Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
e02119d5 CM |
2 | /* |
3 | * Copyright (C) 2008 Oracle. All rights reserved. | |
e02119d5 CM |
4 | */ |
5 | ||
6 | #include <linux/sched.h> | |
5a0e3ad6 | 7 | #include <linux/slab.h> |
c6adc9cc | 8 | #include <linux/blkdev.h> |
5dc562c5 | 9 | #include <linux/list_sort.h> |
c7f88c4e | 10 | #include <linux/iversion.h> |
602cbe91 | 11 | #include "misc.h" |
9678c543 | 12 | #include "ctree.h" |
995946dd | 13 | #include "tree-log.h" |
e02119d5 CM |
14 | #include "disk-io.h" |
15 | #include "locking.h" | |
16 | #include "print-tree.h" | |
f186373f | 17 | #include "backref.h" |
ebb8765b | 18 | #include "compression.h" |
df2c95f3 | 19 | #include "qgroup.h" |
6787bb9f NB |
20 | #include "block-group.h" |
21 | #include "space-info.h" | |
d3575156 | 22 | #include "zoned.h" |
26c2c454 | 23 | #include "inode-item.h" |
c7f13d42 | 24 | #include "fs.h" |
ad1ac501 | 25 | #include "accessors.h" |
e02119d5 | 26 | |
e09d94c9 FM |
27 | #define MAX_CONFLICT_INODES 10 |
28 | ||
e02119d5 CM |
29 | /* magic values for the inode_only field in btrfs_log_inode: |
30 | * | |
31 | * LOG_INODE_ALL means to log everything | |
32 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
33 | * during log replay | |
34 | */ | |
e13976cf DS |
35 | enum { |
36 | LOG_INODE_ALL, | |
37 | LOG_INODE_EXISTS, | |
e13976cf | 38 | }; |
e02119d5 | 39 | |
12fcfd22 CM |
40 | /* |
41 | * directory trouble cases | |
42 | * | |
43 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
44 | * log, we must force a full commit before doing an fsync of the directory | |
45 | * where the unlink was done. | |
46 | * ---> record transid of last unlink/rename per directory | |
47 | * | |
48 | * mkdir foo/some_dir | |
49 | * normal commit | |
50 | * rename foo/some_dir foo2/some_dir | |
51 | * mkdir foo/some_dir | |
52 | * fsync foo/some_dir/some_file | |
53 | * | |
54 | * The fsync above will unlink the original some_dir without recording | |
55 | * it in its new location (foo2). After a crash, some_dir will be gone | |
56 | * unless the fsync of some_file forces a full commit | |
57 | * | |
58 | * 2) we must log any new names for any file or dir that is in the fsync | |
59 | * log. ---> check inode while renaming/linking. | |
60 | * | |
61 | * 2a) we must log any new names for any file or dir during rename | |
62 | * when the directory they are being removed from was logged. | |
63 | * ---> check inode and old parent dir during rename | |
64 | * | |
65 | * 2a is actually the more important variant. With the extra logging | |
66 | * a crash might unlink the old name without recreating the new one | |
67 | * | |
68 | * 3) after a crash, we must go through any directories with a link count | |
69 | * of zero and redo the rm -rf | |
70 | * | |
71 | * mkdir f1/foo | |
72 | * normal commit | |
73 | * rm -rf f1/foo | |
74 | * fsync(f1) | |
75 | * | |
76 | * The directory f1 was fully removed from the FS, but fsync was never | |
77 | * called on f1, only its parent dir. After a crash the rm -rf must | |
78 | * be replayed. This must be able to recurse down the entire | |
79 | * directory tree. The inode link count fixup code takes care of the | |
80 | * ugly details. | |
81 | */ | |
82 | ||
e02119d5 CM |
83 | /* |
84 | * stages for the tree walking. The first | |
85 | * stage (0) is to only pin down the blocks we find | |
86 | * the second stage (1) is to make sure that all the inodes | |
87 | * we find in the log are created in the subvolume. | |
88 | * | |
89 | * The last stage is to deal with directories and links and extents | |
90 | * and all the other fun semantics | |
91 | */ | |
e13976cf DS |
92 | enum { |
93 | LOG_WALK_PIN_ONLY, | |
94 | LOG_WALK_REPLAY_INODES, | |
95 | LOG_WALK_REPLAY_DIR_INDEX, | |
96 | LOG_WALK_REPLAY_ALL, | |
97 | }; | |
e02119d5 | 98 | |
12fcfd22 | 99 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
90d04510 | 100 | struct btrfs_inode *inode, |
49dae1bc | 101 | int inode_only, |
8407f553 | 102 | struct btrfs_log_ctx *ctx); |
ec051c0f YZ |
103 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
104 | struct btrfs_root *root, | |
105 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
106 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
107 | struct btrfs_root *root, | |
108 | struct btrfs_root *log, | |
109 | struct btrfs_path *path, | |
110 | u64 dirid, int del_all); | |
fa1a0f42 | 111 | static void wait_log_commit(struct btrfs_root *root, int transid); |
e02119d5 CM |
112 | |
113 | /* | |
114 | * tree logging is a special write ahead log used to make sure that | |
115 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
116 | * | |
117 | * Full tree commits are expensive because they require commonly | |
118 | * modified blocks to be recowed, creating many dirty pages in the | |
119 | * extent tree an 4x-6x higher write load than ext3. | |
120 | * | |
121 | * Instead of doing a tree commit on every fsync, we use the | |
122 | * key ranges and transaction ids to find items for a given file or directory | |
123 | * that have changed in this transaction. Those items are copied into | |
124 | * a special tree (one per subvolume root), that tree is written to disk | |
125 | * and then the fsync is considered complete. | |
126 | * | |
127 | * After a crash, items are copied out of the log-tree back into the | |
128 | * subvolume tree. Any file data extents found are recorded in the extent | |
129 | * allocation tree, and the log-tree freed. | |
130 | * | |
131 | * The log tree is read three times, once to pin down all the extents it is | |
132 | * using in ram and once, once to create all the inodes logged in the tree | |
133 | * and once to do all the other items. | |
134 | */ | |
135 | ||
e02119d5 CM |
136 | /* |
137 | * start a sub transaction and setup the log tree | |
138 | * this increments the log tree writer count to make the people | |
139 | * syncing the tree wait for us to finish | |
140 | */ | |
141 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
142 | struct btrfs_root *root, |
143 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 144 | { |
0b246afa | 145 | struct btrfs_fs_info *fs_info = root->fs_info; |
47876f7c | 146 | struct btrfs_root *tree_root = fs_info->tree_root; |
fa1a0f42 | 147 | const bool zoned = btrfs_is_zoned(fs_info); |
34eb2a52 | 148 | int ret = 0; |
fa1a0f42 | 149 | bool created = false; |
7237f183 | 150 | |
47876f7c FM |
151 | /* |
152 | * First check if the log root tree was already created. If not, create | |
153 | * it before locking the root's log_mutex, just to keep lockdep happy. | |
154 | */ | |
155 | if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state)) { | |
156 | mutex_lock(&tree_root->log_mutex); | |
157 | if (!fs_info->log_root_tree) { | |
158 | ret = btrfs_init_log_root_tree(trans, fs_info); | |
fa1a0f42 | 159 | if (!ret) { |
47876f7c | 160 | set_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state); |
fa1a0f42 NA |
161 | created = true; |
162 | } | |
47876f7c FM |
163 | } |
164 | mutex_unlock(&tree_root->log_mutex); | |
165 | if (ret) | |
166 | return ret; | |
167 | } | |
168 | ||
7237f183 | 169 | mutex_lock(&root->log_mutex); |
34eb2a52 | 170 | |
fa1a0f42 | 171 | again: |
7237f183 | 172 | if (root->log_root) { |
fa1a0f42 NA |
173 | int index = (root->log_transid + 1) % 2; |
174 | ||
4884b8e8 | 175 | if (btrfs_need_log_full_commit(trans)) { |
f31f09f6 | 176 | ret = BTRFS_LOG_FORCE_COMMIT; |
50471a38 MX |
177 | goto out; |
178 | } | |
34eb2a52 | 179 | |
fa1a0f42 NA |
180 | if (zoned && atomic_read(&root->log_commit[index])) { |
181 | wait_log_commit(root, root->log_transid - 1); | |
182 | goto again; | |
183 | } | |
184 | ||
ff782e0a | 185 | if (!root->log_start_pid) { |
27cdeb70 | 186 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 187 | root->log_start_pid = current->pid; |
ff782e0a | 188 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 189 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 190 | } |
34eb2a52 | 191 | } else { |
fa1a0f42 NA |
192 | /* |
193 | * This means fs_info->log_root_tree was already created | |
194 | * for some other FS trees. Do the full commit not to mix | |
195 | * nodes from multiple log transactions to do sequential | |
196 | * writing. | |
197 | */ | |
198 | if (zoned && !created) { | |
f31f09f6 | 199 | ret = BTRFS_LOG_FORCE_COMMIT; |
fa1a0f42 NA |
200 | goto out; |
201 | } | |
202 | ||
e02119d5 | 203 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 204 | if (ret) |
e87ac136 | 205 | goto out; |
34eb2a52 | 206 | |
e7a79811 | 207 | set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state); |
34eb2a52 Z |
208 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
209 | root->log_start_pid = current->pid; | |
e02119d5 | 210 | } |
34eb2a52 | 211 | |
7237f183 | 212 | atomic_inc(&root->log_writers); |
289cffcb | 213 | if (!ctx->logging_new_name) { |
34eb2a52 | 214 | int index = root->log_transid % 2; |
8b050d35 | 215 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 216 | ctx->log_transid = root->log_transid; |
8b050d35 | 217 | } |
34eb2a52 | 218 | |
e87ac136 | 219 | out: |
7237f183 | 220 | mutex_unlock(&root->log_mutex); |
e87ac136 | 221 | return ret; |
e02119d5 CM |
222 | } |
223 | ||
224 | /* | |
225 | * returns 0 if there was a log transaction running and we were able | |
226 | * to join, or returns -ENOENT if there were not transactions | |
227 | * in progress | |
228 | */ | |
229 | static int join_running_log_trans(struct btrfs_root *root) | |
230 | { | |
fa1a0f42 | 231 | const bool zoned = btrfs_is_zoned(root->fs_info); |
e02119d5 CM |
232 | int ret = -ENOENT; |
233 | ||
e7a79811 FM |
234 | if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state)) |
235 | return ret; | |
236 | ||
7237f183 | 237 | mutex_lock(&root->log_mutex); |
fa1a0f42 | 238 | again: |
e02119d5 | 239 | if (root->log_root) { |
fa1a0f42 NA |
240 | int index = (root->log_transid + 1) % 2; |
241 | ||
e02119d5 | 242 | ret = 0; |
fa1a0f42 NA |
243 | if (zoned && atomic_read(&root->log_commit[index])) { |
244 | wait_log_commit(root, root->log_transid - 1); | |
245 | goto again; | |
246 | } | |
7237f183 | 247 | atomic_inc(&root->log_writers); |
e02119d5 | 248 | } |
7237f183 | 249 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
250 | return ret; |
251 | } | |
252 | ||
12fcfd22 CM |
253 | /* |
254 | * This either makes the current running log transaction wait | |
255 | * until you call btrfs_end_log_trans() or it makes any future | |
256 | * log transactions wait until you call btrfs_end_log_trans() | |
257 | */ | |
45128b08 | 258 | void btrfs_pin_log_trans(struct btrfs_root *root) |
12fcfd22 | 259 | { |
12fcfd22 | 260 | atomic_inc(&root->log_writers); |
12fcfd22 CM |
261 | } |
262 | ||
e02119d5 CM |
263 | /* |
264 | * indicate we're done making changes to the log tree | |
265 | * and wake up anyone waiting to do a sync | |
266 | */ | |
143bede5 | 267 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 268 | { |
7237f183 | 269 | if (atomic_dec_and_test(&root->log_writers)) { |
093258e6 DS |
270 | /* atomic_dec_and_test implies a barrier */ |
271 | cond_wake_up_nomb(&root->log_writer_wait); | |
7237f183 | 272 | } |
e02119d5 CM |
273 | } |
274 | ||
247462a5 DS |
275 | static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf) |
276 | { | |
277 | filemap_fdatawait_range(buf->pages[0]->mapping, | |
278 | buf->start, buf->start + buf->len - 1); | |
279 | } | |
e02119d5 CM |
280 | |
281 | /* | |
282 | * the walk control struct is used to pass state down the chain when | |
283 | * processing the log tree. The stage field tells us which part | |
284 | * of the log tree processing we are currently doing. The others | |
285 | * are state fields used for that specific part | |
286 | */ | |
287 | struct walk_control { | |
288 | /* should we free the extent on disk when done? This is used | |
289 | * at transaction commit time while freeing a log tree | |
290 | */ | |
291 | int free; | |
292 | ||
e02119d5 CM |
293 | /* pin only walk, we record which extents on disk belong to the |
294 | * log trees | |
295 | */ | |
296 | int pin; | |
297 | ||
298 | /* what stage of the replay code we're currently in */ | |
299 | int stage; | |
300 | ||
f2d72f42 FM |
301 | /* |
302 | * Ignore any items from the inode currently being processed. Needs | |
303 | * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in | |
304 | * the LOG_WALK_REPLAY_INODES stage. | |
305 | */ | |
306 | bool ignore_cur_inode; | |
307 | ||
e02119d5 CM |
308 | /* the root we are currently replaying */ |
309 | struct btrfs_root *replay_dest; | |
310 | ||
311 | /* the trans handle for the current replay */ | |
312 | struct btrfs_trans_handle *trans; | |
313 | ||
314 | /* the function that gets used to process blocks we find in the | |
315 | * tree. Note the extent_buffer might not be up to date when it is | |
316 | * passed in, and it must be checked or read if you need the data | |
317 | * inside it | |
318 | */ | |
319 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 320 | struct walk_control *wc, u64 gen, int level); |
e02119d5 CM |
321 | }; |
322 | ||
323 | /* | |
324 | * process_func used to pin down extents, write them or wait on them | |
325 | */ | |
326 | static int process_one_buffer(struct btrfs_root *log, | |
327 | struct extent_buffer *eb, | |
581c1760 | 328 | struct walk_control *wc, u64 gen, int level) |
e02119d5 | 329 | { |
0b246afa | 330 | struct btrfs_fs_info *fs_info = log->fs_info; |
b50c6e25 JB |
331 | int ret = 0; |
332 | ||
8c2a1a30 JB |
333 | /* |
334 | * If this fs is mixed then we need to be able to process the leaves to | |
335 | * pin down any logged extents, so we have to read the block. | |
336 | */ | |
0b246afa | 337 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
6a2e9dc4 | 338 | ret = btrfs_read_extent_buffer(eb, gen, level, NULL); |
8c2a1a30 JB |
339 | if (ret) |
340 | return ret; | |
341 | } | |
342 | ||
c816d705 | 343 | if (wc->pin) { |
9fce5704 | 344 | ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start, |
2ff7e61e | 345 | eb->len); |
c816d705 FM |
346 | if (ret) |
347 | return ret; | |
e02119d5 | 348 | |
c816d705 FM |
349 | if (btrfs_buffer_uptodate(eb, gen, 0) && |
350 | btrfs_header_level(eb) == 0) | |
bcdc428c | 351 | ret = btrfs_exclude_logged_extents(eb); |
e02119d5 | 352 | } |
b50c6e25 | 353 | return ret; |
e02119d5 CM |
354 | } |
355 | ||
086dcbfa FM |
356 | static int do_overwrite_item(struct btrfs_trans_handle *trans, |
357 | struct btrfs_root *root, | |
358 | struct btrfs_path *path, | |
359 | struct extent_buffer *eb, int slot, | |
360 | struct btrfs_key *key) | |
e02119d5 CM |
361 | { |
362 | int ret; | |
363 | u32 item_size; | |
364 | u64 saved_i_size = 0; | |
365 | int save_old_i_size = 0; | |
366 | unsigned long src_ptr; | |
367 | unsigned long dst_ptr; | |
368 | int overwrite_root = 0; | |
4bc4bee4 | 369 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
370 | |
371 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
372 | overwrite_root = 1; | |
373 | ||
3212fa14 | 374 | item_size = btrfs_item_size(eb, slot); |
e02119d5 CM |
375 | src_ptr = btrfs_item_ptr_offset(eb, slot); |
376 | ||
086dcbfa FM |
377 | /* Our caller must have done a search for the key for us. */ |
378 | ASSERT(path->nodes[0] != NULL); | |
379 | ||
380 | /* | |
381 | * And the slot must point to the exact key or the slot where the key | |
382 | * should be at (the first item with a key greater than 'key') | |
383 | */ | |
384 | if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) { | |
385 | struct btrfs_key found_key; | |
386 | ||
387 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
388 | ret = btrfs_comp_cpu_keys(&found_key, key); | |
389 | ASSERT(ret >= 0); | |
390 | } else { | |
391 | ret = 1; | |
392 | } | |
4bc4bee4 | 393 | |
e02119d5 CM |
394 | if (ret == 0) { |
395 | char *src_copy; | |
396 | char *dst_copy; | |
3212fa14 | 397 | u32 dst_size = btrfs_item_size(path->nodes[0], |
e02119d5 CM |
398 | path->slots[0]); |
399 | if (dst_size != item_size) | |
400 | goto insert; | |
401 | ||
402 | if (item_size == 0) { | |
b3b4aa74 | 403 | btrfs_release_path(path); |
e02119d5 CM |
404 | return 0; |
405 | } | |
406 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
407 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 408 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 409 | btrfs_release_path(path); |
2a29edc6 | 410 | kfree(dst_copy); |
411 | kfree(src_copy); | |
412 | return -ENOMEM; | |
413 | } | |
e02119d5 CM |
414 | |
415 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
416 | ||
417 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
418 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
419 | item_size); | |
420 | ret = memcmp(dst_copy, src_copy, item_size); | |
421 | ||
422 | kfree(dst_copy); | |
423 | kfree(src_copy); | |
424 | /* | |
425 | * they have the same contents, just return, this saves | |
426 | * us from cowing blocks in the destination tree and doing | |
427 | * extra writes that may not have been done by a previous | |
428 | * sync | |
429 | */ | |
430 | if (ret == 0) { | |
b3b4aa74 | 431 | btrfs_release_path(path); |
e02119d5 CM |
432 | return 0; |
433 | } | |
434 | ||
4bc4bee4 JB |
435 | /* |
436 | * We need to load the old nbytes into the inode so when we | |
437 | * replay the extents we've logged we get the right nbytes. | |
438 | */ | |
439 | if (inode_item) { | |
440 | struct btrfs_inode_item *item; | |
441 | u64 nbytes; | |
d555438b | 442 | u32 mode; |
4bc4bee4 JB |
443 | |
444 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
445 | struct btrfs_inode_item); | |
446 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
447 | item = btrfs_item_ptr(eb, slot, | |
448 | struct btrfs_inode_item); | |
449 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
450 | |
451 | /* | |
452 | * If this is a directory we need to reset the i_size to | |
453 | * 0 so that we can set it up properly when replaying | |
454 | * the rest of the items in this log. | |
455 | */ | |
456 | mode = btrfs_inode_mode(eb, item); | |
457 | if (S_ISDIR(mode)) | |
458 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
459 | } |
460 | } else if (inode_item) { | |
461 | struct btrfs_inode_item *item; | |
d555438b | 462 | u32 mode; |
4bc4bee4 JB |
463 | |
464 | /* | |
465 | * New inode, set nbytes to 0 so that the nbytes comes out | |
466 | * properly when we replay the extents. | |
467 | */ | |
468 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
469 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
470 | |
471 | /* | |
472 | * If this is a directory we need to reset the i_size to 0 so | |
473 | * that we can set it up properly when replaying the rest of | |
474 | * the items in this log. | |
475 | */ | |
476 | mode = btrfs_inode_mode(eb, item); | |
477 | if (S_ISDIR(mode)) | |
478 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
479 | } |
480 | insert: | |
b3b4aa74 | 481 | btrfs_release_path(path); |
e02119d5 | 482 | /* try to insert the key into the destination tree */ |
df8d116f | 483 | path->skip_release_on_error = 1; |
e02119d5 CM |
484 | ret = btrfs_insert_empty_item(trans, root, path, |
485 | key, item_size); | |
df8d116f | 486 | path->skip_release_on_error = 0; |
e02119d5 CM |
487 | |
488 | /* make sure any existing item is the correct size */ | |
df8d116f | 489 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 | 490 | u32 found_size; |
3212fa14 | 491 | found_size = btrfs_item_size(path->nodes[0], |
e02119d5 | 492 | path->slots[0]); |
143bede5 | 493 | if (found_size > item_size) |
78ac4f9e | 494 | btrfs_truncate_item(path, item_size, 1); |
143bede5 | 495 | else if (found_size < item_size) |
c71dd880 | 496 | btrfs_extend_item(path, item_size - found_size); |
e02119d5 | 497 | } else if (ret) { |
4a500fd1 | 498 | return ret; |
e02119d5 CM |
499 | } |
500 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
501 | path->slots[0]); | |
502 | ||
503 | /* don't overwrite an existing inode if the generation number | |
504 | * was logged as zero. This is done when the tree logging code | |
505 | * is just logging an inode to make sure it exists after recovery. | |
506 | * | |
507 | * Also, don't overwrite i_size on directories during replay. | |
508 | * log replay inserts and removes directory items based on the | |
509 | * state of the tree found in the subvolume, and i_size is modified | |
510 | * as it goes | |
511 | */ | |
512 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
513 | struct btrfs_inode_item *src_item; | |
514 | struct btrfs_inode_item *dst_item; | |
515 | ||
516 | src_item = (struct btrfs_inode_item *)src_ptr; | |
517 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
518 | ||
1a4bcf47 FM |
519 | if (btrfs_inode_generation(eb, src_item) == 0) { |
520 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 521 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 522 | |
2f2ff0ee FM |
523 | /* |
524 | * For regular files an ino_size == 0 is used only when | |
525 | * logging that an inode exists, as part of a directory | |
526 | * fsync, and the inode wasn't fsynced before. In this | |
527 | * case don't set the size of the inode in the fs/subvol | |
528 | * tree, otherwise we would be throwing valid data away. | |
529 | */ | |
1a4bcf47 | 530 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee | 531 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
60d48e2e DS |
532 | ino_size != 0) |
533 | btrfs_set_inode_size(dst_eb, dst_item, ino_size); | |
e02119d5 | 534 | goto no_copy; |
1a4bcf47 | 535 | } |
e02119d5 CM |
536 | |
537 | if (overwrite_root && | |
538 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
539 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
540 | save_old_i_size = 1; | |
541 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
542 | dst_item); | |
543 | } | |
544 | } | |
545 | ||
546 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
547 | src_ptr, item_size); | |
548 | ||
549 | if (save_old_i_size) { | |
550 | struct btrfs_inode_item *dst_item; | |
551 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
552 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
553 | } | |
554 | ||
555 | /* make sure the generation is filled in */ | |
556 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
557 | struct btrfs_inode_item *dst_item; | |
558 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
559 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
560 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
561 | trans->transid); | |
562 | } | |
563 | } | |
564 | no_copy: | |
565 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 566 | btrfs_release_path(path); |
e02119d5 CM |
567 | return 0; |
568 | } | |
569 | ||
086dcbfa FM |
570 | /* |
571 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
572 | * to the src data we are copying out. | |
573 | * | |
574 | * root is the tree we are copying into, and path is a scratch | |
575 | * path for use in this function (it should be released on entry and | |
576 | * will be released on exit). | |
577 | * | |
578 | * If the key is already in the destination tree the existing item is | |
579 | * overwritten. If the existing item isn't big enough, it is extended. | |
580 | * If it is too large, it is truncated. | |
581 | * | |
582 | * If the key isn't in the destination yet, a new item is inserted. | |
583 | */ | |
584 | static int overwrite_item(struct btrfs_trans_handle *trans, | |
585 | struct btrfs_root *root, | |
586 | struct btrfs_path *path, | |
587 | struct extent_buffer *eb, int slot, | |
588 | struct btrfs_key *key) | |
589 | { | |
590 | int ret; | |
591 | ||
592 | /* Look for the key in the destination tree. */ | |
593 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
594 | if (ret < 0) | |
595 | return ret; | |
596 | ||
597 | return do_overwrite_item(trans, root, path, eb, slot, key); | |
598 | } | |
599 | ||
e43eec81 | 600 | static int read_alloc_one_name(struct extent_buffer *eb, void *start, int len, |
6db75318 | 601 | struct fscrypt_str *name) |
e43eec81 STD |
602 | { |
603 | char *buf; | |
604 | ||
605 | buf = kmalloc(len, GFP_NOFS); | |
606 | if (!buf) | |
607 | return -ENOMEM; | |
608 | ||
609 | read_extent_buffer(eb, buf, (unsigned long)start, len); | |
610 | name->name = buf; | |
611 | name->len = len; | |
612 | return 0; | |
613 | } | |
614 | ||
e02119d5 CM |
615 | /* |
616 | * simple helper to read an inode off the disk from a given root | |
617 | * This can only be called for subvolume roots and not for the log | |
618 | */ | |
619 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
620 | u64 objectid) | |
621 | { | |
622 | struct inode *inode; | |
e02119d5 | 623 | |
0202e83f | 624 | inode = btrfs_iget(root->fs_info->sb, objectid, root); |
2e19f1f9 | 625 | if (IS_ERR(inode)) |
5d4f98a2 | 626 | inode = NULL; |
e02119d5 CM |
627 | return inode; |
628 | } | |
629 | ||
630 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
631 | * subvolume 'root'. path is released on entry and should be released | |
632 | * on exit. | |
633 | * | |
634 | * extents in the log tree have not been allocated out of the extent | |
635 | * tree yet. So, this completes the allocation, taking a reference | |
636 | * as required if the extent already exists or creating a new extent | |
637 | * if it isn't in the extent allocation tree yet. | |
638 | * | |
639 | * The extent is inserted into the file, dropping any existing extents | |
640 | * from the file that overlap the new one. | |
641 | */ | |
642 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
643 | struct btrfs_root *root, | |
644 | struct btrfs_path *path, | |
645 | struct extent_buffer *eb, int slot, | |
646 | struct btrfs_key *key) | |
647 | { | |
5893dfb9 | 648 | struct btrfs_drop_extents_args drop_args = { 0 }; |
0b246afa | 649 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 650 | int found_type; |
e02119d5 | 651 | u64 extent_end; |
e02119d5 | 652 | u64 start = key->offset; |
4bc4bee4 | 653 | u64 nbytes = 0; |
e02119d5 CM |
654 | struct btrfs_file_extent_item *item; |
655 | struct inode *inode = NULL; | |
656 | unsigned long size; | |
657 | int ret = 0; | |
658 | ||
659 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
660 | found_type = btrfs_file_extent_type(eb, item); | |
661 | ||
d899e052 | 662 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
663 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
664 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
665 | extent_end = start + nbytes; | |
666 | ||
667 | /* | |
668 | * We don't add to the inodes nbytes if we are prealloc or a | |
669 | * hole. | |
670 | */ | |
671 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
672 | nbytes = 0; | |
673 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 674 | size = btrfs_file_extent_ram_bytes(eb, item); |
4bc4bee4 | 675 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
da17066c | 676 | extent_end = ALIGN(start + size, |
0b246afa | 677 | fs_info->sectorsize); |
e02119d5 CM |
678 | } else { |
679 | ret = 0; | |
680 | goto out; | |
681 | } | |
682 | ||
683 | inode = read_one_inode(root, key->objectid); | |
684 | if (!inode) { | |
685 | ret = -EIO; | |
686 | goto out; | |
687 | } | |
688 | ||
689 | /* | |
690 | * first check to see if we already have this extent in the | |
691 | * file. This must be done before the btrfs_drop_extents run | |
692 | * so we don't try to drop this extent. | |
693 | */ | |
f85b7379 DS |
694 | ret = btrfs_lookup_file_extent(trans, root, path, |
695 | btrfs_ino(BTRFS_I(inode)), start, 0); | |
e02119d5 | 696 | |
d899e052 YZ |
697 | if (ret == 0 && |
698 | (found_type == BTRFS_FILE_EXTENT_REG || | |
699 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
700 | struct btrfs_file_extent_item cmp1; |
701 | struct btrfs_file_extent_item cmp2; | |
702 | struct btrfs_file_extent_item *existing; | |
703 | struct extent_buffer *leaf; | |
704 | ||
705 | leaf = path->nodes[0]; | |
706 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
707 | struct btrfs_file_extent_item); | |
708 | ||
709 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
710 | sizeof(cmp1)); | |
711 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
712 | sizeof(cmp2)); | |
713 | ||
714 | /* | |
715 | * we already have a pointer to this exact extent, | |
716 | * we don't have to do anything | |
717 | */ | |
718 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 719 | btrfs_release_path(path); |
e02119d5 CM |
720 | goto out; |
721 | } | |
722 | } | |
b3b4aa74 | 723 | btrfs_release_path(path); |
e02119d5 CM |
724 | |
725 | /* drop any overlapping extents */ | |
5893dfb9 FM |
726 | drop_args.start = start; |
727 | drop_args.end = extent_end; | |
728 | drop_args.drop_cache = true; | |
