct path *path, struct kstat *stat, |
9314 | u32 request_mask, unsigned int flags) | |
39279cc3 | 9315 | { |
df0af1a5 | 9316 | u64 delalloc_bytes; |
a528d35e | 9317 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 9318 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 YS |
9319 | u32 bi_flags = BTRFS_I(inode)->flags; |
9320 | ||
9321 | stat->result_mask |= STATX_BTIME; | |
9322 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
9323 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
9324 | if (bi_flags & BTRFS_INODE_APPEND) | |
9325 | stat->attributes |= STATX_ATTR_APPEND; | |
9326 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
9327 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
9328 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
9329 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
9330 | if (bi_flags & BTRFS_INODE_NODUMP) | |
9331 | stat->attributes |= STATX_ATTR_NODUMP; | |
9332 | ||
9333 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
9334 | STATX_ATTR_COMPRESSED | | |
9335 | STATX_ATTR_IMMUTABLE | | |
9336 | STATX_ATTR_NODUMP); | |
fadc0d8b | 9337 | |
39279cc3 | 9338 | generic_fillattr(inode, stat); |
0ee5dc67 | 9339 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9340 | |
9341 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 9342 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
df0af1a5 | 9343 | spin_unlock(&BTRFS_I(inode)->lock); |
fadc0d8b | 9344 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9345 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9346 | return 0; |
9347 | } | |
9348 | ||
cdd1fedf DF |
9349 | static int btrfs_rename_exchange(struct inode *old_dir, |
9350 | struct dentry *old_dentry, | |
9351 | struct inode *new_dir, | |
9352 | struct dentry *new_dentry) | |
9353 | { | |
0b246afa | 9354 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
9355 | struct btrfs_trans_handle *trans; |
9356 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9357 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9358 | struct inode *new_inode = new_dentry->d_inode; | |
9359 | struct inode *old_inode = old_dentry->d_inode; | |
95582b00 | 9360 | struct timespec64 ctime = current_time(old_inode); |
cdd1fedf | 9361 | struct dentry *parent; |
4a0cc7ca NB |
9362 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9363 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9364 | u64 old_idx = 0; |
9365 | u64 new_idx = 0; | |
9366 | u64 root_objectid; | |
9367 | int ret; | |
86e8aa0e FM |
9368 | bool root_log_pinned = false; |
9369 | bool dest_log_pinned = false; | |
d4682ba0 FM |
9370 | struct btrfs_log_ctx ctx_root; |
9371 | struct btrfs_log_ctx ctx_dest; | |
9372 | bool sync_log_root = false; | |
9373 | bool sync_log_dest = false; | |
9374 | bool commit_transaction = false; | |
cdd1fedf DF |
9375 | |
9376 | /* we only allow rename subvolume link between subvolumes */ | |
9377 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
9378 | return -EXDEV; | |
9379 | ||
d4682ba0 FM |
9380 | btrfs_init_log_ctx(&ctx_root, old_inode); |
9381 | btrfs_init_log_ctx(&ctx_dest, new_inode); | |
9382 | ||
cdd1fedf DF |
9383 | /* close the race window with snapshot create/destroy ioctl */ |
9384 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9385 | down_read(&fs_info->subvol_sem); |
cdd1fedf | 9386 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9387 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9388 | |
9389 | /* | |
9390 | * We want to reserve the absolute worst case amount of items. So if | |
9391 | * both inodes are subvols and we need to unlink them then that would | |
9392 | * require 4 item modifications, but if they are both normal inodes it | |
9393 | * would require 5 item modifications, so we'll assume their normal | |
9394 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9395 | * should cover the worst case number of items we'll modify. | |
9396 | */ | |
9397 | trans = btrfs_start_transaction(root, 12); | |
9398 | if (IS_ERR(trans)) { | |
9399 | ret = PTR_ERR(trans); | |
9400 | goto out_notrans; | |
9401 | } | |
9402 | ||
9403 | /* | |
9404 | * We need to find a free sequence number both in the source and | |
9405 | * in the destination directory for the exchange. | |
9406 | */ | |
877574e2 | 9407 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
9408 | if (ret) |
9409 | goto out_fail; | |
877574e2 | 9410 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
9411 | if (ret) |
9412 | goto out_fail; | |
9413 | ||
9414 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9415 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9416 | ||
9417 | /* Reference for the source. */ | |
9418 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9419 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9420 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9421 | } else { |
376e5a57 FM |
9422 | btrfs_pin_log_trans(root); |
9423 | root_log_pinned = true; | |
cdd1fedf DF |
9424 | ret = btrfs_insert_inode_ref(trans, dest, |
9425 | new_dentry->d_name.name, | |
9426 | new_dentry->d_name.len, | |
9427 | old_ino, | |
f85b7379 DS |
9428 | btrfs_ino(BTRFS_I(new_dir)), |
9429 | old_idx); | |
cdd1fedf DF |
9430 | if (ret) |
9431 | goto out_fail; | |
cdd1fedf DF |
9432 | } |
9433 | ||
9434 | /* And now for the dest. */ | |
9435 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9436 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9437 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9438 | } else { |
376e5a57 FM |
9439 | btrfs_pin_log_trans(dest); |
9440 | dest_log_pinned = true; | |
cdd1fedf DF |
9441 | ret = btrfs_insert_inode_ref(trans, root, |
9442 | old_dentry->d_name.name, | |
9443 | old_dentry->d_name.len, | |
9444 | new_ino, | |
f85b7379 DS |
9445 | btrfs_ino(BTRFS_I(old_dir)), |
9446 | new_idx); | |
cdd1fedf DF |
9447 | if (ret) |
9448 | goto out_fail; | |
cdd1fedf DF |
9449 | } |
9450 | ||
9451 | /* Update inode version and ctime/mtime. */ | |
9452 | inode_inc_iversion(old_dir); | |
9453 | inode_inc_iversion(new_dir); | |
9454 | inode_inc_iversion(old_inode); | |
9455 | inode_inc_iversion(new_inode); | |
9456 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9457 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9458 | old_inode->i_ctime = ctime; | |
9459 | new_inode->i_ctime = ctime; | |
9460 | ||
9461 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9462 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9463 | BTRFS_I(old_inode), 1); | |
9464 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9465 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9466 | } |
9467 | ||
9468 | /* src is a subvolume */ | |
9469 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9470 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
401b3b19 | 9471 | ret = btrfs_unlink_subvol(trans, old_dir, root_objectid, |
cdd1fedf DF |
9472 | old_dentry->d_name.name, |
9473 | old_dentry->d_name.len); | |
9474 | } else { /* src is an inode */ | |
4ec5934e NB |
9475 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9476 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
9477 | old_dentry->d_name.name, |
9478 | old_dentry->d_name.len); | |
9479 | if (!ret) | |
9480 | ret = btrfs_update_inode(trans, root, old_inode); | |
9481 | } | |
9482 | if (ret) { | |
66642832 | 9483 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9484 | goto out_fail; |
9485 | } | |
9486 | ||
9487 | /* dest is a subvolume */ | |
9488 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9489 | root_objectid = BTRFS_I(new_inode)->root->root_key.objectid; | |
401b3b19 | 9490 | ret = btrfs_unlink_subvol(trans, new_dir, root_objectid, |
cdd1fedf DF |
9491 | new_dentry->d_name.name, |
9492 | new_dentry->d_name.len); | |
9493 | } else { /* dest is an inode */ | |
4ec5934e NB |
9494 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9495 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
9496 | new_dentry->d_name.name, |
