]>
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 CM |
16 | #include <linux/backing-dev.h> |
17 | #include <linux/mpage.h> | |
18 | #include <linux/swap.h> | |
19 | #include <linux/writeback.h> | |
39279cc3 | 20 | #include <linux/compat.h> |
9ebefb18 | 21 | #include <linux/bit_spinlock.h> |
5103e947 | 22 | #include <linux/xattr.h> |
33268eaf | 23 | #include <linux/posix_acl.h> |
d899e052 | 24 | #include <linux/falloc.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
7a36ddec | 26 | #include <linux/ratelimit.h> |
22c44fe6 | 27 | #include <linux/mount.h> |
55e301fd | 28 | #include <linux/btrfs.h> |
53b381b3 | 29 | #include <linux/blkdev.h> |
f23b5a59 | 30 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 31 | #include <linux/uio.h> |
69fe2d75 | 32 | #include <linux/magic.h> |
ae5e165d | 33 | #include <linux/iversion.h> |
92d32170 | 34 | #include <asm/unaligned.h> |
39279cc3 CM |
35 | #include "ctree.h" |
36 | #include "disk-io.h" | |
37 | #include "transaction.h" | |
38 | #include "btrfs_inode.h" | |
39279cc3 | 39 | #include "print-tree.h" |
e6dcd2dc | 40 | #include "ordered-data.h" |
95819c05 | 41 | #include "xattr.h" |
e02119d5 | 42 | #include "tree-log.h" |
4a54c8c1 | 43 | #include "volumes.h" |
c8b97818 | 44 | #include "compression.h" |
b4ce94de | 45 | #include "locking.h" |
dc89e982 | 46 | #include "free-space-cache.h" |
581bb050 | 47 | #include "inode-map.h" |
38c227d8 | 48 | #include "backref.h" |
63541927 | 49 | #include "props.h" |
31193213 | 50 | #include "qgroup.h" |
dda3245e | 51 | #include "dedupe.h" |
39279cc3 CM |
52 | |
53 | struct btrfs_iget_args { | |
90d3e592 | 54 | struct btrfs_key *location; |
39279cc3 CM |
55 | struct btrfs_root *root; |
56 | }; | |
57 | ||
f28a4928 | 58 | struct btrfs_dio_data { |
f28a4928 FM |
59 | u64 reserve; |
60 | u64 unsubmitted_oe_range_start; | |
61 | u64 unsubmitted_oe_range_end; | |
4aaedfb0 | 62 | int overwrite; |
f28a4928 FM |
63 | }; |
64 | ||
6e1d5dcc AD |
65 | static const struct inode_operations btrfs_dir_inode_operations; |
66 | static const struct inode_operations btrfs_symlink_inode_operations; | |
67 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
68 | static const struct inode_operations btrfs_special_inode_operations; | |
69 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
70 | static const struct address_space_operations btrfs_aops; |
71 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 72 | static const struct file_operations btrfs_dir_file_operations; |
20e5506b | 73 | static const struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
74 | |
75 | static struct kmem_cache *btrfs_inode_cachep; | |
76 | struct kmem_cache *btrfs_trans_handle_cachep; | |
39279cc3 | 77 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 78 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
79 | |
80 | #define S_SHIFT 12 | |
4d4ab6d6 | 81 | static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { |
39279cc3 CM |
82 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, |
83 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
84 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
85 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
86 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
87 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
88 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
89 | }; | |
90 | ||
3972f260 | 91 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
213e8c55 | 92 | static int btrfs_truncate(struct inode *inode, bool skip_writeback); |
5fd02043 | 93 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
94 | static noinline int cow_file_range(struct inode *inode, |
95 | struct page *locked_page, | |
dda3245e WX |
96 | u64 start, u64 end, u64 delalloc_end, |
97 | int *page_started, unsigned long *nr_written, | |
98 | int unlock, struct btrfs_dedupe_hash *hash); | |
6f9994db LB |
99 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, |
100 | u64 orig_start, u64 block_start, | |
101 | u64 block_len, u64 orig_block_len, | |
102 | u64 ram_bytes, int compress_type, | |
103 | int type); | |
7b128766 | 104 | |
52427260 QW |
105 | static void __endio_write_update_ordered(struct inode *inode, |
106 | const u64 offset, const u64 bytes, | |
107 | const bool uptodate); | |
108 | ||
109 | /* | |
110 | * Cleanup all submitted ordered extents in specified range to handle errors | |
111 | * from the fill_dellaloc() callback. | |
112 | * | |
113 | * NOTE: caller must ensure that when an error happens, it can not call | |
114 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
115 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
116 | * to be released, which we want to happen only when finishing the ordered | |
117 | * extent (btrfs_finish_ordered_io()). Also note that the caller of the | |
118 | * fill_delalloc() callback already does proper cleanup for the first page of | |
119 | * the range, that is, it invokes the callback writepage_end_io_hook() for the | |
120 | * range of the first page. | |
121 | */ | |
122 | static inline void btrfs_cleanup_ordered_extents(struct inode *inode, | |
123 | const u64 offset, | |
124 | const u64 bytes) | |
125 | { | |
63d71450 NA |
126 | unsigned long index = offset >> PAGE_SHIFT; |
127 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
128 | struct page *page; | |
129 | ||
130 | while (index <= end_index) { | |
131 | page = find_get_page(inode->i_mapping, index); | |
132 | index++; | |
133 | if (!page) | |
134 | continue; | |
135 | ClearPagePrivate2(page); | |
136 | put_page(page); | |
137 | } | |
52427260 QW |
138 | return __endio_write_update_ordered(inode, offset + PAGE_SIZE, |
139 | bytes - PAGE_SIZE, false); | |
140 | } | |
141 | ||
48a3b636 | 142 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 143 | |
6a3891c5 JB |
144 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
145 | void btrfs_test_inode_set_ops(struct inode *inode) | |
146 | { | |
147 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
148 | } | |
149 | #endif | |
150 | ||
f34f57a3 | 151 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
152 | struct inode *inode, struct inode *dir, |
153 | const struct qstr *qstr) | |
0279b4cd JO |
154 | { |
155 | int err; | |
156 | ||
f34f57a3 | 157 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 158 | if (!err) |
2a7dba39 | 159 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
160 | return err; |
161 | } | |
162 | ||
c8b97818 CM |
163 | /* |
164 | * this does all the hard work for inserting an inline extent into | |
165 | * the btree. The caller should have done a btrfs_drop_extents so that | |
166 | * no overlapping inline items exist in the btree | |
167 | */ | |
40f76580 | 168 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 169 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
170 | struct btrfs_root *root, struct inode *inode, |
171 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 172 | int compress_type, |
c8b97818 CM |
173 | struct page **compressed_pages) |
174 | { | |
c8b97818 CM |
175 | struct extent_buffer *leaf; |
176 | struct page *page = NULL; | |
177 | char *kaddr; | |
178 | unsigned long ptr; | |
179 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
180 | int ret; |
181 | size_t cur_size = size; | |
c8b97818 | 182 | unsigned long offset; |
c8b97818 | 183 | |
fe3f566c | 184 | if (compressed_size && compressed_pages) |
c8b97818 | 185 | cur_size = compressed_size; |
c8b97818 | 186 | |
1acae57b | 187 | inode_add_bytes(inode, size); |
c8b97818 | 188 | |
1acae57b FDBM |
189 | if (!extent_inserted) { |
190 | struct btrfs_key key; | |
191 | size_t datasize; | |
c8b97818 | 192 | |
4a0cc7ca | 193 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b | 194 | key.offset = start; |
962a298f | 195 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 196 | |
1acae57b FDBM |
197 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
198 | path->leave_spinning = 1; | |
199 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
200 | datasize); | |
79b4f4c6 | 201 | if (ret) |
1acae57b | 202 | goto fail; |
c8b97818 CM |
203 | } |
204 | leaf = path->nodes[0]; | |
205 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
206 | struct btrfs_file_extent_item); | |
207 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
208 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
209 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
210 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
211 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
212 | ptr = btrfs_file_extent_inline_start(ei); | |
213 | ||
261507a0 | 214 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
215 | struct page *cpage; |
216 | int i = 0; | |
d397712b | 217 | while (compressed_size > 0) { |
c8b97818 | 218 | cpage = compressed_pages[i]; |
5b050f04 | 219 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 220 | PAGE_SIZE); |
c8b97818 | 221 | |
7ac687d9 | 222 | kaddr = kmap_atomic(cpage); |
c8b97818 | 223 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 224 | kunmap_atomic(kaddr); |
c8b97818 CM |
225 | |
226 | i++; | |
227 | ptr += cur_size; | |
228 | compressed_size -= cur_size; | |
229 | } | |
230 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 231 | compress_type); |
c8b97818 CM |
232 | } else { |
233 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 234 | start >> PAGE_SHIFT); |
c8b97818 | 235 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 236 | kaddr = kmap_atomic(page); |
09cbfeaf | 237 | offset = start & (PAGE_SIZE - 1); |
c8b97818 | 238 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 239 | kunmap_atomic(kaddr); |
09cbfeaf | 240 | put_page(page); |
c8b97818 CM |
241 | } |
242 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 243 | btrfs_release_path(path); |
c8b97818 | 244 | |
c2167754 YZ |
245 | /* |
246 | * we're an inline extent, so nobody can | |
247 | * extend the file past i_size without locking | |
248 | * a page we already have locked. | |
249 | * | |
250 | * We must do any isize and inode updates | |
251 | * before we unlock the pages. Otherwise we | |
252 | * could end up racing with unlink. | |
253 | */ | |
c8b97818 | 254 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 255 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 256 | |
c8b97818 | 257 | fail: |
79b4f4c6 | 258 | return ret; |
c8b97818 CM |
259 | } |
260 | ||
261 | ||
262 | /* | |
263 | * conditionally insert an inline extent into the file. This | |
264 | * does the checks required to make sure the data is small enough | |
265 | * to fit as an inline extent. | |
266 | */ | |
d02c0e20 | 267 | static noinline int cow_file_range_inline(struct inode *inode, u64 start, |
00361589 JB |
268 | u64 end, size_t compressed_size, |
269 | int compress_type, | |
270 | struct page **compressed_pages) | |
c8b97818 | 271 | { |
d02c0e20 | 272 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0b246afa | 273 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 274 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
275 | u64 isize = i_size_read(inode); |
276 | u64 actual_end = min(end + 1, isize); | |
277 | u64 inline_len = actual_end - start; | |
0b246afa | 278 | u64 aligned_end = ALIGN(end, fs_info->sectorsize); |
c8b97818 CM |
279 | u64 data_len = inline_len; |
280 | int ret; | |
1acae57b FDBM |
281 | struct btrfs_path *path; |
282 | int extent_inserted = 0; | |
283 | u32 extent_item_size; | |
c8b97818 CM |
284 | |
285 | if (compressed_size) | |
286 | data_len = compressed_size; | |
287 | ||
288 | if (start > 0 || | |
0b246afa JM |
289 | actual_end > fs_info->sectorsize || |
290 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || | |
c8b97818 | 291 | (!compressed_size && |
0b246afa | 292 | (actual_end & (fs_info->sectorsize - 1)) == 0) || |
c8b97818 | 293 | end + 1 < isize || |
0b246afa | 294 | data_len > fs_info->max_inline) { |
c8b97818 CM |
295 | return 1; |
296 | } | |
297 | ||
1acae57b FDBM |
298 | path = btrfs_alloc_path(); |
299 | if (!path) | |
300 | return -ENOMEM; | |
301 | ||
00361589 | 302 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
303 | if (IS_ERR(trans)) { |
304 | btrfs_free_path(path); | |
00361589 | 305 | return PTR_ERR(trans); |
1acae57b | 306 | } |
69fe2d75 | 307 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
00361589 | 308 | |
1acae57b FDBM |
309 | if (compressed_size && compressed_pages) |
310 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
311 | compressed_size); | |
312 | else | |
313 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
314 | inline_len); | |
315 | ||
316 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
317 | start, aligned_end, NULL, | |
318 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 | 319 | if (ret) { |
66642832 | 320 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
321 | goto out; |
322 | } | |
c8b97818 CM |
323 | |
324 | if (isize > actual_end) | |
325 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
326 | ret = insert_inline_extent(trans, path, extent_inserted, |
327 | root, inode, start, | |
c8b97818 | 328 | inline_len, compressed_size, |
fe3f566c | 329 | compress_type, compressed_pages); |
2adcac1a | 330 | if (ret && ret != -ENOSPC) { |
66642832 | 331 | btrfs_abort_transaction(trans, ret); |
00361589 | 332 | goto out; |
2adcac1a | 333 | } else if (ret == -ENOSPC) { |
00361589 JB |
334 | ret = 1; |
335 | goto out; | |
79787eaa | 336 | } |
2adcac1a | 337 | |
bdc20e67 | 338 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
dcdbc059 | 339 | btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0); |
00361589 | 340 | out: |
94ed938a QW |
341 | /* |
342 | * Don't forget to free the reserved space, as for inlined extent | |
343 | * it won't count as data extent, free them directly here. | |
344 | * And at reserve time, it's always aligned to page size, so | |
345 | * just free one page here. | |
346 | */ | |
bc42bda2 | 347 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 348 | btrfs_free_path(path); |
3a45bb20 | 349 | btrfs_end_transaction(trans); |
00361589 | 350 | return ret; |
c8b97818 CM |
351 | } |
352 | ||
771ed689 CM |
353 | struct async_extent { |
354 | u64 start; | |
355 | u64 ram_size; | |
356 | u64 compressed_size; | |
357 | struct page **pages; | |
358 | unsigned long nr_pages; | |
261507a0 | 359 | int compress_type; |
771ed689 CM |
360 | struct list_head list; |
361 | }; | |
362 | ||
363 | struct async_cow { | |
364 | struct inode *inode; | |
365 | struct btrfs_root *root; | |
366 | struct page *locked_page; | |
367 | u64 start; | |
368 | u64 end; | |
f82b7359 | 369 | unsigned int write_flags; |
771ed689 CM |
370 | struct list_head extents; |
371 | struct btrfs_work work; | |
372 | }; | |
373 | ||
374 | static noinline int add_async_extent(struct async_cow *cow, | |
375 | u64 start, u64 ram_size, | |
376 | u64 compressed_size, | |
377 | struct page **pages, | |
261507a0 LZ |
378 | unsigned long nr_pages, |
379 | int compress_type) | |
771ed689 CM |
380 | { |
381 | struct async_extent *async_extent; | |
382 | ||
383 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 384 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
385 | async_extent->start = start; |
386 | async_extent->ram_size = ram_size; | |
387 | async_extent->compressed_size = compressed_size; | |
388 | async_extent->pages = pages; | |
389 | async_extent->nr_pages = nr_pages; | |
261507a0 | 390 | async_extent->compress_type = compress_type; |
771ed689 CM |
391 | list_add_tail(&async_extent->list, &cow->extents); |
392 | return 0; | |
393 | } | |
394 | ||
c2fcdcdf | 395 | static inline int inode_need_compress(struct inode *inode, u64 start, u64 end) |
f79707b0 | 396 | { |
0b246afa | 397 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
f79707b0 WS |
398 | |
399 | /* force compress */ | |
0b246afa | 400 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 401 | return 1; |
eec63c65 DS |
402 | /* defrag ioctl */ |
403 | if (BTRFS_I(inode)->defrag_compress) | |
404 | return 1; | |
f79707b0 WS |
405 | /* bad compression ratios */ |
406 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
407 | return 0; | |
0b246afa | 408 | if (btrfs_test_opt(fs_info, COMPRESS) || |
f79707b0 | 409 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || |
b52aa8c9 | 410 | BTRFS_I(inode)->prop_compress) |
c2fcdcdf | 411 | return btrfs_compress_heuristic(inode, start, end); |
f79707b0 WS |
412 | return 0; |
413 | } | |
414 | ||
6158e1ce | 415 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
26d30f85 AJ |
416 | u64 start, u64 end, u64 num_bytes, u64 small_write) |
417 | { | |
418 | /* If this is a small write inside eof, kick off a defrag */ | |
419 | if (num_bytes < small_write && | |
6158e1ce | 420 | (start > 0 || end + 1 < inode->disk_i_size)) |
26d30f85 AJ |
421 | btrfs_add_inode_defrag(NULL, inode); |
422 | } | |
423 | ||
d352ac68 | 424 | /* |
771ed689 CM |
425 | * we create compressed extents in two phases. The first |
426 | * phase compresses a range of pages that have already been | |
427 | * locked (both pages and state bits are locked). | |
c8b97818 | 428 | * |
771ed689 CM |
429 | * This is done inside an ordered work queue, and the compression |
430 | * is spread across many cpus. The actual IO submission is step | |
431 | * two, and the ordered work queue takes care of making sure that | |
432 | * happens in the same order things were put onto the queue by | |
433 | * writepages and friends. | |
c8b97818 | 434 | * |
771ed689 CM |
435 | * If this code finds it can't get good compression, it puts an |
436 | * entry onto the work queue to write the uncompressed bytes. This | |
437 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
438 | * are written in the same order that the flusher thread sent them |
439 | * down. | |
d352ac68 | 440 | */ |
c44f649e | 441 | static noinline void compress_file_range(struct inode *inode, |
771ed689 CM |
442 | struct page *locked_page, |
443 | u64 start, u64 end, | |
444 | struct async_cow *async_cow, | |
445 | int *num_added) | |
b888db2b | 446 | { |
0b246afa | 447 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
0b246afa | 448 | u64 blocksize = fs_info->sectorsize; |
c8b97818 | 449 | u64 actual_end; |
42dc7bab | 450 | u64 isize = i_size_read(inode); |
e6dcd2dc | 451 | int ret = 0; |
c8b97818 CM |
452 | struct page **pages = NULL; |
453 | unsigned long nr_pages; | |
c8b97818 CM |
454 | unsigned long total_compressed = 0; |
455 | unsigned long total_in = 0; | |
c8b97818 CM |
456 | int i; |
457 | int will_compress; | |
0b246afa | 458 | int compress_type = fs_info->compress_type; |
4adaa611 | 459 | int redirty = 0; |
b888db2b | 460 | |
6158e1ce NB |
461 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
462 | SZ_16K); | |
4cb5300b | 463 | |
42dc7bab | 464 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
465 | again: |
466 | will_compress = 0; | |
09cbfeaf | 467 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
468 | BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0); |
469 | nr_pages = min_t(unsigned long, nr_pages, | |
470 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 471 | |
f03d9301 CM |
472 | /* |
473 | * we don't want to send crud past the end of i_size through | |
474 | * compression, that's just a waste of CPU time. So, if the | |
475 | * end of the file is before the start of our current | |
476 | * requested range of bytes, we bail out to the uncompressed | |
477 | * cleanup code that can deal with all of this. | |
478 | * | |
479 | * It isn't really the fastest way to fix things, but this is a | |
480 | * very uncommon corner. | |
481 | */ | |
482 | if (actual_end <= start) | |
483 | goto cleanup_and_bail_uncompressed; | |
484 | ||
c8b97818 CM |
485 | total_compressed = actual_end - start; |
486 | ||
4bcbb332 SW |
487 | /* |
488 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 489 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
490 | */ |
491 | if (total_compressed <= blocksize && | |
492 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
493 | goto cleanup_and_bail_uncompressed; | |
494 | ||
069eac78 DS |
495 | total_compressed = min_t(unsigned long, total_compressed, |
496 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
497 | total_in = 0; |
498 | ret = 0; | |
db94535d | 499 | |
771ed689 CM |
500 | /* |
501 | * we do compression for mount -o compress and when the | |
502 | * inode has not been flagged as nocompress. This flag can | |
503 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 504 | */ |
c2fcdcdf | 505 | if (inode_need_compress(inode, start, end)) { |
c8b97818 | 506 | WARN_ON(pages); |
31e818fe | 507 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
508 | if (!pages) { |
509 | /* just bail out to the uncompressed code */ | |
510 | goto cont; | |
511 | } | |
c8b97818 | 512 | |
eec63c65 DS |
513 | if (BTRFS_I(inode)->defrag_compress) |
514 | compress_type = BTRFS_I(inode)->defrag_compress; | |
515 | else if (BTRFS_I(inode)->prop_compress) | |
b52aa8c9 | 516 | compress_type = BTRFS_I(inode)->prop_compress; |
261507a0 | 517 | |
4adaa611 CM |
518 | /* |
519 | * we need to call clear_page_dirty_for_io on each | |
520 | * page in the range. Otherwise applications with the file | |
521 | * mmap'd can wander in and change the page contents while | |
522 | * we are compressing them. | |
523 | * | |
524 | * If the compression fails for any reason, we set the pages | |
525 | * dirty again later on. | |
e9679de3 TT |
526 | * |
527 | * Note that the remaining part is redirtied, the start pointer | |
528 | * has moved, the end is the original one. | |
4adaa611 | 529 | */ |
e9679de3 TT |
530 | if (!redirty) { |
531 | extent_range_clear_dirty_for_io(inode, start, end); | |
532 | redirty = 1; | |
533 | } | |
f51d2b59 DS |
534 | |
535 | /* Compression level is applied here and only here */ | |
536 | ret = btrfs_compress_pages( | |
537 | compress_type | (fs_info->compress_level << 4), | |
261507a0 | 538 | inode->i_mapping, start, |
38c31464 | 539 | pages, |
4d3a800e | 540 | &nr_pages, |
261507a0 | 541 | &total_in, |
e5d74902 | 542 | &total_compressed); |
c8b97818 CM |
543 | |
544 | if (!ret) { | |
545 | unsigned long offset = total_compressed & | |
09cbfeaf | 546 | (PAGE_SIZE - 1); |
4d3a800e | 547 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
548 | char *kaddr; |
549 | ||
550 | /* zero the tail end of the last page, we might be | |
551 | * sending it down to disk | |
552 | */ | |
553 | if (offset) { | |
7ac687d9 | 554 | kaddr = kmap_atomic(page); |
c8b97818 | 555 | memset(kaddr + offset, 0, |
09cbfeaf | 556 | PAGE_SIZE - offset); |
7ac687d9 | 557 | kunmap_atomic(kaddr); |
c8b97818 CM |
558 | } |
559 | will_compress = 1; | |
560 | } | |
561 | } | |
560f7d75 | 562 | cont: |
c8b97818 CM |
563 | if (start == 0) { |
564 | /* lets try to make an inline extent */ | |
6018ba0a | 565 | if (ret || total_in < actual_end) { |
c8b97818 | 566 | /* we didn't compress the entire range, try |
771ed689 | 567 | * to make an uncompressed inline extent. |
c8b97818 | 568 | */ |
d02c0e20 NB |
569 | ret = cow_file_range_inline(inode, start, end, 0, |
570 | BTRFS_COMPRESS_NONE, NULL); | |
c8b97818 | 571 | } else { |
771ed689 | 572 | /* try making a compressed inline extent */ |
d02c0e20 | 573 | ret = cow_file_range_inline(inode, start, end, |
fe3f566c LZ |
574 | total_compressed, |
575 | compress_type, pages); | |
c8b97818 | 576 | } |
79787eaa | 577 | if (ret <= 0) { |
151a41bc | 578 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
579 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
580 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 FM |
581 | unsigned long page_error_op; |
582 | ||
e6eb4314 | 583 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 584 | |
771ed689 | 585 | /* |
79787eaa JM |
586 | * inline extent creation worked or returned error, |
587 | * we don't need to create any more async work items. | |
588 | * Unlock and free up our temp pages. | |
8b62f87b JB |
589 | * |
590 | * We use DO_ACCOUNTING here because we need the | |
591 | * delalloc_release_metadata to be done _after_ we drop | |
592 | * our outstanding extent for clearing delalloc for this | |
593 | * range. | |
771ed689 | 594 | */ |
ba8b04c1 QW |
595 | extent_clear_unlock_delalloc(inode, start, end, end, |
596 | NULL, clear_flags, | |
597 | PAGE_UNLOCK | | |
c2790a2e JB |
598 | PAGE_CLEAR_DIRTY | |
599 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 600 | page_error_op | |
c2790a2e | 601 | PAGE_END_WRITEBACK); |
c8b97818 CM |
602 | goto free_pages_out; |
603 | } | |
604 | } | |
605 | ||
606 | if (will_compress) { | |
607 | /* | |
608 | * we aren't doing an inline extent round the compressed size | |
609 | * up to a block size boundary so the allocator does sane | |
610 | * things | |
611 | */ | |
fda2832f | 612 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
613 | |
614 | /* | |
615 | * one last check to make sure the compression is really a | |
170607eb TT |
616 | * win, compare the page count read with the blocks on disk, |
617 | * compression must free at least one sector size | |
c8b97818 | 618 | */ |
09cbfeaf | 619 | total_in = ALIGN(total_in, PAGE_SIZE); |
170607eb | 620 | if (total_compressed + blocksize <= total_in) { |
c8bb0c8b AS |
621 | *num_added += 1; |
622 | ||
623 | /* | |
624 | * The async work queues will take care of doing actual | |
625 | * allocation on disk for these compressed pages, and | |
626 | * will submit them to the elevator. | |
627 | */ | |
1170862d | 628 | add_async_extent(async_cow, start, total_in, |
4d3a800e | 629 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
630 | compress_type); |
631 | ||
1170862d TT |
632 | if (start + total_in < end) { |
633 | start += total_in; | |
c8bb0c8b AS |
634 | pages = NULL; |
635 | cond_resched(); | |
636 | goto again; | |
637 | } | |
638 | return; | |
c8b97818 CM |
639 | } |
640 | } | |
c8bb0c8b | 641 | if (pages) { |
c8b97818 CM |
642 | /* |
643 | * the compression code ran but failed to make things smaller, | |
644 | * free any pages it allocated and our page pointer array | |
645 | */ | |
4d3a800e | 646 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 647 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 648 | put_page(pages[i]); |
c8b97818 CM |
649 | } |
650 | kfree(pages); | |
651 | pages = NULL; | |
652 | total_compressed = 0; | |
4d3a800e | 653 | nr_pages = 0; |
c8b97818 CM |
654 | |
655 | /* flag the file so we don't compress in the future */ | |
0b246afa | 656 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
b52aa8c9 | 657 | !(BTRFS_I(inode)->prop_compress)) { |
a555f810 | 658 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 659 | } |
c8b97818 | 660 | } |
f03d9301 | 661 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
662 | /* |
663 | * No compression, but we still need to write the pages in the file | |
664 | * we've been given so far. redirty the locked page if it corresponds | |
665 | * to our extent and set things up for the async work queue to run | |
666 | * cow_file_range to do the normal delalloc dance. | |
667 | */ | |
668 | if (page_offset(locked_page) >= start && | |
669 | page_offset(locked_page) <= end) | |
670 | __set_page_dirty_nobuffers(locked_page); | |
671 | /* unlocked later on in the async handlers */ | |
672 | ||
673 | if (redirty) | |
674 | extent_range_redirty_for_io(inode, start, end); | |
675 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0, | |
676 | BTRFS_COMPRESS_NONE); | |
677 | *num_added += 1; | |
3b951516 | 678 | |
c44f649e | 679 | return; |
771ed689 CM |
680 | |
681 | free_pages_out: | |
4d3a800e | 682 | for (i = 0; i < nr_pages; i++) { |
771ed689 | 683 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 684 | put_page(pages[i]); |
771ed689 | 685 | } |
d397712b | 686 | kfree(pages); |
771ed689 | 687 | } |
771ed689 | 688 | |
40ae837b FM |
689 | static void free_async_extent_pages(struct async_extent *async_extent) |
690 | { | |
691 | int i; | |
692 | ||
693 | if (!async_extent->pages) | |
694 | return; | |
695 | ||
696 | for (i = 0; i < async_extent->nr_pages; i++) { | |
697 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 698 | put_page(async_extent->pages[i]); |
40ae837b FM |
699 | } |
700 | kfree(async_extent->pages); | |
701 | async_extent->nr_pages = 0; | |
702 | async_extent->pages = NULL; | |
771ed689 CM |
703 | } |
704 | ||
705 | /* | |
706 | * phase two of compressed writeback. This is the ordered portion | |
707 | * of the code, which only gets called in the order the work was | |
708 | * queued. We walk all the async extents created by compress_file_range | |
709 | * and send them down to the disk. | |
710 | */ | |
dec8f175 | 711 | static noinline void submit_compressed_extents(struct inode *inode, |
771ed689 CM |
712 | struct async_cow *async_cow) |
713 | { | |
0b246afa | 714 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
715 | struct async_extent *async_extent; |
716 | u64 alloc_hint = 0; | |
771ed689 CM |
717 | struct btrfs_key ins; |
718 | struct extent_map *em; | |
719 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
771ed689 | 720 | struct extent_io_tree *io_tree; |
f5a84ee3 | 721 | int ret = 0; |
771ed689 | 722 | |
3e04e7f1 | 723 | again: |
d397712b | 724 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
725 | async_extent = list_entry(async_cow->extents.next, |
726 | struct async_extent, list); | |
727 | list_del(&async_extent->list); | |
c8b97818 | 728 | |
771ed689 CM |
729 | io_tree = &BTRFS_I(inode)->io_tree; |
730 | ||
f5a84ee3 | 731 | retry: |
771ed689 CM |
732 | /* did the compression code fall back to uncompressed IO? */ |
733 | if (!async_extent->pages) { | |
734 | int page_started = 0; | |
735 | unsigned long nr_written = 0; | |
736 | ||
737 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 738 | async_extent->start + |
d0082371 | 739 | async_extent->ram_size - 1); |
771ed689 CM |
740 | |
741 | /* allocate blocks */ | |
f5a84ee3 JB |
742 | ret = cow_file_range(inode, async_cow->locked_page, |
743 | async_extent->start, | |
744 | async_extent->start + | |
745 | async_extent->ram_size - 1, | |
dda3245e WX |
746 | async_extent->start + |
747 | async_extent->ram_size - 1, | |
748 | &page_started, &nr_written, 0, | |
749 | NULL); | |
771ed689 | 750 | |
79787eaa JM |
751 | /* JDM XXX */ |
752 | ||
771ed689 CM |
753 | /* |
754 | * if page_started, cow_file_range inserted an | |
755 | * inline extent and took care of all the unlocking | |
756 | * and IO for us. Otherwise, we need to submit | |
757 | * all those pages down to the drive. | |
758 | */ | |
f5a84ee3 | 759 | if (!page_started && !ret) |
5e3ee236 NB |
760 | extent_write_locked_range(inode, |
761 | async_extent->start, | |
d397712b | 762 | async_extent->start + |
771ed689 | 763 | async_extent->ram_size - 1, |
771ed689 | 764 | WB_SYNC_ALL); |
3e04e7f1 JB |
765 | else if (ret) |
766 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
767 | kfree(async_extent); |
768 | cond_resched(); | |
769 | continue; | |
770 | } | |
771 | ||
772 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 773 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 774 | |
18513091 | 775 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
771ed689 CM |
776 | async_extent->compressed_size, |
777 | async_extent->compressed_size, | |
e570fd27 | 778 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 779 | if (ret) { |
40ae837b | 780 | free_async_extent_pages(async_extent); |
3e04e7f1 | 781 | |
fdf8e2ea JB |
782 | if (ret == -ENOSPC) { |
783 | unlock_extent(io_tree, async_extent->start, | |
784 | async_extent->start + | |
785 | async_extent->ram_size - 1); | |
ce62003f LB |
786 | |
787 | /* | |
788 | * we need to redirty the pages if we decide to | |
789 | * fallback to uncompressed IO, otherwise we | |
790 | * will not submit these pages down to lower | |
791 | * layers. | |
792 | */ | |
793 | extent_range_redirty_for_io(inode, | |
794 | async_extent->start, | |
795 | async_extent->start + | |
796 | async_extent->ram_size - 1); | |
797 | ||
79787eaa | 798 | goto retry; |
fdf8e2ea | 799 | } |
3e04e7f1 | 800 | goto out_free; |
f5a84ee3 | 801 | } |
c2167754 YZ |
802 | /* |
803 | * here we're doing allocation and writeback of the | |
804 | * compressed pages | |
805 | */ | |
6f9994db LB |
806 | em = create_io_em(inode, async_extent->start, |
807 | async_extent->ram_size, /* len */ | |
808 | async_extent->start, /* orig_start */ | |
809 | ins.objectid, /* block_start */ | |
810 | ins.offset, /* block_len */ | |
811 | ins.offset, /* orig_block_len */ | |
812 | async_extent->ram_size, /* ram_bytes */ | |
813 | async_extent->compress_type, | |
814 | BTRFS_ORDERED_COMPRESSED); | |
815 | if (IS_ERR(em)) | |
816 | /* ret value is not necessary due to void function */ | |
3e04e7f1 | 817 | goto out_free_reserve; |
6f9994db | 818 | free_extent_map(em); |
3e04e7f1 | 819 | |
261507a0 LZ |
820 | ret = btrfs_add_ordered_extent_compress(inode, |
821 | async_extent->start, | |
822 | ins.objectid, | |
823 | async_extent->ram_size, | |
824 | ins.offset, | |
825 | BTRFS_ORDERED_COMPRESSED, | |
826 | async_extent->compress_type); | |
d9f85963 | 827 | if (ret) { |
dcdbc059 NB |
828 | btrfs_drop_extent_cache(BTRFS_I(inode), |
829 | async_extent->start, | |
d9f85963 FM |
830 | async_extent->start + |
831 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 832 | goto out_free_reserve; |
d9f85963 | 833 | } |
0b246afa | 834 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
771ed689 | 835 | |
771ed689 CM |
836 | /* |
837 | * clear dirty, set writeback and unlock the pages. | |
838 | */ | |
c2790a2e | 839 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
840 | async_extent->start + |
841 | async_extent->ram_size - 1, | |
a791e35e CM |
842 | async_extent->start + |
843 | async_extent->ram_size - 1, | |
151a41bc JB |
844 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
845 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 846 | PAGE_SET_WRITEBACK); |
4e4cbee9 | 847 | if (btrfs_submit_compressed_write(inode, |
d397712b CM |
848 | async_extent->start, |
849 | async_extent->ram_size, | |
850 | ins.objectid, | |
851 | ins.offset, async_extent->pages, | |
f82b7359 LB |
852 | async_extent->nr_pages, |
853 | async_cow->write_flags)) { | |
fce2a4e6 FM |
854 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
855 | struct page *p = async_extent->pages[0]; | |
856 | const u64 start = async_extent->start; | |
857 | const u64 end = start + async_extent->ram_size - 1; | |
858 | ||
859 | p->mapping = inode->i_mapping; | |
860 | tree->ops->writepage_end_io_hook(p, start, end, | |
861 | NULL, 0); | |
862 | p->mapping = NULL; | |
ba8b04c1 QW |
863 | extent_clear_unlock_delalloc(inode, start, end, end, |
864 | NULL, 0, | |
fce2a4e6 FM |
865 | PAGE_END_WRITEBACK | |
866 | PAGE_SET_ERROR); | |
40ae837b | 867 | free_async_extent_pages(async_extent); |
fce2a4e6 | 868 | } |
771ed689 CM |
869 | alloc_hint = ins.objectid + ins.offset; |
870 | kfree(async_extent); | |
871 | cond_resched(); | |
872 | } | |
dec8f175 | 873 | return; |
3e04e7f1 | 874 | out_free_reserve: |
0b246afa | 875 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 876 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 877 | out_free: |
c2790a2e | 878 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
879 | async_extent->start + |
880 | async_extent->ram_size - 1, | |
3e04e7f1 JB |
881 | async_extent->start + |
882 | async_extent->ram_size - 1, | |
c2790a2e | 883 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 884 | EXTENT_DELALLOC_NEW | |
151a41bc JB |
885 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
886 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
887 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
888 | PAGE_SET_ERROR); | |
40ae837b | 889 | free_async_extent_pages(async_extent); |
79787eaa | 890 | kfree(async_extent); |
3e04e7f1 | 891 | goto again; |
771ed689 CM |
892 | } |
893 | ||
4b46fce2 JB |
894 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
895 | u64 num_bytes) | |
896 | { | |
897 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
898 | struct extent_map *em; | |
899 | u64 alloc_hint = 0; | |
900 | ||
901 | read_lock(&em_tree->lock); | |
902 | em = search_extent_mapping(em_tree, start, num_bytes); | |
903 | if (em) { | |
904 | /* | |
905 | * if block start isn't an actual block number then find the | |
906 | * first block in this inode and use that as a hint. If that | |
907 | * block is also bogus then just don't worry about it. | |
908 | */ | |
909 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
910 | free_extent_map(em); | |
911 | em = search_extent_mapping(em_tree, 0, 0); | |
912 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
913 | alloc_hint = em->block_start; | |
914 | if (em) | |
915 | free_extent_map(em); | |
916 | } else { | |
917 | alloc_hint = em->block_start; | |
918 | free_extent_map(em); | |
919 | } | |
920 | } | |
921 | read_unlock(&em_tree->lock); | |
922 | ||
923 | return alloc_hint; | |
924 | } | |
925 | ||
771ed689 CM |
926 | /* |
927 | * when extent_io.c finds a delayed allocation range in the file, | |
928 | * the call backs end up in this code. The basic idea is to | |
929 | * allocate extents on disk for the range, and create ordered data structs | |
930 | * in ram to track those extents. | |
931 | * | |
932 | * locked_page is the page that writepage had locked already. We use | |
933 | * it to make sure we don't do extra locks or unlocks. | |
934 | * | |
935 | * *page_started is set to one if we unlock locked_page and do everything | |
936 | * required to start IO on it. It may be clean and already done with | |
937 | * IO when we return. | |
938 | */ | |
00361589 JB |
939 | static noinline int cow_file_range(struct inode *inode, |
940 | struct page *locked_page, | |
dda3245e WX |
941 | u64 start, u64 end, u64 delalloc_end, |
942 | int *page_started, unsigned long *nr_written, | |
943 | int unlock, struct btrfs_dedupe_hash *hash) | |
771ed689 | 944 | { |
0b246afa | 945 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00361589 | 946 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
947 | u64 alloc_hint = 0; |
948 | u64 num_bytes; | |
949 | unsigned long ram_size; | |
a315e68f | 950 | u64 cur_alloc_size = 0; |
0b246afa | 951 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
952 | struct btrfs_key ins; |
953 | struct extent_map *em; | |
a315e68f FM |
954 | unsigned clear_bits; |
955 | unsigned long page_ops; | |
956 | bool extent_reserved = false; | |
771ed689 CM |
957 | int ret = 0; |
958 | ||
70ddc553 | 959 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
02ecd2c2 | 960 | WARN_ON_ONCE(1); |
29bce2f3 JB |
961 | ret = -EINVAL; |
962 | goto out_unlock; | |
02ecd2c2 | 963 | } |
771ed689 | 964 | |
fda2832f | 965 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 | 966 | num_bytes = max(blocksize, num_bytes); |
566b1760 | 967 | ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 968 | |
6158e1ce | 969 | inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K); |
4cb5300b | 970 | |
771ed689 CM |
971 | if (start == 0) { |
972 | /* lets try to make an inline extent */ | |
d02c0e20 NB |
973 | ret = cow_file_range_inline(inode, start, end, 0, |
974 | BTRFS_COMPRESS_NONE, NULL); | |
771ed689 | 975 | if (ret == 0) { |
8b62f87b JB |
976 | /* |
977 | * We use DO_ACCOUNTING here because we need the | |
978 | * delalloc_release_metadata to be run _after_ we drop | |
979 | * our outstanding extent for clearing delalloc for this | |
980 | * range. | |
981 | */ | |
ba8b04c1 QW |
982 | extent_clear_unlock_delalloc(inode, start, end, |
983 | delalloc_end, NULL, | |
c2790a2e | 984 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
985 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
986 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
c2790a2e JB |
987 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
988 | PAGE_END_WRITEBACK); | |
771ed689 | 989 | *nr_written = *nr_written + |
09cbfeaf | 990 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 991 | *page_started = 1; |
771ed689 | 992 | goto out; |
79787eaa | 993 | } else if (ret < 0) { |
79787eaa | 994 | goto out_unlock; |
771ed689 CM |
995 | } |
996 | } | |
997 | ||
4b46fce2 | 998 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
dcdbc059 NB |
999 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
1000 | start + num_bytes - 1, 0); | |
771ed689 | 1001 | |
3752d22f AJ |
1002 | while (num_bytes > 0) { |
1003 | cur_alloc_size = num_bytes; | |
18513091 | 1004 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
0b246afa | 1005 | fs_info->sectorsize, 0, alloc_hint, |
e570fd27 | 1006 | &ins, 1, 1); |
00361589 | 1007 | if (ret < 0) |
79787eaa | 1008 | goto out_unlock; |
a315e68f FM |
1009 | cur_alloc_size = ins.offset; |
1010 | extent_reserved = true; | |
d397712b | 1011 | |
771ed689 | 1012 | ram_size = ins.offset; |
6f9994db LB |
1013 | em = create_io_em(inode, start, ins.offset, /* len */ |
1014 | start, /* orig_start */ | |
1015 | ins.objectid, /* block_start */ | |
1016 | ins.offset, /* block_len */ | |
1017 | ins.offset, /* orig_block_len */ | |
1018 | ram_size, /* ram_bytes */ | |
1019 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1020 | BTRFS_ORDERED_REGULAR /* type */); |
6f9994db | 1021 | if (IS_ERR(em)) |
ace68bac | 1022 | goto out_reserve; |
6f9994db | 1023 | free_extent_map(em); |
e6dcd2dc | 1024 | |
e6dcd2dc | 1025 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1026 | ram_size, cur_alloc_size, 0); |
ace68bac | 1027 | if (ret) |
d9f85963 | 1028 | goto out_drop_extent_cache; |
c8b97818 | 1029 | |
17d217fe YZ |
1030 | if (root->root_key.objectid == |
1031 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1032 | ret = btrfs_reloc_clone_csums(inode, start, | |
1033 | cur_alloc_size); | |
4dbd80fb QW |
1034 | /* |
1035 | * Only drop cache here, and process as normal. | |
1036 | * | |
1037 | * We must not allow extent_clear_unlock_delalloc() | |
1038 | * at out_unlock label to free meta of this ordered | |
1039 | * extent, as its meta should be freed by | |
1040 | * btrfs_finish_ordered_io(). | |
1041 | * | |
1042 | * So we must continue until @start is increased to | |
1043 | * skip current ordered extent. | |
1044 | */ | |
00361589 | 1045 | if (ret) |
4dbd80fb QW |
1046 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
1047 | start + ram_size - 1, 0); | |
17d217fe YZ |
1048 | } |
1049 | ||
0b246afa | 1050 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1051 | |
c8b97818 CM |
1052 | /* we're not doing compressed IO, don't unlock the first |
1053 | * page (which the caller expects to stay locked), don't | |
1054 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1055 | * |
1056 | * Do set the Private2 bit so we know this page was properly | |
1057 | * setup for writepage | |
c8b97818 | 1058 | */ |
a315e68f FM |
1059 | page_ops = unlock ? PAGE_UNLOCK : 0; |
1060 | page_ops |= PAGE_SET_PRIVATE2; | |
a791e35e | 1061 | |
c2790a2e | 1062 | extent_clear_unlock_delalloc(inode, start, |
ba8b04c1 QW |
1063 | start + ram_size - 1, |
1064 | delalloc_end, locked_page, | |
c2790a2e | 1065 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1066 | page_ops); |
3752d22f AJ |
1067 | if (num_bytes < cur_alloc_size) |
1068 | num_bytes = 0; | |
4dbd80fb | 1069 | else |
3752d22f | 1070 | num_bytes -= cur_alloc_size; |
c59f8951 CM |
1071 | alloc_hint = ins.objectid + ins.offset; |
1072 | start += cur_alloc_size; | |
a315e68f | 1073 | extent_reserved = false; |
4dbd80fb QW |
1074 | |
1075 | /* | |
1076 | * btrfs_reloc_clone_csums() error, since start is increased | |
1077 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1078 | * free metadata of current ordered extent, we're OK to exit. | |
1079 | */ | |
1080 | if (ret) | |
1081 | goto out_unlock; | |
b888db2b | 1082 | } |
79787eaa | 1083 | out: |
be20aa9d | 1084 | return ret; |
b7d5b0a8 | 1085 | |
d9f85963 | 1086 | out_drop_extent_cache: |
dcdbc059 | 1087 | btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0); |
ace68bac | 1088 | out_reserve: |
0b246afa | 1089 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1090 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1091 | out_unlock: |
a7e3b975 FM |
1092 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1093 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
a315e68f FM |
1094 | page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
1095 | PAGE_END_WRITEBACK; | |
1096 | /* | |
1097 | * If we reserved an extent for our delalloc range (or a subrange) and | |
1098 | * failed to create the respective ordered extent, then it means that | |
1099 | * when we reserved the extent we decremented the extent's size from | |
1100 | * the data space_info's bytes_may_use counter and incremented the | |
1101 | * space_info's bytes_reserved counter by the same amount. We must make | |
1102 | * sure extent_clear_unlock_delalloc() does not try to decrement again | |
1103 | * the data space_info's bytes_may_use counter, therefore we do not pass | |
1104 | * it the flag EXTENT_CLEAR_DATA_RESV. | |
1105 | */ | |
1106 | if (extent_reserved) { | |
1107 | extent_clear_unlock_delalloc(inode, start, | |
1108 | start + cur_alloc_size, | |
1109 | start + cur_alloc_size, | |
1110 | locked_page, | |
1111 | clear_bits, | |
1112 | page_ops); | |
1113 | start += cur_alloc_size; | |
1114 | if (start >= end) | |
1115 | goto out; | |
1116 | } | |
ba8b04c1 QW |
1117 | extent_clear_unlock_delalloc(inode, start, end, delalloc_end, |
1118 | locked_page, | |
a315e68f FM |
1119 | clear_bits | EXTENT_CLEAR_DATA_RESV, |
1120 | page_ops); | |
79787eaa | 1121 | goto out; |
771ed689 | 1122 | } |
c8b97818 | 1123 | |
771ed689 CM |
1124 | /* |
1125 | * work queue call back to started compression on a file and pages | |
1126 | */ | |
1127 | static noinline void async_cow_start(struct btrfs_work *work) | |
1128 | { | |
1129 | struct async_cow *async_cow; | |
1130 | int num_added = 0; | |
1131 | async_cow = container_of(work, struct async_cow, work); | |
1132 | ||
1133 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1134 | async_cow->start, async_cow->end, async_cow, | |
1135 | &num_added); | |
8180ef88 | 1136 | if (num_added == 0) { |
cb77fcd8 | 1137 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1138 | async_cow->inode = NULL; |
8180ef88 | 1139 | } |
771ed689 CM |
1140 | } |
1141 | ||
1142 | /* | |
1143 | * work queue call back to submit previously compressed pages | |
1144 | */ | |
1145 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1146 | { | |
0b246afa | 1147 | struct btrfs_fs_info *fs_info; |
771ed689 CM |
1148 | struct async_cow *async_cow; |
1149 | struct btrfs_root *root; | |
1150 | unsigned long nr_pages; | |
1151 | ||
1152 | async_cow = container_of(work, struct async_cow, work); | |
1153 | ||
1154 | root = async_cow->root; | |
0b246afa | 1155 | fs_info = root->fs_info; |
09cbfeaf KS |
1156 | nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >> |
1157 | PAGE_SHIFT; | |
771ed689 | 1158 | |
093258e6 | 1159 | /* atomic_sub_return implies a barrier */ |
0b246afa | 1160 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < |
093258e6 DS |
1161 | 5 * SZ_1M) |
1162 | cond_wake_up_nomb(&fs_info->async_submit_wait); | |
771ed689 | 1163 | |
d397712b | 1164 | if (async_cow->inode) |
771ed689 | 1165 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1166 | } |
c8b97818 | 1167 | |
771ed689 CM |
1168 | static noinline void async_cow_free(struct btrfs_work *work) |
1169 | { | |
1170 | struct async_cow *async_cow; | |
1171 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1172 | if (async_cow->inode) |
cb77fcd8 | 1173 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1174 | kfree(async_cow); |
1175 | } | |
1176 | ||
1177 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1178 | u64 start, u64 end, int *page_started, | |
f82b7359 LB |
1179 | unsigned long *nr_written, |
1180 | unsigned int write_flags) | |
771ed689 | 1181 | { |
0b246afa | 1182 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
1183 | struct async_cow *async_cow; |
1184 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1185 | unsigned long nr_pages; | |
1186 | u64 cur_end; | |
771ed689 | 1187 | |
a3429ab7 | 1188 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
ae0f1625 | 1189 | 1, 0, NULL); |
d397712b | 1190 | while (start < end) { |
771ed689 | 1191 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1192 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1193 | async_cow->inode = igrab(inode); |
771ed689 CM |
1194 | async_cow->root = root; |
1195 | async_cow->locked_page = locked_page; | |
1196 | async_cow->start = start; | |
f82b7359 | 1197 | async_cow->write_flags = write_flags; |
771ed689 | 1198 | |
f79707b0 | 1199 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
0b246afa | 1200 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
771ed689 CM |
1201 | cur_end = end; |
1202 | else | |
ee22184b | 1203 | cur_end = min(end, start + SZ_512K - 1); |
771ed689 CM |
1204 | |
1205 | async_cow->end = cur_end; | |
1206 | INIT_LIST_HEAD(&async_cow->extents); | |
1207 | ||
9e0af237 LB |
1208 | btrfs_init_work(&async_cow->work, |
1209 | btrfs_delalloc_helper, | |
1210 | async_cow_start, async_cow_submit, | |
1211 | async_cow_free); | |
771ed689 | 1212 | |
09cbfeaf KS |
1213 | nr_pages = (cur_end - start + PAGE_SIZE) >> |
1214 | PAGE_SHIFT; | |
0b246afa | 1215 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1216 | |
0b246afa | 1217 | btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work); |
771ed689 | 1218 | |
771ed689 CM |
1219 | *nr_written += nr_pages; |
1220 | start = cur_end + 1; | |
1221 | } | |
1222 | *page_started = 1; | |
1223 | return 0; | |
be20aa9d CM |
1224 | } |
1225 | ||
2ff7e61e | 1226 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1227 | u64 bytenr, u64 num_bytes) |
1228 | { | |
1229 | int ret; | |
1230 | struct btrfs_ordered_sum *sums; | |
1231 | LIST_HEAD(list); | |
1232 | ||
0b246afa | 1233 | ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr, |
a2de733c | 1234 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1235 | if (ret == 0 && list_empty(&list)) |
1236 | return 0; | |
1237 | ||
1238 | while (!list_empty(&list)) { | |
1239 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1240 | list_del(&sums->list); | |
1241 | kfree(sums); | |
1242 | } | |
58113753 LB |
1243 | if (ret < 0) |
1244 | return ret; | |
17d217fe YZ |
1245 | return 1; |
1246 | } | |
1247 | ||
d352ac68 CM |
1248 | /* |
1249 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1250 | * of the extents that exist in the file, and COWs the file as required. | |
1251 | * | |
1252 | * If no cow copies or snapshots exist, we write directly to the existing | |
1253 | * blocks on disk | |
1254 | */ | |
7f366cfe CM |
1255 | static noinline int run_delalloc_nocow(struct inode *inode, |
1256 | struct page *locked_page, | |
771ed689 CM |
1257 | u64 start, u64 end, int *page_started, int force, |
1258 | unsigned long *nr_written) | |
be20aa9d | 1259 | { |
0b246afa | 1260 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
be20aa9d CM |
1261 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1262 | struct extent_buffer *leaf; | |
be20aa9d | 1263 | struct btrfs_path *path; |
80ff3856 | 1264 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1265 | struct btrfs_key found_key; |
6f9994db | 1266 | struct extent_map *em; |
80ff3856 YZ |
1267 | u64 cow_start; |
1268 | u64 cur_offset; | |
1269 | u64 extent_end; | |
5d4f98a2 | 1270 | u64 extent_offset; |
80ff3856 YZ |
1271 | u64 disk_bytenr; |
1272 | u64 num_bytes; | |
b4939680 | 1273 | u64 disk_num_bytes; |
cc95bef6 | 1274 | u64 ram_bytes; |
80ff3856 | 1275 | int extent_type; |
79787eaa | 1276 | int ret, err; |
d899e052 | 1277 | int type; |
80ff3856 YZ |
1278 | int nocow; |
1279 | int check_prev = 1; | |
82d5902d | 1280 | bool nolock; |
4a0cc7ca | 1281 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
be20aa9d CM |
1282 | |
1283 | path = btrfs_alloc_path(); | |
17ca04af | 1284 | if (!path) { |
ba8b04c1 QW |
1285 | extent_clear_unlock_delalloc(inode, start, end, end, |
1286 | locked_page, | |
c2790a2e | 1287 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1288 | EXTENT_DO_ACCOUNTING | |
1289 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1290 | PAGE_CLEAR_DIRTY | |
1291 | PAGE_SET_WRITEBACK | | |
1292 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1293 | return -ENOMEM; |
17ca04af | 1294 | } |
82d5902d | 1295 | |
70ddc553 | 1296 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
82d5902d | 1297 | |
80ff3856 YZ |
1298 | cow_start = (u64)-1; |
1299 | cur_offset = start; | |
1300 | while (1) { | |
e4c3b2dc | 1301 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1302 | cur_offset, 0); |
d788a349 | 1303 | if (ret < 0) |
79787eaa | 1304 | goto error; |
80ff3856 YZ |
1305 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1306 | leaf = path->nodes[0]; | |
1307 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1308 | path->slots[0] - 1); | |
33345d01 | 1309 | if (found_key.objectid == ino && |
80ff3856 YZ |
1310 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1311 | path->slots[0]--; | |
1312 | } | |