729 | ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args); | |
3650860b JB |
730 | if (ret) |
731 | goto out; | |
e02119d5 | 732 | |
07d400a6 YZ |
733 | if (found_type == BTRFS_FILE_EXTENT_REG || |
734 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 735 | u64 offset; |
07d400a6 YZ |
736 | unsigned long dest_offset; |
737 | struct btrfs_key ins; | |
738 | ||
3168021c FM |
739 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0 && |
740 | btrfs_fs_incompat(fs_info, NO_HOLES)) | |
741 | goto update_inode; | |
742 | ||
07d400a6 YZ |
743 | ret = btrfs_insert_empty_item(trans, root, path, key, |
744 | sizeof(*item)); | |
3650860b JB |
745 | if (ret) |
746 | goto out; | |
07d400a6 YZ |
747 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
748 | path->slots[0]); | |
749 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
750 | (unsigned long)item, sizeof(*item)); | |
751 | ||
752 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
753 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
754 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 755 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 756 | |
df2c95f3 QW |
757 | /* |
758 | * Manually record dirty extent, as here we did a shallow | |
759 | * file extent item copy and skip normal backref update, | |
760 | * but modifying extent tree all by ourselves. | |
761 | * So need to manually record dirty extent for qgroup, | |
762 | * as the owner of the file extent changed from log tree | |
763 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
764 | */ | |
a95f3aaf | 765 | ret = btrfs_qgroup_trace_extent(trans, |
df2c95f3 | 766 | btrfs_file_extent_disk_bytenr(eb, item), |
e2896e79 | 767 | btrfs_file_extent_disk_num_bytes(eb, item)); |
df2c95f3 QW |
768 | if (ret < 0) |
769 | goto out; | |
770 | ||
07d400a6 | 771 | if (ins.objectid > 0) { |
82fa113f | 772 | struct btrfs_ref ref = { 0 }; |
07d400a6 YZ |
773 | u64 csum_start; |
774 | u64 csum_end; | |
775 | LIST_HEAD(ordered_sums); | |
82fa113f | 776 | |
07d400a6 YZ |
777 | /* |
778 | * is this extent already allocated in the extent | |
779 | * allocation tree? If so, just add a reference | |
780 | */ | |
2ff7e61e | 781 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 | 782 | ins.offset); |
3736127a MPS |
783 | if (ret < 0) { |
784 | goto out; | |
785 | } else if (ret == 0) { | |
82fa113f QW |
786 | btrfs_init_generic_ref(&ref, |
787 | BTRFS_ADD_DELAYED_REF, | |
788 | ins.objectid, ins.offset, 0); | |
789 | btrfs_init_data_ref(&ref, | |
790 | root->root_key.objectid, | |
f42c5da6 | 791 | key->objectid, offset, 0, false); |
82fa113f | 792 | ret = btrfs_inc_extent_ref(trans, &ref); |
b50c6e25 JB |
793 | if (ret) |
794 | goto out; | |
07d400a6 YZ |
795 | } else { |
796 | /* | |
797 | * insert the extent pointer in the extent | |
798 | * allocation tree | |
799 | */ | |
5d4f98a2 | 800 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e | 801 | root->root_key.objectid, |
5d4f98a2 | 802 | key->objectid, offset, &ins); |
b50c6e25 JB |
803 | if (ret) |
804 | goto out; | |
07d400a6 | 805 | } |
b3b4aa74 | 806 | btrfs_release_path(path); |
07d400a6 YZ |
807 | |
808 | if (btrfs_file_extent_compression(eb, item)) { | |
809 | csum_start = ins.objectid; | |
810 | csum_end = csum_start + ins.offset; | |
811 | } else { | |
812 | csum_start = ins.objectid + | |
813 | btrfs_file_extent_offset(eb, item); | |
814 | csum_end = csum_start + | |
815 | btrfs_file_extent_num_bytes(eb, item); | |
816 | } | |
817 | ||
818 | ret = btrfs_lookup_csums_range(root->log_root, | |
819 | csum_start, csum_end - 1, | |
26ce9114 | 820 | &ordered_sums, 0, false); |
3650860b JB |
821 | if (ret) |
822 | goto out; | |
b84b8390 FM |
823 | /* |
824 | * Now delete all existing cums in the csum root that | |
825 | * cover our range. We do this because we can have an | |
826 | * extent that is completely referenced by one file | |
827 | * extent item and partially referenced by another | |
828 | * file extent item (like after using the clone or | |
829 | * extent_same ioctls). In this case if we end up doing | |
830 | * the replay of the one that partially references the | |
831 | * extent first, and we do not do the csum deletion | |
832 | * below, we can get 2 csum items in the csum tree that | |
833 | * overlap each other. For example, imagine our log has | |
834 | * the two following file extent items: | |
835 | * | |
836 | * key (257 EXTENT_DATA 409600) | |
837 | * extent data disk byte 12845056 nr 102400 | |
838 | * extent data offset 20480 nr 20480 ram 102400 | |
839 | * | |
840 | * key (257 EXTENT_DATA 819200) | |
841 | * extent data disk byte 12845056 nr 102400 | |
842 | * extent data offset 0 nr 102400 ram 102400 | |
843 | * | |
844 | * Where the second one fully references the 100K extent | |
845 | * that starts at disk byte 12845056, and the log tree | |
846 | * has a single csum item that covers the entire range | |
847 | * of the extent: | |
848 | * | |
849 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
850 | * | |
851 | * After the first file extent item is replayed, the | |
852 | * csum tree gets the following csum item: | |
853 | * | |
854 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
855 | * | |
856 | * Which covers the 20K sub-range starting at offset 20K | |
857 | * of our extent. Now when we replay the second file | |
858 | * extent item, if we do not delete existing csum items | |
859 | * that cover any of its blocks, we end up getting two | |
860 | * csum items in our csum tree that overlap each other: | |
861 | * | |
862 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
863 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
864 | * | |
865 | * Which is a problem, because after this anyone trying | |
866 | * to lookup up for the checksum of any block of our | |
867 | * extent starting at an offset of 40K or higher, will | |
868 | * end up looking at the second csum item only, which | |
869 | * does not contain the checksum for any block starting | |
870 | * at offset 40K or higher of our extent. | |
871 | */ | |
07d400a6 YZ |
872 | while (!list_empty(&ordered_sums)) { |
873 | struct btrfs_ordered_sum *sums; | |
fc28b25e JB |
874 | struct btrfs_root *csum_root; |
875 | ||
07d400a6 YZ |
876 | sums = list_entry(ordered_sums.next, |
877 | struct btrfs_ordered_sum, | |
878 | list); | |
fc28b25e JB |
879 | csum_root = btrfs_csum_root(fs_info, |
880 | sums->bytenr); | |
b84b8390 | 881 | if (!ret) |
fc28b25e | 882 | ret = btrfs_del_csums(trans, csum_root, |
5b4aacef JM |
883 | sums->bytenr, |
884 | sums->len); | |
3650860b JB |
885 | if (!ret) |
886 | ret = btrfs_csum_file_blocks(trans, | |
fc28b25e JB |
887 | csum_root, |
888 | sums); | |
07d400a6 YZ |
889 | list_del(&sums->list); |
890 | kfree(sums); | |
891 | } | |
3650860b JB |
892 | if (ret) |
893 | goto out; | |
07d400a6 | 894 | } else { |
b3b4aa74 | 895 | btrfs_release_path(path); |
07d400a6 YZ |
896 | } |
897 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
898 | /* inline extents are easy, we just overwrite them */ | |
899 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
900 | if (ret) |
901 | goto out; | |
07d400a6 | 902 | } |
e02119d5 | 903 | |
9ddc959e JB |
904 | ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, |
905 | extent_end - start); | |
906 | if (ret) | |
907 | goto out; | |
908 | ||
3168021c | 909 | update_inode: |
2766ff61 | 910 | btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found); |
9a56fcd1 | 911 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e02119d5 | 912 | out: |
8aa1e49e | 913 | iput(inode); |
e02119d5 CM |
914 | return ret; |
915 | } | |
916 | ||
313ab753 FM |
917 | static int unlink_inode_for_log_replay(struct btrfs_trans_handle *trans, |
918 | struct btrfs_inode *dir, | |
919 | struct btrfs_inode *inode, | |
6db75318 | 920 | const struct fscrypt_str *name) |
313ab753 FM |
921 | { |
922 | int ret; | |
923 | ||
e43eec81 | 924 | ret = btrfs_unlink_inode(trans, dir, inode, name); |
313ab753 FM |
925 | if (ret) |
926 | return ret; | |
927 | /* | |
928 | * Whenever we need to check if a name exists or not, we check the | |
929 | * fs/subvolume tree. So after an unlink we must run delayed items, so | |
930 | * that future checks for a name during log replay see that the name | |
931 | * does not exists anymore. | |
932 | */ | |
933 | return btrfs_run_delayed_items(trans); | |
934 | } | |
935 | ||
e02119d5 CM |
936 | /* |
937 | * when cleaning up conflicts between the directory names in the | |
938 | * subvolume, directory names in the log and directory names in the | |
939 | * inode back references, we may have to unlink inodes from directories. | |
940 | * | |
941 | * This is a helper function to do the unlink of a specific directory | |
942 | * item | |
943 | */ | |
944 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
e02119d5 | 945 | struct btrfs_path *path, |
207e7d92 | 946 | struct btrfs_inode *dir, |
e02119d5 CM |
947 | struct btrfs_dir_item *di) |
948 | { | |
9798ba24 | 949 | struct btrfs_root *root = dir->root; |
e02119d5 | 950 | struct inode *inode; |
6db75318 | 951 | struct fscrypt_str name; |
e02119d5 CM |
952 | struct extent_buffer *leaf; |
953 | struct btrfs_key location; | |
954 | int ret; | |
955 | ||
956 | leaf = path->nodes[0]; | |
957 | ||
958 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
e43eec81 STD |
959 | ret = read_alloc_one_name(leaf, di + 1, btrfs_dir_name_len(leaf, di), &name); |
960 | if (ret) | |
2a29edc6 | 961 | return -ENOMEM; |
962 | ||
b3b4aa74 | 963 | btrfs_release_path(path); |
e02119d5 CM |
964 | |
965 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 966 | if (!inode) { |
3650860b JB |
967 | ret = -EIO; |
968 | goto out; | |
c00e9493 | 969 | } |
e02119d5 | 970 | |
ec051c0f | 971 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
972 | if (ret) |
973 | goto out; | |
12fcfd22 | 974 | |
e43eec81 | 975 | ret = unlink_inode_for_log_replay(trans, dir, BTRFS_I(inode), &name); |
3650860b | 976 | out: |
e43eec81 | 977 | kfree(name.name); |
e02119d5 CM |
978 | iput(inode); |
979 | return ret; | |
980 | } | |
981 | ||
982 | /* | |
77a5b9e3 FM |
983 | * See if a given name and sequence number found in an inode back reference are |
984 | * already in a directory and correctly point to this inode. | |
985 | * | |
986 | * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it | |
987 | * exists. | |
e02119d5 CM |
988 | */ |
989 | static noinline int inode_in_dir(struct btrfs_root *root, | |
990 | struct btrfs_path *path, | |
991 | u64 dirid, u64 objectid, u64 index, | |
6db75318 | 992 | struct fscrypt_str *name) |
e02119d5 CM |
993 | { |
994 | struct btrfs_dir_item *di; | |
995 | struct btrfs_key location; | |
77a5b9e3 | 996 | int ret = 0; |
e02119d5 CM |
997 | |
998 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
e43eec81 | 999 | index, name, 0); |
77a5b9e3 | 1000 | if (IS_ERR(di)) { |
8dcbc261 | 1001 | ret = PTR_ERR(di); |
77a5b9e3 FM |
1002 | goto out; |
1003 | } else if (di) { | |
e02119d5 CM |
1004 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); |
1005 | if (location.objectid != objectid) | |
1006 | goto out; | |
77a5b9e3 | 1007 | } else { |
e02119d5 | 1008 | goto out; |
77a5b9e3 | 1009 | } |
e02119d5 | 1010 | |
77a5b9e3 | 1011 | btrfs_release_path(path); |
e43eec81 | 1012 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, 0); |
77a5b9e3 FM |
1013 | if (IS_ERR(di)) { |
1014 | ret = PTR_ERR(di); | |
e02119d5 | 1015 | goto out; |
77a5b9e3 FM |
1016 | } else if (di) { |
1017 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
1018 | if (location.objectid == objectid) | |
1019 | ret = 1; | |
1020 | } | |
e02119d5 | 1021 | out: |
b3b4aa74 | 1022 | btrfs_release_path(path); |
77a5b9e3 | 1023 | return ret; |
e02119d5 CM |
1024 | } |
1025 | ||
1026 | /* | |
1027 | * helper function to check a log tree for a named back reference in | |
1028 | * an inode. This is used to decide if a back reference that is | |
1029 | * found in the subvolume conflicts with what we find in the log. | |
1030 | * | |
1031 | * inode backreferences may have multiple refs in a single item, | |
1032 | * during replay we process one reference at a time, and we don't | |
1033 | * want to delete valid links to a file from the subvolume if that | |
1034 | * link is also in the log. | |
1035 | */ | |
1036 | static noinline int backref_in_log(struct btrfs_root *log, | |
1037 | struct btrfs_key *key, | |
f186373f | 1038 | u64 ref_objectid, |
6db75318 | 1039 | const struct fscrypt_str *name) |
e02119d5 CM |
1040 | { |
1041 | struct btrfs_path *path; | |
e02119d5 | 1042 | int ret; |
e02119d5 CM |
1043 | |
1044 | path = btrfs_alloc_path(); | |
2a29edc6 | 1045 | if (!path) |
1046 | return -ENOMEM; | |
1047 | ||
e02119d5 | 1048 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
d3316c82 NB |
1049 | if (ret < 0) { |
1050 | goto out; | |
1051 | } else if (ret == 1) { | |
89cbf5f6 | 1052 | ret = 0; |
f186373f MF |
1053 | goto out; |
1054 | } | |
1055 | ||
89cbf5f6 NB |
1056 | if (key->type == BTRFS_INODE_EXTREF_KEY) |
1057 | ret = !!btrfs_find_name_in_ext_backref(path->nodes[0], | |
1058 | path->slots[0], | |
e43eec81 | 1059 | ref_objectid, name); |
89cbf5f6 NB |
1060 | else |
1061 | ret = !!btrfs_find_name_in_backref(path->nodes[0], | |
e43eec81 | 1062 | path->slots[0], name); |
e02119d5 CM |
1063 | out: |
1064 | btrfs_free_path(path); | |
89cbf5f6 | 1065 | return ret; |
e02119d5 CM |
1066 | } |
1067 | ||
5a1d7843 | 1068 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 1069 | struct btrfs_root *root, |
e02119d5 | 1070 | struct btrfs_path *path, |
5a1d7843 | 1071 | struct btrfs_root *log_root, |
94c91a1f NB |
1072 | struct btrfs_inode *dir, |
1073 | struct btrfs_inode *inode, | |
f186373f | 1074 | u64 inode_objectid, u64 parent_objectid, |
6db75318 | 1075 | u64 ref_index, struct fscrypt_str *name) |
e02119d5 | 1076 | { |
34f3e4f2 | 1077 | int ret; |
f186373f | 1078 | struct extent_buffer *leaf; |
5a1d7843 | 1079 | struct btrfs_dir_item *di; |
f186373f MF |
1080 | struct btrfs_key search_key; |
1081 | struct btrfs_inode_extref *extref; | |
c622ae60 | 1082 | |
f186373f MF |
1083 | again: |
1084 | /* Search old style refs */ | |
1085 | search_key.objectid = inode_objectid; | |
1086 | search_key.type = BTRFS_INODE_REF_KEY; | |
1087 | search_key.offset = parent_objectid; | |
1088 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 1089 | if (ret == 0) { |
e02119d5 CM |
1090 | struct btrfs_inode_ref *victim_ref; |
1091 | unsigned long ptr; | |
1092 | unsigned long ptr_end; | |
f186373f MF |
1093 | |
1094 | leaf = path->nodes[0]; | |
e02119d5 CM |
1095 | |
1096 | /* are we trying to overwrite a back ref for the root directory | |
1097 | * if so, just jump out, we're done | |
1098 | */ | |
f186373f | 1099 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1100 | return 1; |
e02119d5 CM |
1101 | |
1102 | /* check all the names in this back reference to see | |
1103 | * if they are in the log. if so, we allow them to stay | |
1104 | * otherwise they must be unlinked as a conflict | |
1105 | */ | |
1106 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3212fa14 | 1107 | ptr_end = ptr + btrfs_item_size(leaf, path->slots[0]); |
d397712b | 1108 | while (ptr < ptr_end) { |
6db75318 | 1109 | struct fscrypt_str victim_name; |
e02119d5 | 1110 | |
e43eec81 STD |
1111 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1112 | ret = read_alloc_one_name(leaf, (victim_ref + 1), | |
1113 | btrfs_inode_ref_name_len(leaf, victim_ref), | |
1114 | &victim_name); | |
1115 | if (ret) | |
1116 | return ret; | |
e02119d5 | 1117 | |
d3316c82 | 1118 | ret = backref_in_log(log_root, &search_key, |
e43eec81 | 1119 | parent_objectid, &victim_name); |
d3316c82 | 1120 | if (ret < 0) { |
e43eec81 | 1121 | kfree(victim_name.name); |
d3316c82 NB |
1122 | return ret; |
1123 | } else if (!ret) { | |
94c91a1f | 1124 | inc_nlink(&inode->vfs_inode); |
b3b4aa74 | 1125 | btrfs_release_path(path); |
12fcfd22 | 1126 | |
313ab753 | 1127 | ret = unlink_inode_for_log_replay(trans, dir, inode, |
e43eec81 STD |
1128 | &victim_name); |
1129 | kfree(victim_name.name); | |
ada9af21 FDBM |
1130 | if (ret) |
1131 | return ret; | |
f186373f | 1132 | goto again; |
e02119d5 | 1133 | } |
e43eec81 | 1134 | kfree(victim_name.name); |
f186373f | 1135 | |
e43eec81 | 1136 | ptr = (unsigned long)(victim_ref + 1) + victim_name.len; |
e02119d5 | 1137 | } |
e02119d5 | 1138 | } |
b3b4aa74 | 1139 | btrfs_release_path(path); |
e02119d5 | 1140 | |
f186373f | 1141 | /* Same search but for extended refs */ |
e43eec81 | 1142 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, |
f186373f MF |
1143 | inode_objectid, parent_objectid, 0, |
1144 | 0); | |
7a6b75b7 FM |
1145 | if (IS_ERR(extref)) { |
1146 | return PTR_ERR(extref); | |
1147 | } else if (extref) { | |
f186373f MF |
1148 | u32 item_size; |
1149 | u32 cur_offset = 0; | |
1150 | unsigned long base; | |
1151 | struct inode *victim_parent; | |
1152 | ||
1153 | leaf = path->nodes[0]; | |
1154 | ||
3212fa14 | 1155 | item_size = btrfs_item_size(leaf, path->slots[0]); |
f186373f MF |
1156 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); |
1157 | ||
1158 | while (cur_offset < item_size) { | |
6db75318 | 1159 | struct fscrypt_str victim_name; |
f186373f | 1160 | |
e43eec81 | 1161 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1162 | |
1163 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1164 | goto next; | |
1165 | ||
e43eec81 STD |
1166 | ret = read_alloc_one_name(leaf, &extref->name, |
1167 | btrfs_inode_extref_name_len(leaf, extref), | |
1168 | &victim_name); | |
1169 | if (ret) | |
1170 | return ret; | |
f186373f MF |
1171 | |
1172 | search_key.objectid = inode_objectid; | |
1173 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1174 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
e43eec81 STD |
1175 | victim_name.name, |
1176 | victim_name.len); | |
d3316c82 | 1177 | ret = backref_in_log(log_root, &search_key, |
e43eec81 | 1178 | parent_objectid, &victim_name); |
d3316c82 | 1179 | if (ret < 0) { |
e43eec81 | 1180 | kfree(victim_name.name); |
d3316c82 NB |
1181 | return ret; |
1182 | } else if (!ret) { | |
f186373f MF |
1183 | ret = -ENOENT; |
1184 | victim_parent = read_one_inode(root, | |
94c91a1f | 1185 | parent_objectid); |
f186373f | 1186 | if (victim_parent) { |
94c91a1f | 1187 | inc_nlink(&inode->vfs_inode); |
f186373f MF |
1188 | btrfs_release_path(path); |
1189 | ||
313ab753 | 1190 | ret = unlink_inode_for_log_replay(trans, |
4ec5934e | 1191 | BTRFS_I(victim_parent), |
e43eec81 | 1192 | inode, &victim_name); |
f186373f | 1193 | } |
f186373f | 1194 | iput(victim_parent); |
e43eec81 | 1195 | kfree(victim_name.name); |
3650860b JB |
1196 | if (ret) |
1197 | return ret; | |
f186373f MF |
1198 | goto again; |
1199 | } | |
e43eec81 | 1200 | kfree(victim_name.name); |
f186373f | 1201 | next: |
e43eec81 | 1202 | cur_offset += victim_name.len + sizeof(*extref); |
f186373f | 1203 | } |
f186373f MF |
1204 | } |
1205 | btrfs_release_path(path); | |
1206 | ||
34f3e4f2 | 1207 | /* look for a conflicting sequence number */ |
94c91a1f | 1208 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
e43eec81 | 1209 | ref_index, name, 0); |
52db7779 | 1210 | if (IS_ERR(di)) { |
8dcbc261 | 1211 | return PTR_ERR(di); |
52db7779 | 1212 | } else if (di) { |
9798ba24 | 1213 | ret = drop_one_dir_item(trans, path, dir, di); |
3650860b JB |
1214 | if (ret) |
1215 | return ret; | |
34f3e4f2 | 1216 | } |
1217 | btrfs_release_path(path); | |
1218 | ||
52042d8e | 1219 | /* look for a conflicting name */ |
e43eec81 | 1220 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), name, 0); |
52db7779 FM |
1221 | if (IS_ERR(di)) { |
1222 | return PTR_ERR(di); | |
1223 | } else if (di) { | |
9798ba24 | 1224 | ret = drop_one_dir_item(trans, path, dir, di); |
3650860b JB |
1225 | if (ret) |
1226 | return ret; | |
34f3e4f2 | 1227 | } |
1228 | btrfs_release_path(path); | |
1229 | ||
5a1d7843 JS |
1230 | return 0; |
1231 | } | |
e02119d5 | 1232 | |
bae15d95 | 1233 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
6db75318 | 1234 | struct fscrypt_str *name, u64 *index, |
bae15d95 | 1235 | u64 *parent_objectid) |
f186373f MF |
1236 | { |
1237 | struct btrfs_inode_extref *extref; | |
e43eec81 | 1238 | int ret; |
f186373f MF |
1239 | |
1240 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1241 | ||
e43eec81 STD |
1242 | ret = read_alloc_one_name(eb, &extref->name, |
1243 | btrfs_inode_extref_name_len(eb, extref), name); | |
1244 | if (ret) | |
1245 | return ret; | |
f186373f | 1246 | |
1f250e92 FM |
1247 | if (index) |
1248 | *index = btrfs_inode_extref_index(eb, extref); | |
f186373f MF |
1249 | if (parent_objectid) |
1250 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1251 | ||
1252 | return 0; | |
1253 | } | |
1254 | ||
bae15d95 | 1255 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
6db75318 | 1256 | struct fscrypt_str *name, u64 *index) |
f186373f MF |
1257 | { |
1258 | struct btrfs_inode_ref *ref; | |
e43eec81 | 1259 | int ret; |
f186373f MF |
1260 | |
1261 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1262 | ||
e43eec81 STD |
1263 | ret = read_alloc_one_name(eb, ref + 1, btrfs_inode_ref_name_len(eb, ref), |
1264 | name); | |
1265 | if (ret) | |
1266 | return ret; | |
f186373f | 1267 | |
1f250e92 FM |
1268 | if (index) |
1269 | *index = btrfs_inode_ref_index(eb, ref); | |
f186373f MF |
1270 | |
1271 | return 0; | |
1272 | } | |
1273 | ||
1f250e92 FM |
1274 | /* |
1275 | * Take an inode reference item from the log tree and iterate all names from the | |
1276 | * inode reference item in the subvolume tree with the same key (if it exists). | |
1277 | * For any name that is not in the inode reference item from the log tree, do a | |
1278 | * proper unlink of that name (that is, remove its entry from the inode | |
1279 | * reference item and both dir index keys). | |
1280 | */ | |
1281 | static int unlink_old_inode_refs(struct btrfs_trans_handle *trans, | |
1282 | struct btrfs_root *root, | |
1283 | struct btrfs_path *path, | |
1284 | struct btrfs_inode *inode, | |
1285 | struct extent_buffer *log_eb, | |
1286 | int log_slot, | |
1287 | struct btrfs_key *key) | |
1288 | { | |
1289 | int ret; | |
1290 | unsigned long ref_ptr; | |
1291 | unsigned long ref_end; | |
1292 | struct extent_buffer *eb; | |
1293 | ||
1294 | again: | |
1295 | btrfs_release_path(path); | |
1296 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1297 | if (ret > 0) { | |
1298 | ret = 0; | |
1299 | goto out; | |
1300 | } | |
1301 | if (ret < 0) | |
1302 | goto out; | |
1303 | ||
1304 | eb = path->nodes[0]; | |
1305 | ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]); | |
3212fa14 | 1306 | ref_end = ref_ptr + btrfs_item_size(eb, path->slots[0]); |
1f250e92 | 1307 | while (ref_ptr < ref_end) { |
6db75318 | 1308 | struct fscrypt_str name; |
1f250e92 FM |
1309 | u64 parent_id; |
1310 | ||
1311 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
e43eec81 | 1312 | ret = extref_get_fields(eb, ref_ptr, &name, |
1f250e92 FM |
1313 | NULL, &parent_id); |
1314 | } else { | |
1315 | parent_id = key->offset; | |
e43eec81 | 1316 | ret = ref_get_fields(eb, ref_ptr, &name, NULL); |
1f250e92 FM |
1317 | } |
1318 | if (ret) | |
1319 | goto out; | |
1320 | ||
1321 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
6ff49c6a | 1322 | ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot, |
e43eec81 | 1323 | parent_id, &name); |
1f250e92 | 1324 | else |
e43eec81 | 1325 | ret = !!btrfs_find_name_in_backref(log_eb, log_slot, &name); |
1f250e92 FM |
1326 | |
1327 | if (!ret) { | |
1328 | struct inode *dir; | |
1329 | ||
1330 | btrfs_release_path(path); | |
1331 | dir = read_one_inode(root, parent_id); | |
1332 | if (!dir) { | |
1333 | ret = -ENOENT; | |
e43eec81 | 1334 | kfree(name.name); |
1f250e92 FM |
1335 | goto out; |
1336 | } | |
313ab753 | 1337 | ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir), |
e43eec81 STD |
1338 | inode, &name); |
1339 | kfree(name.name); | |
1f250e92 FM |
1340 | iput(dir); |
1341 | if (ret) | |
1342 | goto out; | |
1343 | goto again; | |
1344 | } | |
1345 | ||
e43eec81 STD |
1346 | kfree(name.name); |
1347 | ref_ptr += name.len; | |
1f250e92 FM |
1348 | if (key->type == BTRFS_INODE_EXTREF_KEY) |
1349 | ref_ptr += sizeof(struct btrfs_inode_extref); | |
1350 | else | |
1351 | ref_ptr += sizeof(struct btrfs_inode_ref); | |
1352 | } | |
1353 | ret = 0; | |
1354 | out: | |
1355 | btrfs_release_path(path); | |
1356 | return ret; | |
1357 | } | |
1358 | ||
5a1d7843 JS |
1359 | /* |
1360 | * replay one inode back reference item found in the log tree. | |
1361 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1362 | * root is the destination we are replaying into, and path is for temp | |
1363 | * use by this function. (it should be released on return). | |
1364 | */ | |
1365 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1366 | struct btrfs_root *root, | |
1367 | struct btrfs_root *log, | |
1368 | struct btrfs_path *path, | |
1369 | struct extent_buffer *eb, int slot, | |
1370 | struct btrfs_key *key) | |
1371 | { | |
03b2f08b GB |
1372 | struct inode *dir = NULL; |
1373 | struct inode *inode = NULL; | |
5a1d7843 JS |
1374 | unsigned long ref_ptr; |
1375 | unsigned long ref_end; | |
6db75318 | 1376 | struct fscrypt_str name; |
5a1d7843 | 1377 | int ret; |
f186373f MF |
1378 | int log_ref_ver = 0; |
1379 | u64 parent_objectid; | |
1380 | u64 inode_objectid; | |
f46dbe3d | 1381 | u64 ref_index = 0; |
f186373f MF |
1382 | int ref_struct_size; |
1383 | ||
1384 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
3212fa14 | 1385 | ref_end = ref_ptr + btrfs_item_size(eb, slot); |
f186373f MF |
1386 | |
1387 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1388 | struct btrfs_inode_extref *r; | |
1389 | ||
1390 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1391 | log_ref_ver = 1; | |
1392 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1393 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1394 | } else { | |
1395 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1396 | parent_objectid = key->offset; | |
1397 | } | |
1398 | inode_objectid = key->objectid; | |
e02119d5 | 1399 | |
5a1d7843 JS |
1400 | /* |
1401 | * it is possible that we didn't log all the parent directories | |
1402 | * for a given inode. If we don't find the dir, just don't | |
1403 | * copy the back ref in. The link count fixup code will take | |
1404 | * care of the rest | |
1405 | */ | |
f186373f | 1406 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1407 | if (!dir) { |
1408 | ret = -ENOENT; | |
1409 | goto out; | |
1410 | } | |
5a1d7843 | 1411 | |
f186373f | 1412 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1413 | if (!inode) { |
03b2f08b GB |
1414 | ret = -EIO; |
1415 | goto out; | |
5a1d7843 JS |
1416 | } |
1417 | ||
5a1d7843 | 1418 | while (ref_ptr < ref_end) { |
f186373f | 1419 | if (log_ref_ver) { |
e43eec81 | 1420 | ret = extref_get_fields(eb, ref_ptr, &name, |
bae15d95 | 1421 | &ref_index, &parent_objectid); |
f186373f MF |
1422 | /* |
1423 | * parent object can change from one array | |
1424 | * item to another. | |
1425 | */ | |
1426 | if (!dir) | |
1427 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1428 | if (!dir) { |
1429 | ret = -ENOENT; | |
1430 | goto out; | |
1431 | } | |
f186373f | 1432 | } else { |
e43eec81 | 1433 | ret = ref_get_fields(eb, ref_ptr, &name, &ref_index); |
f186373f MF |
1434 | } |
1435 | if (ret) | |
03b2f08b | 1436 | goto out; |
5a1d7843 | 1437 | |
77a5b9e3 | 1438 | ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), |
e43eec81 | 1439 | btrfs_ino(BTRFS_I(inode)), ref_index, &name); |
77a5b9e3 FM |
1440 | if (ret < 0) { |
1441 | goto out; | |
1442 | } else if (ret == 0) { | |
5a1d7843 JS |
1443 | /* |
1444 | * look for a conflicting back reference in the | |
1445 | * metadata. if we find one we have to unlink that name | |
1446 | * of the file before we add our new link. Later on, we | |
1447 | * overwrite any existing back reference, and we don't | |
1448 | * want to create dangling pointers in the directory. | |
1449 | */ | |
7059c658 FM |
1450 | ret = __add_inode_ref(trans, root, path, log, |
1451 | BTRFS_I(dir), BTRFS_I(inode), | |
1452 | inode_objectid, parent_objectid, | |
e43eec81 | 1453 | ref_index, &name); |
7059c658 FM |
1454 | if (ret) { |
1455 | if (ret == 1) | |
1456 | ret = 0; | |
0d836392 | 1457 | goto out; |
7059c658 | 1458 | } |
0d836392 | 1459 | |
5a1d7843 | 1460 | /* insert our name */ |
7059c658 | 1461 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
e43eec81 | 1462 | &name, 0, ref_index); |
3650860b JB |
1463 | if (ret) |
1464 | goto out; | |
5a1d7843 | 1465 | |
f96d4474 JB |
1466 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