9497 | new_dentry->d_name.len); | |
9498 | if (!ret) | |
9499 | ret = btrfs_update_inode(trans, dest, new_inode); | |
9500 | } | |
9501 | if (ret) { | |
66642832 | 9502 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9503 | goto out_fail; |
9504 | } | |
9505 | ||
db0a669f | 9506 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9507 | new_dentry->d_name.name, |
9508 | new_dentry->d_name.len, 0, old_idx); | |
9509 | if (ret) { | |
66642832 | 9510 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9511 | goto out_fail; |
9512 | } | |
9513 | ||
db0a669f | 9514 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9515 | old_dentry->d_name.name, |
9516 | old_dentry->d_name.len, 0, new_idx); | |
9517 | if (ret) { | |
66642832 | 9518 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9519 | goto out_fail; |
9520 | } | |
9521 | ||
9522 | if (old_inode->i_nlink == 1) | |
9523 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9524 | if (new_inode->i_nlink == 1) | |
9525 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9526 | ||
86e8aa0e | 9527 | if (root_log_pinned) { |
cdd1fedf | 9528 | parent = new_dentry->d_parent; |
d4682ba0 FM |
9529 | ret = btrfs_log_new_name(trans, BTRFS_I(old_inode), |
9530 | BTRFS_I(old_dir), parent, | |
9531 | false, &ctx_root); | |
9532 | if (ret == BTRFS_NEED_LOG_SYNC) | |
9533 | sync_log_root = true; | |
9534 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
9535 | commit_transaction = true; | |
9536 | ret = 0; | |
cdd1fedf | 9537 | btrfs_end_log_trans(root); |
86e8aa0e | 9538 | root_log_pinned = false; |
cdd1fedf | 9539 | } |
86e8aa0e | 9540 | if (dest_log_pinned) { |
d4682ba0 FM |
9541 | if (!commit_transaction) { |
9542 | parent = old_dentry->d_parent; | |
9543 | ret = btrfs_log_new_name(trans, BTRFS_I(new_inode), | |
9544 | BTRFS_I(new_dir), parent, | |
9545 | false, &ctx_dest); | |
9546 | if (ret == BTRFS_NEED_LOG_SYNC) | |
9547 | sync_log_dest = true; | |
9548 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
9549 | commit_transaction = true; | |
9550 | ret = 0; | |
9551 | } | |
cdd1fedf | 9552 | btrfs_end_log_trans(dest); |
86e8aa0e | 9553 | dest_log_pinned = false; |
cdd1fedf DF |
9554 | } |
9555 | out_fail: | |
86e8aa0e FM |
9556 | /* |
9557 | * If we have pinned a log and an error happened, we unpin tasks | |
9558 | * trying to sync the log and force them to fallback to a transaction | |
9559 | * commit if the log currently contains any of the inodes involved in | |
9560 | * this rename operation (to ensure we do not persist a log with an | |
9561 | * inconsistent state for any of these inodes or leading to any | |
9562 | * inconsistencies when replayed). If the transaction was aborted, the | |
9563 | * abortion reason is propagated to userspace when attempting to commit | |
9564 | * the transaction. If the log does not contain any of these inodes, we | |
9565 | * allow the tasks to sync it. | |
9566 | */ | |
9567 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
9568 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9569 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9570 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 9571 | (new_inode && |
0f8939b8 | 9572 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 9573 | btrfs_set_log_full_commit(fs_info, trans); |
86e8aa0e FM |
9574 | |
9575 | if (root_log_pinned) { | |
9576 | btrfs_end_log_trans(root); | |
9577 | root_log_pinned = false; | |
9578 | } | |
9579 | if (dest_log_pinned) { | |
9580 | btrfs_end_log_trans(dest); | |
9581 | dest_log_pinned = false; | |
9582 | } | |
9583 | } | |
d4682ba0 FM |
9584 | if (!ret && sync_log_root && !commit_transaction) { |
9585 | ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, | |
9586 | &ctx_root); | |
9587 | if (ret) | |
9588 | commit_transaction = true; | |
9589 | } | |
9590 | if (!ret && sync_log_dest && !commit_transaction) { | |
9591 | ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root, | |
9592 | &ctx_dest); | |
9593 | if (ret) | |
9594 | commit_transaction = true; | |
9595 | } | |
9596 | if (commit_transaction) { | |
9597 | ret = btrfs_commit_transaction(trans); | |
9598 | } else { | |
9599 | int ret2; | |
9600 | ||
9601 | ret2 = btrfs_end_transaction(trans); | |
9602 | ret = ret ? ret : ret2; | |
9603 | } | |
cdd1fedf DF |
9604 | out_notrans: |
9605 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9606 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 9607 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9608 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9609 | |
9610 | return ret; | |
9611 | } | |
9612 | ||
9613 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9614 | struct btrfs_root *root, | |
9615 | struct inode *dir, | |
9616 | struct dentry *dentry) | |
9617 | { | |
9618 | int ret; | |
9619 | struct inode *inode; | |
9620 | u64 objectid; | |
9621 | u64 index; | |
9622 | ||
9623 | ret = btrfs_find_free_ino(root, &objectid); | |
9624 | if (ret) | |
9625 | return ret; | |
9626 | ||
9627 | inode = btrfs_new_inode(trans, root, dir, | |
9628 | dentry->d_name.name, | |
9629 | dentry->d_name.len, | |
4a0cc7ca | 9630 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9631 | objectid, |
9632 | S_IFCHR | WHITEOUT_MODE, | |
9633 | &index); | |
9634 | ||
9635 | if (IS_ERR(inode)) { | |
9636 | ret = PTR_ERR(inode); | |
9637 | return ret; | |
9638 | } | |
9639 | ||
9640 | inode->i_op = &btrfs_special_inode_operations; | |
9641 | init_special_inode(inode, inode->i_mode, | |
9642 | WHITEOUT_DEV); | |
9643 | ||
9644 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9645 | &dentry->d_name); | |
9646 | if (ret) | |
c9901618 | 9647 | goto out; |
cdd1fedf | 9648 | |
cef415af NB |
9649 | ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9650 | BTRFS_I(inode), 0, index); | |
cdd1fedf | 9651 | if (ret) |
c9901618 | 9652 | goto out; |
cdd1fedf DF |
9653 | |
9654 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 9655 | out: |
cdd1fedf | 9656 | unlock_new_inode(inode); |
c9901618 FM |
9657 | if (ret) |
9658 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9659 | iput(inode); |
9660 | ||
c9901618 | 9661 | return ret; |
cdd1fedf DF |
9662 | } |
9663 | ||
d397712b | 9664 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9665 | struct inode *new_dir, struct dentry *new_dentry, |
9666 | unsigned int flags) | |
39279cc3 | 9667 | { |
0b246afa | 9668 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9669 | struct btrfs_trans_handle *trans; |
5062af35 | 9670 | unsigned int trans_num_items; |
39279cc3 | 9671 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9672 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9673 | struct inode *new_inode = d_inode(new_dentry); |
9674 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9675 | u64 index = 0; |
4df27c4d | 9676 | u64 root_objectid; |
39279cc3 | 9677 | int ret; |
4a0cc7ca | 9678 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 9679 | bool log_pinned = false; |
d4682ba0 FM |
9680 | struct btrfs_log_ctx ctx; |
9681 | bool sync_log = false; | |
9682 | bool commit_transaction = false; | |
39279cc3 | 9683 | |
4a0cc7ca | 9684 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9685 | return -EPERM; |
9686 | ||
4df27c4d | 9687 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9688 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9689 | return -EXDEV; |
9690 | ||
33345d01 | 9691 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9692 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9693 | return -ENOTEMPTY; |
5f39d397 | 9694 | |
4df27c4d YZ |
9695 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9696 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9697 | return -ENOTEMPTY; | |
9c52057c CM |
9698 | |
9699 | ||
9700 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9701 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9702 | new_dentry->d_name.name, |
9703 | new_dentry->d_name.len); | |
9704 | ||