1313 | check_prev = 0; | |
1314 | next_slot: | |
1315 | leaf = path->nodes[0]; | |
1316 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1317 | ret = btrfs_next_leaf(root, path); | |
e8916699 LB |
1318 | if (ret < 0) { |
1319 | if (cow_start != (u64)-1) | |
1320 | cur_offset = cow_start; | |
79787eaa | 1321 | goto error; |
e8916699 | 1322 | } |
80ff3856 YZ |
1323 | if (ret > 0) |
1324 | break; | |
1325 | leaf = path->nodes[0]; | |
1326 | } | |
be20aa9d | 1327 | |
80ff3856 YZ |
1328 | nocow = 0; |
1329 | disk_bytenr = 0; | |
17d217fe | 1330 | num_bytes = 0; |
80ff3856 YZ |
1331 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1332 | ||
1d512cb7 FM |
1333 | if (found_key.objectid > ino) |
1334 | break; | |
1335 | if (WARN_ON_ONCE(found_key.objectid < ino) || | |
1336 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1337 | path->slots[0]++; | |
1338 | goto next_slot; | |
1339 | } | |
1340 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || | |
80ff3856 YZ |
1341 | found_key.offset > end) |
1342 | break; | |
1343 | ||
1344 | if (found_key.offset > cur_offset) { | |
1345 | extent_end = found_key.offset; | |
e9061e21 | 1346 | extent_type = 0; |
80ff3856 YZ |
1347 | goto out_check; |
1348 | } | |
1349 | ||
1350 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1351 | struct btrfs_file_extent_item); | |
1352 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1353 | ||
cc95bef6 | 1354 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1355 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1356 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1357 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1358 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1359 | extent_end = found_key.offset + |
1360 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1361 | disk_num_bytes = |
1362 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1363 | if (extent_end <= start) { |
1364 | path->slots[0]++; | |
1365 | goto next_slot; | |
1366 | } | |
17d217fe YZ |
1367 | if (disk_bytenr == 0) |
1368 | goto out_check; | |
80ff3856 YZ |
1369 | if (btrfs_file_extent_compression(leaf, fi) || |
1370 | btrfs_file_extent_encryption(leaf, fi) || | |
1371 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1372 | goto out_check; | |
d899e052 YZ |
1373 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1374 | goto out_check; | |
2ff7e61e | 1375 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
80ff3856 | 1376 | goto out_check; |
58113753 LB |
1377 | ret = btrfs_cross_ref_exist(root, ino, |
1378 | found_key.offset - | |
1379 | extent_offset, disk_bytenr); | |
1380 | if (ret) { | |
1381 | /* | |
1382 | * ret could be -EIO if the above fails to read | |
1383 | * metadata. | |
1384 | */ | |
1385 | if (ret < 0) { | |
1386 | if (cow_start != (u64)-1) | |
1387 | cur_offset = cow_start; | |
1388 | goto error; | |
1389 | } | |
1390 | ||
1391 | WARN_ON_ONCE(nolock); | |
17d217fe | 1392 | goto out_check; |
58113753 | 1393 | } |
5d4f98a2 | 1394 | disk_bytenr += extent_offset; |
17d217fe YZ |
1395 | disk_bytenr += cur_offset - found_key.offset; |
1396 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1397 | /* |
1398 | * if there are pending snapshots for this root, | |
1399 | * we fall into common COW way. | |
1400 | */ | |
1401 | if (!nolock) { | |
ea14b57f | 1402 | err = btrfs_start_write_no_snapshotting(root); |
e9894fd3 WS |
1403 | if (!err) |
1404 | goto out_check; | |
1405 | } | |
17d217fe YZ |
1406 | /* |
1407 | * force cow if csum exists in the range. | |
1408 | * this ensure that csum for a given extent are | |
1409 | * either valid or do not exist. | |
1410 | */ | |
58113753 LB |
1411 | ret = csum_exist_in_range(fs_info, disk_bytenr, |
1412 | num_bytes); | |
1413 | if (ret) { | |
91e1f56a | 1414 | if (!nolock) |
ea14b57f | 1415 | btrfs_end_write_no_snapshotting(root); |
58113753 LB |
1416 | |
1417 | /* | |
1418 | * ret could be -EIO if the above fails to read | |
1419 | * metadata. | |
1420 | */ | |
1421 | if (ret < 0) { | |
1422 | if (cow_start != (u64)-1) | |
1423 | cur_offset = cow_start; | |
1424 | goto error; | |
1425 | } | |
1426 | WARN_ON_ONCE(nolock); | |
17d217fe | 1427 | goto out_check; |
91e1f56a RK |
1428 | } |
1429 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) { | |
1430 | if (!nolock) | |
ea14b57f | 1431 | btrfs_end_write_no_snapshotting(root); |
f78c436c | 1432 | goto out_check; |
91e1f56a | 1433 | } |
80ff3856 YZ |
1434 | nocow = 1; |
1435 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1436 | extent_end = found_key.offset + | |
514ac8ad CM |
1437 | btrfs_file_extent_inline_len(leaf, |
1438 | path->slots[0], fi); | |
da17066c | 1439 | extent_end = ALIGN(extent_end, |
0b246afa | 1440 | fs_info->sectorsize); |
80ff3856 YZ |
1441 | } else { |
1442 | BUG_ON(1); | |
1443 | } | |
1444 | out_check: | |
1445 | if (extent_end <= start) { | |
1446 | path->slots[0]++; | |
e9894fd3 | 1447 | if (!nolock && nocow) |
ea14b57f | 1448 | btrfs_end_write_no_snapshotting(root); |
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 WS |
1469 | if (ret) { |
1470 | if (!nolock && nocow) | |
ea14b57f | 1471 | btrfs_end_write_no_snapshotting(root); |
f78c436c | 1472 | if (nocow) |
0b246afa | 1473 | btrfs_dec_nocow_writers(fs_info, |
f78c436c | 1474 | disk_bytenr); |
79787eaa | 1475 | goto error; |
e9894fd3 | 1476 | } |
80ff3856 | 1477 | cow_start = (u64)-1; |
7ea394f1 | 1478 | } |
80ff3856 | 1479 | |
d899e052 | 1480 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
6f9994db LB |
1481 | u64 orig_start = found_key.offset - extent_offset; |
1482 | ||
1483 | em = create_io_em(inode, cur_offset, num_bytes, | |
1484 | orig_start, | |
1485 | disk_bytenr, /* block_start */ | |
1486 | num_bytes, /* block_len */ | |
1487 | disk_num_bytes, /* orig_block_len */ | |
1488 | ram_bytes, BTRFS_COMPRESS_NONE, | |
1489 | BTRFS_ORDERED_PREALLOC); | |
1490 | if (IS_ERR(em)) { | |
1491 | if (!nolock && nocow) | |
ea14b57f | 1492 | btrfs_end_write_no_snapshotting(root); |
6f9994db LB |
1493 | if (nocow) |
1494 | btrfs_dec_nocow_writers(fs_info, | |
1495 | disk_bytenr); | |
1496 | ret = PTR_ERR(em); | |
1497 | goto error; | |
d899e052 | 1498 | } |
6f9994db LB |
1499 | free_extent_map(em); |
1500 | } | |
1501 | ||
1502 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
d899e052 YZ |
1503 | type = BTRFS_ORDERED_PREALLOC; |
1504 | } else { | |
1505 | type = BTRFS_ORDERED_NOCOW; | |
1506 | } | |
80ff3856 YZ |
1507 | |
1508 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1509 | num_bytes, num_bytes, type); |
f78c436c | 1510 | if (nocow) |
0b246afa | 1511 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
79787eaa | 1512 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1513 | |
efa56464 | 1514 | if (root->root_key.objectid == |
4dbd80fb QW |
1515 | BTRFS_DATA_RELOC_TREE_OBJECTID) |
1516 | /* | |
1517 | * Error handled later, as we must prevent | |
1518 | * extent_clear_unlock_delalloc() in error handler | |
1519 | * from freeing metadata of created ordered extent. | |
1520 | */ | |
efa56464 YZ |
1521 | ret = btrfs_reloc_clone_csums(inode, cur_offset, |
1522 | num_bytes); | |
efa56464 | 1523 | |
c2790a2e | 1524 | extent_clear_unlock_delalloc(inode, cur_offset, |
ba8b04c1 | 1525 | cur_offset + num_bytes - 1, end, |
c2790a2e | 1526 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1527 | EXTENT_DELALLOC | |
1528 | EXTENT_CLEAR_DATA_RESV, | |
1529 | PAGE_UNLOCK | PAGE_SET_PRIVATE2); | |
1530 | ||
e9894fd3 | 1531 | if (!nolock && nocow) |
ea14b57f | 1532 | btrfs_end_write_no_snapshotting(root); |
80ff3856 | 1533 | cur_offset = extent_end; |
4dbd80fb QW |
1534 | |
1535 | /* | |
1536 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
1537 | * handler, as metadata for created ordered extent will only | |
1538 | * be freed by btrfs_finish_ordered_io(). | |
1539 | */ | |
1540 | if (ret) | |
1541 | goto error; | |
80ff3856 YZ |
1542 | if (cur_offset > end) |
1543 | break; | |
be20aa9d | 1544 | } |
b3b4aa74 | 1545 | btrfs_release_path(path); |
80ff3856 | 1546 | |
17ca04af | 1547 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1548 | cow_start = cur_offset; |
17ca04af JB |
1549 | cur_offset = end; |
1550 | } | |
1551 | ||
80ff3856 | 1552 | if (cow_start != (u64)-1) { |
dda3245e WX |
1553 | ret = cow_file_range(inode, locked_page, cow_start, end, end, |
1554 | page_started, nr_written, 1, NULL); | |
d788a349 | 1555 | if (ret) |
79787eaa | 1556 | goto error; |
80ff3856 YZ |
1557 | } |
1558 | ||
79787eaa | 1559 | error: |
17ca04af | 1560 | if (ret && cur_offset < end) |
ba8b04c1 | 1561 | extent_clear_unlock_delalloc(inode, cur_offset, end, end, |
c2790a2e | 1562 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1563 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1564 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1565 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1566 | PAGE_SET_WRITEBACK | |
1567 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1568 | btrfs_free_path(path); |
79787eaa | 1569 | return ret; |
be20aa9d CM |
1570 | } |
1571 | ||
47059d93 WS |
1572 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1573 | { | |
1574 | ||
1575 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1576 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1577 | return 0; | |
1578 | ||
1579 | /* | |
1580 | * @defrag_bytes is a hint value, no spinlock held here, | |
1581 | * if is not zero, it means the file is defragging. | |
1582 | * Force cow if given extent needs to be defragged. | |
1583 | */ | |
1584 | if (BTRFS_I(inode)->defrag_bytes && | |
1585 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1586 | EXTENT_DEFRAG, 0, NULL)) | |
1587 | return 1; | |
1588 | ||
1589 | return 0; | |
1590 | } | |
1591 | ||
d352ac68 CM |
1592 | /* |
1593 | * extent_io.c call back to do delayed allocation processing | |
1594 | */ | |
c6100a4b | 1595 | static int run_delalloc_range(void *private_data, struct page *locked_page, |
771ed689 | 1596 | u64 start, u64 end, int *page_started, |
f82b7359 LB |
1597 | unsigned long *nr_written, |
1598 | struct writeback_control *wbc) | |
be20aa9d | 1599 | { |
c6100a4b | 1600 | struct inode *inode = private_data; |
be20aa9d | 1601 | int ret; |
47059d93 | 1602 | int force_cow = need_force_cow(inode, start, end); |
f82b7359 | 1603 | unsigned int write_flags = wbc_to_write_flags(wbc); |
a2135011 | 1604 | |
47059d93 | 1605 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1606 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1607 | page_started, 1, nr_written); |
47059d93 | 1608 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1609 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1610 | page_started, 0, nr_written); |
c2fcdcdf | 1611 | } else if (!inode_need_compress(inode, start, end)) { |
dda3245e WX |
1612 | ret = cow_file_range(inode, locked_page, start, end, end, |
1613 | page_started, nr_written, 1, NULL); | |
7ddf5a42 JB |
1614 | } else { |
1615 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1616 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1617 | ret = cow_file_range_async(inode, locked_page, start, end, |
f82b7359 LB |
1618 | page_started, nr_written, |
1619 | write_flags); | |
7ddf5a42 | 1620 | } |
52427260 QW |
1621 | if (ret) |
1622 | btrfs_cleanup_ordered_extents(inode, start, end - start + 1); | |
b888db2b CM |
1623 | return ret; |
1624 | } | |
1625 | ||
c6100a4b | 1626 | static void btrfs_split_extent_hook(void *private_data, |
1bf85046 | 1627 | struct extent_state *orig, u64 split) |
9ed74f2d | 1628 | { |
c6100a4b | 1629 | struct inode *inode = private_data; |
dcab6a3b JB |
1630 | u64 size; |
1631 | ||
0ca1f7ce | 1632 | /* not delalloc, ignore it */ |
9ed74f2d | 1633 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1634 | return; |
9ed74f2d | 1635 | |
dcab6a3b JB |
1636 | size = orig->end - orig->start + 1; |
1637 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 1638 | u32 num_extents; |
dcab6a3b JB |
1639 | u64 new_size; |
1640 | ||
1641 | /* | |
ba117213 JB |
1642 | * See the explanation in btrfs_merge_extent_hook, the same |
1643 | * applies here, just in reverse. | |
dcab6a3b JB |
1644 | */ |
1645 | new_size = orig->end - split + 1; | |
823bb20a | 1646 | num_extents = count_max_extents(new_size); |
ba117213 | 1647 | new_size = split - orig->start; |
823bb20a DS |
1648 | num_extents += count_max_extents(new_size); |
1649 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
1650 | return; |
1651 | } | |
1652 | ||
9e0baf60 | 1653 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1654 | btrfs_mod_outstanding_extents(BTRFS_I(inode), 1); |
9e0baf60 | 1655 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1656 | } |
1657 | ||
1658 | /* | |
1659 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1660 | * extents so we can keep track of new extents that are just merged onto old | |
1661 | * extents, such as when we are doing sequential writes, so we can properly | |
1662 | * account for the metadata space we'll need. | |
1663 | */ | |
c6100a4b | 1664 | static void btrfs_merge_extent_hook(void *private_data, |
1bf85046 JM |
1665 | struct extent_state *new, |
1666 | struct extent_state *other) | |
9ed74f2d | 1667 | { |
c6100a4b | 1668 | struct inode *inode = private_data; |
dcab6a3b | 1669 | u64 new_size, old_size; |
823bb20a | 1670 | u32 num_extents; |
dcab6a3b | 1671 | |
9ed74f2d JB |
1672 | /* not delalloc, ignore it */ |
1673 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1674 | return; |
9ed74f2d | 1675 | |
8461a3de JB |
1676 | if (new->start > other->start) |
1677 | new_size = new->end - other->start + 1; | |
1678 | else | |
1679 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1680 | |
1681 | /* we're not bigger than the max, unreserve the space and go */ | |
1682 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1683 | spin_lock(&BTRFS_I(inode)->lock); | |
8b62f87b | 1684 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
dcab6a3b JB |
1685 | spin_unlock(&BTRFS_I(inode)->lock); |
1686 | return; | |
1687 | } | |
1688 | ||
1689 | /* | |
ba117213 JB |
1690 | * We have to add up either side to figure out how many extents were |
1691 | * accounted for before we merged into one big extent. If the number of | |
1692 | * extents we accounted for is <= the amount we need for the new range | |
1693 | * then we can return, otherwise drop. Think of it like this | |
1694 | * | |
1695 | * [ 4k][MAX_SIZE] | |
1696 | * | |
1697 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1698 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1699 | * we have 1 so they are == and we can return. But in this case | |
1700 | * | |
1701 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1702 | * | |
1703 | * Each range on their own accounts for 2 extents, but merged together | |
1704 | * they are only 3 extents worth of accounting, so we need to drop in | |
1705 | * this case. | |
dcab6a3b | 1706 | */ |
ba117213 | 1707 | old_size = other->end - other->start + 1; |
823bb20a | 1708 | num_extents = count_max_extents(old_size); |
ba117213 | 1709 | old_size = new->end - new->start + 1; |
823bb20a DS |
1710 | num_extents += count_max_extents(old_size); |
1711 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
1712 | return; |
1713 | ||
9e0baf60 | 1714 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1715 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
9e0baf60 | 1716 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1717 | } |
1718 | ||
eb73c1b7 MX |
1719 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1720 | struct inode *inode) | |
1721 | { | |
0b246afa JM |
1722 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1723 | ||
eb73c1b7 MX |
1724 | spin_lock(&root->delalloc_lock); |
1725 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1726 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1727 | &root->delalloc_inodes); | |
1728 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1729 | &BTRFS_I(inode)->runtime_flags); | |
1730 | root->nr_delalloc_inodes++; | |
1731 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 1732 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1733 | BUG_ON(!list_empty(&root->delalloc_root)); |
1734 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
1735 | &fs_info->delalloc_roots); |
1736 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
1737 | } |
1738 | } | |
1739 | spin_unlock(&root->delalloc_lock); | |
1740 | } | |
1741 | ||
2b877331 NB |
1742 | |
1743 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1744 | struct btrfs_inode *inode) | |
eb73c1b7 | 1745 | { |
9e3e97f4 | 1746 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
0b246afa | 1747 | |
9e3e97f4 NB |
1748 | if (!list_empty(&inode->delalloc_inodes)) { |
1749 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 1750 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1751 | &inode->runtime_flags); |
eb73c1b7 MX |
1752 | root->nr_delalloc_inodes--; |
1753 | if (!root->nr_delalloc_inodes) { | |
7c8a0d36 | 1754 | ASSERT(list_empty(&root->delalloc_inodes)); |
0b246afa | 1755 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1756 | BUG_ON(list_empty(&root->delalloc_root)); |
1757 | list_del_init(&root->delalloc_root); | |
0b246afa | 1758 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1759 | } |
1760 | } | |
2b877331 NB |
1761 | } |
1762 | ||
1763 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1764 | struct btrfs_inode *inode) | |
1765 | { | |
1766 | spin_lock(&root->delalloc_lock); | |
1767 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
1768 | spin_unlock(&root->delalloc_lock); |
1769 | } | |
1770 | ||
d352ac68 CM |
1771 | /* |
1772 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1773 | * bytes in this file, and to maintain the list of inodes that | |
1774 | * have pending delalloc work to be done. | |
1775 | */ | |
c6100a4b | 1776 | static void btrfs_set_bit_hook(void *private_data, |
9ee49a04 | 1777 | struct extent_state *state, unsigned *bits) |
291d673e | 1778 | { |
c6100a4b | 1779 | struct inode *inode = private_data; |
9ed74f2d | 1780 | |
0b246afa JM |
1781 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1782 | ||
47059d93 WS |
1783 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1784 | WARN_ON(1); | |
75eff68e CM |
1785 | /* |
1786 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1787 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1788 | * bit, which is only set or cleared with irqs on |
1789 | */ | |
0ca1f7ce | 1790 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1791 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1792 | u64 len = state->end + 1 - state->start; |
8b62f87b | 1793 | u32 num_extents = count_max_extents(len); |
70ddc553 | 1794 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 1795 | |
8b62f87b JB |
1796 | spin_lock(&BTRFS_I(inode)->lock); |
1797 | btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents); | |
1798 | spin_unlock(&BTRFS_I(inode)->lock); | |
287a0ab9 | 1799 | |
6a3891c5 | 1800 | /* For sanity tests */ |
0b246afa | 1801 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1802 | return; |
1803 | ||
104b4e51 NB |
1804 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
1805 | fs_info->delalloc_batch); | |
df0af1a5 | 1806 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1807 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1808 | if (*bits & EXTENT_DEFRAG) |
1809 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1810 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1811 | &BTRFS_I(inode)->runtime_flags)) |
1812 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1813 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1814 | } |
a7e3b975 FM |
1815 | |
1816 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
1817 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1818 | spin_lock(&BTRFS_I(inode)->lock); | |
1819 | BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 - | |
1820 | state->start; | |
1821 | spin_unlock(&BTRFS_I(inode)->lock); | |
1822 | } | |
291d673e CM |
1823 | } |
1824 | ||
d352ac68 CM |
1825 | /* |
1826 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1827 | */ | |
c6100a4b | 1828 | static void btrfs_clear_bit_hook(void *private_data, |
41074888 | 1829 | struct extent_state *state, |
9ee49a04 | 1830 | unsigned *bits) |
291d673e | 1831 | { |
c6100a4b | 1832 | struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data); |
6fc0ef68 | 1833 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
47059d93 | 1834 | u64 len = state->end + 1 - state->start; |
823bb20a | 1835 | u32 num_extents = count_max_extents(len); |
47059d93 | 1836 | |
4a4b964f FM |
1837 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) { |
1838 | spin_lock(&inode->lock); | |
6fc0ef68 | 1839 | inode->defrag_bytes -= len; |
4a4b964f FM |
1840 | spin_unlock(&inode->lock); |
1841 | } | |
47059d93 | 1842 | |
75eff68e CM |
1843 | /* |
1844 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1845 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1846 | * bit, which is only set or cleared with irqs on |
1847 | */ | |
0ca1f7ce | 1848 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 1849 | struct btrfs_root *root = inode->root; |
83eea1f1 | 1850 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1851 | |
8b62f87b JB |
1852 | spin_lock(&inode->lock); |
1853 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
1854 | spin_unlock(&inode->lock); | |
0ca1f7ce | 1855 | |
b6d08f06 JB |
1856 | /* |
1857 | * We don't reserve metadata space for space cache inodes so we | |
1858 | * don't need to call dellalloc_release_metadata if there is an | |
1859 | * error. | |
1860 | */ | |
a315e68f | 1861 | if (*bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 1862 | root != fs_info->tree_root) |
43b18595 | 1863 | btrfs_delalloc_release_metadata(inode, len, false); |
0ca1f7ce | 1864 | |
6a3891c5 | 1865 | /* For sanity tests. */ |
0b246afa | 1866 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1867 | return; |
1868 | ||
a315e68f FM |
1869 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID && |
1870 | do_list && !(state->state & EXTENT_NORESERVE) && | |
1871 | (*bits & EXTENT_CLEAR_DATA_RESV)) | |
6fc0ef68 NB |
1872 | btrfs_free_reserved_data_space_noquota( |
1873 | &inode->vfs_inode, | |
51773bec | 1874 | state->start, len); |
9ed74f2d | 1875 | |
104b4e51 NB |
1876 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
1877 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
1878 | spin_lock(&inode->lock); |
1879 | inode->delalloc_bytes -= len; | |
1880 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 1881 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1882 | &inode->runtime_flags)) |
eb73c1b7 | 1883 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 1884 | spin_unlock(&inode->lock); |
291d673e | 1885 | } |
a7e3b975 FM |
1886 | |
1887 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
1888 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1889 | spin_lock(&inode->lock); | |
1890 | ASSERT(inode->new_delalloc_bytes >= len); | |
1891 | inode->new_delalloc_bytes -= len; | |
1892 | spin_unlock(&inode->lock); | |
1893 | } | |
291d673e CM |
1894 | } |
1895 | ||
d352ac68 CM |
1896 | /* |
1897 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1898 | * we don't create bios that span stripes or chunks | |
6f034ece LB |
1899 | * |
1900 | * return 1 if page cannot be merged to bio | |
1901 | * return 0 if page can be merged to bio | |
1902 | * return error otherwise | |
d352ac68 | 1903 | */ |
81a75f67 | 1904 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1905 | size_t size, struct bio *bio, |
1906 | unsigned long bio_flags) | |
239b14b3 | 1907 | { |
0b246afa JM |
1908 | struct inode *inode = page->mapping->host; |
1909 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
4f024f37 | 1910 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1911 | u64 length = 0; |
1912 | u64 map_length; | |
239b14b3 CM |
1913 | int ret; |
1914 | ||
771ed689 CM |
1915 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1916 | return 0; | |
1917 | ||
4f024f37 | 1918 | length = bio->bi_iter.bi_size; |
239b14b3 | 1919 | map_length = length; |
0b246afa JM |
1920 | ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length, |
1921 | NULL, 0); | |
6f034ece LB |
1922 | if (ret < 0) |
1923 | return ret; | |
d397712b | 1924 | if (map_length < length + size) |
239b14b3 | 1925 | return 1; |
3444a972 | 1926 | return 0; |
239b14b3 CM |
1927 | } |
1928 | ||
d352ac68 CM |
1929 | /* |
1930 | * in order to insert checksums into the metadata in large chunks, | |
1931 | * we wait until bio submission time. All the pages in the bio are | |
1932 | * checksummed and sums are attached onto the ordered extent record. | |
1933 | * | |
1934 | * At IO completion time the cums attached on the ordered extent record | |
1935 | * are inserted into the btree | |
1936 | */ | |
d0ee3934 | 1937 | static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio, |
eaf25d93 | 1938 | u64 bio_offset) |
065631f6 | 1939 | { |
c6100a4b | 1940 | struct inode *inode = private_data; |
4e4cbee9 | 1941 | blk_status_t ret = 0; |
e015640f | 1942 | |
2ff7e61e | 1943 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
79787eaa | 1944 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1945 | return 0; |
1946 | } | |
e015640f | 1947 | |
4a69a410 CM |
1948 | /* |
1949 | * in order to insert checksums into the metadata in large chunks, | |
1950 | * we wait until bio submission time. All the pages in the bio are | |
1951 | * checksummed and sums are attached onto the ordered extent record. | |
1952 | * | |
1953 | * At IO completion time the cums attached on the ordered extent record | |
1954 | * are inserted into the btree | |
1955 | */ | |
d0ee3934 | 1956 | static blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio, |
6c553435 | 1957 | int mirror_num) |
4a69a410 | 1958 | { |
c6100a4b | 1959 | struct inode *inode = private_data; |
2ff7e61e | 1960 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4e4cbee9 | 1961 | blk_status_t ret; |
61891923 | 1962 | |
2ff7e61e | 1963 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 1); |
4246a0b6 | 1964 | if (ret) { |
4e4cbee9 | 1965 | bio->bi_status = ret; |
4246a0b6 CH |
1966 | bio_endio(bio); |
1967 | } | |
61891923 | 1968 | return ret; |
44b8bd7e CM |
1969 | } |
1970 | ||
d352ac68 | 1971 | /* |
cad321ad | 1972 | * extent_io.c submission hook. This does the right thing for csum calculation |
4c274bc6 LB |
1973 | * on write, or reading the csums from the tree before a read. |
1974 | * | |
1975 | * Rules about async/sync submit, | |
1976 | * a) read: sync submit | |
1977 | * | |
1978 | * b) write without checksum: sync submit | |
1979 | * | |
1980 | * c) write with checksum: | |
1981 | * c-1) if bio is issued by fsync: sync submit | |
1982 | * (sync_writers != 0) | |
1983 | * | |
1984 | * c-2) if root is reloc root: sync submit | |
1985 | * (only in case of buffered IO) | |
1986 | * | |
1987 | * c-3) otherwise: async submit | |
d352ac68 | 1988 | */ |
8c27cb35 | 1989 | static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio, |
c6100a4b JB |
1990 | int mirror_num, unsigned long bio_flags, |
1991 | u64 bio_offset) | |
44b8bd7e | 1992 | { |
c6100a4b | 1993 | struct inode *inode = private_data; |
0b246afa | 1994 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 1995 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 1996 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
4e4cbee9 | 1997 | blk_status_t ret = 0; |
19b9bdb0 | 1998 | int skip_sum; |
b812ce28 | 1999 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 2000 | |
6cbff00f | 2001 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 2002 | |
70ddc553 | 2003 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 2004 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 2005 | |
37226b21 | 2006 | if (bio_op(bio) != REQ_OP_WRITE) { |
0b246afa | 2007 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 2008 | if (ret) |
61891923 | 2009 | goto out; |
5fd02043 | 2010 | |
d20f7043 | 2011 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
2012 | ret = btrfs_submit_compressed_read(inode, bio, |
2013 | mirror_num, | |
2014 | bio_flags); | |
2015 | goto out; | |
c2db1073 | 2016 | } else if (!skip_sum) { |
2ff7e61e | 2017 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 2018 | if (ret) |
61891923 | 2019 | goto out; |
c2db1073 | 2020 | } |
4d1b5fb4 | 2021 | goto mapit; |
b812ce28 | 2022 | } else if (async && !skip_sum) { |
17d217fe YZ |
2023 | /* csum items have already been cloned */ |
2024 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
2025 | goto mapit; | |
19b9bdb0 | 2026 | /* we're doing a write, do the async checksumming */ |
c6100a4b JB |
2027 | ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags, |
2028 | bio_offset, inode, | |
d0ee3934 DS |
2029 | btrfs_submit_bio_start, |
2030 | btrfs_submit_bio_done); | |
61891923 | 2031 | goto out; |
b812ce28 | 2032 | } else if (!skip_sum) { |
2ff7e61e | 2033 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
b812ce28 JB |
2034 | if (ret) |
2035 | goto out; | |
19b9bdb0 CM |
2036 | } |
2037 | ||
0b86a832 | 2038 | mapit: |
2ff7e61e | 2039 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
61891923 SB |
2040 | |
2041 | out: | |
4e4cbee9 CH |
2042 | if (ret) { |
2043 | bio->bi_status = ret; | |
4246a0b6 CH |
2044 | bio_endio(bio); |
2045 | } | |
61891923 | 2046 | return ret; |
065631f6 | 2047 | } |
6885f308 | 2048 | |
d352ac68 CM |
2049 | /* |
2050 | * given a list of ordered sums record them in the inode. This happens | |
2051 | * at IO completion time based on sums calculated at bio submission time. | |
2052 | */ | |
ba1da2f4 | 2053 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
df9f628e | 2054 | struct inode *inode, struct list_head *list) |
e6dcd2dc | 2055 | { |
e6dcd2dc | 2056 | struct btrfs_ordered_sum *sum; |
ac01f26a | 2057 | int ret; |
e6dcd2dc | 2058 | |
c6e30871 | 2059 | list_for_each_entry(sum, list, list) { |
7c2871a2 | 2060 | trans->adding_csums = true; |
ac01f26a | 2061 | ret = btrfs_csum_file_blocks(trans, |
d20f7043 | 2062 | BTRFS_I(inode)->root->fs_info->csum_root, sum); |
7c2871a2 | 2063 | trans->adding_csums = false; |
ac01f26a NB |
2064 | if (ret) |
2065 | return ret; | |
e6dcd2dc CM |
2066 | } |
2067 | return 0; | |
2068 | } | |
2069 | ||
2ac55d41 | 2070 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
e3b8a485 | 2071 | unsigned int extra_bits, |
ba8b04c1 | 2072 | struct extent_state **cached_state, int dedupe) |
ea8c2819 | 2073 | { |
09cbfeaf | 2074 | WARN_ON((end & (PAGE_SIZE - 1)) == 0); |
ea8c2819 | 2075 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
e3b8a485 | 2076 | extra_bits, cached_state); |
ea8c2819 CM |
2077 | } |
2078 | ||
d352ac68 | 2079 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2080 | struct btrfs_writepage_fixup { |
2081 | struct page *page; | |
2082 | struct btrfs_work work; | |
2083 | }; | |
2084 | ||
b2950863 | 2085 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2086 | { |
2087 | struct btrfs_writepage_fixup *fixup; | |
2088 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2089 | struct extent_state *cached_state = NULL; |
364ecf36 | 2090 | struct extent_changeset *data_reserved = NULL; |
247e743c CM |
2091 | struct page *page; |
2092 | struct inode *inode; | |
2093 | u64 page_start; | |
2094 | u64 page_end; | |
87826df0 | 2095 | int ret; |
247e743c CM |
2096 | |
2097 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2098 | page = fixup->page; | |
4a096752 | 2099 | again: |
247e743c CM |
2100 | lock_page(page); |
2101 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
2102 | ClearPageChecked(page); | |
2103 | goto out_page; | |
2104 | } | |
2105 | ||
2106 | inode = page->mapping->host; | |
2107 | page_start = page_offset(page); | |
09cbfeaf | 2108 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 2109 | |
ff13db41 | 2110 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 2111 | &cached_state); |
4a096752 CM |
2112 | |
2113 | /* already ordered? We're done */ | |
8b62b72b | 2114 | if (PagePrivate2(page)) |
247e743c | 2115 | goto out; |
4a096752 | 2116 | |
a776c6fa | 2117 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
09cbfeaf | 2118 | PAGE_SIZE); |
4a096752 | 2119 | if (ordered) { |
2ac55d41 | 2120 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
e43bbe5e | 2121 | page_end, &cached_state); |
4a096752 CM |
2122 | unlock_page(page); |
2123 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 2124 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2125 | goto again; |
2126 | } | |
247e743c | 2127 | |
364ecf36 | 2128 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
09cbfeaf | 2129 | PAGE_SIZE); |
87826df0 JM |
2130 | if (ret) { |
2131 | mapping_set_error(page->mapping, ret); | |
2132 | end_extent_writepage(page, ret, page_start, page_end); | |
2133 | ClearPageChecked(page); | |
2134 | goto out; | |
2135 | } | |
2136 | ||
f3038ee3 NB |
2137 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, |
2138 | &cached_state, 0); | |
2139 | if (ret) { | |
2140 | mapping_set_error(page->mapping, ret); | |
2141 | end_extent_writepage(page, ret, page_start, page_end); | |
2142 | ClearPageChecked(page); | |
2143 | goto out; | |
2144 | } | |
2145 | ||
247e743c | 2146 | ClearPageChecked(page); |
87826df0 | 2147 | set_page_dirty(page); |
43b18595 | 2148 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false); |
247e743c | 2149 | out: |
2ac55d41 | 2150 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
e43bbe5e | 2151 | &cached_state); |
247e743c CM |
2152 | out_page: |
2153 | unlock_page(page); | |
09cbfeaf | 2154 | put_page(page); |
b897abec | 2155 | kfree(fixup); |
364ecf36 | 2156 | extent_changeset_free(data_reserved); |
247e743c CM |
2157 | } |
2158 | ||
2159 | /* | |
2160 | * There are a few paths in the higher layers of the kernel that directly | |
2161 | * set the page dirty bit without asking the filesystem if it is a | |
2162 | * good idea. This causes problems because we want to make sure COW | |
2163 | * properly happens and the data=ordered rules are followed. | |
2164 | * | |
c8b97818 | 2165 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2166 | * hasn't been properly setup for IO. We kick off an async process |
2167 | * to fix it up. The async helper will wait for ordered extents, set | |
2168 | * the delalloc bit and make it safe to write the page. | |
2169 | */ | |
b2950863 | 2170 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
2171 | { |
2172 | struct inode *inode = page->mapping->host; | |
0b246afa | 2173 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2174 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2175 | |
8b62b72b CM |
2176 | /* this page is properly in the ordered list */ |
2177 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2178 | return 0; |
2179 | ||
2180 | if (PageChecked(page)) | |
2181 | return -EAGAIN; | |
2182 | ||
2183 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2184 | if (!fixup) | |
2185 | return -EAGAIN; | |
f421950f | 2186 | |
247e743c | 2187 | SetPageChecked(page); |
09cbfeaf | 2188 | get_page(page); |
9e0af237 LB |
2189 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2190 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2191 | fixup->page = page; |
0b246afa | 2192 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
87826df0 | 2193 | return -EBUSY; |
247e743c CM |
2194 | } |
2195 | ||
d899e052 YZ |
2196 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2197 | struct inode *inode, u64 file_pos, | |
2198 | u64 disk_bytenr, u64 disk_num_bytes, | |
2199 | u64 num_bytes, u64 ram_bytes, | |
2200 | u8 compression, u8 encryption, | |
2201 | u16 other_encoding, int extent_type) | |
2202 | { | |
2203 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2204 | struct btrfs_file_extent_item *fi; | |
2205 | struct btrfs_path *path; | |
2206 | struct extent_buffer *leaf; | |
2207 | struct btrfs_key ins; | |
a12b877b | 2208 | u64 qg_released; |
1acae57b | 2209 | int extent_inserted = 0; |
d899e052 YZ |
2210 | int ret; |
2211 | ||
2212 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2213 | if (!path) |
2214 | return -ENOMEM; | |
d899e052 | 2215 | |
a1ed835e CM |
2216 | /* |
2217 | * we may be replacing one extent in the tree with another. | |
2218 | * The new extent is pinned in the extent map, and we don't want | |
2219 | * to drop it from the cache until it is completely in the btree. | |
2220 | * | |
2221 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2222 | * the caller is expected to unpin it and allow it to be merged | |
2223 | * with the others. | |
2224 | */ | |
1acae57b FDBM |
2225 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2226 | file_pos + num_bytes, NULL, 0, | |
2227 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2228 | if (ret) |
2229 | goto out; | |
d899e052 | 2230 | |
1acae57b | 2231 | if (!extent_inserted) { |
4a0cc7ca | 2232 | ins.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b FDBM |
2233 | ins.offset = file_pos; |
2234 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2235 | ||
2236 | path->leave_spinning = 1; | |
2237 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2238 | sizeof(*fi)); | |
2239 | if (ret) | |
2240 | goto out; | |
2241 | } | |
d899e052 YZ |
2242 | leaf = path->nodes[0]; |
2243 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2244 | struct btrfs_file_extent_item); | |
2245 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2246 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2247 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2248 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2249 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2250 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2251 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2252 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2253 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2254 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2255 | |
d899e052 | 2256 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2257 | btrfs_release_path(path); |
d899e052 YZ |
2258 | |
2259 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2260 | |
2261 | ins.objectid = disk_bytenr; | |
2262 | ins.offset = disk_num_bytes; | |
2263 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2264 | |
297d750b | 2265 | /* |
5846a3c2 QW |
2266 | * Release the reserved range from inode dirty range map, as it is |
2267 | * already moved into delayed_ref_head | |
297d750b | 2268 | */ |
a12b877b QW |
2269 | ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes); |
2270 | if (ret < 0) | |
2271 | goto out; | |
2272 | qg_released = ret; | |
84f7d8e6 JB |
2273 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2274 | btrfs_ino(BTRFS_I(inode)), | |
2275 | file_pos, qg_released, &ins); | |
79787eaa | 2276 | out: |
d899e052 | 2277 | btrfs_free_path(path); |
b9473439 | 2278 | |
79787eaa | 2279 | return ret; |
d899e052 YZ |
2280 | } |
2281 | ||
38c227d8 LB |
2282 | /* snapshot-aware defrag */ |
2283 | struct sa_defrag_extent_backref { | |
2284 | struct rb_node node; | |
2285 | struct old_sa_defrag_extent *old; | |
2286 | u64 root_id; | |
2287 | u64 inum; | |
2288 | u64 file_pos; | |
2289 | u64 extent_offset; | |
2290 | u64 num_bytes; | |
2291 | u64 generation; | |
2292 | }; | |
2293 | ||
2294 | struct old_sa_defrag_extent { | |
2295 | struct list_head list; | |
2296 | struct new_sa_defrag_extent *new; | |
2297 | ||
2298 | u64 extent_offset; | |
2299 | u64 bytenr; | |
2300 | u64 offset; | |
2301 | u64 len; | |
2302 | int count; | |
2303 | }; | |
2304 | ||
2305 | struct new_sa_defrag_extent { | |
2306 | struct rb_root root; | |
2307 | struct list_head head; | |
2308 | struct btrfs_path *path; | |
2309 | struct inode *inode; | |
2310 | u64 file_pos; | |
2311 | u64 len; | |
2312 | u64 bytenr; | |
2313 | u64 disk_len; | |
2314 | u8 compress_type; | |
2315 | }; | |
2316 | ||
2317 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2318 | struct sa_defrag_extent_backref *b2) | |
2319 | { | |
2320 | if (b1->root_id < b2->root_id) | |
2321 | return -1; | |
2322 | else if (b1->root_id > b2->root_id) | |
2323 | return 1; | |
2324 | ||
2325 | if (b1->inum < b2->inum) | |
2326 | return -1; | |
2327 | else if (b1->inum > b2->inum) | |
2328 | return 1; | |
2329 | ||
2330 | if (b1->file_pos < b2->file_pos) | |
2331 | return -1; | |
2332 | else if (b1->file_pos > b2->file_pos) | |
2333 | return 1; | |
2334 | ||
2335 | /* | |
2336 | * [------------------------------] ===> (a range of space) | |
2337 | * |<--->| |<---->| =============> (fs/file tree A) | |
2338 | * |<---------------------------->| ===> (fs/file tree B) | |
2339 | * | |
2340 | * A range of space can refer to two file extents in one tree while | |
2341 | * refer to only one file extent in another tree. | |
2342 | * | |
2343 | * So we may process a disk offset more than one time(two extents in A) | |
2344 | * and locate at the same extent(one extent in B), then insert two same | |
2345 | * backrefs(both refer to the extent in B). | |
2346 | */ | |
2347 | return 0; | |
2348 | } | |
2349 | ||
2350 | static void backref_insert(struct rb_root *root, | |
2351 | struct sa_defrag_extent_backref *backref) | |
2352 | { | |
2353 | struct rb_node **p = &root->rb_node; | |
2354 | struct rb_node *parent = NULL; | |
2355 | struct sa_defrag_extent_backref *entry; | |
2356 | int ret; | |
2357 | ||
2358 | while (*p) { | |
2359 | parent = *p; | |
2360 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2361 | ||
2362 | ret = backref_comp(backref, entry); | |
2363 | if (ret < 0) | |
2364 | p = &(*p)->rb_left; | |
2365 | else | |
2366 | p = &(*p)->rb_right; | |
2367 | } | |
2368 | ||
2369 | rb_link_node(&backref->node, parent, p); | |
2370 | rb_insert_color(&backref->node, root); | |
2371 | } | |
2372 | ||
2373 | /* | |
2374 | * Note the backref might has changed, and in this case we just return 0. | |
2375 | */ | |
2376 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2377 | void *ctx) | |
2378 | { | |
2379 | struct btrfs_file_extent_item *extent; | |
38c227d8 LB |
2380 | struct old_sa_defrag_extent *old = ctx; |
2381 | struct new_sa_defrag_extent *new = old->new; | |
2382 | struct btrfs_path *path = new->path; | |
2383 | struct btrfs_key key; | |
2384 | struct btrfs_root *root; | |
2385 | struct sa_defrag_extent_backref *backref; | |
2386 | struct extent_buffer *leaf; | |
2387 | struct inode *inode = new->inode; | |
0b246afa | 2388 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2389 | int slot; |
2390 | int ret; | |
2391 | u64 extent_offset; | |
2392 | u64 num_bytes; | |
2393 | ||
2394 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
4a0cc7ca | 2395 | inum == btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2396 | return 0; |
2397 | ||
2398 | key.objectid = root_id; | |
2399 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2400 | key.offset = (u64)-1; | |
2401 | ||
38c227d8 LB |
2402 | root = btrfs_read_fs_root_no_name(fs_info, &key); |
2403 | if (IS_ERR(root)) { | |
2404 | if (PTR_ERR(root) == -ENOENT) | |
2405 | return 0; | |
2406 | WARN_ON(1); | |
ab8d0fc4 | 2407 | btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu", |
38c227d8 LB |
2408 | inum, offset, root_id); |
2409 | return PTR_ERR(root); | |
2410 | } | |
2411 | ||
2412 | key.objectid = inum; | |
2413 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2414 | if (offset > (u64)-1 << 32) | |
2415 | key.offset = 0; | |
2416 | else | |
2417 | key.offset = offset; | |
2418 | ||
2419 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2420 | if (WARN_ON(ret < 0)) |
38c227d8 | 2421 | return ret; |
50f1319c | 2422 | ret = 0; |
38c227d8 LB |
2423 | |
2424 | while (1) { | |
2425 | cond_resched(); | |
2426 | ||
2427 | leaf = path->nodes[0]; | |
2428 | slot = path->slots[0]; | |
2429 | ||
2430 | if (slot >= btrfs_header_nritems(leaf)) { | |
2431 | ret = btrfs_next_leaf(root, path); | |
2432 | if (ret < 0) { | |
2433 | goto out; | |
2434 | } else if (ret > 0) { | |
2435 | ret = 0; | |
2436 | goto out; | |
2437 | } | |
2438 | continue; | |
2439 | } | |
2440 | ||
2441 | path->slots[0]++; | |
2442 | ||
2443 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2444 | ||
2445 | if (key.objectid > inum) | |
2446 | goto out; | |
2447 | ||
2448 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2449 | continue; | |
2450 | ||
2451 | extent = btrfs_item_ptr(leaf, slot, | |
2452 | struct btrfs_file_extent_item); | |
2453 | ||
2454 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2455 | continue; | |
2456 | ||
e68afa49 LB |
2457 | /* |
2458 | * 'offset' refers to the exact key.offset, | |
2459 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2460 | * (key.offset - extent_offset). | |
2461 | */ | |
2462 | if (key.offset != offset) | |
38c227d8 LB |
2463 | continue; |
2464 | ||
e68afa49 | 2465 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2466 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2467 | |
38c227d8 LB |
2468 | if (extent_offset >= old->extent_offset + old->offset + |
2469 | old->len || extent_offset + num_bytes <= | |
2470 | old->extent_offset + old->offset) | |
2471 | continue; | |
38c227d8 LB |
2472 | break; |
2473 | } | |
2474 | ||
2475 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2476 | if (!backref) { | |
2477 | ret = -ENOENT; | |
2478 | goto out; | |
2479 | } | |
2480 | ||
2481 | backref->root_id = root_id; | |
2482 | backref->inum = inum; | |
e68afa49 | 2483 | backref->file_pos = offset; |
38c227d8 LB |
2484 | backref->num_bytes = num_bytes; |
2485 | backref->extent_offset = extent_offset; | |
2486 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2487 | backref->old = old; | |
2488 | backref_insert(&new->root, backref); | |
2489 | old->count++; | |
2490 | out: | |
2491 | btrfs_release_path(path); | |
2492 | WARN_ON(ret); | |
2493 | return ret; | |
2494 | } | |
2495 | ||
2496 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2497 | struct new_sa_defrag_extent *new) | |
2498 | { | |
0b246afa | 2499 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2500 | struct old_sa_defrag_extent *old, *tmp; |
2501 | int ret; | |
2502 | ||
2503 | new->path = path; | |
2504 | ||
2505 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2506 | ret = iterate_inodes_from_logical(old->bytenr + |
2507 | old->extent_offset, fs_info, | |
38c227d8 | 2508 | path, record_one_backref, |
c995ab3c | 2509 | old, false); |
4724b106 JB |
2510 | if (ret < 0 && ret != -ENOENT) |
2511 | return false; | |
38c227d8 LB |
2512 | |
2513 | /* no backref to be processed for this extent */ | |
2514 | if (!old->count) { | |
2515 | list_del(&old->list); | |
2516 | kfree(old); | |
2517 | } | |
2518 | } | |
2519 | ||
2520 | if (list_empty(&new->head)) | |
2521 | return false; | |
2522 | ||
2523 | return true; | |
2524 | } | |
2525 | ||
2526 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2527 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2528 | struct new_sa_defrag_extent *new) |
38c227d8 | 2529 | { |
116e0024 | 2530 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2531 | return 0; |
2532 | ||
2533 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2534 | return 0; | |
2535 | ||
116e0024 LB |
2536 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2537 | return 0; | |
2538 | ||
2539 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2540 | btrfs_file_extent_other_encoding(leaf, fi)) |
2541 | return 0; | |
2542 | ||
2543 | return 1; | |
2544 | } | |
2545 | ||
2546 | /* | |
2547 | * Note the backref might has changed, and in this case we just return 0. | |
2548 | */ | |
2549 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2550 | struct sa_defrag_extent_backref *prev, | |
2551 | struct sa_defrag_extent_backref *backref) | |
2552 | { | |
2553 | struct btrfs_file_extent_item *extent; | |
2554 | struct btrfs_file_extent_item *item; | |
2555 | struct btrfs_ordered_extent *ordered; | |
2556 | struct btrfs_trans_handle *trans; | |
38c227d8 LB |
2557 | struct btrfs_root *root; |
2558 | struct btrfs_key key; | |
2559 | struct extent_buffer *leaf; | |
2560 | struct old_sa_defrag_extent *old = backref->old; | |
2561 | struct new_sa_defrag_extent *new = old->new; | |
0b246afa | 2562 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2563 | struct inode *inode; |
2564 | struct extent_state *cached = NULL; | |
2565 | int ret = 0; | |
2566 | u64 start; | |
2567 | u64 len; | |
2568 | u64 lock_start; | |
2569 | u64 lock_end; | |
2570 | bool merge = false; | |
2571 | int index; | |
2572 | ||
2573 | if (prev && prev->root_id == backref->root_id && | |
2574 | prev->inum == backref->inum && | |
2575 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2576 | merge = true; | |
2577 | ||
2578 | /* step 1: get root */ | |
2579 | key.objectid = backref->root_id; | |
2580 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2581 | key.offset = (u64)-1; | |
2582 | ||
38c227d8 LB |
2583 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2584 | ||
2585 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2586 | if (IS_ERR(root)) { | |
2587 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2588 | if (PTR_ERR(root) == -ENOENT) | |
2589 | return 0; | |
2590 | return PTR_ERR(root); | |
2591 | } | |
38c227d8 | 2592 | |
bcbba5e6 WS |
2593 | if (btrfs_root_readonly(root)) { |
2594 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2595 | return 0; | |
2596 | } | |
2597 | ||
38c227d8 LB |
2598 | /* step 2: get inode */ |
2599 | key.objectid = backref->inum; | |
2600 | key.type = BTRFS_INODE_ITEM_KEY; | |
2601 | key.offset = 0; | |
2602 | ||
2603 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2604 | if (IS_ERR(inode)) { | |
2605 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2606 | return 0; | |
2607 | } | |
2608 | ||
2609 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2610 | ||
2611 | /* step 3: relink backref */ | |
2612 | lock_start = backref->file_pos; | |
2613 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2614 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2615 | &cached); |
38c227d8 LB |
2616 | |
2617 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2618 | if (ordered) { | |
2619 | btrfs_put_ordered_extent(ordered); | |
2620 | goto out_unlock; | |
2621 | } | |
2622 | ||
2623 | trans = btrfs_join_transaction(root); | |
2624 | if (IS_ERR(trans)) { | |
2625 | ret = PTR_ERR(trans); | |
2626 | goto out_unlock; | |
2627 | } | |
2628 | ||
2629 | key.objectid = backref->inum; | |
2630 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2631 | key.offset = backref->file_pos; | |
2632 | ||
2633 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2634 | if (ret < 0) { | |
2635 | goto out_free_path; | |
2636 | } else if (ret > 0) { | |
2637 | ret = 0; | |
2638 | goto out_free_path; | |
2639 | } | |
2640 | ||
2641 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2642 | struct btrfs_file_extent_item); | |
2643 | ||
2644 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2645 | backref->generation) | |
2646 | goto out_free_path; | |
2647 | ||
2648 | btrfs_release_path(path); | |
2649 | ||
2650 | start = backref->file_pos; | |
2651 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2652 | start += old->extent_offset + old->offset - | |
2653 | backref->extent_offset; | |
2654 | ||
2655 | len = min(backref->extent_offset + backref->num_bytes, | |
2656 | old->extent_offset + old->offset + old->len); | |
2657 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2658 | ||
2659 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2660 | start + len, 1); | |
2661 | if (ret) | |
2662 | goto out_free_path; | |
2663 | again: | |
4a0cc7ca | 2664 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2665 | key.type = BTRFS_EXTENT_DATA_KEY; |
2666 | key.offset = start; | |
2667 | ||
a09a0a70 | 2668 | path->leave_spinning = 1; |
38c227d8 LB |
2669 | if (merge) { |
2670 | struct btrfs_file_extent_item *fi; | |
2671 | u64 extent_len; | |
2672 | struct btrfs_key found_key; | |
2673 | ||
3c9665df | 2674 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2675 | if (ret < 0) |
2676 | goto out_free_path; | |
2677 | ||
2678 | path->slots[0]--; | |
2679 | leaf = path->nodes[0]; | |
2680 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2681 | ||
2682 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2683 | struct btrfs_file_extent_item); | |
2684 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2685 | ||
116e0024 LB |
2686 | if (extent_len + found_key.offset == start && |
2687 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2688 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2689 | extent_len + len); | |
2690 | btrfs_mark_buffer_dirty(leaf); | |
2691 | inode_add_bytes(inode, len); | |
2692 | ||
2693 | ret = 1; | |
2694 | goto out_free_path; | |
2695 | } else { | |
2696 | merge = false; | |
2697 | btrfs_release_path(path); | |
2698 | goto again; | |
2699 | } | |
2700 | } | |
2701 | ||
2702 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2703 | sizeof(*extent)); | |
2704 | if (ret) { | |