1467 | if (ret) | |
1468 | goto out; | |
5a1d7843 | 1469 | } |
77a5b9e3 | 1470 | /* Else, ret == 1, we already have a perfect match, we're done. */ |
5a1d7843 | 1471 | |
e43eec81 STD |
1472 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + name.len; |
1473 | kfree(name.name); | |
1474 | name.name = NULL; | |
f186373f MF |
1475 | if (log_ref_ver) { |
1476 | iput(dir); | |
1477 | dir = NULL; | |
1478 | } | |
5a1d7843 | 1479 | } |
e02119d5 | 1480 | |
1f250e92 FM |
1481 | /* |
1482 | * Before we overwrite the inode reference item in the subvolume tree | |
1483 | * with the item from the log tree, we must unlink all names from the | |
1484 | * parent directory that are in the subvolume's tree inode reference | |
1485 | * item, otherwise we end up with an inconsistent subvolume tree where | |
1486 | * dir index entries exist for a name but there is no inode reference | |
1487 | * item with the same name. | |
1488 | */ | |
1489 | ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot, | |
1490 | key); | |
1491 | if (ret) | |
1492 | goto out; | |
1493 | ||
e02119d5 CM |
1494 | /* finally write the back reference in the inode */ |
1495 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1496 | out: |
b3b4aa74 | 1497 | btrfs_release_path(path); |
e43eec81 | 1498 | kfree(name.name); |
e02119d5 CM |
1499 | iput(dir); |
1500 | iput(inode); | |
3650860b | 1501 | return ret; |
e02119d5 CM |
1502 | } |
1503 | ||
f186373f | 1504 | static int count_inode_extrefs(struct btrfs_root *root, |
36283658 | 1505 | struct btrfs_inode *inode, struct btrfs_path *path) |
f186373f MF |
1506 | { |
1507 | int ret = 0; | |
1508 | int name_len; | |
1509 | unsigned int nlink = 0; | |
1510 | u32 item_size; | |
1511 | u32 cur_offset = 0; | |
36283658 | 1512 | u64 inode_objectid = btrfs_ino(inode); |
f186373f MF |
1513 | u64 offset = 0; |
1514 | unsigned long ptr; | |
1515 | struct btrfs_inode_extref *extref; | |
1516 | struct extent_buffer *leaf; | |
1517 | ||
1518 | while (1) { | |
1519 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1520 | &extref, &offset); | |
1521 | if (ret) | |
1522 | break; | |
c71bf099 | 1523 | |
f186373f | 1524 | leaf = path->nodes[0]; |
3212fa14 | 1525 | item_size = btrfs_item_size(leaf, path->slots[0]); |
f186373f | 1526 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); |
2c2c452b | 1527 | cur_offset = 0; |
f186373f MF |
1528 | |
1529 | while (cur_offset < item_size) { | |
1530 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1531 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1532 | ||
1533 | nlink++; | |
1534 | ||
1535 | cur_offset += name_len + sizeof(*extref); | |
1536 | } | |
1537 | ||
1538 | offset++; | |
1539 | btrfs_release_path(path); | |
1540 | } | |
1541 | btrfs_release_path(path); | |
1542 | ||
2c2c452b | 1543 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1544 | return ret; |
1545 | return nlink; | |
1546 | } | |
1547 | ||
1548 | static int count_inode_refs(struct btrfs_root *root, | |
f329e319 | 1549 | struct btrfs_inode *inode, struct btrfs_path *path) |
e02119d5 | 1550 | { |
e02119d5 CM |
1551 | int ret; |
1552 | struct btrfs_key key; | |
f186373f | 1553 | unsigned int nlink = 0; |
e02119d5 CM |
1554 | unsigned long ptr; |
1555 | unsigned long ptr_end; | |
1556 | int name_len; | |
f329e319 | 1557 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1558 | |
33345d01 | 1559 | key.objectid = ino; |
e02119d5 CM |
1560 | key.type = BTRFS_INODE_REF_KEY; |
1561 | key.offset = (u64)-1; | |
1562 | ||
d397712b | 1563 | while (1) { |
e02119d5 CM |
1564 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1565 | if (ret < 0) | |
1566 | break; | |
1567 | if (ret > 0) { | |
1568 | if (path->slots[0] == 0) | |
1569 | break; | |
1570 | path->slots[0]--; | |
1571 | } | |
e93ae26f | 1572 | process_slot: |
e02119d5 CM |
1573 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1574 | path->slots[0]); | |
33345d01 | 1575 | if (key.objectid != ino || |
e02119d5 CM |
1576 | key.type != BTRFS_INODE_REF_KEY) |
1577 | break; | |
1578 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
3212fa14 | 1579 | ptr_end = ptr + btrfs_item_size(path->nodes[0], |
e02119d5 | 1580 | path->slots[0]); |
d397712b | 1581 | while (ptr < ptr_end) { |
e02119d5 CM |
1582 | struct btrfs_inode_ref *ref; |
1583 | ||
1584 | ref = (struct btrfs_inode_ref *)ptr; | |
1585 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1586 | ref); | |
1587 | ptr = (unsigned long)(ref + 1) + name_len; | |
1588 | nlink++; | |
1589 | } | |
1590 | ||
1591 | if (key.offset == 0) | |
1592 | break; | |
e93ae26f FDBM |
1593 | if (path->slots[0] > 0) { |
1594 | path->slots[0]--; | |
1595 | goto process_slot; | |
1596 | } | |
e02119d5 | 1597 | key.offset--; |
b3b4aa74 | 1598 | btrfs_release_path(path); |
e02119d5 | 1599 | } |
b3b4aa74 | 1600 | btrfs_release_path(path); |
f186373f MF |
1601 | |
1602 | return nlink; | |
1603 | } | |
1604 | ||
1605 | /* | |
1606 | * There are a few corners where the link count of the file can't | |
1607 | * be properly maintained during replay. So, instead of adding | |
1608 | * lots of complexity to the log code, we just scan the backrefs | |
1609 | * for any file that has been through replay. | |
1610 | * | |
1611 | * The scan will update the link count on the inode to reflect the | |
1612 | * number of back refs found. If it goes down to zero, the iput | |
1613 | * will free the inode. | |
1614 | */ | |
1615 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1616 | struct btrfs_root *root, | |
1617 | struct inode *inode) | |
1618 | { | |
1619 | struct btrfs_path *path; | |
1620 | int ret; | |
1621 | u64 nlink = 0; | |
4a0cc7ca | 1622 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1623 | |
1624 | path = btrfs_alloc_path(); | |
1625 | if (!path) | |
1626 | return -ENOMEM; | |
1627 | ||
f329e319 | 1628 | ret = count_inode_refs(root, BTRFS_I(inode), path); |
f186373f MF |
1629 | if (ret < 0) |
1630 | goto out; | |
1631 | ||
1632 | nlink = ret; | |
1633 | ||
36283658 | 1634 | ret = count_inode_extrefs(root, BTRFS_I(inode), path); |
f186373f MF |
1635 | if (ret < 0) |
1636 | goto out; | |
1637 | ||
1638 | nlink += ret; | |
1639 | ||
1640 | ret = 0; | |
1641 | ||
e02119d5 | 1642 | if (nlink != inode->i_nlink) { |
bfe86848 | 1643 | set_nlink(inode, nlink); |
f96d4474 JB |
1644 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
1645 | if (ret) | |
1646 | goto out; | |
e02119d5 | 1647 | } |
8d5bf1cb | 1648 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1649 | |
c71bf099 YZ |
1650 | if (inode->i_nlink == 0) { |
1651 | if (S_ISDIR(inode->i_mode)) { | |
1652 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1653 | ino, 1); |
3650860b JB |
1654 | if (ret) |
1655 | goto out; | |
c71bf099 | 1656 | } |
ecdcf3c2 NB |
1657 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1658 | if (ret == -EEXIST) | |
1659 | ret = 0; | |
12fcfd22 | 1660 | } |
12fcfd22 | 1661 | |
f186373f MF |
1662 | out: |
1663 | btrfs_free_path(path); | |
1664 | return ret; | |
e02119d5 CM |
1665 | } |
1666 | ||
1667 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1668 | struct btrfs_root *root, | |
1669 | struct btrfs_path *path) | |
1670 | { | |
1671 | int ret; | |
1672 | struct btrfs_key key; | |
1673 | struct inode *inode; | |
1674 | ||
1675 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1676 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1677 | key.offset = (u64)-1; | |
d397712b | 1678 | while (1) { |
e02119d5 CM |
1679 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1680 | if (ret < 0) | |
1681 | break; | |
1682 | ||
1683 | if (ret == 1) { | |
011b28ac | 1684 | ret = 0; |
e02119d5 CM |
1685 | if (path->slots[0] == 0) |
1686 | break; | |
1687 | path->slots[0]--; | |
1688 | } | |
1689 | ||
1690 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1691 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1692 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1693 | break; | |
1694 | ||
1695 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 | 1696 | if (ret) |
011b28ac | 1697 | break; |
e02119d5 | 1698 | |
b3b4aa74 | 1699 | btrfs_release_path(path); |
e02119d5 | 1700 | inode = read_one_inode(root, key.offset); |
011b28ac JB |
1701 | if (!inode) { |
1702 | ret = -EIO; | |
1703 | break; | |
1704 | } | |
e02119d5 CM |
1705 | |
1706 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1707 | iput(inode); |
3650860b | 1708 | if (ret) |
011b28ac | 1709 | break; |
e02119d5 | 1710 | |
12fcfd22 CM |
1711 | /* |
1712 | * fixup on a directory may create new entries, | |
1713 | * make sure we always look for the highset possible | |
1714 | * offset | |
1715 | */ | |
1716 | key.offset = (u64)-1; | |
e02119d5 | 1717 | } |
b3b4aa74 | 1718 | btrfs_release_path(path); |
65a246c5 | 1719 | return ret; |
e02119d5 CM |
1720 | } |
1721 | ||
1722 | ||
1723 | /* | |
1724 | * record a given inode in the fixup dir so we can check its link | |
1725 | * count when replay is done. The link count is incremented here | |
1726 | * so the inode won't go away until we check it | |
1727 | */ | |
1728 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1729 | struct btrfs_root *root, | |
1730 | struct btrfs_path *path, | |
1731 | u64 objectid) | |
1732 | { | |
1733 | struct btrfs_key key; | |
1734 | int ret = 0; | |
1735 | struct inode *inode; | |
1736 | ||
1737 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1738 | if (!inode) |
1739 | return -EIO; | |
e02119d5 CM |
1740 | |
1741 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1742 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1743 | key.offset = objectid; |
1744 | ||
1745 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1746 | ||
b3b4aa74 | 1747 | btrfs_release_path(path); |
e02119d5 | 1748 | if (ret == 0) { |
9bf7a489 JB |
1749 | if (!inode->i_nlink) |
1750 | set_nlink(inode, 1); | |
1751 | else | |
8b558c5f | 1752 | inc_nlink(inode); |
9a56fcd1 | 1753 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e02119d5 CM |
1754 | } else if (ret == -EEXIST) { |
1755 | ret = 0; | |
e02119d5 CM |
1756 | } |
1757 | iput(inode); | |
1758 | ||
1759 | return ret; | |
1760 | } | |
1761 | ||
1762 | /* | |
1763 | * when replaying the log for a directory, we only insert names | |
1764 | * for inodes that actually exist. This means an fsync on a directory | |
1765 | * does not implicitly fsync all the new files in it | |
1766 | */ | |
1767 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1768 | struct btrfs_root *root, | |
e02119d5 | 1769 | u64 dirid, u64 index, |
6db75318 | 1770 | const struct fscrypt_str *name, |
e02119d5 CM |
1771 | struct btrfs_key *location) |
1772 | { | |
1773 | struct inode *inode; | |
1774 | struct inode *dir; | |
1775 | int ret; | |
1776 | ||
1777 | inode = read_one_inode(root, location->objectid); | |
1778 | if (!inode) | |
1779 | return -ENOENT; | |
1780 | ||
1781 | dir = read_one_inode(root, dirid); | |
1782 | if (!dir) { | |
1783 | iput(inode); | |
1784 | return -EIO; | |
1785 | } | |
d555438b | 1786 | |
db0a669f | 1787 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, |
e43eec81 | 1788 | 1, index); |
e02119d5 CM |
1789 | |
1790 | /* FIXME, put inode into FIXUP list */ | |
1791 | ||
1792 | iput(inode); | |
1793 | iput(dir); | |
1794 | return ret; | |
1795 | } | |
1796 | ||
339d0354 FM |
1797 | static int delete_conflicting_dir_entry(struct btrfs_trans_handle *trans, |
1798 | struct btrfs_inode *dir, | |
1799 | struct btrfs_path *path, | |
1800 | struct btrfs_dir_item *dst_di, | |
1801 | const struct btrfs_key *log_key, | |
94a48aef | 1802 | u8 log_flags, |
339d0354 FM |
1803 | bool exists) |
1804 | { | |
1805 | struct btrfs_key found_key; | |
1806 | ||
1807 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1808 | /* The existing dentry points to the same inode, don't delete it. */ | |
1809 | if (found_key.objectid == log_key->objectid && | |
1810 | found_key.type == log_key->type && | |
1811 | found_key.offset == log_key->offset && | |
94a48aef | 1812 | btrfs_dir_flags(path->nodes[0], dst_di) == log_flags) |
339d0354 FM |
1813 | return 1; |
1814 | ||
1815 | /* | |
1816 | * Don't drop the conflicting directory entry if the inode for the new | |
1817 | * entry doesn't exist. | |
1818 | */ | |
1819 | if (!exists) | |
1820 | return 0; | |
1821 | ||
1822 | return drop_one_dir_item(trans, path, dir, dst_di); | |
1823 | } | |
1824 | ||
e02119d5 CM |
1825 | /* |
1826 | * take a single entry in a log directory item and replay it into | |
1827 | * the subvolume. | |
1828 | * | |
1829 | * if a conflicting item exists in the subdirectory already, | |
1830 | * the inode it points to is unlinked and put into the link count | |
1831 | * fix up tree. | |
1832 | * | |
1833 | * If a name from the log points to a file or directory that does | |
1834 | * not exist in the FS, it is skipped. fsyncs on directories | |
1835 | * do not force down inodes inside that directory, just changes to the | |
1836 | * names or unlinks in a directory. | |
bb53eda9 FM |
1837 | * |
1838 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1839 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1840 | */ |
1841 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1842 | struct btrfs_root *root, | |
1843 | struct btrfs_path *path, | |
1844 | struct extent_buffer *eb, | |
1845 | struct btrfs_dir_item *di, | |
1846 | struct btrfs_key *key) | |
1847 | { | |
6db75318 | 1848 | struct fscrypt_str name; |
339d0354 FM |
1849 | struct btrfs_dir_item *dir_dst_di; |
1850 | struct btrfs_dir_item *index_dst_di; | |
1851 | bool dir_dst_matches = false; | |
1852 | bool index_dst_matches = false; | |
e02119d5 | 1853 | struct btrfs_key log_key; |
339d0354 | 1854 | struct btrfs_key search_key; |
e02119d5 | 1855 | struct inode *dir; |
94a48aef | 1856 | u8 log_flags; |
cfd31269 FM |
1857 | bool exists; |
1858 | int ret; | |
339d0354 | 1859 | bool update_size = true; |
bb53eda9 | 1860 | bool name_added = false; |
e02119d5 CM |
1861 | |
1862 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1863 | if (!dir) |
1864 | return -EIO; | |
e02119d5 | 1865 | |
e43eec81 STD |
1866 | ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name); |
1867 | if (ret) | |
2bac325e | 1868 | goto out; |
2a29edc6 | 1869 | |
94a48aef | 1870 | log_flags = btrfs_dir_flags(eb, di); |
e02119d5 | 1871 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); |
cfd31269 | 1872 | ret = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
b3b4aa74 | 1873 | btrfs_release_path(path); |
cfd31269 FM |
1874 | if (ret < 0) |
1875 | goto out; | |
1876 | exists = (ret == 0); | |
1877 | ret = 0; | |
4bef0848 | 1878 | |
339d0354 | 1879 | dir_dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, |
e43eec81 | 1880 | &name, 1); |
339d0354 FM |
1881 | if (IS_ERR(dir_dst_di)) { |
1882 | ret = PTR_ERR(dir_dst_di); | |
3650860b | 1883 | goto out; |
339d0354 FM |
1884 | } else if (dir_dst_di) { |
1885 | ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path, | |
94a48aef OS |
1886 | dir_dst_di, &log_key, |
1887 | log_flags, exists); | |
339d0354 FM |
1888 | if (ret < 0) |
1889 | goto out; | |
1890 | dir_dst_matches = (ret == 1); | |
e02119d5 | 1891 | } |
e15ac641 | 1892 | |
339d0354 FM |
1893 | btrfs_release_path(path); |
1894 | ||
1895 | index_dst_di = btrfs_lookup_dir_index_item(trans, root, path, | |
1896 | key->objectid, key->offset, | |
e43eec81 | 1897 | &name, 1); |
339d0354 FM |
1898 | if (IS_ERR(index_dst_di)) { |
1899 | ret = PTR_ERR(index_dst_di); | |
e15ac641 | 1900 | goto out; |
339d0354 FM |
1901 | } else if (index_dst_di) { |
1902 | ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path, | |
1903 | index_dst_di, &log_key, | |
94a48aef | 1904 | log_flags, exists); |
339d0354 | 1905 | if (ret < 0) |
e02119d5 | 1906 | goto out; |
339d0354 | 1907 | index_dst_matches = (ret == 1); |
e02119d5 CM |
1908 | } |
1909 | ||
339d0354 FM |
1910 | btrfs_release_path(path); |
1911 | ||
1912 | if (dir_dst_matches && index_dst_matches) { | |
1913 | ret = 0; | |
a2cc11db | 1914 | update_size = false; |
e02119d5 CM |
1915 | goto out; |
1916 | } | |
1917 | ||
725af92a NB |
1918 | /* |
1919 | * Check if the inode reference exists in the log for the given name, | |
1920 | * inode and parent inode | |
1921 | */ | |
339d0354 FM |
1922 | search_key.objectid = log_key.objectid; |
1923 | search_key.type = BTRFS_INODE_REF_KEY; | |
1924 | search_key.offset = key->objectid; | |
e43eec81 | 1925 | ret = backref_in_log(root->log_root, &search_key, 0, &name); |
725af92a NB |
1926 | if (ret < 0) { |
1927 | goto out; | |
1928 | } else if (ret) { | |
1929 | /* The dentry will be added later. */ | |
1930 | ret = 0; | |
1931 | update_size = false; | |
1932 | goto out; | |
1933 | } | |
1934 | ||
339d0354 FM |
1935 | search_key.objectid = log_key.objectid; |
1936 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1937 | search_key.offset = key->objectid; | |
e43eec81 | 1938 | ret = backref_in_log(root->log_root, &search_key, key->objectid, &name); |
725af92a NB |
1939 | if (ret < 0) { |
1940 | goto out; | |
1941 | } else if (ret) { | |
df8d116f FM |
1942 | /* The dentry will be added later. */ |
1943 | ret = 0; | |
1944 | update_size = false; | |
1945 | goto out; | |
1946 | } | |
b3b4aa74 | 1947 | btrfs_release_path(path); |
60d53eb3 | 1948 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
e43eec81 | 1949 | &name, &log_key); |
df8d116f | 1950 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 1951 | goto out; |
bb53eda9 FM |
1952 | if (!ret) |
1953 | name_added = true; | |
d555438b | 1954 | update_size = false; |
3650860b | 1955 | ret = 0; |
339d0354 FM |
1956 | |
1957 | out: | |
1958 | if (!ret && update_size) { | |
e43eec81 | 1959 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name.len * 2); |
339d0354 FM |
1960 | ret = btrfs_update_inode(trans, root, BTRFS_I(dir)); |
1961 | } | |
e43eec81 | 1962 | kfree(name.name); |
339d0354 FM |
1963 | iput(dir); |
1964 | if (!ret && name_added) | |
1965 | ret = 1; | |
1966 | return ret; | |
e02119d5 CM |
1967 | } |
1968 | ||
339d0354 | 1969 | /* Replay one dir item from a BTRFS_DIR_INDEX_KEY key. */ |
e02119d5 CM |
1970 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, |
1971 | struct btrfs_root *root, | |
1972 | struct btrfs_path *path, | |
1973 | struct extent_buffer *eb, int slot, | |
1974 | struct btrfs_key *key) | |
1975 | { | |
339d0354 | 1976 | int ret; |
e02119d5 | 1977 | struct btrfs_dir_item *di; |
e02119d5 | 1978 | |
339d0354 FM |
1979 | /* We only log dir index keys, which only contain a single dir item. */ |
1980 | ASSERT(key->type == BTRFS_DIR_INDEX_KEY); | |
bb53eda9 | 1981 | |
339d0354 FM |
1982 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); |
1983 | ret = replay_one_name(trans, root, path, eb, di, key); | |
1984 | if (ret < 0) | |
1985 | return ret; | |
bb53eda9 | 1986 | |
339d0354 FM |
1987 | /* |
1988 | * If this entry refers to a non-directory (directories can not have a | |
1989 | * link count > 1) and it was added in the transaction that was not | |
1990 | * committed, make sure we fixup the link count of the inode the entry | |
1991 | * points to. Otherwise something like the following would result in a | |
1992 | * directory pointing to an inode with a wrong link that does not account | |
1993 | * for this dir entry: | |
1994 | * | |
1995 | * mkdir testdir | |
1996 | * touch testdir/foo | |
1997 | * touch testdir/bar | |
1998 | * sync | |
1999 | * | |
2000 | * ln testdir/bar testdir/bar_link | |
2001 | * ln testdir/foo testdir/foo_link | |
2002 | * xfs_io -c "fsync" testdir/bar | |
2003 | * | |
2004 | * <power failure> | |
2005 | * | |
2006 | * mount fs, log replay happens | |
2007 | * | |
2008 | * File foo would remain with a link count of 1 when it has two entries | |
2009 | * pointing to it in the directory testdir. This would make it impossible | |
2010 | * to ever delete the parent directory has it would result in stale | |
2011 | * dentries that can never be deleted. | |
2012 | */ | |
94a48aef | 2013 | if (ret == 1 && btrfs_dir_ftype(eb, di) != BTRFS_FT_DIR) { |
339d0354 FM |
2014 | struct btrfs_path *fixup_path; |
2015 | struct btrfs_key di_key; | |
bb53eda9 | 2016 | |
339d0354 FM |
2017 | fixup_path = btrfs_alloc_path(); |
2018 | if (!fixup_path) | |
2019 | return -ENOMEM; | |
2020 | ||
2021 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2022 | ret = link_to_fixup_dir(trans, root, fixup_path, di_key.objectid); | |
2023 | btrfs_free_path(fixup_path); | |
e02119d5 | 2024 | } |
339d0354 | 2025 | |
bb53eda9 | 2026 | return ret; |
e02119d5 CM |
2027 | } |
2028 | ||
2029 | /* | |
2030 | * directory replay has two parts. There are the standard directory | |
2031 | * items in the log copied from the subvolume, and range items | |
2032 | * created in the log while the subvolume was logged. | |
2033 | * | |
2034 | * The range items tell us which parts of the key space the log | |
2035 | * is authoritative for. During replay, if a key in the subvolume | |
2036 | * directory is in a logged range item, but not actually in the log | |
2037 | * that means it was deleted from the directory before the fsync | |
2038 | * and should be removed. | |
2039 | */ | |
2040 | static noinline int find_dir_range(struct btrfs_root *root, | |
2041 | struct btrfs_path *path, | |
ccae4a19 | 2042 | u64 dirid, |
e02119d5 CM |
2043 | u64 *start_ret, u64 *end_ret) |
2044 | { | |
2045 | struct btrfs_key key; | |
2046 | u64 found_end; | |
2047 | struct btrfs_dir_log_item *item; | |
2048 | int ret; | |
2049 | int nritems; | |
2050 | ||
2051 | if (*start_ret == (u64)-1) | |
2052 | return 1; | |
2053 | ||
2054 | key.objectid = dirid; | |
ccae4a19 | 2055 | key.type = BTRFS_DIR_LOG_INDEX_KEY; |
e02119d5 CM |
2056 | key.offset = *start_ret; |
2057 | ||
2058 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2059 | if (ret < 0) | |
2060 | goto out; | |
2061 | if (ret > 0) { | |
2062 | if (path->slots[0] == 0) | |
2063 | goto out; | |
2064 | path->slots[0]--; | |
2065 | } | |
2066 | if (ret != 0) | |
2067 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2068 | ||
ccae4a19 | 2069 | if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) { |
e02119d5 CM |
2070 | ret = 1; |
2071 | goto next; | |
2072 | } | |
2073 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2074 | struct btrfs_dir_log_item); | |
2075 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2076 | ||
2077 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
2078 | ret = 0; | |
2079 | *start_ret = key.offset; | |
2080 | *end_ret = found_end; | |
2081 | goto out; | |
2082 | } | |
2083 | ret = 1; | |
2084 | next: | |
2085 | /* check the next slot in the tree to see if it is a valid item */ | |
2086 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 2087 | path->slots[0]++; |
e02119d5 CM |
2088 | if (path->slots[0] >= nritems) { |
2089 | ret = btrfs_next_leaf(root, path); | |
2090 | if (ret) | |
2091 | goto out; | |
e02119d5 CM |
2092 | } |
2093 | ||
2094 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2095 | ||
ccae4a19 | 2096 | if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) { |
e02119d5 CM |
2097 | ret = 1; |
2098 | goto out; | |
2099 | } | |
2100 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2101 | struct btrfs_dir_log_item); | |
2102 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2103 | *start_ret = key.offset; | |
2104 | *end_ret = found_end; | |
2105 | ret = 0; | |
2106 | out: | |
b3b4aa74 | 2107 | btrfs_release_path(path); |
e02119d5 CM |
2108 | return ret; |
2109 | } | |
2110 | ||
2111 | /* | |
2112 | * this looks for a given directory item in the log. If the directory | |
2113 | * item is not in the log, the item is removed and the inode it points | |
2114 | * to is unlinked | |
2115 | */ | |
2116 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
e02119d5 CM |
2117 | struct btrfs_root *log, |
2118 | struct btrfs_path *path, | |
2119 | struct btrfs_path *log_path, | |
2120 | struct inode *dir, | |
2121 | struct btrfs_key *dir_key) | |
2122 | { | |
d1ed82f3 | 2123 | struct btrfs_root *root = BTRFS_I(dir)->root; |
e02119d5 CM |
2124 | int ret; |
2125 | struct extent_buffer *eb; | |
2126 | int slot; | |
e02119d5 | 2127 | struct btrfs_dir_item *di; |
6db75318 | 2128 | struct fscrypt_str name; |
ccae4a19 | 2129 | struct inode *inode = NULL; |
e02119d5 CM |
2130 | struct btrfs_key location; |
2131 | ||
ccae4a19 | 2132 | /* |
143823cf | 2133 | * Currently we only log dir index keys. Even if we replay a log created |
ccae4a19 FM |
2134 | * by an older kernel that logged both dir index and dir item keys, all |
2135 | * we need to do is process the dir index keys, we (and our caller) can | |
2136 | * safely ignore dir item keys (key type BTRFS_DIR_ITEM_KEY). | |
2137 | */ | |
2138 | ASSERT(dir_key->type == BTRFS_DIR_INDEX_KEY); | |
2139 | ||
e02119d5 CM |
2140 | eb = path->nodes[0]; |
2141 | slot = path->slots[0]; | |
ccae4a19 | 2142 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); |
e43eec81 STD |
2143 | ret = read_alloc_one_name(eb, di + 1, btrfs_dir_name_len(eb, di), &name); |
2144 | if (ret) | |
ccae4a19 | 2145 | goto out; |
3650860b | 2146 | |
ccae4a19 FM |
2147 | if (log) { |
2148 | struct btrfs_dir_item *log_di; | |
e02119d5 | 2149 | |
ccae4a19 FM |
2150 | log_di = btrfs_lookup_dir_index_item(trans, log, log_path, |
2151 | dir_key->objectid, | |
e43eec81 | 2152 | dir_key->offset, &name, 0); |
ccae4a19 FM |
2153 | if (IS_ERR(log_di)) { |
2154 | ret = PTR_ERR(log_di); | |
2155 | goto out; | |
2156 | } else if (log_di) { | |
2157 | /* The dentry exists in the log, we have nothing to do. */ | |
e02119d5 CM |
2158 | ret = 0; |
2159 | goto out; | |
2160 | } | |
ccae4a19 | 2161 | } |
e02119d5 | 2162 | |
ccae4a19 FM |
2163 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
2164 | btrfs_release_path(path); | |
2165 | btrfs_release_path(log_path); | |
2166 | inode = read_one_inode(root, location.objectid); | |
2167 | if (!inode) { | |
2168 | ret = -EIO; | |
2169 | goto out; | |
e02119d5 | 2170 | } |
ccae4a19 FM |
2171 | |
2172 | ret = link_to_fixup_dir(trans, root, path, location.objectid); | |
2173 | if (ret) | |
2174 | goto out; | |
2175 | ||
2176 | inc_nlink(inode); | |
313ab753 | 2177 | ret = unlink_inode_for_log_replay(trans, BTRFS_I(dir), BTRFS_I(inode), |
e43eec81 | 2178 | &name); |
ccae4a19 FM |
2179 | /* |
2180 | * Unlike dir item keys, dir index keys can only have one name (entry) in | |
2181 | * them, as there are no key collisions since each key has a unique offset | |
2182 | * (an index number), so we're done. | |
2183 | */ | |
e02119d5 | 2184 | out: |
b3b4aa74 DS |
2185 | btrfs_release_path(path); |
2186 | btrfs_release_path(log_path); | |
e43eec81 | 2187 | kfree(name.name); |
ccae4a19 | 2188 | iput(inode); |
e02119d5 CM |
2189 | return ret; |
2190 | } | |
2191 | ||
4f764e51 FM |
2192 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2193 | struct btrfs_root *root, | |
2194 | struct btrfs_root *log, | |
2195 | struct btrfs_path *path, | |
2196 | const u64 ino) | |
2197 | { | |
2198 | struct btrfs_key search_key; | |
2199 | struct btrfs_path *log_path; | |
2200 | int i; | |
2201 | int nritems; | |
2202 | int ret; | |
2203 | ||
2204 | log_path = btrfs_alloc_path(); | |
2205 | if (!log_path) | |
2206 | return -ENOMEM; | |
2207 | ||
2208 | search_key.objectid = ino; | |
2209 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2210 | search_key.offset = 0; | |
2211 | again: | |
2212 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2213 | if (ret < 0) | |
2214 | goto out; | |
2215 | process_leaf: | |
2216 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2217 | for (i = path->slots[0]; i < nritems; i++) { | |
2218 | struct btrfs_key key; | |
2219 | struct btrfs_dir_item *di; | |
2220 | struct btrfs_dir_item *log_di; | |
2221 | u32 total_size; | |
2222 | u32 cur; | |
2223 | ||
2224 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2225 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2226 | ret = 0; | |
2227 | goto out; | |
2228 | } | |
2229 | ||
2230 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
3212fa14 | 2231 | total_size = btrfs_item_size(path->nodes[0], i); |
4f764e51 FM |
2232 | cur = 0; |
2233 | while (cur < total_size) { | |
2234 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2235 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2236 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2237 | char *name; | |
2238 | ||
2239 | name = kmalloc(name_len, GFP_NOFS); | |
2240 | if (!name) { | |
2241 | ret = -ENOMEM; | |
2242 | goto out; | |
2243 | } | |
2244 | read_extent_buffer(path->nodes[0], name, | |
2245 | (unsigned long)(di + 1), name_len); | |
2246 | ||
2247 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2248 | name, name_len, 0); | |
2249 | btrfs_release_path(log_path); | |