9705 | if (ret) { | |
9706 | if (ret == -EEXIST) { | |
9707 | /* we shouldn't get | |
9708 | * eexist without a new_inode */ | |
fae7f21c | 9709 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9710 | return ret; |
9711 | } | |
9712 | } else { | |
9713 | /* maybe -EOVERFLOW */ | |
9714 | return ret; | |
9715 | } | |
9716 | } | |
9717 | ret = 0; | |
9718 | ||
5a3f23d5 | 9719 | /* |
8d875f95 CM |
9720 | * we're using rename to replace one file with another. Start IO on it |
9721 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9722 | */ |
8d875f95 | 9723 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9724 | filemap_flush(old_inode->i_mapping); |
9725 | ||
76dda93c | 9726 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9727 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9728 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
9729 | /* |
9730 | * We want to reserve the absolute worst case amount of items. So if | |
9731 | * both inodes are subvols and we need to unlink them then that would | |
9732 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9733 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9734 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9735 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9736 | * If our rename has the whiteout flag, we need more 5 units for the |
9737 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9738 | * when selinux is enabled). | |
a22285a6 | 9739 | */ |
5062af35 FM |
9740 | trans_num_items = 11; |
9741 | if (flags & RENAME_WHITEOUT) | |
9742 | trans_num_items += 5; | |
9743 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9744 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9745 | ret = PTR_ERR(trans); |
9746 | goto out_notrans; | |
9747 | } | |
76dda93c | 9748 | |
4df27c4d YZ |
9749 | if (dest != root) |
9750 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9751 | |
877574e2 | 9752 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9753 | if (ret) |
9754 | goto out_fail; | |
5a3f23d5 | 9755 | |
67de1176 | 9756 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9757 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9758 | /* force full log commit if subvolume involved. */ |
0b246afa | 9759 | btrfs_set_log_full_commit(fs_info, trans); |
4df27c4d | 9760 | } else { |
c4aba954 FM |
9761 | btrfs_pin_log_trans(root); |
9762 | log_pinned = true; | |
a5719521 YZ |
9763 | ret = btrfs_insert_inode_ref(trans, dest, |
9764 | new_dentry->d_name.name, | |
9765 | new_dentry->d_name.len, | |
33345d01 | 9766 | old_ino, |
4a0cc7ca | 9767 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
9768 | if (ret) |
9769 | goto out_fail; | |
4df27c4d | 9770 | } |
5a3f23d5 | 9771 | |
0c4d2d95 JB |
9772 | inode_inc_iversion(old_dir); |
9773 | inode_inc_iversion(new_dir); | |
9774 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9775 | old_dir->i_ctime = old_dir->i_mtime = |
9776 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 9777 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 9778 | |
12fcfd22 | 9779 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
9780 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9781 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 9782 | |
33345d01 | 9783 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9784 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
401b3b19 | 9785 | ret = btrfs_unlink_subvol(trans, old_dir, root_objectid, |
4df27c4d YZ |
9786 | old_dentry->d_name.name, |
9787 | old_dentry->d_name.len); | |
9788 | } else { | |
4ec5934e NB |
9789 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9790 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
9791 | old_dentry->d_name.name, |
9792 | old_dentry->d_name.len); | |
9793 | if (!ret) | |
9794 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 9795 | } |
79787eaa | 9796 | if (ret) { |
66642832 | 9797 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9798 | goto out_fail; |
9799 | } | |
39279cc3 CM |
9800 | |
9801 | if (new_inode) { | |
0c4d2d95 | 9802 | inode_inc_iversion(new_inode); |
c2050a45 | 9803 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9804 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d YZ |
9805 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
9806 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
401b3b19 | 9807 | ret = btrfs_unlink_subvol(trans, new_dir, root_objectid, |
4df27c4d YZ |
9808 | new_dentry->d_name.name, |
9809 | new_dentry->d_name.len); | |
9810 | BUG_ON(new_inode->i_nlink == 0); | |
9811 | } else { | |
4ec5934e NB |
9812 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9813 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
9814 | new_dentry->d_name.name, |
9815 | new_dentry->d_name.len); | |
9816 | } | |
4ef31a45 | 9817 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
9818 | ret = btrfs_orphan_add(trans, |
9819 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 9820 | if (ret) { |
66642832 | 9821 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9822 | goto out_fail; |
9823 | } | |
39279cc3 | 9824 | } |
aec7477b | 9825 | |
db0a669f | 9826 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 9827 | new_dentry->d_name.name, |
a5719521 | 9828 | new_dentry->d_name.len, 0, index); |
79787eaa | 9829 | if (ret) { |
66642832 | 9830 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9831 | goto out_fail; |
9832 | } | |
39279cc3 | 9833 | |
67de1176 MX |
9834 | if (old_inode->i_nlink == 1) |
9835 | BTRFS_I(old_inode)->dir_index = index; | |
9836 | ||
3dc9e8f7 | 9837 | if (log_pinned) { |
10d9f309 | 9838 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 9839 | |
d4682ba0 FM |
9840 | btrfs_init_log_ctx(&ctx, old_inode); |
9841 | ret = btrfs_log_new_name(trans, BTRFS_I(old_inode), | |
9842 | BTRFS_I(old_dir), parent, | |
9843 | false, &ctx); | |
9844 | if (ret == BTRFS_NEED_LOG_SYNC) | |
9845 | sync_log = true; | |
9846 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
9847 | commit_transaction = true; | |
9848 | ret = 0; | |
4df27c4d | 9849 | btrfs_end_log_trans(root); |
3dc9e8f7 | 9850 | log_pinned = false; |
4df27c4d | 9851 | } |
cdd1fedf DF |
9852 | |
9853 | if (flags & RENAME_WHITEOUT) { | |
9854 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
9855 | old_dentry); | |
9856 | ||
9857 | if (ret) { | |
66642832 | 9858 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9859 | goto out_fail; |
9860 | } | |
4df27c4d | 9861 | } |
39279cc3 | 9862 | out_fail: |
3dc9e8f7 FM |
9863 | /* |
9864 | * If we have pinned the log and an error happened, we unpin tasks | |
9865 | * trying to sync the log and force them to fallback to a transaction | |
9866 | * commit if the log currently contains any of the inodes involved in | |
9867 | * this rename operation (to ensure we do not persist a log with an | |
9868 | * inconsistent state for any of these inodes or leading to any | |
9869 | * inconsistencies when replayed). If the transaction was aborted, the | |
9870 | * abortion reason is propagated to userspace when attempting to commit | |
9871 | * the transaction. If the log does not contain any of these inodes, we | |
9872 | * allow the tasks to sync it. | |
9873 | */ | |
9874 | if (ret && log_pinned) { | |
0f8939b8 NB |
9875 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9876 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9877 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 9878 | (new_inode && |
0f8939b8 | 9879 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 9880 | btrfs_set_log_full_commit(fs_info, trans); |
3dc9e8f7 FM |
9881 | |
9882 | btrfs_end_log_trans(root); | |
9883 | log_pinned = false; | |
9884 | } | |
d4682ba0 FM |
9885 | if (!ret && sync_log) { |
9886 | ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx); | |
9887 | if (ret) | |
9888 | commit_transaction = true; | |
9889 | } | |
9890 | if (commit_transaction) { | |
9891 | ret = btrfs_commit_transaction(trans); | |
9892 | } else { | |