66642832 | 2705 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2706 | goto out_free_path; |
2707 | } | |
2708 | ||
2709 | leaf = path->nodes[0]; | |
2710 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2711 | struct btrfs_file_extent_item); | |
2712 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2713 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2714 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2715 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2716 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2717 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2718 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2719 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2720 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2721 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2722 | ||
2723 | btrfs_mark_buffer_dirty(leaf); | |
2724 | inode_add_bytes(inode, len); | |
a09a0a70 | 2725 | btrfs_release_path(path); |
38c227d8 | 2726 | |
84f7d8e6 | 2727 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, |
38c227d8 LB |
2728 | new->disk_len, 0, |
2729 | backref->root_id, backref->inum, | |
b06c4bf5 | 2730 | new->file_pos); /* start - extent_offset */ |
38c227d8 | 2731 | if (ret) { |
66642832 | 2732 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2733 | goto out_free_path; |
2734 | } | |
2735 | ||
2736 | ret = 1; | |
2737 | out_free_path: | |
2738 | btrfs_release_path(path); | |
a09a0a70 | 2739 | path->leave_spinning = 0; |
3a45bb20 | 2740 | btrfs_end_transaction(trans); |
38c227d8 LB |
2741 | out_unlock: |
2742 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
e43bbe5e | 2743 | &cached); |
38c227d8 LB |
2744 | iput(inode); |
2745 | return ret; | |
2746 | } | |
2747 | ||
6f519564 LB |
2748 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2749 | { | |
2750 | struct old_sa_defrag_extent *old, *tmp; | |
2751 | ||
2752 | if (!new) | |
2753 | return; | |
2754 | ||
2755 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2756 | kfree(old); |
2757 | } | |
2758 | kfree(new); | |
2759 | } | |
2760 | ||
38c227d8 LB |
2761 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2762 | { | |
0b246afa | 2763 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 | 2764 | struct btrfs_path *path; |
38c227d8 LB |
2765 | struct sa_defrag_extent_backref *backref; |
2766 | struct sa_defrag_extent_backref *prev = NULL; | |
2767 | struct inode *inode; | |
38c227d8 LB |
2768 | struct rb_node *node; |
2769 | int ret; | |
2770 | ||
2771 | inode = new->inode; | |
38c227d8 LB |
2772 | |
2773 | path = btrfs_alloc_path(); | |
2774 | if (!path) | |
2775 | return; | |
2776 | ||
2777 | if (!record_extent_backrefs(path, new)) { | |
2778 | btrfs_free_path(path); | |
2779 | goto out; | |
2780 | } | |
2781 | btrfs_release_path(path); | |
2782 | ||
2783 | while (1) { | |
2784 | node = rb_first(&new->root); | |
2785 | if (!node) | |
2786 | break; | |
2787 | rb_erase(node, &new->root); | |
2788 | ||
2789 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2790 | ||
2791 | ret = relink_extent_backref(path, prev, backref); | |
2792 | WARN_ON(ret < 0); | |
2793 | ||
2794 | kfree(prev); | |
2795 | ||
2796 | if (ret == 1) | |
2797 | prev = backref; | |
2798 | else | |
2799 | prev = NULL; | |
2800 | cond_resched(); | |
2801 | } | |
2802 | kfree(prev); | |
2803 | ||
2804 | btrfs_free_path(path); | |
38c227d8 | 2805 | out: |
6f519564 LB |
2806 | free_sa_defrag_extent(new); |
2807 | ||
0b246afa JM |
2808 | atomic_dec(&fs_info->defrag_running); |
2809 | wake_up(&fs_info->transaction_wait); | |
38c227d8 LB |
2810 | } |
2811 | ||
2812 | static struct new_sa_defrag_extent * | |
2813 | record_old_file_extents(struct inode *inode, | |
2814 | struct btrfs_ordered_extent *ordered) | |
2815 | { | |
0b246afa | 2816 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2817 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2818 | struct btrfs_path *path; | |
2819 | struct btrfs_key key; | |
6f519564 | 2820 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2821 | struct new_sa_defrag_extent *new; |
2822 | int ret; | |
2823 | ||
2824 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2825 | if (!new) | |
2826 | return NULL; | |
2827 | ||
2828 | new->inode = inode; | |
2829 | new->file_pos = ordered->file_offset; | |
2830 | new->len = ordered->len; | |
2831 | new->bytenr = ordered->start; | |
2832 | new->disk_len = ordered->disk_len; | |
2833 | new->compress_type = ordered->compress_type; | |
2834 | new->root = RB_ROOT; | |
2835 | INIT_LIST_HEAD(&new->head); | |
2836 | ||
2837 | path = btrfs_alloc_path(); | |
2838 | if (!path) | |
2839 | goto out_kfree; | |
2840 | ||
4a0cc7ca | 2841 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2842 | key.type = BTRFS_EXTENT_DATA_KEY; |
2843 | key.offset = new->file_pos; | |
2844 | ||
2845 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2846 | if (ret < 0) | |
2847 | goto out_free_path; | |
2848 | if (ret > 0 && path->slots[0] > 0) | |
2849 | path->slots[0]--; | |
2850 | ||
2851 | /* find out all the old extents for the file range */ | |
2852 | while (1) { | |
2853 | struct btrfs_file_extent_item *extent; | |
2854 | struct extent_buffer *l; | |
2855 | int slot; | |
2856 | u64 num_bytes; | |
2857 | u64 offset; | |
2858 | u64 end; | |
2859 | u64 disk_bytenr; | |
2860 | u64 extent_offset; | |
2861 | ||
2862 | l = path->nodes[0]; | |
2863 | slot = path->slots[0]; | |
2864 | ||
2865 | if (slot >= btrfs_header_nritems(l)) { | |
2866 | ret = btrfs_next_leaf(root, path); | |
2867 | if (ret < 0) | |
6f519564 | 2868 | goto out_free_path; |
38c227d8 LB |
2869 | else if (ret > 0) |
2870 | break; | |
2871 | continue; | |
2872 | } | |
2873 | ||
2874 | btrfs_item_key_to_cpu(l, &key, slot); | |
2875 | ||
4a0cc7ca | 2876 | if (key.objectid != btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2877 | break; |
2878 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2879 | break; | |
2880 | if (key.offset >= new->file_pos + new->len) | |
2881 | break; | |
2882 | ||
2883 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2884 | ||
2885 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2886 | if (key.offset + num_bytes < new->file_pos) | |
2887 | goto next; | |
2888 | ||
2889 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2890 | if (!disk_bytenr) | |
2891 | goto next; | |
2892 | ||
2893 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2894 | ||
2895 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2896 | if (!old) | |
6f519564 | 2897 | goto out_free_path; |
38c227d8 LB |
2898 | |
2899 | offset = max(new->file_pos, key.offset); | |
2900 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2901 | ||
2902 | old->bytenr = disk_bytenr; | |
2903 | old->extent_offset = extent_offset; | |
2904 | old->offset = offset - key.offset; | |
2905 | old->len = end - offset; | |
2906 | old->new = new; | |
2907 | old->count = 0; | |
2908 | list_add_tail(&old->list, &new->head); | |
2909 | next: | |
2910 | path->slots[0]++; | |
2911 | cond_resched(); | |
2912 | } | |
2913 | ||
2914 | btrfs_free_path(path); | |
0b246afa | 2915 | atomic_inc(&fs_info->defrag_running); |
38c227d8 LB |
2916 | |
2917 | return new; | |
2918 | ||
38c227d8 LB |
2919 | out_free_path: |
2920 | btrfs_free_path(path); | |
2921 | out_kfree: | |
6f519564 | 2922 | free_sa_defrag_extent(new); |
38c227d8 LB |
2923 | return NULL; |
2924 | } | |
2925 | ||
2ff7e61e | 2926 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2927 | u64 start, u64 len) |
2928 | { | |
2929 | struct btrfs_block_group_cache *cache; | |
2930 | ||
0b246afa | 2931 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2932 | ASSERT(cache); |
2933 | ||
2934 | spin_lock(&cache->lock); | |
2935 | cache->delalloc_bytes -= len; | |
2936 | spin_unlock(&cache->lock); | |
2937 | ||
2938 | btrfs_put_block_group(cache); | |
2939 | } | |
2940 | ||
d352ac68 CM |
2941 | /* as ordered data IO finishes, this gets called so we can finish |
2942 | * an ordered extent if the range of bytes in the file it covers are | |
2943 | * fully written. | |
2944 | */ | |
5fd02043 | 2945 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2946 | { |
5fd02043 | 2947 | struct inode *inode = ordered_extent->inode; |
0b246afa | 2948 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 2949 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2950 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2951 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2952 | struct extent_state *cached_state = NULL; |
38c227d8 | 2953 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2954 | int compress_type = 0; |
77cef2ec JB |
2955 | int ret = 0; |
2956 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2957 | bool nolock; |
77cef2ec | 2958 | bool truncated = false; |
a7e3b975 FM |
2959 | bool range_locked = false; |
2960 | bool clear_new_delalloc_bytes = false; | |
2961 | ||
2962 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
2963 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
2964 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags)) | |
2965 | clear_new_delalloc_bytes = true; | |
e6dcd2dc | 2966 | |
70ddc553 | 2967 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
0cb59c99 | 2968 | |
5fd02043 JB |
2969 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2970 | ret = -EIO; | |
2971 | goto out; | |
2972 | } | |
2973 | ||
7ab7956e NB |
2974 | btrfs_free_io_failure_record(BTRFS_I(inode), |
2975 | ordered_extent->file_offset, | |
2976 | ordered_extent->file_offset + | |
2977 | ordered_extent->len - 1); | |
f612496b | 2978 | |
77cef2ec JB |
2979 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2980 | truncated = true; | |
2981 | logical_len = ordered_extent->truncated_len; | |
2982 | /* Truncated the entire extent, don't bother adding */ | |
2983 | if (!logical_len) | |
2984 | goto out; | |
2985 | } | |
2986 | ||
c2167754 | 2987 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2988 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
2989 | |
2990 | /* | |
2991 | * For mwrite(mmap + memset to write) case, we still reserve | |
2992 | * space for NOCOW range. | |
2993 | * As NOCOW won't cause a new delayed ref, just free the space | |
2994 | */ | |
bc42bda2 | 2995 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
94ed938a | 2996 | ordered_extent->len); |
6c760c07 JB |
2997 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2998 | if (nolock) | |
2999 | trans = btrfs_join_transaction_nolock(root); | |
3000 | else | |
3001 | trans = btrfs_join_transaction(root); | |
3002 | if (IS_ERR(trans)) { | |
3003 | ret = PTR_ERR(trans); | |
3004 | trans = NULL; | |
3005 | goto out; | |
c2167754 | 3006 | } |
69fe2d75 | 3007 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
6c760c07 JB |
3008 | ret = btrfs_update_inode_fallback(trans, root, inode); |
3009 | if (ret) /* -ENOMEM or corruption */ | |
66642832 | 3010 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
3011 | goto out; |
3012 | } | |
e6dcd2dc | 3013 | |
a7e3b975 | 3014 | range_locked = true; |
2ac55d41 JB |
3015 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
3016 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 3017 | &cached_state); |
e6dcd2dc | 3018 | |
38c227d8 LB |
3019 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
3020 | ordered_extent->file_offset + ordered_extent->len - 1, | |
452e62b7 | 3021 | EXTENT_DEFRAG, 0, cached_state); |
38c227d8 LB |
3022 | if (ret) { |
3023 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 3024 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
3025 | /* the inode is shared */ |
3026 | new = record_old_file_extents(inode, ordered_extent); | |
3027 | ||
3028 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
3029 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ae0f1625 | 3030 | EXTENT_DEFRAG, 0, 0, &cached_state); |
38c227d8 LB |
3031 | } |
3032 | ||
0cb59c99 | 3033 | if (nolock) |
7a7eaa40 | 3034 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 3035 | else |
7a7eaa40 | 3036 | trans = btrfs_join_transaction(root); |
79787eaa JM |
3037 | if (IS_ERR(trans)) { |
3038 | ret = PTR_ERR(trans); | |
3039 | trans = NULL; | |
a7e3b975 | 3040 | goto out; |
79787eaa | 3041 | } |
a79b7d4b | 3042 | |
69fe2d75 | 3043 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
c2167754 | 3044 | |
c8b97818 | 3045 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3046 | compress_type = ordered_extent->compress_type; |
d899e052 | 3047 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3048 | BUG_ON(compress_type); |
b430b775 JM |
3049 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
3050 | ordered_extent->len); | |
7a6d7067 | 3051 | ret = btrfs_mark_extent_written(trans, BTRFS_I(inode), |
d899e052 YZ |
3052 | ordered_extent->file_offset, |
3053 | ordered_extent->file_offset + | |
77cef2ec | 3054 | logical_len); |
d899e052 | 3055 | } else { |
0b246afa | 3056 | BUG_ON(root == fs_info->tree_root); |
d899e052 YZ |
3057 | ret = insert_reserved_file_extent(trans, inode, |
3058 | ordered_extent->file_offset, | |
3059 | ordered_extent->start, | |
3060 | ordered_extent->disk_len, | |
77cef2ec | 3061 | logical_len, logical_len, |
261507a0 | 3062 | compress_type, 0, 0, |
d899e052 | 3063 | BTRFS_FILE_EXTENT_REG); |
e570fd27 | 3064 | if (!ret) |
2ff7e61e | 3065 | btrfs_release_delalloc_bytes(fs_info, |
e570fd27 MX |
3066 | ordered_extent->start, |
3067 | ordered_extent->disk_len); | |
d899e052 | 3068 | } |
5dc562c5 JB |
3069 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
3070 | ordered_extent->file_offset, ordered_extent->len, | |
3071 | trans->transid); | |
79787eaa | 3072 | if (ret < 0) { |
66642832 | 3073 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3074 | goto out; |
79787eaa | 3075 | } |
2ac55d41 | 3076 | |
ac01f26a NB |
3077 | ret = add_pending_csums(trans, inode, &ordered_extent->list); |
3078 | if (ret) { | |
3079 | btrfs_abort_transaction(trans, ret); | |
3080 | goto out; | |
3081 | } | |
e6dcd2dc | 3082 | |
6c760c07 JB |
3083 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
3084 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
3085 | if (ret) { /* -ENOMEM or corruption */ | |
66642832 | 3086 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3087 | goto out; |
1ef30be1 JB |
3088 | } |
3089 | ret = 0; | |
c2167754 | 3090 | out: |
a7e3b975 FM |
3091 | if (range_locked || clear_new_delalloc_bytes) { |
3092 | unsigned int clear_bits = 0; | |
3093 | ||
3094 | if (range_locked) | |
3095 | clear_bits |= EXTENT_LOCKED; | |
3096 | if (clear_new_delalloc_bytes) | |
3097 | clear_bits |= EXTENT_DELALLOC_NEW; | |
3098 | clear_extent_bit(&BTRFS_I(inode)->io_tree, | |
3099 | ordered_extent->file_offset, | |
3100 | ordered_extent->file_offset + | |
3101 | ordered_extent->len - 1, | |
3102 | clear_bits, | |
3103 | (clear_bits & EXTENT_LOCKED) ? 1 : 0, | |
ae0f1625 | 3104 | 0, &cached_state); |
a7e3b975 FM |
3105 | } |
3106 | ||
a698d075 | 3107 | if (trans) |
3a45bb20 | 3108 | btrfs_end_transaction(trans); |
0cb59c99 | 3109 | |
77cef2ec JB |
3110 | if (ret || truncated) { |
3111 | u64 start, end; | |
3112 | ||
3113 | if (truncated) | |
3114 | start = ordered_extent->file_offset + logical_len; | |
3115 | else | |
3116 | start = ordered_extent->file_offset; | |
3117 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
f08dc36f | 3118 | clear_extent_uptodate(io_tree, start, end, NULL); |
77cef2ec JB |
3119 | |
3120 | /* Drop the cache for the part of the extent we didn't write. */ | |
dcdbc059 | 3121 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); |
5fd02043 | 3122 | |
0bec9ef5 JB |
3123 | /* |
3124 | * If the ordered extent had an IOERR or something else went | |
3125 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3126 | * back to the allocator. We only free the extent in the |
3127 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 3128 | */ |
77cef2ec JB |
3129 | if ((ret || !logical_len) && |
3130 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 | 3131 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2ff7e61e JM |
3132 | btrfs_free_reserved_extent(fs_info, |
3133 | ordered_extent->start, | |
e570fd27 | 3134 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
3135 | } |
3136 | ||
3137 | ||
5fd02043 | 3138 | /* |
8bad3c02 LB |
3139 | * This needs to be done to make sure anybody waiting knows we are done |
3140 | * updating everything for this ordered extent. | |
5fd02043 JB |
3141 | */ |
3142 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
3143 | ||
38c227d8 | 3144 | /* for snapshot-aware defrag */ |
6f519564 LB |
3145 | if (new) { |
3146 | if (ret) { | |
3147 | free_sa_defrag_extent(new); | |
0b246afa | 3148 | atomic_dec(&fs_info->defrag_running); |
6f519564 LB |
3149 | } else { |
3150 | relink_file_extents(new); | |
3151 | } | |
3152 | } | |
38c227d8 | 3153 | |
e6dcd2dc CM |
3154 | /* once for us */ |
3155 | btrfs_put_ordered_extent(ordered_extent); | |
3156 | /* once for the tree */ | |
3157 | btrfs_put_ordered_extent(ordered_extent); | |
3158 | ||
5fd02043 JB |
3159 | return ret; |
3160 | } | |
3161 | ||
3162 | static void finish_ordered_fn(struct btrfs_work *work) | |
3163 | { | |
3164 | struct btrfs_ordered_extent *ordered_extent; | |
3165 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3166 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3167 | } |
3168 | ||
c3988d63 | 3169 | static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
3170 | struct extent_state *state, int uptodate) |
3171 | { | |
5fd02043 | 3172 | struct inode *inode = page->mapping->host; |
0b246afa | 3173 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5fd02043 | 3174 | struct btrfs_ordered_extent *ordered_extent = NULL; |
9e0af237 LB |
3175 | struct btrfs_workqueue *wq; |
3176 | btrfs_work_func_t func; | |
5fd02043 | 3177 | |
1abe9b8a | 3178 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
3179 | ||
8b62b72b | 3180 | ClearPagePrivate2(page); |
5fd02043 JB |
3181 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
3182 | end - start + 1, uptodate)) | |
c3988d63 | 3183 | return; |
5fd02043 | 3184 | |
70ddc553 | 3185 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
0b246afa | 3186 | wq = fs_info->endio_freespace_worker; |
9e0af237 LB |
3187 | func = btrfs_freespace_write_helper; |
3188 | } else { | |
0b246afa | 3189 | wq = fs_info->endio_write_workers; |
9e0af237 LB |
3190 | func = btrfs_endio_write_helper; |
3191 | } | |
5fd02043 | 3192 | |
9e0af237 LB |
3193 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3194 | NULL); | |
3195 | btrfs_queue_work(wq, &ordered_extent->work); | |
211f90e6 CM |
3196 | } |
3197 | ||
dc380aea MX |
3198 | static int __readpage_endio_check(struct inode *inode, |
3199 | struct btrfs_io_bio *io_bio, | |
3200 | int icsum, struct page *page, | |
3201 | int pgoff, u64 start, size_t len) | |
3202 | { | |
3203 | char *kaddr; | |
3204 | u32 csum_expected; | |
3205 | u32 csum = ~(u32)0; | |
dc380aea MX |
3206 | |
3207 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3208 | ||
3209 | kaddr = kmap_atomic(page); | |
3210 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
0b5e3daf | 3211 | btrfs_csum_final(csum, (u8 *)&csum); |
dc380aea MX |
3212 | if (csum != csum_expected) |
3213 | goto zeroit; | |
3214 | ||
3215 | kunmap_atomic(kaddr); | |
3216 | return 0; | |
3217 | zeroit: | |
0970a22e | 3218 | btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected, |
6f6b643e | 3219 | io_bio->mirror_num); |
dc380aea MX |
3220 | memset(kaddr + pgoff, 1, len); |
3221 | flush_dcache_page(page); | |
3222 | kunmap_atomic(kaddr); | |
dc380aea MX |
3223 | return -EIO; |
3224 | } | |
3225 | ||
d352ac68 CM |
3226 | /* |
3227 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3228 | * if there's a match, we allow the bio to finish. If not, the code in |
3229 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3230 | */ |
facc8a22 MX |
3231 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3232 | u64 phy_offset, struct page *page, | |
3233 | u64 start, u64 end, int mirror) | |
07157aac | 3234 | { |
4eee4fa4 | 3235 | size_t offset = start - page_offset(page); |
07157aac | 3236 | struct inode *inode = page->mapping->host; |
d1310b2e | 3237 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3238 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3239 | |
d20f7043 CM |
3240 | if (PageChecked(page)) { |
3241 | ClearPageChecked(page); | |
dc380aea | 3242 | return 0; |
d20f7043 | 3243 | } |
6cbff00f CH |
3244 | |
3245 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3246 | return 0; |
17d217fe YZ |
3247 | |
3248 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3249 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3250 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3251 | return 0; |
17d217fe | 3252 | } |
d20f7043 | 3253 | |
facc8a22 | 3254 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3255 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3256 | start, (size_t)(end - start + 1)); | |
07157aac | 3257 | } |
b888db2b | 3258 | |
c1c3fac2 NB |
3259 | /* |
3260 | * btrfs_add_delayed_iput - perform a delayed iput on @inode | |
3261 | * | |
3262 | * @inode: The inode we want to perform iput on | |
3263 | * | |
3264 | * This function uses the generic vfs_inode::i_count to track whether we should | |
3265 | * just decrement it (in case it's > 1) or if this is the last iput then link | |
3266 | * the inode to the delayed iput machinery. Delayed iputs are processed at | |
3267 | * transaction commit time/superblock commit/cleaner kthread. | |
3268 | */ | |
24bbcf04 YZ |
3269 | void btrfs_add_delayed_iput(struct inode *inode) |
3270 | { | |
0b246afa | 3271 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3272 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3273 | |
3274 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3275 | return; | |
3276 | ||
24bbcf04 | 3277 | spin_lock(&fs_info->delayed_iput_lock); |
c1c3fac2 NB |
3278 | ASSERT(list_empty(&binode->delayed_iput)); |
3279 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
24bbcf04 YZ |
3280 | spin_unlock(&fs_info->delayed_iput_lock); |
3281 | } | |
3282 | ||
2ff7e61e | 3283 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3284 | { |
24bbcf04 | 3285 | |
24bbcf04 | 3286 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3287 | while (!list_empty(&fs_info->delayed_iputs)) { |
3288 | struct btrfs_inode *inode; | |
3289 | ||
3290 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3291 | struct btrfs_inode, delayed_iput); | |
c1c3fac2 | 3292 | list_del_init(&inode->delayed_iput); |
8089fe62 DS |
3293 | spin_unlock(&fs_info->delayed_iput_lock); |
3294 | iput(&inode->vfs_inode); | |
3295 | spin_lock(&fs_info->delayed_iput_lock); | |
24bbcf04 | 3296 | } |
8089fe62 | 3297 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3298 | } |
3299 | ||
d68fc57b | 3300 | /* |
42b2aa86 | 3301 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3302 | * files in the subvolume, it removes orphan item and frees block_rsv |
3303 | * structure. | |
3304 | */ | |
3305 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3306 | struct btrfs_root *root) | |
3307 | { | |
0b246afa | 3308 | struct btrfs_fs_info *fs_info = root->fs_info; |
90290e19 | 3309 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3310 | int ret; |
3311 | ||
8a35d95f | 3312 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3313 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3314 | return; | |
3315 | ||
90290e19 | 3316 | spin_lock(&root->orphan_lock); |
8a35d95f | 3317 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3318 | spin_unlock(&root->orphan_lock); |
3319 | return; | |
3320 | } | |
3321 | ||
3322 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3323 | spin_unlock(&root->orphan_lock); | |
3324 | return; | |
3325 | } | |
3326 | ||
3327 | block_rsv = root->orphan_block_rsv; | |
3328 | root->orphan_block_rsv = NULL; | |
3329 | spin_unlock(&root->orphan_lock); | |
3330 | ||
27cdeb70 | 3331 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b | 3332 | btrfs_root_refs(&root->root_item) > 0) { |
0b246afa | 3333 | ret = btrfs_del_orphan_item(trans, fs_info->tree_root, |
d68fc57b | 3334 | root->root_key.objectid); |
4ef31a45 | 3335 | if (ret) |
66642832 | 3336 | btrfs_abort_transaction(trans, ret); |
4ef31a45 | 3337 | else |
27cdeb70 MX |
3338 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3339 | &root->state); | |
d68fc57b YZ |
3340 | } |
3341 | ||
90290e19 JB |
3342 | if (block_rsv) { |
3343 | WARN_ON(block_rsv->size > 0); | |
2ff7e61e | 3344 | btrfs_free_block_rsv(fs_info, block_rsv); |
d68fc57b YZ |
3345 | } |
3346 | } | |
3347 | ||
7b128766 | 3348 | /* |
f7e9e8fc OS |
3349 | * This creates an orphan entry for the given inode in case something goes wrong |
3350 | * in the middle of an unlink. | |
d68fc57b YZ |
3351 | * |
3352 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3353 | * this function. | |
7b128766 | 3354 | */ |
73f2e545 NB |
3355 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
3356 | struct btrfs_inode *inode) | |
7b128766 | 3357 | { |
73f2e545 NB |
3358 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
3359 | struct btrfs_root *root = inode->root; | |
d68fc57b YZ |
3360 | struct btrfs_block_rsv *block_rsv = NULL; |
3361 | int reserve = 0; | |
0a0d4415 | 3362 | bool insert = false; |
d68fc57b | 3363 | int ret; |
7b128766 | 3364 | |
d68fc57b | 3365 | if (!root->orphan_block_rsv) { |
2ff7e61e JM |
3366 | block_rsv = btrfs_alloc_block_rsv(fs_info, |
3367 | BTRFS_BLOCK_RSV_TEMP); | |
b532402e TI |
3368 | if (!block_rsv) |
3369 | return -ENOMEM; | |
d68fc57b | 3370 | } |
7b128766 | 3371 | |
8a35d95f | 3372 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
0a0d4415 OS |
3373 | &inode->runtime_flags)) |
3374 | insert = true; | |
7b128766 | 3375 | |
72ac3c0d | 3376 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
73f2e545 | 3377 | &inode->runtime_flags)) |
d68fc57b | 3378 | reserve = 1; |
3d5addaf LB |
3379 | |
3380 | spin_lock(&root->orphan_lock); | |
3381 | /* If someone has created ->orphan_block_rsv, be happy to use it. */ | |
3382 | if (!root->orphan_block_rsv) { | |
3383 | root->orphan_block_rsv = block_rsv; | |
3384 | } else if (block_rsv) { | |
3385 | btrfs_free_block_rsv(fs_info, block_rsv); | |
3386 | block_rsv = NULL; | |
3387 | } | |
3388 | ||
3389 | if (insert) | |
3390 | atomic_inc(&root->orphan_inodes); | |
d68fc57b | 3391 | spin_unlock(&root->orphan_lock); |
7b128766 | 3392 | |
d68fc57b YZ |
3393 | /* grab metadata reservation from transaction handle */ |
3394 | if (reserve) { | |
3395 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
3b6571c1 JB |
3396 | ASSERT(!ret); |
3397 | if (ret) { | |
1a932ef4 LB |
3398 | /* |
3399 | * dec doesn't need spin_lock as ->orphan_block_rsv | |
3400 | * would be released only if ->orphan_inodes is | |
3401 | * zero. | |
3402 | */ | |
3b6571c1 JB |
3403 | atomic_dec(&root->orphan_inodes); |
3404 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
73f2e545 | 3405 | &inode->runtime_flags); |
3b6571c1 JB |
3406 | if (insert) |
3407 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, | |
73f2e545 | 3408 | &inode->runtime_flags); |
3b6571c1 JB |
3409 | return ret; |
3410 | } | |
d68fc57b | 3411 | } |
7b128766 | 3412 | |
f7e9e8fc | 3413 | /* insert an orphan item to track this unlinked file */ |
0a0d4415 | 3414 | if (insert) { |
73f2e545 | 3415 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3416 | if (ret) { |
4ef31a45 JB |
3417 | if (reserve) { |
3418 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
73f2e545 | 3419 | &inode->runtime_flags); |
4ef31a45 JB |
3420 | btrfs_orphan_release_metadata(inode); |
3421 | } | |
1a932ef4 LB |
3422 | /* |
3423 | * btrfs_orphan_commit_root may race with us and set | |
3424 | * ->orphan_block_rsv to zero, in order to avoid that, | |
3425 | * decrease ->orphan_inodes after everything is done. | |
3426 | */ | |
3427 | atomic_dec(&root->orphan_inodes); | |
4ef31a45 | 3428 | if (ret != -EEXIST) { |
e8e7cff6 | 3429 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
73f2e545 | 3430 | &inode->runtime_flags); |
66642832 | 3431 | btrfs_abort_transaction(trans, ret); |
4ef31a45 JB |
3432 | return ret; |
3433 | } | |
79787eaa JM |
3434 | } |
3435 | ret = 0; | |
d68fc57b YZ |
3436 | } |
3437 | ||
d68fc57b | 3438 | return 0; |
7b128766 JB |
3439 | } |
3440 | ||
3441 | /* | |
f7e9e8fc OS |
3442 | * We have done the delete so we can go ahead and remove the orphan item for |
3443 | * this particular inode. | |
7b128766 | 3444 | */ |
48a3b636 | 3445 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3446 | struct btrfs_inode *inode) |
7b128766 | 3447 | { |
3d6ae7bb | 3448 | struct btrfs_root *root = inode->root; |
d68fc57b | 3449 | int delete_item = 0; |
7b128766 JB |
3450 | int ret = 0; |
3451 | ||
8a35d95f | 3452 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3d6ae7bb | 3453 | &inode->runtime_flags)) |
d68fc57b | 3454 | delete_item = 1; |
7b128766 | 3455 | |
1a932ef4 LB |
3456 | if (delete_item && trans) |
3457 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); | |
3458 | ||
72ac3c0d | 3459 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3d6ae7bb | 3460 | &inode->runtime_flags)) |
1a932ef4 | 3461 | btrfs_orphan_release_metadata(inode); |
7b128766 | 3462 | |
1a932ef4 LB |
3463 | /* |
3464 | * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv | |
3465 | * to zero, in order to avoid that, decrease ->orphan_inodes after | |
3466 | * everything is done. | |
3467 | */ | |
3468 | if (delete_item) | |
8a35d95f | 3469 | atomic_dec(&root->orphan_inodes); |
703c88e0 | 3470 | |
4ef31a45 | 3471 | return ret; |
7b128766 JB |
3472 | } |
3473 | ||
3474 | /* | |
3475 | * this cleans up any orphans that may be left on the list from the last use | |
3476 | * of this root. | |
3477 | */ | |
66b4ffd1 | 3478 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3479 | { |
0b246afa | 3480 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3481 | struct btrfs_path *path; |
3482 | struct extent_buffer *leaf; | |
7b128766 JB |
3483 | struct btrfs_key key, found_key; |
3484 | struct btrfs_trans_handle *trans; | |
3485 | struct inode *inode; | |
8f6d7f4f | 3486 | u64 last_objectid = 0; |
f7e9e8fc | 3487 | int ret = 0, nr_unlink = 0; |
7b128766 | 3488 | |
d68fc57b | 3489 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3490 | return 0; |
c71bf099 YZ |
3491 | |
3492 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3493 | if (!path) { |
3494 | ret = -ENOMEM; | |
3495 | goto out; | |
3496 | } | |
e4058b54 | 3497 | path->reada = READA_BACK; |
7b128766 JB |
3498 | |
3499 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3500 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3501 | key.offset = (u64)-1; |
3502 | ||
7b128766 JB |
3503 | while (1) { |
3504 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3505 | if (ret < 0) |
3506 | goto out; | |
7b128766 JB |
3507 | |
3508 | /* | |
3509 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3510 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3511 | * find the key and see if we have stuff that matches |
3512 | */ | |
3513 | if (ret > 0) { | |
66b4ffd1 | 3514 | ret = 0; |
7b128766 JB |
3515 | if (path->slots[0] == 0) |
3516 | break; | |
3517 | path->slots[0]--; | |
3518 | } | |
3519 | ||
3520 | /* pull out the item */ | |
3521 | leaf = path->nodes[0]; | |
7b128766 JB |
3522 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3523 | ||
3524 | /* make sure the item matches what we want */ | |
3525 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3526 | break; | |
962a298f | 3527 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3528 | break; |
3529 | ||
3530 | /* release the path since we're done with it */ | |
b3b4aa74 | 3531 | btrfs_release_path(path); |
7b128766 JB |
3532 | |
3533 | /* | |
3534 | * this is where we are basically btrfs_lookup, without the | |
3535 | * crossing root thing. we store the inode number in the | |
3536 | * offset of the orphan item. | |
3537 | */ | |
8f6d7f4f JB |
3538 | |
3539 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3540 | btrfs_err(fs_info, |
3541 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3542 | ret = -EINVAL; |
3543 | goto out; | |
3544 | } | |
3545 | ||
3546 | last_objectid = found_key.offset; | |
3547 | ||
5d4f98a2 YZ |
3548 | found_key.objectid = found_key.offset; |
3549 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3550 | found_key.offset = 0; | |
0b246afa | 3551 | inode = btrfs_iget(fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3552 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3553 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3554 | goto out; |
7b128766 | 3555 | |
0b246afa | 3556 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 AJ |
3557 | struct btrfs_root *dead_root; |
3558 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3559 | int is_dead_root = 0; | |
3560 | ||
3561 | /* | |
3562 | * this is an orphan in the tree root. Currently these | |
3563 | * could come from 2 sources: | |
3564 | * a) a snapshot deletion in progress | |
3565 | * b) a free space cache inode | |
3566 | * We need to distinguish those two, as the snapshot | |
3567 | * orphan must not get deleted. | |
3568 | * find_dead_roots already ran before us, so if this | |
3569 | * is a snapshot deletion, we should find the root | |
3570 | * in the dead_roots list | |
3571 | */ | |
3572 | spin_lock(&fs_info->trans_lock); | |
3573 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3574 | root_list) { | |
3575 | if (dead_root->root_key.objectid == | |
3576 | found_key.objectid) { | |
3577 | is_dead_root = 1; | |
3578 | break; | |
3579 | } | |
3580 | } | |
3581 | spin_unlock(&fs_info->trans_lock); | |
3582 | if (is_dead_root) { | |
3583 | /* prevent this orphan from being found again */ | |
3584 | key.offset = found_key.objectid - 1; | |
3585 | continue; | |
3586 | } | |
f7e9e8fc | 3587 | |
f8e9e0b0 | 3588 | } |
f7e9e8fc | 3589 | |
7b128766 | 3590 | /* |
f7e9e8fc OS |
3591 | * If we have an inode with links, there are a couple of |
3592 | * possibilities. Old kernels (before v3.12) used to create an | |
3593 | * orphan item for truncate indicating that there were possibly | |
3594 | * extent items past i_size that needed to be deleted. In v3.12, | |
3595 | * truncate was changed to update i_size in sync with the extent | |
3596 | * items, but the (useless) orphan item was still created. Since | |
3597 | * v4.18, we don't create the orphan item for truncate at all. | |
3598 | * | |
3599 | * So, this item could mean that we need to do a truncate, but | |
3600 | * only if this filesystem was last used on a pre-v3.12 kernel | |
3601 | * and was not cleanly unmounted. The odds of that are quite | |
3602 | * slim, and it's a pain to do the truncate now, so just delete | |
3603 | * the orphan item. | |
3604 | * | |
3605 | * It's also possible that this orphan item was supposed to be | |
3606 | * deleted but wasn't. The inode number may have been reused, | |
3607 | * but either way, we can delete the orphan item. | |
7b128766 | 3608 | */ |
f7e9e8fc OS |
3609 | if (ret == -ENOENT || inode->i_nlink) { |
3610 | if (!ret) | |
3611 | iput(inode); | |
a8c9e576 | 3612 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3613 | if (IS_ERR(trans)) { |
3614 | ret = PTR_ERR(trans); | |
3615 | goto out; | |
3616 | } | |
0b246afa JM |
3617 | btrfs_debug(fs_info, "auto deleting %Lu", |
3618 | found_key.objectid); | |
a8c9e576 JB |
3619 | ret = btrfs_del_orphan_item(trans, root, |
3620 | found_key.objectid); | |
3a45bb20 | 3621 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3622 | if (ret) |
3623 | goto out; | |
7b128766 JB |
3624 | continue; |
3625 | } | |
3626 | ||
a8c9e576 JB |
3627 | /* |
3628 | * add this inode to the orphan list so btrfs_orphan_del does | |
3629 | * the proper thing when we hit it | |
3630 | */ | |
8a35d95f JB |
3631 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3632 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3633 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3634 | |
f7e9e8fc | 3635 | nr_unlink++; |
7b128766 JB |
3636 | |
3637 | /* this will do delete_inode and everything for us */ | |
3638 | iput(inode); | |
66b4ffd1 JB |
3639 | if (ret) |
3640 | goto out; | |
7b128766 | 3641 | } |
3254c876 MX |
3642 | /* release the path since we're done with it */ |
3643 | btrfs_release_path(path); | |
3644 | ||
d68fc57b YZ |
3645 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3646 | ||
3647 | if (root->orphan_block_rsv) | |
2ff7e61e | 3648 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, |
d68fc57b YZ |
3649 | (u64)-1); |
3650 | ||
27cdeb70 MX |
3651 | if (root->orphan_block_rsv || |
3652 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3653 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3654 | if (!IS_ERR(trans)) |
3a45bb20 | 3655 | btrfs_end_transaction(trans); |
d68fc57b | 3656 | } |
7b128766 JB |
3657 | |
3658 | if (nr_unlink) | |
0b246afa | 3659 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
66b4ffd1 JB |
3660 | |
3661 | out: | |
3662 | if (ret) | |
0b246afa | 3663 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3664 | btrfs_free_path(path); |
3665 | return ret; | |
7b128766 JB |
3666 | } |
3667 | ||
46a53cca CM |
3668 | /* |
3669 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3670 | * don't find any xattrs, we know there can't be any acls. | |
3671 | * | |
3672 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3673 | */ | |
3674 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3675 | int slot, u64 objectid, |
3676 | int *first_xattr_slot) | |
46a53cca CM |
3677 | { |
3678 | u32 nritems = btrfs_header_nritems(leaf); | |
3679 | struct btrfs_key found_key; | |
f23b5a59 JB |
3680 | static u64 xattr_access = 0; |
3681 | static u64 xattr_default = 0; | |
46a53cca CM |
3682 | int scanned = 0; |
3683 | ||
f23b5a59 | 3684 | if (!xattr_access) { |
97d79299 AG |
3685 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3686 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3687 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3688 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3689 | } |
3690 | ||
46a53cca | 3691 | slot++; |
63541927 | 3692 | *first_xattr_slot = -1; |
46a53cca CM |
3693 | while (slot < nritems) { |
3694 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3695 | ||
3696 | /* we found a different objectid, there must not be acls */ | |
3697 | if (found_key.objectid != objectid) | |
3698 | return 0; | |
3699 | ||
3700 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3701 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3702 | if (*first_xattr_slot == -1) |
3703 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3704 | if (found_key.offset == xattr_access || |
3705 | found_key.offset == xattr_default) | |
3706 | return 1; | |
3707 | } | |
46a53cca CM |
3708 | |
3709 | /* | |
3710 | * we found a key greater than an xattr key, there can't | |
3711 | * be any acls later on | |
3712 | */ | |
3713 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3714 | return 0; | |
3715 | ||
3716 | slot++; | |
3717 | scanned++; | |
3718 | ||
3719 | /* | |
3720 | * it goes inode, inode backrefs, xattrs, extents, | |
3721 | * so if there are a ton of hard links to an inode there can | |
3722 | * be a lot of backrefs. Don't waste time searching too hard, | |
3723 | * this is just an optimization | |
3724 | */ | |
3725 | if (scanned >= 8) | |
3726 | break; | |
3727 | } | |
3728 | /* we hit the end of the leaf before we found an xattr or | |
3729 | * something larger than an xattr. We have to assume the inode | |
3730 | * has acls | |
3731 | */ | |
63541927 FDBM |
3732 | if (*first_xattr_slot == -1) |
3733 | *first_xattr_slot = slot; | |
46a53cca CM |
3734 | return 1; |
3735 | } | |
3736 | ||
d352ac68 CM |
3737 | /* |
3738 | * read an inode from the btree into the in-memory inode | |
3739 | */ | |
67710892 | 3740 | static int btrfs_read_locked_inode(struct inode *inode) |
39279cc3 | 3741 | { |
0b246afa | 3742 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 3743 | struct btrfs_path *path; |
5f39d397 | 3744 | struct extent_buffer *leaf; |
39279cc3 CM |
3745 | struct btrfs_inode_item *inode_item; |
3746 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3747 | struct btrfs_key location; | |
67de1176 | 3748 | unsigned long ptr; |
46a53cca | 3749 | int maybe_acls; |
618e21d5 | 3750 | u32 rdev; |
39279cc3 | 3751 | int ret; |
2f7e33d4 | 3752 | bool filled = false; |
63541927 | 3753 | int first_xattr_slot; |
2f7e33d4 MX |
3754 | |
3755 | ret = btrfs_fill_inode(inode, &rdev); | |
3756 | if (!ret) | |
3757 | filled = true; | |
39279cc3 CM |
3758 | |
3759 | path = btrfs_alloc_path(); | |
67710892 FM |
3760 | if (!path) { |
3761 | ret = -ENOMEM; | |
1748f843 | 3762 | goto make_bad; |
67710892 | 3763 | } |
1748f843 | 3764 | |
39279cc3 | 3765 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3766 | |
39279cc3 | 3767 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 FM |
3768 | if (ret) { |
3769 | if (ret > 0) | |
3770 | ret = -ENOENT; | |
39279cc3 | 3771 | goto make_bad; |
67710892 | 3772 | } |
39279cc3 | 3773 | |
5f39d397 | 3774 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3775 | |
3776 | if (filled) | |
67de1176 | 3777 | goto cache_index; |
2f7e33d4 | 3778 | |
5f39d397 CM |
3779 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3780 | struct btrfs_inode_item); | |
5f39d397 | 3781 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3782 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3783 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3784 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3785 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3786 | |
a937b979 DS |
3787 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3788 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3789 | |
a937b979 DS |
3790 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3791 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3792 | |
a937b979 DS |
3793 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3794 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3795 | |
9cc97d64 | 3796 | BTRFS_I(inode)->i_otime.tv_sec = |
3797 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3798 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3799 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3800 | |
a76a3cd4 | 3801 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3802 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3803 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3804 | ||
c7f88c4e JL |
3805 | inode_set_iversion_queried(inode, |
3806 | btrfs_inode_sequence(leaf, inode_item)); | |
6e17d30b YD |
3807 | inode->i_generation = BTRFS_I(inode)->generation; |
3808 | inode->i_rdev = 0; | |
3809 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3810 | ||
3811 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3812 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3813 | ||
3814 | cache_index: | |
5dc562c5 JB |
3815 | /* |
3816 | * If we were modified in the current generation and evicted from memory | |
3817 | * and then re-read we need to do a full sync since we don't have any | |
3818 | * idea about which extents were modified before we were evicted from | |
3819 | * cache. | |
6e17d30b YD |
3820 | * |
3821 | * This is required for both inode re-read from disk and delayed inode | |
3822 | * in delayed_nodes_tree. | |
5dc562c5 | 3823 | */ |
0b246afa | 3824 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3825 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3826 | &BTRFS_I(inode)->runtime_flags); | |
3827 | ||
bde6c242 FM |
3828 | /* |
3829 | * We don't persist the id of the transaction where an unlink operation | |
3830 | * against the inode was last made. So here we assume the inode might | |
3831 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3832 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3833 | * between the inode and its parent if the inode is fsync'ed and the log | |
3834 | * replayed. For example, in the scenario: | |
3835 | * | |
3836 | * touch mydir/foo | |
3837 | * ln mydir/foo mydir/bar | |
3838 | * sync | |
3839 | * unlink mydir/bar | |
3840 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3841 | * xfs_io -c fsync mydir/foo | |
3842 | * <power failure> | |
3843 | * mount fs, triggers fsync log replay | |
3844 | * | |
3845 | * We must make sure that when we fsync our inode foo we also log its | |
3846 | * parent inode, otherwise after log replay the parent still has the | |
3847 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3848 | * and doesn't have an inode ref with the name "bar" anymore. | |
3849 | * | |
3850 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3851 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3852 | * transaction commits on fsync if our inode is a directory, or if our |
3853 | * inode is not a directory, logging its parent unnecessarily. | |
3854 | */ | |
3855 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3856 | ||
67de1176 MX |
3857 | path->slots[0]++; |
3858 | if (inode->i_nlink != 1 || | |
3859 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3860 | goto cache_acl; | |
3861 | ||
3862 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3863 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3864 | goto cache_acl; |
3865 | ||
3866 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3867 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3868 | struct btrfs_inode_ref *ref; | |
3869 | ||
3870 | ref = (struct btrfs_inode_ref *)ptr; | |
3871 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3872 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3873 | struct btrfs_inode_extref *extref; | |
3874 | ||
3875 | extref = (struct btrfs_inode_extref *)ptr; | |
3876 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3877 | extref); | |
3878 | } | |
2f7e33d4 | 3879 | cache_acl: |
46a53cca CM |
3880 | /* |
3881 | * try to precache a NULL acl entry for files that don't have | |
3882 | * any xattrs or acls | |
3883 | */ | |
33345d01 | 3884 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3885 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3886 | if (first_xattr_slot != -1) { |
3887 | path->slots[0] = first_xattr_slot; | |
3888 | ret = btrfs_load_inode_props(inode, path); | |
3889 | if (ret) | |
0b246afa | 3890 | btrfs_err(fs_info, |
351fd353 | 3891 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3892 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3893 | root->root_key.objectid, ret); |
3894 | } | |
3895 | btrfs_free_path(path); | |
3896 | ||
72c04902 AV |
3897 | if (!maybe_acls) |
3898 | cache_no_acl(inode); | |
46a53cca | 3899 | |
39279cc3 | 3900 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3901 | case S_IFREG: |
3902 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3903 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3904 | inode->i_fop = &btrfs_file_operations; |
3905 | inode->i_op = &btrfs_file_inode_operations; | |
3906 | break; | |
3907 | case S_IFDIR: | |
3908 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3909 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3910 | break; |
3911 | case S_IFLNK: | |
3912 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3913 | inode_nohighmem(inode); |
39279cc3 CM |
3914 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
3915 | break; | |
618e21d5 | 3916 | default: |
0279b4cd | 3917 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3918 | init_special_inode(inode, inode->i_mode, rdev); |
3919 | break; | |
39279cc3 | 3920 | } |
6cbff00f | 3921 | |
7b6a221e | 3922 | btrfs_sync_inode_flags_to_i_flags(inode); |
67710892 | 3923 | return 0; |
39279cc3 CM |
3924 | |
3925 | make_bad: | |
39279cc3 | 3926 | btrfs_free_path(path); |
39279cc3 | 3927 | make_bad_inode(inode); |
67710892 | 3928 | return ret; |
39279cc3 CM |
3929 | } |
3930 | ||
d352ac68 CM |
3931 | /* |
3932 | * given a leaf and an inode, copy the inode fields into the leaf | |
3933 | */ | |
e02119d5 CM |
3934 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3935 | struct extent_buffer *leaf, | |
5f39d397 | 3936 | struct btrfs_inode_item *item, |
39279cc3 CM |
3937 | struct inode *inode) |
3938 | { | |
51fab693 LB |
3939 | struct btrfs_map_token token; |
3940 | ||
3941 | btrfs_init_map_token(&token); | |
5f39d397 | 3942 | |
51fab693 LB |
3943 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3944 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3945 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3946 | &token); | |
3947 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3948 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3949 | |
a937b979 | 3950 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3951 | inode->i_atime.tv_sec, &token); |
a937b979 | 3952 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3953 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3954 | |
a937b979 | 3955 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3956 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3957 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3958 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3959 | |
a937b979 | 3960 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3961 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3962 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3963 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3964 | |
9cc97d64 | 3965 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3966 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3967 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3968 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3969 | ||
51fab693 LB |
3970 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3971 | &token); | |
3972 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3973 | &token); | |
c7f88c4e JL |
3974 | btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode), |
3975 | &token); | |
51fab693 LB |
3976 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); |
3977 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3978 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3979 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3980 | } |
3981 | ||
d352ac68 CM |
3982 | /* |
3983 | * copy everything in the in-memory inode into the btree. | |
3984 | */ | |
2115133f | 3985 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3986 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3987 | { |
3988 | struct btrfs_inode_item *inode_item; | |
3989 | struct btrfs_path *path; | |
5f39d397 | 3990 | struct extent_buffer *leaf; |
39279cc3 CM |
3991 | int ret; |
3992 | ||
3993 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3994 | if (!path) |
3995 | return -ENOMEM; | |
3996 | ||
b9473439 | 3997 | path->leave_spinning = 1; |
16cdcec7 MX |
3998 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3999 | 1); | |
39279cc3 CM |
4000 | if (ret) { |
4001 | if (ret > 0) | |
4002 | ret = -ENOENT; | |
4003 | goto failed; | |
4004 | } | |
4005 | ||
5f39d397 CM |
4006 | leaf = path->nodes[0]; |
4007 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 4008 | struct btrfs_inode_item); |
39279cc3 | 4009 | |
e02119d5 | 4010 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 4011 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 4012 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
4013 | ret = 0; |
4014 | failed: | |
39279cc3 CM |
4015 | btrfs_free_path(path); |
4016 | return ret; | |
4017 | } | |
4018 | ||
2115133f CM |
4019 | /* |
4020 | * copy everything in the in-memory inode into the btree. | |
4021 | */ | |
4022 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
4023 | struct btrfs_root *root, struct inode *inode) | |
4024 | { | |
0b246afa | 4025 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
4026 | int ret; |
4027 | ||
4028 | /* | |
4029 | * If the inode is a free space inode, we can deadlock during commit | |
4030 | * if we put it into the delayed code. | |
4031 | * | |
4032 | * The data relocation inode should also be directly updated | |
4033 | * without delay | |
4034 | */ | |
70ddc553 | 4035 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) |
1d52c78a | 4036 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 4037 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
4038 | btrfs_update_root_times(trans, root); |
4039 | ||
2115133f CM |
4040 | ret = btrfs_delayed_update_inode(trans, root, inode); |
4041 | if (!ret) | |
4042 | btrfs_set_inode_last_trans(trans, inode); | |
4043 | return ret; | |
4044 | } | |
4045 | ||
4046 | return btrfs_update_inode_item(trans, root, inode); | |
4047 | } | |
4048 | ||
be6aef60 JB |
4049 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
4050 | struct btrfs_root *root, | |
4051 | struct inode *inode) | |
2115133f CM |
4052 | { |
4053 | int ret; | |
4054 | ||
4055 | ret = btrfs_update_inode(trans, root, inode); | |
4056 | if (ret == -ENOSPC) | |
4057 | return btrfs_update_inode_item(trans, root, inode); | |
4058 | return ret; | |
4059 | } | |
4060 | ||
d352ac68 CM |
4061 | /* |