2250 | if (!log_di) { | |
2251 | /* Doesn't exist in log tree, so delete it. */ | |
2252 | btrfs_release_path(path); | |
2253 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2254 | name, name_len, -1); | |
2255 | kfree(name); | |
2256 | if (IS_ERR(di)) { | |
2257 | ret = PTR_ERR(di); | |
2258 | goto out; | |
2259 | } | |
2260 | ASSERT(di); | |
2261 | ret = btrfs_delete_one_dir_name(trans, root, | |
2262 | path, di); | |
2263 | if (ret) | |
2264 | goto out; | |
2265 | btrfs_release_path(path); | |
2266 | search_key = key; | |
2267 | goto again; | |
2268 | } | |
2269 | kfree(name); | |
2270 | if (IS_ERR(log_di)) { | |
2271 | ret = PTR_ERR(log_di); | |
2272 | goto out; | |
2273 | } | |
2274 | cur += this_len; | |
2275 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2276 | } | |
2277 | } | |
2278 | ret = btrfs_next_leaf(root, path); | |
2279 | if (ret > 0) | |
2280 | ret = 0; | |
2281 | else if (ret == 0) | |
2282 | goto process_leaf; | |
2283 | out: | |
2284 | btrfs_free_path(log_path); | |
2285 | btrfs_release_path(path); | |
2286 | return ret; | |
2287 | } | |
2288 | ||
2289 | ||
e02119d5 CM |
2290 | /* |
2291 | * deletion replay happens before we copy any new directory items | |
2292 | * out of the log or out of backreferences from inodes. It | |
2293 | * scans the log to find ranges of keys that log is authoritative for, | |
2294 | * and then scans the directory to find items in those ranges that are | |
2295 | * not present in the log. | |
2296 | * | |
2297 | * Anything we don't find in the log is unlinked and removed from the | |
2298 | * directory. | |
2299 | */ | |
2300 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2301 | struct btrfs_root *root, | |
2302 | struct btrfs_root *log, | |
2303 | struct btrfs_path *path, | |
12fcfd22 | 2304 | u64 dirid, int del_all) |
e02119d5 CM |
2305 | { |
2306 | u64 range_start; | |
2307 | u64 range_end; | |
e02119d5 CM |
2308 | int ret = 0; |
2309 | struct btrfs_key dir_key; | |
2310 | struct btrfs_key found_key; | |
2311 | struct btrfs_path *log_path; | |
2312 | struct inode *dir; | |
2313 | ||
2314 | dir_key.objectid = dirid; | |
ccae4a19 | 2315 | dir_key.type = BTRFS_DIR_INDEX_KEY; |
e02119d5 CM |
2316 | log_path = btrfs_alloc_path(); |
2317 | if (!log_path) | |
2318 | return -ENOMEM; | |
2319 | ||
2320 | dir = read_one_inode(root, dirid); | |
2321 | /* it isn't an error if the inode isn't there, that can happen | |
2322 | * because we replay the deletes before we copy in the inode item | |
2323 | * from the log | |
2324 | */ | |
2325 | if (!dir) { | |
2326 | btrfs_free_path(log_path); | |
2327 | return 0; | |
2328 | } | |
ccae4a19 | 2329 | |
e02119d5 CM |
2330 | range_start = 0; |
2331 | range_end = 0; | |
d397712b | 2332 | while (1) { |
12fcfd22 CM |
2333 | if (del_all) |
2334 | range_end = (u64)-1; | |
2335 | else { | |
ccae4a19 | 2336 | ret = find_dir_range(log, path, dirid, |
12fcfd22 | 2337 | &range_start, &range_end); |
10adb115 FM |
2338 | if (ret < 0) |
2339 | goto out; | |
2340 | else if (ret > 0) | |
12fcfd22 CM |
2341 | break; |
2342 | } | |
e02119d5 CM |
2343 | |
2344 | dir_key.offset = range_start; | |
d397712b | 2345 | while (1) { |
e02119d5 CM |
2346 | int nritems; |
2347 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2348 | 0, 0); | |
2349 | if (ret < 0) | |
2350 | goto out; | |
2351 | ||
2352 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2353 | if (path->slots[0] >= nritems) { | |
2354 | ret = btrfs_next_leaf(root, path); | |
b98def7c | 2355 | if (ret == 1) |
e02119d5 | 2356 | break; |
b98def7c LB |
2357 | else if (ret < 0) |
2358 | goto out; | |
e02119d5 CM |
2359 | } |
2360 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2361 | path->slots[0]); | |
2362 | if (found_key.objectid != dirid || | |
ccae4a19 FM |
2363 | found_key.type != dir_key.type) { |
2364 | ret = 0; | |
2365 | goto out; | |
2366 | } | |
e02119d5 CM |
2367 | |
2368 | if (found_key.offset > range_end) | |
2369 | break; | |
2370 | ||
d1ed82f3 | 2371 | ret = check_item_in_log(trans, log, path, |
12fcfd22 CM |
2372 | log_path, dir, |
2373 | &found_key); | |
3650860b JB |
2374 | if (ret) |
2375 | goto out; | |
e02119d5 CM |
2376 | if (found_key.offset == (u64)-1) |
2377 | break; | |
2378 | dir_key.offset = found_key.offset + 1; | |
2379 | } | |
b3b4aa74 | 2380 | btrfs_release_path(path); |
e02119d5 CM |
2381 | if (range_end == (u64)-1) |
2382 | break; | |
2383 | range_start = range_end + 1; | |
2384 | } | |
e02119d5 | 2385 | ret = 0; |
e02119d5 | 2386 | out: |
b3b4aa74 | 2387 | btrfs_release_path(path); |
e02119d5 CM |
2388 | btrfs_free_path(log_path); |
2389 | iput(dir); | |
2390 | return ret; | |
2391 | } | |
2392 | ||
2393 | /* | |
2394 | * the process_func used to replay items from the log tree. This | |
2395 | * gets called in two different stages. The first stage just looks | |
2396 | * for inodes and makes sure they are all copied into the subvolume. | |
2397 | * | |
2398 | * The second stage copies all the other item types from the log into | |
2399 | * the subvolume. The two stage approach is slower, but gets rid of | |
2400 | * lots of complexity around inodes referencing other inodes that exist | |
2401 | * only in the log (references come from either directory items or inode | |
2402 | * back refs). | |
2403 | */ | |
2404 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 2405 | struct walk_control *wc, u64 gen, int level) |
e02119d5 CM |
2406 | { |
2407 | int nritems; | |
2408 | struct btrfs_path *path; | |
2409 | struct btrfs_root *root = wc->replay_dest; | |
2410 | struct btrfs_key key; | |
e02119d5 CM |
2411 | int i; |
2412 | int ret; | |
2413 | ||
6a2e9dc4 | 2414 | ret = btrfs_read_extent_buffer(eb, gen, level, NULL); |
018642a1 TI |
2415 | if (ret) |
2416 | return ret; | |
e02119d5 CM |
2417 | |
2418 | level = btrfs_header_level(eb); | |
2419 | ||
2420 | if (level != 0) | |
2421 | return 0; | |
2422 | ||
2423 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2424 | if (!path) |
2425 | return -ENOMEM; | |
e02119d5 CM |
2426 | |
2427 | nritems = btrfs_header_nritems(eb); | |
2428 | for (i = 0; i < nritems; i++) { | |
2429 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2430 | |
2431 | /* inode keys are done during the first stage */ | |
2432 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2433 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2434 | struct btrfs_inode_item *inode_item; |
2435 | u32 mode; | |
2436 | ||
2437 | inode_item = btrfs_item_ptr(eb, i, | |
2438 | struct btrfs_inode_item); | |
f2d72f42 FM |
2439 | /* |
2440 | * If we have a tmpfile (O_TMPFILE) that got fsync'ed | |
2441 | * and never got linked before the fsync, skip it, as | |
2442 | * replaying it is pointless since it would be deleted | |
2443 | * later. We skip logging tmpfiles, but it's always | |
2444 | * possible we are replaying a log created with a kernel | |
2445 | * that used to log tmpfiles. | |
2446 | */ | |
2447 | if (btrfs_inode_nlink(eb, inode_item) == 0) { | |
2448 | wc->ignore_cur_inode = true; | |
2449 | continue; | |
2450 | } else { | |
2451 | wc->ignore_cur_inode = false; | |
2452 | } | |
4f764e51 FM |
2453 | ret = replay_xattr_deletes(wc->trans, root, log, |
2454 | path, key.objectid); | |
2455 | if (ret) | |
2456 | break; | |
e02119d5 CM |
2457 | mode = btrfs_inode_mode(eb, inode_item); |
2458 | if (S_ISDIR(mode)) { | |
2459 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2460 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2461 | if (ret) |
2462 | break; | |
e02119d5 CM |
2463 | } |
2464 | ret = overwrite_item(wc->trans, root, path, | |
2465 | eb, i, &key); | |
b50c6e25 JB |
2466 | if (ret) |
2467 | break; | |
e02119d5 | 2468 | |
471d557a FM |
2469 | /* |
2470 | * Before replaying extents, truncate the inode to its | |
2471 | * size. We need to do it now and not after log replay | |
2472 | * because before an fsync we can have prealloc extents | |
2473 | * added beyond the inode's i_size. If we did it after, | |
2474 | * through orphan cleanup for example, we would drop | |
2475 | * those prealloc extents just after replaying them. | |
e02119d5 CM |
2476 | */ |
2477 | if (S_ISREG(mode)) { | |
5893dfb9 | 2478 | struct btrfs_drop_extents_args drop_args = { 0 }; |
471d557a FM |
2479 | struct inode *inode; |
2480 | u64 from; | |
2481 | ||
2482 | inode = read_one_inode(root, key.objectid); | |
2483 | if (!inode) { | |
2484 | ret = -EIO; | |
2485 | break; | |
2486 | } | |
2487 | from = ALIGN(i_size_read(inode), | |
2488 | root->fs_info->sectorsize); | |
5893dfb9 FM |
2489 | drop_args.start = from; |
2490 | drop_args.end = (u64)-1; | |
2491 | drop_args.drop_cache = true; | |
2492 | ret = btrfs_drop_extents(wc->trans, root, | |
2493 | BTRFS_I(inode), | |
2494 | &drop_args); | |
471d557a | 2495 | if (!ret) { |
2766ff61 FM |
2496 | inode_sub_bytes(inode, |
2497 | drop_args.bytes_found); | |
f2d72f42 | 2498 | /* Update the inode's nbytes. */ |
471d557a | 2499 | ret = btrfs_update_inode(wc->trans, |
9a56fcd1 | 2500 | root, BTRFS_I(inode)); |
471d557a FM |
2501 | } |
2502 | iput(inode); | |
b50c6e25 JB |
2503 | if (ret) |
2504 | break; | |
e02119d5 | 2505 | } |
c71bf099 | 2506 | |
e02119d5 CM |
2507 | ret = link_to_fixup_dir(wc->trans, root, |
2508 | path, key.objectid); | |
b50c6e25 JB |
2509 | if (ret) |
2510 | break; | |
e02119d5 | 2511 | } |
dd8e7217 | 2512 | |
f2d72f42 FM |
2513 | if (wc->ignore_cur_inode) |
2514 | continue; | |
2515 | ||
dd8e7217 JB |
2516 | if (key.type == BTRFS_DIR_INDEX_KEY && |
2517 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2518 | ret = replay_one_dir_item(wc->trans, root, path, | |
2519 | eb, i, &key); | |
2520 | if (ret) | |
2521 | break; | |
2522 | } | |
2523 | ||
e02119d5 CM |
2524 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2525 | continue; | |
2526 | ||
2527 | /* these keys are simply copied */ | |
2528 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2529 | ret = overwrite_item(wc->trans, root, path, | |
2530 | eb, i, &key); | |
b50c6e25 JB |
2531 | if (ret) |
2532 | break; | |
2da1c669 LB |
2533 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2534 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2535 | ret = add_inode_ref(wc->trans, root, log, path, |
2536 | eb, i, &key); | |
b50c6e25 JB |
2537 | if (ret && ret != -ENOENT) |
2538 | break; | |
2539 | ret = 0; | |
e02119d5 CM |
2540 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2541 | ret = replay_one_extent(wc->trans, root, path, | |
2542 | eb, i, &key); | |
b50c6e25 JB |
2543 | if (ret) |
2544 | break; | |
e02119d5 | 2545 | } |
339d0354 FM |
2546 | /* |
2547 | * We don't log BTRFS_DIR_ITEM_KEY keys anymore, only the | |
2548 | * BTRFS_DIR_INDEX_KEY items which we use to derive the | |
2549 | * BTRFS_DIR_ITEM_KEY items. If we are replaying a log from an | |
2550 | * older kernel with such keys, ignore them. | |
2551 | */ | |
e02119d5 CM |
2552 | } |
2553 | btrfs_free_path(path); | |
b50c6e25 | 2554 | return ret; |
e02119d5 CM |
2555 | } |
2556 | ||
6787bb9f NB |
2557 | /* |
2558 | * Correctly adjust the reserved bytes occupied by a log tree extent buffer | |
2559 | */ | |
2560 | static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start) | |
2561 | { | |
2562 | struct btrfs_block_group *cache; | |
2563 | ||
2564 | cache = btrfs_lookup_block_group(fs_info, start); | |
2565 | if (!cache) { | |
2566 | btrfs_err(fs_info, "unable to find block group for %llu", start); | |
2567 | return; | |
2568 | } | |
2569 | ||
2570 | spin_lock(&cache->space_info->lock); | |
2571 | spin_lock(&cache->lock); | |
2572 | cache->reserved -= fs_info->nodesize; | |
2573 | cache->space_info->bytes_reserved -= fs_info->nodesize; | |
2574 | spin_unlock(&cache->lock); | |
2575 | spin_unlock(&cache->space_info->lock); | |
2576 | ||
2577 | btrfs_put_block_group(cache); | |
2578 | } | |
2579 | ||
d397712b | 2580 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2581 | struct btrfs_root *root, |
2582 | struct btrfs_path *path, int *level, | |
2583 | struct walk_control *wc) | |
2584 | { | |
0b246afa | 2585 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2586 | u64 bytenr; |
2587 | u64 ptr_gen; | |
2588 | struct extent_buffer *next; | |
2589 | struct extent_buffer *cur; | |
e02119d5 CM |
2590 | u32 blocksize; |
2591 | int ret = 0; | |
2592 | ||
d397712b | 2593 | while (*level > 0) { |
581c1760 QW |
2594 | struct btrfs_key first_key; |
2595 | ||
e02119d5 CM |
2596 | cur = path->nodes[*level]; |
2597 | ||
fae7f21c | 2598 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2599 | |
2600 | if (path->slots[*level] >= | |
2601 | btrfs_header_nritems(cur)) | |
2602 | break; | |
2603 | ||
2604 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2605 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
581c1760 | 2606 | btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]); |
0b246afa | 2607 | blocksize = fs_info->nodesize; |
e02119d5 | 2608 | |
3fbaf258 JB |
2609 | next = btrfs_find_create_tree_block(fs_info, bytenr, |
2610 | btrfs_header_owner(cur), | |
2611 | *level - 1); | |
c871b0f2 LB |
2612 | if (IS_ERR(next)) |
2613 | return PTR_ERR(next); | |
e02119d5 | 2614 | |
e02119d5 | 2615 | if (*level == 1) { |
581c1760 QW |
2616 | ret = wc->process_func(root, next, wc, ptr_gen, |
2617 | *level - 1); | |
b50c6e25 JB |
2618 | if (ret) { |
2619 | free_extent_buffer(next); | |
1e5063d0 | 2620 | return ret; |
b50c6e25 | 2621 | } |
4a500fd1 | 2622 | |
e02119d5 CM |
2623 | path->slots[*level]++; |
2624 | if (wc->free) { | |
6a2e9dc4 | 2625 | ret = btrfs_read_extent_buffer(next, ptr_gen, |
581c1760 | 2626 | *level - 1, &first_key); |
018642a1 TI |
2627 | if (ret) { |
2628 | free_extent_buffer(next); | |
2629 | return ret; | |
2630 | } | |
e02119d5 | 2631 | |
681ae509 JB |
2632 | if (trans) { |
2633 | btrfs_tree_lock(next); | |
6a884d7d | 2634 | btrfs_clean_tree_block(next); |
681ae509 JB |
2635 | btrfs_wait_tree_block_writeback(next); |
2636 | btrfs_tree_unlock(next); | |
7bfc1007 | 2637 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2638 | bytenr, blocksize); |
2639 | if (ret) { | |
2640 | free_extent_buffer(next); | |
2641 | return ret; | |
2642 | } | |
d3575156 NA |
2643 | btrfs_redirty_list_add( |
2644 | trans->transaction, next); | |
1846430c LB |
2645 | } else { |
2646 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2647 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2648 | unaccount_log_buffer(fs_info, bytenr); |
3650860b | 2649 | } |
e02119d5 CM |
2650 | } |
2651 | free_extent_buffer(next); | |
2652 | continue; | |
2653 | } | |
6a2e9dc4 | 2654 | ret = btrfs_read_extent_buffer(next, ptr_gen, *level - 1, &first_key); |
018642a1 TI |
2655 | if (ret) { |
2656 | free_extent_buffer(next); | |
2657 | return ret; | |
2658 | } | |
e02119d5 | 2659 | |
e02119d5 CM |
2660 | if (path->nodes[*level-1]) |
2661 | free_extent_buffer(path->nodes[*level-1]); | |
2662 | path->nodes[*level-1] = next; | |
2663 | *level = btrfs_header_level(next); | |
2664 | path->slots[*level] = 0; | |
2665 | cond_resched(); | |
2666 | } | |
4a500fd1 | 2667 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2668 | |
2669 | cond_resched(); | |
2670 | return 0; | |
2671 | } | |
2672 | ||
d397712b | 2673 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2674 | struct btrfs_root *root, |
2675 | struct btrfs_path *path, int *level, | |
2676 | struct walk_control *wc) | |
2677 | { | |
0b246afa | 2678 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2679 | int i; |
2680 | int slot; | |
2681 | int ret; | |
2682 | ||
d397712b | 2683 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2684 | slot = path->slots[i]; |
4a500fd1 | 2685 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2686 | path->slots[i]++; |
2687 | *level = i; | |
2688 | WARN_ON(*level == 0); | |
2689 | return 0; | |
2690 | } else { | |
1e5063d0 | 2691 | ret = wc->process_func(root, path->nodes[*level], wc, |
581c1760 QW |
2692 | btrfs_header_generation(path->nodes[*level]), |
2693 | *level); | |
1e5063d0 MF |
2694 | if (ret) |
2695 | return ret; | |
2696 | ||
e02119d5 CM |
2697 | if (wc->free) { |
2698 | struct extent_buffer *next; | |
2699 | ||
2700 | next = path->nodes[*level]; | |
2701 | ||
681ae509 JB |
2702 | if (trans) { |
2703 | btrfs_tree_lock(next); | |
6a884d7d | 2704 | btrfs_clean_tree_block(next); |
681ae509 JB |
2705 | btrfs_wait_tree_block_writeback(next); |
2706 | btrfs_tree_unlock(next); | |
7bfc1007 | 2707 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2708 | path->nodes[*level]->start, |
2709 | path->nodes[*level]->len); | |
2710 | if (ret) | |
2711 | return ret; | |
84c25448 NA |
2712 | btrfs_redirty_list_add(trans->transaction, |
2713 | next); | |
1846430c LB |
2714 | } else { |
2715 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2716 | clear_extent_buffer_dirty(next); | |
e02119d5 | 2717 | |
10e958d5 NB |
2718 | unaccount_log_buffer(fs_info, |
2719 | path->nodes[*level]->start); | |
2720 | } | |
e02119d5 CM |
2721 | } |
2722 | free_extent_buffer(path->nodes[*level]); | |
2723 | path->nodes[*level] = NULL; | |
2724 | *level = i + 1; | |
2725 | } | |
2726 | } | |
2727 | return 1; | |
2728 | } | |
2729 | ||
2730 | /* | |
2731 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2732 | * the tree freeing any blocks that have a ref count of zero after being | |
2733 | * decremented. | |
2734 | */ | |
2735 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2736 | struct btrfs_root *log, struct walk_control *wc) | |
2737 | { | |
2ff7e61e | 2738 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2739 | int ret = 0; |
2740 | int wret; | |
2741 | int level; | |
2742 | struct btrfs_path *path; | |
e02119d5 CM |
2743 | int orig_level; |
2744 | ||
2745 | path = btrfs_alloc_path(); | |
db5b493a TI |
2746 | if (!path) |
2747 | return -ENOMEM; | |
e02119d5 CM |
2748 | |
2749 | level = btrfs_header_level(log->node); | |
2750 | orig_level = level; | |
2751 | path->nodes[level] = log->node; | |
67439dad | 2752 | atomic_inc(&log->node->refs); |
e02119d5 CM |
2753 | path->slots[level] = 0; |
2754 | ||
d397712b | 2755 | while (1) { |
e02119d5 CM |
2756 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2757 | if (wret > 0) | |
2758 | break; | |
79787eaa | 2759 | if (wret < 0) { |
e02119d5 | 2760 | ret = wret; |
79787eaa JM |
2761 | goto out; |
2762 | } | |
e02119d5 CM |
2763 | |
2764 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2765 | if (wret > 0) | |
2766 | break; | |
79787eaa | 2767 | if (wret < 0) { |
e02119d5 | 2768 | ret = wret; |
79787eaa JM |
2769 | goto out; |
2770 | } | |
e02119d5 CM |
2771 | } |
2772 | ||
2773 | /* was the root node processed? if not, catch it here */ | |
2774 | if (path->nodes[orig_level]) { | |
79787eaa | 2775 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
581c1760 QW |
2776 | btrfs_header_generation(path->nodes[orig_level]), |
2777 | orig_level); | |
79787eaa JM |
2778 | if (ret) |
2779 | goto out; | |
e02119d5 CM |
2780 | if (wc->free) { |
2781 | struct extent_buffer *next; | |
2782 | ||
2783 | next = path->nodes[orig_level]; | |
2784 | ||
681ae509 JB |
2785 | if (trans) { |
2786 | btrfs_tree_lock(next); | |
6a884d7d | 2787 | btrfs_clean_tree_block(next); |
681ae509 JB |
2788 | btrfs_wait_tree_block_writeback(next); |
2789 | btrfs_tree_unlock(next); | |
7bfc1007 | 2790 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2791 | next->start, next->len); |
2792 | if (ret) | |
2793 | goto out; | |
84c25448 | 2794 | btrfs_redirty_list_add(trans->transaction, next); |
1846430c LB |
2795 | } else { |
2796 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2797 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2798 | unaccount_log_buffer(fs_info, next->start); |
681ae509 | 2799 | } |
e02119d5 CM |
2800 | } |
2801 | } | |
2802 | ||
79787eaa | 2803 | out: |
e02119d5 | 2804 | btrfs_free_path(path); |
e02119d5 CM |
2805 | return ret; |
2806 | } | |
2807 | ||
7237f183 YZ |
2808 | /* |
2809 | * helper function to update the item for a given subvolumes log root | |
2810 | * in the tree of log roots | |
2811 | */ | |
2812 | static int update_log_root(struct btrfs_trans_handle *trans, | |
4203e968 JB |
2813 | struct btrfs_root *log, |
2814 | struct btrfs_root_item *root_item) | |
7237f183 | 2815 | { |
0b246afa | 2816 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
2817 | int ret; |
2818 | ||
2819 | if (log->log_transid == 1) { | |
2820 | /* insert root item on the first sync */ | |
0b246afa | 2821 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
4203e968 | 2822 | &log->root_key, root_item); |
7237f183 | 2823 | } else { |
0b246afa | 2824 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
4203e968 | 2825 | &log->root_key, root_item); |
7237f183 YZ |
2826 | } |
2827 | return ret; | |
2828 | } | |
2829 | ||
60d53eb3 | 2830 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2831 | { |
2832 | DEFINE_WAIT(wait); | |
7237f183 | 2833 | int index = transid % 2; |
e02119d5 | 2834 | |
7237f183 YZ |
2835 | /* |
2836 | * we only allow two pending log transactions at a time, | |
2837 | * so we know that if ours is more than 2 older than the | |
2838 | * current transaction, we're done | |
2839 | */ | |
49e83f57 | 2840 | for (;;) { |
7237f183 YZ |
2841 | prepare_to_wait(&root->log_commit_wait[index], |
2842 | &wait, TASK_UNINTERRUPTIBLE); | |
12fcfd22 | 2843 | |
49e83f57 LB |
2844 | if (!(root->log_transid_committed < transid && |
2845 | atomic_read(&root->log_commit[index]))) | |
2846 | break; | |
12fcfd22 | 2847 | |
49e83f57 LB |
2848 | mutex_unlock(&root->log_mutex); |
2849 | schedule(); | |
7237f183 | 2850 | mutex_lock(&root->log_mutex); |
49e83f57 LB |
2851 | } |
2852 | finish_wait(&root->log_commit_wait[index], &wait); | |
7237f183 YZ |
2853 | } |
2854 | ||
60d53eb3 | 2855 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
2856 | { |
2857 | DEFINE_WAIT(wait); | |
8b050d35 | 2858 | |
49e83f57 LB |
2859 | for (;;) { |
2860 | prepare_to_wait(&root->log_writer_wait, &wait, | |
2861 | TASK_UNINTERRUPTIBLE); | |
2862 | if (!atomic_read(&root->log_writers)) | |
2863 | break; | |
2864 | ||
7237f183 | 2865 | mutex_unlock(&root->log_mutex); |
49e83f57 | 2866 | schedule(); |
575849ec | 2867 | mutex_lock(&root->log_mutex); |
7237f183 | 2868 | } |
49e83f57 | 2869 | finish_wait(&root->log_writer_wait, &wait); |
e02119d5 CM |
2870 | } |
2871 | ||
8b050d35 MX |
2872 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2873 | struct btrfs_log_ctx *ctx) | |
2874 | { | |
8b050d35 MX |
2875 | mutex_lock(&root->log_mutex); |
2876 | list_del_init(&ctx->list); | |
2877 | mutex_unlock(&root->log_mutex); | |
2878 | } | |
2879 | ||
2880 | /* | |
2881 | * Invoked in log mutex context, or be sure there is no other task which | |
2882 | * can access the list. | |
2883 | */ | |
2884 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2885 | int index, int error) | |
2886 | { | |
2887 | struct btrfs_log_ctx *ctx; | |
570dd450 | 2888 | struct btrfs_log_ctx *safe; |
8b050d35 | 2889 | |
570dd450 CM |
2890 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
2891 | list_del_init(&ctx->list); | |
8b050d35 | 2892 | ctx->log_ret = error; |
570dd450 | 2893 | } |
8b050d35 MX |
2894 | } |
2895 | ||
e02119d5 CM |
2896 | /* |
2897 | * btrfs_sync_log does sends a given tree log down to the disk and | |
2898 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
2899 | * you know that any inodes previously logged are safely on disk only |
2900 | * if it returns 0. | |
2901 | * | |
2902 | * Any other return value means you need to call btrfs_commit_transaction. | |
2903 | * Some of the edge cases for fsyncing directories that have had unlinks | |
2904 | * or renames done in the past mean that sometimes the only safe | |
2905 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
2906 | * that has happened. | |
e02119d5 CM |
2907 | */ |
2908 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 2909 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 2910 | { |
7237f183 YZ |
2911 | int index1; |
2912 | int index2; | |
8cef4e16 | 2913 | int mark; |
e02119d5 | 2914 | int ret; |
0b246afa | 2915 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2916 | struct btrfs_root *log = root->log_root; |
0b246afa | 2917 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
4203e968 | 2918 | struct btrfs_root_item new_root_item; |
bb14a59b | 2919 | int log_transid = 0; |
8b050d35 | 2920 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 2921 | struct blk_plug plug; |
47876f7c FM |
2922 | u64 log_root_start; |
2923 | u64 log_root_level; | |
e02119d5 | 2924 | |
7237f183 | 2925 | mutex_lock(&root->log_mutex); |
d1433deb MX |
2926 | log_transid = ctx->log_transid; |
2927 | if (root->log_transid_committed >= log_transid) { | |
2928 | mutex_unlock(&root->log_mutex); | |
2929 | return ctx->log_ret; | |
2930 | } | |
2931 | ||
2932 | index1 = log_transid % 2; | |
7237f183 | 2933 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 2934 | wait_log_commit(root, log_transid); |
7237f183 | 2935 | mutex_unlock(&root->log_mutex); |
8b050d35 | 2936 | return ctx->log_ret; |
e02119d5 | 2937 | } |
d1433deb | 2938 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
2939 | atomic_set(&root->log_commit[index1], 1); |
2940 | ||
2941 | /* wait for previous tree log sync to complete */ | |
2942 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 2943 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 2944 | |
86df7eb9 | 2945 | while (1) { |
2ecb7923 | 2946 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 2947 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 2948 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 2949 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
2950 | mutex_unlock(&root->log_mutex); |
2951 | schedule_timeout_uninterruptible(1); | |
2952 | mutex_lock(&root->log_mutex); | |
2953 | } | |
60d53eb3 | 2954 | wait_for_writer(root); |
2ecb7923 | 2955 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
2956 | break; |
2957 | } | |
e02119d5 | 2958 | |
12fcfd22 | 2959 | /* bail out if we need to do a full commit */ |
4884b8e8 | 2960 | if (btrfs_need_log_full_commit(trans)) { |
f31f09f6 | 2961 | ret = BTRFS_LOG_FORCE_COMMIT; |
12fcfd22 CM |
2962 | mutex_unlock(&root->log_mutex); |
2963 | goto out; | |
2964 | } | |
2965 | ||
8cef4e16 YZ |
2966 | if (log_transid % 2 == 0) |
2967 | mark = EXTENT_DIRTY; | |
2968 | else | |
2969 | mark = EXTENT_NEW; | |
2970 | ||
690587d1 CM |
2971 | /* we start IO on all the marked extents here, but we don't actually |
2972 | * wait for them until later. | |
2973 | */ | |
c6adc9cc | 2974 | blk_start_plug(&plug); |
2ff7e61e | 2975 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
b528f467 NA |
2976 | /* |
2977 | * -EAGAIN happens when someone, e.g., a concurrent transaction | |
2978 | * commit, writes a dirty extent in this tree-log commit. This | |
2979 | * concurrent write will create a hole writing out the extents, | |
2980 | * and we cannot proceed on a zoned filesystem, requiring | |
2981 | * sequential writing. While we can bail out to a full commit | |
2982 | * here, but we can continue hoping the concurrent writing fills | |
2983 | * the hole. | |
2984 | */ | |
2985 | if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) | |
2986 | ret = 0; | |
79787eaa | 2987 | if (ret) { |
c6adc9cc | 2988 | blk_finish_plug(&plug); |
66642832 | 2989 | btrfs_abort_transaction(trans, ret); |
90787766 | 2990 | btrfs_set_log_full_commit(trans); |
79787eaa JM |
2991 | mutex_unlock(&root->log_mutex); |
2992 | goto out; | |
2993 | } | |
7237f183 | 2994 | |
4203e968 JB |
2995 | /* |
2996 | * We _must_ update under the root->log_mutex in order to make sure we | |
2997 | * have a consistent view of the log root we are trying to commit at | |
2998 | * this moment. | |
2999 | * | |
3000 | * We _must_ copy this into a local copy, because we are not holding the | |
3001 | * log_root_tree->log_mutex yet. This is important because when we | |
3002 | * commit the log_root_tree we must have a consistent view of the | |
3003 | * log_root_tree when we update the super block to point at the | |
3004 | * log_root_tree bytenr. If we update the log_root_tree here we'll race | |
3005 | * with the commit and possibly point at the new block which we may not | |
3006 | * have written out. | |
3007 | */ | |
5d4f98a2 | 3008 | btrfs_set_root_node(&log->root_item, log->node); |
4203e968 | 3009 | memcpy(&new_root_item, &log->root_item, sizeof(new_root_item)); |
7237f183 | 3010 | |
7237f183 YZ |
3011 | root->log_transid++; |
3012 | log->log_transid = root->log_transid; | |
ff782e0a | 3013 | root->log_start_pid = 0; |
7237f183 | 3014 | /* |
8cef4e16 YZ |
3015 | * IO has been started, blocks of the log tree have WRITTEN flag set |
3016 | * in their headers. new modifications of the log will be written to | |