9893 | int ret2; | |
9894 | ||
9895 | ret2 = btrfs_end_transaction(trans); | |
9896 | ret = ret ? ret : ret2; | |
9897 | } | |
b44c59a8 | 9898 | out_notrans: |
33345d01 | 9899 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9900 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 9901 | |
39279cc3 CM |
9902 | return ret; |
9903 | } | |
9904 | ||
80ace85c MS |
9905 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
9906 | struct inode *new_dir, struct dentry *new_dentry, | |
9907 | unsigned int flags) | |
9908 | { | |
cdd1fedf | 9909 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9910 | return -EINVAL; |
9911 | ||
cdd1fedf DF |
9912 | if (flags & RENAME_EXCHANGE) |
9913 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
9914 | new_dentry); | |
9915 | ||
9916 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
9917 | } |
9918 | ||
3a2f8c07 NB |
9919 | struct btrfs_delalloc_work { |
9920 | struct inode *inode; | |
9921 | struct completion completion; | |
9922 | struct list_head list; | |
9923 | struct btrfs_work work; | |
9924 | }; | |
9925 | ||
8ccf6f19 MX |
9926 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9927 | { | |
9928 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9929 | struct inode *inode; |
8ccf6f19 MX |
9930 | |
9931 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9932 | work); | |
9f23e289 | 9933 | inode = delalloc_work->inode; |
30424601 DS |
9934 | filemap_flush(inode->i_mapping); |
9935 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9936 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9937 | filemap_flush(inode->i_mapping); |
8ccf6f19 | 9938 | |
076da91c | 9939 | iput(inode); |
8ccf6f19 MX |
9940 | complete(&delalloc_work->completion); |
9941 | } | |
9942 | ||
3a2f8c07 | 9943 | static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode) |
8ccf6f19 MX |
9944 | { |
9945 | struct btrfs_delalloc_work *work; | |
9946 | ||
100d5702 | 9947 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9948 | if (!work) |
9949 | return NULL; | |
9950 | ||
9951 | init_completion(&work->completion); | |
9952 | INIT_LIST_HEAD(&work->list); | |
9953 | work->inode = inode; | |
9e0af237 LB |
9954 | WARN_ON_ONCE(!inode); |
9955 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
9956 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
9957 | |
9958 | return work; | |
9959 | } | |
9960 | ||
d352ac68 CM |
9961 | /* |
9962 | * some fairly slow code that needs optimization. This walks the list | |
9963 | * of all the inodes with pending delalloc and forces them to disk. | |
9964 | */ | |
3cd24c69 | 9965 | static int start_delalloc_inodes(struct btrfs_root *root, int nr, bool snapshot) |
ea8c2819 | 9966 | { |
ea8c2819 | 9967 | struct btrfs_inode *binode; |
5b21f2ed | 9968 | struct inode *inode; |
8ccf6f19 MX |
9969 | struct btrfs_delalloc_work *work, *next; |
9970 | struct list_head works; | |
1eafa6c7 | 9971 | struct list_head splice; |
8ccf6f19 | 9972 | int ret = 0; |
ea8c2819 | 9973 | |
8ccf6f19 | 9974 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9975 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9976 | |
573bfb72 | 9977 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9978 | spin_lock(&root->delalloc_lock); |
9979 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9980 | while (!list_empty(&splice)) { |
9981 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9982 | delalloc_inodes); |
1eafa6c7 | 9983 | |
eb73c1b7 MX |
9984 | list_move_tail(&binode->delalloc_inodes, |
9985 | &root->delalloc_inodes); | |
5b21f2ed | 9986 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9987 | if (!inode) { |
eb73c1b7 | 9988 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9989 | continue; |
df0af1a5 | 9990 | } |
eb73c1b7 | 9991 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 9992 | |
3cd24c69 EL |
9993 | if (snapshot) |
9994 | set_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
9995 | &binode->runtime_flags); | |
076da91c | 9996 | work = btrfs_alloc_delalloc_work(inode); |
5d99a998 | 9997 | if (!work) { |
4fbb5147 | 9998 | iput(inode); |
1eafa6c7 | 9999 | ret = -ENOMEM; |
a1ecaabb | 10000 | goto out; |
5b21f2ed | 10001 | } |
1eafa6c7 | 10002 | list_add_tail(&work->list, &works); |
a44903ab QW |
10003 | btrfs_queue_work(root->fs_info->flush_workers, |
10004 | &work->work); | |
6c255e67 MX |
10005 | ret++; |
10006 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 10007 | goto out; |
5b21f2ed | 10008 | cond_resched(); |
eb73c1b7 | 10009 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 10010 | } |
eb73c1b7 | 10011 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 10012 | |
a1ecaabb | 10013 | out: |
eb73c1b7 MX |
10014 | list_for_each_entry_safe(work, next, &works, list) { |
10015 | list_del_init(&work->list); | |
40012f96 NB |
10016 | wait_for_completion(&work->completion); |
10017 | kfree(work); | |
eb73c1b7 MX |
10018 | } |
10019 | ||
81f1d390 | 10020 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
10021 | spin_lock(&root->delalloc_lock); |
10022 | list_splice_tail(&splice, &root->delalloc_inodes); | |
10023 | spin_unlock(&root->delalloc_lock); | |
10024 | } | |
573bfb72 | 10025 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
10026 | return ret; |
10027 | } | |
1eafa6c7 | 10028 | |
3cd24c69 | 10029 | int btrfs_start_delalloc_snapshot(struct btrfs_root *root) |
eb73c1b7 | 10030 | { |
0b246afa | 10031 | struct btrfs_fs_info *fs_info = root->fs_info; |
eb73c1b7 | 10032 | int ret; |
1eafa6c7 | 10033 | |
0b246afa | 10034 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10035 | return -EROFS; |
10036 | ||
3cd24c69 | 10037 | ret = start_delalloc_inodes(root, -1, true); |
6c255e67 MX |
10038 | if (ret > 0) |
10039 | ret = 0; | |
eb73c1b7 MX |
10040 | return ret; |
10041 | } | |
10042 | ||
82b3e53b | 10043 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr) |
eb73c1b7 MX |
10044 | { |
10045 | struct btrfs_root *root; | |
10046 | struct list_head splice; | |
10047 | int ret; | |
10048 | ||
2c21b4d7 | 10049 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10050 | return -EROFS; |
10051 | ||
10052 | INIT_LIST_HEAD(&splice); | |
10053 | ||
573bfb72 | 10054 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
10055 | spin_lock(&fs_info->delalloc_root_lock); |
10056 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 10057 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
10058 | root = list_first_entry(&splice, struct btrfs_root, |
10059 | delalloc_root); | |
10060 | root = btrfs_grab_fs_root(root); | |
10061 | BUG_ON(!root); | |
10062 | list_move_tail(&root->delalloc_root, | |
10063 | &fs_info->delalloc_roots); | |
10064 | spin_unlock(&fs_info->delalloc_root_lock); | |
10065 | ||
3cd24c69 | 10066 | ret = start_delalloc_inodes(root, nr, false); |
eb73c1b7 | 10067 | btrfs_put_fs_root(root); |
6c255e67 | 10068 | if (ret < 0) |
eb73c1b7 MX |
10069 | goto out; |
10070 | ||
6c255e67 MX |
10071 | if (nr != -1) { |
10072 | nr -= ret; | |
10073 | WARN_ON(nr < 0); | |
10074 | } | |
eb73c1b7 | 10075 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 10076 | } |
eb73c1b7 | 10077 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 10078 | |
6c255e67 | 10079 | ret = 0; |
eb73c1b7 | 10080 | out: |
81f1d390 | 10081 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
10082 | spin_lock(&fs_info->delalloc_root_lock); |
10083 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
10084 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 10085 | } |
573bfb72 | 10086 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 10087 | return ret; |
ea8c2819 CM |
10088 | } |
10089 | ||
39279cc3 CM |
10090 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
10091 | const char *symname) | |
10092 | { | |
0b246afa | 10093 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
10094 | struct btrfs_trans_handle *trans; |
10095 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10096 | struct btrfs_path *path; | |