4062 | * unlink helper that gets used here in inode.c and in the tree logging | |
4063 | * recovery code. It remove a link in a directory with a given name, and | |
4064 | * also drops the back refs in the inode to the directory | |
4065 | */ | |
92986796 AV |
4066 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4067 | struct btrfs_root *root, | |
4ec5934e NB |
4068 | struct btrfs_inode *dir, |
4069 | struct btrfs_inode *inode, | |
92986796 | 4070 | const char *name, int name_len) |
39279cc3 | 4071 | { |
0b246afa | 4072 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4073 | struct btrfs_path *path; |
39279cc3 | 4074 | int ret = 0; |
5f39d397 | 4075 | struct extent_buffer *leaf; |
39279cc3 | 4076 | struct btrfs_dir_item *di; |
5f39d397 | 4077 | struct btrfs_key key; |
aec7477b | 4078 | u64 index; |
33345d01 LZ |
4079 | u64 ino = btrfs_ino(inode); |
4080 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
4081 | |
4082 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
4083 | if (!path) { |
4084 | ret = -ENOMEM; | |
554233a6 | 4085 | goto out; |
54aa1f4d CM |
4086 | } |
4087 | ||
b9473439 | 4088 | path->leave_spinning = 1; |
33345d01 | 4089 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
4090 | name, name_len, -1); |
4091 | if (IS_ERR(di)) { | |
4092 | ret = PTR_ERR(di); | |
4093 | goto err; | |
4094 | } | |
4095 | if (!di) { | |
4096 | ret = -ENOENT; | |
4097 | goto err; | |
4098 | } | |
5f39d397 CM |
4099 | leaf = path->nodes[0]; |
4100 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 4101 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
4102 | if (ret) |
4103 | goto err; | |
b3b4aa74 | 4104 | btrfs_release_path(path); |
39279cc3 | 4105 | |
67de1176 MX |
4106 | /* |
4107 | * If we don't have dir index, we have to get it by looking up | |
4108 | * the inode ref, since we get the inode ref, remove it directly, | |
4109 | * it is unnecessary to do delayed deletion. | |
4110 | * | |
4111 | * But if we have dir index, needn't search inode ref to get it. | |
4112 | * Since the inode ref is close to the inode item, it is better | |
4113 | * that we delay to delete it, and just do this deletion when | |
4114 | * we update the inode item. | |
4115 | */ | |
4ec5934e | 4116 | if (inode->dir_index) { |
67de1176 MX |
4117 | ret = btrfs_delayed_delete_inode_ref(inode); |
4118 | if (!ret) { | |
4ec5934e | 4119 | index = inode->dir_index; |
67de1176 MX |
4120 | goto skip_backref; |
4121 | } | |
4122 | } | |
4123 | ||
33345d01 LZ |
4124 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
4125 | dir_ino, &index); | |
aec7477b | 4126 | if (ret) { |
0b246afa | 4127 | btrfs_info(fs_info, |
c2cf52eb | 4128 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 4129 | name_len, name, ino, dir_ino); |
66642832 | 4130 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
4131 | goto err; |
4132 | } | |
67de1176 | 4133 | skip_backref: |
2ff7e61e | 4134 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index); |
79787eaa | 4135 | if (ret) { |
66642832 | 4136 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 4137 | goto err; |
79787eaa | 4138 | } |
39279cc3 | 4139 | |
4ec5934e NB |
4140 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
4141 | dir_ino); | |
79787eaa | 4142 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 4143 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4144 | goto err; |
4145 | } | |
e02119d5 | 4146 | |
4ec5934e NB |
4147 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
4148 | index); | |
6418c961 CM |
4149 | if (ret == -ENOENT) |
4150 | ret = 0; | |
d4e3991b | 4151 | else if (ret) |
66642832 | 4152 | btrfs_abort_transaction(trans, ret); |
39279cc3 CM |
4153 | err: |
4154 | btrfs_free_path(path); | |
e02119d5 CM |
4155 | if (ret) |
4156 | goto out; | |
4157 | ||
6ef06d27 | 4158 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
4159 | inode_inc_iversion(&inode->vfs_inode); |
4160 | inode_inc_iversion(&dir->vfs_inode); | |
4161 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
4162 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
4163 | ret = btrfs_update_inode(trans, root, &dir->vfs_inode); | |
e02119d5 | 4164 | out: |
39279cc3 CM |
4165 | return ret; |
4166 | } | |
4167 | ||
92986796 AV |
4168 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4169 | struct btrfs_root *root, | |
4ec5934e | 4170 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
4171 | const char *name, int name_len) |
4172 | { | |
4173 | int ret; | |
4174 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4175 | if (!ret) { | |
4ec5934e NB |
4176 | drop_nlink(&inode->vfs_inode); |
4177 | ret = btrfs_update_inode(trans, root, &inode->vfs_inode); | |
92986796 AV |
4178 | } |
4179 | return ret; | |
4180 | } | |
39279cc3 | 4181 | |
a22285a6 YZ |
4182 | /* |
4183 | * helper to start transaction for unlink and rmdir. | |
4184 | * | |
d52be818 JB |
4185 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4186 | * if we cannot make our reservations the normal way try and see if there is | |
4187 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4188 | * allow the unlink to occur. | |
a22285a6 | 4189 | */ |
d52be818 | 4190 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4191 | { |
a22285a6 | 4192 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4193 | |
e70bea5f JB |
4194 | /* |
4195 | * 1 for the possible orphan item | |
4196 | * 1 for the dir item | |
4197 | * 1 for the dir index | |
4198 | * 1 for the inode ref | |
e70bea5f JB |
4199 | * 1 for the inode |
4200 | */ | |
8eab77ff | 4201 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4202 | } |
4203 | ||
4204 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4205 | { | |
4206 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4207 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4208 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4209 | int ret; |
a22285a6 | 4210 | |
d52be818 | 4211 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4212 | if (IS_ERR(trans)) |
4213 | return PTR_ERR(trans); | |
5f39d397 | 4214 | |
4ec5934e NB |
4215 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4216 | 0); | |
12fcfd22 | 4217 | |
4ec5934e NB |
4218 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4219 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4220 | dentry->d_name.len); | |
b532402e TI |
4221 | if (ret) |
4222 | goto out; | |
7b128766 | 4223 | |
a22285a6 | 4224 | if (inode->i_nlink == 0) { |
73f2e545 | 4225 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
4226 | if (ret) |
4227 | goto out; | |
a22285a6 | 4228 | } |
7b128766 | 4229 | |
b532402e | 4230 | out: |
3a45bb20 | 4231 | btrfs_end_transaction(trans); |
2ff7e61e | 4232 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
4233 | return ret; |
4234 | } | |
4235 | ||
f60a2364 | 4236 | static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
4df27c4d YZ |
4237 | struct btrfs_root *root, |
4238 | struct inode *dir, u64 objectid, | |
4239 | const char *name, int name_len) | |
4240 | { | |
0b246afa | 4241 | struct btrfs_fs_info *fs_info = root->fs_info; |
4df27c4d YZ |
4242 | struct btrfs_path *path; |
4243 | struct extent_buffer *leaf; | |
4244 | struct btrfs_dir_item *di; | |
4245 | struct btrfs_key key; | |
4246 | u64 index; | |
4247 | int ret; | |
4a0cc7ca | 4248 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d YZ |
4249 | |
4250 | path = btrfs_alloc_path(); | |
4251 | if (!path) | |
4252 | return -ENOMEM; | |
4253 | ||
33345d01 | 4254 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4255 | name, name_len, -1); |
79787eaa JM |
4256 | if (IS_ERR_OR_NULL(di)) { |
4257 | if (!di) | |
4258 | ret = -ENOENT; | |
4259 | else | |
4260 | ret = PTR_ERR(di); | |
4261 | goto out; | |
4262 | } | |
4df27c4d YZ |
4263 | |
4264 | leaf = path->nodes[0]; | |
4265 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4266 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4267 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4268 | if (ret) { |
66642832 | 4269 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4270 | goto out; |
4271 | } | |
b3b4aa74 | 4272 | btrfs_release_path(path); |
4df27c4d | 4273 | |
0b246afa JM |
4274 | ret = btrfs_del_root_ref(trans, fs_info, objectid, |
4275 | root->root_key.objectid, dir_ino, | |
4276 | &index, name, name_len); | |
4df27c4d | 4277 | if (ret < 0) { |
79787eaa | 4278 | if (ret != -ENOENT) { |
66642832 | 4279 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4280 | goto out; |
4281 | } | |
33345d01 | 4282 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4283 | name, name_len); |
79787eaa JM |
4284 | if (IS_ERR_OR_NULL(di)) { |
4285 | if (!di) | |
4286 | ret = -ENOENT; | |
4287 | else | |
4288 | ret = PTR_ERR(di); | |
66642832 | 4289 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4290 | goto out; |
4291 | } | |
4df27c4d YZ |
4292 | |
4293 | leaf = path->nodes[0]; | |
4294 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4295 | btrfs_release_path(path); |
4df27c4d YZ |
4296 | index = key.offset; |
4297 | } | |
945d8962 | 4298 | btrfs_release_path(path); |
4df27c4d | 4299 | |
e67bbbb9 | 4300 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index); |
79787eaa | 4301 | if (ret) { |
66642832 | 4302 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4303 | goto out; |
4304 | } | |
4df27c4d | 4305 | |
6ef06d27 | 4306 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 4307 | inode_inc_iversion(dir); |
c2050a45 | 4308 | dir->i_mtime = dir->i_ctime = current_time(dir); |
5a24e84c | 4309 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa | 4310 | if (ret) |
66642832 | 4311 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4312 | out: |
71d7aed0 | 4313 | btrfs_free_path(path); |
79787eaa | 4314 | return ret; |
4df27c4d YZ |
4315 | } |
4316 | ||
ec42f167 MT |
4317 | /* |
4318 | * Helper to check if the subvolume references other subvolumes or if it's | |
4319 | * default. | |
4320 | */ | |
f60a2364 | 4321 | static noinline int may_destroy_subvol(struct btrfs_root *root) |
ec42f167 MT |
4322 | { |
4323 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4324 | struct btrfs_path *path; | |
4325 | struct btrfs_dir_item *di; | |
4326 | struct btrfs_key key; | |
4327 | u64 dir_id; | |
4328 | int ret; | |
4329 | ||
4330 | path = btrfs_alloc_path(); | |
4331 | if (!path) | |
4332 | return -ENOMEM; | |
4333 | ||
4334 | /* Make sure this root isn't set as the default subvol */ | |
4335 | dir_id = btrfs_super_root_dir(fs_info->super_copy); | |
4336 | di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path, | |
4337 | dir_id, "default", 7, 0); | |
4338 | if (di && !IS_ERR(di)) { | |
4339 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
4340 | if (key.objectid == root->root_key.objectid) { | |
4341 | ret = -EPERM; | |
4342 | btrfs_err(fs_info, | |
4343 | "deleting default subvolume %llu is not allowed", | |
4344 | key.objectid); | |
4345 | goto out; | |
4346 | } | |
4347 | btrfs_release_path(path); | |
4348 | } | |
4349 | ||
4350 | key.objectid = root->root_key.objectid; | |
4351 | key.type = BTRFS_ROOT_REF_KEY; | |
4352 | key.offset = (u64)-1; | |
4353 | ||
4354 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
4355 | if (ret < 0) | |
4356 | goto out; | |
4357 | BUG_ON(ret == 0); | |
4358 | ||
4359 | ret = 0; | |
4360 | if (path->slots[0] > 0) { | |
4361 | path->slots[0]--; | |
4362 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
4363 | if (key.objectid == root->root_key.objectid && | |
4364 | key.type == BTRFS_ROOT_REF_KEY) | |
4365 | ret = -ENOTEMPTY; | |
4366 | } | |
4367 | out: | |
4368 | btrfs_free_path(path); | |
4369 | return ret; | |
4370 | } | |
4371 | ||
20a68004 NB |
4372 | /* Delete all dentries for inodes belonging to the root */ |
4373 | static void btrfs_prune_dentries(struct btrfs_root *root) | |
4374 | { | |
4375 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4376 | struct rb_node *node; | |
4377 | struct rb_node *prev; | |
4378 | struct btrfs_inode *entry; | |
4379 | struct inode *inode; | |
4380 | u64 objectid = 0; | |
4381 | ||
4382 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) | |
4383 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4384 | ||
4385 | spin_lock(&root->inode_lock); | |
4386 | again: | |
4387 | node = root->inode_tree.rb_node; | |
4388 | prev = NULL; | |
4389 | while (node) { | |
4390 | prev = node; | |
4391 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4392 | ||
4393 | if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode))) | |
4394 | node = node->rb_left; | |
4395 | else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode))) | |
4396 | node = node->rb_right; | |
4397 | else | |
4398 | break; | |
4399 | } | |
4400 | if (!node) { | |
4401 | while (prev) { | |
4402 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
4403 | if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) { | |
4404 | node = prev; | |
4405 | break; | |
4406 | } | |
4407 | prev = rb_next(prev); | |
4408 | } | |
4409 | } | |
4410 | while (node) { | |
4411 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4412 | objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1; | |
4413 | inode = igrab(&entry->vfs_inode); | |
4414 | if (inode) { | |
4415 | spin_unlock(&root->inode_lock); | |
4416 | if (atomic_read(&inode->i_count) > 1) | |
4417 | d_prune_aliases(inode); | |
4418 | /* | |
4419 | * btrfs_drop_inode will have it removed from the inode | |
4420 | * cache when its usage count hits zero. | |
4421 | */ | |
4422 | iput(inode); | |
4423 | cond_resched(); | |
4424 | spin_lock(&root->inode_lock); | |
4425 | goto again; | |
4426 | } | |
4427 | ||
4428 | if (cond_resched_lock(&root->inode_lock)) | |
4429 | goto again; | |
4430 | ||
4431 | node = rb_next(node); | |
4432 | } | |
4433 | spin_unlock(&root->inode_lock); | |
4434 | } | |
4435 | ||
f60a2364 MT |
4436 | int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry) |
4437 | { | |
4438 | struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb); | |
4439 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4440 | struct inode *inode = d_inode(dentry); | |
4441 | struct btrfs_root *dest = BTRFS_I(inode)->root; | |
4442 | struct btrfs_trans_handle *trans; | |
4443 | struct btrfs_block_rsv block_rsv; | |
4444 | u64 root_flags; | |
4445 | u64 qgroup_reserved; | |
4446 | int ret; | |
4447 | int err; | |
4448 | ||
4449 | /* | |
4450 | * Don't allow to delete a subvolume with send in progress. This is | |
4451 | * inside the inode lock so the error handling that has to drop the bit | |
4452 | * again is not run concurrently. | |
4453 | */ | |
4454 | spin_lock(&dest->root_item_lock); | |
4455 | root_flags = btrfs_root_flags(&dest->root_item); | |
4456 | if (dest->send_in_progress == 0) { | |
4457 | btrfs_set_root_flags(&dest->root_item, | |
4458 | root_flags | BTRFS_ROOT_SUBVOL_DEAD); | |
4459 | spin_unlock(&dest->root_item_lock); | |
4460 | } else { | |
4461 | spin_unlock(&dest->root_item_lock); | |
4462 | btrfs_warn(fs_info, | |
4463 | "attempt to delete subvolume %llu during send", | |
4464 | dest->root_key.objectid); | |
4465 | return -EPERM; | |
4466 | } | |
4467 | ||
4468 | down_write(&fs_info->subvol_sem); | |
4469 | ||
4470 | err = may_destroy_subvol(dest); | |
4471 | if (err) | |
4472 | goto out_up_write; | |
4473 | ||
4474 | btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); | |
4475 | /* | |
4476 | * One for dir inode, | |
4477 | * two for dir entries, | |
4478 | * two for root ref/backref. | |
4479 | */ | |
4480 | err = btrfs_subvolume_reserve_metadata(root, &block_rsv, | |
4481 | 5, &qgroup_reserved, true); | |
4482 | if (err) | |
4483 | goto out_up_write; | |
4484 | ||
4485 | trans = btrfs_start_transaction(root, 0); | |
4486 | if (IS_ERR(trans)) { | |
4487 | err = PTR_ERR(trans); | |
4488 | goto out_release; | |
4489 | } | |
4490 | trans->block_rsv = &block_rsv; | |
4491 | trans->bytes_reserved = block_rsv.size; | |
4492 | ||
4493 | btrfs_record_snapshot_destroy(trans, BTRFS_I(dir)); | |
4494 | ||
4495 | ret = btrfs_unlink_subvol(trans, root, dir, | |
4496 | dest->root_key.objectid, | |
4497 | dentry->d_name.name, | |
4498 | dentry->d_name.len); | |
4499 | if (ret) { | |
4500 | err = ret; | |
4501 | btrfs_abort_transaction(trans, ret); | |
4502 | goto out_end_trans; | |
4503 | } | |
4504 | ||
4505 | btrfs_record_root_in_trans(trans, dest); | |
4506 | ||
4507 | memset(&dest->root_item.drop_progress, 0, | |
4508 | sizeof(dest->root_item.drop_progress)); | |
4509 | dest->root_item.drop_level = 0; | |
4510 | btrfs_set_root_refs(&dest->root_item, 0); | |
4511 | ||
4512 | if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { | |
4513 | ret = btrfs_insert_orphan_item(trans, | |
4514 | fs_info->tree_root, | |
4515 | dest->root_key.objectid); | |
4516 | if (ret) { | |
4517 | btrfs_abort_transaction(trans, ret); | |
4518 | err = ret; | |
4519 | goto out_end_trans; | |
4520 | } | |
4521 | } | |
4522 | ||
4523 | ret = btrfs_uuid_tree_rem(trans, fs_info, dest->root_item.uuid, | |
4524 | BTRFS_UUID_KEY_SUBVOL, | |
4525 | dest->root_key.objectid); | |
4526 | if (ret && ret != -ENOENT) { | |
4527 | btrfs_abort_transaction(trans, ret); | |
4528 | err = ret; | |
4529 | goto out_end_trans; | |
4530 | } | |
4531 | if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { | |
4532 | ret = btrfs_uuid_tree_rem(trans, fs_info, | |
4533 | dest->root_item.received_uuid, | |
4534 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4535 | dest->root_key.objectid); | |
4536 | if (ret && ret != -ENOENT) { | |
4537 | btrfs_abort_transaction(trans, ret); | |
4538 | err = ret; | |
4539 | goto out_end_trans; | |
4540 | } | |
4541 | } | |
4542 | ||
4543 | out_end_trans: | |
4544 | trans->block_rsv = NULL; | |
4545 | trans->bytes_reserved = 0; | |
4546 | ret = btrfs_end_transaction(trans); | |
4547 | if (ret && !err) | |
4548 | err = ret; | |
4549 | inode->i_flags |= S_DEAD; | |
4550 | out_release: | |
4551 | btrfs_subvolume_release_metadata(fs_info, &block_rsv); | |
4552 | out_up_write: | |
4553 | up_write(&fs_info->subvol_sem); | |
4554 | if (err) { | |
4555 | spin_lock(&dest->root_item_lock); | |
4556 | root_flags = btrfs_root_flags(&dest->root_item); | |
4557 | btrfs_set_root_flags(&dest->root_item, | |
4558 | root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); | |
4559 | spin_unlock(&dest->root_item_lock); | |
4560 | } else { | |
4561 | d_invalidate(dentry); | |
20a68004 | 4562 | btrfs_prune_dentries(dest); |
f60a2364 MT |
4563 | ASSERT(dest->send_in_progress == 0); |
4564 | ||
4565 | /* the last ref */ | |
4566 | if (dest->ino_cache_inode) { | |
4567 | iput(dest->ino_cache_inode); | |
4568 | dest->ino_cache_inode = NULL; | |
4569 | } | |
4570 | } | |
4571 | ||
4572 | return err; | |
4573 | } | |
4574 | ||
39279cc3 CM |
4575 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4576 | { | |
2b0143b5 | 4577 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4578 | int err = 0; |
39279cc3 | 4579 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4580 | struct btrfs_trans_handle *trans; |
44f714da | 4581 | u64 last_unlink_trans; |
39279cc3 | 4582 | |
b3ae244e | 4583 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4584 | return -ENOTEMPTY; |
4a0cc7ca | 4585 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
a79a464d | 4586 | return btrfs_delete_subvolume(dir, dentry); |
134d4512 | 4587 | |
d52be818 | 4588 | trans = __unlink_start_trans(dir); |
a22285a6 | 4589 | if (IS_ERR(trans)) |
5df6a9f6 | 4590 | return PTR_ERR(trans); |
5df6a9f6 | 4591 | |
4a0cc7ca | 4592 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4593 | err = btrfs_unlink_subvol(trans, root, dir, |
4594 | BTRFS_I(inode)->location.objectid, | |
4595 | dentry->d_name.name, | |
4596 | dentry->d_name.len); | |
4597 | goto out; | |
4598 | } | |
4599 | ||
73f2e545 | 4600 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4601 | if (err) |
4df27c4d | 4602 | goto out; |
7b128766 | 4603 | |
44f714da FM |
4604 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4605 | ||
39279cc3 | 4606 | /* now the directory is empty */ |
4ec5934e NB |
4607 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4608 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4609 | dentry->d_name.len); | |
44f714da | 4610 | if (!err) { |
6ef06d27 | 4611 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4612 | /* |
4613 | * Propagate the last_unlink_trans value of the deleted dir to | |
4614 | * its parent directory. This is to prevent an unrecoverable | |
4615 | * log tree in the case we do something like this: | |
4616 | * 1) create dir foo | |
4617 | * 2) create snapshot under dir foo | |
4618 | * 3) delete the snapshot | |
4619 | * 4) rmdir foo | |
4620 | * 5) mkdir foo | |
4621 | * 6) fsync foo or some file inside foo | |
4622 | */ | |
4623 | if (last_unlink_trans >= trans->transid) | |
4624 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4625 | } | |
4df27c4d | 4626 | out: |
3a45bb20 | 4627 | btrfs_end_transaction(trans); |
2ff7e61e | 4628 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4629 | |
39279cc3 CM |
4630 | return err; |
4631 | } | |
4632 | ||
28f75a0e CM |
4633 | static int truncate_space_check(struct btrfs_trans_handle *trans, |
4634 | struct btrfs_root *root, | |
4635 | u64 bytes_deleted) | |
4636 | { | |
0b246afa | 4637 | struct btrfs_fs_info *fs_info = root->fs_info; |
28f75a0e CM |
4638 | int ret; |
4639 | ||
dc95f7bf JB |
4640 | /* |
4641 | * This is only used to apply pressure to the enospc system, we don't | |
4642 | * intend to use this reservation at all. | |
4643 | */ | |
2ff7e61e | 4644 | bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted); |
0b246afa JM |
4645 | bytes_deleted *= fs_info->nodesize; |
4646 | ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv, | |
28f75a0e | 4647 | bytes_deleted, BTRFS_RESERVE_NO_FLUSH); |
dc95f7bf | 4648 | if (!ret) { |
0b246afa | 4649 | trace_btrfs_space_reservation(fs_info, "transaction", |
dc95f7bf JB |
4650 | trans->transid, |
4651 | bytes_deleted, 1); | |
28f75a0e | 4652 | trans->bytes_reserved += bytes_deleted; |
dc95f7bf | 4653 | } |
28f75a0e CM |
4654 | return ret; |
4655 | ||
4656 | } | |
4657 | ||
ddfae63c JB |
4658 | /* |
4659 | * Return this if we need to call truncate_block for the last bit of the | |
4660 | * truncate. | |
4661 | */ | |
4662 | #define NEED_TRUNCATE_BLOCK 1 | |
0305cd5f | 4663 | |
39279cc3 CM |
4664 | /* |
4665 | * this can truncate away extent items, csum items and directory items. | |
4666 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4667 | * any higher than new_size |
39279cc3 CM |
4668 | * |
4669 | * csum items that cross the new i_size are truncated to the new size | |
4670 | * as well. | |
7b128766 JB |
4671 | * |
4672 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4673 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4674 | */ |
8082510e YZ |
4675 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4676 | struct btrfs_root *root, | |
4677 | struct inode *inode, | |
4678 | u64 new_size, u32 min_type) | |
39279cc3 | 4679 | { |
0b246afa | 4680 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4681 | struct btrfs_path *path; |
5f39d397 | 4682 | struct extent_buffer *leaf; |
39279cc3 | 4683 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4684 | struct btrfs_key key; |
4685 | struct btrfs_key found_key; | |
39279cc3 | 4686 | u64 extent_start = 0; |
db94535d | 4687 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4688 | u64 extent_offset = 0; |
39279cc3 | 4689 | u64 item_end = 0; |
c1aa4575 | 4690 | u64 last_size = new_size; |
8082510e | 4691 | u32 found_type = (u8)-1; |
39279cc3 CM |
4692 | int found_extent; |
4693 | int del_item; | |
85e21bac CM |
4694 | int pending_del_nr = 0; |
4695 | int pending_del_slot = 0; | |
179e29e4 | 4696 | int extent_type = -1; |
8082510e | 4697 | int ret; |
4a0cc7ca | 4698 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
28ed1345 | 4699 | u64 bytes_deleted = 0; |
897ca819 TM |
4700 | bool be_nice = false; |
4701 | bool should_throttle = false; | |
4702 | bool should_end = false; | |
8082510e YZ |
4703 | |
4704 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4705 | |
28ed1345 CM |
4706 | /* |
4707 | * for non-free space inodes and ref cows, we want to back off from | |
4708 | * time to time | |
4709 | */ | |
70ddc553 | 4710 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) && |
28ed1345 | 4711 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
897ca819 | 4712 | be_nice = true; |
28ed1345 | 4713 | |
0eb0e19c MF |
4714 | path = btrfs_alloc_path(); |
4715 | if (!path) | |
4716 | return -ENOMEM; | |
e4058b54 | 4717 | path->reada = READA_BACK; |
0eb0e19c | 4718 | |
5dc562c5 JB |
4719 | /* |
4720 | * We want to drop from the next block forward in case this new size is | |
4721 | * not block aligned since we will be keeping the last block of the | |
4722 | * extent just the way it is. | |
4723 | */ | |
27cdeb70 | 4724 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 4725 | root == fs_info->tree_root) |
dcdbc059 | 4726 | btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size, |
0b246afa | 4727 | fs_info->sectorsize), |
da17066c | 4728 | (u64)-1, 0); |
8082510e | 4729 | |
16cdcec7 MX |
4730 | /* |
4731 | * This function is also used to drop the items in the log tree before | |
4732 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4733 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4734 | * items. | |
4735 | */ | |
4736 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4ccb5c72 | 4737 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); |
16cdcec7 | 4738 | |
33345d01 | 4739 | key.objectid = ino; |
39279cc3 | 4740 | key.offset = (u64)-1; |
5f39d397 CM |
4741 | key.type = (u8)-1; |
4742 | ||
85e21bac | 4743 | search_again: |
28ed1345 CM |
4744 | /* |
4745 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4746 | * up a huge file in a single leaf. Most of the time that | |
4747 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4748 | */ | |
fd86a3a3 OS |
4749 | if (be_nice && bytes_deleted > SZ_32M && |
4750 | btrfs_should_end_transaction(trans)) { | |
4751 | ret = -EAGAIN; | |
4752 | goto out; | |
28ed1345 CM |
4753 | } |
4754 | ||
b9473439 | 4755 | path->leave_spinning = 1; |
85e21bac | 4756 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
fd86a3a3 | 4757 | if (ret < 0) |
8082510e | 4758 | goto out; |
d397712b | 4759 | |
85e21bac | 4760 | if (ret > 0) { |
fd86a3a3 | 4761 | ret = 0; |
e02119d5 CM |
4762 | /* there are no items in the tree for us to truncate, we're |
4763 | * done | |
4764 | */ | |
8082510e YZ |
4765 | if (path->slots[0] == 0) |
4766 | goto out; | |
85e21bac CM |
4767 | path->slots[0]--; |
4768 | } | |
4769 | ||
d397712b | 4770 | while (1) { |
39279cc3 | 4771 | fi = NULL; |
5f39d397 CM |
4772 | leaf = path->nodes[0]; |
4773 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4774 | found_type = found_key.type; |
39279cc3 | 4775 | |
33345d01 | 4776 | if (found_key.objectid != ino) |
39279cc3 | 4777 | break; |
5f39d397 | 4778 | |
85e21bac | 4779 | if (found_type < min_type) |
39279cc3 CM |
4780 | break; |
4781 | ||
5f39d397 | 4782 | item_end = found_key.offset; |
39279cc3 | 4783 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4784 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4785 | struct btrfs_file_extent_item); |
179e29e4 CM |
4786 | extent_type = btrfs_file_extent_type(leaf, fi); |
4787 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4788 | item_end += |
db94535d | 4789 | btrfs_file_extent_num_bytes(leaf, fi); |
09ed2f16 LB |
4790 | |
4791 | trace_btrfs_truncate_show_fi_regular( | |
4792 | BTRFS_I(inode), leaf, fi, | |
4793 | found_key.offset); | |
179e29e4 | 4794 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4795 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4796 | path->slots[0], fi); |
09ed2f16 LB |
4797 | |
4798 | trace_btrfs_truncate_show_fi_inline( | |
4799 | BTRFS_I(inode), leaf, fi, path->slots[0], | |
4800 | found_key.offset); | |
39279cc3 | 4801 | } |
008630c1 | 4802 | item_end--; |
39279cc3 | 4803 | } |
8082510e YZ |
4804 | if (found_type > min_type) { |
4805 | del_item = 1; | |
4806 | } else { | |
76b42abb | 4807 | if (item_end < new_size) |
b888db2b | 4808 | break; |
8082510e YZ |
4809 | if (found_key.offset >= new_size) |
4810 | del_item = 1; | |
4811 | else | |
4812 | del_item = 0; | |
39279cc3 | 4813 | } |
39279cc3 | 4814 | found_extent = 0; |
39279cc3 | 4815 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4816 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4817 | goto delete; | |
4818 | ||
4819 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 4820 | u64 num_dec; |
db94535d | 4821 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4822 | if (!del_item) { |
db94535d CM |
4823 | u64 orig_num_bytes = |
4824 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4825 | extent_num_bytes = ALIGN(new_size - |
4826 | found_key.offset, | |
0b246afa | 4827 | fs_info->sectorsize); |
db94535d CM |
4828 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4829 | extent_num_bytes); | |
4830 | num_dec = (orig_num_bytes - | |
9069218d | 4831 | extent_num_bytes); |
27cdeb70 MX |
4832 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4833 | &root->state) && | |
4834 | extent_start != 0) | |
a76a3cd4 | 4835 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4836 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4837 | } else { |
db94535d CM |
4838 | extent_num_bytes = |
4839 | btrfs_file_extent_disk_num_bytes(leaf, | |
4840 | fi); | |
5d4f98a2 YZ |
4841 | extent_offset = found_key.offset - |
4842 | btrfs_file_extent_offset(leaf, fi); | |
4843 | ||
39279cc3 | 4844 | /* FIXME blocksize != 4096 */ |
9069218d | 4845 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4846 | if (extent_start != 0) { |
4847 | found_extent = 1; | |
27cdeb70 MX |
4848 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4849 | &root->state)) | |
a76a3cd4 | 4850 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4851 | } |
39279cc3 | 4852 | } |
9069218d | 4853 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4854 | /* |
4855 | * we can't truncate inline items that have had | |
4856 | * special encodings | |
4857 | */ | |
4858 | if (!del_item && | |
c8b97818 | 4859 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
ddfae63c JB |
4860 | btrfs_file_extent_other_encoding(leaf, fi) == 0 && |
4861 | btrfs_file_extent_compression(leaf, fi) == 0) { | |
4862 | u32 size = (u32)(new_size - found_key.offset); | |
4863 | ||
4864 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4865 | size = btrfs_file_extent_calc_inline_size(size); | |
4866 | btrfs_truncate_item(root->fs_info, path, size, 1); | |
4867 | } else if (!del_item) { | |
514ac8ad | 4868 | /* |
ddfae63c JB |
4869 | * We have to bail so the last_size is set to |
4870 | * just before this extent. | |
514ac8ad | 4871 | */ |
fd86a3a3 | 4872 | ret = NEED_TRUNCATE_BLOCK; |
ddfae63c JB |
4873 | break; |
4874 | } | |
0305cd5f | 4875 | |
ddfae63c | 4876 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
0305cd5f | 4877 | inode_sub_bytes(inode, item_end + 1 - new_size); |
39279cc3 | 4878 | } |
179e29e4 | 4879 | delete: |
ddfae63c JB |
4880 | if (del_item) |
4881 | last_size = found_key.offset; | |
4882 | else | |
4883 | last_size = new_size; | |
39279cc3 | 4884 | if (del_item) { |
85e21bac CM |
4885 | if (!pending_del_nr) { |
4886 | /* no pending yet, add ourselves */ | |
4887 | pending_del_slot = path->slots[0]; | |
4888 | pending_del_nr = 1; | |
4889 | } else if (pending_del_nr && | |
4890 | path->slots[0] + 1 == pending_del_slot) { | |
4891 | /* hop on the pending chunk */ | |
4892 | pending_del_nr++; | |
4893 | pending_del_slot = path->slots[0]; | |
4894 | } else { | |
d397712b | 4895 | BUG(); |
85e21bac | 4896 | } |
39279cc3 CM |
4897 | } else { |
4898 | break; | |
4899 | } | |
897ca819 | 4900 | should_throttle = false; |
28f75a0e | 4901 | |
27cdeb70 MX |
4902 | if (found_extent && |
4903 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
0b246afa | 4904 | root == fs_info->tree_root)) { |
b9473439 | 4905 | btrfs_set_path_blocking(path); |
28ed1345 | 4906 | bytes_deleted += extent_num_bytes; |
84f7d8e6 | 4907 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4908 | extent_num_bytes, 0, |
4909 | btrfs_header_owner(leaf), | |
b06c4bf5 | 4910 | ino, extent_offset); |
05522109 OS |
4911 | if (ret) { |
4912 | btrfs_abort_transaction(trans, ret); | |
4913 | break; | |
4914 | } | |
2ff7e61e JM |
4915 | if (btrfs_should_throttle_delayed_refs(trans, fs_info)) |
4916 | btrfs_async_run_delayed_refs(fs_info, | |
dd4b857a WX |
4917 | trans->delayed_ref_updates * 2, |
4918 | trans->transid, 0); | |
28f75a0e CM |
4919 | if (be_nice) { |
4920 | if (truncate_space_check(trans, root, | |
4921 | extent_num_bytes)) { | |
897ca819 | 4922 | should_end = true; |
28f75a0e CM |
4923 | } |
4924 | if (btrfs_should_throttle_delayed_refs(trans, | |
2ff7e61e | 4925 | fs_info)) |
897ca819 | 4926 | should_throttle = true; |
28f75a0e | 4927 | } |
39279cc3 | 4928 | } |
85e21bac | 4929 | |
8082510e YZ |
4930 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4931 | break; | |
4932 | ||
4933 | if (path->slots[0] == 0 || | |
1262133b | 4934 | path->slots[0] != pending_del_slot || |
28f75a0e | 4935 | should_throttle || should_end) { |
8082510e YZ |
4936 | if (pending_del_nr) { |
4937 | ret = btrfs_del_items(trans, root, path, | |
4938 | pending_del_slot, | |
4939 | pending_del_nr); | |
79787eaa | 4940 | if (ret) { |
66642832 | 4941 | btrfs_abort_transaction(trans, ret); |
fd86a3a3 | 4942 | break; |
79787eaa | 4943 | } |
8082510e YZ |
4944 | pending_del_nr = 0; |
4945 | } | |
b3b4aa74 | 4946 | btrfs_release_path(path); |
28f75a0e | 4947 | if (should_throttle) { |
1262133b JB |
4948 | unsigned long updates = trans->delayed_ref_updates; |
4949 | if (updates) { | |
4950 | trans->delayed_ref_updates = 0; | |
2ff7e61e | 4951 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 4952 | updates * 2); |
fd86a3a3 OS |
4953 | if (ret) |
4954 | break; | |
1262133b JB |
4955 | } |
4956 | } | |
28f75a0e CM |
4957 | /* |
4958 | * if we failed to refill our space rsv, bail out | |
4959 | * and let the transaction restart | |
4960 | */ | |
4961 | if (should_end) { | |
fd86a3a3 OS |
4962 | ret = -EAGAIN; |
4963 | break; | |
28f75a0e | 4964 | } |
85e21bac | 4965 | goto search_again; |
8082510e YZ |
4966 | } else { |
4967 | path->slots[0]--; | |
85e21bac | 4968 | } |
39279cc3 | 4969 | } |
8082510e | 4970 | out: |
fd86a3a3 OS |
4971 | if (ret >= 0 && pending_del_nr) { |
4972 | int err; | |
4973 | ||
4974 | err = btrfs_del_items(trans, root, path, pending_del_slot, | |
85e21bac | 4975 | pending_del_nr); |
fd86a3a3 OS |
4976 | if (err) { |
4977 | btrfs_abort_transaction(trans, err); | |
4978 | ret = err; | |
4979 | } | |
85e21bac | 4980 | } |
76b42abb FM |
4981 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
4982 | ASSERT(last_size >= new_size); | |
fd86a3a3 | 4983 | if (!ret && last_size > new_size) |
76b42abb | 4984 | last_size = new_size; |
7f4f6e0a | 4985 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
76b42abb | 4986 | } |
28ed1345 | 4987 | |
39279cc3 | 4988 | btrfs_free_path(path); |
28ed1345 | 4989 | |
fd86a3a3 | 4990 | if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) { |
28ed1345 | 4991 | unsigned long updates = trans->delayed_ref_updates; |
fd86a3a3 OS |
4992 | int err; |
4993 | ||
28ed1345 CM |
4994 | if (updates) { |
4995 | trans->delayed_ref_updates = 0; | |
fd86a3a3 OS |
4996 | err = btrfs_run_delayed_refs(trans, updates * 2); |
4997 | if (err) | |
4998 | ret = err; | |
28ed1345 CM |
4999 | } |
5000 | } | |
fd86a3a3 | 5001 | return ret; |
39279cc3 CM |
5002 | } |
5003 | ||
5004 | /* | |
9703fefe | 5005 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
5006 | * @inode - inode that we're zeroing |
5007 | * @from - the offset to start zeroing | |
5008 | * @len - the length to zero, 0 to zero the entire range respective to the | |
5009 | * offset | |
5010 | * @front - zero up to the offset instead of from the offset on | |
5011 | * | |
9703fefe | 5012 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 5013 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 5014 | */ |
9703fefe | 5015 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 5016 | int front) |
39279cc3 | 5017 | { |
0b246afa | 5018 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 | 5019 | struct address_space *mapping = inode->i_mapping; |
e6dcd2dc CM |
5020 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
5021 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 5022 | struct extent_state *cached_state = NULL; |
364ecf36 | 5023 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 5024 | char *kaddr; |
0b246afa | 5025 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 5026 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 5027 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 5028 | struct page *page; |
3b16a4e3 | 5029 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 5030 | int ret = 0; |
9703fefe CR |
5031 | u64 block_start; |
5032 | u64 block_end; | |
39279cc3 | 5033 | |
b03ebd99 NB |
5034 | if (IS_ALIGNED(offset, blocksize) && |
5035 | (!len || IS_ALIGNED(len, blocksize))) | |
39279cc3 | 5036 | goto out; |
9703fefe | 5037 | |
8b62f87b JB |
5038 | block_start = round_down(from, blocksize); |
5039 | block_end = block_start + blocksize - 1; | |
5040 | ||
364ecf36 | 5041 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8b62f87b | 5042 | block_start, blocksize); |
5d5e103a JB |
5043 | if (ret) |
5044 | goto out; | |
39279cc3 | 5045 | |
211c17f5 | 5046 | again: |
3b16a4e3 | 5047 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 5048 | if (!page) { |
bc42bda2 | 5049 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
5050 | block_start, blocksize, true); |
5051 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true); | |
ac6a2b36 | 5052 | ret = -ENOMEM; |
39279cc3 | 5053 | goto out; |
5d5e103a | 5054 | } |
e6dcd2dc | 5055 | |
39279cc3 | 5056 | if (!PageUptodate(page)) { |
9ebefb18 | 5057 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 5058 | lock_page(page); |
211c17f5 CM |
5059 | if (page->mapping != mapping) { |
5060 | unlock_page(page); | |
09cbfeaf | 5061 | put_page(page); |
211c17f5 CM |
5062 | goto again; |
5063 | } | |
39279cc3 CM |
5064 | if (!PageUptodate(page)) { |
5065 | ret = -EIO; | |
89642229 | 5066 | goto out_unlock; |
39279cc3 CM |
5067 | } |
5068 | } | |
211c17f5 | 5069 | wait_on_page_writeback(page); |
e6dcd2dc | 5070 | |
9703fefe | 5071 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
5072 | set_page_extent_mapped(page); |
5073 | ||
9703fefe | 5074 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 5075 | if (ordered) { |
9703fefe | 5076 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 5077 | &cached_state); |
e6dcd2dc | 5078 | unlock_page(page); |
09cbfeaf | 5079 | put_page(page); |
eb84ae03 | 5080 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
5081 | btrfs_put_ordered_extent(ordered); |
5082 | goto again; | |
5083 | } | |
5084 | ||
9703fefe | 5085 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
5086 | EXTENT_DIRTY | EXTENT_DELALLOC | |
5087 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 5088 | 0, 0, &cached_state); |
5d5e103a | 5089 | |
e3b8a485 | 5090 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
ba8b04c1 | 5091 | &cached_state, 0); |
9ed74f2d | 5092 | if (ret) { |
9703fefe | 5093 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 5094 | &cached_state); |
9ed74f2d JB |
5095 | goto out_unlock; |
5096 | } | |
5097 | ||
9703fefe | 5098 | if (offset != blocksize) { |
2aaa6655 | 5099 | if (!len) |
9703fefe | 5100 | len = blocksize - offset; |
e6dcd2dc | 5101 | kaddr = kmap(page); |
2aaa6655 | 5102 | if (front) |
9703fefe CR |
5103 | memset(kaddr + (block_start - page_offset(page)), |
5104 | 0, offset); | |
2aaa6655 | 5105 | else |
9703fefe CR |
5106 | memset(kaddr + (block_start - page_offset(page)) + offset, |
5107 | 0, len); | |
e6dcd2dc CM |
5108 | flush_dcache_page(page); |
5109 | kunmap(page); | |
5110 | } | |
247e743c | 5111 | ClearPageChecked(page); |
e6dcd2dc | 5112 | set_page_dirty(page); |
e43bbe5e | 5113 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state); |
39279cc3 | 5114 | |
89642229 | 5115 | out_unlock: |
5d5e103a | 5116 | if (ret) |
bc42bda2 | 5117 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
43b18595 QW |
5118 | blocksize, true); |
5119 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0)); | |
39279cc3 | 5120 | unlock_page(page); |
09cbfeaf | 5121 | put_page(page); |
39279cc3 | 5122 | out: |
364ecf36 | 5123 | extent_changeset_free(data_reserved); |
39279cc3 CM |
5124 | return ret; |
5125 | } | |
5126 | ||
16e7549f JB |
5127 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
5128 | u64 offset, u64 len) | |
5129 | { | |
0b246afa | 5130 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
16e7549f JB |
5131 | struct btrfs_trans_handle *trans; |
5132 | int ret; | |
5133 | ||
5134 | /* | |
5135 | * Still need to make sure the inode looks like it's been updated so | |
5136 | * that any holes get logged if we fsync. | |
5137 | */ | |
0b246afa JM |
5138 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) { |
5139 | BTRFS_I(inode)->last_trans = fs_info->generation; | |
16e7549f JB |
5140 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
5141 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
5142 | return 0; | |
5143 | } | |
5144 | ||
5145 | /* | |
5146 | * 1 - for the one we're dropping | |
5147 | * 1 - for the one we're adding | |
5148 | * 1 - for updating the inode. | |
5149 | */ | |
5150 | trans = btrfs_start_transaction(root, 3); | |
5151 | if (IS_ERR(trans)) | |
5152 | return PTR_ERR(trans); | |
5153 | ||
5154 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
5155 | if (ret) { | |
66642832 | 5156 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 5157 | btrfs_end_transaction(trans); |
16e7549f JB |
5158 | return ret; |
5159 | } | |
5160 | ||
f85b7379 DS |
5161 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)), |
5162 | offset, 0, 0, len, 0, len, 0, 0, 0); | |
16e7549f | 5163 | if (ret) |
66642832 | 5164 | btrfs_abort_transaction(trans, ret); |
16e7549f JB |
5165 | else |
5166 | btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 5167 | btrfs_end_transaction(trans); |
16e7549f JB |
5168 | return ret; |
5169 | } | |
5170 | ||
695a0d0d JB |
5171 | /* |
5172 | * This function puts in dummy file extents for the area we're creating a hole | |
5173 | * for. So if we are truncating this file to a larger size we need to insert | |
5174 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
5175 | * the range between oldsize and size | |
5176 | */ | |
a41ad394 | 5177 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 5178 | { |
0b246afa | 5179 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9036c102 YZ |
5180 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5181 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 5182 | struct extent_map *em = NULL; |
2ac55d41 | 5183 | struct extent_state *cached_state = NULL; |
5dc562c5 | 5184 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
0b246afa JM |
5185 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
5186 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
5187 | u64 last_byte; |
5188 | u64 cur_offset; | |
5189 | u64 hole_size; | |
9ed74f2d | 5190 | int err = 0; |
39279cc3 | 5191 | |
a71754fc | 5192 | /* |
9703fefe CR |
5193 | * If our size started in the middle of a block we need to zero out the |
5194 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
5195 | * expose stale data. |
5196 | */ | |
9703fefe | 5197 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
5198 | if (err) |
5199 | return err; | |
5200 | ||
9036c102 YZ |
5201 | if (size <= hole_start) |
5202 | return 0; | |
5203 | ||
9036c102 YZ |
5204 | while (1) { |
5205 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 5206 | |
ff13db41 | 5207 | lock_extent_bits(io_tree, hole_start, block_end - 1, |
d0082371 | 5208 | &cached_state); |
a776c6fa | 5209 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start, |
fa7c1494 | 5210 | block_end - hole_start); |
9036c102 YZ |
5211 | if (!ordered) |
5212 | break; | |
2ac55d41 | 5213 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
e43bbe5e | 5214 | &cached_state); |
fa7c1494 | 5215 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
5216 | btrfs_put_ordered_extent(ordered); |
5217 | } | |
39279cc3 | 5218 | |
9036c102 YZ |
5219 | cur_offset = hole_start; |
5220 | while (1) { | |
fc4f21b1 | 5221 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
9036c102 | 5222 | block_end - cur_offset, 0); |
79787eaa JM |
5223 | if (IS_ERR(em)) { |
5224 | err = PTR_ERR(em); | |
f2767956 | 5225 | em = NULL; |
79787eaa JM |
5226 | break; |
5227 | } | |
9036c102 | 5228 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 5229 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
8082510e | 5230 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 5231 | struct extent_map *hole_em; |
9036c102 | 5232 | hole_size = last_byte - cur_offset; |
9ed74f2d | 5233 | |
16e7549f JB |
5234 | err = maybe_insert_hole(root, inode, cur_offset, |
5235 | hole_size); | |
5236 | if (err) | |
3893e33b | 5237 | break; |
dcdbc059 | 5238 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
5239 | cur_offset + hole_size - 1, 0); |
5240 | hole_em = alloc_extent_map(); | |
5241 | if (!hole_em) { | |
5242 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
5243 | &BTRFS_I(inode)->runtime_flags); | |
5244 | goto next; | |
5245 | } | |
5246 | hole_em->start = cur_offset; | |
5247 | hole_em->len = hole_size; | |
5248 | hole_em->orig_start = cur_offset; | |
8082510e | 5249 | |
5dc562c5 JB |
5250 | hole_em->block_start = EXTENT_MAP_HOLE; |
5251 | hole_em->block_len = 0; | |
b4939680 | 5252 | hole_em->orig_block_len = 0; |
cc95bef6 | 5253 | hole_em->ram_bytes = hole_size; |
0b246afa | 5254 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 | 5255 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 5256 | hole_em->generation = fs_info->generation; |
8082510e | 5257 | |
5dc562c5 JB |
5258 | while (1) { |
5259 | write_lock(&em_tree->lock); | |
09a2a8f9 | 5260 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
5261 | write_unlock(&em_tree->lock); |
5262 | if (err != -EEXIST) | |
5263 | break; | |
dcdbc059 NB |
5264 | btrfs_drop_extent_cache(BTRFS_I(inode), |
5265 | cur_offset, | |
5dc562c5 JB |
5266 | cur_offset + |
5267 | hole_size - 1, 0); | |
5268 | } | |
5269 | free_extent_map(hole_em); | |
9036c102 | 5270 | } |
16e7549f | 5271 | next: |
9036c102 | 5272 | free_extent_map(em); |
a22285a6 | 5273 | em = NULL; |
9036c102 | 5274 | cur_offset = last_byte; |
8082510e | 5275 | if (cur_offset >= block_end) |
9036c102 YZ |
5276 | break; |
5277 | } | |
a22285a6 | 5278 | free_extent_map(em); |
e43bbe5e | 5279 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state); |
9036c102 YZ |
5280 | return err; |
5281 | } | |
39279cc3 | 5282 | |
3972f260 | 5283 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 5284 | { |
f4a2f4c5 MX |
5285 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5286 | struct btrfs_trans_handle *trans; | |
a41ad394 | 5287 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
5288 | loff_t newsize = attr->ia_size; |
5289 | int mask = attr->ia_valid; | |
8082510e YZ |
5290 | int ret; |
5291 | ||
3972f260 ES |
5292 | /* |
5293 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
5294 | * special case where we need to update the times despite not having | |
5295 | * these flags set. For all other operations the VFS set these flags | |
5296 | * explicitly if it wants a timestamp update. | |
5297 | */ | |
dff6efc3 CH |
5298 | if (newsize != oldsize) { |
5299 | inode_inc_iversion(inode); | |
5300 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
5301 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 5302 | current_time(inode); |
dff6efc3 | 5303 | } |
3972f260 | 5304 | |
a41ad394 | 5305 | if (newsize > oldsize) { |
9ea24bbe | 5306 | /* |
ea14b57f | 5307 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
5308 | * This is to ensure the snapshot captures a fully consistent |
5309 | * state of this file - if the snapshot captures this expanding | |
5310 | * truncation, it must capture all writes that happened before | |
5311 | * this truncation. | |
5312 | */ | |
0bc19f90 | 5313 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 5314 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe | 5315 | if (ret) { |
ea14b57f | 5316 | btrfs_end_write_no_snapshotting(root); |
8082510e | 5317 | return ret; |
9ea24bbe | 5318 | } |
8082510e | 5319 | |
f4a2f4c5 | 5320 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 5321 | if (IS_ERR(trans)) { |
ea14b57f | 5322 | btrfs_end_write_no_snapshotting(root); |
f4a2f4c5 | 5323 | return PTR_ERR(trans); |
9ea24bbe | 5324 | } |
f4a2f4c5 MX |
5325 | |
5326 | i_size_write(inode, newsize); | |
5327 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 5328 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 5329 | ret = btrfs_update_inode(trans, root, inode); |
ea14b57f | 5330 | btrfs_end_write_no_snapshotting(root); |
3a45bb20 | 5331 | btrfs_end_transaction(trans); |
a41ad394 | 5332 | } else { |
8082510e | 5333 | |
a41ad394 JB |
5334 | /* |
5335 | * We're truncating a file that used to have good data down to | |
5336 | * zero. Make sure it gets into the ordered flush list so that | |
5337 | * any new writes get down to disk quickly. | |
5338 | */ | |
5339 | if (newsize == 0) | |
72ac3c0d JB |
5340 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
5341 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 5342 | |
a41ad394 | 5343 | truncate_setsize(inode, newsize); |
2e60a51e MX |
5344 | |
5345 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
abcefb1e | 5346 | btrfs_inode_block_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5347 | inode_dio_wait(inode); |
0b581701 | 5348 | btrfs_inode_resume_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5349 | |
213e8c55 | 5350 | ret = btrfs_truncate(inode, newsize == oldsize); |
7f4f6e0a JB |
5351 | if (ret && inode->i_nlink) { |
5352 | int err; | |
5353 | ||
5354 | /* | |
f7e9e8fc OS |
5355 | * Truncate failed, so fix up the in-memory size. We |
5356 | * adjusted disk_i_size down as we removed extents, so | |
5357 | * wait for disk_i_size to be stable and then update the | |
5358 | * in-memory size to match. | |
7f4f6e0a | 5359 | */ |
f7e9e8fc | 5360 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
7f4f6e0a | 5361 | if (err) |
f7e9e8fc OS |
5362 | return err; |
5363 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
7f4f6e0a | 5364 | } |
8082510e YZ |
5365 | } |
5366 | ||
a41ad394 | 5367 | return ret; |
8082510e YZ |
5368 | } |
5369 | ||
9036c102 YZ |
5370 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5371 | { | |
2b0143b5 | 5372 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5373 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5374 | int err; |
39279cc3 | 5375 | |
b83cc969 LZ |
5376 | if (btrfs_root_readonly(root)) |
5377 | return -EROFS; | |
5378 | ||
31051c85 | 5379 | err = setattr_prepare(dentry, attr); |
9036c102 YZ |
5380 | if (err) |
5381 | return err; | |
2bf5a725 | 5382 | |
5a3f23d5 | 5383 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5384 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5385 | if (err) |
5386 | return err; | |
39279cc3 | 5387 | } |
9036c102 | 5388 | |
1025774c CH |
5389 | if (attr->ia_valid) { |
5390 | setattr_copy(inode, attr); | |
0c4d2d95 | 5391 | inode_inc_iversion(inode); |
22c44fe6 | 5392 | err = btrfs_dirty_inode(inode); |
1025774c | 5393 | |
22c44fe6 | 5394 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5395 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5396 | } |
33268eaf | 5397 | |
39279cc3 CM |
5398 | return err; |
5399 | } | |
61295eb8 | 5400 | |
131e404a FDBM |
5401 | /* |
5402 | * While truncating the inode pages during eviction, we get the VFS calling | |
5403 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5404 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5405 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5406 | * extent_state structures over and over, wasting lots of time. | |
5407 | * | |
5408 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5409 | * those expensive operations on a per page basis and do only the ordered io | |
5410 | * finishing, while we release here the extent_map and extent_state structures, | |
5411 | * without the excessive merging and splitting. | |
5412 | */ | |
5413 | static void evict_inode_truncate_pages(struct inode *inode) | |
5414 | { | |
5415 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5416 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5417 | struct rb_node *node; | |
5418 | ||
5419 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5420 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5421 | |
5422 | write_lock(&map_tree->lock); | |
5423 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
5424 | struct extent_map *em; | |
5425 | ||