3017 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
3018 | */ |
3019 | mutex_unlock(&root->log_mutex); | |
3020 | ||
3ddebf27 | 3021 | if (btrfs_is_zoned(fs_info)) { |
e75f9fd1 | 3022 | mutex_lock(&fs_info->tree_root->log_mutex); |
3ddebf27 NA |
3023 | if (!log_root_tree->node) { |
3024 | ret = btrfs_alloc_log_tree_node(trans, log_root_tree); | |
3025 | if (ret) { | |
ea32af47 | 3026 | mutex_unlock(&fs_info->tree_root->log_mutex); |
50ff5788 | 3027 | blk_finish_plug(&plug); |
3ddebf27 NA |
3028 | goto out; |
3029 | } | |
3030 | } | |
e75f9fd1 | 3031 | mutex_unlock(&fs_info->tree_root->log_mutex); |
3ddebf27 NA |
3032 | } |
3033 | ||
e75f9fd1 NA |
3034 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
3035 | ||
3036 | mutex_lock(&log_root_tree->log_mutex); | |
3037 | ||
e3d3b415 FM |
3038 | index2 = log_root_tree->log_transid % 2; |
3039 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
3040 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
3041 | ||
4203e968 JB |
3042 | /* |
3043 | * Now we are safe to update the log_root_tree because we're under the | |
3044 | * log_mutex, and we're a current writer so we're holding the commit | |
3045 | * open until we drop the log_mutex. | |
3046 | */ | |
3047 | ret = update_log_root(trans, log, &new_root_item); | |
4a500fd1 | 3048 | if (ret) { |
d1433deb MX |
3049 | if (!list_empty(&root_log_ctx.list)) |
3050 | list_del_init(&root_log_ctx.list); | |
3051 | ||
c6adc9cc | 3052 | blk_finish_plug(&plug); |
90787766 | 3053 | btrfs_set_log_full_commit(trans); |
995946dd | 3054 | |
79787eaa | 3055 | if (ret != -ENOSPC) { |
66642832 | 3056 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3057 | mutex_unlock(&log_root_tree->log_mutex); |
3058 | goto out; | |
3059 | } | |
bf89d38f | 3060 | btrfs_wait_tree_log_extents(log, mark); |
4a500fd1 | 3061 | mutex_unlock(&log_root_tree->log_mutex); |
f31f09f6 | 3062 | ret = BTRFS_LOG_FORCE_COMMIT; |
4a500fd1 YZ |
3063 | goto out; |
3064 | } | |
3065 | ||
d1433deb | 3066 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 3067 | blk_finish_plug(&plug); |
cbd60aa7 | 3068 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
3069 | mutex_unlock(&log_root_tree->log_mutex); |
3070 | ret = root_log_ctx.log_ret; | |
3071 | goto out; | |
3072 | } | |
8b050d35 | 3073 | |
d1433deb | 3074 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 3075 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 3076 | blk_finish_plug(&plug); |
bf89d38f | 3077 | ret = btrfs_wait_tree_log_extents(log, mark); |
60d53eb3 | 3078 | wait_log_commit(log_root_tree, |
d1433deb | 3079 | root_log_ctx.log_transid); |
7237f183 | 3080 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
3081 | if (!ret) |
3082 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
3083 | goto out; |
3084 | } | |
d1433deb | 3085 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
3086 | atomic_set(&log_root_tree->log_commit[index2], 1); |
3087 | ||
12fcfd22 | 3088 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 3089 | wait_log_commit(log_root_tree, |
d1433deb | 3090 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
3091 | } |
3092 | ||
12fcfd22 CM |
3093 | /* |
3094 | * now that we've moved on to the tree of log tree roots, | |
3095 | * check the full commit flag again | |
3096 | */ | |
4884b8e8 | 3097 | if (btrfs_need_log_full_commit(trans)) { |
c6adc9cc | 3098 | blk_finish_plug(&plug); |
bf89d38f | 3099 | btrfs_wait_tree_log_extents(log, mark); |
12fcfd22 | 3100 | mutex_unlock(&log_root_tree->log_mutex); |
f31f09f6 | 3101 | ret = BTRFS_LOG_FORCE_COMMIT; |
12fcfd22 CM |
3102 | goto out_wake_log_root; |
3103 | } | |
7237f183 | 3104 | |
2ff7e61e | 3105 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
3106 | &log_root_tree->dirty_log_pages, |
3107 | EXTENT_DIRTY | EXTENT_NEW); | |
3108 | blk_finish_plug(&plug); | |
b528f467 NA |
3109 | /* |
3110 | * As described above, -EAGAIN indicates a hole in the extents. We | |
3111 | * cannot wait for these write outs since the waiting cause a | |
3112 | * deadlock. Bail out to the full commit instead. | |
3113 | */ | |
3114 | if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) { | |
3115 | btrfs_set_log_full_commit(trans); | |
3116 | btrfs_wait_tree_log_extents(log, mark); | |
3117 | mutex_unlock(&log_root_tree->log_mutex); | |
3118 | goto out_wake_log_root; | |
3119 | } else if (ret) { | |
90787766 | 3120 | btrfs_set_log_full_commit(trans); |
66642832 | 3121 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3122 | mutex_unlock(&log_root_tree->log_mutex); |
3123 | goto out_wake_log_root; | |
3124 | } | |
bf89d38f | 3125 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 3126 | if (!ret) |
bf89d38f JM |
3127 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
3128 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 3129 | if (ret) { |
90787766 | 3130 | btrfs_set_log_full_commit(trans); |
5ab5e44a FM |
3131 | mutex_unlock(&log_root_tree->log_mutex); |
3132 | goto out_wake_log_root; | |
3133 | } | |
e02119d5 | 3134 | |
47876f7c FM |
3135 | log_root_start = log_root_tree->node->start; |
3136 | log_root_level = btrfs_header_level(log_root_tree->node); | |
7237f183 | 3137 | log_root_tree->log_transid++; |
7237f183 YZ |
3138 | mutex_unlock(&log_root_tree->log_mutex); |
3139 | ||
3140 | /* | |
47876f7c FM |
3141 | * Here we are guaranteed that nobody is going to write the superblock |
3142 | * for the current transaction before us and that neither we do write | |
3143 | * our superblock before the previous transaction finishes its commit | |
3144 | * and writes its superblock, because: | |
3145 | * | |
3146 | * 1) We are holding a handle on the current transaction, so no body | |
3147 | * can commit it until we release the handle; | |
3148 | * | |
3149 | * 2) Before writing our superblock we acquire the tree_log_mutex, so | |
3150 | * if the previous transaction is still committing, and hasn't yet | |
3151 | * written its superblock, we wait for it to do it, because a | |
3152 | * transaction commit acquires the tree_log_mutex when the commit | |
3153 | * begins and releases it only after writing its superblock. | |
7237f183 | 3154 | */ |
47876f7c | 3155 | mutex_lock(&fs_info->tree_log_mutex); |
165ea85f JB |
3156 | |
3157 | /* | |
3158 | * The previous transaction writeout phase could have failed, and thus | |
3159 | * marked the fs in an error state. We must not commit here, as we | |
3160 | * could have updated our generation in the super_for_commit and | |
3161 | * writing the super here would result in transid mismatches. If there | |
3162 | * is an error here just bail. | |
3163 | */ | |
84961539 | 3164 | if (BTRFS_FS_ERROR(fs_info)) { |
165ea85f JB |
3165 | ret = -EIO; |
3166 | btrfs_set_log_full_commit(trans); | |
3167 | btrfs_abort_transaction(trans, ret); | |
3168 | mutex_unlock(&fs_info->tree_log_mutex); | |
3169 | goto out_wake_log_root; | |
3170 | } | |
3171 | ||
47876f7c FM |
3172 | btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start); |
3173 | btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level); | |
eece6a9c | 3174 | ret = write_all_supers(fs_info, 1); |
47876f7c | 3175 | mutex_unlock(&fs_info->tree_log_mutex); |
5af3e8cc | 3176 | if (ret) { |
90787766 | 3177 | btrfs_set_log_full_commit(trans); |
66642832 | 3178 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
3179 | goto out_wake_log_root; |
3180 | } | |
7237f183 | 3181 | |
e1a6d264 FM |
3182 | /* |
3183 | * We know there can only be one task here, since we have not yet set | |
3184 | * root->log_commit[index1] to 0 and any task attempting to sync the | |
3185 | * log must wait for the previous log transaction to commit if it's | |
3186 | * still in progress or wait for the current log transaction commit if | |
3187 | * someone else already started it. We use <= and not < because the | |
3188 | * first log transaction has an ID of 0. | |
3189 | */ | |
3190 | ASSERT(root->last_log_commit <= log_transid); | |
3191 | root->last_log_commit = log_transid; | |
257c62e1 | 3192 | |
12fcfd22 | 3193 | out_wake_log_root: |
570dd450 | 3194 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
3195 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
3196 | ||
d1433deb | 3197 | log_root_tree->log_transid_committed++; |
7237f183 | 3198 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
3199 | mutex_unlock(&log_root_tree->log_mutex); |
3200 | ||
33a9eca7 | 3201 | /* |
093258e6 DS |
3202 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3203 | * all the updates above are seen by the woken threads. It might not be | |
3204 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3205 | */ |
093258e6 | 3206 | cond_wake_up(&log_root_tree->log_commit_wait[index2]); |
e02119d5 | 3207 | out: |
d1433deb | 3208 | mutex_lock(&root->log_mutex); |
570dd450 | 3209 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 3210 | root->log_transid_committed++; |
7237f183 | 3211 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 3212 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3213 | |
33a9eca7 | 3214 | /* |
093258e6 DS |
3215 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3216 | * all the updates above are seen by the woken threads. It might not be | |
3217 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3218 | */ |
093258e6 | 3219 | cond_wake_up(&root->log_commit_wait[index1]); |
b31eabd8 | 3220 | return ret; |
e02119d5 CM |
3221 | } |
3222 | ||
4a500fd1 YZ |
3223 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3224 | struct btrfs_root *log) | |
e02119d5 CM |
3225 | { |
3226 | int ret; | |
e02119d5 CM |
3227 | struct walk_control wc = { |
3228 | .free = 1, | |
3229 | .process_func = process_one_buffer | |
3230 | }; | |
3231 | ||
3ddebf27 NA |
3232 | if (log->node) { |
3233 | ret = walk_log_tree(trans, log, &wc); | |
3234 | if (ret) { | |
40cdc509 FM |
3235 | /* |
3236 | * We weren't able to traverse the entire log tree, the | |
3237 | * typical scenario is getting an -EIO when reading an | |
3238 | * extent buffer of the tree, due to a previous writeback | |
3239 | * failure of it. | |
3240 | */ | |
3241 | set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR, | |
3242 | &log->fs_info->fs_state); | |
3243 | ||
3244 | /* | |
3245 | * Some extent buffers of the log tree may still be dirty | |
3246 | * and not yet written back to storage, because we may | |
3247 | * have updates to a log tree without syncing a log tree, | |
3248 | * such as during rename and link operations. So flush | |
3249 | * them out and wait for their writeback to complete, so | |
3250 | * that we properly cleanup their state and pages. | |
3251 | */ | |
3252 | btrfs_write_marked_extents(log->fs_info, | |
3253 | &log->dirty_log_pages, | |
3254 | EXTENT_DIRTY | EXTENT_NEW); | |
3255 | btrfs_wait_tree_log_extents(log, | |
3256 | EXTENT_DIRTY | EXTENT_NEW); | |
3257 | ||
3ddebf27 NA |
3258 | if (trans) |
3259 | btrfs_abort_transaction(trans, ret); | |
3260 | else | |
3261 | btrfs_handle_fs_error(log->fs_info, ret, NULL); | |
3262 | } | |
374b0e2d | 3263 | } |
e02119d5 | 3264 | |
59b0713a FM |
3265 | clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1, |
3266 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT); | |
e289f03e | 3267 | extent_io_tree_release(&log->log_csum_range); |
d3575156 | 3268 | |
00246528 | 3269 | btrfs_put_root(log); |
4a500fd1 YZ |
3270 | } |
3271 | ||
3272 | /* | |
3273 | * free all the extents used by the tree log. This should be called | |
3274 | * at commit time of the full transaction | |
3275 | */ | |
3276 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3277 | { | |
3278 | if (root->log_root) { | |
3279 | free_log_tree(trans, root->log_root); | |
3280 | root->log_root = NULL; | |
e7a79811 | 3281 | clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state); |
4a500fd1 YZ |
3282 | } |
3283 | return 0; | |
3284 | } | |
3285 | ||
3286 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3287 | struct btrfs_fs_info *fs_info) | |
3288 | { | |
3289 | if (fs_info->log_root_tree) { | |
3290 | free_log_tree(trans, fs_info->log_root_tree); | |
3291 | fs_info->log_root_tree = NULL; | |
47876f7c | 3292 | clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state); |
4a500fd1 | 3293 | } |
e02119d5 CM |
3294 | return 0; |
3295 | } | |
3296 | ||
803f0f64 | 3297 | /* |
0f8ce498 FM |
3298 | * Check if an inode was logged in the current transaction. This correctly deals |
3299 | * with the case where the inode was logged but has a logged_trans of 0, which | |
3300 | * happens if the inode is evicted and loaded again, as logged_trans is an in | |
3301 | * memory only field (not persisted). | |
3302 | * | |
3303 | * Returns 1 if the inode was logged before in the transaction, 0 if it was not, | |
3304 | * and < 0 on error. | |
803f0f64 | 3305 | */ |
0f8ce498 FM |
3306 | static int inode_logged(struct btrfs_trans_handle *trans, |
3307 | struct btrfs_inode *inode, | |
3308 | struct btrfs_path *path_in) | |
803f0f64 | 3309 | { |
0f8ce498 FM |
3310 | struct btrfs_path *path = path_in; |
3311 | struct btrfs_key key; | |
3312 | int ret; | |
3313 | ||
803f0f64 | 3314 | if (inode->logged_trans == trans->transid) |
0f8ce498 | 3315 | return 1; |
803f0f64 | 3316 | |
0f8ce498 FM |
3317 | /* |
3318 | * If logged_trans is not 0, then we know the inode logged was not logged | |
3319 | * in this transaction, so we can return false right away. | |
3320 | */ | |
3321 | if (inode->logged_trans > 0) | |
3322 | return 0; | |
3323 | ||
3324 | /* | |
3325 | * If no log tree was created for this root in this transaction, then | |
3326 | * the inode can not have been logged in this transaction. In that case | |
3327 | * set logged_trans to anything greater than 0 and less than the current | |
3328 | * transaction's ID, to avoid the search below in a future call in case | |
3329 | * a log tree gets created after this. | |
3330 | */ | |
3331 | if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &inode->root->state)) { | |
3332 | inode->logged_trans = trans->transid - 1; | |
3333 | return 0; | |
3334 | } | |
3335 | ||
3336 | /* | |
3337 | * We have a log tree and the inode's logged_trans is 0. We can't tell | |
3338 | * for sure if the inode was logged before in this transaction by looking | |
3339 | * only at logged_trans. We could be pessimistic and assume it was, but | |
3340 | * that can lead to unnecessarily logging an inode during rename and link | |
3341 | * operations, and then further updating the log in followup rename and | |
3342 | * link operations, specially if it's a directory, which adds latency | |
3343 | * visible to applications doing a series of rename or link operations. | |
3344 | * | |
3345 | * A logged_trans of 0 here can mean several things: | |
3346 | * | |
3347 | * 1) The inode was never logged since the filesystem was mounted, and may | |
3348 | * or may have not been evicted and loaded again; | |
3349 | * | |
3350 | * 2) The inode was logged in a previous transaction, then evicted and | |
3351 | * then loaded again; | |
3352 | * | |
3353 | * 3) The inode was logged in the current transaction, then evicted and | |
3354 | * then loaded again. | |
3355 | * | |
3356 | * For cases 1) and 2) we don't want to return true, but we need to detect | |
3357 | * case 3) and return true. So we do a search in the log root for the inode | |
3358 | * item. | |
3359 | */ | |
3360 | key.objectid = btrfs_ino(inode); | |
3361 | key.type = BTRFS_INODE_ITEM_KEY; | |
3362 | key.offset = 0; | |
3363 | ||
3364 | if (!path) { | |
3365 | path = btrfs_alloc_path(); | |
3366 | if (!path) | |
3367 | return -ENOMEM; | |
3368 | } | |
3369 | ||
3370 | ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0); | |
3371 | ||
3372 | if (path_in) | |
3373 | btrfs_release_path(path); | |
3374 | else | |
3375 | btrfs_free_path(path); | |
1e0860f3 | 3376 | |
6e8e777d | 3377 | /* |
0f8ce498 FM |
3378 | * Logging an inode always results in logging its inode item. So if we |
3379 | * did not find the item we know the inode was not logged for sure. | |
6e8e777d | 3380 | */ |
0f8ce498 FM |
3381 | if (ret < 0) { |
3382 | return ret; | |
3383 | } else if (ret > 0) { | |
3384 | /* | |
3385 | * Set logged_trans to a value greater than 0 and less then the | |
3386 | * current transaction to avoid doing the search in future calls. | |
3387 | */ | |
3388 | inode->logged_trans = trans->transid - 1; | |
3389 | return 0; | |
3390 | } | |
3391 | ||
3392 | /* | |
3393 | * The inode was previously logged and then evicted, set logged_trans to | |
3394 | * the current transacion's ID, to avoid future tree searches as long as | |
3395 | * the inode is not evicted again. | |
3396 | */ | |
3397 | inode->logged_trans = trans->transid; | |
3398 | ||
3399 | /* | |
3400 | * If it's a directory, then we must set last_dir_index_offset to the | |
3401 | * maximum possible value, so that the next attempt to log the inode does | |
3402 | * not skip checking if dir index keys found in modified subvolume tree | |
3403 | * leaves have been logged before, otherwise it would result in attempts | |
3404 | * to insert duplicate dir index keys in the log tree. This must be done | |
3405 | * because last_dir_index_offset is an in-memory only field, not persisted | |
3406 | * in the inode item or any other on-disk structure, so its value is lost | |
3407 | * once the inode is evicted. | |
3408 | */ | |
3409 | if (S_ISDIR(inode->vfs_inode.i_mode)) | |
3410 | inode->last_dir_index_offset = (u64)-1; | |
803f0f64 | 3411 | |
0f8ce498 | 3412 | return 1; |
803f0f64 FM |
3413 | } |
3414 | ||
839061fe FM |
3415 | /* |
3416 | * Delete a directory entry from the log if it exists. | |
3417 | * | |
3418 | * Returns < 0 on error | |
3419 | * 1 if the entry does not exists | |
3420 | * 0 if the entry existed and was successfully deleted | |
3421 | */ | |
3422 | static int del_logged_dentry(struct btrfs_trans_handle *trans, | |
3423 | struct btrfs_root *log, | |
3424 | struct btrfs_path *path, | |
3425 | u64 dir_ino, | |
6db75318 | 3426 | const struct fscrypt_str *name, |
839061fe FM |
3427 | u64 index) |
3428 | { | |
3429 | struct btrfs_dir_item *di; | |
3430 | ||
3431 | /* | |
3432 | * We only log dir index items of a directory, so we don't need to look | |
3433 | * for dir item keys. | |
3434 | */ | |
3435 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, | |
e43eec81 | 3436 | index, name, -1); |
839061fe FM |
3437 | if (IS_ERR(di)) |
3438 | return PTR_ERR(di); | |
3439 | else if (!di) | |
3440 | return 1; | |
3441 | ||
3442 | /* | |
3443 | * We do not need to update the size field of the directory's | |
3444 | * inode item because on log replay we update the field to reflect | |
3445 | * all existing entries in the directory (see overwrite_item()). | |
3446 | */ | |
3447 | return btrfs_delete_one_dir_name(trans, log, path, di); | |
3448 | } | |
3449 | ||
e02119d5 CM |
3450 | /* |
3451 | * If both a file and directory are logged, and unlinks or renames are | |
3452 | * mixed in, we have a few interesting corners: | |
3453 | * | |
3454 | * create file X in dir Y | |
3455 | * link file X to X.link in dir Y | |
3456 | * fsync file X | |
3457 | * unlink file X but leave X.link | |
3458 | * fsync dir Y | |
3459 | * | |
3460 | * After a crash we would expect only X.link to exist. But file X | |
3461 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3462 | * | |
3463 | * We solve this by removing directory entries and inode backrefs from the | |
3464 | * log when a file that was logged in the current transaction is | |
3465 | * unlinked. Any later fsync will include the updated log entries, and | |
3466 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3467 | * | |
3468 | * This optimizations allows us to avoid relogging the entire inode | |
3469 | * or the entire directory. | |
3470 | */ | |
9a35fc95 JB |
3471 | void btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, |
3472 | struct btrfs_root *root, | |
6db75318 | 3473 | const struct fscrypt_str *name, |
9a35fc95 | 3474 | struct btrfs_inode *dir, u64 index) |
e02119d5 | 3475 | { |
e02119d5 CM |
3476 | struct btrfs_path *path; |
3477 | int ret; | |
e02119d5 | 3478 | |
0f8ce498 FM |
3479 | ret = inode_logged(trans, dir, NULL); |
3480 | if (ret == 0) | |
3481 | return; | |
3482 | else if (ret < 0) { | |
3483 | btrfs_set_log_full_commit(trans); | |
9a35fc95 | 3484 | return; |
0f8ce498 | 3485 | } |
3a5f1d45 | 3486 | |
e02119d5 CM |
3487 | ret = join_running_log_trans(root); |
3488 | if (ret) | |
9a35fc95 | 3489 | return; |
e02119d5 | 3490 | |
49f34d1f | 3491 | mutex_lock(&dir->log_mutex); |
e02119d5 | 3492 | |
e02119d5 | 3493 | path = btrfs_alloc_path(); |
a62f44a5 | 3494 | if (!path) { |
839061fe | 3495 | ret = -ENOMEM; |
a62f44a5 TI |
3496 | goto out_unlock; |
3497 | } | |
2a29edc6 | 3498 | |
839061fe | 3499 | ret = del_logged_dentry(trans, root->log_root, path, btrfs_ino(dir), |
e43eec81 | 3500 | name, index); |
e02119d5 | 3501 | btrfs_free_path(path); |
a62f44a5 | 3502 | out_unlock: |
49f34d1f | 3503 | mutex_unlock(&dir->log_mutex); |
839061fe | 3504 | if (ret < 0) |
90787766 | 3505 | btrfs_set_log_full_commit(trans); |
12fcfd22 | 3506 | btrfs_end_log_trans(root); |
e02119d5 CM |
3507 | } |
3508 | ||
3509 | /* see comments for btrfs_del_dir_entries_in_log */ | |
9a35fc95 JB |
3510 | void btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, |
3511 | struct btrfs_root *root, | |
6db75318 | 3512 | const struct fscrypt_str *name, |
9a35fc95 | 3513 | struct btrfs_inode *inode, u64 dirid) |
e02119d5 CM |
3514 | { |
3515 | struct btrfs_root *log; | |
3516 | u64 index; | |
3517 | int ret; | |
3518 | ||
0f8ce498 FM |
3519 | ret = inode_logged(trans, inode, NULL); |
3520 | if (ret == 0) | |
9a35fc95 | 3521 | return; |
0f8ce498 FM |
3522 | else if (ret < 0) { |
3523 | btrfs_set_log_full_commit(trans); | |
3524 | return; | |
3525 | } | |
3a5f1d45 | 3526 | |
e02119d5 CM |
3527 | ret = join_running_log_trans(root); |
3528 | if (ret) | |
9a35fc95 | 3529 | return; |
e02119d5 | 3530 | log = root->log_root; |
a491abb2 | 3531 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3532 | |
e43eec81 | 3533 | ret = btrfs_del_inode_ref(trans, log, name, btrfs_ino(inode), |
e02119d5 | 3534 | dirid, &index); |
a491abb2 | 3535 | mutex_unlock(&inode->log_mutex); |
9a35fc95 | 3536 | if (ret < 0 && ret != -ENOENT) |
90787766 | 3537 | btrfs_set_log_full_commit(trans); |
12fcfd22 | 3538 | btrfs_end_log_trans(root); |
e02119d5 CM |
3539 | } |
3540 | ||
3541 | /* | |
3542 | * creates a range item in the log for 'dirid'. first_offset and | |
3543 | * last_offset tell us which parts of the key space the log should | |
3544 | * be considered authoritative for. | |
3545 | */ | |
3546 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3547 | struct btrfs_root *log, | |
3548 | struct btrfs_path *path, | |
339d0354 | 3549 | u64 dirid, |
e02119d5 CM |
3550 | u64 first_offset, u64 last_offset) |
3551 | { | |
3552 | int ret; | |
3553 | struct btrfs_key key; | |
3554 | struct btrfs_dir_log_item *item; | |
3555 | ||
3556 | key.objectid = dirid; | |
3557 | key.offset = first_offset; | |
339d0354 | 3558 | key.type = BTRFS_DIR_LOG_INDEX_KEY; |
e02119d5 | 3559 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); |
750ee454 FM |
3560 | /* |
3561 | * -EEXIST is fine and can happen sporadically when we are logging a | |
3562 | * directory and have concurrent insertions in the subvolume's tree for | |
3563 | * items from other inodes and that result in pushing off some dir items | |
3564 | * from one leaf to another in order to accommodate for the new items. | |
3565 | * This results in logging the same dir index range key. | |
3566 | */ | |
3567 | if (ret && ret != -EEXIST) | |
4a500fd1 | 3568 | return ret; |
e02119d5 CM |
3569 | |
3570 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3571 | struct btrfs_dir_log_item); | |
750ee454 FM |
3572 | if (ret == -EEXIST) { |
3573 | const u64 curr_end = btrfs_dir_log_end(path->nodes[0], item); | |
3574 | ||
3575 | /* | |
3576 | * btrfs_del_dir_entries_in_log() might have been called during | |
3577 | * an unlink between the initial insertion of this key and the | |
3578 | * current update, or we might be logging a single entry deletion | |
3579 | * during a rename, so set the new last_offset to the max value. | |
3580 | */ | |
3581 | last_offset = max(last_offset, curr_end); | |
3582 | } | |
e02119d5 CM |
3583 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); |
3584 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3585 | btrfs_release_path(path); |
e02119d5 CM |
3586 | return 0; |
3587 | } | |
3588 | ||
086dcbfa FM |
3589 | static int flush_dir_items_batch(struct btrfs_trans_handle *trans, |
3590 | struct btrfs_root *log, | |
3591 | struct extent_buffer *src, | |
3592 | struct btrfs_path *dst_path, | |
3593 | int start_slot, | |
3594 | int count) | |
3595 | { | |
3596 | char *ins_data = NULL; | |
b7ef5f3a | 3597 | struct btrfs_item_batch batch; |
086dcbfa | 3598 | struct extent_buffer *dst; |
da1b811f FM |
3599 | unsigned long src_offset; |
3600 | unsigned long dst_offset; | |
086dcbfa FM |
3601 | struct btrfs_key key; |
3602 | u32 item_size; | |
3603 | int ret; | |
3604 | int i; | |
3605 | ||
3606 | ASSERT(count > 0); | |
b7ef5f3a | 3607 | batch.nr = count; |
086dcbfa FM |
3608 | |
3609 | if (count == 1) { | |
3610 | btrfs_item_key_to_cpu(src, &key, start_slot); | |
3212fa14 | 3611 | item_size = btrfs_item_size(src, start_slot); |
b7ef5f3a FM |
3612 | batch.keys = &key; |
3613 | batch.data_sizes = &item_size; | |
3614 | batch.total_data_size = item_size; | |
086dcbfa | 3615 | } else { |
b7ef5f3a FM |
3616 | struct btrfs_key *ins_keys; |
3617 | u32 *ins_sizes; | |
3618 | ||
086dcbfa FM |
3619 | ins_data = kmalloc(count * sizeof(u32) + |
3620 | count * sizeof(struct btrfs_key), GFP_NOFS); | |
3621 | if (!ins_data) | |
3622 | return -ENOMEM; | |
3623 | ||
3624 | ins_sizes = (u32 *)ins_data; | |
3625 | ins_keys = (struct btrfs_key *)(ins_data + count * sizeof(u32)); | |
b7ef5f3a FM |
3626 | batch.keys = ins_keys; |
3627 | batch.data_sizes = ins_sizes; | |
3628 | batch.total_data_size = 0; | |
086dcbfa FM |
3629 | |
3630 | for (i = 0; i < count; i++) { | |
3631 | const int slot = start_slot + i; | |
3632 | ||
3633 | btrfs_item_key_to_cpu(src, &ins_keys[i], slot); | |
3212fa14 | 3634 | ins_sizes[i] = btrfs_item_size(src, slot); |
b7ef5f3a | 3635 | batch.total_data_size += ins_sizes[i]; |
086dcbfa FM |
3636 | } |
3637 | } | |
3638 | ||
b7ef5f3a | 3639 | ret = btrfs_insert_empty_items(trans, log, dst_path, &batch); |
086dcbfa FM |
3640 | if (ret) |
3641 | goto out; | |
3642 | ||
3643 | dst = dst_path->nodes[0]; | |
da1b811f FM |
3644 | /* |
3645 | * Copy all the items in bulk, in a single copy operation. Item data is | |
3646 | * organized such that it's placed at the end of a leaf and from right | |
3647 | * to left. For example, the data for the second item ends at an offset | |
3648 | * that matches the offset where the data for the first item starts, the | |
3649 | * data for the third item ends at an offset that matches the offset | |
3650 | * where the data of the second items starts, and so on. | |
3651 | * Therefore our source and destination start offsets for copy match the | |
3652 | * offsets of the last items (highest slots). | |
3653 | */ | |
3654 | dst_offset = btrfs_item_ptr_offset(dst, dst_path->slots[0] + count - 1); | |
3655 | src_offset = btrfs_item_ptr_offset(src, start_slot + count - 1); | |
3656 | copy_extent_buffer(dst, src, dst_offset, src_offset, batch.total_data_size); | |
086dcbfa FM |
3657 | btrfs_release_path(dst_path); |
3658 | out: | |
3659 | kfree(ins_data); | |
3660 | ||
3661 | return ret; | |
3662 | } | |
3663 | ||
eb10d85e FM |
3664 | static int process_dir_items_leaf(struct btrfs_trans_handle *trans, |
3665 | struct btrfs_inode *inode, | |
3666 | struct btrfs_path *path, | |
3667 | struct btrfs_path *dst_path, | |
732d591a FM |
3668 | struct btrfs_log_ctx *ctx, |
3669 | u64 *last_old_dentry_offset) | |
eb10d85e FM |
3670 | { |
3671 | struct btrfs_root *log = inode->root->log_root; | |
796787c9 FM |
3672 | struct extent_buffer *src; |
3673 | const int nritems = btrfs_header_nritems(path->nodes[0]); | |
eb10d85e | 3674 | const u64 ino = btrfs_ino(inode); |
086dcbfa FM |
3675 | bool last_found = false; |
3676 | int batch_start = 0; | |
3677 | int batch_size = 0; | |
eb10d85e FM |
3678 | int i; |
3679 | ||
796787c9 FM |
3680 | /* |
3681 | * We need to clone the leaf, release the read lock on it, and use the | |
3682 | * clone before modifying the log tree. See the comment at copy_items() | |
3683 | * about why we need to do this. | |
3684 | */ | |
3685 | src = btrfs_clone_extent_buffer(path->nodes[0]); | |
3686 | if (!src) | |
3687 | return -ENOMEM; | |