10097 | struct btrfs_key key; | |
1832a6d5 | 10098 | struct inode *inode = NULL; |
39279cc3 | 10099 | int err; |
39279cc3 | 10100 | u64 objectid; |
67871254 | 10101 | u64 index = 0; |
39279cc3 CM |
10102 | int name_len; |
10103 | int datasize; | |
5f39d397 | 10104 | unsigned long ptr; |
39279cc3 | 10105 | struct btrfs_file_extent_item *ei; |
5f39d397 | 10106 | struct extent_buffer *leaf; |
39279cc3 | 10107 | |
f06becc4 | 10108 | name_len = strlen(symname); |
0b246afa | 10109 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 10110 | return -ENAMETOOLONG; |
1832a6d5 | 10111 | |
9ed74f2d JB |
10112 | /* |
10113 | * 2 items for inode item and ref | |
10114 | * 2 items for dir items | |
9269d12b FM |
10115 | * 1 item for updating parent inode item |
10116 | * 1 item for the inline extent item | |
9ed74f2d JB |
10117 | * 1 item for xattr if selinux is on |
10118 | */ | |
9269d12b | 10119 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10120 | if (IS_ERR(trans)) |
10121 | return PTR_ERR(trans); | |
1832a6d5 | 10122 | |
581bb050 LZ |
10123 | err = btrfs_find_free_ino(root, &objectid); |
10124 | if (err) | |
10125 | goto out_unlock; | |
10126 | ||
aec7477b | 10127 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
10128 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
10129 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
10130 | if (IS_ERR(inode)) { |
10131 | err = PTR_ERR(inode); | |
32955c54 | 10132 | inode = NULL; |
39279cc3 | 10133 | goto out_unlock; |
7cf96da3 | 10134 | } |
39279cc3 | 10135 | |
ad19db71 CS |
10136 | /* |
10137 | * If the active LSM wants to access the inode during | |
10138 | * d_instantiate it needs these. Smack checks to see | |
10139 | * if the filesystem supports xattrs by looking at the | |
10140 | * ops vector. | |
10141 | */ | |
10142 | inode->i_fop = &btrfs_file_operations; | |
10143 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10144 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10145 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10146 | ||
10147 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10148 | if (err) | |
32955c54 | 10149 | goto out_unlock; |
ad19db71 | 10150 | |
39279cc3 | 10151 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10152 | if (!path) { |
10153 | err = -ENOMEM; | |
32955c54 | 10154 | goto out_unlock; |
d8926bb3 | 10155 | } |
4a0cc7ca | 10156 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 10157 | key.offset = 0; |
962a298f | 10158 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10159 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10160 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10161 | datasize); | |
54aa1f4d | 10162 | if (err) { |
b0839166 | 10163 | btrfs_free_path(path); |
32955c54 | 10164 | goto out_unlock; |
54aa1f4d | 10165 | } |
5f39d397 CM |
10166 | leaf = path->nodes[0]; |
10167 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10168 | struct btrfs_file_extent_item); | |
10169 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10170 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10171 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10172 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10173 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10174 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10175 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10176 | ||
39279cc3 | 10177 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10178 | write_extent_buffer(leaf, symname, ptr, name_len); |
10179 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10180 | btrfs_free_path(path); |
5f39d397 | 10181 | |
39279cc3 | 10182 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10183 | inode_nohighmem(inode); |
4779cc04 | 10184 | inode->i_mapping->a_ops = &btrfs_aops; |
d899e052 | 10185 | inode_set_bytes(inode, name_len); |
6ef06d27 | 10186 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
54aa1f4d | 10187 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
10188 | /* |
10189 | * Last step, add directory indexes for our symlink inode. This is the | |
10190 | * last step to avoid extra cleanup of these indexes if an error happens | |
10191 | * elsewhere above. | |
10192 | */ | |
10193 | if (!err) | |
cef415af NB |
10194 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
10195 | BTRFS_I(inode), 0, index); | |
32955c54 AV |
10196 | if (err) |
10197 | goto out_unlock; | |
b0d5d10f | 10198 | |
1e2e547a | 10199 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
10200 | |
10201 | out_unlock: | |
3a45bb20 | 10202 | btrfs_end_transaction(trans); |
32955c54 | 10203 | if (err && inode) { |
39279cc3 | 10204 | inode_dec_link_count(inode); |
32955c54 | 10205 | discard_new_inode(inode); |
39279cc3 | 10206 | } |
2ff7e61e | 10207 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
10208 | return err; |
10209 | } | |
16432985 | 10210 | |
0af3d00b JB |
10211 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10212 | u64 start, u64 num_bytes, u64 min_size, | |
10213 | loff_t actual_len, u64 *alloc_hint, | |
10214 | struct btrfs_trans_handle *trans) | |
d899e052 | 10215 | { |
0b246afa | 10216 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
10217 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10218 | struct extent_map *em; | |
d899e052 YZ |
10219 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10220 | struct btrfs_key ins; | |
d899e052 | 10221 | u64 cur_offset = start; |
55a61d1d | 10222 | u64 i_size; |
154ea289 | 10223 | u64 cur_bytes; |
0b670dc4 | 10224 | u64 last_alloc = (u64)-1; |
d899e052 | 10225 | int ret = 0; |
0af3d00b | 10226 | bool own_trans = true; |
18513091 | 10227 | u64 end = start + num_bytes - 1; |
d899e052 | 10228 | |
0af3d00b JB |
10229 | if (trans) |
10230 | own_trans = false; | |
d899e052 | 10231 | while (num_bytes > 0) { |
0af3d00b JB |
10232 | if (own_trans) { |
10233 | trans = btrfs_start_transaction(root, 3); | |
10234 | if (IS_ERR(trans)) { | |
10235 | ret = PTR_ERR(trans); | |
10236 | break; | |
10237 | } | |
5a303d5d YZ |
10238 | } |
10239 | ||
ee22184b | 10240 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10241 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10242 | /* |
10243 | * If we are severely fragmented we could end up with really | |
10244 | * small allocations, so if the allocator is returning small | |
10245 | * chunks lets make its job easier by only searching for those | |
10246 | * sized chunks. | |
10247 | */ | |
10248 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
10249 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
10250 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
5a303d5d | 10251 | if (ret) { |
0af3d00b | 10252 | if (own_trans) |
3a45bb20 | 10253 | btrfs_end_transaction(trans); |
a22285a6 | 10254 | break; |
d899e052 | 10255 | } |
0b246afa | 10256 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5a303d5d | 10257 | |
0b670dc4 | 10258 | last_alloc = ins.offset; |
d899e052 YZ |
10259 | ret = insert_reserved_file_extent(trans, inode, |
10260 | cur_offset, ins.objectid, | |
10261 | ins.offset, ins.offset, | |
920bbbfb | 10262 | ins.offset, 0, 0, 0, |
d899e052 | 10263 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 10264 | if (ret) { |
2ff7e61e | 10265 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 10266 | ins.offset, 0); |
66642832 | 10267 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10268 | if (own_trans) |
3a45bb20 | 10269 | btrfs_end_transaction(trans); |
79787eaa JM |
10270 | break; |
10271 | } | |
31193213 | 10272 | |
dcdbc059 | 10273 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 10274 | cur_offset + ins.offset -1, 0); |
5a303d5d | 10275 | |
5dc562c5 JB |
10276 | em = alloc_extent_map(); |
10277 | if (!em) { | |
10278 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10279 | &BTRFS_I(inode)->runtime_flags); | |
10280 | goto next; | |
10281 | } | |
10282 | ||
10283 | em->start = cur_offset; | |