5426 | node = rb_first(&map_tree->map); | |
5427 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
5428 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5429 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5430 | remove_extent_mapping(map_tree, em); |
5431 | free_extent_map(em); | |
7064dd5c FM |
5432 | if (need_resched()) { |
5433 | write_unlock(&map_tree->lock); | |
5434 | cond_resched(); | |
5435 | write_lock(&map_tree->lock); | |
5436 | } | |
131e404a FDBM |
5437 | } |
5438 | write_unlock(&map_tree->lock); | |
5439 | ||
6ca07097 FM |
5440 | /* |
5441 | * Keep looping until we have no more ranges in the io tree. | |
5442 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5443 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5444 | * still in progress (unlocked the pages in the bio but did not yet | |
5445 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5446 | * ranges can still be locked and eviction started because before |
5447 | * submitting those bios, which are executed by a separate task (work | |
5448 | * queue kthread), inode references (inode->i_count) were not taken | |
5449 | * (which would be dropped in the end io callback of each bio). | |
5450 | * Therefore here we effectively end up waiting for those bios and | |
5451 | * anyone else holding locked ranges without having bumped the inode's | |
5452 | * reference count - if we don't do it, when they access the inode's | |
5453 | * io_tree to unlock a range it may be too late, leading to an | |
5454 | * use-after-free issue. | |
5455 | */ | |
131e404a FDBM |
5456 | spin_lock(&io_tree->lock); |
5457 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5458 | struct extent_state *state; | |
5459 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5460 | u64 start; |
5461 | u64 end; | |
131e404a FDBM |
5462 | |
5463 | node = rb_first(&io_tree->state); | |
5464 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5465 | start = state->start; |
5466 | end = state->end; | |
131e404a FDBM |
5467 | spin_unlock(&io_tree->lock); |
5468 | ||
ff13db41 | 5469 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5470 | |
5471 | /* | |
5472 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5473 | * and its reserved space won't be freed by delayed_ref. | |
5474 | * So we need to free its reserved space here. | |
5475 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5476 | * | |
5477 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5478 | */ | |
5479 | if (state->state & EXTENT_DELALLOC) | |
bc42bda2 | 5480 | btrfs_qgroup_free_data(inode, NULL, start, end - start + 1); |
b9d0b389 | 5481 | |
6ca07097 | 5482 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5483 | EXTENT_LOCKED | EXTENT_DIRTY | |
5484 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
ae0f1625 | 5485 | EXTENT_DEFRAG, 1, 1, &cached_state); |
131e404a | 5486 | |
7064dd5c | 5487 | cond_resched(); |
131e404a FDBM |
5488 | spin_lock(&io_tree->lock); |
5489 | } | |
5490 | spin_unlock(&io_tree->lock); | |
5491 | } | |
5492 | ||
bd555975 | 5493 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5494 | { |
0b246afa | 5495 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5496 | struct btrfs_trans_handle *trans; |
5497 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 5498 | struct btrfs_block_rsv *rsv, *global_rsv; |
3bce876f | 5499 | int steal_from_global = 0; |
3d48d981 | 5500 | u64 min_size; |
39279cc3 CM |
5501 | int ret; |
5502 | ||
1abe9b8a | 5503 | trace_btrfs_inode_evict(inode); |
5504 | ||
3d48d981 | 5505 | if (!root) { |
e8f1bc14 | 5506 | clear_inode(inode); |
3d48d981 NB |
5507 | return; |
5508 | } | |
5509 | ||
0b246afa | 5510 | min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); |
3d48d981 | 5511 | |
131e404a FDBM |
5512 | evict_inode_truncate_pages(inode); |
5513 | ||
69e9c6c6 SB |
5514 | if (inode->i_nlink && |
5515 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5516 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5517 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5518 | goto no_delete; |
5519 | ||
39279cc3 | 5520 | if (is_bad_inode(inode)) { |
3d6ae7bb | 5521 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
39279cc3 CM |
5522 | goto no_delete; |
5523 | } | |
bd555975 | 5524 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
a30e577c JM |
5525 | if (!special_file(inode->i_mode)) |
5526 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5f39d397 | 5527 | |
7ab7956e | 5528 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5529 | |
0b246afa | 5530 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
6bf02314 | 5531 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 5532 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
5533 | goto no_delete; |
5534 | } | |
5535 | ||
76dda93c | 5536 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5537 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5538 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5539 | goto no_delete; |
5540 | } | |
5541 | ||
aa79021f | 5542 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
0e8c36a9 | 5543 | if (ret) { |
3d6ae7bb | 5544 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
0e8c36a9 MX |
5545 | goto no_delete; |
5546 | } | |
5547 | ||
2ff7e61e | 5548 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
4289a667 | 5549 | if (!rsv) { |
3d6ae7bb | 5550 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
4289a667 JB |
5551 | goto no_delete; |
5552 | } | |
4a338542 | 5553 | rsv->size = min_size; |
ca7e70f5 | 5554 | rsv->failfast = 1; |
0b246afa | 5555 | global_rsv = &fs_info->global_block_rsv; |
4289a667 | 5556 | |
6ef06d27 | 5557 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5558 | |
4289a667 | 5559 | /* |
8407aa46 MX |
5560 | * This is a bit simpler than btrfs_truncate since we've already |
5561 | * reserved our space for our orphan item in the unlink, so we just | |
5562 | * need to reserve some slack space in case we add bytes and update | |
5563 | * inode item when doing the truncate. | |
4289a667 | 5564 | */ |
8082510e | 5565 | while (1) { |
08e007d2 MX |
5566 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
5567 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
5568 | |
5569 | /* | |
5570 | * Try and steal from the global reserve since we will | |
5571 | * likely not use this space anyway, we want to try as | |
5572 | * hard as possible to get this to work. | |
5573 | */ | |
5574 | if (ret) | |
3bce876f JB |
5575 | steal_from_global++; |
5576 | else | |
5577 | steal_from_global = 0; | |
5578 | ret = 0; | |
d68fc57b | 5579 | |
3bce876f JB |
5580 | /* |
5581 | * steal_from_global == 0: we reserved stuff, hooray! | |
5582 | * steal_from_global == 1: we didn't reserve stuff, boo! | |
5583 | * steal_from_global == 2: we've committed, still not a lot of | |
5584 | * room but maybe we'll have room in the global reserve this | |
5585 | * time. | |
5586 | * steal_from_global == 3: abandon all hope! | |
5587 | */ | |
5588 | if (steal_from_global > 2) { | |
0b246afa JM |
5589 | btrfs_warn(fs_info, |
5590 | "Could not get space for a delete, will truncate on mount %d", | |
5591 | ret); | |
3d6ae7bb | 5592 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5593 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5594 | goto no_delete; |
d68fc57b | 5595 | } |
7b128766 | 5596 | |
0e8c36a9 | 5597 | trans = btrfs_join_transaction(root); |
4289a667 | 5598 | if (IS_ERR(trans)) { |
3d6ae7bb | 5599 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5600 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5601 | goto no_delete; |
d68fc57b | 5602 | } |
7b128766 | 5603 | |
3bce876f | 5604 | /* |
01327610 | 5605 | * We can't just steal from the global reserve, we need to make |
3bce876f JB |
5606 | * sure there is room to do it, if not we need to commit and try |
5607 | * again. | |
5608 | */ | |
5609 | if (steal_from_global) { | |
2ff7e61e | 5610 | if (!btrfs_check_space_for_delayed_refs(trans, fs_info)) |
3bce876f | 5611 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, |
25d609f8 | 5612 | min_size, 0); |
3bce876f JB |
5613 | else |
5614 | ret = -ENOSPC; | |
5615 | } | |
5616 | ||
5617 | /* | |
5618 | * Couldn't steal from the global reserve, we have too much | |
5619 | * pending stuff built up, commit the transaction and try it | |
5620 | * again. | |
5621 | */ | |
5622 | if (ret) { | |
3a45bb20 | 5623 | ret = btrfs_commit_transaction(trans); |
3bce876f | 5624 | if (ret) { |
3d6ae7bb | 5625 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5626 | btrfs_free_block_rsv(fs_info, rsv); |
3bce876f JB |
5627 | goto no_delete; |
5628 | } | |
5629 | continue; | |
5630 | } else { | |
5631 | steal_from_global = 0; | |
5632 | } | |
5633 | ||
4289a667 JB |
5634 | trans->block_rsv = rsv; |
5635 | ||
d68fc57b | 5636 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
28ed1345 | 5637 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 5638 | break; |
85e21bac | 5639 | |
0b246afa | 5640 | trans->block_rsv = &fs_info->trans_block_rsv; |
3a45bb20 | 5641 | btrfs_end_transaction(trans); |
8082510e | 5642 | trans = NULL; |
2ff7e61e | 5643 | btrfs_btree_balance_dirty(fs_info); |
8082510e | 5644 | } |
5f39d397 | 5645 | |
2ff7e61e | 5646 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5647 | |
4ef31a45 JB |
5648 | /* |
5649 | * Errors here aren't a big deal, it just means we leave orphan items | |
5650 | * in the tree. They will be cleaned up on the next mount. | |
5651 | */ | |
8082510e | 5652 | if (ret == 0) { |
4289a667 | 5653 | trans->block_rsv = root->orphan_block_rsv; |
3d6ae7bb | 5654 | btrfs_orphan_del(trans, BTRFS_I(inode)); |
4ef31a45 | 5655 | } else { |
3d6ae7bb | 5656 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
8082510e | 5657 | } |
54aa1f4d | 5658 | |
0b246afa JM |
5659 | trans->block_rsv = &fs_info->trans_block_rsv; |
5660 | if (!(root == fs_info->tree_root || | |
581bb050 | 5661 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) |
4a0cc7ca | 5662 | btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode))); |
581bb050 | 5663 | |
3a45bb20 | 5664 | btrfs_end_transaction(trans); |
2ff7e61e | 5665 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 5666 | no_delete: |
f48d1cf5 | 5667 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5668 | clear_inode(inode); |
39279cc3 CM |
5669 | } |
5670 | ||
5671 | /* | |
5672 | * this returns the key found in the dir entry in the location pointer. | |
005d6712 SY |
5673 | * If no dir entries were found, returns -ENOENT. |
5674 | * If found a corrupted location in dir entry, returns -EUCLEAN. | |
39279cc3 CM |
5675 | */ |
5676 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5677 | struct btrfs_key *location) | |
5678 | { | |
5679 | const char *name = dentry->d_name.name; | |
5680 | int namelen = dentry->d_name.len; | |
5681 | struct btrfs_dir_item *di; | |
5682 | struct btrfs_path *path; | |
5683 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5684 | int ret = 0; |
39279cc3 CM |
5685 | |
5686 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5687 | if (!path) |
5688 | return -ENOMEM; | |
3954401f | 5689 | |
f85b7379 DS |
5690 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5691 | name, namelen, 0); | |
005d6712 SY |
5692 | if (!di) { |
5693 | ret = -ENOENT; | |
5694 | goto out; | |
5695 | } | |
5696 | if (IS_ERR(di)) { | |
0d9f7f3e | 5697 | ret = PTR_ERR(di); |
005d6712 SY |
5698 | goto out; |
5699 | } | |
d397712b | 5700 | |
5f39d397 | 5701 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5702 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5703 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
005d6712 | 5704 | ret = -EUCLEAN; |
56a0e706 LB |
5705 | btrfs_warn(root->fs_info, |
5706 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5707 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5708 | location->objectid, location->type, location->offset); | |
56a0e706 | 5709 | } |
39279cc3 | 5710 | out: |
39279cc3 CM |
5711 | btrfs_free_path(path); |
5712 | return ret; | |
5713 | } | |
5714 | ||
5715 | /* | |
5716 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5717 | * needs to be changed to reflect the root directory of the tree root. This | |
5718 | * is kind of like crossing a mount point. | |
5719 | */ | |
2ff7e61e | 5720 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5721 | struct inode *dir, |
5722 | struct dentry *dentry, | |
5723 | struct btrfs_key *location, | |
5724 | struct btrfs_root **sub_root) | |
39279cc3 | 5725 | { |
4df27c4d YZ |
5726 | struct btrfs_path *path; |
5727 | struct btrfs_root *new_root; | |
5728 | struct btrfs_root_ref *ref; | |
5729 | struct extent_buffer *leaf; | |
1d4c08e0 | 5730 | struct btrfs_key key; |
4df27c4d YZ |
5731 | int ret; |
5732 | int err = 0; | |
39279cc3 | 5733 | |
4df27c4d YZ |
5734 | path = btrfs_alloc_path(); |
5735 | if (!path) { | |
5736 | err = -ENOMEM; | |
5737 | goto out; | |
5738 | } | |
39279cc3 | 5739 | |
4df27c4d | 5740 | err = -ENOENT; |
1d4c08e0 DS |
5741 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5742 | key.type = BTRFS_ROOT_REF_KEY; | |
5743 | key.offset = location->objectid; | |
5744 | ||
0b246afa | 5745 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5746 | if (ret) { |
5747 | if (ret < 0) | |
5748 | err = ret; | |
5749 | goto out; | |
5750 | } | |
39279cc3 | 5751 | |
4df27c4d YZ |
5752 | leaf = path->nodes[0]; |
5753 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5754 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5755 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5756 | goto out; | |
39279cc3 | 5757 | |
4df27c4d YZ |
5758 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5759 | (unsigned long)(ref + 1), | |
5760 | dentry->d_name.len); | |
5761 | if (ret) | |
5762 | goto out; | |
5763 | ||
b3b4aa74 | 5764 | btrfs_release_path(path); |
4df27c4d | 5765 | |
0b246afa | 5766 | new_root = btrfs_read_fs_root_no_name(fs_info, location); |
4df27c4d YZ |
5767 | if (IS_ERR(new_root)) { |
5768 | err = PTR_ERR(new_root); | |
5769 | goto out; | |
5770 | } | |
5771 | ||
4df27c4d YZ |
5772 | *sub_root = new_root; |
5773 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5774 | location->type = BTRFS_INODE_ITEM_KEY; | |
5775 | location->offset = 0; | |
5776 | err = 0; | |
5777 | out: | |
5778 | btrfs_free_path(path); | |
5779 | return err; | |
39279cc3 CM |
5780 | } |
5781 | ||
5d4f98a2 YZ |
5782 | static void inode_tree_add(struct inode *inode) |
5783 | { | |
5784 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5785 | struct btrfs_inode *entry; | |
03e860bd NP |
5786 | struct rb_node **p; |
5787 | struct rb_node *parent; | |
cef21937 | 5788 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5789 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5790 | |
1d3382cb | 5791 | if (inode_unhashed(inode)) |
76dda93c | 5792 | return; |
e1409cef | 5793 | parent = NULL; |
5d4f98a2 | 5794 | spin_lock(&root->inode_lock); |
e1409cef | 5795 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5796 | while (*p) { |
5797 | parent = *p; | |
5798 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5799 | ||
4a0cc7ca | 5800 | if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5801 | p = &parent->rb_left; |
4a0cc7ca | 5802 | else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5803 | p = &parent->rb_right; |
5d4f98a2 YZ |
5804 | else { |
5805 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5806 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5807 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd NP |
5808 | RB_CLEAR_NODE(parent); |
5809 | spin_unlock(&root->inode_lock); | |
cef21937 | 5810 | return; |
5d4f98a2 YZ |
5811 | } |
5812 | } | |
cef21937 FDBM |
5813 | rb_link_node(new, parent, p); |
5814 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5815 | spin_unlock(&root->inode_lock); |
5816 | } | |
5817 | ||
5818 | static void inode_tree_del(struct inode *inode) | |
5819 | { | |
0b246afa | 5820 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5d4f98a2 | 5821 | struct btrfs_root *root = BTRFS_I(inode)->root; |
76dda93c | 5822 | int empty = 0; |
5d4f98a2 | 5823 | |
03e860bd | 5824 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5825 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5826 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5827 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5828 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5829 | } |
03e860bd | 5830 | spin_unlock(&root->inode_lock); |
76dda93c | 5831 | |
69e9c6c6 | 5832 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
0b246afa | 5833 | synchronize_srcu(&fs_info->subvol_srcu); |
76dda93c YZ |
5834 | spin_lock(&root->inode_lock); |
5835 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5836 | spin_unlock(&root->inode_lock); | |
5837 | if (empty) | |
5838 | btrfs_add_dead_root(root); | |
5839 | } | |
5840 | } | |
5841 | ||
5d4f98a2 | 5842 | |
e02119d5 CM |
5843 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5844 | { | |
5845 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5846 | inode->i_ino = args->location->objectid; |
5847 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5848 | sizeof(*args->location)); | |
e02119d5 | 5849 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5850 | return 0; |
5851 | } | |
5852 | ||
5853 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5854 | { | |
5855 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5856 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5857 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5858 | } |
5859 | ||
5d4f98a2 | 5860 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5861 | struct btrfs_key *location, |
5d4f98a2 | 5862 | struct btrfs_root *root) |
39279cc3 CM |
5863 | { |
5864 | struct inode *inode; | |
5865 | struct btrfs_iget_args args; | |
90d3e592 | 5866 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5867 | |
90d3e592 | 5868 | args.location = location; |
39279cc3 CM |
5869 | args.root = root; |
5870 | ||
778ba82b | 5871 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5872 | btrfs_init_locked_inode, |
5873 | (void *)&args); | |
5874 | return inode; | |
5875 | } | |
5876 | ||
1a54ef8c BR |
5877 | /* Get an inode object given its location and corresponding root. |
5878 | * Returns in *is_new if the inode was read from disk | |
5879 | */ | |
5880 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5881 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5882 | { |
5883 | struct inode *inode; | |
5884 | ||
90d3e592 | 5885 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5886 | if (!inode) |
5d4f98a2 | 5887 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5888 | |
5889 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5890 | int ret; |
5891 | ||
5892 | ret = btrfs_read_locked_inode(inode); | |
1748f843 MF |
5893 | if (!is_bad_inode(inode)) { |
5894 | inode_tree_add(inode); | |
5895 | unlock_new_inode(inode); | |
5896 | if (new) | |
5897 | *new = 1; | |
5898 | } else { | |
e0b6d65b ST |
5899 | unlock_new_inode(inode); |
5900 | iput(inode); | |
67710892 FM |
5901 | ASSERT(ret < 0); |
5902 | inode = ERR_PTR(ret < 0 ? ret : -ESTALE); | |
1748f843 MF |
5903 | } |
5904 | } | |
5905 | ||
1a54ef8c BR |
5906 | return inode; |
5907 | } | |
5908 | ||
4df27c4d YZ |
5909 | static struct inode *new_simple_dir(struct super_block *s, |
5910 | struct btrfs_key *key, | |
5911 | struct btrfs_root *root) | |
5912 | { | |
5913 | struct inode *inode = new_inode(s); | |
5914 | ||
5915 | if (!inode) | |
5916 | return ERR_PTR(-ENOMEM); | |
5917 | ||
4df27c4d YZ |
5918 | BTRFS_I(inode)->root = root; |
5919 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5920 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5921 | |
5922 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5923 | inode->i_op = &btrfs_dir_ro_inode_operations; |
1fdf4194 | 5924 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5925 | inode->i_fop = &simple_dir_operations; |
5926 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5927 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5928 | inode->i_atime = inode->i_mtime; |
5929 | inode->i_ctime = inode->i_mtime; | |
5930 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
4df27c4d YZ |
5931 | |
5932 | return inode; | |
5933 | } | |
5934 | ||
3de4586c | 5935 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5936 | { |
0b246afa | 5937 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5938 | struct inode *inode; |
4df27c4d | 5939 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5940 | struct btrfs_root *sub_root = root; |
5941 | struct btrfs_key location; | |
76dda93c | 5942 | int index; |
b4aff1f8 | 5943 | int ret = 0; |
39279cc3 CM |
5944 | |
5945 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5946 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5947 | |
39e3c955 | 5948 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5949 | if (ret < 0) |
5950 | return ERR_PTR(ret); | |
5f39d397 | 5951 | |
4df27c4d | 5952 | if (location.type == BTRFS_INODE_ITEM_KEY) { |
73f73415 | 5953 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5954 | return inode; |
5955 | } | |
5956 | ||
0b246afa | 5957 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2ff7e61e | 5958 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5959 | &location, &sub_root); |
5960 | if (ret < 0) { | |
5961 | if (ret != -ENOENT) | |
5962 | inode = ERR_PTR(ret); | |
5963 | else | |
5964 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5965 | } else { | |
73f73415 | 5966 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5967 | } |
0b246afa | 5968 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
76dda93c | 5969 | |
34d19bad | 5970 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5971 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5972 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5973 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5974 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5975 | if (ret) { |
5976 | iput(inode); | |
66b4ffd1 | 5977 | inode = ERR_PTR(ret); |
01cd3367 | 5978 | } |
c71bf099 YZ |
5979 | } |
5980 | ||
3de4586c CM |
5981 | return inode; |
5982 | } | |
5983 | ||
fe15ce44 | 5984 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5985 | { |
5986 | struct btrfs_root *root; | |
2b0143b5 | 5987 | struct inode *inode = d_inode(dentry); |
76dda93c | 5988 | |
848cce0d | 5989 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5990 | inode = d_inode(dentry->d_parent); |
76dda93c | 5991 | |
848cce0d LZ |
5992 | if (inode) { |
5993 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5994 | if (btrfs_root_refs(&root->root_item) == 0) |
5995 | return 1; | |
848cce0d | 5996 | |
4a0cc7ca | 5997 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5998 | return 1; |
efefb143 | 5999 | } |
76dda93c YZ |
6000 | return 0; |
6001 | } | |
6002 | ||
3de4586c | 6003 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 6004 | unsigned int flags) |
3de4586c | 6005 | { |
5662344b | 6006 | struct inode *inode; |
a66e7cc6 | 6007 | |
5662344b TI |
6008 | inode = btrfs_lookup_dentry(dir, dentry); |
6009 | if (IS_ERR(inode)) { | |
6010 | if (PTR_ERR(inode) == -ENOENT) | |
6011 | inode = NULL; | |
6012 | else | |
6013 | return ERR_CAST(inode); | |
6014 | } | |
6015 | ||
41d28bca | 6016 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
6017 | } |
6018 | ||
16cdcec7 | 6019 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
6020 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
6021 | }; | |
6022 | ||
23b5ec74 JB |
6023 | /* |
6024 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
6025 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
6026 | * our information into that, and then dir_emit from the buffer. This is | |
6027 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
6028 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
6029 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
6030 | * tree lock. | |
6031 | */ | |
6032 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
6033 | { | |
6034 | struct btrfs_file_private *private; | |
6035 | ||
6036 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
6037 | if (!private) | |
6038 | return -ENOMEM; | |
6039 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
6040 | if (!private->filldir_buf) { | |
6041 | kfree(private); | |
6042 | return -ENOMEM; | |
6043 | } | |
6044 | file->private_data = private; | |
6045 | return 0; | |
6046 | } | |
6047 | ||
6048 | struct dir_entry { | |
6049 | u64 ino; | |
6050 | u64 offset; | |
6051 | unsigned type; | |
6052 | int name_len; | |
6053 | }; | |
6054 | ||
6055 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
6056 | { | |
6057 | while (entries--) { | |
6058 | struct dir_entry *entry = addr; | |
6059 | char *name = (char *)(entry + 1); | |
6060 | ||
92d32170 DS |
6061 | ctx->pos = get_unaligned(&entry->offset); |
6062 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
6063 | get_unaligned(&entry->ino), | |
6064 | get_unaligned(&entry->type))) | |
23b5ec74 | 6065 | return 1; |
92d32170 DS |
6066 | addr += sizeof(struct dir_entry) + |
6067 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
6068 | ctx->pos++; |
6069 | } | |
6070 | return 0; | |
6071 | } | |
6072 | ||
9cdda8d3 | 6073 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 6074 | { |
9cdda8d3 | 6075 | struct inode *inode = file_inode(file); |
39279cc3 | 6076 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 6077 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
6078 | struct btrfs_dir_item *di; |
6079 | struct btrfs_key key; | |
5f39d397 | 6080 | struct btrfs_key found_key; |
39279cc3 | 6081 | struct btrfs_path *path; |
23b5ec74 | 6082 | void *addr; |
16cdcec7 MX |
6083 | struct list_head ins_list; |
6084 | struct list_head del_list; | |
39279cc3 | 6085 | int ret; |
5f39d397 | 6086 | struct extent_buffer *leaf; |
39279cc3 | 6087 | int slot; |
5f39d397 CM |
6088 | char *name_ptr; |
6089 | int name_len; | |
23b5ec74 JB |
6090 | int entries = 0; |
6091 | int total_len = 0; | |
02dbfc99 | 6092 | bool put = false; |
c2951f32 | 6093 | struct btrfs_key location; |
5f39d397 | 6094 | |
9cdda8d3 AV |
6095 | if (!dir_emit_dots(file, ctx)) |
6096 | return 0; | |
6097 | ||
49593bfa | 6098 | path = btrfs_alloc_path(); |
16cdcec7 MX |
6099 | if (!path) |
6100 | return -ENOMEM; | |
ff5714cc | 6101 | |
23b5ec74 | 6102 | addr = private->filldir_buf; |
e4058b54 | 6103 | path->reada = READA_FORWARD; |
49593bfa | 6104 | |
c2951f32 JM |
6105 | INIT_LIST_HEAD(&ins_list); |
6106 | INIT_LIST_HEAD(&del_list); | |
6107 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 6108 | |
23b5ec74 | 6109 | again: |
c2951f32 | 6110 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 6111 | key.offset = ctx->pos; |
4a0cc7ca | 6112 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 6113 | |
39279cc3 CM |
6114 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6115 | if (ret < 0) | |
6116 | goto err; | |
49593bfa DW |
6117 | |
6118 | while (1) { | |
23b5ec74 JB |
6119 | struct dir_entry *entry; |
6120 | ||
5f39d397 | 6121 | leaf = path->nodes[0]; |
39279cc3 | 6122 | slot = path->slots[0]; |
b9e03af0 LZ |
6123 | if (slot >= btrfs_header_nritems(leaf)) { |
6124 | ret = btrfs_next_leaf(root, path); | |
6125 | if (ret < 0) | |
6126 | goto err; | |
6127 | else if (ret > 0) | |
6128 | break; | |
6129 | continue; | |
39279cc3 | 6130 | } |
3de4586c | 6131 | |
5f39d397 CM |
6132 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
6133 | ||
6134 | if (found_key.objectid != key.objectid) | |
39279cc3 | 6135 | break; |
c2951f32 | 6136 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 6137 | break; |
9cdda8d3 | 6138 | if (found_key.offset < ctx->pos) |
b9e03af0 | 6139 | goto next; |
c2951f32 | 6140 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 6141 | goto next; |
39279cc3 | 6142 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
c2951f32 | 6143 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
6144 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
6145 | PAGE_SIZE) { | |
6146 | btrfs_release_path(path); | |
6147 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6148 | if (ret) | |
6149 | goto nopos; | |
6150 | addr = private->filldir_buf; | |
6151 | entries = 0; | |
6152 | total_len = 0; | |
6153 | goto again; | |
c2951f32 | 6154 | } |
23b5ec74 JB |
6155 | |
6156 | entry = addr; | |
92d32170 | 6157 | put_unaligned(name_len, &entry->name_len); |
23b5ec74 | 6158 | name_ptr = (char *)(entry + 1); |
c2951f32 JM |
6159 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
6160 | name_len); | |
92d32170 DS |
6161 | put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)], |
6162 | &entry->type); | |
c2951f32 | 6163 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
92d32170 DS |
6164 | put_unaligned(location.objectid, &entry->ino); |
6165 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
6166 | entries++; |
6167 | addr += sizeof(struct dir_entry) + name_len; | |
6168 | total_len += sizeof(struct dir_entry) + name_len; | |
b9e03af0 LZ |
6169 | next: |
6170 | path->slots[0]++; | |
39279cc3 | 6171 | } |
23b5ec74 JB |
6172 | btrfs_release_path(path); |
6173 | ||
6174 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6175 | if (ret) | |
6176 | goto nopos; | |
49593bfa | 6177 | |
d2fbb2b5 | 6178 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 6179 | if (ret) |
bc4ef759 DS |
6180 | goto nopos; |
6181 | ||
db62efbb ZB |
6182 | /* |
6183 | * Stop new entries from being returned after we return the last | |
6184 | * entry. | |
6185 | * | |
6186 | * New directory entries are assigned a strictly increasing | |
6187 | * offset. This means that new entries created during readdir | |
6188 | * are *guaranteed* to be seen in the future by that readdir. | |
6189 | * This has broken buggy programs which operate on names as | |
6190 | * they're returned by readdir. Until we re-use freed offsets | |
6191 | * we have this hack to stop new entries from being returned | |
6192 | * under the assumption that they'll never reach this huge | |
6193 | * offset. | |
6194 | * | |
6195 | * This is being careful not to overflow 32bit loff_t unless the | |
6196 | * last entry requires it because doing so has broken 32bit apps | |
6197 | * in the past. | |
6198 | */ | |
c2951f32 JM |
6199 | if (ctx->pos >= INT_MAX) |
6200 | ctx->pos = LLONG_MAX; | |
6201 | else | |
6202 | ctx->pos = INT_MAX; | |
39279cc3 CM |
6203 | nopos: |
6204 | ret = 0; | |
6205 | err: | |
02dbfc99 OS |
6206 | if (put) |
6207 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 6208 | btrfs_free_path(path); |
39279cc3 CM |
6209 | return ret; |
6210 | } | |
6211 | ||
a9185b41 | 6212 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
6213 | { |
6214 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6215 | struct btrfs_trans_handle *trans; | |
6216 | int ret = 0; | |
0af3d00b | 6217 | bool nolock = false; |
39279cc3 | 6218 | |
72ac3c0d | 6219 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
6220 | return 0; |
6221 | ||
70ddc553 NB |
6222 | if (btrfs_fs_closing(root->fs_info) && |
6223 | btrfs_is_free_space_inode(BTRFS_I(inode))) | |
82d5902d | 6224 | nolock = true; |
0af3d00b | 6225 | |
a9185b41 | 6226 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 6227 | if (nolock) |
7a7eaa40 | 6228 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 6229 | else |
7a7eaa40 | 6230 | trans = btrfs_join_transaction(root); |
3612b495 TI |
6231 | if (IS_ERR(trans)) |
6232 | return PTR_ERR(trans); | |
3a45bb20 | 6233 | ret = btrfs_commit_transaction(trans); |
39279cc3 CM |
6234 | } |
6235 | return ret; | |
6236 | } | |
6237 | ||
6238 | /* | |
54aa1f4d | 6239 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
6240 | * inode changes. But, it is most likely to find the inode in cache. |
6241 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
6242 | * to keep or drop this code. | |
6243 | */ | |
48a3b636 | 6244 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 6245 | { |
2ff7e61e | 6246 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
6247 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6248 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
6249 | int ret; |
6250 | ||
72ac3c0d | 6251 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 6252 | return 0; |
39279cc3 | 6253 | |
7a7eaa40 | 6254 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
6255 | if (IS_ERR(trans)) |
6256 | return PTR_ERR(trans); | |
8929ecfa YZ |
6257 | |
6258 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
6259 | if (ret && ret == -ENOSPC) { |
6260 | /* whoops, lets try again with the full transaction */ | |
3a45bb20 | 6261 | btrfs_end_transaction(trans); |
94b60442 | 6262 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
6263 | if (IS_ERR(trans)) |
6264 | return PTR_ERR(trans); | |
8929ecfa | 6265 | |
94b60442 | 6266 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 6267 | } |
3a45bb20 | 6268 | btrfs_end_transaction(trans); |
16cdcec7 | 6269 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 6270 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6271 | |
6272 | return ret; | |
6273 | } | |
6274 | ||
6275 | /* | |
6276 | * This is a copy of file_update_time. We need this so we can return error on | |
6277 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6278 | */ | |
e41f941a JB |
6279 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
6280 | int flags) | |
22c44fe6 | 6281 | { |
2bc55652 | 6282 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3a8c7231 | 6283 | bool dirty = flags & ~S_VERSION; |
2bc55652 AB |
6284 | |
6285 | if (btrfs_root_readonly(root)) | |
6286 | return -EROFS; | |
6287 | ||
e41f941a | 6288 | if (flags & S_VERSION) |
3a8c7231 | 6289 | dirty |= inode_maybe_inc_iversion(inode, dirty); |
e41f941a JB |
6290 | if (flags & S_CTIME) |
6291 | inode->i_ctime = *now; | |
6292 | if (flags & S_MTIME) | |
6293 | inode->i_mtime = *now; | |
6294 | if (flags & S_ATIME) | |
6295 | inode->i_atime = *now; | |
3a8c7231 | 6296 | return dirty ? btrfs_dirty_inode(inode) : 0; |
39279cc3 CM |
6297 | } |
6298 | ||
d352ac68 CM |
6299 | /* |
6300 | * find the highest existing sequence number in a directory | |
6301 | * and then set the in-memory index_cnt variable to reflect | |
6302 | * free sequence numbers | |
6303 | */ | |
4c570655 | 6304 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 6305 | { |
4c570655 | 6306 | struct btrfs_root *root = inode->root; |
aec7477b JB |
6307 | struct btrfs_key key, found_key; |
6308 | struct btrfs_path *path; | |
6309 | struct extent_buffer *leaf; | |
6310 | int ret; | |
6311 | ||
4c570655 | 6312 | key.objectid = btrfs_ino(inode); |
962a298f | 6313 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6314 | key.offset = (u64)-1; |
6315 | ||
6316 | path = btrfs_alloc_path(); | |
6317 | if (!path) | |
6318 | return -ENOMEM; | |
6319 | ||
6320 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6321 | if (ret < 0) | |
6322 | goto out; | |
6323 | /* FIXME: we should be able to handle this */ | |
6324 | if (ret == 0) | |
6325 | goto out; | |
6326 | ret = 0; | |
6327 | ||
6328 | /* | |
6329 | * MAGIC NUMBER EXPLANATION: | |
6330 | * since we search a directory based on f_pos we have to start at 2 | |
6331 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6332 | * else has to start at 2 | |
6333 | */ | |
6334 | if (path->slots[0] == 0) { | |
4c570655 | 6335 | inode->index_cnt = 2; |
aec7477b JB |
6336 | goto out; |
6337 | } | |
6338 | ||
6339 | path->slots[0]--; | |
6340 | ||
6341 | leaf = path->nodes[0]; | |
6342 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6343 | ||
4c570655 | 6344 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6345 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
4c570655 | 6346 | inode->index_cnt = 2; |
aec7477b JB |
6347 | goto out; |
6348 | } | |
6349 | ||
4c570655 | 6350 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6351 | out: |
6352 | btrfs_free_path(path); | |
6353 | return ret; | |
6354 | } | |
6355 | ||
d352ac68 CM |
6356 | /* |
6357 | * helper to find a free sequence number in a given directory. This current | |
6358 | * code is very simple, later versions will do smarter things in the btree | |
6359 | */ | |
877574e2 | 6360 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6361 | { |
6362 | int ret = 0; | |
6363 | ||
877574e2 NB |
6364 | if (dir->index_cnt == (u64)-1) { |
6365 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6366 | if (ret) { |
6367 | ret = btrfs_set_inode_index_count(dir); | |
6368 | if (ret) | |
6369 | return ret; | |
6370 | } | |
aec7477b JB |
6371 | } |
6372 | ||
877574e2 NB |
6373 | *index = dir->index_cnt; |
6374 | dir->index_cnt++; | |
aec7477b JB |
6375 | |
6376 | return ret; | |
6377 | } | |
6378 | ||
b0d5d10f CM |
6379 | static int btrfs_insert_inode_locked(struct inode *inode) |
6380 | { | |
6381 | struct btrfs_iget_args args; | |
6382 | args.location = &BTRFS_I(inode)->location; | |
6383 | args.root = BTRFS_I(inode)->root; | |
6384 | ||
6385 | return insert_inode_locked4(inode, | |
6386 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6387 | btrfs_find_actor, &args); | |
6388 | } | |
6389 | ||
19aee8de AJ |
6390 | /* |
6391 | * Inherit flags from the parent inode. | |
6392 | * | |
6393 | * Currently only the compression flags and the cow flags are inherited. | |
6394 | */ | |
6395 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
6396 | { | |
6397 | unsigned int flags; | |
6398 | ||
6399 | if (!dir) | |
6400 | return; | |
6401 | ||
6402 | flags = BTRFS_I(dir)->flags; | |
6403 | ||
6404 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
6405 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
6406 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
6407 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
6408 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
6409 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
6410 | } | |
6411 | ||
6412 | if (flags & BTRFS_INODE_NODATACOW) { | |
6413 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
6414 | if (S_ISREG(inode->i_mode)) | |
6415 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6416 | } | |
6417 | ||
7b6a221e | 6418 | btrfs_sync_inode_flags_to_i_flags(inode); |
19aee8de AJ |
6419 | } |
6420 | ||
39279cc3 CM |
6421 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6422 | struct btrfs_root *root, | |
aec7477b | 6423 | struct inode *dir, |
9c58309d | 6424 | const char *name, int name_len, |
175a4eb7 AV |
6425 | u64 ref_objectid, u64 objectid, |
6426 | umode_t mode, u64 *index) | |
39279cc3 | 6427 | { |
0b246afa | 6428 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6429 | struct inode *inode; |
5f39d397 | 6430 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6431 | struct btrfs_key *location; |
5f39d397 | 6432 | struct btrfs_path *path; |
9c58309d CM |
6433 | struct btrfs_inode_ref *ref; |
6434 | struct btrfs_key key[2]; | |
6435 | u32 sizes[2]; | |
ef3b9af5 | 6436 | int nitems = name ? 2 : 1; |
9c58309d | 6437 | unsigned long ptr; |
39279cc3 | 6438 | int ret; |
39279cc3 | 6439 | |
5f39d397 | 6440 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6441 | if (!path) |
6442 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6443 | |
0b246afa | 6444 | inode = new_inode(fs_info->sb); |
8fb27640 YS |
6445 | if (!inode) { |
6446 | btrfs_free_path(path); | |
39279cc3 | 6447 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6448 | } |
39279cc3 | 6449 | |
5762b5c9 FM |
6450 | /* |
6451 | * O_TMPFILE, set link count to 0, so that after this point, | |
6452 | * we fill in an inode item with the correct link count. | |
6453 | */ | |
6454 | if (!name) | |
6455 | set_nlink(inode, 0); | |
6456 | ||
581bb050 LZ |
6457 | /* |
6458 | * we have to initialize this early, so we can reclaim the inode | |
6459 | * number if we fail afterwards in this function. | |
6460 | */ | |
6461 | inode->i_ino = objectid; | |
6462 | ||
ef3b9af5 | 6463 | if (dir && name) { |
1abe9b8a | 6464 | trace_btrfs_inode_request(dir); |
6465 | ||
877574e2 | 6466 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 6467 | if (ret) { |
8fb27640 | 6468 | btrfs_free_path(path); |
09771430 | 6469 | iput(inode); |
aec7477b | 6470 | return ERR_PTR(ret); |
09771430 | 6471 | } |
ef3b9af5 FM |
6472 | } else if (dir) { |
6473 | *index = 0; | |
aec7477b JB |
6474 | } |
6475 | /* | |
6476 | * index_cnt is ignored for everything but a dir, | |
df6703e1 | 6477 | * btrfs_set_inode_index_count has an explanation for the magic |
aec7477b JB |
6478 | * number |
6479 | */ | |
6480 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6481 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6482 | BTRFS_I(inode)->root = root; |
e02119d5 | 6483 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6484 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6485 | |
5dc562c5 JB |
6486 | /* |
6487 | * We could have gotten an inode number from somebody who was fsynced | |
6488 | * and then removed in this same transaction, so let's just set full | |
6489 | * sync since it will be a full sync anyway and this will blow away the | |
6490 | * old info in the log. | |
6491 | */ | |
6492 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6493 | ||
9c58309d | 6494 | key[0].objectid = objectid; |
962a298f | 6495 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6496 | key[0].offset = 0; |
6497 | ||
9c58309d | 6498 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6499 | |
6500 | if (name) { | |
6501 | /* | |
6502 | * Start new inodes with an inode_ref. This is slightly more | |
6503 | * efficient for small numbers of hard links since they will | |
6504 | * be packed into one item. Extended refs will kick in if we | |
6505 | * add more hard links than can fit in the ref item. | |
6506 | */ | |
6507 | key[1].objectid = objectid; | |
962a298f | 6508 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6509 | key[1].offset = ref_objectid; |
6510 | ||
6511 | sizes[1] = name_len + sizeof(*ref); | |
6512 | } | |
9c58309d | 6513 | |
b0d5d10f CM |
6514 | location = &BTRFS_I(inode)->location; |
6515 | location->objectid = objectid; | |
6516 | location->offset = 0; | |
962a298f | 6517 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6518 | |
6519 | ret = btrfs_insert_inode_locked(inode); | |
6520 | if (ret < 0) | |
6521 | goto fail; | |
6522 | ||
b9473439 | 6523 | path->leave_spinning = 1; |
ef3b9af5 | 6524 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6525 | if (ret != 0) |
b0d5d10f | 6526 | goto fail_unlock; |
5f39d397 | 6527 | |
ecc11fab | 6528 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6529 | inode_set_bytes(inode, 0); |
9cc97d64 | 6530 | |
c2050a45 | 6531 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6532 | inode->i_atime = inode->i_mtime; |
6533 | inode->i_ctime = inode->i_mtime; | |
6534 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
6535 | ||
5f39d397 CM |
6536 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6537 | struct btrfs_inode_item); | |
b159fa28 | 6538 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6539 | sizeof(*inode_item)); |
e02119d5 | 6540 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6541 | |
ef3b9af5 FM |
6542 | if (name) { |
6543 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6544 | struct btrfs_inode_ref); | |
6545 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6546 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6547 | ptr = (unsigned long)(ref + 1); | |
6548 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6549 | } | |
9c58309d | 6550 | |
5f39d397 CM |
6551 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6552 | btrfs_free_path(path); | |
6553 | ||
6cbff00f CH |
6554 | btrfs_inherit_iflags(inode, dir); |
6555 | ||
569254b0 | 6556 | if (S_ISREG(mode)) { |
0b246afa | 6557 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6558 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6559 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6560 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6561 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6562 | } |
6563 | ||
5d4f98a2 | 6564 | inode_tree_add(inode); |
1abe9b8a | 6565 | |
6566 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6567 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6568 | |
8ea05e3a AB |
6569 | btrfs_update_root_times(trans, root); |
6570 | ||
63541927 FDBM |
6571 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6572 | if (ret) | |
0b246afa | 6573 | btrfs_err(fs_info, |
63541927 | 6574 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6575 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6576 | |
39279cc3 | 6577 | return inode; |
b0d5d10f CM |
6578 | |
6579 | fail_unlock: | |
6580 | unlock_new_inode(inode); | |
5f39d397 | 6581 | fail: |
ef3b9af5 | 6582 | if (dir && name) |
aec7477b | 6583 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 6584 | btrfs_free_path(path); |
09771430 | 6585 | iput(inode); |
5f39d397 | 6586 | return ERR_PTR(ret); |
39279cc3 CM |
6587 | } |
6588 | ||
6589 | static inline u8 btrfs_inode_type(struct inode *inode) | |
6590 | { | |
6591 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
6592 | } | |
6593 | ||
d352ac68 CM |
6594 | /* |
6595 | * utility function to add 'inode' into 'parent_inode' with | |
6596 | * a give name and a given sequence number. | |
6597 | * if 'add_backref' is true, also insert a backref from the | |
6598 | * inode to the parent directory. | |
6599 | */ | |
e02119d5 | 6600 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6601 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6602 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6603 | { |
db0a669f | 6604 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
4df27c4d | 6605 | int ret = 0; |
39279cc3 | 6606 | struct btrfs_key key; |
db0a669f NB |
6607 | struct btrfs_root *root = parent_inode->root; |
6608 | u64 ino = btrfs_ino(inode); | |
6609 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6610 | |
33345d01 | 6611 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6612 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6613 | } else { |
33345d01 | 6614 | key.objectid = ino; |
962a298f | 6615 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6616 | key.offset = 0; |
6617 | } | |
6618 | ||
33345d01 | 6619 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
0b246afa JM |
6620 | ret = btrfs_add_root_ref(trans, fs_info, key.objectid, |
6621 | root->root_key.objectid, parent_ino, | |
6622 | index, name, name_len); | |
4df27c4d | 6623 | } else if (add_backref) { |
33345d01 LZ |
6624 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6625 | parent_ino, index); | |
4df27c4d | 6626 | } |
39279cc3 | 6627 | |
79787eaa JM |
6628 | /* Nothing to clean up yet */ |
6629 | if (ret) | |
6630 | return ret; | |
4df27c4d | 6631 | |
79787eaa JM |
6632 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
6633 | parent_inode, &key, | |
db0a669f | 6634 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6635 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6636 | goto fail_dir_item; |
6637 | else if (ret) { | |
66642832 | 6638 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6639 | return ret; |
39279cc3 | 6640 | } |
79787eaa | 6641 | |
db0a669f | 6642 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6643 | name_len * 2); |
db0a669f NB |
6644 | inode_inc_iversion(&parent_inode->vfs_inode); |
6645 | parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime = | |
6646 | current_time(&parent_inode->vfs_inode); | |
6647 | ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode); | |
79787eaa | 6648 | if (ret) |
66642832 | 6649 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6650 | return ret; |
fe66a05a CM |
6651 | |
6652 | fail_dir_item: | |
6653 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6654 | u64 local_index; | |
6655 | int err; | |
0b246afa JM |
6656 | err = btrfs_del_root_ref(trans, fs_info, key.objectid, |
6657 | root->root_key.objectid, parent_ino, | |
6658 | &local_index, name, name_len); | |
fe66a05a CM |
6659 | |
6660 | } else if (add_backref) { | |
6661 | u64 local_index; | |
6662 | int err; | |
6663 | ||
6664 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6665 | ino, parent_ino, &local_index); | |
6666 | } | |
6667 | return ret; | |
39279cc3 CM |
6668 | } |
6669 | ||
6670 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
cef415af NB |
6671 | struct btrfs_inode *dir, struct dentry *dentry, |
6672 | struct btrfs_inode *inode, int backref, u64 index) | |
39279cc3 | 6673 | { |
a1b075d2 JB |
6674 | int err = btrfs_add_link(trans, dir, inode, |
6675 | dentry->d_name.name, dentry->d_name.len, | |
6676 | backref, index); | |
39279cc3 CM |
6677 | if (err > 0) |
6678 | err = -EEXIST; | |
6679 | return err; | |
6680 | } | |
6681 | ||
618e21d5 | 6682 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6683 | umode_t mode, dev_t rdev) |
618e21d5 | 6684 | { |
2ff7e61e | 6685 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6686 | struct btrfs_trans_handle *trans; |
6687 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6688 | struct inode *inode = NULL; |
618e21d5 JB |
6689 | int err; |
6690 | int drop_inode = 0; | |
6691 | u64 objectid; | |
00e4e6b3 | 6692 | u64 index = 0; |
618e21d5 | 6693 | |
9ed74f2d JB |
6694 | /* |
6695 | * 2 for inode item and ref | |
6696 | * 2 for dir items | |
6697 | * 1 for xattr if selinux is on | |
6698 | */ | |
a22285a6 YZ |
6699 | trans = btrfs_start_transaction(root, 5); |
6700 | if (IS_ERR(trans)) | |
6701 | return PTR_ERR(trans); | |
1832a6d5 | 6702 | |
581bb050 LZ |
6703 | err = btrfs_find_free_ino(root, &objectid); |
6704 | if (err) | |
6705 | goto out_unlock; | |
6706 | ||
aec7477b | 6707 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6708 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6709 | mode, &index); | |
7cf96da3 TI |
6710 | if (IS_ERR(inode)) { |
6711 | err = PTR_ERR(inode); | |
618e21d5 | 6712 | goto out_unlock; |
7cf96da3 | 6713 | } |
618e21d5 | 6714 | |
ad19db71 CS |
6715 | /* |
6716 | * If the active LSM wants to access the inode during | |
6717 | * d_instantiate it needs these. Smack checks to see | |
6718 | * if the filesystem supports xattrs by looking at the | |
6719 | * ops vector. | |
6720 | */ | |
ad19db71 | 6721 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6722 | init_special_inode(inode, inode->i_mode, rdev); |
6723 | ||
6724 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6725 | if (err) |
b0d5d10f CM |
6726 | goto out_unlock_inode; |
6727 | ||
cef415af NB |
6728 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6729 | 0, index); | |
b0d5d10f CM |
6730 | if (err) { |
6731 | goto out_unlock_inode; | |
6732 | } else { | |
1b4ab1bb | 6733 | btrfs_update_inode(trans, root, inode); |
1e2e547a | 6734 | d_instantiate_new(dentry, inode); |
618e21d5 | 6735 | } |
b0d5d10f | 6736 | |
618e21d5 | 6737 | out_unlock: |
3a45bb20 | 6738 | btrfs_end_transaction(trans); |
2ff7e61e | 6739 | btrfs_btree_balance_dirty(fs_info); |
618e21d5 JB |
6740 | if (drop_inode) { |
6741 | inode_dec_link_count(inode); | |
6742 | iput(inode); | |
6743 | } | |
618e21d5 | 6744 | return err; |
b0d5d10f CM |
6745 | |
6746 | out_unlock_inode: | |
6747 | drop_inode = 1; | |
6748 | unlock_new_inode(inode); | |
6749 | goto out_unlock; | |
6750 | ||
618e21d5 JB |
6751 | } |
6752 | ||
39279cc3 | 6753 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6754 | umode_t mode, bool excl) |
39279cc3 | 6755 | { |
2ff7e61e | 6756 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6757 | struct btrfs_trans_handle *trans; |
6758 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6759 | struct inode *inode = NULL; |
43baa579 | 6760 | int drop_inode_on_err = 0; |
a22285a6 | 6761 | int err; |
39279cc3 | 6762 | u64 objectid; |
00e4e6b3 | 6763 | u64 index = 0; |
39279cc3 | 6764 | |
9ed74f2d JB |
6765 | /* |
6766 | * 2 for inode item and ref | |
6767 | * 2 for dir items | |
6768 | * 1 for xattr if selinux is on | |
6769 | */ | |
a22285a6 YZ |
6770 | trans = btrfs_start_transaction(root, 5); |
6771 | if (IS_ERR(trans)) | |
6772 | return PTR_ERR(trans); | |
9ed74f2d | 6773 | |
581bb050 LZ |
6774 | err = btrfs_find_free_ino(root, &objectid); |
6775 | if (err) | |
6776 | goto out_unlock; | |
6777 | ||
aec7477b | 6778 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6779 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6780 | mode, &index); | |
7cf96da3 TI |
6781 | if (IS_ERR(inode)) { |
6782 | err = PTR_ERR(inode); | |
39279cc3 | 6783 | goto out_unlock; |
7cf96da3 | 6784 | } |
43baa579 | 6785 | drop_inode_on_err = 1; |
ad19db71 CS |
6786 | /* |
6787 | * If the active LSM wants to access the inode during | |
6788 | * d_instantiate it needs these. Smack checks to see | |
6789 | * if the filesystem supports xattrs by looking at the | |
6790 | * ops vector. | |
6791 | */ | |
6792 | inode->i_fop = &btrfs_file_operations; | |
6793 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6794 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6795 | |
6796 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6797 | if (err) | |
6798 | goto out_unlock_inode; | |