3688 | ||
3689 | i = path->slots[0]; | |
3690 | btrfs_release_path(path); | |
3691 | path->nodes[0] = src; | |
3692 | path->slots[0] = i; | |
3693 | ||
3694 | for (; i < nritems; i++) { | |
732d591a | 3695 | struct btrfs_dir_item *di; |
eb10d85e | 3696 | struct btrfs_key key; |
eb10d85e FM |
3697 | int ret; |
3698 | ||
3699 | btrfs_item_key_to_cpu(src, &key, i); | |
3700 | ||
339d0354 | 3701 | if (key.objectid != ino || key.type != BTRFS_DIR_INDEX_KEY) { |
086dcbfa FM |
3702 | last_found = true; |
3703 | break; | |
3704 | } | |
eb10d85e | 3705 | |
732d591a | 3706 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); |
dc287224 | 3707 | ctx->last_dir_item_offset = key.offset; |
732d591a FM |
3708 | |
3709 | /* | |
3710 | * Skip ranges of items that consist only of dir item keys created | |
3711 | * in past transactions. However if we find a gap, we must log a | |
3712 | * dir index range item for that gap, so that index keys in that | |
3713 | * gap are deleted during log replay. | |
3714 | */ | |
3715 | if (btrfs_dir_transid(src, di) < trans->transid) { | |
3716 | if (key.offset > *last_old_dentry_offset + 1) { | |
3717 | ret = insert_dir_log_key(trans, log, dst_path, | |
3718 | ino, *last_old_dentry_offset + 1, | |
3719 | key.offset - 1); | |
732d591a FM |
3720 | if (ret < 0) |
3721 | return ret; | |
3722 | } | |
3723 | ||
3724 | *last_old_dentry_offset = key.offset; | |
3725 | continue; | |
3726 | } | |
193df624 FM |
3727 | |
3728 | /* If we logged this dir index item before, we can skip it. */ | |
3729 | if (key.offset <= inode->last_dir_index_offset) | |
3730 | continue; | |
3731 | ||
eb10d85e FM |
3732 | /* |
3733 | * We must make sure that when we log a directory entry, the | |
3734 | * corresponding inode, after log replay, has a matching link | |
3735 | * count. For example: | |
3736 | * | |
3737 | * touch foo | |
3738 | * mkdir mydir | |
3739 | * sync | |
3740 | * ln foo mydir/bar | |
3741 | * xfs_io -c "fsync" mydir | |
3742 | * <crash> | |
3743 | * <mount fs and log replay> | |
3744 | * | |
3745 | * Would result in a fsync log that when replayed, our file inode | |
3746 | * would have a link count of 1, but we get two directory entries | |
3747 | * pointing to the same inode. After removing one of the names, | |
3748 | * it would not be possible to remove the other name, which | |
3749 | * resulted always in stale file handle errors, and would not be | |
3750 | * possible to rmdir the parent directory, since its i_size could | |
3751 | * never be decremented to the value BTRFS_EMPTY_DIR_SIZE, | |
3752 | * resulting in -ENOTEMPTY errors. | |
3753 | */ | |
086dcbfa | 3754 | if (!ctx->log_new_dentries) { |
086dcbfa FM |
3755 | struct btrfs_key di_key; |
3756 | ||
086dcbfa | 3757 | btrfs_dir_item_key_to_cpu(src, di, &di_key); |
732d591a | 3758 | if (di_key.type != BTRFS_ROOT_ITEM_KEY) |
086dcbfa FM |
3759 | ctx->log_new_dentries = true; |
3760 | } | |
3761 | ||
086dcbfa FM |
3762 | if (batch_size == 0) |
3763 | batch_start = i; | |
3764 | batch_size++; | |
eb10d85e FM |
3765 | } |
3766 | ||
086dcbfa FM |
3767 | if (batch_size > 0) { |
3768 | int ret; | |
3769 | ||
3770 | ret = flush_dir_items_batch(trans, log, src, dst_path, | |
3771 | batch_start, batch_size); | |
3772 | if (ret < 0) | |
3773 | return ret; | |
3774 | } | |
3775 | ||
3776 | return last_found ? 1 : 0; | |
eb10d85e FM |
3777 | } |
3778 | ||
e02119d5 CM |
3779 | /* |
3780 | * log all the items included in the current transaction for a given | |
3781 | * directory. This also creates the range items in the log tree required | |
3782 | * to replay anything deleted before the fsync | |
3783 | */ | |
3784 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
90d04510 | 3785 | struct btrfs_inode *inode, |
e02119d5 | 3786 | struct btrfs_path *path, |
339d0354 | 3787 | struct btrfs_path *dst_path, |
2f2ff0ee | 3788 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3789 | u64 min_offset, u64 *last_offset_ret) |
3790 | { | |
3791 | struct btrfs_key min_key; | |
90d04510 | 3792 | struct btrfs_root *root = inode->root; |
e02119d5 | 3793 | struct btrfs_root *log = root->log_root; |
4a500fd1 | 3794 | int err = 0; |
e02119d5 | 3795 | int ret; |
732d591a | 3796 | u64 last_old_dentry_offset = min_offset - 1; |
e02119d5 | 3797 | u64 last_offset = (u64)-1; |
684a5773 | 3798 | u64 ino = btrfs_ino(inode); |
e02119d5 | 3799 | |
33345d01 | 3800 | min_key.objectid = ino; |
339d0354 | 3801 | min_key.type = BTRFS_DIR_INDEX_KEY; |
e02119d5 CM |
3802 | min_key.offset = min_offset; |
3803 | ||
6174d3cb | 3804 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3805 | |
3806 | /* | |
3807 | * we didn't find anything from this transaction, see if there | |
3808 | * is anything at all | |
3809 | */ | |
339d0354 FM |
3810 | if (ret != 0 || min_key.objectid != ino || |
3811 | min_key.type != BTRFS_DIR_INDEX_KEY) { | |
33345d01 | 3812 | min_key.objectid = ino; |
339d0354 | 3813 | min_key.type = BTRFS_DIR_INDEX_KEY; |
e02119d5 | 3814 | min_key.offset = (u64)-1; |
b3b4aa74 | 3815 | btrfs_release_path(path); |
e02119d5 CM |
3816 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3817 | if (ret < 0) { | |
b3b4aa74 | 3818 | btrfs_release_path(path); |
e02119d5 CM |
3819 | return ret; |
3820 | } | |
339d0354 | 3821 | ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY); |
e02119d5 CM |
3822 | |
3823 | /* if ret == 0 there are items for this type, | |
3824 | * create a range to tell us the last key of this type. | |
3825 | * otherwise, there are no items in this directory after | |
3826 | * *min_offset, and we create a range to indicate that. | |
3827 | */ | |
3828 | if (ret == 0) { | |
3829 | struct btrfs_key tmp; | |
732d591a | 3830 | |
e02119d5 CM |
3831 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, |
3832 | path->slots[0]); | |
339d0354 | 3833 | if (tmp.type == BTRFS_DIR_INDEX_KEY) |
732d591a | 3834 | last_old_dentry_offset = tmp.offset; |
e02119d5 CM |
3835 | } |
3836 | goto done; | |
3837 | } | |
3838 | ||
3839 | /* go backward to find any previous key */ | |
339d0354 | 3840 | ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY); |
e02119d5 CM |
3841 | if (ret == 0) { |
3842 | struct btrfs_key tmp; | |
a450a4af | 3843 | |
e02119d5 | 3844 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); |
a450a4af FM |
3845 | /* |
3846 | * The dir index key before the first one we found that needs to | |
3847 | * be logged might be in a previous leaf, and there might be a | |
3848 | * gap between these keys, meaning that we had deletions that | |
3849 | * happened. So the key range item we log (key type | |
3850 | * BTRFS_DIR_LOG_INDEX_KEY) must cover a range that starts at the | |
3851 | * previous key's offset plus 1, so that those deletes are replayed. | |
3852 | */ | |
3853 | if (tmp.type == BTRFS_DIR_INDEX_KEY) | |
732d591a | 3854 | last_old_dentry_offset = tmp.offset; |
e02119d5 | 3855 | } |
b3b4aa74 | 3856 | btrfs_release_path(path); |
e02119d5 | 3857 | |
2cc83342 JB |
3858 | /* |
3859 | * Find the first key from this transaction again. See the note for | |
3860 | * log_new_dir_dentries, if we're logging a directory recursively we | |
3861 | * won't be holding its i_mutex, which means we can modify the directory | |
3862 | * while we're logging it. If we remove an entry between our first | |
3863 | * search and this search we'll not find the key again and can just | |
3864 | * bail. | |
3865 | */ | |
bb56f02f | 3866 | search: |
e02119d5 | 3867 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
2cc83342 | 3868 | if (ret != 0) |
e02119d5 | 3869 | goto done; |
e02119d5 CM |
3870 | |
3871 | /* | |
3872 | * we have a block from this transaction, log every item in it | |
3873 | * from our directory | |
3874 | */ | |
d397712b | 3875 | while (1) { |
732d591a FM |
3876 | ret = process_dir_items_leaf(trans, inode, path, dst_path, ctx, |
3877 | &last_old_dentry_offset); | |
eb10d85e FM |
3878 | if (ret != 0) { |
3879 | if (ret < 0) | |
4a500fd1 | 3880 | err = ret; |
eb10d85e | 3881 | goto done; |
e02119d5 | 3882 | } |
eb10d85e | 3883 | path->slots[0] = btrfs_header_nritems(path->nodes[0]); |
e02119d5 CM |
3884 | |
3885 | /* | |
3886 | * look ahead to the next item and see if it is also | |
3887 | * from this directory and from this transaction | |
3888 | */ | |
3889 | ret = btrfs_next_leaf(root, path); | |
80c0b421 LB |
3890 | if (ret) { |
3891 | if (ret == 1) | |
3892 | last_offset = (u64)-1; | |
3893 | else | |
3894 | err = ret; | |
e02119d5 CM |
3895 | goto done; |
3896 | } | |
eb10d85e | 3897 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, path->slots[0]); |
339d0354 | 3898 | if (min_key.objectid != ino || min_key.type != BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
3899 | last_offset = (u64)-1; |
3900 | goto done; | |
3901 | } | |
3902 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
a450a4af FM |
3903 | /* |
3904 | * The next leaf was not changed in the current transaction | |
3905 | * and has at least one dir index key. | |
3906 | * We check for the next key because there might have been | |
3907 | * one or more deletions between the last key we logged and | |
3908 | * that next key. So the key range item we log (key type | |
3909 | * BTRFS_DIR_LOG_INDEX_KEY) must end at the next key's | |
3910 | * offset minus 1, so that those deletes are replayed. | |
3911 | */ | |
3912 | last_offset = min_key.offset - 1; | |
e02119d5 CM |
3913 | goto done; |
3914 | } | |
eb10d85e FM |
3915 | if (need_resched()) { |
3916 | btrfs_release_path(path); | |
3917 | cond_resched(); | |
3918 | goto search; | |
3919 | } | |
e02119d5 CM |
3920 | } |
3921 | done: | |
b3b4aa74 DS |
3922 | btrfs_release_path(path); |
3923 | btrfs_release_path(dst_path); | |
e02119d5 | 3924 | |
4a500fd1 YZ |
3925 | if (err == 0) { |
3926 | *last_offset_ret = last_offset; | |
3927 | /* | |
732d591a FM |
3928 | * In case the leaf was changed in the current transaction but |
3929 | * all its dir items are from a past transaction, the last item | |
3930 | * in the leaf is a dir item and there's no gap between that last | |
3931 | * dir item and the first one on the next leaf (which did not | |
3932 | * change in the current transaction), then we don't need to log | |
3933 | * a range, last_old_dentry_offset is == to last_offset. | |
4a500fd1 | 3934 | */ |
732d591a FM |
3935 | ASSERT(last_old_dentry_offset <= last_offset); |
3936 | if (last_old_dentry_offset < last_offset) { | |
3937 | ret = insert_dir_log_key(trans, log, path, ino, | |
3938 | last_old_dentry_offset + 1, | |
3939 | last_offset); | |
3940 | if (ret) | |
3941 | err = ret; | |
3942 | } | |
4a500fd1 YZ |
3943 | } |
3944 | return err; | |
e02119d5 CM |
3945 | } |
3946 | ||
193df624 FM |
3947 | /* |
3948 | * If the inode was logged before and it was evicted, then its | |
3949 | * last_dir_index_offset is (u64)-1, so we don't the value of the last index | |
3950 | * key offset. If that's the case, search for it and update the inode. This | |
3951 | * is to avoid lookups in the log tree every time we try to insert a dir index | |
3952 | * key from a leaf changed in the current transaction, and to allow us to always | |
3953 | * do batch insertions of dir index keys. | |
3954 | */ | |
3955 | static int update_last_dir_index_offset(struct btrfs_inode *inode, | |
3956 | struct btrfs_path *path, | |
3957 | const struct btrfs_log_ctx *ctx) | |
3958 | { | |
3959 | const u64 ino = btrfs_ino(inode); | |
3960 | struct btrfs_key key; | |
3961 | int ret; | |
3962 | ||
3963 | lockdep_assert_held(&inode->log_mutex); | |
3964 | ||
3965 | if (inode->last_dir_index_offset != (u64)-1) | |
3966 | return 0; | |
3967 | ||
3968 | if (!ctx->logged_before) { | |
3969 | inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1; | |
3970 | return 0; | |
3971 | } | |
3972 | ||
3973 | key.objectid = ino; | |
3974 | key.type = BTRFS_DIR_INDEX_KEY; | |
3975 | key.offset = (u64)-1; | |
3976 | ||
3977 | ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0); | |
3978 | /* | |
3979 | * An error happened or we actually have an index key with an offset | |
3980 | * value of (u64)-1. Bail out, we're done. | |
3981 | */ | |
3982 | if (ret <= 0) | |
3983 | goto out; | |
3984 | ||
3985 | ret = 0; | |
3986 | inode->last_dir_index_offset = BTRFS_DIR_START_INDEX - 1; | |
3987 | ||
3988 | /* | |
3989 | * No dir index items, bail out and leave last_dir_index_offset with | |
3990 | * the value right before the first valid index value. | |
3991 | */ | |
3992 | if (path->slots[0] == 0) | |
3993 | goto out; | |
3994 | ||
3995 | /* | |
3996 | * btrfs_search_slot() left us at one slot beyond the slot with the last | |
3997 | * index key, or beyond the last key of the directory that is not an | |
3998 | * index key. If we have an index key before, set last_dir_index_offset | |
3999 | * to its offset value, otherwise leave it with a value right before the | |
4000 | * first valid index value, as it means we have an empty directory. | |
4001 | */ | |
4002 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
4003 | if (key.objectid == ino && key.type == BTRFS_DIR_INDEX_KEY) | |
4004 | inode->last_dir_index_offset = key.offset; | |
4005 | ||
4006 | out: | |
4007 | btrfs_release_path(path); | |
4008 | ||
4009 | return ret; | |
4010 | } | |
4011 | ||
e02119d5 CM |
4012 | /* |
4013 | * logging directories is very similar to logging inodes, We find all the items | |
4014 | * from the current transaction and write them to the log. | |
4015 | * | |
4016 | * The recovery code scans the directory in the subvolume, and if it finds a | |
4017 | * key in the range logged that is not present in the log tree, then it means | |
4018 | * that dir entry was unlinked during the transaction. | |
4019 | * | |
4020 | * In order for that scan to work, we must include one key smaller than | |
4021 | * the smallest logged by this transaction and one key larger than the largest | |
4022 | * key logged by this transaction. | |
4023 | */ | |
4024 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
90d04510 | 4025 | struct btrfs_inode *inode, |
e02119d5 | 4026 | struct btrfs_path *path, |
2f2ff0ee FM |
4027 | struct btrfs_path *dst_path, |
4028 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
4029 | { |
4030 | u64 min_key; | |
4031 | u64 max_key; | |
4032 | int ret; | |
e02119d5 | 4033 | |
193df624 FM |
4034 | ret = update_last_dir_index_offset(inode, path, ctx); |
4035 | if (ret) | |
4036 | return ret; | |
4037 | ||
732d591a | 4038 | min_key = BTRFS_DIR_START_INDEX; |
e02119d5 | 4039 | max_key = 0; |
339d0354 | 4040 | ctx->last_dir_item_offset = inode->last_dir_index_offset; |
dc287224 | 4041 | |
d397712b | 4042 | while (1) { |
339d0354 | 4043 | ret = log_dir_items(trans, inode, path, dst_path, |
dbf39ea4 | 4044 | ctx, min_key, &max_key); |
4a500fd1 YZ |
4045 | if (ret) |
4046 | return ret; | |
e02119d5 CM |
4047 | if (max_key == (u64)-1) |
4048 | break; | |
4049 | min_key = max_key + 1; | |
4050 | } | |
4051 | ||
339d0354 FM |
4052 | inode->last_dir_index_offset = ctx->last_dir_item_offset; |
4053 | ||
e02119d5 CM |
4054 | return 0; |
4055 | } | |
4056 | ||
4057 | /* | |
4058 | * a helper function to drop items from the log before we relog an | |
4059 | * inode. max_key_type indicates the highest item type to remove. | |
4060 | * This cannot be run for file data extents because it does not | |
4061 | * free the extents they point to. | |
4062 | */ | |
88e221cd | 4063 | static int drop_inode_items(struct btrfs_trans_handle *trans, |
e02119d5 CM |
4064 | struct btrfs_root *log, |
4065 | struct btrfs_path *path, | |
88e221cd FM |
4066 | struct btrfs_inode *inode, |
4067 | int max_key_type) | |
e02119d5 CM |
4068 | { |
4069 | int ret; | |
4070 | struct btrfs_key key; | |
4071 | struct btrfs_key found_key; | |
18ec90d6 | 4072 | int start_slot; |
e02119d5 | 4073 | |
88e221cd | 4074 | key.objectid = btrfs_ino(inode); |
e02119d5 CM |
4075 | key.type = max_key_type; |
4076 | key.offset = (u64)-1; | |
4077 | ||
d397712b | 4078 | while (1) { |
e02119d5 | 4079 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 4080 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 4081 | if (ret < 0) |
e02119d5 CM |
4082 | break; |
4083 | ||
4084 | if (path->slots[0] == 0) | |
4085 | break; | |
4086 | ||
4087 | path->slots[0]--; | |
4088 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
4089 | path->slots[0]); | |
4090 | ||
88e221cd | 4091 | if (found_key.objectid != key.objectid) |
e02119d5 CM |
4092 | break; |
4093 | ||
18ec90d6 JB |
4094 | found_key.offset = 0; |
4095 | found_key.type = 0; | |
e3b83361 | 4096 | ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot); |
cbca7d59 FM |
4097 | if (ret < 0) |
4098 | break; | |
18ec90d6 JB |
4099 | |
4100 | ret = btrfs_del_items(trans, log, path, start_slot, | |
4101 | path->slots[0] - start_slot + 1); | |
4102 | /* | |
4103 | * If start slot isn't 0 then we don't need to re-search, we've | |
4104 | * found the last guy with the objectid in this tree. | |
4105 | */ | |
4106 | if (ret || start_slot != 0) | |
65a246c5 | 4107 | break; |
b3b4aa74 | 4108 | btrfs_release_path(path); |
e02119d5 | 4109 | } |
b3b4aa74 | 4110 | btrfs_release_path(path); |
5bdbeb21 JB |
4111 | if (ret > 0) |
4112 | ret = 0; | |
4a500fd1 | 4113 | return ret; |
e02119d5 CM |
4114 | } |
4115 | ||
8a2b3da1 FM |
4116 | static int truncate_inode_items(struct btrfs_trans_handle *trans, |
4117 | struct btrfs_root *log_root, | |
4118 | struct btrfs_inode *inode, | |
4119 | u64 new_size, u32 min_type) | |
4120 | { | |
d9ac19c3 JB |
4121 | struct btrfs_truncate_control control = { |
4122 | .new_size = new_size, | |
487e81d2 | 4123 | .ino = btrfs_ino(inode), |
d9ac19c3 | 4124 | .min_type = min_type, |
5caa490e | 4125 | .skip_ref_updates = true, |
d9ac19c3 | 4126 | }; |
8a2b3da1 | 4127 | |
8697b8f8 | 4128 | return btrfs_truncate_inode_items(trans, log_root, &control); |
8a2b3da1 FM |
4129 | } |
4130 | ||
94edf4ae JB |
4131 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
4132 | struct extent_buffer *leaf, | |
4133 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
4134 | struct inode *inode, int log_inode_only, |
4135 | u64 logged_isize) | |
94edf4ae | 4136 | { |
0b1c6cca | 4137 | struct btrfs_map_token token; |
77eea05e | 4138 | u64 flags; |
0b1c6cca | 4139 | |
c82f823c | 4140 | btrfs_init_map_token(&token, leaf); |
94edf4ae JB |
4141 | |
4142 | if (log_inode_only) { | |
4143 | /* set the generation to zero so the recover code | |
4144 | * can tell the difference between an logging | |
4145 | * just to say 'this inode exists' and a logging | |
4146 | * to say 'update this inode with these values' | |
4147 | */ | |
cc4c13d5 DS |
4148 | btrfs_set_token_inode_generation(&token, item, 0); |
4149 | btrfs_set_token_inode_size(&token, item, logged_isize); | |
94edf4ae | 4150 | } else { |
cc4c13d5 DS |
4151 | btrfs_set_token_inode_generation(&token, item, |
4152 | BTRFS_I(inode)->generation); | |
4153 | btrfs_set_token_inode_size(&token, item, inode->i_size); | |
0b1c6cca JB |
4154 | } |
4155 | ||
cc4c13d5 DS |
4156 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
4157 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
4158 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
4159 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
4160 | ||
4161 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
4162 | inode->i_atime.tv_sec); | |
4163 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
4164 | inode->i_atime.tv_nsec); | |
4165 | ||
4166 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
4167 | inode->i_mtime.tv_sec); | |
4168 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
4169 | inode->i_mtime.tv_nsec); | |
4170 | ||
4171 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
4172 | inode->i_ctime.tv_sec); | |
4173 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
4174 | inode->i_ctime.tv_nsec); | |
4175 | ||
e593e54e FM |
4176 | /* |
4177 | * We do not need to set the nbytes field, in fact during a fast fsync | |
4178 | * its value may not even be correct, since a fast fsync does not wait | |
4179 | * for ordered extent completion, which is where we update nbytes, it | |
4180 | * only waits for writeback to complete. During log replay as we find | |
4181 | * file extent items and replay them, we adjust the nbytes field of the | |
4182 | * inode item in subvolume tree as needed (see overwrite_item()). | |
4183 | */ | |
cc4c13d5 DS |
4184 | |
4185 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
4186 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
4187 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
77eea05e BB |
4188 | flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags, |
4189 | BTRFS_I(inode)->ro_flags); | |
4190 | btrfs_set_token_inode_flags(&token, item, flags); | |
cc4c13d5 | 4191 | btrfs_set_token_inode_block_group(&token, item, 0); |
94edf4ae JB |
4192 | } |
4193 | ||
a95249b3 JB |
4194 | static int log_inode_item(struct btrfs_trans_handle *trans, |
4195 | struct btrfs_root *log, struct btrfs_path *path, | |
2ac691d8 | 4196 | struct btrfs_inode *inode, bool inode_item_dropped) |
a95249b3 JB |
4197 | { |
4198 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
4199 | int ret; |
4200 | ||
2ac691d8 FM |
4201 | /* |
4202 | * If we are doing a fast fsync and the inode was logged before in the | |
4203 | * current transaction, then we know the inode was previously logged and | |
4204 | * it exists in the log tree. For performance reasons, in this case use | |
4205 | * btrfs_search_slot() directly with ins_len set to 0 so that we never | |
4206 | * attempt a write lock on the leaf's parent, which adds unnecessary lock | |
4207 | * contention in case there are concurrent fsyncs for other inodes of the | |
4208 | * same subvolume. Using btrfs_insert_empty_item() when the inode item | |
4209 | * already exists can also result in unnecessarily splitting a leaf. | |
4210 | */ | |
4211 | if (!inode_item_dropped && inode->logged_trans == trans->transid) { | |
4212 | ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1); | |
4213 | ASSERT(ret <= 0); | |
4214 | if (ret > 0) | |
4215 | ret = -ENOENT; | |
4216 | } else { | |
4217 | /* | |
4218 | * This means it is the first fsync in the current transaction, | |
4219 | * so the inode item is not in the log and we need to insert it. | |
4220 | * We can never get -EEXIST because we are only called for a fast | |
4221 | * fsync and in case an inode eviction happens after the inode was | |
4222 | * logged before in the current transaction, when we load again | |
4223 | * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime | |
4224 | * flags and set ->logged_trans to 0. | |
4225 | */ | |
4226 | ret = btrfs_insert_empty_item(trans, log, path, &inode->location, | |
4227 | sizeof(*inode_item)); | |
4228 | ASSERT(ret != -EEXIST); | |
4229 | } | |
4230 | if (ret) | |
a95249b3 JB |
4231 | return ret; |
4232 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4233 | struct btrfs_inode_item); | |
6d889a3b NB |
4234 | fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode, |
4235 | 0, 0); | |
a95249b3 JB |
4236 | btrfs_release_path(path); |
4237 | return 0; | |
4238 | } | |
4239 | ||
40e046ac | 4240 | static int log_csums(struct btrfs_trans_handle *trans, |
3ebac17c | 4241 | struct btrfs_inode *inode, |
40e046ac FM |
4242 | struct btrfs_root *log_root, |
4243 | struct btrfs_ordered_sum *sums) | |
4244 | { | |
e289f03e FM |
4245 | const u64 lock_end = sums->bytenr + sums->len - 1; |
4246 | struct extent_state *cached_state = NULL; | |
40e046ac FM |
4247 | int ret; |
4248 | ||
3ebac17c FM |
4249 | /* |
4250 | * If this inode was not used for reflink operations in the current | |
4251 | * transaction with new extents, then do the fast path, no need to | |
4252 | * worry about logging checksum items with overlapping ranges. | |
4253 | */ | |
4254 | if (inode->last_reflink_trans < trans->transid) | |
4255 | return btrfs_csum_file_blocks(trans, log_root, sums); | |
4256 | ||
e289f03e FM |
4257 | /* |
4258 | * Serialize logging for checksums. This is to avoid racing with the | |
4259 | * same checksum being logged by another task that is logging another | |
4260 | * file which happens to refer to the same extent as well. Such races | |
4261 | * can leave checksum items in the log with overlapping ranges. | |
4262 | */ | |
570eb97b JB |
4263 | ret = lock_extent(&log_root->log_csum_range, sums->bytenr, lock_end, |
4264 | &cached_state); | |
e289f03e FM |
4265 | if (ret) |
4266 | return ret; | |
40e046ac FM |
4267 | /* |
4268 | * Due to extent cloning, we might have logged a csum item that covers a | |
4269 | * subrange of a cloned extent, and later we can end up logging a csum | |
4270 | * item for a larger subrange of the same extent or the entire range. | |
4271 | * This would leave csum items in the log tree that cover the same range | |
4272 | * and break the searches for checksums in the log tree, resulting in | |
4273 | * some checksums missing in the fs/subvolume tree. So just delete (or | |
4274 | * trim and adjust) any existing csum items in the log for this range. | |
4275 | */ | |
4276 | ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len); | |
e289f03e FM |
4277 | if (!ret) |
4278 | ret = btrfs_csum_file_blocks(trans, log_root, sums); | |
40e046ac | 4279 | |
570eb97b JB |
4280 | unlock_extent(&log_root->log_csum_range, sums->bytenr, lock_end, |
4281 | &cached_state); | |
e289f03e FM |
4282 | |
4283 | return ret; | |
40e046ac FM |
4284 | } |
4285 | ||
31ff1cd2 | 4286 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 4287 | struct btrfs_inode *inode, |
31ff1cd2 | 4288 | struct btrfs_path *dst_path, |
0e56315c | 4289 | struct btrfs_path *src_path, |
1a4bcf47 FM |
4290 | int start_slot, int nr, int inode_only, |
4291 | u64 logged_isize) | |
31ff1cd2 | 4292 | { |
44d70e19 | 4293 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 | 4294 | struct btrfs_file_extent_item *extent; |
796787c9 | 4295 | struct extent_buffer *src; |
7f30c072 | 4296 | int ret = 0; |
31ff1cd2 CM |
4297 | struct btrfs_key *ins_keys; |
4298 | u32 *ins_sizes; | |
b7ef5f3a | 4299 | struct btrfs_item_batch batch; |
31ff1cd2 CM |
4300 | char *ins_data; |
4301 | int i; | |
7f30c072 | 4302 | int dst_index; |
7f30c072 FM |
4303 | const bool skip_csum = (inode->flags & BTRFS_INODE_NODATASUM); |
4304 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
d20f7043 | 4305 | |
796787c9 FM |
4306 | /* |
4307 | * To keep lockdep happy and avoid deadlocks, clone the source leaf and | |
4308 | * use the clone. This is because otherwise we would be changing the log | |
4309 | * tree, to insert items from the subvolume tree or insert csum items, | |
4310 | * while holding a read lock on a leaf from the subvolume tree, which | |
4311 | * creates a nasty lock dependency when COWing log tree nodes/leaves: | |
4312 | * | |
4313 | * 1) Modifying the log tree triggers an extent buffer allocation while | |
4314 | * holding a write lock on a parent extent buffer from the log tree. | |
4315 | * Allocating the pages for an extent buffer, or the extent buffer | |
4316 | * struct, can trigger inode eviction and finally the inode eviction | |
4317 | * will trigger a release/remove of a delayed node, which requires | |
4318 | * taking the delayed node's mutex; | |
4319 | * | |
4320 | * 2) Allocating a metadata extent for a log tree can trigger the async | |
4321 | * reclaim thread and make us wait for it to release enough space and | |
4322 | * unblock our reservation ticket. The reclaim thread can start | |
4323 | * flushing delayed items, and that in turn results in the need to | |
4324 | * lock delayed node mutexes and in the need to write lock extent | |
4325 | * buffers of a subvolume tree - all this while holding a write lock | |
4326 | * on the parent extent buffer in the log tree. | |
4327 | * | |
4328 | * So one task in scenario 1) running in parallel with another task in | |
4329 | * scenario 2) could lead to a deadlock, one wanting to lock a delayed | |
4330 | * node mutex while having a read lock on a leaf from the subvolume, | |
4331 | * while the other is holding the delayed node's mutex and wants to | |
4332 | * write lock the same subvolume leaf for flushing delayed items. | |
4333 | */ | |
4334 | src = btrfs_clone_extent_buffer(src_path->nodes[0]); | |
4335 | if (!src) | |
4336 | return -ENOMEM; | |
4337 | ||
4338 | i = src_path->slots[0]; | |
4339 | btrfs_release_path(src_path); | |
4340 | src_path->nodes[0] = src; | |
4341 | src_path->slots[0] = i; | |
4342 | ||
31ff1cd2 CM |
4343 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + |
4344 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 4345 | if (!ins_data) |
4346 | return -ENOMEM; | |
4347 | ||
31ff1cd2 CM |