10284 | em->orig_start = cur_offset; | |
10285 | em->len = ins.offset; | |
10286 | em->block_start = ins.objectid; | |
10287 | em->block_len = ins.offset; | |
b4939680 | 10288 | em->orig_block_len = ins.offset; |
cc95bef6 | 10289 | em->ram_bytes = ins.offset; |
0b246afa | 10290 | em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 JB |
10291 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
10292 | em->generation = trans->transid; | |
10293 | ||
10294 | while (1) { | |
10295 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10296 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10297 | write_unlock(&em_tree->lock); |
10298 | if (ret != -EEXIST) | |
10299 | break; | |
dcdbc059 | 10300 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
10301 | cur_offset + ins.offset - 1, |
10302 | 0); | |
10303 | } | |
10304 | free_extent_map(em); | |
10305 | next: | |
d899e052 YZ |
10306 | num_bytes -= ins.offset; |
10307 | cur_offset += ins.offset; | |
efa56464 | 10308 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10309 | |
0c4d2d95 | 10310 | inode_inc_iversion(inode); |
c2050a45 | 10311 | inode->i_ctime = current_time(inode); |
6cbff00f | 10312 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10313 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10314 | (actual_len > inode->i_size) && |
10315 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10316 | if (cur_offset > actual_len) |
55a61d1d | 10317 | i_size = actual_len; |
d1ea6a61 | 10318 | else |
55a61d1d JB |
10319 | i_size = cur_offset; |
10320 | i_size_write(inode, i_size); | |
10321 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10322 | } |
10323 | ||
d899e052 | 10324 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10325 | |
10326 | if (ret) { | |
66642832 | 10327 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10328 | if (own_trans) |
3a45bb20 | 10329 | btrfs_end_transaction(trans); |
79787eaa JM |
10330 | break; |
10331 | } | |
d899e052 | 10332 | |
0af3d00b | 10333 | if (own_trans) |
3a45bb20 | 10334 | btrfs_end_transaction(trans); |
5a303d5d | 10335 | } |
18513091 | 10336 | if (cur_offset < end) |
bc42bda2 | 10337 | btrfs_free_reserved_data_space(inode, NULL, cur_offset, |
18513091 | 10338 | end - cur_offset + 1); |
d899e052 YZ |
10339 | return ret; |
10340 | } | |
10341 | ||
0af3d00b JB |
10342 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10343 | u64 start, u64 num_bytes, u64 min_size, | |
10344 | loff_t actual_len, u64 *alloc_hint) | |
10345 | { | |
10346 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10347 | min_size, actual_len, alloc_hint, | |
10348 | NULL); | |
10349 | } | |
10350 | ||
10351 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10352 | struct btrfs_trans_handle *trans, int mode, | |
10353 | u64 start, u64 num_bytes, u64 min_size, | |
10354 | loff_t actual_len, u64 *alloc_hint) | |
10355 | { | |
10356 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10357 | min_size, actual_len, alloc_hint, trans); | |
10358 | } | |
10359 | ||
e6dcd2dc CM |
10360 | static int btrfs_set_page_dirty(struct page *page) |
10361 | { | |
e6dcd2dc CM |
10362 | return __set_page_dirty_nobuffers(page); |
10363 | } | |
10364 | ||
10556cb2 | 10365 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10366 | { |
b83cc969 | 10367 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10368 | umode_t mode = inode->i_mode; |
b83cc969 | 10369 | |
cb6db4e5 JM |
10370 | if (mask & MAY_WRITE && |
10371 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10372 | if (btrfs_root_readonly(root)) | |
10373 | return -EROFS; | |
10374 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10375 | return -EACCES; | |
10376 | } | |
2830ba7f | 10377 | return generic_permission(inode, mask); |
fdebe2bd | 10378 | } |
39279cc3 | 10379 | |
ef3b9af5 FM |
10380 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10381 | { | |
2ff7e61e | 10382 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10383 | struct btrfs_trans_handle *trans; |
10384 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10385 | struct inode *inode = NULL; | |
10386 | u64 objectid; | |
10387 | u64 index; | |
10388 | int ret = 0; | |
10389 | ||
10390 | /* | |
10391 | * 5 units required for adding orphan entry | |
10392 | */ | |
10393 | trans = btrfs_start_transaction(root, 5); | |
10394 | if (IS_ERR(trans)) | |
10395 | return PTR_ERR(trans); | |
10396 | ||
10397 | ret = btrfs_find_free_ino(root, &objectid); | |
10398 | if (ret) | |
10399 | goto out; | |
10400 | ||
10401 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 10402 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
10403 | if (IS_ERR(inode)) { |
10404 | ret = PTR_ERR(inode); | |
10405 | inode = NULL; | |
10406 | goto out; | |
10407 | } | |
10408 | ||
ef3b9af5 FM |
10409 | inode->i_fop = &btrfs_file_operations; |
10410 | inode->i_op = &btrfs_file_inode_operations; | |
10411 | ||
10412 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10413 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10414 | ||
b0d5d10f CM |
10415 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10416 | if (ret) | |
32955c54 | 10417 | goto out; |
b0d5d10f CM |
10418 | |
10419 | ret = btrfs_update_inode(trans, root, inode); | |
10420 | if (ret) | |
32955c54 | 10421 | goto out; |
73f2e545 | 10422 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 10423 | if (ret) |
32955c54 | 10424 | goto out; |
ef3b9af5 | 10425 | |
5762b5c9 FM |
10426 | /* |
10427 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10428 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10429 | * through: | |
10430 | * | |
10431 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10432 | */ | |
10433 | set_nlink(inode, 1); | |
ef3b9af5 | 10434 | d_tmpfile(dentry, inode); |
32955c54 | 10435 | unlock_new_inode(inode); |
ef3b9af5 | 10436 | mark_inode_dirty(inode); |
ef3b9af5 | 10437 | out: |
3a45bb20 | 10438 | btrfs_end_transaction(trans); |
32955c54 AV |
10439 | if (ret && inode) |
10440 | discard_new_inode(inode); | |
2ff7e61e | 10441 | btrfs_btree_balance_dirty(fs_info); |
ef3b9af5 FM |
10442 | return ret; |
10443 | } | |
10444 | ||
5cdc84bf | 10445 | void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
c6100a4b | 10446 | { |
5cdc84bf | 10447 | struct inode *inode = tree->private_data; |
c6100a4b JB |
10448 | unsigned long index = start >> PAGE_SHIFT; |
10449 | unsigned long end_index = end >> PAGE_SHIFT; | |
10450 | struct page *page; | |
10451 | ||
10452 | while (index <= end_index) { | |
10453 | page = find_get_page(inode->i_mapping, index); | |
10454 | ASSERT(page); /* Pages should be in the extent_io_tree */ | |
10455 | set_page_writeback(page); | |
10456 | put_page(page); | |
10457 | index++; | |
10458 | } | |
10459 | } | |
10460 | ||
ed46ff3d OS |
10461 | #ifdef CONFIG_SWAP |
10462 | /* | |
10463 | * Add an entry indicating a block group or device which is pinned by a | |
10464 | * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a | |
10465 | * negative errno on failure. | |
10466 | */ | |
10467 | static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr, | |
10468 | bool is_block_group) | |
10469 | { | |
10470 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10471 | struct btrfs_swapfile_pin *sp, *entry; | |
10472 | struct rb_node **p; | |
10473 | struct rb_node *parent = NULL; | |
10474 | ||
10475 | sp = kmalloc(sizeof(*sp), GFP_NOFS); | |
10476 | if (!sp) | |
10477 | return -ENOMEM; | |
10478 | sp->ptr = ptr; | |
10479 | sp->inode = inode; | |
10480 | sp->is_block_group = is_block_group; | |
10481 | ||
10482 | spin_lock(&fs_info->swapfile_pins_lock); | |
10483 | p = &fs_info->swapfile_pins.rb_node; | |
10484 | while (*p) { | |
10485 | parent = *p; | |
10486 | entry = rb_entry(parent, struct btrfs_swapfile_pin, node); | |
10487 | if (sp->ptr < entry->ptr || | |
10488 | (sp->ptr == entry->ptr && sp->inode < entry->inode)) { | |
10489 | p = &(*p)->rb_left; | |