6799 | ||
6800 | err = btrfs_update_inode(trans, root, inode); | |
6801 | if (err) | |
6802 | goto out_unlock_inode; | |
ad19db71 | 6803 | |
cef415af NB |
6804 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6805 | 0, index); | |
39279cc3 | 6806 | if (err) |
b0d5d10f | 6807 | goto out_unlock_inode; |
43baa579 | 6808 | |
43baa579 | 6809 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
1e2e547a | 6810 | d_instantiate_new(dentry, inode); |
43baa579 | 6811 | |
39279cc3 | 6812 | out_unlock: |
3a45bb20 | 6813 | btrfs_end_transaction(trans); |
43baa579 | 6814 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6815 | inode_dec_link_count(inode); |
6816 | iput(inode); | |
6817 | } | |
2ff7e61e | 6818 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 6819 | return err; |
b0d5d10f CM |
6820 | |
6821 | out_unlock_inode: | |
6822 | unlock_new_inode(inode); | |
6823 | goto out_unlock; | |
6824 | ||
39279cc3 CM |
6825 | } |
6826 | ||
6827 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6828 | struct dentry *dentry) | |
6829 | { | |
271dba45 | 6830 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6831 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6832 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6833 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6834 | u64 index; |
39279cc3 CM |
6835 | int err; |
6836 | int drop_inode = 0; | |
6837 | ||
4a8be425 TH |
6838 | /* do not allow sys_link's with other subvols of the same device */ |
6839 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6840 | return -EXDEV; |
4a8be425 | 6841 | |
f186373f | 6842 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6843 | return -EMLINK; |
4a8be425 | 6844 | |
877574e2 | 6845 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6846 | if (err) |
6847 | goto fail; | |
6848 | ||
a22285a6 | 6849 | /* |
7e6b6465 | 6850 | * 2 items for inode and inode ref |
a22285a6 | 6851 | * 2 items for dir items |
7e6b6465 | 6852 | * 1 item for parent inode |
a22285a6 | 6853 | */ |
7e6b6465 | 6854 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6855 | if (IS_ERR(trans)) { |
6856 | err = PTR_ERR(trans); | |
271dba45 | 6857 | trans = NULL; |
a22285a6 YZ |
6858 | goto fail; |
6859 | } | |
5f39d397 | 6860 | |
67de1176 MX |
6861 | /* There are several dir indexes for this inode, clear the cache. */ |
6862 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6863 | inc_nlink(inode); |
0c4d2d95 | 6864 | inode_inc_iversion(inode); |
c2050a45 | 6865 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6866 | ihold(inode); |
e9976151 | 6867 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6868 | |
cef415af NB |
6869 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6870 | 1, index); | |
5f39d397 | 6871 | |
a5719521 | 6872 | if (err) { |
54aa1f4d | 6873 | drop_inode = 1; |
a5719521 | 6874 | } else { |
10d9f309 | 6875 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6876 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6877 | if (err) |
6878 | goto fail; | |
ef3b9af5 FM |
6879 | if (inode->i_nlink == 1) { |
6880 | /* | |
6881 | * If new hard link count is 1, it's a file created | |
6882 | * with open(2) O_TMPFILE flag. | |
6883 | */ | |
3d6ae7bb | 6884 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6885 | if (err) |
6886 | goto fail; | |
6887 | } | |
08c422c2 | 6888 | d_instantiate(dentry, inode); |
9ca5fbfb | 6889 | btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent); |
a5719521 | 6890 | } |
39279cc3 | 6891 | |
1832a6d5 | 6892 | fail: |
271dba45 | 6893 | if (trans) |
3a45bb20 | 6894 | btrfs_end_transaction(trans); |
39279cc3 CM |
6895 | if (drop_inode) { |
6896 | inode_dec_link_count(inode); | |
6897 | iput(inode); | |
6898 | } | |
2ff7e61e | 6899 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6900 | return err; |
6901 | } | |
6902 | ||
18bb1db3 | 6903 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6904 | { |
2ff7e61e | 6905 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6906 | struct inode *inode = NULL; |
39279cc3 CM |
6907 | struct btrfs_trans_handle *trans; |
6908 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6909 | int err = 0; | |
6910 | int drop_on_err = 0; | |
b9d86667 | 6911 | u64 objectid = 0; |
00e4e6b3 | 6912 | u64 index = 0; |
39279cc3 | 6913 | |
9ed74f2d JB |
6914 | /* |
6915 | * 2 items for inode and ref | |
6916 | * 2 items for dir items | |
6917 | * 1 for xattr if selinux is on | |
6918 | */ | |
a22285a6 YZ |
6919 | trans = btrfs_start_transaction(root, 5); |
6920 | if (IS_ERR(trans)) | |
6921 | return PTR_ERR(trans); | |
39279cc3 | 6922 | |
581bb050 LZ |
6923 | err = btrfs_find_free_ino(root, &objectid); |
6924 | if (err) | |
6925 | goto out_fail; | |
6926 | ||
aec7477b | 6927 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6928 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6929 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6930 | if (IS_ERR(inode)) { |
6931 | err = PTR_ERR(inode); | |
6932 | goto out_fail; | |
6933 | } | |
5f39d397 | 6934 | |
39279cc3 | 6935 | drop_on_err = 1; |
b0d5d10f CM |
6936 | /* these must be set before we unlock the inode */ |
6937 | inode->i_op = &btrfs_dir_inode_operations; | |
6938 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6939 | |
2a7dba39 | 6940 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6941 | if (err) |
b0d5d10f | 6942 | goto out_fail_inode; |
39279cc3 | 6943 | |
6ef06d27 | 6944 | btrfs_i_size_write(BTRFS_I(inode), 0); |
39279cc3 CM |
6945 | err = btrfs_update_inode(trans, root, inode); |
6946 | if (err) | |
b0d5d10f | 6947 | goto out_fail_inode; |
5f39d397 | 6948 | |
db0a669f NB |
6949 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6950 | dentry->d_name.name, | |
6951 | dentry->d_name.len, 0, index); | |
39279cc3 | 6952 | if (err) |
b0d5d10f | 6953 | goto out_fail_inode; |
5f39d397 | 6954 | |
1e2e547a | 6955 | d_instantiate_new(dentry, inode); |
39279cc3 | 6956 | drop_on_err = 0; |
39279cc3 CM |
6957 | |
6958 | out_fail: | |
3a45bb20 | 6959 | btrfs_end_transaction(trans); |
c7cfb8a5 WS |
6960 | if (drop_on_err) { |
6961 | inode_dec_link_count(inode); | |
39279cc3 | 6962 | iput(inode); |
c7cfb8a5 | 6963 | } |
2ff7e61e | 6964 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 6965 | return err; |
b0d5d10f CM |
6966 | |
6967 | out_fail_inode: | |
6968 | unlock_new_inode(inode); | |
6969 | goto out_fail; | |
39279cc3 CM |
6970 | } |
6971 | ||
c8b97818 | 6972 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6973 | struct page *page, |
c8b97818 CM |
6974 | size_t pg_offset, u64 extent_offset, |
6975 | struct btrfs_file_extent_item *item) | |
6976 | { | |
6977 | int ret; | |
6978 | struct extent_buffer *leaf = path->nodes[0]; | |
6979 | char *tmp; | |
6980 | size_t max_size; | |
6981 | unsigned long inline_size; | |
6982 | unsigned long ptr; | |
261507a0 | 6983 | int compress_type; |
c8b97818 CM |
6984 | |
6985 | WARN_ON(pg_offset != 0); | |
261507a0 | 6986 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6987 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6988 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6989 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6990 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6991 | if (!tmp) |
6992 | return -ENOMEM; | |
c8b97818 CM |
6993 | ptr = btrfs_file_extent_inline_start(item); |
6994 | ||
6995 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6996 | ||
09cbfeaf | 6997 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6998 | ret = btrfs_decompress(compress_type, tmp, page, |
6999 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
7000 | |
7001 | /* | |
7002 | * decompression code contains a memset to fill in any space between the end | |
7003 | * of the uncompressed data and the end of max_size in case the decompressed | |
7004 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
7005 | * the end of an inline extent and the beginning of the next block, so we | |
7006 | * cover that region here. | |
7007 | */ | |
7008 | ||
7009 | if (max_size + pg_offset < PAGE_SIZE) { | |
7010 | char *map = kmap(page); | |
7011 | memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset); | |
7012 | kunmap(page); | |
7013 | } | |
c8b97818 | 7014 | kfree(tmp); |
166ae5a4 | 7015 | return ret; |
c8b97818 CM |
7016 | } |
7017 | ||
d352ac68 CM |
7018 | /* |
7019 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
7020 | * the ugly parts come from merging extents from the disk with the in-ram |
7021 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
7022 | * where the in-ram extents might be locked pending data=ordered completion. |
7023 | * | |
7024 | * This also copies inline extents directly into the page. | |
7025 | */ | |
fc4f21b1 NB |
7026 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
7027 | struct page *page, | |
7028 | size_t pg_offset, u64 start, u64 len, | |
7029 | int create) | |
a52d9a80 | 7030 | { |
fc4f21b1 | 7031 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
a52d9a80 CM |
7032 | int ret; |
7033 | int err = 0; | |
a52d9a80 CM |
7034 | u64 extent_start = 0; |
7035 | u64 extent_end = 0; | |
fc4f21b1 | 7036 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 7037 | u32 found_type; |
f421950f | 7038 | struct btrfs_path *path = NULL; |
fc4f21b1 | 7039 | struct btrfs_root *root = inode->root; |
a52d9a80 | 7040 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
7041 | struct extent_buffer *leaf; |
7042 | struct btrfs_key found_key; | |
a52d9a80 | 7043 | struct extent_map *em = NULL; |
fc4f21b1 NB |
7044 | struct extent_map_tree *em_tree = &inode->extent_tree; |
7045 | struct extent_io_tree *io_tree = &inode->io_tree; | |
7ffbb598 | 7046 | const bool new_inline = !page || create; |
a52d9a80 | 7047 | |
890871be | 7048 | read_lock(&em_tree->lock); |
d1310b2e | 7049 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d | 7050 | if (em) |
0b246afa | 7051 | em->bdev = fs_info->fs_devices->latest_bdev; |
890871be | 7052 | read_unlock(&em_tree->lock); |
d1310b2e | 7053 | |
a52d9a80 | 7054 | if (em) { |
e1c4b745 CM |
7055 | if (em->start > start || em->start + em->len <= start) |
7056 | free_extent_map(em); | |
7057 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
7058 | free_extent_map(em); |
7059 | else | |
7060 | goto out; | |
a52d9a80 | 7061 | } |
172ddd60 | 7062 | em = alloc_extent_map(); |
a52d9a80 | 7063 | if (!em) { |
d1310b2e CM |
7064 | err = -ENOMEM; |
7065 | goto out; | |
a52d9a80 | 7066 | } |
0b246afa | 7067 | em->bdev = fs_info->fs_devices->latest_bdev; |
d1310b2e | 7068 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 7069 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 7070 | em->len = (u64)-1; |
c8b97818 | 7071 | em->block_len = (u64)-1; |
f421950f CM |
7072 | |
7073 | if (!path) { | |
7074 | path = btrfs_alloc_path(); | |
026fd317 JB |
7075 | if (!path) { |
7076 | err = -ENOMEM; | |
7077 | goto out; | |
7078 | } | |
7079 | /* | |
7080 | * Chances are we'll be called again, so go ahead and do | |
7081 | * readahead | |
7082 | */ | |
e4058b54 | 7083 | path->reada = READA_FORWARD; |
f421950f CM |
7084 | } |
7085 | ||
5c9a702e | 7086 | ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0); |
a52d9a80 CM |
7087 | if (ret < 0) { |
7088 | err = ret; | |
7089 | goto out; | |
7090 | } | |
7091 | ||
7092 | if (ret != 0) { | |
7093 | if (path->slots[0] == 0) | |
7094 | goto not_found; | |
7095 | path->slots[0]--; | |
7096 | } | |
7097 | ||
5f39d397 CM |
7098 | leaf = path->nodes[0]; |
7099 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 7100 | struct btrfs_file_extent_item); |
a52d9a80 | 7101 | /* are we inside the extent that was found? */ |
5f39d397 | 7102 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 7103 | found_type = found_key.type; |
5f39d397 | 7104 | if (found_key.objectid != objectid || |
a52d9a80 | 7105 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
7106 | /* |
7107 | * If we backup past the first extent we want to move forward | |
7108 | * and see if there is an extent in front of us, otherwise we'll | |
7109 | * say there is a hole for our whole search range which can | |
7110 | * cause problems. | |
7111 | */ | |
7112 | extent_end = start; | |
7113 | goto next; | |
a52d9a80 CM |
7114 | } |
7115 | ||
5f39d397 CM |
7116 | found_type = btrfs_file_extent_type(leaf, item); |
7117 | extent_start = found_key.offset; | |
d899e052 YZ |
7118 | if (found_type == BTRFS_FILE_EXTENT_REG || |
7119 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 7120 | extent_end = extent_start + |
db94535d | 7121 | btrfs_file_extent_num_bytes(leaf, item); |
09ed2f16 LB |
7122 | |
7123 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, | |
7124 | extent_start); | |
9036c102 YZ |
7125 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
7126 | size_t size; | |
514ac8ad | 7127 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
da17066c | 7128 | extent_end = ALIGN(extent_start + size, |
0b246afa | 7129 | fs_info->sectorsize); |
09ed2f16 LB |
7130 | |
7131 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, | |
7132 | path->slots[0], | |
7133 | extent_start); | |
9036c102 | 7134 | } |
25a50341 | 7135 | next: |
9036c102 YZ |
7136 | if (start >= extent_end) { |
7137 | path->slots[0]++; | |
7138 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
7139 | ret = btrfs_next_leaf(root, path); | |
7140 | if (ret < 0) { | |
7141 | err = ret; | |
7142 | goto out; | |
a52d9a80 | 7143 | } |
9036c102 YZ |
7144 | if (ret > 0) |
7145 | goto not_found; | |
7146 | leaf = path->nodes[0]; | |
a52d9a80 | 7147 | } |
9036c102 YZ |
7148 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
7149 | if (found_key.objectid != objectid || | |
7150 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
7151 | goto not_found; | |
7152 | if (start + len <= found_key.offset) | |
7153 | goto not_found; | |
e2eca69d WS |
7154 | if (start > found_key.offset) |
7155 | goto next; | |
9036c102 | 7156 | em->start = start; |
70c8a91c | 7157 | em->orig_start = start; |
9036c102 YZ |
7158 | em->len = found_key.offset - start; |
7159 | goto not_found_em; | |
7160 | } | |
7161 | ||
fc4f21b1 | 7162 | btrfs_extent_item_to_extent_map(inode, path, item, |
9cdc5124 | 7163 | new_inline, em); |
7ffbb598 | 7164 | |
d899e052 YZ |
7165 | if (found_type == BTRFS_FILE_EXTENT_REG || |
7166 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
7167 | goto insert; |
7168 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 7169 | unsigned long ptr; |
a52d9a80 | 7170 | char *map; |
3326d1b0 CM |
7171 | size_t size; |
7172 | size_t extent_offset; | |
7173 | size_t copy_size; | |
a52d9a80 | 7174 | |
7ffbb598 | 7175 | if (new_inline) |
689f9346 | 7176 | goto out; |
5f39d397 | 7177 | |
514ac8ad | 7178 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 7179 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
7180 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
7181 | size - extent_offset); | |
3326d1b0 | 7182 | em->start = extent_start + extent_offset; |
0b246afa | 7183 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 7184 | em->orig_block_len = em->len; |
70c8a91c | 7185 | em->orig_start = em->start; |
689f9346 | 7186 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
bf46f52d | 7187 | if (!PageUptodate(page)) { |
261507a0 LZ |
7188 | if (btrfs_file_extent_compression(leaf, item) != |
7189 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 7190 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 7191 | extent_offset, item); |
166ae5a4 ZB |
7192 | if (ret) { |
7193 | err = ret; | |
7194 | goto out; | |
7195 | } | |
c8b97818 CM |
7196 | } else { |
7197 | map = kmap(page); | |
7198 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
7199 | copy_size); | |
09cbfeaf | 7200 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 7201 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 7202 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
7203 | copy_size); |
7204 | } | |
c8b97818 CM |
7205 | kunmap(page); |
7206 | } | |
179e29e4 | 7207 | flush_dcache_page(page); |
a52d9a80 | 7208 | } |
d1310b2e | 7209 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 7210 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 7211 | goto insert; |
a52d9a80 CM |
7212 | } |
7213 | not_found: | |
7214 | em->start = start; | |
70c8a91c | 7215 | em->orig_start = start; |
d1310b2e | 7216 | em->len = len; |
a52d9a80 | 7217 | not_found_em: |
5f39d397 | 7218 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 | 7219 | insert: |
b3b4aa74 | 7220 | btrfs_release_path(path); |
d1310b2e | 7221 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 7222 | btrfs_err(fs_info, |
5d163e0e JM |
7223 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
7224 | em->start, em->len, start, len); | |
a52d9a80 CM |
7225 | err = -EIO; |
7226 | goto out; | |
7227 | } | |
d1310b2e CM |
7228 | |
7229 | err = 0; | |
890871be | 7230 | write_lock(&em_tree->lock); |
f46b24c9 | 7231 | err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len); |
890871be | 7232 | write_unlock(&em_tree->lock); |
a52d9a80 | 7233 | out: |
1abe9b8a | 7234 | |
fc4f21b1 | 7235 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 7236 | |
527afb44 | 7237 | btrfs_free_path(path); |
a52d9a80 CM |
7238 | if (err) { |
7239 | free_extent_map(em); | |
a52d9a80 CM |
7240 | return ERR_PTR(err); |
7241 | } | |
79787eaa | 7242 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
7243 | return em; |
7244 | } | |
7245 | ||
fc4f21b1 NB |
7246 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
7247 | struct page *page, | |
7248 | size_t pg_offset, u64 start, u64 len, | |
7249 | int create) | |
ec29ed5b CM |
7250 | { |
7251 | struct extent_map *em; | |
7252 | struct extent_map *hole_em = NULL; | |
7253 | u64 range_start = start; | |
7254 | u64 end; | |
7255 | u64 found; | |
7256 | u64 found_end; | |
7257 | int err = 0; | |
7258 | ||
7259 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
7260 | if (IS_ERR(em)) | |
7261 | return em; | |
9986277e DC |
7262 | /* |
7263 | * If our em maps to: | |
7264 | * - a hole or | |
7265 | * - a pre-alloc extent, | |
7266 | * there might actually be delalloc bytes behind it. | |
7267 | */ | |
7268 | if (em->block_start != EXTENT_MAP_HOLE && | |
7269 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7270 | return em; | |
7271 | else | |
7272 | hole_em = em; | |
ec29ed5b CM |
7273 | |
7274 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7275 | end = start + len; | |
7276 | if (end < start) | |
7277 | end = (u64)-1; | |
7278 | else | |
7279 | end -= 1; | |
7280 | ||
7281 | em = NULL; | |
7282 | ||
7283 | /* ok, we didn't find anything, lets look for delalloc */ | |
fc4f21b1 | 7284 | found = count_range_bits(&inode->io_tree, &range_start, |
ec29ed5b CM |
7285 | end, len, EXTENT_DELALLOC, 1); |
7286 | found_end = range_start + found; | |
7287 | if (found_end < range_start) | |
7288 | found_end = (u64)-1; | |
7289 | ||
7290 | /* | |
7291 | * we didn't find anything useful, return | |
7292 | * the original results from get_extent() | |
7293 | */ | |
7294 | if (range_start > end || found_end <= start) { | |
7295 | em = hole_em; | |
7296 | hole_em = NULL; | |
7297 | goto out; | |
7298 | } | |
7299 | ||
7300 | /* adjust the range_start to make sure it doesn't | |
7301 | * go backwards from the start they passed in | |
7302 | */ | |
67871254 | 7303 | range_start = max(start, range_start); |
ec29ed5b CM |
7304 | found = found_end - range_start; |
7305 | ||
7306 | if (found > 0) { | |
7307 | u64 hole_start = start; | |
7308 | u64 hole_len = len; | |
7309 | ||
172ddd60 | 7310 | em = alloc_extent_map(); |
ec29ed5b CM |
7311 | if (!em) { |
7312 | err = -ENOMEM; | |
7313 | goto out; | |
7314 | } | |
7315 | /* | |
7316 | * when btrfs_get_extent can't find anything it | |
7317 | * returns one huge hole | |
7318 | * | |
7319 | * make sure what it found really fits our range, and | |
7320 | * adjust to make sure it is based on the start from | |
7321 | * the caller | |
7322 | */ | |
7323 | if (hole_em) { | |
7324 | u64 calc_end = extent_map_end(hole_em); | |
7325 | ||
7326 | if (calc_end <= start || (hole_em->start > end)) { | |
7327 | free_extent_map(hole_em); | |
7328 | hole_em = NULL; | |
7329 | } else { | |
7330 | hole_start = max(hole_em->start, start); | |
7331 | hole_len = calc_end - hole_start; | |
7332 | } | |
7333 | } | |
7334 | em->bdev = NULL; | |
7335 | if (hole_em && range_start > hole_start) { | |
7336 | /* our hole starts before our delalloc, so we | |
7337 | * have to return just the parts of the hole | |
7338 | * that go until the delalloc starts | |
7339 | */ | |
7340 | em->len = min(hole_len, | |
7341 | range_start - hole_start); | |
7342 | em->start = hole_start; | |
7343 | em->orig_start = hole_start; | |
7344 | /* | |
7345 | * don't adjust block start at all, | |
7346 | * it is fixed at EXTENT_MAP_HOLE | |
7347 | */ | |
7348 | em->block_start = hole_em->block_start; | |
7349 | em->block_len = hole_len; | |
f9e4fb53 LB |
7350 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7351 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
7352 | } else { |
7353 | em->start = range_start; | |
7354 | em->len = found; | |
7355 | em->orig_start = range_start; | |
7356 | em->block_start = EXTENT_MAP_DELALLOC; | |
7357 | em->block_len = found; | |
7358 | } | |
bf8d32b9 | 7359 | } else { |
ec29ed5b CM |
7360 | return hole_em; |
7361 | } | |
7362 | out: | |
7363 | ||
7364 | free_extent_map(hole_em); | |
7365 | if (err) { | |
7366 | free_extent_map(em); | |
7367 | return ERR_PTR(err); | |
7368 | } | |
7369 | return em; | |
7370 | } | |
7371 | ||
5f9a8a51 FM |
7372 | static struct extent_map *btrfs_create_dio_extent(struct inode *inode, |
7373 | const u64 start, | |
7374 | const u64 len, | |
7375 | const u64 orig_start, | |
7376 | const u64 block_start, | |
7377 | const u64 block_len, | |
7378 | const u64 orig_block_len, | |
7379 | const u64 ram_bytes, | |
7380 | const int type) | |
7381 | { | |
7382 | struct extent_map *em = NULL; | |
7383 | int ret; | |
7384 | ||
5f9a8a51 | 7385 | if (type != BTRFS_ORDERED_NOCOW) { |
6f9994db LB |
7386 | em = create_io_em(inode, start, len, orig_start, |
7387 | block_start, block_len, orig_block_len, | |
7388 | ram_bytes, | |
7389 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
7390 | type); | |
5f9a8a51 FM |
7391 | if (IS_ERR(em)) |
7392 | goto out; | |
7393 | } | |
7394 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, | |
7395 | len, block_len, type); | |
7396 | if (ret) { | |
7397 | if (em) { | |
7398 | free_extent_map(em); | |
dcdbc059 | 7399 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
5f9a8a51 FM |
7400 | start + len - 1, 0); |
7401 | } | |
7402 | em = ERR_PTR(ret); | |
7403 | } | |
7404 | out: | |
5f9a8a51 FM |
7405 | |
7406 | return em; | |
7407 | } | |
7408 | ||
4b46fce2 JB |
7409 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7410 | u64 start, u64 len) | |
7411 | { | |
0b246afa | 7412 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7413 | struct btrfs_root *root = BTRFS_I(inode)->root; |
70c8a91c | 7414 | struct extent_map *em; |
4b46fce2 JB |
7415 | struct btrfs_key ins; |
7416 | u64 alloc_hint; | |
7417 | int ret; | |
4b46fce2 | 7418 | |
4b46fce2 | 7419 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7420 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7421 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7422 | if (ret) |
7423 | return ERR_PTR(ret); | |
4b46fce2 | 7424 | |
5f9a8a51 FM |
7425 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
7426 | ins.objectid, ins.offset, ins.offset, | |
6288d6ea | 7427 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7428 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7429 | if (IS_ERR(em)) |
2ff7e61e JM |
7430 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
7431 | ins.offset, 1); | |
de0ee0ed | 7432 | |
4b46fce2 JB |
7433 | return em; |
7434 | } | |
7435 | ||
46bfbb5c CM |
7436 | /* |
7437 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7438 | * block must be cow'd | |
7439 | */ | |
00361589 | 7440 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7441 | u64 *orig_start, u64 *orig_block_len, |
7442 | u64 *ram_bytes) | |
46bfbb5c | 7443 | { |
2ff7e61e | 7444 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
7445 | struct btrfs_path *path; |
7446 | int ret; | |
7447 | struct extent_buffer *leaf; | |
7448 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7449 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7450 | struct btrfs_file_extent_item *fi; |
7451 | struct btrfs_key key; | |
7452 | u64 disk_bytenr; | |
7453 | u64 backref_offset; | |
7454 | u64 extent_end; | |
7455 | u64 num_bytes; | |
7456 | int slot; | |
7457 | int found_type; | |
7ee9e440 | 7458 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7459 | |
46bfbb5c CM |
7460 | path = btrfs_alloc_path(); |
7461 | if (!path) | |
7462 | return -ENOMEM; | |
7463 | ||
f85b7379 DS |
7464 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7465 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7466 | if (ret < 0) |
7467 | goto out; | |
7468 | ||
7469 | slot = path->slots[0]; | |
7470 | if (ret == 1) { | |
7471 | if (slot == 0) { | |
7472 | /* can't find the item, must cow */ | |
7473 | ret = 0; | |
7474 | goto out; | |
7475 | } | |
7476 | slot--; | |
7477 | } | |
7478 | ret = 0; | |
7479 | leaf = path->nodes[0]; | |
7480 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7481 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7482 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7483 | /* not our file or wrong item type, must cow */ | |
7484 | goto out; | |
7485 | } | |
7486 | ||
7487 | if (key.offset > offset) { | |
7488 | /* Wrong offset, must cow */ | |
7489 | goto out; | |
7490 | } | |
7491 | ||
7492 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7493 | found_type = btrfs_file_extent_type(leaf, fi); | |
7494 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7495 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7496 | /* not a regular extent, must cow */ | |
7497 | goto out; | |
7498 | } | |
7ee9e440 JB |
7499 | |
7500 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7501 | goto out; | |
7502 | ||
e77751aa MX |
7503 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7504 | if (extent_end <= offset) | |
7505 | goto out; | |
7506 | ||
46bfbb5c | 7507 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7508 | if (disk_bytenr == 0) |
7509 | goto out; | |
7510 | ||
7511 | if (btrfs_file_extent_compression(leaf, fi) || | |
7512 | btrfs_file_extent_encryption(leaf, fi) || | |
7513 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7514 | goto out; | |
7515 | ||
46bfbb5c CM |
7516 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7517 | ||
7ee9e440 JB |
7518 | if (orig_start) { |
7519 | *orig_start = key.offset - backref_offset; | |
7520 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7521 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7522 | } | |
eb384b55 | 7523 | |
2ff7e61e | 7524 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7525 | goto out; |
7b2b7085 MX |
7526 | |
7527 | num_bytes = min(offset + *len, extent_end) - offset; | |
7528 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7529 | u64 range_end; | |
7530 | ||
da17066c JM |
7531 | range_end = round_up(offset + num_bytes, |
7532 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7533 | ret = test_range_bit(io_tree, offset, range_end, |
7534 | EXTENT_DELALLOC, 0, NULL); | |
7535 | if (ret) { | |
7536 | ret = -EAGAIN; | |
7537 | goto out; | |
7538 | } | |
7539 | } | |
7540 | ||
1bda19eb | 7541 | btrfs_release_path(path); |
46bfbb5c CM |
7542 | |
7543 | /* | |
7544 | * look for other files referencing this extent, if we | |
7545 | * find any we must cow | |
7546 | */ | |
00361589 | 7547 | |
e4c3b2dc | 7548 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
00361589 | 7549 | key.offset - backref_offset, disk_bytenr); |
00361589 JB |
7550 | if (ret) { |
7551 | ret = 0; | |
7552 | goto out; | |
7553 | } | |
46bfbb5c CM |
7554 | |
7555 | /* | |
7556 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7557 | * in this extent we are about to write. If there | |
7558 | * are any csums in that range we have to cow in order | |
7559 | * to keep the csums correct | |
7560 | */ | |
7561 | disk_bytenr += backref_offset; | |
7562 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7563 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7564 | goto out; | |
46bfbb5c CM |
7565 | /* |
7566 | * all of the above have passed, it is safe to overwrite this extent | |
7567 | * without cow | |
7568 | */ | |
eb384b55 | 7569 | *len = num_bytes; |
46bfbb5c CM |
7570 | ret = 1; |
7571 | out: | |
7572 | btrfs_free_path(path); | |
7573 | return ret; | |
7574 | } | |
7575 | ||
eb838e73 JB |
7576 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7577 | struct extent_state **cached_state, int writing) | |
7578 | { | |
7579 | struct btrfs_ordered_extent *ordered; | |
7580 | int ret = 0; | |
7581 | ||
7582 | while (1) { | |
7583 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7584 | cached_state); |
eb838e73 JB |
7585 | /* |
7586 | * We're concerned with the entire range that we're going to be | |
01327610 | 7587 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7588 | * extents in this range. |
7589 | */ | |
a776c6fa | 7590 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7591 | lockend - lockstart + 1); |
7592 | ||
7593 | /* | |
7594 | * We need to make sure there are no buffered pages in this | |
7595 | * range either, we could have raced between the invalidate in | |
7596 | * generic_file_direct_write and locking the extent. The | |
7597 | * invalidate needs to happen so that reads after a write do not | |
7598 | * get stale data. | |
7599 | */ | |
fc4adbff | 7600 | if (!ordered && |
051c98eb DS |
7601 | (!writing || !filemap_range_has_page(inode->i_mapping, |
7602 | lockstart, lockend))) | |
eb838e73 JB |
7603 | break; |
7604 | ||
7605 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
e43bbe5e | 7606 | cached_state); |
eb838e73 JB |
7607 | |
7608 | if (ordered) { | |
ade77029 FM |
7609 | /* |
7610 | * If we are doing a DIO read and the ordered extent we | |
7611 | * found is for a buffered write, we can not wait for it | |
7612 | * to complete and retry, because if we do so we can | |
7613 | * deadlock with concurrent buffered writes on page | |
7614 | * locks. This happens only if our DIO read covers more | |
7615 | * than one extent map, if at this point has already | |
7616 | * created an ordered extent for a previous extent map | |
7617 | * and locked its range in the inode's io tree, and a | |
7618 | * concurrent write against that previous extent map's | |
7619 | * range and this range started (we unlock the ranges | |
7620 | * in the io tree only when the bios complete and | |
7621 | * buffered writes always lock pages before attempting | |
7622 | * to lock range in the io tree). | |
7623 | */ | |
7624 | if (writing || | |
7625 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7626 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7627 | else | |
7628 | ret = -ENOTBLK; | |
eb838e73 JB |
7629 | btrfs_put_ordered_extent(ordered); |
7630 | } else { | |
eb838e73 | 7631 | /* |
b850ae14 FM |
7632 | * We could trigger writeback for this range (and wait |
7633 | * for it to complete) and then invalidate the pages for | |
7634 | * this range (through invalidate_inode_pages2_range()), | |
7635 | * but that can lead us to a deadlock with a concurrent | |
7636 | * call to readpages() (a buffered read or a defrag call | |
7637 | * triggered a readahead) on a page lock due to an | |
7638 | * ordered dio extent we created before but did not have | |
7639 | * yet a corresponding bio submitted (whence it can not | |
7640 | * complete), which makes readpages() wait for that | |
7641 | * ordered extent to complete while holding a lock on | |
7642 | * that page. | |
eb838e73 | 7643 | */ |
b850ae14 | 7644 | ret = -ENOTBLK; |
eb838e73 JB |
7645 | } |
7646 | ||
ade77029 FM |
7647 | if (ret) |
7648 | break; | |
7649 | ||
eb838e73 JB |
7650 | cond_resched(); |
7651 | } | |
7652 | ||
7653 | return ret; | |
7654 | } | |
7655 | ||
6f9994db LB |
7656 | /* The callers of this must take lock_extent() */ |
7657 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, | |
7658 | u64 orig_start, u64 block_start, | |
7659 | u64 block_len, u64 orig_block_len, | |
7660 | u64 ram_bytes, int compress_type, | |
7661 | int type) | |
69ffb543 JB |
7662 | { |
7663 | struct extent_map_tree *em_tree; | |
7664 | struct extent_map *em; | |
7665 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7666 | int ret; | |
7667 | ||
6f9994db LB |
7668 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7669 | type == BTRFS_ORDERED_COMPRESSED || | |
7670 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7671 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7672 | |
69ffb543 JB |
7673 | em_tree = &BTRFS_I(inode)->extent_tree; |
7674 | em = alloc_extent_map(); | |
7675 | if (!em) | |
7676 | return ERR_PTR(-ENOMEM); | |
7677 | ||
7678 | em->start = start; | |
7679 | em->orig_start = orig_start; | |
7680 | em->len = len; | |
7681 | em->block_len = block_len; | |
7682 | em->block_start = block_start; | |
7683 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7684 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7685 | em->ram_bytes = ram_bytes; |
70c8a91c | 7686 | em->generation = -1; |
69ffb543 | 7687 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7688 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7689 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7690 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7691 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7692 | em->compress_type = compress_type; | |
7693 | } | |
69ffb543 JB |
7694 | |
7695 | do { | |
dcdbc059 | 7696 | btrfs_drop_extent_cache(BTRFS_I(inode), em->start, |
69ffb543 JB |
7697 | em->start + em->len - 1, 0); |
7698 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7699 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7700 | write_unlock(&em_tree->lock); |
6f9994db LB |
7701 | /* |
7702 | * The caller has taken lock_extent(), who could race with us | |
7703 | * to add em? | |
7704 | */ | |
69ffb543 JB |
7705 | } while (ret == -EEXIST); |
7706 | ||
7707 | if (ret) { | |
7708 | free_extent_map(em); | |
7709 | return ERR_PTR(ret); | |
7710 | } | |
7711 | ||
6f9994db | 7712 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7713 | return em; |
7714 | } | |
7715 | ||
4b46fce2 JB |
7716 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7717 | struct buffer_head *bh_result, int create) | |
7718 | { | |
0b246afa | 7719 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7720 | struct extent_map *em; |
eb838e73 | 7721 | struct extent_state *cached_state = NULL; |
50745b0a | 7722 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7723 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7724 | u64 lockstart, lockend; |
4b46fce2 | 7725 | u64 len = bh_result->b_size; |
eb838e73 | 7726 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7727 | int ret = 0; |
eb838e73 | 7728 | |
172a5049 | 7729 | if (create) |
3266789f | 7730 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7731 | else |
0b246afa | 7732 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7733 | |
c329861d JB |
7734 | lockstart = start; |
7735 | lockend = start + len - 1; | |
7736 | ||
e1cbbfa5 JB |
7737 | if (current->journal_info) { |
7738 | /* | |
7739 | * Need to pull our outstanding extents and set journal_info to NULL so | |
01327610 | 7740 | * that anything that needs to check if there's a transaction doesn't get |
e1cbbfa5 JB |
7741 | * confused. |
7742 | */ | |
50745b0a | 7743 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7744 | current->journal_info = NULL; |
7745 | } | |
7746 | ||
eb838e73 JB |
7747 | /* |
7748 | * If this errors out it's because we couldn't invalidate pagecache for | |
7749 | * this range and we need to fallback to buffered. | |
7750 | */ | |
9c9464cc FM |
7751 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7752 | create)) { | |
7753 | ret = -ENOTBLK; | |
7754 | goto err; | |
7755 | } | |
eb838e73 | 7756 | |
fc4f21b1 | 7757 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); |
eb838e73 JB |
7758 | if (IS_ERR(em)) { |
7759 | ret = PTR_ERR(em); | |
7760 | goto unlock_err; | |
7761 | } | |
4b46fce2 JB |
7762 | |
7763 | /* | |
7764 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7765 | * io. INLINE is special, and we could probably kludge it in here, but | |
7766 | * it's still buffered so for safety lets just fall back to the generic | |
7767 | * buffered path. | |
7768 | * | |
7769 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7770 | * decompress it, so there will be buffering required no matter what we | |
7771 | * do, so go ahead and fallback to buffered. | |
7772 | * | |
01327610 | 7773 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7774 | * to buffered IO. Don't blame me, this is the price we pay for using |
7775 | * the generic code. | |
7776 | */ | |
7777 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7778 | em->block_start == EXTENT_MAP_INLINE) { | |
7779 | free_extent_map(em); | |
eb838e73 JB |
7780 | ret = -ENOTBLK; |
7781 | goto unlock_err; | |
4b46fce2 JB |
7782 | } |
7783 | ||
7784 | /* Just a good old fashioned hole, return */ | |
7785 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7786 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7787 | free_extent_map(em); | |
eb838e73 | 7788 | goto unlock_err; |
4b46fce2 JB |
7789 | } |
7790 | ||
7791 | /* | |
7792 | * We don't allocate a new extent in the following cases | |
7793 | * | |
7794 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7795 | * existing extent. | |
7796 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7797 | * just use the extent. | |
7798 | * | |
7799 | */ | |
46bfbb5c | 7800 | if (!create) { |
eb838e73 JB |
7801 | len = min(len, em->len - (start - em->start)); |
7802 | lockstart = start + len; | |
7803 | goto unlock; | |
46bfbb5c | 7804 | } |
4b46fce2 JB |
7805 | |
7806 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7807 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7808 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 | 7809 | int type; |
eb384b55 | 7810 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7811 | |
7812 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7813 | type = BTRFS_ORDERED_PREALLOC; | |
7814 | else | |
7815 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7816 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7817 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7818 | |
00361589 | 7819 | if (can_nocow_extent(inode, start, &len, &orig_start, |
f78c436c | 7820 | &orig_block_len, &ram_bytes) == 1 && |
0b246afa | 7821 | btrfs_inc_nocow_writers(fs_info, block_start)) { |
5f9a8a51 | 7822 | struct extent_map *em2; |
0b901916 | 7823 | |
5f9a8a51 FM |
7824 | em2 = btrfs_create_dio_extent(inode, start, len, |
7825 | orig_start, block_start, | |
7826 | len, orig_block_len, | |
7827 | ram_bytes, type); | |
0b246afa | 7828 | btrfs_dec_nocow_writers(fs_info, block_start); |
69ffb543 JB |
7829 | if (type == BTRFS_ORDERED_PREALLOC) { |
7830 | free_extent_map(em); | |
5f9a8a51 | 7831 | em = em2; |
69ffb543 | 7832 | } |
5f9a8a51 FM |
7833 | if (em2 && IS_ERR(em2)) { |
7834 | ret = PTR_ERR(em2); | |
eb838e73 | 7835 | goto unlock_err; |
46bfbb5c | 7836 | } |
18513091 WX |
7837 | /* |
7838 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7839 | * use the existing or preallocated extent, so does not | |
7840 | * need to adjust btrfs_space_info's bytes_may_use. | |
7841 | */ | |
7842 | btrfs_free_reserved_data_space_noquota(inode, | |
7843 | start, len); | |
46bfbb5c | 7844 | goto unlock; |
4b46fce2 | 7845 | } |
4b46fce2 | 7846 | } |
00361589 | 7847 | |
46bfbb5c CM |
7848 | /* |
7849 | * this will cow the extent, reset the len in case we changed | |
7850 | * it above | |
7851 | */ | |
7852 | len = bh_result->b_size; | |
70c8a91c JB |
7853 | free_extent_map(em); |
7854 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7855 | if (IS_ERR(em)) { |
7856 | ret = PTR_ERR(em); | |
7857 | goto unlock_err; | |
7858 | } | |
46bfbb5c CM |
7859 | len = min(len, em->len - (start - em->start)); |
7860 | unlock: | |
4b46fce2 JB |
7861 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7862 | inode->i_blkbits; | |
46bfbb5c | 7863 | bh_result->b_size = len; |
4b46fce2 JB |
7864 | bh_result->b_bdev = em->bdev; |
7865 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7866 | if (create) { |
7867 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7868 | set_buffer_new(bh_result); | |
7869 | ||
7870 | /* | |
7871 | * Need to update the i_size under the extent lock so buffered | |
7872 | * readers will get the updated i_size when we unlock. | |
7873 | */ | |
4aaedfb0 | 7874 | if (!dio_data->overwrite && start + len > i_size_read(inode)) |
c3473e83 | 7875 | i_size_write(inode, start + len); |
0934856d | 7876 | |
50745b0a | 7877 | WARN_ON(dio_data->reserve < len); |
7878 | dio_data->reserve -= len; | |
f28a4928 | 7879 | dio_data->unsubmitted_oe_range_end = start + len; |
50745b0a | 7880 | current->journal_info = dio_data; |
c3473e83 | 7881 | } |
4b46fce2 | 7882 | |
eb838e73 JB |
7883 | /* |
7884 | * In the case of write we need to clear and unlock the entire range, | |
7885 | * in the case of read we need to unlock only the end area that we | |
7886 | * aren't using if there is any left over space. | |
7887 | */ | |
24c03fa5 | 7888 | if (lockstart < lockend) { |
0934856d MX |
7889 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7890 | lockend, unlock_bits, 1, 0, | |
ae0f1625 | 7891 | &cached_state); |
24c03fa5 | 7892 | } else { |
eb838e73 | 7893 | free_extent_state(cached_state); |
24c03fa5 | 7894 | } |
eb838e73 | 7895 | |
4b46fce2 JB |
7896 | free_extent_map(em); |
7897 | ||
7898 | return 0; | |
eb838e73 JB |
7899 | |
7900 | unlock_err: | |
eb838e73 | 7901 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
ae0f1625 | 7902 | unlock_bits, 1, 0, &cached_state); |
9c9464cc | 7903 | err: |
50745b0a | 7904 | if (dio_data) |
7905 | current->journal_info = dio_data; | |
eb838e73 | 7906 | return ret; |
4b46fce2 JB |
7907 | } |
7908 | ||
58efbc9f OS |
7909 | static inline blk_status_t submit_dio_repair_bio(struct inode *inode, |
7910 | struct bio *bio, | |
7911 | int mirror_num) | |
8b110e39 | 7912 | { |
2ff7e61e | 7913 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
58efbc9f | 7914 | blk_status_t ret; |
8b110e39 | 7915 | |
37226b21 | 7916 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 | 7917 | |
2ff7e61e | 7918 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR); |
8b110e39 | 7919 | if (ret) |
ea057f6d | 7920 | return ret; |
8b110e39 | 7921 | |
2ff7e61e | 7922 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
ea057f6d | 7923 | |
8b110e39 MX |
7924 | return ret; |
7925 | } | |
7926 | ||
7927 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7928 | struct bio *failed_bio, | |
7929 | struct io_failure_record *failrec, | |
7930 | int failed_mirror) | |
7931 | { | |
ab8d0fc4 | 7932 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 MX |
7933 | int num_copies; |
7934 | ||
ab8d0fc4 | 7935 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
8b110e39 MX |
7936 | if (num_copies == 1) { |
7937 | /* | |
7938 | * we only have a single copy of the data, so don't bother with | |
7939 | * all the retry and error correction code that follows. no | |
7940 | * matter what the error is, it is very likely to persist. | |
7941 | */ | |
ab8d0fc4 JM |
7942 | btrfs_debug(fs_info, |
7943 | "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
7944 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7945 | return 0; |
7946 | } | |
7947 | ||
7948 | failrec->failed_mirror = failed_mirror; | |
7949 | failrec->this_mirror++; | |
7950 | if (failrec->this_mirror == failed_mirror) | |
7951 | failrec->this_mirror++; | |
7952 | ||
7953 | if (failrec->this_mirror > num_copies) { | |
ab8d0fc4 JM |
7954 | btrfs_debug(fs_info, |
7955 | "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
7956 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7957 | return 0; |
7958 | } | |
7959 | ||
7960 | return 1; | |
7961 | } | |
7962 | ||
58efbc9f OS |
7963 | static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio, |
7964 | struct page *page, unsigned int pgoff, | |
7965 | u64 start, u64 end, int failed_mirror, | |
7966 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
7967 | { |
7968 | struct io_failure_record *failrec; | |
7870d082 JB |
7969 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
7970 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8b110e39 MX |
7971 | struct bio *bio; |
7972 | int isector; | |
f1c77c55 | 7973 | unsigned int read_mode = 0; |
17347cec | 7974 | int segs; |
8b110e39 | 7975 | int ret; |
58efbc9f | 7976 | blk_status_t status; |
c16a8ac3 | 7977 | struct bio_vec bvec; |
8b110e39 | 7978 | |
37226b21 | 7979 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
8b110e39 MX |
7980 | |
7981 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7982 | if (ret) | |
58efbc9f | 7983 | return errno_to_blk_status(ret); |
8b110e39 MX |
7984 | |
7985 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7986 | failed_mirror); | |
7987 | if (!ret) { | |
7870d082 | 7988 | free_io_failure(failure_tree, io_tree, failrec); |
58efbc9f | 7989 | return BLK_STS_IOERR; |
8b110e39 MX |
7990 | } |
7991 | ||
17347cec | 7992 | segs = bio_segments(failed_bio); |
c16a8ac3 | 7993 | bio_get_first_bvec(failed_bio, &bvec); |
17347cec | 7994 | if (segs > 1 || |
c16a8ac3 | 7995 | (bvec.bv_len > btrfs_inode_sectorsize(inode))) |
70fd7614 | 7996 | read_mode |= REQ_FAILFAST_DEV; |
8b110e39 MX |
7997 | |
7998 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7999 | isector >>= inode->i_sb->s_blocksize_bits; | |
8000 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 8001 | pgoff, isector, repair_endio, repair_arg); |
37226b21 | 8002 | bio_set_op_attrs(bio, REQ_OP_READ, read_mode); |
8b110e39 MX |
8003 | |
8004 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
913e1535 | 8005 | "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d", |
8b110e39 MX |
8006 | read_mode, failrec->this_mirror, failrec->in_validation); |
8007 | ||
58efbc9f OS |
8008 | status = submit_dio_repair_bio(inode, bio, failrec->this_mirror); |
8009 | if (status) { | |
7870d082 | 8010 | free_io_failure(failure_tree, io_tree, failrec); |
8b110e39 MX |
8011 | bio_put(bio); |
8012 | } | |
8013 | ||
58efbc9f | 8014 | return status; |
8b110e39 MX |
8015 | } |
8016 | ||
8017 | struct btrfs_retry_complete { | |
8018 | struct completion done; | |
8019 | struct inode *inode; | |
8020 | u64 start; | |
8021 | int uptodate; | |
8022 | }; | |
8023 | ||
4246a0b6 | 8024 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
8025 | { |
8026 | struct btrfs_retry_complete *done = bio->bi_private; | |
7870d082 | 8027 | struct inode *inode = done->inode; |
8b110e39 | 8028 | struct bio_vec *bvec; |
7870d082 | 8029 | struct extent_io_tree *io_tree, *failure_tree; |
8b110e39 MX |
8030 | int i; |
8031 | ||
4e4cbee9 | 8032 | if (bio->bi_status) |
8b110e39 MX |
8033 | goto end; |
8034 | ||
2dabb324 | 8035 | ASSERT(bio->bi_vcnt == 1); |
7870d082 JB |
8036 | io_tree = &BTRFS_I(inode)->io_tree; |
8037 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
263663cd | 8038 | ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode)); |
2dabb324 | 8039 | |
8b110e39 | 8040 | done->uptodate = 1; |
c09abff8 | 8041 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
8b110e39 | 8042 | bio_for_each_segment_all(bvec, bio, i) |
7870d082 JB |
8043 | clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree, |
8044 | io_tree, done->start, bvec->bv_page, | |
8045 | btrfs_ino(BTRFS_I(inode)), 0); | |
8b110e39 MX |
8046 | end: |
8047 | complete(&done->done); | |
8048 | bio_put(bio); | |
8049 | } | |
8050 | ||
58efbc9f OS |
8051 | static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode, |
8052 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 8053 | { |
2dabb324 | 8054 | struct btrfs_fs_info *fs_info; |
17347cec LB |
8055 | struct bio_vec bvec; |
8056 | struct bvec_iter iter; | |
8b110e39 | 8057 | struct btrfs_retry_complete done; |
4b46fce2 | 8058 | u64 start; |
2dabb324 CR |
8059 | unsigned int pgoff; |
8060 | u32 sectorsize; | |
8061 | int nr_sectors; | |
58efbc9f OS |
8062 | blk_status_t ret; |
8063 | blk_status_t err = BLK_STS_OK; | |
4b46fce2 | 8064 | |
2dabb324 | 8065 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8066 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8067 | |
8b110e39 MX |
8068 | start = io_bio->logical; |
8069 | done.inode = inode; | |
17347cec | 8070 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 8071 | |
17347cec LB |
8072 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
8073 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
8074 | pgoff = bvec.bv_offset; | |
2dabb324 CR |
8075 | |
8076 | next_block_or_try_again: | |
8b110e39 MX |
8077 | done.uptodate = 0; |
8078 | done.start = start; | |
8079 | init_completion(&done.done); | |
8080 | ||
17347cec | 8081 | ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
2dabb324 CR |
8082 | pgoff, start, start + sectorsize - 1, |
8083 | io_bio->mirror_num, | |
8084 | btrfs_retry_endio_nocsum, &done); | |
629ebf4f LB |
8085 | if (ret) { |
8086 | err = ret; | |
8087 | goto next; | |
8088 | } | |
8b110e39 | 8089 | |
9c17f6cd | 8090 | wait_for_completion_io(&done.done); |
8b110e39 MX |
8091 | |
8092 | if (!done.uptodate) { | |
8093 | /* We might have another mirror, so try again */ | |
2dabb324 | 8094 | goto next_block_or_try_again; |
8b110e39 MX |
8095 | } |
8096 | ||
629ebf4f | 8097 | next: |
2dabb324 CR |
8098 | start += sectorsize; |
8099 | ||
97bf5a55 LB |
8100 | nr_sectors--; |
8101 | if (nr_sectors) { | |
2dabb324 | 8102 | pgoff += sectorsize; |
97bf5a55 | 8103 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8104 | goto next_block_or_try_again; |
8105 | } | |
8b110e39 MX |
8106 | } |
8107 | ||
629ebf4f | 8108 | return err; |
8b110e39 MX |
8109 | } |
8110 | ||
4246a0b6 | 8111 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
8112 | { |
8113 | struct btrfs_retry_complete *done = bio->bi_private; | |
8114 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7870d082 JB |
8115 | struct extent_io_tree *io_tree, *failure_tree; |
8116 | struct inode *inode = done->inode; | |
8b110e39 MX |
8117 | struct bio_vec *bvec; |
8118 | int uptodate; | |
8119 | int ret; | |
8120 | int i; | |
8121 | ||
4e4cbee9 | 8122 | if (bio->bi_status) |
8b110e39 MX |
8123 | goto end; |
8124 | ||
8125 | uptodate = 1; | |
2dabb324 | 8126 | |
2dabb324 | 8127 | ASSERT(bio->bi_vcnt == 1); |
263663cd | 8128 | ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode)); |
2dabb324 | 8129 | |
7870d082 JB |
8130 | io_tree = &BTRFS_I(inode)->io_tree; |
8131 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8132 | ||
c09abff8 | 8133 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
8b110e39 | 8134 | bio_for_each_segment_all(bvec, bio, i) { |
7870d082 JB |
8135 | ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page, |
8136 | bvec->bv_offset, done->start, | |
8137 | bvec->bv_len); | |
8b110e39 | 8138 | if (!ret) |
7870d082 JB |
8139 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
8140 | failure_tree, io_tree, done->start, | |
8141 | bvec->bv_page, | |
8142 | btrfs_ino(BTRFS_I(inode)), | |
8143 | bvec->bv_offset); | |
8b110e39 MX |
8144 | else |
8145 | uptodate = 0; | |
8146 | } | |
8147 | ||
8148 | done->uptodate = uptodate; | |
8149 | end: | |
8150 | complete(&done->done); | |
8151 | bio_put(bio); | |
8152 | } | |
8153 | ||
4e4cbee9 CH |
8154 | static blk_status_t __btrfs_subio_endio_read(struct inode *inode, |
8155 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 | 8156 | { |
2dabb324 | 8157 | struct btrfs_fs_info *fs_info; |
17347cec LB |
8158 | struct bio_vec bvec; |
8159 | struct bvec_iter iter; | |
8b110e39 MX |
8160 | struct btrfs_retry_complete done; |
8161 | u64 start; | |
8162 | u64 offset = 0; | |
2dabb324 CR |
8163 | u32 sectorsize; |
8164 | int nr_sectors; | |
8165 | unsigned int pgoff; | |
8166 | int csum_pos; | |
ef7cdac1 | 8167 | bool uptodate = (err == 0); |
8b110e39 | 8168 | int ret; |
58efbc9f | 8169 | blk_status_t status; |
dc380aea | 8170 | |
2dabb324 | 8171 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8172 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8173 | |
58efbc9f | 8174 | err = BLK_STS_OK; |
c1dc0896 | 8175 | start = io_bio->logical; |
8b110e39 | 8176 | done.inode = inode; |
17347cec | 8177 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 8178 | |
17347cec LB |
8179 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
8180 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
2dabb324 | 8181 | |
17347cec | 8182 | pgoff = bvec.bv_offset; |
2dabb324 | 8183 | next_block: |
ef7cdac1 LB |
8184 | if (uptodate) { |
8185 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