4348 | ins_sizes = (u32 *)ins_data; |
4349 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
b7ef5f3a FM |
4350 | batch.keys = ins_keys; |
4351 | batch.data_sizes = ins_sizes; | |
4352 | batch.total_data_size = 0; | |
7f30c072 | 4353 | batch.nr = 0; |
31ff1cd2 | 4354 | |
7f30c072 | 4355 | dst_index = 0; |
31ff1cd2 | 4356 | for (i = 0; i < nr; i++) { |
7f30c072 FM |
4357 | const int src_slot = start_slot + i; |
4358 | struct btrfs_root *csum_root; | |
5b7ce5e2 FM |
4359 | struct btrfs_ordered_sum *sums; |
4360 | struct btrfs_ordered_sum *sums_next; | |
4361 | LIST_HEAD(ordered_sums); | |
7f30c072 FM |
4362 | u64 disk_bytenr; |
4363 | u64 disk_num_bytes; | |
4364 | u64 extent_offset; | |
4365 | u64 extent_num_bytes; | |
4366 | bool is_old_extent; | |
4367 | ||
4368 | btrfs_item_key_to_cpu(src, &ins_keys[dst_index], src_slot); | |
4369 | ||
4370 | if (ins_keys[dst_index].type != BTRFS_EXTENT_DATA_KEY) | |
4371 | goto add_to_batch; | |
4372 | ||
4373 | extent = btrfs_item_ptr(src, src_slot, | |
4374 | struct btrfs_file_extent_item); | |
4375 | ||
4376 | is_old_extent = (btrfs_file_extent_generation(src, extent) < | |
4377 | trans->transid); | |
4378 | ||
4379 | /* | |
4380 | * Don't copy extents from past generations. That would make us | |
4381 | * log a lot more metadata for common cases like doing only a | |
4382 | * few random writes into a file and then fsync it for the first | |
4383 | * time or after the full sync flag is set on the inode. We can | |
4384 | * get leaves full of extent items, most of which are from past | |
4385 | * generations, so we can skip them - as long as the inode has | |
4386 | * not been the target of a reflink operation in this transaction, | |
4387 | * as in that case it might have had file extent items with old | |
4388 | * generations copied into it. We also must always log prealloc | |
4389 | * extents that start at or beyond eof, otherwise we would lose | |
4390 | * them on log replay. | |
4391 | */ | |
4392 | if (is_old_extent && | |
4393 | ins_keys[dst_index].offset < i_size && | |
4394 | inode->last_reflink_trans < trans->transid) | |
4395 | continue; | |
4396 | ||
4397 | if (skip_csum) | |
4398 | goto add_to_batch; | |
4399 | ||
4400 | /* Only regular extents have checksums. */ | |
4401 | if (btrfs_file_extent_type(src, extent) != BTRFS_FILE_EXTENT_REG) | |
4402 | goto add_to_batch; | |
4403 | ||
4404 | /* | |
4405 | * If it's an extent created in a past transaction, then its | |
4406 | * checksums are already accessible from the committed csum tree, | |
4407 | * no need to log them. | |
4408 | */ | |
4409 | if (is_old_extent) | |
4410 | goto add_to_batch; | |
4411 | ||
4412 | disk_bytenr = btrfs_file_extent_disk_bytenr(src, extent); | |
4413 | /* If it's an explicit hole, there are no checksums. */ | |
4414 | if (disk_bytenr == 0) | |
4415 | goto add_to_batch; | |
4416 | ||
4417 | disk_num_bytes = btrfs_file_extent_disk_num_bytes(src, extent); | |
4418 | ||
4419 | if (btrfs_file_extent_compression(src, extent)) { | |
4420 | extent_offset = 0; | |
4421 | extent_num_bytes = disk_num_bytes; | |
4422 | } else { | |
4423 | extent_offset = btrfs_file_extent_offset(src, extent); | |
4424 | extent_num_bytes = btrfs_file_extent_num_bytes(src, extent); | |
4425 | } | |
4426 | ||
4427 | csum_root = btrfs_csum_root(trans->fs_info, disk_bytenr); | |
4428 | disk_bytenr += extent_offset; | |
4429 | ret = btrfs_lookup_csums_range(csum_root, disk_bytenr, | |
4430 | disk_bytenr + extent_num_bytes - 1, | |
26ce9114 | 4431 | &ordered_sums, 0, false); |
7f30c072 FM |
4432 | if (ret) |
4433 | goto out; | |
4434 | ||
5b7ce5e2 FM |
4435 | list_for_each_entry_safe(sums, sums_next, &ordered_sums, list) { |
4436 | if (!ret) | |
4437 | ret = log_csums(trans, inode, log, sums); | |
4438 | list_del(&sums->list); | |
4439 | kfree(sums); | |
4440 | } | |
4441 | if (ret) | |
4442 | goto out; | |
4443 | ||
7f30c072 FM |
4444 | add_to_batch: |
4445 | ins_sizes[dst_index] = btrfs_item_size(src, src_slot); | |
4446 | batch.total_data_size += ins_sizes[dst_index]; | |
4447 | batch.nr++; | |
4448 | dst_index++; | |
31ff1cd2 | 4449 | } |
7f30c072 FM |
4450 | |
4451 | /* | |
4452 | * We have a leaf full of old extent items that don't need to be logged, | |
4453 | * so we don't need to do anything. | |
4454 | */ | |
4455 | if (batch.nr == 0) | |
4456 | goto out; | |
4457 | ||
b7ef5f3a | 4458 | ret = btrfs_insert_empty_items(trans, log, dst_path, &batch); |
7f30c072 FM |
4459 | if (ret) |
4460 | goto out; | |
4461 | ||
4462 | dst_index = 0; | |
4463 | for (i = 0; i < nr; i++) { | |
4464 | const int src_slot = start_slot + i; | |
4465 | const int dst_slot = dst_path->slots[0] + dst_index; | |
4466 | struct btrfs_key key; | |
4467 | unsigned long src_offset; | |
4468 | unsigned long dst_offset; | |
4469 | ||
4470 | /* | |
4471 | * We're done, all the remaining items in the source leaf | |
4472 | * correspond to old file extent items. | |
4473 | */ | |
4474 | if (dst_index >= batch.nr) | |
4475 | break; | |
4476 | ||
4477 | btrfs_item_key_to_cpu(src, &key, src_slot); | |
4478 | ||
4479 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
4480 | goto copy_item; | |
31ff1cd2 | 4481 | |
7f30c072 FM |
4482 | extent = btrfs_item_ptr(src, src_slot, |
4483 | struct btrfs_file_extent_item); | |
31ff1cd2 | 4484 | |
7f30c072 FM |
4485 | /* See the comment in the previous loop, same logic. */ |
4486 | if (btrfs_file_extent_generation(src, extent) < trans->transid && | |
4487 | key.offset < i_size && | |
4488 | inode->last_reflink_trans < trans->transid) | |
4489 | continue; | |
4490 | ||
4491 | copy_item: | |
4492 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], dst_slot); | |
4493 | src_offset = btrfs_item_ptr_offset(src, src_slot); | |
31ff1cd2 | 4494 | |
7f30c072 FM |
4495 | if (key.type == BTRFS_INODE_ITEM_KEY) { |
4496 | struct btrfs_inode_item *inode_item; | |
4497 | ||
4498 | inode_item = btrfs_item_ptr(dst_path->nodes[0], dst_slot, | |
31ff1cd2 | 4499 | struct btrfs_inode_item); |
94edf4ae | 4500 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
f85b7379 DS |
4501 | &inode->vfs_inode, |
4502 | inode_only == LOG_INODE_EXISTS, | |
1a4bcf47 | 4503 | logged_isize); |
94edf4ae JB |
4504 | } else { |
4505 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
7f30c072 | 4506 | src_offset, ins_sizes[dst_index]); |
31ff1cd2 | 4507 | } |
94edf4ae | 4508 | |
7f30c072 | 4509 | dst_index++; |
31ff1cd2 CM |
4510 | } |
4511 | ||
4512 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 4513 | btrfs_release_path(dst_path); |
7f30c072 | 4514 | out: |
31ff1cd2 | 4515 | kfree(ins_data); |
d20f7043 | 4516 | |
4a500fd1 | 4517 | return ret; |
31ff1cd2 CM |
4518 | } |
4519 | ||
4f0f586b ST |
4520 | static int extent_cmp(void *priv, const struct list_head *a, |
4521 | const struct list_head *b) | |
5dc562c5 | 4522 | { |
214cc184 | 4523 | const struct extent_map *em1, *em2; |
5dc562c5 JB |
4524 | |
4525 | em1 = list_entry(a, struct extent_map, list); | |
4526 | em2 = list_entry(b, struct extent_map, list); | |
4527 | ||
4528 | if (em1->start < em2->start) | |
4529 | return -1; | |
4530 | else if (em1->start > em2->start) | |
4531 | return 1; | |
4532 | return 0; | |
4533 | } | |
4534 | ||
e7175a69 JB |
4535 | static int log_extent_csums(struct btrfs_trans_handle *trans, |
4536 | struct btrfs_inode *inode, | |
a9ecb653 | 4537 | struct btrfs_root *log_root, |
48778179 FM |
4538 | const struct extent_map *em, |
4539 | struct btrfs_log_ctx *ctx) | |
5dc562c5 | 4540 | { |
48778179 | 4541 | struct btrfs_ordered_extent *ordered; |
fc28b25e | 4542 | struct btrfs_root *csum_root; |
2ab28f32 JB |
4543 | u64 csum_offset; |
4544 | u64 csum_len; | |
48778179 FM |
4545 | u64 mod_start = em->mod_start; |
4546 | u64 mod_len = em->mod_len; | |
8407f553 FM |
4547 | LIST_HEAD(ordered_sums); |
4548 | int ret = 0; | |
0aa4a17d | 4549 | |
e7175a69 JB |
4550 | if (inode->flags & BTRFS_INODE_NODATASUM || |
4551 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
8407f553 | 4552 | em->block_start == EXTENT_MAP_HOLE) |
70c8a91c | 4553 | return 0; |
5dc562c5 | 4554 | |
48778179 FM |
4555 | list_for_each_entry(ordered, &ctx->ordered_extents, log_list) { |
4556 | const u64 ordered_end = ordered->file_offset + ordered->num_bytes; | |
4557 | const u64 mod_end = mod_start + mod_len; | |
4558 | struct btrfs_ordered_sum *sums; | |
4559 | ||
4560 | if (mod_len == 0) | |
4561 | break; | |
4562 | ||
4563 | if (ordered_end <= mod_start) | |
4564 | continue; | |
4565 | if (mod_end <= ordered->file_offset) | |
4566 | break; | |
4567 | ||
4568 | /* | |
4569 | * We are going to copy all the csums on this ordered extent, so | |
4570 | * go ahead and adjust mod_start and mod_len in case this ordered | |
4571 | * extent has already been logged. | |
4572 | */ | |
4573 | if (ordered->file_offset > mod_start) { | |
4574 | if (ordered_end >= mod_end) | |
4575 | mod_len = ordered->file_offset - mod_start; | |
4576 | /* | |
4577 | * If we have this case | |
4578 | * | |
4579 | * |--------- logged extent ---------| | |
4580 | * |----- ordered extent ----| | |
4581 | * | |
4582 | * Just don't mess with mod_start and mod_len, we'll | |
4583 | * just end up logging more csums than we need and it | |
4584 | * will be ok. | |
4585 | */ | |
4586 | } else { | |
4587 | if (ordered_end < mod_end) { | |
4588 | mod_len = mod_end - ordered_end; | |
4589 | mod_start = ordered_end; | |
4590 | } else { | |
4591 | mod_len = 0; | |
4592 | } | |
4593 | } | |
4594 | ||
4595 | /* | |
4596 | * To keep us from looping for the above case of an ordered | |
4597 | * extent that falls inside of the logged extent. | |
4598 | */ | |
4599 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags)) | |
4600 | continue; | |
4601 | ||
4602 | list_for_each_entry(sums, &ordered->list, list) { | |
4603 | ret = log_csums(trans, inode, log_root, sums); | |
4604 | if (ret) | |
4605 | return ret; | |
4606 | } | |
4607 | } | |
4608 | ||
4609 | /* We're done, found all csums in the ordered extents. */ | |
4610 | if (mod_len == 0) | |
4611 | return 0; | |
4612 | ||
e7175a69 | 4613 | /* If we're compressed we have to save the entire range of csums. */ |
488111aa FDBM |
4614 | if (em->compress_type) { |
4615 | csum_offset = 0; | |
8407f553 | 4616 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa | 4617 | } else { |
48778179 FM |
4618 | csum_offset = mod_start - em->start; |
4619 | csum_len = mod_len; | |
488111aa | 4620 | } |
2ab28f32 | 4621 | |
70c8a91c | 4622 | /* block start is already adjusted for the file extent offset. */ |
fc28b25e JB |
4623 | csum_root = btrfs_csum_root(trans->fs_info, em->block_start); |
4624 | ret = btrfs_lookup_csums_range(csum_root, | |
70c8a91c JB |
4625 | em->block_start + csum_offset, |
4626 | em->block_start + csum_offset + | |
26ce9114 | 4627 | csum_len - 1, &ordered_sums, 0, false); |
70c8a91c JB |
4628 | if (ret) |
4629 | return ret; | |
5dc562c5 | 4630 | |
70c8a91c JB |
4631 | while (!list_empty(&ordered_sums)) { |
4632 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4633 | struct btrfs_ordered_sum, | |
4634 | list); | |
4635 | if (!ret) | |
3ebac17c | 4636 | ret = log_csums(trans, inode, log_root, sums); |
70c8a91c JB |
4637 | list_del(&sums->list); |
4638 | kfree(sums); | |
5dc562c5 JB |
4639 | } |
4640 | ||
70c8a91c | 4641 | return ret; |
5dc562c5 JB |
4642 | } |
4643 | ||
8407f553 | 4644 | static int log_one_extent(struct btrfs_trans_handle *trans, |
90d04510 | 4645 | struct btrfs_inode *inode, |
8407f553 FM |
4646 | const struct extent_map *em, |
4647 | struct btrfs_path *path, | |
8407f553 FM |
4648 | struct btrfs_log_ctx *ctx) |
4649 | { | |
5893dfb9 | 4650 | struct btrfs_drop_extents_args drop_args = { 0 }; |
90d04510 | 4651 | struct btrfs_root *log = inode->root->log_root; |
e1f53ed8 | 4652 | struct btrfs_file_extent_item fi = { 0 }; |
8407f553 | 4653 | struct extent_buffer *leaf; |
8407f553 FM |
4654 | struct btrfs_key key; |
4655 | u64 extent_offset = em->start - em->orig_start; | |
4656 | u64 block_len; | |
4657 | int ret; | |
8407f553 | 4658 | |
e1f53ed8 FM |
4659 | btrfs_set_stack_file_extent_generation(&fi, trans->transid); |
4660 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4661 | btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_PREALLOC); | |
4662 | else | |
4663 | btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_REG); | |
4664 | ||
4665 | block_len = max(em->block_len, em->orig_block_len); | |
4666 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
4667 | btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start); | |
4668 | btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len); | |
4669 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
4670 | btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start - | |
4671 | extent_offset); | |
4672 | btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len); | |
4673 | } | |
4674 | ||
4675 | btrfs_set_stack_file_extent_offset(&fi, extent_offset); | |
4676 | btrfs_set_stack_file_extent_num_bytes(&fi, em->len); | |
4677 | btrfs_set_stack_file_extent_ram_bytes(&fi, em->ram_bytes); | |
4678 | btrfs_set_stack_file_extent_compression(&fi, em->compress_type); | |
4679 | ||
48778179 | 4680 | ret = log_extent_csums(trans, inode, log, em, ctx); |
8407f553 FM |
4681 | if (ret) |
4682 | return ret; | |
4683 | ||
5328b2a7 FM |
4684 | /* |
4685 | * If this is the first time we are logging the inode in the current | |
4686 | * transaction, we can avoid btrfs_drop_extents(), which is expensive | |
4687 | * because it does a deletion search, which always acquires write locks | |
4688 | * for extent buffers at levels 2, 1 and 0. This not only wastes time | |
4689 | * but also adds significant contention in a log tree, since log trees | |
4690 | * are small, with a root at level 2 or 3 at most, due to their short | |
4691 | * life span. | |
4692 | */ | |
0f8ce498 | 4693 | if (ctx->logged_before) { |
5328b2a7 FM |
4694 | drop_args.path = path; |
4695 | drop_args.start = em->start; | |
4696 | drop_args.end = em->start + em->len; | |
4697 | drop_args.replace_extent = true; | |
e1f53ed8 | 4698 | drop_args.extent_item_size = sizeof(fi); |
5328b2a7 FM |
4699 | ret = btrfs_drop_extents(trans, log, inode, &drop_args); |
4700 | if (ret) | |
4701 | return ret; | |
4702 | } | |
8407f553 | 4703 | |
5893dfb9 | 4704 | if (!drop_args.extent_inserted) { |
9d122629 | 4705 | key.objectid = btrfs_ino(inode); |
8407f553 FM |
4706 | key.type = BTRFS_EXTENT_DATA_KEY; |
4707 | key.offset = em->start; | |
4708 | ||
4709 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
e1f53ed8 | 4710 | sizeof(fi)); |
8407f553 FM |
4711 | if (ret) |
4712 | return ret; | |
4713 | } | |
4714 | leaf = path->nodes[0]; | |
e1f53ed8 FM |
4715 | write_extent_buffer(leaf, &fi, |
4716 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4717 | sizeof(fi)); | |
8407f553 FM |
4718 | btrfs_mark_buffer_dirty(leaf); |
4719 | ||
4720 | btrfs_release_path(path); | |
4721 | ||
4722 | return ret; | |
4723 | } | |
4724 | ||
31d11b83 FM |
4725 | /* |
4726 | * Log all prealloc extents beyond the inode's i_size to make sure we do not | |
d9947887 | 4727 | * lose them after doing a full/fast fsync and replaying the log. We scan the |
31d11b83 FM |
4728 | * subvolume's root instead of iterating the inode's extent map tree because |
4729 | * otherwise we can log incorrect extent items based on extent map conversion. | |
4730 | * That can happen due to the fact that extent maps are merged when they | |
4731 | * are not in the extent map tree's list of modified extents. | |
4732 | */ | |
4733 | static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans, | |
4734 | struct btrfs_inode *inode, | |
4735 | struct btrfs_path *path) | |
4736 | { | |
4737 | struct btrfs_root *root = inode->root; | |
4738 | struct btrfs_key key; | |
4739 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
4740 | const u64 ino = btrfs_ino(inode); | |
4741 | struct btrfs_path *dst_path = NULL; | |
0e56315c | 4742 | bool dropped_extents = false; |
f135cea3 FM |
4743 | u64 truncate_offset = i_size; |
4744 | struct extent_buffer *leaf; | |
4745 | int slot; | |
31d11b83 FM |
4746 | int ins_nr = 0; |
4747 | int start_slot; | |
4748 | int ret; | |
4749 | ||
4750 | if (!(inode->flags & BTRFS_INODE_PREALLOC)) | |
4751 | return 0; | |
4752 | ||
4753 | key.objectid = ino; | |
4754 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4755 | key.offset = i_size; | |
4756 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4757 | if (ret < 0) | |
4758 | goto out; | |
4759 | ||
f135cea3 FM |
4760 | /* |
4761 | * We must check if there is a prealloc extent that starts before the | |
4762 | * i_size and crosses the i_size boundary. This is to ensure later we | |
4763 | * truncate down to the end of that extent and not to the i_size, as | |
4764 | * otherwise we end up losing part of the prealloc extent after a log | |
4765 | * replay and with an implicit hole if there is another prealloc extent | |
4766 | * that starts at an offset beyond i_size. | |
4767 | */ | |
4768 | ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY); | |
4769 | if (ret < 0) | |
4770 | goto out; | |
4771 | ||
4772 | if (ret == 0) { | |
4773 | struct btrfs_file_extent_item *ei; | |
4774 | ||
4775 | leaf = path->nodes[0]; | |
4776 | slot = path->slots[0]; | |
4777 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4778 | ||
4779 | if (btrfs_file_extent_type(leaf, ei) == | |
4780 | BTRFS_FILE_EXTENT_PREALLOC) { | |
4781 | u64 extent_end; | |
4782 | ||
4783 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4784 | extent_end = key.offset + | |
4785 | btrfs_file_extent_num_bytes(leaf, ei); | |
4786 | ||
4787 | if (extent_end > i_size) | |
4788 | truncate_offset = extent_end; | |
4789 | } | |
4790 | } else { | |
4791 | ret = 0; | |
4792 | } | |
4793 | ||
31d11b83 | 4794 | while (true) { |
f135cea3 FM |
4795 | leaf = path->nodes[0]; |
4796 | slot = path->slots[0]; | |
31d11b83 FM |
4797 | |
4798 | if (slot >= btrfs_header_nritems(leaf)) { | |
4799 | if (ins_nr > 0) { | |
4800 | ret = copy_items(trans, inode, dst_path, path, | |
0e56315c | 4801 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4802 | if (ret < 0) |
4803 | goto out; | |
4804 | ins_nr = 0; | |
4805 | } | |
4806 | ret = btrfs_next_leaf(root, path); | |
4807 | if (ret < 0) | |
4808 | goto out; | |
4809 | if (ret > 0) { | |
4810 | ret = 0; | |
4811 | break; | |
4812 | } | |
4813 | continue; | |
4814 | } | |
4815 | ||
4816 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4817 | if (key.objectid > ino) | |
4818 | break; | |
4819 | if (WARN_ON_ONCE(key.objectid < ino) || | |
4820 | key.type < BTRFS_EXTENT_DATA_KEY || | |
4821 | key.offset < i_size) { | |
4822 | path->slots[0]++; | |
4823 | continue; | |
4824 | } | |
0e56315c | 4825 | if (!dropped_extents) { |
31d11b83 FM |
4826 | /* |
4827 | * Avoid logging extent items logged in past fsync calls | |
4828 | * and leading to duplicate keys in the log tree. | |
4829 | */ | |
8a2b3da1 FM |
4830 | ret = truncate_inode_items(trans, root->log_root, inode, |
4831 | truncate_offset, | |
4832 | BTRFS_EXTENT_DATA_KEY); | |
31d11b83 FM |
4833 | if (ret) |
4834 | goto out; | |
0e56315c | 4835 | dropped_extents = true; |
31d11b83 FM |
4836 | } |
4837 | if (ins_nr == 0) | |
4838 | start_slot = slot; | |
4839 | ins_nr++; | |
4840 | path->slots[0]++; | |
4841 | if (!dst_path) { | |
4842 | dst_path = btrfs_alloc_path(); | |
4843 | if (!dst_path) { | |
4844 | ret = -ENOMEM; | |
4845 | goto out; | |
4846 | } | |
4847 | } | |
4848 | } | |
0bc2d3c0 | 4849 | if (ins_nr > 0) |
0e56315c | 4850 | ret = copy_items(trans, inode, dst_path, path, |
31d11b83 | 4851 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4852 | out: |
4853 | btrfs_release_path(path); | |
4854 | btrfs_free_path(dst_path); | |
4855 | return ret; | |
4856 | } | |
4857 | ||
5dc562c5 | 4858 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
9d122629 | 4859 | struct btrfs_inode *inode, |
827463c4 | 4860 | struct btrfs_path *path, |
48778179 | 4861 | struct btrfs_log_ctx *ctx) |
5dc562c5 | 4862 | { |
48778179 FM |
4863 | struct btrfs_ordered_extent *ordered; |
4864 | struct btrfs_ordered_extent *tmp; | |
5dc562c5 JB |
4865 | struct extent_map *em, *n; |
4866 | struct list_head extents; | |
9d122629 | 4867 | struct extent_map_tree *tree = &inode->extent_tree; |
5dc562c5 | 4868 | int ret = 0; |
2ab28f32 | 4869 | int num = 0; |
5dc562c5 JB |
4870 | |
4871 | INIT_LIST_HEAD(&extents); | |
4872 | ||
5dc562c5 | 4873 | write_lock(&tree->lock); |
5dc562c5 JB |
4874 | |
4875 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
4876 | list_del_init(&em->list); | |
2ab28f32 JB |
4877 | /* |
4878 | * Just an arbitrary number, this can be really CPU intensive | |
4879 | * once we start getting a lot of extents, and really once we | |
4880 | * have a bunch of extents we just want to commit since it will | |
4881 | * be faster. | |
4882 | */ | |
4883 | if (++num > 32768) { | |
4884 | list_del_init(&tree->modified_extents); | |
4885 | ret = -EFBIG; | |
4886 | goto process; | |
4887 | } | |
4888 | ||
5f96bfb7 | 4889 | if (em->generation < trans->transid) |
5dc562c5 | 4890 | continue; |
8c6c5928 | 4891 | |
31d11b83 FM |
4892 | /* We log prealloc extents beyond eof later. */ |
4893 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && | |
4894 | em->start >= i_size_read(&inode->vfs_inode)) | |
4895 | continue; | |
4896 | ||
ff44c6e3 | 4897 | /* Need a ref to keep it from getting evicted from cache */ |
490b54d6 | 4898 | refcount_inc(&em->refs); |
ff44c6e3 | 4899 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); |
5dc562c5 | 4900 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4901 | num++; |
5dc562c5 JB |
4902 | } |
4903 | ||
4904 | list_sort(NULL, &extents, extent_cmp); | |
2ab28f32 | 4905 | process: |
5dc562c5 JB |
4906 | while (!list_empty(&extents)) { |
4907 | em = list_entry(extents.next, struct extent_map, list); | |
4908 | ||
4909 | list_del_init(&em->list); | |
4910 | ||
4911 | /* | |
4912 | * If we had an error we just need to delete everybody from our | |
4913 | * private list. | |
4914 | */ | |
ff44c6e3 | 4915 | if (ret) { |
201a9038 | 4916 | clear_em_logging(tree, em); |
ff44c6e3 | 4917 | free_extent_map(em); |
5dc562c5 | 4918 | continue; |
ff44c6e3 JB |
4919 | } |
4920 | ||
4921 | write_unlock(&tree->lock); | |
5dc562c5 | 4922 | |
90d04510 | 4923 | ret = log_one_extent(trans, inode, em, path, ctx); |
ff44c6e3 | 4924 | write_lock(&tree->lock); |
201a9038 JB |
4925 | clear_em_logging(tree, em); |
4926 | free_extent_map(em); | |
5dc562c5 | 4927 | } |
ff44c6e3 JB |
4928 | WARN_ON(!list_empty(&extents)); |
4929 | write_unlock(&tree->lock); | |
5dc562c5 | 4930 | |
31d11b83 FM |
4931 | if (!ret) |
4932 | ret = btrfs_log_prealloc_extents(trans, inode, path); | |
48778179 FM |
4933 | if (ret) |
4934 | return ret; | |
31d11b83 | 4935 | |
48778179 FM |
4936 | /* |
4937 | * We have logged all extents successfully, now make sure the commit of | |
4938 | * the current transaction waits for the ordered extents to complete | |
4939 | * before it commits and wipes out the log trees, otherwise we would | |
4940 | * lose data if an ordered extents completes after the transaction | |
4941 | * commits and a power failure happens after the transaction commit. | |
4942 | */ | |
4943 | list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) { | |
4944 | list_del_init(&ordered->log_list); | |
4945 | set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags); | |
4946 | ||
4947 | if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) { | |
4948 | spin_lock_irq(&inode->ordered_tree.lock); | |
4949 | if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) { | |
4950 | set_bit(BTRFS_ORDERED_PENDING, &ordered->flags); | |
4951 | atomic_inc(&trans->transaction->pending_ordered); | |
4952 | } | |
4953 | spin_unlock_irq(&inode->ordered_tree.lock); | |
4954 | } | |
4955 | btrfs_put_ordered_extent(ordered); | |
4956 | } | |
4957 | ||
4958 | return 0; | |
5dc562c5 JB |
4959 | } |
4960 | ||
481b01c0 | 4961 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
4962 | struct btrfs_path *path, u64 *size_ret) |
4963 | { | |
4964 | struct btrfs_key key; | |
4965 | int ret; | |
4966 | ||
481b01c0 | 4967 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
4968 | key.type = BTRFS_INODE_ITEM_KEY; |
4969 | key.offset = 0; | |
4970 | ||
4971 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4972 | if (ret < 0) { | |
4973 | return ret; | |
4974 | } else if (ret > 0) { | |
2f2ff0ee | 4975 | *size_ret = 0; |
1a4bcf47 FM |
4976 | } else { |
4977 | struct btrfs_inode_item *item; | |
4978 | ||
4979 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4980 | struct btrfs_inode_item); | |
4981 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
bf504110 FM |
4982 | /* |
4983 | * If the in-memory inode's i_size is smaller then the inode | |
4984 | * size stored in the btree, return the inode's i_size, so | |
4985 | * that we get a correct inode size after replaying the log | |
4986 | * when before a power failure we had a shrinking truncate | |
4987 | * followed by addition of a new name (rename / new hard link). | |
4988 | * Otherwise return the inode size from the btree, to avoid | |
4989 | * data loss when replaying a log due to previously doing a | |
4990 | * write that expands the inode's size and logging a new name | |
4991 | * immediately after. | |
4992 | */ | |
4993 | if (*size_ret > inode->vfs_inode.i_size) | |
4994 | *size_ret = inode->vfs_inode.i_size; | |
1a4bcf47 FM |
4995 | } |
4996 | ||
4997 | btrfs_release_path(path); | |
4998 | return 0; | |
4999 | } | |
5000 | ||
36283bf7 FM |
5001 | /* |
5002 | * At the moment we always log all xattrs. This is to figure out at log replay | |
5003 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
5004 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
5005 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
5006 | * happens, causing the log to be replayed the next time the fs is mounted, | |
5007 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
5008 | * with a journal, ext3/4, xfs, f2fs, etc). | |
5009 | */ | |
5010 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
1a93c36a | 5011 | struct btrfs_inode *inode, |
36283bf7 FM |
5012 | struct btrfs_path *path, |
5013 | struct btrfs_path *dst_path) | |
5014 | { | |
90d04510 | 5015 | struct btrfs_root *root = inode->root; |
36283bf7 FM |
5016 | int ret; |
5017 | struct btrfs_key key; | |
1a93c36a | 5018 | const u64 ino = btrfs_ino(inode); |
36283bf7 FM |
5019 | int ins_nr = 0; |
5020 | int start_slot = 0; | |
f2f121ab FM |
5021 | bool found_xattrs = false; |
5022 | ||
5023 | if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags)) | |
5024 | return 0; | |
36283bf7 FM |
5025 | |
5026 | key.objectid = ino; | |
5027 | key.type = BTRFS_XATTR_ITEM_KEY; | |
5028 | key.offset = 0; | |
5029 | ||
5030 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5031 | if (ret < 0) | |
5032 | return ret; | |
5033 | ||
5034 | while (true) { | |
5035 | int slot = path->slots[0]; | |
5036 | struct extent_buffer *leaf = path->nodes[0]; | |
5037 | int nritems = btrfs_header_nritems(leaf); | |
5038 | ||
5039 | if (slot >= nritems) { | |
5040 | if (ins_nr > 0) { | |
1a93c36a | 5041 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 5042 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
5043 | if (ret < 0) |
5044 | return ret; | |
5045 | ins_nr = 0; | |
5046 | } | |
5047 | ret = btrfs_next_leaf(root, path); | |
5048 | if (ret < 0) | |
5049 | return ret; | |
5050 | else if (ret > 0) | |
5051 | break; | |
5052 | continue; | |
5053 | } | |
5054 | ||
5055 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5056 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
5057 | break; | |
5058 | ||
5059 | if (ins_nr == 0) | |
5060 | start_slot = slot; | |
5061 | ins_nr++; | |
5062 | path->slots[0]++; | |
f2f121ab | 5063 | found_xattrs = true; |
36283bf7 FM |
5064 | cond_resched(); |
5065 | } | |
5066 | if (ins_nr > 0) { | |
1a93c36a | 5067 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 5068 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
5069 | if (ret < 0) |
5070 | return ret; | |
5071 | } | |
5072 | ||
f2f121ab FM |
5073 | if (!found_xattrs) |
5074 | set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags); | |
5075 | ||
36283bf7 FM |
5076 | return 0; |
5077 | } | |
5078 | ||
a89ca6f2 | 5079 | /* |
0e56315c FM |
5080 | * When using the NO_HOLES feature if we punched a hole that causes the |
5081 | * deletion of entire leafs or all the extent items of the first leaf (the one | |
5082 | * that contains the inode item and references) we may end up not processing | |
5083 | * any extents, because there are no leafs with a generation matching the | |