10490 | } else if (sp->ptr > entry->ptr || | |
10491 | (sp->ptr == entry->ptr && sp->inode > entry->inode)) { | |
10492 | p = &(*p)->rb_right; | |
10493 | } else { | |
10494 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10495 | kfree(sp); | |
10496 | return 1; | |
10497 | } | |
10498 | } | |
10499 | rb_link_node(&sp->node, parent, p); | |
10500 | rb_insert_color(&sp->node, &fs_info->swapfile_pins); | |
10501 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10502 | return 0; | |
10503 | } | |
10504 | ||
10505 | /* Free all of the entries pinned by this swapfile. */ | |
10506 | static void btrfs_free_swapfile_pins(struct inode *inode) | |
10507 | { | |
10508 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10509 | struct btrfs_swapfile_pin *sp; | |
10510 | struct rb_node *node, *next; | |
10511 | ||
10512 | spin_lock(&fs_info->swapfile_pins_lock); | |
10513 | node = rb_first(&fs_info->swapfile_pins); | |
10514 | while (node) { | |
10515 | next = rb_next(node); | |
10516 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
10517 | if (sp->inode == inode) { | |
10518 | rb_erase(&sp->node, &fs_info->swapfile_pins); | |
10519 | if (sp->is_block_group) | |
10520 | btrfs_put_block_group(sp->ptr); | |
10521 | kfree(sp); | |
10522 | } | |
10523 | node = next; | |
10524 | } | |
10525 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10526 | } | |
10527 | ||
10528 | struct btrfs_swap_info { | |
10529 | u64 start; | |
10530 | u64 block_start; | |
10531 | u64 block_len; | |
10532 | u64 lowest_ppage; | |
10533 | u64 highest_ppage; | |
10534 | unsigned long nr_pages; | |
10535 | int nr_extents; | |
10536 | }; | |
10537 | ||
10538 | static int btrfs_add_swap_extent(struct swap_info_struct *sis, | |
10539 | struct btrfs_swap_info *bsi) | |
10540 | { | |
10541 | unsigned long nr_pages; | |
10542 | u64 first_ppage, first_ppage_reported, next_ppage; | |
10543 | int ret; | |
10544 | ||
10545 | first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT; | |
10546 | next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len, | |
10547 | PAGE_SIZE) >> PAGE_SHIFT; | |
10548 | ||
10549 | if (first_ppage >= next_ppage) | |
10550 | return 0; | |
10551 | nr_pages = next_ppage - first_ppage; | |
10552 | ||
10553 | first_ppage_reported = first_ppage; | |
10554 | if (bsi->start == 0) | |
10555 | first_ppage_reported++; | |
10556 | if (bsi->lowest_ppage > first_ppage_reported) | |
10557 | bsi->lowest_ppage = first_ppage_reported; | |
10558 | if (bsi->highest_ppage < (next_ppage - 1)) | |
10559 | bsi->highest_ppage = next_ppage - 1; | |
10560 | ||
10561 | ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage); | |
10562 | if (ret < 0) | |
10563 | return ret; | |
10564 | bsi->nr_extents += ret; | |
10565 | bsi->nr_pages += nr_pages; | |
10566 | return 0; | |
10567 | } | |
10568 | ||
10569 | static void btrfs_swap_deactivate(struct file *file) | |
10570 | { | |
10571 | struct inode *inode = file_inode(file); | |
10572 | ||
10573 | btrfs_free_swapfile_pins(inode); | |
10574 | atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles); | |
10575 | } | |
10576 | ||
10577 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10578 | sector_t *span) | |
10579 | { | |
10580 | struct inode *inode = file_inode(file); | |
10581 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10582 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
10583 | struct extent_state *cached_state = NULL; | |
10584 | struct extent_map *em = NULL; | |
10585 | struct btrfs_device *device = NULL; | |
10586 | struct btrfs_swap_info bsi = { | |
10587 | .lowest_ppage = (sector_t)-1ULL, | |
10588 | }; | |
10589 | int ret = 0; | |
10590 | u64 isize; | |
10591 | u64 start; | |
10592 | ||
10593 | /* | |
10594 | * If the swap file was just created, make sure delalloc is done. If the | |
10595 | * file changes again after this, the user is doing something stupid and | |
10596 | * we don't really care. | |
10597 | */ | |
10598 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
10599 | if (ret) | |
10600 | return ret; | |
10601 | ||
10602 | /* | |
10603 | * The inode is locked, so these flags won't change after we check them. | |
10604 | */ | |
10605 | if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) { | |
10606 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10607 | return -EINVAL; | |
10608 | } | |
10609 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) { | |
10610 | btrfs_warn(fs_info, "swapfile must not be copy-on-write"); | |
10611 | return -EINVAL; | |
10612 | } | |
10613 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
10614 | btrfs_warn(fs_info, "swapfile must not be checksummed"); | |
10615 | return -EINVAL; | |
10616 | } | |
10617 | ||
10618 | /* | |
10619 | * Balance or device remove/replace/resize can move stuff around from | |
10620 | * under us. The EXCL_OP flag makes sure they aren't running/won't run | |
10621 | * concurrently while we are mapping the swap extents, and | |
10622 | * fs_info->swapfile_pins prevents them from running while the swap file | |
10623 | * is active and moving the extents. Note that this also prevents a | |
10624 | * concurrent device add which isn't actually necessary, but it's not | |
10625 | * really worth the trouble to allow it. | |
10626 | */ | |
10627 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) { | |
10628 | btrfs_warn(fs_info, | |
10629 | "cannot activate swapfile while exclusive operation is running"); | |
10630 | return -EBUSY; | |
10631 | } | |
10632 | /* | |
10633 | * Snapshots can create extents which require COW even if NODATACOW is | |
10634 | * set. We use this counter to prevent snapshots. We must increment it | |
10635 | * before walking the extents because we don't want a concurrent | |
10636 | * snapshot to run after we've already checked the extents. | |
10637 | */ | |
10638 | atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles); | |
10639 | ||
10640 | isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize); | |
10641 | ||
10642 | lock_extent_bits(io_tree, 0, isize - 1, &cached_state); | |
10643 | start = 0; | |
10644 | while (start < isize) { | |
10645 | u64 logical_block_start, physical_block_start; | |
10646 | struct btrfs_block_group_cache *bg; | |
10647 | u64 len = isize - start; | |
10648 | ||
10649 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); | |
10650 | if (IS_ERR(em)) { | |
10651 | ret = PTR_ERR(em); | |
10652 | goto out; | |
10653 | } | |
10654 | ||
10655 | if (em->block_start == EXTENT_MAP_HOLE) { | |
10656 | btrfs_warn(fs_info, "swapfile must not have holes"); | |
10657 | ret = -EINVAL; | |
10658 | goto out; | |
10659 | } | |
10660 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10661 | /* | |
10662 | * It's unlikely we'll ever actually find ourselves | |
10663 | * here, as a file small enough to fit inline won't be | |
10664 | * big enough to store more than the swap header, but in | |
10665 | * case something changes in the future, let's catch it | |
10666 | * here rather than later. | |
10667 | */ | |
10668 | btrfs_warn(fs_info, "swapfile must not be inline"); | |
10669 | ret = -EINVAL; | |
10670 | goto out; | |
10671 | } | |
10672 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10673 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10674 | ret = -EINVAL; | |
10675 | goto out; | |
10676 | } | |
10677 | ||
10678 | logical_block_start = em->block_start + (start - em->start); | |
10679 | len = min(len, em->len - (start - em->start)); | |
10680 | free_extent_map(em); | |
10681 | em = NULL; | |
10682 | ||
10683 | ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL); | |
10684 | if (ret < 0) { | |
10685 | goto out; | |
10686 | } else if (ret) { | |
10687 | ret = 0; | |
10688 | } else { | |
10689 | btrfs_warn(fs_info, | |
10690 | "swapfile must not be copy-on-write"); | |
10691 | ret = -EINVAL; | |
10692 | goto out; | |
10693 | } | |
10694 | ||
10695 | em = btrfs_get_chunk_map(fs_info, logical_block_start, len); | |