8186 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
8187 | bvec.bv_page, pgoff, start, sectorsize); | |
8188 | if (likely(!ret)) | |
8189 | goto next; | |
8190 | } | |
8b110e39 MX |
8191 | try_again: |
8192 | done.uptodate = 0; | |
8193 | done.start = start; | |
8194 | init_completion(&done.done); | |
8195 | ||
58efbc9f OS |
8196 | status = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
8197 | pgoff, start, start + sectorsize - 1, | |
8198 | io_bio->mirror_num, btrfs_retry_endio, | |
8199 | &done); | |
8200 | if (status) { | |
8201 | err = status; | |
8b110e39 MX |
8202 | goto next; |
8203 | } | |
8204 | ||
9c17f6cd | 8205 | wait_for_completion_io(&done.done); |
8b110e39 MX |
8206 | |
8207 | if (!done.uptodate) { | |
8208 | /* We might have another mirror, so try again */ | |
8209 | goto try_again; | |
8210 | } | |
8211 | next: | |
2dabb324 CR |
8212 | offset += sectorsize; |
8213 | start += sectorsize; | |
8214 | ||
8215 | ASSERT(nr_sectors); | |
8216 | ||
97bf5a55 LB |
8217 | nr_sectors--; |
8218 | if (nr_sectors) { | |
2dabb324 | 8219 | pgoff += sectorsize; |
97bf5a55 | 8220 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8221 | goto next_block; |
8222 | } | |
2c30c71b | 8223 | } |
c1dc0896 MX |
8224 | |
8225 | return err; | |
8226 | } | |
8227 | ||
4e4cbee9 CH |
8228 | static blk_status_t btrfs_subio_endio_read(struct inode *inode, |
8229 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 MX |
8230 | { |
8231 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8232 | ||
8233 | if (skip_csum) { | |
8234 | if (unlikely(err)) | |
8235 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8236 | else | |
58efbc9f | 8237 | return BLK_STS_OK; |
8b110e39 MX |
8238 | } else { |
8239 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8240 | } | |
8241 | } | |
8242 | ||
4246a0b6 | 8243 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8244 | { |
8245 | struct btrfs_dio_private *dip = bio->bi_private; | |
8246 | struct inode *inode = dip->inode; | |
8247 | struct bio *dio_bio; | |
8248 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4e4cbee9 | 8249 | blk_status_t err = bio->bi_status; |
c1dc0896 | 8250 | |
99c4e3b9 | 8251 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8b110e39 | 8252 | err = btrfs_subio_endio_read(inode, io_bio, err); |
c1dc0896 | 8253 | |
4b46fce2 | 8254 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8255 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8256 | dio_bio = dip->dio_bio; |
4b46fce2 | 8257 | |
4b46fce2 | 8258 | kfree(dip); |
c0da7aa1 | 8259 | |
99c4e3b9 | 8260 | dio_bio->bi_status = err; |
4055351c | 8261 | dio_end_io(dio_bio); |
23ea8e5a MX |
8262 | |
8263 | if (io_bio->end_io) | |
4e4cbee9 | 8264 | io_bio->end_io(io_bio, blk_status_to_errno(err)); |
9be3395b | 8265 | bio_put(bio); |
4b46fce2 JB |
8266 | } |
8267 | ||
52427260 QW |
8268 | static void __endio_write_update_ordered(struct inode *inode, |
8269 | const u64 offset, const u64 bytes, | |
8270 | const bool uptodate) | |
4b46fce2 | 8271 | { |
0b246afa | 8272 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 8273 | struct btrfs_ordered_extent *ordered = NULL; |
52427260 QW |
8274 | struct btrfs_workqueue *wq; |
8275 | btrfs_work_func_t func; | |
14543774 FM |
8276 | u64 ordered_offset = offset; |
8277 | u64 ordered_bytes = bytes; | |
67c003f9 | 8278 | u64 last_offset; |
4b46fce2 | 8279 | |
52427260 QW |
8280 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
8281 | wq = fs_info->endio_freespace_worker; | |
8282 | func = btrfs_freespace_write_helper; | |
8283 | } else { | |
8284 | wq = fs_info->endio_write_workers; | |
8285 | func = btrfs_endio_write_helper; | |
8286 | } | |
8287 | ||
b25f0d00 NB |
8288 | while (ordered_offset < offset + bytes) { |
8289 | last_offset = ordered_offset; | |
8290 | if (btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
8291 | &ordered_offset, | |
8292 | ordered_bytes, | |
8293 | uptodate)) { | |
8294 | btrfs_init_work(&ordered->work, func, | |
8295 | finish_ordered_fn, | |
8296 | NULL, NULL); | |
8297 | btrfs_queue_work(wq, &ordered->work); | |
8298 | } | |
8299 | /* | |
8300 | * If btrfs_dec_test_ordered_pending does not find any ordered | |
8301 | * extent in the range, we can exit. | |
8302 | */ | |
8303 | if (ordered_offset == last_offset) | |
8304 | return; | |
8305 | /* | |
8306 | * Our bio might span multiple ordered extents. In this case | |
8307 | * we keep goin until we have accounted the whole dio. | |
8308 | */ | |
8309 | if (ordered_offset < offset + bytes) { | |
8310 | ordered_bytes = offset + bytes - ordered_offset; | |
8311 | ordered = NULL; | |
8312 | } | |
163cf09c | 8313 | } |
14543774 FM |
8314 | } |
8315 | ||
8316 | static void btrfs_endio_direct_write(struct bio *bio) | |
8317 | { | |
8318 | struct btrfs_dio_private *dip = bio->bi_private; | |
8319 | struct bio *dio_bio = dip->dio_bio; | |
8320 | ||
52427260 | 8321 | __endio_write_update_ordered(dip->inode, dip->logical_offset, |
4e4cbee9 | 8322 | dip->bytes, !bio->bi_status); |
4b46fce2 | 8323 | |
4b46fce2 | 8324 | kfree(dip); |
c0da7aa1 | 8325 | |
4e4cbee9 | 8326 | dio_bio->bi_status = bio->bi_status; |
4055351c | 8327 | dio_end_io(dio_bio); |
9be3395b | 8328 | bio_put(bio); |
4b46fce2 JB |
8329 | } |
8330 | ||
d0ee3934 | 8331 | static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data, |
d0779291 | 8332 | struct bio *bio, u64 offset) |
eaf25d93 | 8333 | { |
c6100a4b | 8334 | struct inode *inode = private_data; |
4e4cbee9 | 8335 | blk_status_t ret; |
2ff7e61e | 8336 | ret = btrfs_csum_one_bio(inode, bio, offset, 1); |
79787eaa | 8337 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8338 | return 0; |
8339 | } | |
8340 | ||
4246a0b6 | 8341 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8342 | { |
8343 | struct btrfs_dio_private *dip = bio->bi_private; | |
4e4cbee9 | 8344 | blk_status_t err = bio->bi_status; |
e65e1535 | 8345 | |
8b110e39 MX |
8346 | if (err) |
8347 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 8348 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 DS |
8349 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
8350 | bio->bi_opf, | |
8b110e39 MX |
8351 | (unsigned long long)bio->bi_iter.bi_sector, |
8352 | bio->bi_iter.bi_size, err); | |
8353 | ||
8354 | if (dip->subio_endio) | |
8355 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8356 | |
8357 | if (err) { | |
e65e1535 | 8358 | /* |
de224b7c NB |
8359 | * We want to perceive the errors flag being set before |
8360 | * decrementing the reference count. We don't need a barrier | |
8361 | * since atomic operations with a return value are fully | |
8362 | * ordered as per atomic_t.txt | |
e65e1535 | 8363 | */ |
de224b7c | 8364 | dip->errors = 1; |
e65e1535 MX |
8365 | } |
8366 | ||
8367 | /* if there are more bios still pending for this dio, just exit */ | |
8368 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8369 | goto out; | |
8370 | ||
9be3395b | 8371 | if (dip->errors) { |
e65e1535 | 8372 | bio_io_error(dip->orig_bio); |
9be3395b | 8373 | } else { |
2dbe0c77 | 8374 | dip->dio_bio->bi_status = BLK_STS_OK; |
4246a0b6 | 8375 | bio_endio(dip->orig_bio); |
e65e1535 MX |
8376 | } |
8377 | out: | |
8378 | bio_put(bio); | |
8379 | } | |
8380 | ||
4e4cbee9 | 8381 | static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode, |
c1dc0896 MX |
8382 | struct btrfs_dio_private *dip, |
8383 | struct bio *bio, | |
8384 | u64 file_offset) | |
8385 | { | |
8386 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8387 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
4e4cbee9 | 8388 | blk_status_t ret; |
c1dc0896 MX |
8389 | |
8390 | /* | |
8391 | * We load all the csum data we need when we submit | |
8392 | * the first bio to reduce the csum tree search and | |
8393 | * contention. | |
8394 | */ | |
8395 | if (dip->logical_offset == file_offset) { | |
2ff7e61e | 8396 | ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio, |
c1dc0896 MX |
8397 | file_offset); |
8398 | if (ret) | |
8399 | return ret; | |
8400 | } | |
8401 | ||
8402 | if (bio == dip->orig_bio) | |
8403 | return 0; | |
8404 | ||
8405 | file_offset -= dip->logical_offset; | |
8406 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8407 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8408 | ||
8409 | return 0; | |
8410 | } | |
8411 | ||
d0ee3934 DS |
8412 | static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio, |
8413 | struct inode *inode, u64 file_offset, int async_submit) | |
e65e1535 | 8414 | { |
0b246afa | 8415 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 8416 | struct btrfs_dio_private *dip = bio->bi_private; |
37226b21 | 8417 | bool write = bio_op(bio) == REQ_OP_WRITE; |
4e4cbee9 | 8418 | blk_status_t ret; |
e65e1535 | 8419 | |
4c274bc6 | 8420 | /* Check btrfs_submit_bio_hook() for rules about async submit. */ |
b812ce28 JB |
8421 | if (async_submit) |
8422 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8423 | ||
5fd02043 | 8424 | if (!write) { |
0b246afa | 8425 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
8426 | if (ret) |
8427 | goto err; | |
8428 | } | |
e65e1535 | 8429 | |
e6961cac | 8430 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
8431 | goto map; |
8432 | ||
8433 | if (write && async_submit) { | |
c6100a4b JB |
8434 | ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0, |
8435 | file_offset, inode, | |
d0ee3934 DS |
8436 | btrfs_submit_bio_start_direct_io, |
8437 | btrfs_submit_bio_done); | |
e65e1535 | 8438 | goto err; |
1ae39938 JB |
8439 | } else if (write) { |
8440 | /* | |
8441 | * If we aren't doing async submit, calculate the csum of the | |
8442 | * bio now. | |
8443 | */ | |
2ff7e61e | 8444 | ret = btrfs_csum_one_bio(inode, bio, file_offset, 1); |
1ae39938 JB |
8445 | if (ret) |
8446 | goto err; | |
23ea8e5a | 8447 | } else { |
2ff7e61e | 8448 | ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio, |
c1dc0896 | 8449 | file_offset); |
c2db1073 TI |
8450 | if (ret) |
8451 | goto err; | |
8452 | } | |
1ae39938 | 8453 | map: |
9b4a9b28 | 8454 | ret = btrfs_map_bio(fs_info, bio, 0, 0); |
e65e1535 | 8455 | err: |
e65e1535 MX |
8456 | return ret; |
8457 | } | |
8458 | ||
e6961cac | 8459 | static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip) |
e65e1535 MX |
8460 | { |
8461 | struct inode *inode = dip->inode; | |
0b246afa | 8462 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e65e1535 MX |
8463 | struct bio *bio; |
8464 | struct bio *orig_bio = dip->orig_bio; | |
4f024f37 | 8465 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 | 8466 | u64 file_offset = dip->logical_offset; |
e65e1535 | 8467 | u64 map_length; |
1ae39938 | 8468 | int async_submit = 0; |
725130ba LB |
8469 | u64 submit_len; |
8470 | int clone_offset = 0; | |
8471 | int clone_len; | |
5f4dc8fc | 8472 | int ret; |
58efbc9f | 8473 | blk_status_t status; |
e65e1535 | 8474 | |
4f024f37 | 8475 | map_length = orig_bio->bi_iter.bi_size; |
725130ba | 8476 | submit_len = map_length; |
0b246afa JM |
8477 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9, |
8478 | &map_length, NULL, 0); | |
7a5c3c9b | 8479 | if (ret) |
e65e1535 | 8480 | return -EIO; |
facc8a22 | 8481 | |
725130ba | 8482 | if (map_length >= submit_len) { |
02f57c7a | 8483 | bio = orig_bio; |
c1dc0896 | 8484 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8485 | goto submit; |
8486 | } | |
8487 | ||
53b381b3 | 8488 | /* async crcs make it difficult to collect full stripe writes. */ |
1b86826d | 8489 | if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8490 | async_submit = 0; |
8491 | else | |
8492 | async_submit = 1; | |
8493 | ||
725130ba LB |
8494 | /* bio split */ |
8495 | ASSERT(map_length <= INT_MAX); | |
02f57c7a | 8496 | atomic_inc(&dip->pending_bios); |
3c91ee69 | 8497 | do { |
725130ba | 8498 | clone_len = min_t(int, submit_len, map_length); |
02f57c7a | 8499 | |
725130ba LB |
8500 | /* |
8501 | * This will never fail as it's passing GPF_NOFS and | |
8502 | * the allocation is backed by btrfs_bioset. | |
8503 | */ | |
e477094f | 8504 | bio = btrfs_bio_clone_partial(orig_bio, clone_offset, |
725130ba LB |
8505 | clone_len); |
8506 | bio->bi_private = dip; | |
8507 | bio->bi_end_io = btrfs_end_dio_bio; | |
8508 | btrfs_io_bio(bio)->logical = file_offset; | |
8509 | ||
8510 | ASSERT(submit_len >= clone_len); | |
8511 | submit_len -= clone_len; | |
8512 | if (submit_len == 0) | |
8513 | break; | |
e65e1535 | 8514 | |
725130ba LB |
8515 | /* |
8516 | * Increase the count before we submit the bio so we know | |
8517 | * the end IO handler won't happen before we increase the | |
8518 | * count. Otherwise, the dip might get freed before we're | |
8519 | * done setting it up. | |
8520 | */ | |
8521 | atomic_inc(&dip->pending_bios); | |
e65e1535 | 8522 | |
d0ee3934 | 8523 | status = btrfs_submit_dio_bio(bio, inode, file_offset, |
58efbc9f OS |
8524 | async_submit); |
8525 | if (status) { | |
725130ba LB |
8526 | bio_put(bio); |
8527 | atomic_dec(&dip->pending_bios); | |
8528 | goto out_err; | |
8529 | } | |
e65e1535 | 8530 | |
725130ba LB |
8531 | clone_offset += clone_len; |
8532 | start_sector += clone_len >> 9; | |
8533 | file_offset += clone_len; | |
5f4dc8fc | 8534 | |
725130ba LB |
8535 | map_length = submit_len; |
8536 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), | |
8537 | start_sector << 9, &map_length, NULL, 0); | |
8538 | if (ret) | |
8539 | goto out_err; | |
3c91ee69 | 8540 | } while (submit_len > 0); |
e65e1535 | 8541 | |
02f57c7a | 8542 | submit: |
d0ee3934 | 8543 | status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit); |
58efbc9f | 8544 | if (!status) |
e65e1535 MX |
8545 | return 0; |
8546 | ||
8547 | bio_put(bio); | |
8548 | out_err: | |
8549 | dip->errors = 1; | |
8550 | /* | |
de224b7c NB |
8551 | * Before atomic variable goto zero, we must make sure dip->errors is |
8552 | * perceived to be set. This ordering is ensured by the fact that an | |
8553 | * atomic operations with a return value are fully ordered as per | |
8554 | * atomic_t.txt | |
e65e1535 | 8555 | */ |
e65e1535 MX |
8556 | if (atomic_dec_and_test(&dip->pending_bios)) |
8557 | bio_io_error(dip->orig_bio); | |
8558 | ||
8559 | /* bio_end_io() will handle error, so we needn't return it */ | |
8560 | return 0; | |
8561 | } | |
8562 | ||
8a4c1e42 MC |
8563 | static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode, |
8564 | loff_t file_offset) | |
4b46fce2 | 8565 | { |
61de718f | 8566 | struct btrfs_dio_private *dip = NULL; |
3892ac90 LB |
8567 | struct bio *bio = NULL; |
8568 | struct btrfs_io_bio *io_bio; | |
8a4c1e42 | 8569 | bool write = (bio_op(dio_bio) == REQ_OP_WRITE); |
4b46fce2 JB |
8570 | int ret = 0; |
8571 | ||
8b6c1d56 | 8572 | bio = btrfs_bio_clone(dio_bio); |
9be3395b | 8573 | |
c1dc0896 | 8574 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8575 | if (!dip) { |
8576 | ret = -ENOMEM; | |
61de718f | 8577 | goto free_ordered; |
4b46fce2 | 8578 | } |
4b46fce2 | 8579 | |
9be3395b | 8580 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8581 | dip->inode = inode; |
8582 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8583 | dip->bytes = dio_bio->bi_iter.bi_size; |
8584 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
3892ac90 LB |
8585 | bio->bi_private = dip; |
8586 | dip->orig_bio = bio; | |
9be3395b | 8587 | dip->dio_bio = dio_bio; |
e65e1535 | 8588 | atomic_set(&dip->pending_bios, 0); |
3892ac90 LB |
8589 | io_bio = btrfs_io_bio(bio); |
8590 | io_bio->logical = file_offset; | |
4b46fce2 | 8591 | |
c1dc0896 | 8592 | if (write) { |
3892ac90 | 8593 | bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8594 | } else { |
3892ac90 | 8595 | bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8596 | dip->subio_endio = btrfs_subio_endio_read; |
8597 | } | |
4b46fce2 | 8598 | |
f28a4928 FM |
8599 | /* |
8600 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8601 | * even if we fail to submit a bio, because in such case we do the | |
8602 | * corresponding error handling below and it must not be done a second | |
8603 | * time by btrfs_direct_IO(). | |
8604 | */ | |
8605 | if (write) { | |
8606 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8607 | ||
8608 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8609 | dip->bytes; | |
8610 | dio_data->unsubmitted_oe_range_start = | |
8611 | dio_data->unsubmitted_oe_range_end; | |
8612 | } | |
8613 | ||
e6961cac | 8614 | ret = btrfs_submit_direct_hook(dip); |
e65e1535 | 8615 | if (!ret) |
eaf25d93 | 8616 | return; |
9be3395b | 8617 | |
3892ac90 LB |
8618 | if (io_bio->end_io) |
8619 | io_bio->end_io(io_bio, ret); | |
9be3395b | 8620 | |
4b46fce2 JB |
8621 | free_ordered: |
8622 | /* | |
61de718f FM |
8623 | * If we arrived here it means either we failed to submit the dip |
8624 | * or we either failed to clone the dio_bio or failed to allocate the | |
8625 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8626 | * call bio_endio against our io_bio so that we get proper resource | |
8627 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8628 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8629 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8630 | */ |
3892ac90 | 8631 | if (bio && dip) { |
054ec2f6 | 8632 | bio_io_error(bio); |
61de718f | 8633 | /* |
3892ac90 | 8634 | * The end io callbacks free our dip, do the final put on bio |
61de718f FM |
8635 | * and all the cleanup and final put for dio_bio (through |
8636 | * dio_end_io()). | |
8637 | */ | |
8638 | dip = NULL; | |
3892ac90 | 8639 | bio = NULL; |
61de718f | 8640 | } else { |
14543774 | 8641 | if (write) |
52427260 | 8642 | __endio_write_update_ordered(inode, |
14543774 FM |
8643 | file_offset, |
8644 | dio_bio->bi_iter.bi_size, | |
52427260 | 8645 | false); |
14543774 | 8646 | else |
61de718f FM |
8647 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8648 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8649 | |
4e4cbee9 | 8650 | dio_bio->bi_status = BLK_STS_IOERR; |
61de718f FM |
8651 | /* |
8652 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8653 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8654 | */ | |
4055351c | 8655 | dio_end_io(dio_bio); |
4b46fce2 | 8656 | } |
3892ac90 LB |
8657 | if (bio) |
8658 | bio_put(bio); | |
61de718f | 8659 | kfree(dip); |
4b46fce2 JB |
8660 | } |
8661 | ||
2ff7e61e | 8662 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
2ff7e61e | 8663 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
8664 | { |
8665 | int seg; | |
a1b75f7d | 8666 | int i; |
0b246afa | 8667 | unsigned int blocksize_mask = fs_info->sectorsize - 1; |
5a5f79b5 | 8668 | ssize_t retval = -EINVAL; |
5a5f79b5 CM |
8669 | |
8670 | if (offset & blocksize_mask) | |
8671 | goto out; | |
8672 | ||
28060d5d AV |
8673 | if (iov_iter_alignment(iter) & blocksize_mask) |
8674 | goto out; | |
a1b75f7d | 8675 | |
28060d5d | 8676 | /* If this is a write we don't need to check anymore */ |
cd27e455 | 8677 | if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter)) |
28060d5d AV |
8678 | return 0; |
8679 | /* | |
8680 | * Check to make sure we don't have duplicate iov_base's in this | |
8681 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8682 | * when reading back. | |
8683 | */ | |
8684 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8685 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8686 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8687 | goto out; |
8688 | } | |
5a5f79b5 CM |
8689 | } |
8690 | retval = 0; | |
8691 | out: | |
8692 | return retval; | |
8693 | } | |
eb838e73 | 8694 | |
c8b8e32d | 8695 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8696 | { |
4b46fce2 JB |
8697 | struct file *file = iocb->ki_filp; |
8698 | struct inode *inode = file->f_mapping->host; | |
0b246afa | 8699 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
50745b0a | 8700 | struct btrfs_dio_data dio_data = { 0 }; |
364ecf36 | 8701 | struct extent_changeset *data_reserved = NULL; |
c8b8e32d | 8702 | loff_t offset = iocb->ki_pos; |
0934856d | 8703 | size_t count = 0; |
2e60a51e | 8704 | int flags = 0; |
38851cc1 MX |
8705 | bool wakeup = true; |
8706 | bool relock = false; | |
0934856d | 8707 | ssize_t ret; |
4b46fce2 | 8708 | |
8c70c9f8 | 8709 | if (check_direct_IO(fs_info, iter, offset)) |
5a5f79b5 | 8710 | return 0; |
3f7c579c | 8711 | |
fe0f07d0 | 8712 | inode_dio_begin(inode); |
38851cc1 | 8713 | |
0e267c44 | 8714 | /* |
41bd9ca4 MX |
8715 | * The generic stuff only does filemap_write_and_wait_range, which |
8716 | * isn't enough if we've written compressed pages to this area, so | |
8717 | * we need to flush the dirty pages again to make absolutely sure | |
8718 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8719 | */ |
a6cbcd4a | 8720 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8721 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8722 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8723 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8724 | offset + count - 1); | |
0e267c44 | 8725 | |
6f673763 | 8726 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8727 | /* |
8728 | * If the write DIO is beyond the EOF, we need update | |
8729 | * the isize, but it is protected by i_mutex. So we can | |
8730 | * not unlock the i_mutex at this case. | |
8731 | */ | |
8732 | if (offset + count <= inode->i_size) { | |
4aaedfb0 | 8733 | dio_data.overwrite = 1; |
5955102c | 8734 | inode_unlock(inode); |
38851cc1 | 8735 | relock = true; |
edf064e7 GR |
8736 | } else if (iocb->ki_flags & IOCB_NOWAIT) { |
8737 | ret = -EAGAIN; | |
8738 | goto out; | |
38851cc1 | 8739 | } |
364ecf36 QW |
8740 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8741 | offset, count); | |
0934856d | 8742 | if (ret) |
38851cc1 | 8743 | goto out; |
e1cbbfa5 JB |
8744 | |
8745 | /* | |
8746 | * We need to know how many extents we reserved so that we can | |
8747 | * do the accounting properly if we go over the number we | |
8748 | * originally calculated. Abuse current->journal_info for this. | |
8749 | */ | |
da17066c | 8750 | dio_data.reserve = round_up(count, |
0b246afa | 8751 | fs_info->sectorsize); |
f28a4928 FM |
8752 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8753 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8754 | current->journal_info = &dio_data; |
97dcdea0 | 8755 | down_read(&BTRFS_I(inode)->dio_sem); |
ee39b432 DS |
8756 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8757 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8758 | inode_dio_end(inode); |
38851cc1 MX |
8759 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8760 | wakeup = false; | |
0934856d MX |
8761 | } |
8762 | ||
17f8c842 | 8763 | ret = __blockdev_direct_IO(iocb, inode, |
0b246afa | 8764 | fs_info->fs_devices->latest_bdev, |
c8b8e32d | 8765 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8766 | btrfs_submit_direct, flags); |
6f673763 | 8767 | if (iov_iter_rw(iter) == WRITE) { |
97dcdea0 | 8768 | up_read(&BTRFS_I(inode)->dio_sem); |
e1cbbfa5 | 8769 | current->journal_info = NULL; |
ddba1bfc | 8770 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8771 | if (dio_data.reserve) |
bc42bda2 | 8772 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 | 8773 | offset, dio_data.reserve, true); |
f28a4928 FM |
8774 | /* |
8775 | * On error we might have left some ordered extents | |
8776 | * without submitting corresponding bios for them, so | |
8777 | * cleanup them up to avoid other tasks getting them | |
8778 | * and waiting for them to complete forever. | |
8779 | */ | |
8780 | if (dio_data.unsubmitted_oe_range_start < | |
8781 | dio_data.unsubmitted_oe_range_end) | |
52427260 | 8782 | __endio_write_update_ordered(inode, |
f28a4928 FM |
8783 | dio_data.unsubmitted_oe_range_start, |
8784 | dio_data.unsubmitted_oe_range_end - | |
8785 | dio_data.unsubmitted_oe_range_start, | |
52427260 | 8786 | false); |
ddba1bfc | 8787 | } else if (ret >= 0 && (size_t)ret < count) |
bc42bda2 | 8788 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
8789 | offset, count - (size_t)ret, true); |
8790 | btrfs_delalloc_release_extents(BTRFS_I(inode), count, false); | |
0934856d | 8791 | } |
38851cc1 | 8792 | out: |
2e60a51e | 8793 | if (wakeup) |
fe0f07d0 | 8794 | inode_dio_end(inode); |
38851cc1 | 8795 | if (relock) |
5955102c | 8796 | inode_lock(inode); |
0934856d | 8797 | |
364ecf36 | 8798 | extent_changeset_free(data_reserved); |
0934856d | 8799 | return ret; |
16432985 CM |
8800 | } |
8801 | ||
05dadc09 TI |
8802 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8803 | ||
1506fcc8 YS |
8804 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8805 | __u64 start, __u64 len) | |
8806 | { | |
05dadc09 TI |
8807 | int ret; |
8808 | ||
8809 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8810 | if (ret) | |
8811 | return ret; | |
8812 | ||
2135fb9b | 8813 | return extent_fiemap(inode, fieinfo, start, len); |
1506fcc8 YS |
8814 | } |
8815 | ||
a52d9a80 | 8816 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8817 | { |
d1310b2e CM |
8818 | struct extent_io_tree *tree; |
8819 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8820 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8821 | } |
1832a6d5 | 8822 | |
a52d9a80 | 8823 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8824 | { |
be7bd730 JB |
8825 | struct inode *inode = page->mapping->host; |
8826 | int ret; | |
b888db2b CM |
8827 | |
8828 | if (current->flags & PF_MEMALLOC) { | |
8829 | redirty_page_for_writepage(wbc, page); | |
8830 | unlock_page(page); | |
8831 | return 0; | |
8832 | } | |
be7bd730 JB |
8833 | |
8834 | /* | |
8835 | * If we are under memory pressure we will call this directly from the | |
8836 | * VM, we need to make sure we have the inode referenced for the ordered | |
8837 | * extent. If not just return like we didn't do anything. | |
8838 | */ | |
8839 | if (!igrab(inode)) { | |
8840 | redirty_page_for_writepage(wbc, page); | |
8841 | return AOP_WRITEPAGE_ACTIVATE; | |
8842 | } | |
0a9b0e53 | 8843 | ret = extent_write_full_page(page, wbc); |
be7bd730 JB |
8844 | btrfs_add_delayed_iput(inode); |
8845 | return ret; | |
9ebefb18 CM |
8846 | } |
8847 | ||
48a3b636 ES |
8848 | static int btrfs_writepages(struct address_space *mapping, |
8849 | struct writeback_control *wbc) | |
b293f02e | 8850 | { |
8ae225a8 | 8851 | return extent_writepages(mapping, wbc); |
b293f02e CM |
8852 | } |
8853 | ||
3ab2fb5a CM |
8854 | static int |
8855 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8856 | struct list_head *pages, unsigned nr_pages) | |
8857 | { | |
2a3ff0ad | 8858 | return extent_readpages(mapping, pages, nr_pages); |
3ab2fb5a | 8859 | } |
2a3ff0ad | 8860 | |
e6dcd2dc | 8861 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8862 | { |
477a30ba | 8863 | int ret = try_release_extent_mapping(page, gfp_flags); |
a52d9a80 CM |
8864 | if (ret == 1) { |
8865 | ClearPagePrivate(page); | |
8866 | set_page_private(page, 0); | |
09cbfeaf | 8867 | put_page(page); |
39279cc3 | 8868 | } |
a52d9a80 | 8869 | return ret; |
39279cc3 CM |
8870 | } |
8871 | ||
e6dcd2dc CM |
8872 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8873 | { | |
98509cfc CM |
8874 | if (PageWriteback(page) || PageDirty(page)) |
8875 | return 0; | |
3ba7ab22 | 8876 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8877 | } |
8878 | ||
d47992f8 LC |
8879 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8880 | unsigned int length) | |
39279cc3 | 8881 | { |
5fd02043 | 8882 | struct inode *inode = page->mapping->host; |
d1310b2e | 8883 | struct extent_io_tree *tree; |
e6dcd2dc | 8884 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8885 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8886 | u64 page_start = page_offset(page); |
09cbfeaf | 8887 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8888 | u64 start; |
8889 | u64 end; | |
131e404a | 8890 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8891 | |
8b62b72b CM |
8892 | /* |
8893 | * we have the page locked, so new writeback can't start, | |
8894 | * and the dirty bit won't be cleared while we are here. | |
8895 | * | |
8896 | * Wait for IO on this page so that we can safely clear | |
8897 | * the PagePrivate2 bit and do ordered accounting | |
8898 | */ | |
e6dcd2dc | 8899 | wait_on_page_writeback(page); |
8b62b72b | 8900 | |
5fd02043 | 8901 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8902 | if (offset) { |
8903 | btrfs_releasepage(page, GFP_NOFS); | |
8904 | return; | |
8905 | } | |
131e404a FDBM |
8906 | |
8907 | if (!inode_evicting) | |
ff13db41 | 8908 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8909 | again: |
8910 | start = page_start; | |
a776c6fa | 8911 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
dbfdb6d1 | 8912 | page_end - start + 1); |
e6dcd2dc | 8913 | if (ordered) { |
dbfdb6d1 | 8914 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
8915 | /* |
8916 | * IO on this page will never be started, so we need | |
8917 | * to account for any ordered extents now | |
8918 | */ | |
131e404a | 8919 | if (!inode_evicting) |
dbfdb6d1 | 8920 | clear_extent_bit(tree, start, end, |
131e404a | 8921 | EXTENT_DIRTY | EXTENT_DELALLOC | |
a7e3b975 | 8922 | EXTENT_DELALLOC_NEW | |
131e404a | 8923 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
ae0f1625 | 8924 | EXTENT_DEFRAG, 1, 0, &cached_state); |
8b62b72b CM |
8925 | /* |
8926 | * whoever cleared the private bit is responsible | |
8927 | * for the finish_ordered_io | |
8928 | */ | |
77cef2ec JB |
8929 | if (TestClearPagePrivate2(page)) { |
8930 | struct btrfs_ordered_inode_tree *tree; | |
8931 | u64 new_len; | |
8932 | ||
8933 | tree = &BTRFS_I(inode)->ordered_tree; | |
8934 | ||
8935 | spin_lock_irq(&tree->lock); | |
8936 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8937 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8938 | if (new_len < ordered->truncated_len) |
8939 | ordered->truncated_len = new_len; | |
8940 | spin_unlock_irq(&tree->lock); | |
8941 | ||
8942 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
8943 | start, |
8944 | end - start + 1, 1)) | |
77cef2ec | 8945 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 8946 | } |
e6dcd2dc | 8947 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8948 | if (!inode_evicting) { |
8949 | cached_state = NULL; | |
dbfdb6d1 | 8950 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
8951 | &cached_state); |
8952 | } | |
dbfdb6d1 CR |
8953 | |
8954 | start = end + 1; | |
8955 | if (start < page_end) | |
8956 | goto again; | |
131e404a FDBM |
8957 | } |
8958 | ||
b9d0b389 QW |
8959 | /* |
8960 | * Qgroup reserved space handler | |
8961 | * Page here will be either | |
8962 | * 1) Already written to disk | |
8963 | * In this case, its reserved space is released from data rsv map | |
8964 | * and will be freed by delayed_ref handler finally. | |
8965 | * So even we call qgroup_free_data(), it won't decrease reserved | |
8966 | * space. | |
8967 | * 2) Not written to disk | |
0b34c261 GR |
8968 | * This means the reserved space should be freed here. However, |
8969 | * if a truncate invalidates the page (by clearing PageDirty) | |
8970 | * and the page is accounted for while allocating extent | |
8971 | * in btrfs_check_data_free_space() we let delayed_ref to | |
8972 | * free the entire extent. | |
b9d0b389 | 8973 | */ |
0b34c261 | 8974 | if (PageDirty(page)) |
bc42bda2 | 8975 | btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE); |
131e404a FDBM |
8976 | if (!inode_evicting) { |
8977 | clear_extent_bit(tree, page_start, page_end, | |
8978 | EXTENT_LOCKED | EXTENT_DIRTY | | |
a7e3b975 FM |
8979 | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
8980 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
ae0f1625 | 8981 | &cached_state); |
131e404a FDBM |
8982 | |
8983 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8984 | } |
e6dcd2dc | 8985 | |
4a096752 | 8986 | ClearPageChecked(page); |
9ad6b7bc | 8987 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8988 | ClearPagePrivate(page); |
8989 | set_page_private(page, 0); | |
09cbfeaf | 8990 | put_page(page); |
9ad6b7bc | 8991 | } |
39279cc3 CM |
8992 | } |
8993 | ||
9ebefb18 CM |
8994 | /* |
8995 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8996 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8997 | * be careful to check for EOF conditions here. We set the page up correctly | |
8998 | * for a written page which means we get ENOSPC checking when writing into | |
8999 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
9000 | * support these features. | |
9001 | * | |
9002 | * We are not allowed to take the i_mutex here so we have to play games to | |
9003 | * protect against truncate races as the page could now be beyond EOF. Because | |
d1342aad OS |
9004 | * truncate_setsize() writes the inode size before removing pages, once we have |
9005 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
9ebefb18 CM |
9006 | * beyond EOF, then the page is guaranteed safe against truncation until we |
9007 | * unlock the page. | |
9008 | */ | |
11bac800 | 9009 | int btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 9010 | { |
c2ec175c | 9011 | struct page *page = vmf->page; |
11bac800 | 9012 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 9013 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
9014 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
9015 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 9016 | struct extent_state *cached_state = NULL; |
364ecf36 | 9017 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc CM |
9018 | char *kaddr; |
9019 | unsigned long zero_start; | |
9ebefb18 | 9020 | loff_t size; |
1832a6d5 | 9021 | int ret; |
9998eb70 | 9022 | int reserved = 0; |
d0b7da88 | 9023 | u64 reserved_space; |
a52d9a80 | 9024 | u64 page_start; |
e6dcd2dc | 9025 | u64 page_end; |
d0b7da88 CR |
9026 | u64 end; |
9027 | ||
09cbfeaf | 9028 | reserved_space = PAGE_SIZE; |
9ebefb18 | 9029 | |
b2b5ef5c | 9030 | sb_start_pagefault(inode->i_sb); |
df480633 | 9031 | page_start = page_offset(page); |
09cbfeaf | 9032 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 9033 | end = page_end; |
df480633 | 9034 | |
d0b7da88 CR |
9035 | /* |
9036 | * Reserving delalloc space after obtaining the page lock can lead to | |
9037 | * deadlock. For example, if a dirty page is locked by this function | |
9038 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
9039 | * dirty page write out, then the btrfs_writepage() function could | |
9040 | * end up waiting indefinitely to get a lock on the page currently | |
9041 | * being processed by btrfs_page_mkwrite() function. | |
9042 | */ | |
364ecf36 | 9043 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
d0b7da88 | 9044 | reserved_space); |
9998eb70 | 9045 | if (!ret) { |
11bac800 | 9046 | ret = file_update_time(vmf->vma->vm_file); |
9998eb70 CM |
9047 | reserved = 1; |
9048 | } | |
56a76f82 NP |
9049 | if (ret) { |
9050 | if (ret == -ENOMEM) | |
9051 | ret = VM_FAULT_OOM; | |
9052 | else /* -ENOSPC, -EIO, etc */ | |
9053 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
9054 | if (reserved) |
9055 | goto out; | |
9056 | goto out_noreserve; | |
56a76f82 | 9057 | } |
1832a6d5 | 9058 | |
56a76f82 | 9059 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 9060 | again: |
9ebefb18 | 9061 | lock_page(page); |
9ebefb18 | 9062 | size = i_size_read(inode); |
a52d9a80 | 9063 | |
9ebefb18 | 9064 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 9065 | (page_start >= size)) { |
9ebefb18 CM |
9066 | /* page got truncated out from underneath us */ |
9067 | goto out_unlock; | |
9068 | } | |
e6dcd2dc CM |
9069 | wait_on_page_writeback(page); |
9070 | ||
ff13db41 | 9071 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
9072 | set_page_extent_mapped(page); |
9073 | ||
eb84ae03 CM |
9074 | /* |
9075 | * we can't set the delalloc bits if there are pending ordered | |
9076 | * extents. Drop our locks and wait for them to finish | |
9077 | */ | |
a776c6fa NB |
9078 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
9079 | PAGE_SIZE); | |
e6dcd2dc | 9080 | if (ordered) { |
2ac55d41 | 9081 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 9082 | &cached_state); |
e6dcd2dc | 9083 | unlock_page(page); |
eb84ae03 | 9084 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
9085 | btrfs_put_ordered_extent(ordered); |
9086 | goto again; | |
9087 | } | |
9088 | ||
09cbfeaf | 9089 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 9090 | reserved_space = round_up(size - page_start, |
0b246afa | 9091 | fs_info->sectorsize); |
09cbfeaf | 9092 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 9093 | end = page_start + reserved_space - 1; |
bc42bda2 | 9094 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
9095 | page_start, PAGE_SIZE - reserved_space, |
9096 | true); | |
d0b7da88 CR |
9097 | } |
9098 | } | |
9099 | ||
fbf19087 | 9100 | /* |
5416034f LB |
9101 | * page_mkwrite gets called when the page is firstly dirtied after it's |
9102 | * faulted in, but write(2) could also dirty a page and set delalloc | |
9103 | * bits, thus in this case for space account reason, we still need to | |
9104 | * clear any delalloc bits within this page range since we have to | |
9105 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 9106 | */ |
d0b7da88 | 9107 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
9108 | EXTENT_DIRTY | EXTENT_DELALLOC | |
9109 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 9110 | 0, 0, &cached_state); |
fbf19087 | 9111 | |
e3b8a485 | 9112 | ret = btrfs_set_extent_delalloc(inode, page_start, end, 0, |
ba8b04c1 | 9113 | &cached_state, 0); |
9ed74f2d | 9114 | if (ret) { |
2ac55d41 | 9115 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 9116 | &cached_state); |
9ed74f2d JB |
9117 | ret = VM_FAULT_SIGBUS; |
9118 | goto out_unlock; | |
9119 | } | |
e6dcd2dc | 9120 | ret = 0; |
9ebefb18 CM |
9121 | |
9122 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
9123 | if (page_start + PAGE_SIZE > size) |
9124 | zero_start = size & ~PAGE_MASK; | |
9ebefb18 | 9125 | else |
09cbfeaf | 9126 | zero_start = PAGE_SIZE; |
9ebefb18 | 9127 | |
09cbfeaf | 9128 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 9129 | kaddr = kmap(page); |
09cbfeaf | 9130 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
9131 | flush_dcache_page(page); |
9132 | kunmap(page); | |
9133 | } | |
247e743c | 9134 | ClearPageChecked(page); |
e6dcd2dc | 9135 | set_page_dirty(page); |
50a9b214 | 9136 | SetPageUptodate(page); |
5a3f23d5 | 9137 | |
0b246afa | 9138 | BTRFS_I(inode)->last_trans = fs_info->generation; |
257c62e1 | 9139 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; |
46d8bc34 | 9140 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 9141 | |
e43bbe5e | 9142 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state); |
9ebefb18 CM |
9143 | |
9144 | out_unlock: | |
b2b5ef5c | 9145 | if (!ret) { |
43b18595 | 9146 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true); |
b2b5ef5c | 9147 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 9148 | extent_changeset_free(data_reserved); |
50a9b214 | 9149 | return VM_FAULT_LOCKED; |
b2b5ef5c | 9150 | } |
9ebefb18 | 9151 | unlock_page(page); |
1832a6d5 | 9152 | out: |
43b18595 | 9153 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0)); |
bc42bda2 | 9154 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
43b18595 | 9155 | reserved_space, (ret != 0)); |
9998eb70 | 9156 | out_noreserve: |
b2b5ef5c | 9157 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 9158 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
9159 | return ret; |
9160 | } | |
9161 | ||
213e8c55 | 9162 | static int btrfs_truncate(struct inode *inode, bool skip_writeback) |
39279cc3 | 9163 | { |
0b246afa | 9164 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 9165 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 9166 | struct btrfs_block_rsv *rsv; |
a71754fc | 9167 | int ret = 0; |
3893e33b | 9168 | int err = 0; |
39279cc3 | 9169 | struct btrfs_trans_handle *trans; |
0b246afa JM |
9170 | u64 mask = fs_info->sectorsize - 1; |
9171 | u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); | |
39279cc3 | 9172 | |
213e8c55 FM |
9173 | if (!skip_writeback) { |
9174 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), | |
9175 | (u64)-1); | |
9176 | if (ret) | |
9177 | return ret; | |
9178 | } | |
39279cc3 | 9179 | |
fcb80c2a | 9180 | /* |
f7e9e8fc OS |
9181 | * Yes ladies and gentlemen, this is indeed ugly. We have a couple of |
9182 | * things going on here: | |
fcb80c2a | 9183 | * |
f7e9e8fc | 9184 | * 1) We need to reserve space to update our inode. |
fcb80c2a | 9185 | * |
f7e9e8fc | 9186 | * 2) We need to have something to cache all the space that is going to |
fcb80c2a JB |
9187 | * be free'd up by the truncate operation, but also have some slack |
9188 | * space reserved in case it uses space during the truncate (thank you | |
9189 | * very much snapshotting). | |
9190 | * | |
f7e9e8fc | 9191 | * And we need these to be separate. The fact is we can use a lot of |
fcb80c2a | 9192 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 9193 | * we will use, so we need the truncate reservation to be separate so it |
f7e9e8fc OS |
9194 | * doesn't end up using space reserved for updating the inode. We also |
9195 | * need to be able to stop the transaction and start a new one, which | |
9196 | * means we need to be able to update the inode several times, and we | |
9197 | * have no idea of knowing how many times that will be, so we can't just | |
9198 | * reserve 1 item for the entirety of the operation, so that has to be | |
9199 | * done separately as well. | |
fcb80c2a JB |
9200 | * |
9201 | * So that leaves us with | |
9202 | * | |
f7e9e8fc | 9203 | * 1) rsv - for the truncate reservation, which we will steal from the |
fcb80c2a | 9204 | * transaction reservation. |
f7e9e8fc | 9205 | * 2) fs_info->trans_block_rsv - this will have 1 items worth left for |
fcb80c2a JB |
9206 | * updating the inode. |
9207 | */ | |
2ff7e61e | 9208 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
9209 | if (!rsv) |
9210 | return -ENOMEM; | |
4a338542 | 9211 | rsv->size = min_size; |
ca7e70f5 | 9212 | rsv->failfast = 1; |
f0cd846e | 9213 | |
907cbceb | 9214 | /* |
07127184 | 9215 | * 1 for the truncate slack space |
907cbceb JB |
9216 | * 1 for updating the inode. |
9217 | */ | |
f3fe820c | 9218 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
9219 | if (IS_ERR(trans)) { |
9220 | err = PTR_ERR(trans); | |
9221 | goto out; | |
9222 | } | |
f0cd846e | 9223 | |
907cbceb | 9224 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 9225 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
25d609f8 | 9226 | min_size, 0); |
fcb80c2a | 9227 | BUG_ON(ret); |
f0cd846e | 9228 | |
5dc562c5 JB |
9229 | /* |
9230 | * So if we truncate and then write and fsync we normally would just | |
9231 | * write the extents that changed, which is a problem if we need to | |
9232 | * first truncate that entire inode. So set this flag so we write out | |
9233 | * all of the extents in the inode to the sync log so we're completely | |
9234 | * safe. | |
9235 | */ | |
9236 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9237 | trans->block_rsv = rsv; |
907cbceb | 9238 | |
8082510e YZ |
9239 | while (1) { |
9240 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9241 | inode->i_size, | |
9242 | BTRFS_EXTENT_DATA_KEY); | |
ddfae63c | 9243 | trans->block_rsv = &fs_info->trans_block_rsv; |
28ed1345 | 9244 | if (ret != -ENOSPC && ret != -EAGAIN) { |
d5014738 OS |
9245 | if (ret < 0) |
9246 | err = ret; | |
8082510e | 9247 | break; |
3893e33b | 9248 | } |
39279cc3 | 9249 | |
8082510e | 9250 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
9251 | if (ret) { |
9252 | err = ret; | |
9253 | break; | |
9254 | } | |
ca7e70f5 | 9255 | |
3a45bb20 | 9256 | btrfs_end_transaction(trans); |
2ff7e61e | 9257 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
9258 | |
9259 | trans = btrfs_start_transaction(root, 2); | |
9260 | if (IS_ERR(trans)) { | |
9261 | ret = err = PTR_ERR(trans); | |
9262 | trans = NULL; | |
9263 | break; | |
9264 | } | |
9265 | ||
47b5d646 | 9266 | btrfs_block_rsv_release(fs_info, rsv, -1); |
0b246afa | 9267 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
25d609f8 | 9268 | rsv, min_size, 0); |
ca7e70f5 JB |
9269 | BUG_ON(ret); /* shouldn't happen */ |
9270 | trans->block_rsv = rsv; | |
8082510e YZ |
9271 | } |
9272 | ||
ddfae63c JB |
9273 | /* |
9274 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
9275 | * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know | |
9276 | * we've truncated everything except the last little bit, and can do | |
9277 | * btrfs_truncate_block and then update the disk_i_size. | |
9278 | */ | |
9279 | if (ret == NEED_TRUNCATE_BLOCK) { | |
9280 | btrfs_end_transaction(trans); | |
9281 | btrfs_btree_balance_dirty(fs_info); | |
9282 | ||
9283 | ret = btrfs_truncate_block(inode, inode->i_size, 0, 0); | |
9284 | if (ret) | |
9285 | goto out; | |
9286 | trans = btrfs_start_transaction(root, 1); | |
9287 | if (IS_ERR(trans)) { | |
9288 | ret = PTR_ERR(trans); | |
9289 | goto out; | |
9290 | } | |
9291 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
9292 | } | |
9293 | ||
917c16b2 | 9294 | if (trans) { |
0b246afa | 9295 | trans->block_rsv = &fs_info->trans_block_rsv; |
917c16b2 CM |
9296 | ret = btrfs_update_inode(trans, root, inode); |
9297 | if (ret && !err) | |
9298 | err = ret; | |
7b128766 | 9299 | |
3a45bb20 | 9300 | ret = btrfs_end_transaction(trans); |
2ff7e61e | 9301 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 9302 | } |
fcb80c2a | 9303 | out: |
2ff7e61e | 9304 | btrfs_free_block_rsv(fs_info, rsv); |
fcb80c2a | 9305 | |
3893e33b JB |
9306 | if (ret && !err) |
9307 | err = ret; | |
a41ad394 | 9308 | |
3893e33b | 9309 | return err; |
39279cc3 CM |
9310 | } |
9311 | ||
d352ac68 CM |
9312 | /* |
9313 | * create a new subvolume directory/inode (helper for the ioctl). | |
9314 | */ | |
d2fb3437 | 9315 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9316 | struct btrfs_root *new_root, |
9317 | struct btrfs_root *parent_root, | |
9318 | u64 new_dirid) | |
39279cc3 | 9319 | { |
39279cc3 | 9320 | struct inode *inode; |
76dda93c | 9321 | int err; |
00e4e6b3 | 9322 | u64 index = 0; |
39279cc3 | 9323 | |
12fc9d09 FA |
9324 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9325 | new_dirid, new_dirid, | |
9326 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9327 | &index); | |
54aa1f4d | 9328 | if (IS_ERR(inode)) |
f46b5a66 | 9329 | return PTR_ERR(inode); |
39279cc3 CM |
9330 | inode->i_op = &btrfs_dir_inode_operations; |
9331 | inode->i_fop = &btrfs_dir_file_operations; | |
9332 | ||
bfe86848 | 9333 | set_nlink(inode, 1); |
6ef06d27 | 9334 | btrfs_i_size_write(BTRFS_I(inode), 0); |
b0d5d10f | 9335 | unlock_new_inode(inode); |
3b96362c | 9336 | |
63541927 FDBM |
9337 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9338 | if (err) | |
9339 | btrfs_err(new_root->fs_info, | |
351fd353 | 9340 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9341 | new_root->root_key.objectid, err); |
9342 | ||
76dda93c | 9343 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9344 | |
76dda93c | 9345 | iput(inode); |
ce598979 | 9346 | return err; |
39279cc3 CM |
9347 | } |
9348 | ||
39279cc3 CM |
9349 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9350 | { | |
69fe2d75 | 9351 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 9352 | struct btrfs_inode *ei; |
2ead6ae7 | 9353 | struct inode *inode; |
39279cc3 | 9354 | |
712e36c5 | 9355 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL); |
39279cc3 CM |
9356 | if (!ei) |
9357 | return NULL; | |
2ead6ae7 YZ |
9358 | |
9359 | ei->root = NULL; | |
2ead6ae7 | 9360 | ei->generation = 0; |
15ee9bc7 | 9361 | ei->last_trans = 0; |
257c62e1 | 9362 | ei->last_sub_trans = 0; |
e02119d5 | 9363 | ei->logged_trans = 0; |
2ead6ae7 | 9364 | ei->delalloc_bytes = 0; |
a7e3b975 | 9365 | ei->new_delalloc_bytes = 0; |
47059d93 | 9366 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9367 | ei->disk_i_size = 0; |
9368 | ei->flags = 0; | |
7709cde3 | 9369 | ei->csum_bytes = 0; |
2ead6ae7 | 9370 | ei->index_cnt = (u64)-1; |
67de1176 | 9371 | ei->dir_index = 0; |
2ead6ae7 | 9372 | ei->last_unlink_trans = 0; |
46d8bc34 | 9373 | ei->last_log_commit = 0; |
2ead6ae7 | 9374 | |
9e0baf60 JB |
9375 | spin_lock_init(&ei->lock); |
9376 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
9377 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
9378 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
9379 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 9380 | ei->runtime_flags = 0; |
b52aa8c9 | 9381 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 9382 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9383 | |
16cdcec7 MX |
9384 | ei->delayed_node = NULL; |
9385 | ||
9cc97d64 | 9386 | ei->i_otime.tv_sec = 0; |
9387 | ei->i_otime.tv_nsec = 0; | |
9388 | ||
2ead6ae7 | 9389 | inode = &ei->vfs_inode; |
a8067e02 | 9390 | extent_map_tree_init(&ei->extent_tree); |
c6100a4b JB |
9391 | extent_io_tree_init(&ei->io_tree, inode); |
9392 | extent_io_tree_init(&ei->io_failure_tree, inode); | |
0b32f4bb JB |
9393 | ei->io_tree.track_uptodate = 1; |
9394 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 9395 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9396 | mutex_init(&ei->log_mutex); |
f248679e | 9397 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9398 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9399 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9400 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9401 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 9402 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
9403 | |
9404 | return inode; | |
39279cc3 CM |
9405 | } |
9406 | ||
aaedb55b JB |
9407 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9408 | void btrfs_test_destroy_inode(struct inode *inode) | |
9409 | { | |
dcdbc059 | 9410 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
9411 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9412 | } | |
9413 | #endif | |
9414 | ||
fa0d7e3d NP |
9415 | static void btrfs_i_callback(struct rcu_head *head) |
9416 | { | |
9417 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
9418 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9419 | } | |
9420 | ||
39279cc3 CM |
9421 | void btrfs_destroy_inode(struct inode *inode) |
9422 | { | |
0b246afa | 9423 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 9424 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9425 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9426 | ||
b3d9b7a3 | 9427 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9428 | WARN_ON(inode->i_data.nrpages); |
69fe2d75 JB |
9429 | WARN_ON(BTRFS_I(inode)->block_rsv.reserved); |
9430 | WARN_ON(BTRFS_I(inode)->block_rsv.size); | |
9e0baf60 | 9431 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7709cde3 | 9432 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
a7e3b975 | 9433 | WARN_ON(BTRFS_I(inode)->new_delalloc_bytes); |
7709cde3 | 9434 | WARN_ON(BTRFS_I(inode)->csum_bytes); |
47059d93 | 9435 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9436 | |
a6dbd429 JB |
9437 | /* |
9438 | * This can happen where we create an inode, but somebody else also | |
9439 | * created the same inode and we need to destroy the one we already | |
9440 | * created. | |
9441 | */ | |
9442 | if (!root) | |
9443 | goto free; | |
9444 | ||
8a35d95f JB |
9445 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
9446 | &BTRFS_I(inode)->runtime_flags)) { | |
0b246afa | 9447 | btrfs_info(fs_info, "inode %llu still on the orphan list", |
4a0cc7ca | 9448 | btrfs_ino(BTRFS_I(inode))); |
8a35d95f | 9449 | atomic_dec(&root->orphan_inodes); |
7b128766 | 9450 | } |
7b128766 | 9451 | |
d397712b | 9452 | while (1) { |
e6dcd2dc CM |
9453 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9454 | if (!ordered) | |
9455 | break; | |
9456 | else { | |
0b246afa | 9457 | btrfs_err(fs_info, |
5d163e0e JM |
9458 | "found ordered extent %llu %llu on inode cleanup", |
9459 | ordered->file_offset, ordered->len); | |
e6dcd2dc CM |
9460 | btrfs_remove_ordered_extent(inode, ordered); |
9461 | btrfs_put_ordered_extent(ordered); | |
9462 | btrfs_put_ordered_extent(ordered); | |
9463 | } | |
9464 | } | |
56fa9d07 | 9465 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9466 | inode_tree_del(inode); |
dcdbc059 | 9467 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
a6dbd429 | 9468 | free: |
fa0d7e3d | 9469 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
9470 | } |
9471 | ||
45321ac5 | 9472 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9473 | { |
9474 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9475 | |
6379ef9f NA |
9476 | if (root == NULL) |
9477 | return 1; | |
9478 | ||
fa6ac876 | 9479 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9480 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9481 | return 1; |
76dda93c | 9482 | else |
45321ac5 | 9483 | return generic_drop_inode(inode); |
76dda93c YZ |
9484 | } |
9485 | ||
0ee0fda0 | 9486 | static void init_once(void *foo) |