10696 | if (IS_ERR(em)) { | |
10697 | ret = PTR_ERR(em); | |
10698 | goto out; | |
10699 | } | |
10700 | ||
10701 | if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
10702 | btrfs_warn(fs_info, | |
10703 | "swapfile must have single data profile"); | |
10704 | ret = -EINVAL; | |
10705 | goto out; | |
10706 | } | |
10707 | ||
10708 | if (device == NULL) { | |
10709 | device = em->map_lookup->stripes[0].dev; | |
10710 | ret = btrfs_add_swapfile_pin(inode, device, false); | |
10711 | if (ret == 1) | |
10712 | ret = 0; | |
10713 | else if (ret) | |
10714 | goto out; | |
10715 | } else if (device != em->map_lookup->stripes[0].dev) { | |
10716 | btrfs_warn(fs_info, "swapfile must be on one device"); | |
10717 | ret = -EINVAL; | |
10718 | goto out; | |
10719 | } | |
10720 | ||
10721 | physical_block_start = (em->map_lookup->stripes[0].physical + | |
10722 | (logical_block_start - em->start)); | |
10723 | len = min(len, em->len - (logical_block_start - em->start)); | |
10724 | free_extent_map(em); | |
10725 | em = NULL; | |
10726 | ||
10727 | bg = btrfs_lookup_block_group(fs_info, logical_block_start); | |
10728 | if (!bg) { | |
10729 | btrfs_warn(fs_info, | |
10730 | "could not find block group containing swapfile"); | |
10731 | ret = -EINVAL; | |
10732 | goto out; | |
10733 | } | |
10734 | ||
10735 | ret = btrfs_add_swapfile_pin(inode, bg, true); | |
10736 | if (ret) { | |
10737 | btrfs_put_block_group(bg); | |
10738 | if (ret == 1) | |
10739 | ret = 0; | |
10740 | else | |
10741 | goto out; | |
10742 | } | |
10743 | ||
10744 | if (bsi.block_len && | |
10745 | bsi.block_start + bsi.block_len == physical_block_start) { | |
10746 | bsi.block_len += len; | |
10747 | } else { | |
10748 | if (bsi.block_len) { | |
10749 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10750 | if (ret) | |
10751 | goto out; | |
10752 | } | |
10753 | bsi.start = start; | |
10754 | bsi.block_start = physical_block_start; | |
10755 | bsi.block_len = len; | |
10756 | } | |
10757 | ||
10758 | start += len; | |
10759 | } | |
10760 | ||
10761 | if (bsi.block_len) | |
10762 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10763 | ||
10764 | out: | |
10765 | if (!IS_ERR_OR_NULL(em)) | |
10766 | free_extent_map(em); | |
10767 | ||
10768 | unlock_extent_cached(io_tree, 0, isize - 1, &cached_state); | |
10769 | ||
10770 | if (ret) | |
10771 | btrfs_swap_deactivate(file); | |
10772 | ||
10773 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); | |
10774 | ||
10775 | if (ret) | |
10776 | return ret; | |
10777 | ||
10778 | if (device) | |
10779 | sis->bdev = device->bdev; | |
10780 | *span = bsi.highest_ppage - bsi.lowest_ppage + 1; | |
10781 | sis->max = bsi.nr_pages; | |
10782 | sis->pages = bsi.nr_pages - 1; | |
10783 | sis->highest_bit = bsi.nr_pages - 1; | |
10784 | return bsi.nr_extents; | |
10785 | } | |
10786 | #else | |
10787 | static void btrfs_swap_deactivate(struct file *file) | |
10788 | { | |
10789 | } | |
10790 | ||
10791 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10792 | sector_t *span) | |
10793 | { | |
10794 | return -EOPNOTSUPP; | |
10795 | } | |
10796 | #endif | |
10797 | ||
6e1d5dcc | 10798 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10799 | .getattr = btrfs_getattr, |
39279cc3 CM |
10800 | .lookup = btrfs_lookup, |
10801 | .create = btrfs_create, | |
10802 | .unlink = btrfs_unlink, | |
10803 | .link = btrfs_link, | |
10804 | .mkdir = btrfs_mkdir, | |
10805 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10806 | .rename = btrfs_rename2, |
39279cc3 CM |
10807 | .symlink = btrfs_symlink, |
10808 | .setattr = btrfs_setattr, | |
618e21d5 | 10809 | .mknod = btrfs_mknod, |
5103e947 | 10810 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10811 | .permission = btrfs_permission, |
4e34e719 | 10812 | .get_acl = btrfs_get_acl, |
996a710d | 10813 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10814 | .update_time = btrfs_update_time, |
ef3b9af5 | 10815 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10816 | }; |
6e1d5dcc | 10817 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10818 | .lookup = btrfs_lookup, |
fdebe2bd | 10819 | .permission = btrfs_permission, |
93fd63c2 | 10820 | .update_time = btrfs_update_time, |
39279cc3 | 10821 | }; |
76dda93c | 10822 | |
828c0950 | 10823 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10824 | .llseek = generic_file_llseek, |
10825 | .read = generic_read_dir, | |
02dbfc99 | 10826 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 10827 | .open = btrfs_opendir, |
34287aa3 | 10828 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10829 | #ifdef CONFIG_COMPAT |
4c63c245 | 10830 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10831 | #endif |
6bf13c0c | 10832 | .release = btrfs_release_file, |
e02119d5 | 10833 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10834 | }; |
10835 | ||
20e5506b | 10836 | static const struct extent_io_ops btrfs_extent_io_ops = { |
4d53dddb | 10837 | /* mandatory callbacks */ |
065631f6 | 10838 | .submit_bio_hook = btrfs_submit_bio_hook, |
07157aac CM |
10839 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
10840 | }; | |
10841 | ||
35054394 CM |
10842 | /* |
10843 | * btrfs doesn't support the bmap operation because swapfiles | |
10844 | * use bmap to make a mapping of extents in the file. They assume | |
10845 | * these extents won't change over the life of the file and they | |
10846 | * use the bmap result to do IO directly to the drive. | |
10847 | * | |
10848 | * the btrfs bmap call would return logical addresses that aren't | |
10849 | * suitable for IO and they also will change frequently as COW | |
10850 | * operations happen. So, swapfile + btrfs == corruption. | |
10851 | * | |
10852 | * For now we're avoiding this by dropping bmap. | |
10853 | */ | |
7f09410b | 10854 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10855 | .readpage = btrfs_readpage, |
10856 | .writepage = btrfs_writepage, | |
b293f02e | 10857 | .writepages = btrfs_writepages, |
3ab2fb5a | 10858 | .readpages = btrfs_readpages, |
16432985 | 10859 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10860 | .invalidatepage = btrfs_invalidatepage, |
10861 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 10862 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10863 | .error_remove_page = generic_error_remove_page, |
ed46ff3d OS |
10864 | .swap_activate = btrfs_swap_activate, |
10865 | .swap_deactivate = btrfs_swap_deactivate, | |
39279cc3 CM |
10866 | }; |
10867 | ||
6e1d5dcc | 10868 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10869 | .getattr = btrfs_getattr, |
10870 | .setattr = btrfs_setattr, | |
5103e947 | 10871 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10872 | .permission = btrfs_permission, |
1506fcc8 | 10873 | .fiemap = btrfs_fiemap, |
4e34e719 | 10874 | .get_acl = btrfs_get_acl, |
996a710d | 10875 | .set_acl = btrfs_set_acl, |
e41f941a | 10876 | .update_time = btrfs_update_time, |
39279cc3 | 10877 | }; |
6e1d5dcc | 10878 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10879 | .getattr = btrfs_getattr, |
10880 | .setattr = btrfs_setattr, | |
fdebe2bd | 10881 | .permission = btrfs_permission, |
33268eaf | 10882 | .listxattr = btrfs_listxattr, |
4e34e719 | 10883 | .get_acl = btrfs_get_acl, |
996a710d | 10884 | .set_acl = btrfs_set_acl, |
e41f941a | 10885 | .update_time = btrfs_update_time, |
618e21d5 | 10886 | }; |
6e1d5dcc | 10887 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10888 | .get_link = page_get_link, |
f209561a | 10889 | .getattr = btrfs_getattr, |
22c44fe6 | 10890 | .setattr = btrfs_setattr, |
fdebe2bd | 10891 | .permission = btrfs_permission, |
0279b4cd | 10892 | .listxattr = btrfs_listxattr, |
e41f941a | 10893 | .update_time = btrfs_update_time, |
39279cc3 | 10894 | }; |
76dda93c | 10895 | |
82d339d9 | 10896 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
10897 | .d_delete = btrfs_dentry_delete, |
10898 | }; |