39279cc3 CM |
9487 | { |
9488 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9489 | ||
9490 | inode_init_once(&ei->vfs_inode); | |
9491 | } | |
9492 | ||
e67c718b | 9493 | void __cold btrfs_destroy_cachep(void) |
39279cc3 | 9494 | { |
8c0a8537 KS |
9495 | /* |
9496 | * Make sure all delayed rcu free inodes are flushed before we | |
9497 | * destroy cache. | |
9498 | */ | |
9499 | rcu_barrier(); | |
5598e900 KM |
9500 | kmem_cache_destroy(btrfs_inode_cachep); |
9501 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
9502 | kmem_cache_destroy(btrfs_path_cachep); |
9503 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
9504 | } |
9505 | ||
f5c29bd9 | 9506 | int __init btrfs_init_cachep(void) |
39279cc3 | 9507 | { |
837e1972 | 9508 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9509 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9510 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9511 | init_once); | |
39279cc3 CM |
9512 | if (!btrfs_inode_cachep) |
9513 | goto fail; | |
9601e3f6 | 9514 | |
837e1972 | 9515 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 9516 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 9517 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9518 | if (!btrfs_trans_handle_cachep) |
9519 | goto fail; | |
9601e3f6 | 9520 | |
837e1972 | 9521 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 9522 | sizeof(struct btrfs_path), 0, |
fba4b697 | 9523 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9524 | if (!btrfs_path_cachep) |
9525 | goto fail; | |
9601e3f6 | 9526 | |
837e1972 | 9527 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 9528 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 9529 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
9530 | if (!btrfs_free_space_cachep) |
9531 | goto fail; | |
9532 | ||
39279cc3 CM |
9533 | return 0; |
9534 | fail: | |
9535 | btrfs_destroy_cachep(); | |
9536 | return -ENOMEM; | |
9537 | } | |
9538 | ||
a528d35e DH |
9539 | static int btrfs_getattr(const struct path *path, struct kstat *stat, |
9540 | u32 request_mask, unsigned int flags) | |
39279cc3 | 9541 | { |
df0af1a5 | 9542 | u64 delalloc_bytes; |
a528d35e | 9543 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 9544 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 YS |
9545 | u32 bi_flags = BTRFS_I(inode)->flags; |
9546 | ||
9547 | stat->result_mask |= STATX_BTIME; | |
9548 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
9549 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
9550 | if (bi_flags & BTRFS_INODE_APPEND) | |
9551 | stat->attributes |= STATX_ATTR_APPEND; | |
9552 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
9553 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
9554 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
9555 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
9556 | if (bi_flags & BTRFS_INODE_NODUMP) | |
9557 | stat->attributes |= STATX_ATTR_NODUMP; | |
9558 | ||
9559 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
9560 | STATX_ATTR_COMPRESSED | | |
9561 | STATX_ATTR_IMMUTABLE | | |
9562 | STATX_ATTR_NODUMP); | |
fadc0d8b | 9563 | |
39279cc3 | 9564 | generic_fillattr(inode, stat); |
0ee5dc67 | 9565 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9566 | |
9567 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 9568 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
df0af1a5 | 9569 | spin_unlock(&BTRFS_I(inode)->lock); |
fadc0d8b | 9570 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9571 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9572 | return 0; |
9573 | } | |
9574 | ||
cdd1fedf DF |
9575 | static int btrfs_rename_exchange(struct inode *old_dir, |
9576 | struct dentry *old_dentry, | |
9577 | struct inode *new_dir, | |
9578 | struct dentry *new_dentry) | |
9579 | { | |
0b246afa | 9580 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
9581 | struct btrfs_trans_handle *trans; |
9582 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9583 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9584 | struct inode *new_inode = new_dentry->d_inode; | |
9585 | struct inode *old_inode = old_dentry->d_inode; | |
c2050a45 | 9586 | struct timespec ctime = current_time(old_inode); |
cdd1fedf | 9587 | struct dentry *parent; |
4a0cc7ca NB |
9588 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9589 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9590 | u64 old_idx = 0; |
9591 | u64 new_idx = 0; | |
9592 | u64 root_objectid; | |
9593 | int ret; | |
86e8aa0e FM |
9594 | bool root_log_pinned = false; |
9595 | bool dest_log_pinned = false; | |
cdd1fedf DF |
9596 | |
9597 | /* we only allow rename subvolume link between subvolumes */ | |
9598 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
9599 | return -EXDEV; | |
9600 | ||
9601 | /* close the race window with snapshot create/destroy ioctl */ | |
9602 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9603 | down_read(&fs_info->subvol_sem); |
cdd1fedf | 9604 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9605 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9606 | |
9607 | /* | |
9608 | * We want to reserve the absolute worst case amount of items. So if | |
9609 | * both inodes are subvols and we need to unlink them then that would | |
9610 | * require 4 item modifications, but if they are both normal inodes it | |
9611 | * would require 5 item modifications, so we'll assume their normal | |
9612 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9613 | * should cover the worst case number of items we'll modify. | |
9614 | */ | |
9615 | trans = btrfs_start_transaction(root, 12); | |
9616 | if (IS_ERR(trans)) { | |
9617 | ret = PTR_ERR(trans); | |
9618 | goto out_notrans; | |
9619 | } | |
9620 | ||
9621 | /* | |
9622 | * We need to find a free sequence number both in the source and | |
9623 | * in the destination directory for the exchange. | |
9624 | */ | |
877574e2 | 9625 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
9626 | if (ret) |
9627 | goto out_fail; | |
877574e2 | 9628 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
9629 | if (ret) |
9630 | goto out_fail; | |
9631 | ||
9632 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9633 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9634 | ||
9635 | /* Reference for the source. */ | |
9636 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9637 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9638 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9639 | } else { |
376e5a57 FM |
9640 | btrfs_pin_log_trans(root); |
9641 | root_log_pinned = true; | |
cdd1fedf DF |
9642 | ret = btrfs_insert_inode_ref(trans, dest, |
9643 | new_dentry->d_name.name, | |
9644 | new_dentry->d_name.len, | |
9645 | old_ino, | |
f85b7379 DS |
9646 | btrfs_ino(BTRFS_I(new_dir)), |
9647 | old_idx); | |
cdd1fedf DF |
9648 | if (ret) |
9649 | goto out_fail; | |
cdd1fedf DF |
9650 | } |
9651 | ||
9652 | /* And now for the dest. */ | |
9653 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9654 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9655 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9656 | } else { |
376e5a57 FM |
9657 | btrfs_pin_log_trans(dest); |
9658 | dest_log_pinned = true; | |
cdd1fedf DF |
9659 | ret = btrfs_insert_inode_ref(trans, root, |
9660 | old_dentry->d_name.name, | |
9661 | old_dentry->d_name.len, | |
9662 | new_ino, | |
f85b7379 DS |
9663 | btrfs_ino(BTRFS_I(old_dir)), |
9664 | new_idx); | |
cdd1fedf DF |
9665 | if (ret) |
9666 | goto out_fail; | |
cdd1fedf DF |
9667 | } |
9668 | ||
9669 | /* Update inode version and ctime/mtime. */ | |
9670 | inode_inc_iversion(old_dir); | |
9671 | inode_inc_iversion(new_dir); | |
9672 | inode_inc_iversion(old_inode); | |
9673 | inode_inc_iversion(new_inode); | |
9674 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9675 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9676 | old_inode->i_ctime = ctime; | |
9677 | new_inode->i_ctime = ctime; | |
9678 | ||
9679 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9680 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9681 | BTRFS_I(old_inode), 1); | |
9682 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9683 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9684 | } |
9685 | ||
9686 | /* src is a subvolume */ | |
9687 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9688 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
9689 | ret = btrfs_unlink_subvol(trans, root, old_dir, | |
9690 | root_objectid, | |
9691 | old_dentry->d_name.name, | |
9692 | old_dentry->d_name.len); | |
9693 | } else { /* src is an inode */ | |
4ec5934e NB |
9694 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9695 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
9696 | old_dentry->d_name.name, |
9697 | old_dentry->d_name.len); | |
9698 | if (!ret) | |
9699 | ret = btrfs_update_inode(trans, root, old_inode); | |
9700 | } | |
9701 | if (ret) { | |
66642832 | 9702 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9703 | goto out_fail; |
9704 | } | |
9705 | ||
9706 | /* dest is a subvolume */ | |
9707 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9708 | root_objectid = BTRFS_I(new_inode)->root->root_key.objectid; | |
9709 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9710 | root_objectid, | |
9711 | new_dentry->d_name.name, | |
9712 | new_dentry->d_name.len); | |
9713 | } else { /* dest is an inode */ | |
4ec5934e NB |
9714 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9715 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
9716 | new_dentry->d_name.name, |
9717 | new_dentry->d_name.len); | |
9718 | if (!ret) | |
9719 | ret = btrfs_update_inode(trans, dest, new_inode); | |
9720 | } | |
9721 | if (ret) { | |
66642832 | 9722 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9723 | goto out_fail; |
9724 | } | |
9725 | ||
db0a669f | 9726 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9727 | new_dentry->d_name.name, |
9728 | new_dentry->d_name.len, 0, old_idx); | |
9729 | if (ret) { | |
66642832 | 9730 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9731 | goto out_fail; |
9732 | } | |
9733 | ||
db0a669f | 9734 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9735 | old_dentry->d_name.name, |
9736 | old_dentry->d_name.len, 0, new_idx); | |
9737 | if (ret) { | |
66642832 | 9738 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9739 | goto out_fail; |
9740 | } | |
9741 | ||
9742 | if (old_inode->i_nlink == 1) | |
9743 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9744 | if (new_inode->i_nlink == 1) | |
9745 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9746 | ||
86e8aa0e | 9747 | if (root_log_pinned) { |
cdd1fedf | 9748 | parent = new_dentry->d_parent; |
f85b7379 DS |
9749 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9750 | parent); | |
cdd1fedf | 9751 | btrfs_end_log_trans(root); |
86e8aa0e | 9752 | root_log_pinned = false; |
cdd1fedf | 9753 | } |
86e8aa0e | 9754 | if (dest_log_pinned) { |
cdd1fedf | 9755 | parent = old_dentry->d_parent; |
f85b7379 DS |
9756 | btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir), |
9757 | parent); | |
cdd1fedf | 9758 | btrfs_end_log_trans(dest); |
86e8aa0e | 9759 | dest_log_pinned = false; |
cdd1fedf DF |
9760 | } |
9761 | out_fail: | |
86e8aa0e FM |
9762 | /* |
9763 | * If we have pinned a log and an error happened, we unpin tasks | |
9764 | * trying to sync the log and force them to fallback to a transaction | |
9765 | * commit if the log currently contains any of the inodes involved in | |
9766 | * this rename operation (to ensure we do not persist a log with an | |
9767 | * inconsistent state for any of these inodes or leading to any | |
9768 | * inconsistencies when replayed). If the transaction was aborted, the | |
9769 | * abortion reason is propagated to userspace when attempting to commit | |
9770 | * the transaction. If the log does not contain any of these inodes, we | |
9771 | * allow the tasks to sync it. | |
9772 | */ | |
9773 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
9774 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9775 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9776 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 9777 | (new_inode && |
0f8939b8 | 9778 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 9779 | btrfs_set_log_full_commit(fs_info, trans); |
86e8aa0e FM |
9780 | |
9781 | if (root_log_pinned) { | |
9782 | btrfs_end_log_trans(root); | |
9783 | root_log_pinned = false; | |
9784 | } | |
9785 | if (dest_log_pinned) { | |
9786 | btrfs_end_log_trans(dest); | |
9787 | dest_log_pinned = false; | |
9788 | } | |
9789 | } | |
3a45bb20 | 9790 | ret = btrfs_end_transaction(trans); |
cdd1fedf DF |
9791 | out_notrans: |
9792 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9793 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 9794 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9795 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9796 | |
9797 | return ret; | |
9798 | } | |
9799 | ||
9800 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9801 | struct btrfs_root *root, | |
9802 | struct inode *dir, | |
9803 | struct dentry *dentry) | |
9804 | { | |
9805 | int ret; | |
9806 | struct inode *inode; | |
9807 | u64 objectid; | |
9808 | u64 index; | |
9809 | ||
9810 | ret = btrfs_find_free_ino(root, &objectid); | |
9811 | if (ret) | |
9812 | return ret; | |
9813 | ||
9814 | inode = btrfs_new_inode(trans, root, dir, | |
9815 | dentry->d_name.name, | |
9816 | dentry->d_name.len, | |
4a0cc7ca | 9817 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9818 | objectid, |
9819 | S_IFCHR | WHITEOUT_MODE, | |
9820 | &index); | |
9821 | ||
9822 | if (IS_ERR(inode)) { | |
9823 | ret = PTR_ERR(inode); | |
9824 | return ret; | |
9825 | } | |
9826 | ||
9827 | inode->i_op = &btrfs_special_inode_operations; | |
9828 | init_special_inode(inode, inode->i_mode, | |
9829 | WHITEOUT_DEV); | |
9830 | ||
9831 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9832 | &dentry->d_name); | |
9833 | if (ret) | |
c9901618 | 9834 | goto out; |
cdd1fedf | 9835 | |
cef415af NB |
9836 | ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9837 | BTRFS_I(inode), 0, index); | |
cdd1fedf | 9838 | if (ret) |
c9901618 | 9839 | goto out; |
cdd1fedf DF |
9840 | |
9841 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 9842 | out: |
cdd1fedf | 9843 | unlock_new_inode(inode); |
c9901618 FM |
9844 | if (ret) |
9845 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9846 | iput(inode); |
9847 | ||
c9901618 | 9848 | return ret; |
cdd1fedf DF |
9849 | } |
9850 | ||
d397712b | 9851 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9852 | struct inode *new_dir, struct dentry *new_dentry, |
9853 | unsigned int flags) | |
39279cc3 | 9854 | { |
0b246afa | 9855 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9856 | struct btrfs_trans_handle *trans; |
5062af35 | 9857 | unsigned int trans_num_items; |
39279cc3 | 9858 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9859 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9860 | struct inode *new_inode = d_inode(new_dentry); |
9861 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9862 | u64 index = 0; |
4df27c4d | 9863 | u64 root_objectid; |
39279cc3 | 9864 | int ret; |
4a0cc7ca | 9865 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 9866 | bool log_pinned = false; |
39279cc3 | 9867 | |
4a0cc7ca | 9868 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9869 | return -EPERM; |
9870 | ||
4df27c4d | 9871 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9872 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9873 | return -EXDEV; |
9874 | ||
33345d01 | 9875 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9876 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9877 | return -ENOTEMPTY; |
5f39d397 | 9878 | |
4df27c4d YZ |
9879 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9880 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9881 | return -ENOTEMPTY; | |
9c52057c CM |
9882 | |
9883 | ||
9884 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9885 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9886 | new_dentry->d_name.name, |
9887 | new_dentry->d_name.len); | |
9888 | ||
9889 | if (ret) { | |
9890 | if (ret == -EEXIST) { | |
9891 | /* we shouldn't get | |
9892 | * eexist without a new_inode */ | |
fae7f21c | 9893 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9894 | return ret; |
9895 | } | |
9896 | } else { | |
9897 | /* maybe -EOVERFLOW */ | |
9898 | return ret; | |
9899 | } | |
9900 | } | |
9901 | ret = 0; | |
9902 | ||
5a3f23d5 | 9903 | /* |
8d875f95 CM |
9904 | * we're using rename to replace one file with another. Start IO on it |
9905 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9906 | */ |
8d875f95 | 9907 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9908 | filemap_flush(old_inode->i_mapping); |
9909 | ||
76dda93c | 9910 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9911 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9912 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
9913 | /* |
9914 | * We want to reserve the absolute worst case amount of items. So if | |
9915 | * both inodes are subvols and we need to unlink them then that would | |
9916 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9917 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9918 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9919 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9920 | * If our rename has the whiteout flag, we need more 5 units for the |
9921 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9922 | * when selinux is enabled). | |
a22285a6 | 9923 | */ |
5062af35 FM |
9924 | trans_num_items = 11; |
9925 | if (flags & RENAME_WHITEOUT) | |
9926 | trans_num_items += 5; | |
9927 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9928 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9929 | ret = PTR_ERR(trans); |
9930 | goto out_notrans; | |
9931 | } | |
76dda93c | 9932 | |
4df27c4d YZ |
9933 | if (dest != root) |
9934 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9935 | |
877574e2 | 9936 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9937 | if (ret) |
9938 | goto out_fail; | |
5a3f23d5 | 9939 | |
67de1176 | 9940 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9941 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9942 | /* force full log commit if subvolume involved. */ |
0b246afa | 9943 | btrfs_set_log_full_commit(fs_info, trans); |
4df27c4d | 9944 | } else { |
c4aba954 FM |
9945 | btrfs_pin_log_trans(root); |
9946 | log_pinned = true; | |
a5719521 YZ |
9947 | ret = btrfs_insert_inode_ref(trans, dest, |
9948 | new_dentry->d_name.name, | |
9949 | new_dentry->d_name.len, | |
33345d01 | 9950 | old_ino, |
4a0cc7ca | 9951 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
9952 | if (ret) |
9953 | goto out_fail; | |
4df27c4d | 9954 | } |
5a3f23d5 | 9955 | |
0c4d2d95 JB |
9956 | inode_inc_iversion(old_dir); |
9957 | inode_inc_iversion(new_dir); | |
9958 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9959 | old_dir->i_ctime = old_dir->i_mtime = |
9960 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 9961 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 9962 | |
12fcfd22 | 9963 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
9964 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9965 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 9966 | |
33345d01 | 9967 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
9968 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
9969 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
9970 | old_dentry->d_name.name, | |
9971 | old_dentry->d_name.len); | |
9972 | } else { | |
4ec5934e NB |
9973 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9974 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
9975 | old_dentry->d_name.name, |
9976 | old_dentry->d_name.len); | |
9977 | if (!ret) | |
9978 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 9979 | } |
79787eaa | 9980 | if (ret) { |
66642832 | 9981 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9982 | goto out_fail; |
9983 | } | |
39279cc3 CM |
9984 | |
9985 | if (new_inode) { | |
0c4d2d95 | 9986 | inode_inc_iversion(new_inode); |
c2050a45 | 9987 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9988 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d YZ |
9989 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
9990 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
9991 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9992 | root_objectid, | |
9993 | new_dentry->d_name.name, | |
9994 | new_dentry->d_name.len); | |
9995 | BUG_ON(new_inode->i_nlink == 0); | |
9996 | } else { | |
4ec5934e NB |
9997 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9998 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
9999 | new_dentry->d_name.name, |
10000 | new_dentry->d_name.len); | |
10001 | } | |
4ef31a45 | 10002 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
10003 | ret = btrfs_orphan_add(trans, |
10004 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 10005 | if (ret) { |
66642832 | 10006 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
10007 | goto out_fail; |
10008 | } | |
39279cc3 | 10009 | } |
aec7477b | 10010 | |
db0a669f | 10011 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 10012 | new_dentry->d_name.name, |
a5719521 | 10013 | new_dentry->d_name.len, 0, index); |
79787eaa | 10014 | if (ret) { |
66642832 | 10015 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
10016 | goto out_fail; |
10017 | } | |
39279cc3 | 10018 | |
67de1176 MX |
10019 | if (old_inode->i_nlink == 1) |
10020 | BTRFS_I(old_inode)->dir_index = index; | |
10021 | ||
3dc9e8f7 | 10022 | if (log_pinned) { |
10d9f309 | 10023 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 10024 | |
f85b7379 DS |
10025 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
10026 | parent); | |
4df27c4d | 10027 | btrfs_end_log_trans(root); |
3dc9e8f7 | 10028 | log_pinned = false; |
4df27c4d | 10029 | } |
cdd1fedf DF |
10030 | |
10031 | if (flags & RENAME_WHITEOUT) { | |
10032 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
10033 | old_dentry); | |
10034 | ||
10035 | if (ret) { | |
66642832 | 10036 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
10037 | goto out_fail; |
10038 | } | |
4df27c4d | 10039 | } |
39279cc3 | 10040 | out_fail: |
3dc9e8f7 FM |
10041 | /* |
10042 | * If we have pinned the log and an error happened, we unpin tasks | |
10043 | * trying to sync the log and force them to fallback to a transaction | |
10044 | * commit if the log currently contains any of the inodes involved in | |
10045 | * this rename operation (to ensure we do not persist a log with an | |
10046 | * inconsistent state for any of these inodes or leading to any | |
10047 | * inconsistencies when replayed). If the transaction was aborted, the | |
10048 | * abortion reason is propagated to userspace when attempting to commit | |
10049 | * the transaction. If the log does not contain any of these inodes, we | |
10050 | * allow the tasks to sync it. | |
10051 | */ | |
10052 | if (ret && log_pinned) { | |
0f8939b8 NB |
10053 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
10054 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
10055 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 10056 | (new_inode && |
0f8939b8 | 10057 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 10058 | btrfs_set_log_full_commit(fs_info, trans); |
3dc9e8f7 FM |
10059 | |
10060 | btrfs_end_log_trans(root); | |
10061 | log_pinned = false; | |
10062 | } | |
3a45bb20 | 10063 | btrfs_end_transaction(trans); |
b44c59a8 | 10064 | out_notrans: |
33345d01 | 10065 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 10066 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 10067 | |
39279cc3 CM |
10068 | return ret; |
10069 | } | |
10070 | ||
80ace85c MS |
10071 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
10072 | struct inode *new_dir, struct dentry *new_dentry, | |
10073 | unsigned int flags) | |
10074 | { | |
cdd1fedf | 10075 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
10076 | return -EINVAL; |
10077 | ||
cdd1fedf DF |
10078 | if (flags & RENAME_EXCHANGE) |
10079 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
10080 | new_dentry); | |
10081 | ||
10082 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
10083 | } |
10084 | ||
3a2f8c07 NB |
10085 | struct btrfs_delalloc_work { |
10086 | struct inode *inode; | |
10087 | struct completion completion; | |
10088 | struct list_head list; | |
10089 | struct btrfs_work work; | |
10090 | }; | |
10091 | ||
8ccf6f19 MX |
10092 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
10093 | { | |
10094 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 10095 | struct inode *inode; |
8ccf6f19 MX |
10096 | |
10097 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
10098 | work); | |
9f23e289 | 10099 | inode = delalloc_work->inode; |
30424601 DS |
10100 | filemap_flush(inode->i_mapping); |
10101 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
10102 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 10103 | filemap_flush(inode->i_mapping); |
8ccf6f19 | 10104 | |
076da91c | 10105 | iput(inode); |
8ccf6f19 MX |
10106 | complete(&delalloc_work->completion); |
10107 | } | |
10108 | ||
3a2f8c07 | 10109 | static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode) |
8ccf6f19 MX |
10110 | { |
10111 | struct btrfs_delalloc_work *work; | |
10112 | ||
100d5702 | 10113 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
10114 | if (!work) |
10115 | return NULL; | |
10116 | ||
10117 | init_completion(&work->completion); | |
10118 | INIT_LIST_HEAD(&work->list); | |
10119 | work->inode = inode; | |
9e0af237 LB |
10120 | WARN_ON_ONCE(!inode); |
10121 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
10122 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
10123 | |
10124 | return work; | |
10125 | } | |
10126 | ||
d352ac68 CM |
10127 | /* |
10128 | * some fairly slow code that needs optimization. This walks the list | |
10129 | * of all the inodes with pending delalloc and forces them to disk. | |
10130 | */ | |
4fbb5147 | 10131 | static int start_delalloc_inodes(struct btrfs_root *root, int nr) |
ea8c2819 | 10132 | { |
ea8c2819 | 10133 | struct btrfs_inode *binode; |
5b21f2ed | 10134 | struct inode *inode; |
8ccf6f19 MX |
10135 | struct btrfs_delalloc_work *work, *next; |
10136 | struct list_head works; | |
1eafa6c7 | 10137 | struct list_head splice; |
8ccf6f19 | 10138 | int ret = 0; |
ea8c2819 | 10139 | |
8ccf6f19 | 10140 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 10141 | INIT_LIST_HEAD(&splice); |
63607cc8 | 10142 | |
573bfb72 | 10143 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
10144 | spin_lock(&root->delalloc_lock); |
10145 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
10146 | while (!list_empty(&splice)) { |
10147 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 10148 | delalloc_inodes); |
1eafa6c7 | 10149 | |
eb73c1b7 MX |
10150 | list_move_tail(&binode->delalloc_inodes, |
10151 | &root->delalloc_inodes); | |
5b21f2ed | 10152 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 10153 | if (!inode) { |
eb73c1b7 | 10154 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 10155 | continue; |
df0af1a5 | 10156 | } |
eb73c1b7 | 10157 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 10158 | |
076da91c | 10159 | work = btrfs_alloc_delalloc_work(inode); |
5d99a998 | 10160 | if (!work) { |
4fbb5147 | 10161 | iput(inode); |
1eafa6c7 | 10162 | ret = -ENOMEM; |
a1ecaabb | 10163 | goto out; |
5b21f2ed | 10164 | } |
1eafa6c7 | 10165 | list_add_tail(&work->list, &works); |
a44903ab QW |
10166 | btrfs_queue_work(root->fs_info->flush_workers, |
10167 | &work->work); | |
6c255e67 MX |
10168 | ret++; |
10169 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 10170 | goto out; |
5b21f2ed | 10171 | cond_resched(); |
eb73c1b7 | 10172 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 10173 | } |
eb73c1b7 | 10174 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 10175 | |
a1ecaabb | 10176 | out: |
eb73c1b7 MX |
10177 | list_for_each_entry_safe(work, next, &works, list) { |
10178 | list_del_init(&work->list); | |
40012f96 NB |
10179 | wait_for_completion(&work->completion); |
10180 | kfree(work); | |
eb73c1b7 MX |
10181 | } |
10182 | ||
81f1d390 | 10183 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
10184 | spin_lock(&root->delalloc_lock); |
10185 | list_splice_tail(&splice, &root->delalloc_inodes); | |
10186 | spin_unlock(&root->delalloc_lock); | |
10187 | } | |
573bfb72 | 10188 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
10189 | return ret; |
10190 | } | |
1eafa6c7 | 10191 | |
76f32e24 | 10192 | int btrfs_start_delalloc_inodes(struct btrfs_root *root) |
eb73c1b7 | 10193 | { |
0b246afa | 10194 | struct btrfs_fs_info *fs_info = root->fs_info; |
eb73c1b7 | 10195 | int ret; |
1eafa6c7 | 10196 | |
0b246afa | 10197 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10198 | return -EROFS; |
10199 | ||
4fbb5147 | 10200 | ret = start_delalloc_inodes(root, -1); |
6c255e67 MX |
10201 | if (ret > 0) |
10202 | ret = 0; | |
eb73c1b7 MX |
10203 | return ret; |
10204 | } | |
10205 | ||
82b3e53b | 10206 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr) |
eb73c1b7 MX |
10207 | { |
10208 | struct btrfs_root *root; | |
10209 | struct list_head splice; | |
10210 | int ret; | |
10211 | ||
2c21b4d7 | 10212 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10213 | return -EROFS; |
10214 | ||
10215 | INIT_LIST_HEAD(&splice); | |
10216 | ||
573bfb72 | 10217 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
10218 | spin_lock(&fs_info->delalloc_root_lock); |
10219 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 10220 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
10221 | root = list_first_entry(&splice, struct btrfs_root, |
10222 | delalloc_root); | |
10223 | root = btrfs_grab_fs_root(root); | |
10224 | BUG_ON(!root); | |
10225 | list_move_tail(&root->delalloc_root, | |
10226 | &fs_info->delalloc_roots); | |
10227 | spin_unlock(&fs_info->delalloc_root_lock); | |
10228 | ||
4fbb5147 | 10229 | ret = start_delalloc_inodes(root, nr); |
eb73c1b7 | 10230 | btrfs_put_fs_root(root); |
6c255e67 | 10231 | if (ret < 0) |
eb73c1b7 MX |
10232 | goto out; |
10233 | ||
6c255e67 MX |
10234 | if (nr != -1) { |
10235 | nr -= ret; | |
10236 | WARN_ON(nr < 0); | |
10237 | } | |
eb73c1b7 | 10238 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 10239 | } |
eb73c1b7 | 10240 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 10241 | |
6c255e67 | 10242 | ret = 0; |
eb73c1b7 | 10243 | out: |
81f1d390 | 10244 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
10245 | spin_lock(&fs_info->delalloc_root_lock); |
10246 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
10247 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 10248 | } |
573bfb72 | 10249 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 10250 | return ret; |
ea8c2819 CM |
10251 | } |
10252 | ||
39279cc3 CM |
10253 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
10254 | const char *symname) | |
10255 | { | |
0b246afa | 10256 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
10257 | struct btrfs_trans_handle *trans; |
10258 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10259 | struct btrfs_path *path; | |
10260 | struct btrfs_key key; | |
1832a6d5 | 10261 | struct inode *inode = NULL; |
39279cc3 CM |
10262 | int err; |
10263 | int drop_inode = 0; | |
10264 | u64 objectid; | |
67871254 | 10265 | u64 index = 0; |
39279cc3 CM |
10266 | int name_len; |
10267 | int datasize; | |
5f39d397 | 10268 | unsigned long ptr; |
39279cc3 | 10269 | struct btrfs_file_extent_item *ei; |
5f39d397 | 10270 | struct extent_buffer *leaf; |
39279cc3 | 10271 | |
f06becc4 | 10272 | name_len = strlen(symname); |
0b246afa | 10273 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 10274 | return -ENAMETOOLONG; |
1832a6d5 | 10275 | |
9ed74f2d JB |
10276 | /* |
10277 | * 2 items for inode item and ref | |
10278 | * 2 items for dir items | |
9269d12b FM |
10279 | * 1 item for updating parent inode item |
10280 | * 1 item for the inline extent item | |
9ed74f2d JB |
10281 | * 1 item for xattr if selinux is on |
10282 | */ | |
9269d12b | 10283 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10284 | if (IS_ERR(trans)) |
10285 | return PTR_ERR(trans); | |
1832a6d5 | 10286 | |
581bb050 LZ |
10287 | err = btrfs_find_free_ino(root, &objectid); |
10288 | if (err) | |
10289 | goto out_unlock; | |
10290 | ||
aec7477b | 10291 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
10292 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
10293 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
10294 | if (IS_ERR(inode)) { |
10295 | err = PTR_ERR(inode); | |
39279cc3 | 10296 | goto out_unlock; |
7cf96da3 | 10297 | } |
39279cc3 | 10298 | |
ad19db71 CS |
10299 | /* |
10300 | * If the active LSM wants to access the inode during | |
10301 | * d_instantiate it needs these. Smack checks to see | |
10302 | * if the filesystem supports xattrs by looking at the | |
10303 | * ops vector. | |
10304 | */ | |
10305 | inode->i_fop = &btrfs_file_operations; | |
10306 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10307 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10308 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10309 | ||
10310 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10311 | if (err) | |
10312 | goto out_unlock_inode; | |
ad19db71 | 10313 | |
39279cc3 | 10314 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10315 | if (!path) { |
10316 | err = -ENOMEM; | |
b0d5d10f | 10317 | goto out_unlock_inode; |
d8926bb3 | 10318 | } |
4a0cc7ca | 10319 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 10320 | key.offset = 0; |
962a298f | 10321 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10322 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10323 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10324 | datasize); | |
54aa1f4d | 10325 | if (err) { |
b0839166 | 10326 | btrfs_free_path(path); |
b0d5d10f | 10327 | goto out_unlock_inode; |
54aa1f4d | 10328 | } |
5f39d397 CM |
10329 | leaf = path->nodes[0]; |
10330 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10331 | struct btrfs_file_extent_item); | |
10332 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10333 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10334 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10335 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10336 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10337 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10338 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10339 | ||
39279cc3 | 10340 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10341 | write_extent_buffer(leaf, symname, ptr, name_len); |
10342 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10343 | btrfs_free_path(path); |
5f39d397 | 10344 | |
39279cc3 | 10345 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10346 | inode_nohighmem(inode); |
39279cc3 | 10347 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
d899e052 | 10348 | inode_set_bytes(inode, name_len); |
6ef06d27 | 10349 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
54aa1f4d | 10350 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
10351 | /* |
10352 | * Last step, add directory indexes for our symlink inode. This is the | |
10353 | * last step to avoid extra cleanup of these indexes if an error happens | |
10354 | * elsewhere above. | |
10355 | */ | |
10356 | if (!err) | |
cef415af NB |
10357 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
10358 | BTRFS_I(inode), 0, index); | |
b0d5d10f | 10359 | if (err) { |
54aa1f4d | 10360 | drop_inode = 1; |
b0d5d10f CM |
10361 | goto out_unlock_inode; |
10362 | } | |
10363 | ||
1e2e547a | 10364 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
10365 | |
10366 | out_unlock: | |
3a45bb20 | 10367 | btrfs_end_transaction(trans); |
39279cc3 CM |
10368 | if (drop_inode) { |
10369 | inode_dec_link_count(inode); | |
10370 | iput(inode); | |
10371 | } | |
2ff7e61e | 10372 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 10373 | return err; |
b0d5d10f CM |
10374 | |
10375 | out_unlock_inode: | |
10376 | drop_inode = 1; | |
10377 | unlock_new_inode(inode); | |
10378 | goto out_unlock; | |
39279cc3 | 10379 | } |
16432985 | 10380 | |
0af3d00b JB |
10381 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10382 | u64 start, u64 num_bytes, u64 min_size, | |
10383 | loff_t actual_len, u64 *alloc_hint, | |
10384 | struct btrfs_trans_handle *trans) | |
d899e052 | 10385 | { |
0b246afa | 10386 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
10387 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10388 | struct extent_map *em; | |
d899e052 YZ |
10389 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10390 | struct btrfs_key ins; | |
d899e052 | 10391 | u64 cur_offset = start; |
55a61d1d | 10392 | u64 i_size; |
154ea289 | 10393 | u64 cur_bytes; |
0b670dc4 | 10394 | u64 last_alloc = (u64)-1; |
d899e052 | 10395 | int ret = 0; |
0af3d00b | 10396 | bool own_trans = true; |
18513091 | 10397 | u64 end = start + num_bytes - 1; |
d899e052 | 10398 | |
0af3d00b JB |
10399 | if (trans) |
10400 | own_trans = false; | |
d899e052 | 10401 | while (num_bytes > 0) { |
0af3d00b JB |
10402 | if (own_trans) { |
10403 | trans = btrfs_start_transaction(root, 3); | |
10404 | if (IS_ERR(trans)) { | |
10405 | ret = PTR_ERR(trans); | |
10406 | break; | |
10407 | } | |
5a303d5d YZ |
10408 | } |
10409 | ||
ee22184b | 10410 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10411 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10412 | /* |
10413 | * If we are severely fragmented we could end up with really | |
10414 | * small allocations, so if the allocator is returning small | |
10415 | * chunks lets make its job easier by only searching for those | |
10416 | * sized chunks. | |
10417 | */ | |
10418 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
10419 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
10420 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
5a303d5d | 10421 | if (ret) { |
0af3d00b | 10422 | if (own_trans) |
3a45bb20 | 10423 | btrfs_end_transaction(trans); |
a22285a6 | 10424 | break; |
d899e052 | 10425 | } |
0b246afa | 10426 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5a303d5d | 10427 | |
0b670dc4 | 10428 | last_alloc = ins.offset; |
d899e052 YZ |
10429 | ret = insert_reserved_file_extent(trans, inode, |
10430 | cur_offset, ins.objectid, | |
10431 | ins.offset, ins.offset, | |
920bbbfb | 10432 | ins.offset, 0, 0, 0, |
d899e052 | 10433 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 10434 | if (ret) { |
2ff7e61e | 10435 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 10436 | ins.offset, 0); |
66642832 | 10437 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10438 | if (own_trans) |
3a45bb20 | 10439 | btrfs_end_transaction(trans); |
79787eaa JM |
10440 | break; |
10441 | } | |
31193213 | 10442 | |
dcdbc059 | 10443 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 10444 | cur_offset + ins.offset -1, 0); |
5a303d5d | 10445 | |
5dc562c5 JB |
10446 | em = alloc_extent_map(); |
10447 | if (!em) { | |
10448 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10449 | &BTRFS_I(inode)->runtime_flags); | |
10450 | goto next; | |
10451 | } | |
10452 | ||
10453 | em->start = cur_offset; | |
10454 | em->orig_start = cur_offset; | |
10455 | em->len = ins.offset; | |
10456 | em->block_start = ins.objectid; | |
10457 | em->block_len = ins.offset; | |
b4939680 | 10458 | em->orig_block_len = ins.offset; |
cc95bef6 | 10459 | em->ram_bytes = ins.offset; |
0b246afa | 10460 | em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 JB |
10461 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
10462 | em->generation = trans->transid; | |
10463 | ||
10464 | while (1) { | |
10465 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10466 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10467 | write_unlock(&em_tree->lock); |
10468 | if (ret != -EEXIST) | |
10469 | break; | |
dcdbc059 | 10470 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
10471 | cur_offset + ins.offset - 1, |
10472 | 0); | |
10473 | } | |
10474 | free_extent_map(em); | |
10475 | next: | |
d899e052 YZ |
10476 | num_bytes -= ins.offset; |
10477 | cur_offset += ins.offset; | |
efa56464 | 10478 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10479 | |
0c4d2d95 | 10480 | inode_inc_iversion(inode); |
c2050a45 | 10481 | inode->i_ctime = current_time(inode); |
6cbff00f | 10482 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10483 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10484 | (actual_len > inode->i_size) && |
10485 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10486 | if (cur_offset > actual_len) |
55a61d1d | 10487 | i_size = actual_len; |
d1ea6a61 | 10488 | else |
55a61d1d JB |
10489 | i_size = cur_offset; |
10490 | i_size_write(inode, i_size); | |
10491 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10492 | } |
10493 | ||
d899e052 | 10494 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10495 | |
10496 | if (ret) { | |
66642832 | 10497 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10498 | if (own_trans) |
3a45bb20 | 10499 | btrfs_end_transaction(trans); |
79787eaa JM |
10500 | break; |
10501 | } | |
d899e052 | 10502 | |
0af3d00b | 10503 | if (own_trans) |
3a45bb20 | 10504 | btrfs_end_transaction(trans); |
5a303d5d | 10505 | } |
18513091 | 10506 | if (cur_offset < end) |
bc42bda2 | 10507 | btrfs_free_reserved_data_space(inode, NULL, cur_offset, |
18513091 | 10508 | end - cur_offset + 1); |
d899e052 YZ |
10509 | return ret; |
10510 | } | |
10511 | ||
0af3d00b JB |
10512 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10513 | u64 start, u64 num_bytes, u64 min_size, | |
10514 | loff_t actual_len, u64 *alloc_hint) | |
10515 | { | |
10516 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10517 | min_size, actual_len, alloc_hint, | |
10518 | NULL); | |
10519 | } | |
10520 | ||
10521 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10522 | struct btrfs_trans_handle *trans, int mode, | |
10523 | u64 start, u64 num_bytes, u64 min_size, | |
10524 | loff_t actual_len, u64 *alloc_hint) | |
10525 | { | |
10526 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10527 | min_size, actual_len, alloc_hint, trans); | |
10528 | } | |
10529 | ||
e6dcd2dc CM |
10530 | static int btrfs_set_page_dirty(struct page *page) |
10531 | { | |
e6dcd2dc CM |
10532 | return __set_page_dirty_nobuffers(page); |
10533 | } | |
10534 | ||
10556cb2 | 10535 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10536 | { |
b83cc969 | 10537 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10538 | umode_t mode = inode->i_mode; |
b83cc969 | 10539 | |
cb6db4e5 JM |
10540 | if (mask & MAY_WRITE && |
10541 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10542 | if (btrfs_root_readonly(root)) | |
10543 | return -EROFS; | |
10544 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10545 | return -EACCES; | |
10546 | } | |
2830ba7f | 10547 | return generic_permission(inode, mask); |
fdebe2bd | 10548 | } |
39279cc3 | 10549 | |
ef3b9af5 FM |
10550 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10551 | { | |
2ff7e61e | 10552 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10553 | struct btrfs_trans_handle *trans; |
10554 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10555 | struct inode *inode = NULL; | |
10556 | u64 objectid; | |
10557 | u64 index; | |
10558 | int ret = 0; | |
10559 | ||
10560 | /* | |
10561 | * 5 units required for adding orphan entry | |
10562 | */ | |
10563 | trans = btrfs_start_transaction(root, 5); | |
10564 | if (IS_ERR(trans)) | |
10565 | return PTR_ERR(trans); | |
10566 | ||
10567 | ret = btrfs_find_free_ino(root, &objectid); | |
10568 | if (ret) | |
10569 | goto out; | |
10570 | ||
10571 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 10572 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
10573 | if (IS_ERR(inode)) { |
10574 | ret = PTR_ERR(inode); | |
10575 | inode = NULL; | |
10576 | goto out; | |
10577 | } | |
10578 | ||
ef3b9af5 FM |
10579 | inode->i_fop = &btrfs_file_operations; |
10580 | inode->i_op = &btrfs_file_inode_operations; | |
10581 | ||
10582 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10583 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10584 | ||
b0d5d10f CM |
10585 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10586 | if (ret) | |
10587 | goto out_inode; | |
10588 | ||
10589 | ret = btrfs_update_inode(trans, root, inode); | |
10590 | if (ret) | |
10591 | goto out_inode; | |
73f2e545 | 10592 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 10593 | if (ret) |
b0d5d10f | 10594 | goto out_inode; |
ef3b9af5 | 10595 | |
5762b5c9 FM |
10596 | /* |
10597 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10598 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10599 | * through: | |
10600 | * | |
10601 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10602 | */ | |
10603 | set_nlink(inode, 1); | |
b0d5d10f | 10604 | unlock_new_inode(inode); |
ef3b9af5 FM |
10605 | d_tmpfile(dentry, inode); |
10606 | mark_inode_dirty(inode); | |
10607 | ||
10608 | out: | |
3a45bb20 | 10609 | btrfs_end_transaction(trans); |
ef3b9af5 FM |
10610 | if (ret) |
10611 | iput(inode); | |
2ff7e61e | 10612 | btrfs_btree_balance_dirty(fs_info); |
ef3b9af5 | 10613 | return ret; |
b0d5d10f CM |
10614 | |
10615 | out_inode: | |
10616 | unlock_new_inode(inode); | |
10617 | goto out; | |
10618 | ||
ef3b9af5 FM |
10619 | } |
10620 | ||
20a7db8a | 10621 | __attribute__((const)) |
9d0d1c8b | 10622 | static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror) |
20a7db8a | 10623 | { |
9d0d1c8b | 10624 | return -EAGAIN; |
20a7db8a DS |
10625 | } |
10626 | ||
c6100a4b JB |
10627 | static struct btrfs_fs_info *iotree_fs_info(void *private_data) |
10628 | { | |
10629 | struct inode *inode = private_data; | |
10630 | return btrfs_sb(inode->i_sb); | |
10631 | } | |
10632 | ||
10633 | static void btrfs_check_extent_io_range(void *private_data, const char *caller, | |
10634 | u64 start, u64 end) | |
10635 | { | |
10636 | struct inode *inode = private_data; | |
10637 | u64 isize; | |
10638 | ||
10639 | isize = i_size_read(inode); | |
10640 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
10641 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
10642 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
10643 | caller, btrfs_ino(BTRFS_I(inode)), isize, start, end); | |
10644 | } | |
10645 | } | |
10646 | ||
10647 | void btrfs_set_range_writeback(void *private_data, u64 start, u64 end) | |
10648 | { | |
10649 | struct inode *inode = private_data; | |
10650 | unsigned long index = start >> PAGE_SHIFT; | |
10651 | unsigned long end_index = end >> PAGE_SHIFT; | |
10652 | struct page *page; | |
10653 | ||
10654 | while (index <= end_index) { | |
10655 | page = find_get_page(inode->i_mapping, index); | |
10656 | ASSERT(page); /* Pages should be in the extent_io_tree */ | |
10657 | set_page_writeback(page); | |
10658 | put_page(page); | |
10659 | index++; | |
10660 | } | |
10661 | } | |
10662 | ||
6e1d5dcc | 10663 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10664 | .getattr = btrfs_getattr, |
39279cc3 CM |
10665 | .lookup = btrfs_lookup, |
10666 | .create = btrfs_create, | |
10667 | .unlink = btrfs_unlink, | |
10668 | .link = btrfs_link, | |
10669 | .mkdir = btrfs_mkdir, | |
10670 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10671 | .rename = btrfs_rename2, |
39279cc3 CM |
10672 | .symlink = btrfs_symlink, |
10673 | .setattr = btrfs_setattr, | |
618e21d5 | 10674 | .mknod = btrfs_mknod, |
5103e947 | 10675 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10676 | .permission = btrfs_permission, |
4e34e719 | 10677 | .get_acl = btrfs_get_acl, |
996a710d | 10678 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10679 | .update_time = btrfs_update_time, |
ef3b9af5 | 10680 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10681 | }; |
6e1d5dcc | 10682 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10683 | .lookup = btrfs_lookup, |
fdebe2bd | 10684 | .permission = btrfs_permission, |
93fd63c2 | 10685 | .update_time = btrfs_update_time, |
39279cc3 | 10686 | }; |
76dda93c | 10687 | |
828c0950 | 10688 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10689 | .llseek = generic_file_llseek, |
10690 | .read = generic_read_dir, | |
02dbfc99 | 10691 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 10692 | .open = btrfs_opendir, |
34287aa3 | 10693 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10694 | #ifdef CONFIG_COMPAT |
4c63c245 | 10695 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10696 | #endif |
6bf13c0c | 10697 | .release = btrfs_release_file, |
e02119d5 | 10698 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10699 | }; |
10700 | ||
20e5506b | 10701 | static const struct extent_io_ops btrfs_extent_io_ops = { |
4d53dddb | 10702 | /* mandatory callbacks */ |
065631f6 | 10703 | .submit_bio_hook = btrfs_submit_bio_hook, |
07157aac | 10704 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
4d53dddb | 10705 | .merge_bio_hook = btrfs_merge_bio_hook, |
9d0d1c8b | 10706 | .readpage_io_failed_hook = btrfs_readpage_io_failed_hook, |
c6100a4b JB |
10707 | .tree_fs_info = iotree_fs_info, |
10708 | .set_range_writeback = btrfs_set_range_writeback, | |
4d53dddb DS |
10709 | |
10710 | /* optional callbacks */ | |
10711 | .fill_delalloc = run_delalloc_range, | |
e6dcd2dc | 10712 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 10713 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
10714 | .set_bit_hook = btrfs_set_bit_hook, |
10715 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
10716 | .merge_extent_hook = btrfs_merge_extent_hook, |
10717 | .split_extent_hook = btrfs_split_extent_hook, | |
c6100a4b | 10718 | .check_extent_io_range = btrfs_check_extent_io_range, |
07157aac CM |
10719 | }; |
10720 | ||
35054394 CM |
10721 | /* |
10722 | * btrfs doesn't support the bmap operation because swapfiles | |
10723 | * use bmap to make a mapping of extents in the file. They assume | |
10724 | * these extents won't change over the life of the file and they | |
10725 | * use the bmap result to do IO directly to the drive. | |
10726 | * | |
10727 | * the btrfs bmap call would return logical addresses that aren't | |
10728 | * suitable for IO and they also will change frequently as COW | |
10729 | * operations happen. So, swapfile + btrfs == corruption. | |
10730 | * | |
10731 | * For now we're avoiding this by dropping bmap. | |
10732 | */ | |
7f09410b | 10733 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10734 | .readpage = btrfs_readpage, |
10735 | .writepage = btrfs_writepage, | |
b293f02e | 10736 | .writepages = btrfs_writepages, |
3ab2fb5a | 10737 | .readpages = btrfs_readpages, |
16432985 | 10738 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10739 | .invalidatepage = btrfs_invalidatepage, |
10740 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 10741 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10742 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
10743 | }; |
10744 | ||
7f09410b | 10745 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
10746 | .readpage = btrfs_readpage, |
10747 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
10748 | .invalidatepage = btrfs_invalidatepage, |
10749 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
10750 | }; |
10751 | ||
6e1d5dcc | 10752 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10753 | .getattr = btrfs_getattr, |
10754 | .setattr = btrfs_setattr, | |
5103e947 | 10755 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10756 | .permission = btrfs_permission, |
1506fcc8 | 10757 | .fiemap = btrfs_fiemap, |
4e34e719 | 10758 | .get_acl = btrfs_get_acl, |
996a710d | 10759 | .set_acl = btrfs_set_acl, |
e41f941a | 10760 | .update_time = btrfs_update_time, |
39279cc3 | 10761 | }; |
6e1d5dcc | 10762 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10763 | .getattr = btrfs_getattr, |
10764 | .setattr = btrfs_setattr, | |
fdebe2bd | 10765 | .permission = btrfs_permission, |
33268eaf | 10766 | .listxattr = btrfs_listxattr, |
4e34e719 | 10767 | .get_acl = btrfs_get_acl, |
996a710d | 10768 | .set_acl = btrfs_set_acl, |
e41f941a | 10769 | .update_time = btrfs_update_time, |
618e21d5 | 10770 | }; |
6e1d5dcc | 10771 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10772 | .get_link = page_get_link, |
f209561a | 10773 | .getattr = btrfs_getattr, |
22c44fe6 | 10774 | .setattr = btrfs_setattr, |
fdebe2bd | 10775 | .permission = btrfs_permission, |
0279b4cd | 10776 | .listxattr = btrfs_listxattr, |
e41f941a | 10777 | .update_time = btrfs_update_time, |
39279cc3 | 10778 | }; |
76dda93c | 10779 | |
82d339d9 | 10780 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
10781 | .d_delete = btrfs_dentry_delete, |
10782 | }; |