5084 | * current transaction that have extent items for our inode. So we need to find | |
5085 | * if any holes exist and then log them. We also need to log holes after any | |
5086 | * truncate operation that changes the inode's size. | |
a89ca6f2 | 5087 | */ |
0e56315c | 5088 | static int btrfs_log_holes(struct btrfs_trans_handle *trans, |
0e56315c | 5089 | struct btrfs_inode *inode, |
7af59743 | 5090 | struct btrfs_path *path) |
a89ca6f2 | 5091 | { |
90d04510 | 5092 | struct btrfs_root *root = inode->root; |
0b246afa | 5093 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 | 5094 | struct btrfs_key key; |
a0308dd7 NB |
5095 | const u64 ino = btrfs_ino(inode); |
5096 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
7af59743 | 5097 | u64 prev_extent_end = 0; |
0e56315c | 5098 | int ret; |
a89ca6f2 | 5099 | |
0e56315c | 5100 | if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0) |
a89ca6f2 FM |
5101 | return 0; |
5102 | ||
5103 | key.objectid = ino; | |
5104 | key.type = BTRFS_EXTENT_DATA_KEY; | |
7af59743 | 5105 | key.offset = 0; |
a89ca6f2 FM |
5106 | |
5107 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
a89ca6f2 FM |
5108 | if (ret < 0) |
5109 | return ret; | |
5110 | ||
0e56315c | 5111 | while (true) { |
0e56315c | 5112 | struct extent_buffer *leaf = path->nodes[0]; |
a89ca6f2 | 5113 | |
0e56315c FM |
5114 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
5115 | ret = btrfs_next_leaf(root, path); | |
5116 | if (ret < 0) | |
5117 | return ret; | |
5118 | if (ret > 0) { | |
5119 | ret = 0; | |
5120 | break; | |
5121 | } | |
5122 | leaf = path->nodes[0]; | |
5123 | } | |
5124 | ||
5125 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
5126 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
5127 | break; | |
5128 | ||
5129 | /* We have a hole, log it. */ | |
5130 | if (prev_extent_end < key.offset) { | |
7af59743 | 5131 | const u64 hole_len = key.offset - prev_extent_end; |
0e56315c FM |
5132 | |
5133 | /* | |
5134 | * Release the path to avoid deadlocks with other code | |
5135 | * paths that search the root while holding locks on | |
5136 | * leafs from the log root. | |
5137 | */ | |
5138 | btrfs_release_path(path); | |
d1f68ba0 OS |
5139 | ret = btrfs_insert_hole_extent(trans, root->log_root, |
5140 | ino, prev_extent_end, | |
5141 | hole_len); | |
0e56315c FM |
5142 | if (ret < 0) |
5143 | return ret; | |
5144 | ||
5145 | /* | |
5146 | * Search for the same key again in the root. Since it's | |
5147 | * an extent item and we are holding the inode lock, the | |
5148 | * key must still exist. If it doesn't just emit warning | |
5149 | * and return an error to fall back to a transaction | |
5150 | * commit. | |
5151 | */ | |
5152 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5153 | if (ret < 0) | |
5154 | return ret; | |
5155 | if (WARN_ON(ret > 0)) | |
5156 | return -ENOENT; | |
5157 | leaf = path->nodes[0]; | |
5158 | } | |
a89ca6f2 | 5159 | |
7af59743 | 5160 | prev_extent_end = btrfs_file_extent_end(path); |
0e56315c FM |
5161 | path->slots[0]++; |
5162 | cond_resched(); | |
a89ca6f2 | 5163 | } |
a89ca6f2 | 5164 | |
7af59743 | 5165 | if (prev_extent_end < i_size) { |
0e56315c | 5166 | u64 hole_len; |
a89ca6f2 | 5167 | |
0e56315c | 5168 | btrfs_release_path(path); |
7af59743 | 5169 | hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize); |
d1f68ba0 OS |
5170 | ret = btrfs_insert_hole_extent(trans, root->log_root, ino, |
5171 | prev_extent_end, hole_len); | |
0e56315c FM |
5172 | if (ret < 0) |
5173 | return ret; | |
5174 | } | |
5175 | ||
5176 | return 0; | |
a89ca6f2 FM |
5177 | } |
5178 | ||
56f23fdb FM |
5179 | /* |
5180 | * When we are logging a new inode X, check if it doesn't have a reference that | |
5181 | * matches the reference from some other inode Y created in a past transaction | |
5182 | * and that was renamed in the current transaction. If we don't do this, then at | |
5183 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
5184 | * | |
5185 | * mkdir /mnt/x | |
5186 | * echo "hello world" > /mnt/x/foobar | |
5187 | * sync | |
5188 | * mv /mnt/x /mnt/y | |
5189 | * mkdir /mnt/x # or touch /mnt/x | |
5190 | * xfs_io -c fsync /mnt/x | |
5191 | * <power fail> | |
5192 | * mount fs, trigger log replay | |
5193 | * | |
5194 | * After the log replay procedure, we would lose the first directory and all its | |
5195 | * files (file foobar). | |
5196 | * For the case where inode Y is not a directory we simply end up losing it: | |
5197 | * | |
5198 | * echo "123" > /mnt/foo | |
5199 | * sync | |
5200 | * mv /mnt/foo /mnt/bar | |
5201 | * echo "abc" > /mnt/foo | |
5202 | * xfs_io -c fsync /mnt/foo | |
5203 | * <power fail> | |
5204 | * | |
5205 | * We also need this for cases where a snapshot entry is replaced by some other | |
5206 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
5207 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
5208 | * if it were a regular entry: | |
5209 | * | |
5210 | * mkdir /mnt/x | |
5211 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
5212 | * btrfs subvolume delete /mnt/x/snap | |
5213 | * rmdir /mnt/x | |
5214 | * mkdir /mnt/x | |
5215 | * fsync /mnt/x or fsync some new file inside it | |
5216 | * <power fail> | |
5217 | * | |
5218 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
5219 | * the same transaction. | |
5220 | */ | |
5221 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
5222 | const int slot, | |
5223 | const struct btrfs_key *key, | |
4791c8f1 | 5224 | struct btrfs_inode *inode, |
a3baaf0d | 5225 | u64 *other_ino, u64 *other_parent) |
56f23fdb FM |
5226 | { |
5227 | int ret; | |
5228 | struct btrfs_path *search_path; | |
5229 | char *name = NULL; | |
5230 | u32 name_len = 0; | |
3212fa14 | 5231 | u32 item_size = btrfs_item_size(eb, slot); |
56f23fdb FM |
5232 | u32 cur_offset = 0; |
5233 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
5234 | ||
5235 | search_path = btrfs_alloc_path(); | |
5236 | if (!search_path) | |
5237 | return -ENOMEM; | |
5238 | search_path->search_commit_root = 1; | |
5239 | search_path->skip_locking = 1; | |
5240 | ||
5241 | while (cur_offset < item_size) { | |
5242 | u64 parent; | |
5243 | u32 this_name_len; | |
5244 | u32 this_len; | |
5245 | unsigned long name_ptr; | |
5246 | struct btrfs_dir_item *di; | |
6db75318 | 5247 | struct fscrypt_str name_str; |
56f23fdb FM |
5248 | |
5249 | if (key->type == BTRFS_INODE_REF_KEY) { | |
5250 | struct btrfs_inode_ref *iref; | |
5251 | ||
5252 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
5253 | parent = key->offset; | |
5254 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
5255 | name_ptr = (unsigned long)(iref + 1); | |
5256 | this_len = sizeof(*iref) + this_name_len; | |
5257 | } else { | |
5258 | struct btrfs_inode_extref *extref; | |
5259 | ||
5260 | extref = (struct btrfs_inode_extref *)(ptr + | |
5261 | cur_offset); | |
5262 | parent = btrfs_inode_extref_parent(eb, extref); | |
5263 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
5264 | name_ptr = (unsigned long)&extref->name; | |
5265 | this_len = sizeof(*extref) + this_name_len; | |
5266 | } | |
5267 | ||
5268 | if (this_name_len > name_len) { | |
5269 | char *new_name; | |
5270 | ||
5271 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
5272 | if (!new_name) { | |
5273 | ret = -ENOMEM; | |
5274 | goto out; | |
5275 | } | |
5276 | name_len = this_name_len; | |
5277 | name = new_name; | |
5278 | } | |
5279 | ||
5280 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
e43eec81 STD |
5281 | |
5282 | name_str.name = name; | |
5283 | name_str.len = this_name_len; | |
4791c8f1 | 5284 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
e43eec81 | 5285 | parent, &name_str, 0); |
56f23fdb | 5286 | if (di && !IS_ERR(di)) { |
44f714da FM |
5287 | struct btrfs_key di_key; |
5288 | ||
5289 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
5290 | di, &di_key); | |
5291 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
6b5fc433 FM |
5292 | if (di_key.objectid != key->objectid) { |
5293 | ret = 1; | |
5294 | *other_ino = di_key.objectid; | |
a3baaf0d | 5295 | *other_parent = parent; |
6b5fc433 FM |
5296 | } else { |
5297 | ret = 0; | |
5298 | } | |
44f714da FM |
5299 | } else { |
5300 | ret = -EAGAIN; | |
5301 | } | |
56f23fdb FM |
5302 | goto out; |
5303 | } else if (IS_ERR(di)) { | |
5304 | ret = PTR_ERR(di); | |
5305 | goto out; | |
5306 | } | |
5307 | btrfs_release_path(search_path); | |
5308 | ||
5309 | cur_offset += this_len; | |
5310 | } | |
5311 | ret = 0; | |
5312 | out: | |
5313 | btrfs_free_path(search_path); | |
5314 | kfree(name); | |
5315 | return ret; | |
5316 | } | |
5317 | ||
a3751024 FM |
5318 | /* |
5319 | * Check if we need to log an inode. This is used in contexts where while | |
5320 | * logging an inode we need to log another inode (either that it exists or in | |
5321 | * full mode). This is used instead of btrfs_inode_in_log() because the later | |
5322 | * requires the inode to be in the log and have the log transaction committed, | |
5323 | * while here we do not care if the log transaction was already committed - our | |
5324 | * caller will commit the log later - and we want to avoid logging an inode | |
5325 | * multiple times when multiple tasks have joined the same log transaction. | |
5326 | */ | |
5327 | static bool need_log_inode(const struct btrfs_trans_handle *trans, | |
5328 | const struct btrfs_inode *inode) | |
5329 | { | |
5330 | /* | |
5331 | * If a directory was not modified, no dentries added or removed, we can | |
5332 | * and should avoid logging it. | |
5333 | */ | |
5334 | if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid) | |
5335 | return false; | |
5336 | ||
5337 | /* | |
5338 | * If this inode does not have new/updated/deleted xattrs since the last | |
5339 | * time it was logged and is flagged as logged in the current transaction, | |
5340 | * we can skip logging it. As for new/deleted names, those are updated in | |
5341 | * the log by link/unlink/rename operations. | |
5342 | * In case the inode was logged and then evicted and reloaded, its | |
5343 | * logged_trans will be 0, in which case we have to fully log it since | |
5344 | * logged_trans is a transient field, not persisted. | |
5345 | */ | |
5346 | if (inode->logged_trans == trans->transid && | |
5347 | !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags)) | |
5348 | return false; | |
5349 | ||
5350 | return true; | |
5351 | } | |
5352 | ||
f6d86dbe FM |
5353 | struct btrfs_dir_list { |
5354 | u64 ino; | |
5355 | struct list_head list; | |
5356 | }; | |
5357 | ||
5358 | /* | |
5359 | * Log the inodes of the new dentries of a directory. | |
5360 | * See process_dir_items_leaf() for details about why it is needed. | |
5361 | * This is a recursive operation - if an existing dentry corresponds to a | |
5362 | * directory, that directory's new entries are logged too (same behaviour as | |
5363 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5364 | * the dentries point to we do not acquire their VFS lock, otherwise lockdep | |
5365 | * complains about the following circular lock dependency / possible deadlock: | |
5366 | * | |
5367 | * CPU0 CPU1 | |
5368 | * ---- ---- | |
5369 | * lock(&type->i_mutex_dir_key#3/2); | |
5370 | * lock(sb_internal#2); | |
5371 | * lock(&type->i_mutex_dir_key#3/2); | |
5372 | * lock(&sb->s_type->i_mutex_key#14); | |
5373 | * | |
5374 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5375 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5376 | * Not acquiring the VFS lock of the inodes is still safe because: | |
5377 | * | |
5378 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5379 | * that while logging the inode new references (names) are added or removed | |
5380 | * from the inode, leaving the logged inode item with a link count that does | |
5381 | * not match the number of logged inode reference items. This is fine because | |
5382 | * at log replay time we compute the real number of links and correct the | |
5383 | * link count in the inode item (see replay_one_buffer() and | |
5384 | * link_to_fixup_dir()); | |
5385 | * | |
5386 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5387 | * while logging the inode's items new index items (key type | |
5388 | * BTRFS_DIR_INDEX_KEY) are added to fs/subvol tree and the logged inode item | |
5389 | * has a size that doesn't match the sum of the lengths of all the logged | |
5390 | * names - this is ok, not a problem, because at log replay time we set the | |
5391 | * directory's i_size to the correct value (see replay_one_name() and | |
5392 | * do_overwrite_item()). | |
5393 | */ | |
5394 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5395 | struct btrfs_inode *start_inode, | |
5396 | struct btrfs_log_ctx *ctx) | |
5397 | { | |
5398 | struct btrfs_root *root = start_inode->root; | |
5399 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5400 | struct btrfs_path *path; | |
5401 | LIST_HEAD(dir_list); | |
5402 | struct btrfs_dir_list *dir_elem; | |
5403 | u64 ino = btrfs_ino(start_inode); | |
5404 | int ret = 0; | |
5405 | ||
5406 | /* | |
5407 | * If we are logging a new name, as part of a link or rename operation, | |
5408 | * don't bother logging new dentries, as we just want to log the names | |
5409 | * of an inode and that any new parents exist. | |
5410 | */ | |
5411 | if (ctx->logging_new_name) | |
5412 | return 0; | |
5413 | ||
5414 | path = btrfs_alloc_path(); | |
5415 | if (!path) | |
5416 | return -ENOMEM; | |
5417 | ||
5418 | while (true) { | |
5419 | struct extent_buffer *leaf; | |
5420 | struct btrfs_key min_key; | |
5421 | bool continue_curr_inode = true; | |
5422 | int nritems; | |
5423 | int i; | |
5424 | ||
5425 | min_key.objectid = ino; | |
5426 | min_key.type = BTRFS_DIR_INDEX_KEY; | |
5427 | min_key.offset = 0; | |
5428 | again: | |
5429 | btrfs_release_path(path); | |
5430 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); | |
5431 | if (ret < 0) { | |
5432 | break; | |
5433 | } else if (ret > 0) { | |
5434 | ret = 0; | |
5435 | goto next; | |
5436 | } | |
5437 | ||
5438 | leaf = path->nodes[0]; | |
5439 | nritems = btrfs_header_nritems(leaf); | |
5440 | for (i = path->slots[0]; i < nritems; i++) { | |
5441 | struct btrfs_dir_item *di; | |
5442 | struct btrfs_key di_key; | |
5443 | struct inode *di_inode; | |
5444 | int log_mode = LOG_INODE_EXISTS; | |
5445 | int type; | |
5446 | ||
5447 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5448 | if (min_key.objectid != ino || | |
5449 | min_key.type != BTRFS_DIR_INDEX_KEY) { | |
5450 | continue_curr_inode = false; | |
5451 | break; | |
5452 | } | |
5453 | ||
5454 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
94a48aef | 5455 | type = btrfs_dir_ftype(leaf, di); |
f6d86dbe FM |
5456 | if (btrfs_dir_transid(leaf, di) < trans->transid) |
5457 | continue; | |
5458 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5459 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5460 | continue; | |
5461 | ||
5462 | btrfs_release_path(path); | |
5463 | di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root); | |
5464 | if (IS_ERR(di_inode)) { | |
5465 | ret = PTR_ERR(di_inode); | |
5466 | goto out; | |
5467 | } | |
5468 | ||
5469 | if (!need_log_inode(trans, BTRFS_I(di_inode))) { | |
5470 | btrfs_add_delayed_iput(di_inode); | |
5471 | break; | |
5472 | } | |
5473 | ||
5474 | ctx->log_new_dentries = false; | |
5475 | if (type == BTRFS_FT_DIR) | |
5476 | log_mode = LOG_INODE_ALL; | |
5477 | ret = btrfs_log_inode(trans, BTRFS_I(di_inode), | |
5478 | log_mode, ctx); | |
5479 | btrfs_add_delayed_iput(di_inode); | |
5480 | if (ret) | |
5481 | goto out; | |
5482 | if (ctx->log_new_dentries) { | |
5483 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5484 | if (!dir_elem) { | |
5485 | ret = -ENOMEM; | |
5486 | goto out; | |
5487 | } | |
5488 | dir_elem->ino = di_key.objectid; | |
5489 | list_add_tail(&dir_elem->list, &dir_list); | |
5490 | } | |
5491 | break; | |
5492 | } | |
5493 | ||
5494 | if (continue_curr_inode && min_key.offset < (u64)-1) { | |
5495 | min_key.offset++; | |
5496 | goto again; | |
5497 | } | |
5498 | ||
5499 | next: | |
5500 | if (list_empty(&dir_list)) | |
5501 | break; | |
5502 | ||
5503 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, list); | |
5504 | ino = dir_elem->ino; | |
5505 | list_del(&dir_elem->list); | |
5506 | kfree(dir_elem); | |
5507 | } | |
5508 | out: | |
5509 | btrfs_free_path(path); | |
5510 | if (ret) { | |
5511 | struct btrfs_dir_list *next; | |
5512 | ||
5513 | list_for_each_entry_safe(dir_elem, next, &dir_list, list) | |
5514 | kfree(dir_elem); | |
5515 | } | |
5516 | ||
5517 | return ret; | |
5518 | } | |
5519 | ||
6b5fc433 FM |
5520 | struct btrfs_ino_list { |
5521 | u64 ino; | |
a3baaf0d | 5522 | u64 parent; |
6b5fc433 FM |
5523 | struct list_head list; |
5524 | }; | |
5525 | ||
e09d94c9 FM |
5526 | static void free_conflicting_inodes(struct btrfs_log_ctx *ctx) |
5527 | { | |
5528 | struct btrfs_ino_list *curr; | |
5529 | struct btrfs_ino_list *next; | |
5530 | ||
5531 | list_for_each_entry_safe(curr, next, &ctx->conflict_inodes, list) { | |
5532 | list_del(&curr->list); | |
5533 | kfree(curr); | |
5534 | } | |
5535 | } | |
5536 | ||
5557a069 FM |
5537 | static int conflicting_inode_is_dir(struct btrfs_root *root, u64 ino, |
5538 | struct btrfs_path *path) | |
5539 | { | |
5540 | struct btrfs_key key; | |
5541 | int ret; | |
5542 | ||
5543 | key.objectid = ino; | |
5544 | key.type = BTRFS_INODE_ITEM_KEY; | |
5545 | key.offset = 0; | |
5546 | ||
5547 | path->search_commit_root = 1; | |
5548 | path->skip_locking = 1; | |
5549 | ||
5550 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5551 | if (WARN_ON_ONCE(ret > 0)) { | |
5552 | /* | |
5553 | * We have previously found the inode through the commit root | |
5554 | * so this should not happen. If it does, just error out and | |
5555 | * fallback to a transaction commit. | |
5556 | */ | |
5557 | ret = -ENOENT; | |
5558 | } else if (ret == 0) { | |
5559 | struct btrfs_inode_item *item; | |
5560 | ||
5561 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5562 | struct btrfs_inode_item); | |
5563 | if (S_ISDIR(btrfs_inode_mode(path->nodes[0], item))) | |
5564 | ret = 1; | |
5565 | } | |
5566 | ||
5567 | btrfs_release_path(path); | |
5568 | path->search_commit_root = 0; | |
5569 | path->skip_locking = 0; | |
5570 | ||
5571 | return ret; | |
5572 | } | |
5573 | ||
e09d94c9 FM |
5574 | static int add_conflicting_inode(struct btrfs_trans_handle *trans, |
5575 | struct btrfs_root *root, | |
5557a069 | 5576 | struct btrfs_path *path, |
e09d94c9 FM |
5577 | u64 ino, u64 parent, |
5578 | struct btrfs_log_ctx *ctx) | |
6b5fc433 FM |
5579 | { |
5580 | struct btrfs_ino_list *ino_elem; | |
e09d94c9 FM |
5581 | struct inode *inode; |
5582 | ||
5583 | /* | |
5584 | * It's rare to have a lot of conflicting inodes, in practice it is not | |
5585 | * common to have more than 1 or 2. We don't want to collect too many, | |
5586 | * as we could end up logging too many inodes (even if only in | |
5587 | * LOG_INODE_EXISTS mode) and slow down other fsyncs or transaction | |
5588 | * commits. | |
5589 | */ | |
5590 | if (ctx->num_conflict_inodes >= MAX_CONFLICT_INODES) | |
5591 | return BTRFS_LOG_FORCE_COMMIT; | |
5592 | ||
5593 | inode = btrfs_iget(root->fs_info->sb, ino, root); | |
5594 | /* | |
5595 | * If the other inode that had a conflicting dir entry was deleted in | |
5557a069 FM |
5596 | * the current transaction then we either: |
5597 | * | |
5598 | * 1) Log the parent directory (later after adding it to the list) if | |
5599 | * the inode is a directory. This is because it may be a deleted | |
5600 | * subvolume/snapshot or it may be a regular directory that had | |
5601 | * deleted subvolumes/snapshots (or subdirectories that had them), | |
5602 | * and at the moment we can't deal with dropping subvolumes/snapshots | |
5603 | * during log replay. So we just log the parent, which will result in | |
5604 | * a fallback to a transaction commit if we are dealing with those | |
5605 | * cases (last_unlink_trans will match the current transaction); | |
5606 | * | |
5607 | * 2) Do nothing if it's not a directory. During log replay we simply | |
5608 | * unlink the conflicting dentry from the parent directory and then | |
5609 | * add the dentry for our inode. Like this we can avoid logging the | |
5610 | * parent directory (and maybe fallback to a transaction commit in | |
5611 | * case it has a last_unlink_trans == trans->transid, due to moving | |
5612 | * some inode from it to some other directory). | |
e09d94c9 FM |
5613 | */ |
5614 | if (IS_ERR(inode)) { | |
5615 | int ret = PTR_ERR(inode); | |
5616 | ||
5617 | if (ret != -ENOENT) | |
5618 | return ret; | |
5619 | ||
5557a069 FM |
5620 | ret = conflicting_inode_is_dir(root, ino, path); |
5621 | /* Not a directory or we got an error. */ | |
5622 | if (ret <= 0) | |
5623 | return ret; | |
5624 | ||
5625 | /* Conflicting inode is a directory, so we'll log its parent. */ | |
e09d94c9 FM |
5626 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); |
5627 | if (!ino_elem) | |
5628 | return -ENOMEM; | |
5629 | ino_elem->ino = ino; | |
5630 | ino_elem->parent = parent; | |
5631 | list_add_tail(&ino_elem->list, &ctx->conflict_inodes); | |
5632 | ctx->num_conflict_inodes++; | |
5633 | ||
5634 | return 0; | |
5635 | } | |
5636 | ||
5637 | /* | |
5638 | * If the inode was already logged skip it - otherwise we can hit an | |
5639 | * infinite loop. Example: | |
5640 | * | |
5641 | * From the commit root (previous transaction) we have the following | |
5642 | * inodes: | |
5643 | * | |
5644 | * inode 257 a directory | |
5645 | * inode 258 with references "zz" and "zz_link" on inode 257 | |
5646 | * inode 259 with reference "a" on inode 257 | |
5647 | * | |
5648 | * And in the current (uncommitted) transaction we have: | |
5649 | * | |
5650 | * inode 257 a directory, unchanged | |
5651 | * inode 258 with references "a" and "a2" on inode 257 | |
5652 | * inode 259 with reference "zz_link" on inode 257 | |
5653 | * inode 261 with reference "zz" on inode 257 | |
5654 | * | |
5655 | * When logging inode 261 the following infinite loop could | |
5656 | * happen if we don't skip already logged inodes: | |
5657 | * | |
5658 | * - we detect inode 258 as a conflicting inode, with inode 261 | |
5659 | * on reference "zz", and log it; | |
5660 | * | |
5661 | * - we detect inode 259 as a conflicting inode, with inode 258 | |
5662 | * on reference "a", and log it; | |
5663 | * | |
5664 | * - we detect inode 258 as a conflicting inode, with inode 259 | |
5665 | * on reference "zz_link", and log it - again! After this we | |
5666 | * repeat the above steps forever. | |
5667 | * | |
5668 | * Here we can use need_log_inode() because we only need to log the | |
5669 | * inode in LOG_INODE_EXISTS mode and rename operations update the log, | |
5670 | * so that the log ends up with the new name and without the old name. | |
5671 | */ | |
5672 | if (!need_log_inode(trans, BTRFS_I(inode))) { | |
5673 | btrfs_add_delayed_iput(inode); | |
5674 | return 0; | |
5675 | } | |
5676 | ||
5677 | btrfs_add_delayed_iput(inode); | |
6b5fc433 FM |
5678 | |
5679 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
5680 | if (!ino_elem) | |
5681 | return -ENOMEM; | |
5682 | ino_elem->ino = ino; | |
a3baaf0d | 5683 | ino_elem->parent = parent; |
e09d94c9 FM |
5684 | list_add_tail(&ino_elem->list, &ctx->conflict_inodes); |
5685 | ctx->num_conflict_inodes++; | |
6b5fc433 | 5686 | |
e09d94c9 FM |
5687 | return 0; |
5688 | } | |
6b5fc433 | 5689 | |
e09d94c9 FM |
5690 | static int log_conflicting_inodes(struct btrfs_trans_handle *trans, |
5691 | struct btrfs_root *root, | |
5692 | struct btrfs_log_ctx *ctx) | |
5693 | { | |
5694 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5695 | int ret = 0; | |
6b5fc433 | 5696 | |
e09d94c9 FM |
5697 | /* |
5698 | * Conflicting inodes are logged by the first call to btrfs_log_inode(), | |
5699 | * otherwise we could have unbounded recursion of btrfs_log_inode() | |
5700 | * calls. This check guarantees we can have only 1 level of recursion. | |
5701 | */ | |
5702 | if (ctx->logging_conflict_inodes) | |
5703 | return 0; | |
5704 | ||
5705 | ctx->logging_conflict_inodes = true; | |
5706 | ||
5707 | /* | |
5708 | * New conflicting inodes may be found and added to the list while we | |
5709 | * are logging a conflicting inode, so keep iterating while the list is | |
5710 | * not empty. | |
5711 | */ | |
5712 | while (!list_empty(&ctx->conflict_inodes)) { | |
5713 | struct btrfs_ino_list *curr; | |
5714 | struct inode *inode; | |
5715 | u64 ino; | |
5716 | u64 parent; | |
5717 | ||
5718 | curr = list_first_entry(&ctx->conflict_inodes, | |
5719 | struct btrfs_ino_list, list); | |
5720 | ino = curr->ino; | |
5721 | parent = curr->parent; | |
5722 | list_del(&curr->list); | |
5723 | kfree(curr); | |
6b5fc433 | 5724 | |
0202e83f | 5725 | inode = btrfs_iget(fs_info->sb, ino, root); |
6b5fc433 FM |
5726 | /* |
5727 | * If the other inode that had a conflicting dir entry was | |
a3baaf0d | 5728 | * deleted in the current transaction, we need to log its parent |
e09d94c9 | 5729 | * directory. See the comment at add_conflicting_inode(). |
6b5fc433 FM |
5730 | */ |
5731 | if (IS_ERR(inode)) { | |
5732 | ret = PTR_ERR(inode); | |
e09d94c9 FM |
5733 | if (ret != -ENOENT) |
5734 | break; | |
5735 | ||
5736 | inode = btrfs_iget(fs_info->sb, parent, root); | |
5737 | if (IS_ERR(inode)) { | |
5738 | ret = PTR_ERR(inode); | |
5739 | break; | |
a3baaf0d | 5740 | } |
e09d94c9 FM |
5741 | |
5742 | /* | |
5743 | * Always log the directory, we cannot make this | |
5744 | * conditional on need_log_inode() because the directory | |
5745 | * might have been logged in LOG_INODE_EXISTS mode or | |
5746 | * the dir index of the conflicting inode is not in a | |
5747 | * dir index key range logged for the directory. So we | |
5748 | * must make sure the deletion is recorded. | |
5749 | */ | |
5750 | ret = btrfs_log_inode(trans, BTRFS_I(inode), | |
5751 | LOG_INODE_ALL, ctx); | |
5752 | btrfs_add_delayed_iput(inode); | |
5753 | if (ret) | |
5754 | break; | |
6b5fc433 FM |
5755 | continue; |
5756 | } | |
e09d94c9 | 5757 | |
b5e4ff9d | 5758 | /* |
e09d94c9 FM |
5759 | * Here we can use need_log_inode() because we only need to log |
5760 | * the inode in LOG_INODE_EXISTS mode and rename operations | |
5761 | * update the log, so that the log ends up with the new name and | |
5762 | * without the old name. | |
b5e4ff9d | 5763 | * |
e09d94c9 FM |
5764 | * We did this check at add_conflicting_inode(), but here we do |
5765 | * it again because if some other task logged the inode after | |
5766 | * that, we can avoid doing it again. | |
b5e4ff9d | 5767 | */ |
e09d94c9 | 5768 | if (!need_log_inode(trans, BTRFS_I(inode))) { |
b5e4ff9d FM |
5769 | btrfs_add_delayed_iput(inode); |
5770 | continue; | |
5771 | } | |
e09d94c9 | 5772 | |
6b5fc433 FM |
5773 | /* |
5774 | * We are safe logging the other inode without acquiring its | |
5775 | * lock as long as we log with the LOG_INODE_EXISTS mode. We | |
5776 | * are safe against concurrent renames of the other inode as | |
5777 | * well because during a rename we pin the log and update the | |
5778 | * log with the new name before we unpin it. | |
5779 | */ | |
e09d94c9 | 5780 | ret = btrfs_log_inode(trans, BTRFS_I(inode), LOG_INODE_EXISTS, ctx); |
410f954c | 5781 | btrfs_add_delayed_iput(inode); |
e09d94c9 FM |
5782 | if (ret) |
5783 | break; | |
6b5fc433 FM |
5784 | } |
5785 | ||
e09d94c9 FM |
5786 | ctx->logging_conflict_inodes = false; |
5787 | if (ret) | |
5788 | free_conflicting_inodes(ctx); | |
5789 | ||
6b5fc433 FM |
5790 | return ret; |
5791 | } | |
5792 | ||
da447009 FM |
5793 | static int copy_inode_items_to_log(struct btrfs_trans_handle *trans, |
5794 | struct btrfs_inode *inode, | |
5795 | struct btrfs_key *min_key, | |
5796 | const struct btrfs_key *max_key, | |
5797 | struct btrfs_path *path, | |
5798 | struct btrfs_path *dst_path, | |
5799 | const u64 logged_isize, | |
da447009 FM |
5800 | const int inode_only, |
5801 | struct btrfs_log_ctx *ctx, | |
5802 | bool *need_log_inode_item) | |
5803 | { | |
d9947887 | 5804 | const u64 i_size = i_size_read(&inode->vfs_inode); |
da447009 FM |
5805 | struct btrfs_root *root = inode->root; |
5806 | int ins_start_slot = 0; | |
5807 | int ins_nr = 0; | |
5808 | int ret; | |
5809 | ||
5810 | while (1) { | |
5811 | ret = btrfs_search_forward(root, min_key, path, trans->transid); | |
5812 | if (ret < 0) | |
5813 | return ret; | |
5814 | if (ret > 0) { | |
5815 | ret = 0; | |
5816 | break; | |
5817 | } | |
5818 | again: | |
5819 | /* Note, ins_nr might be > 0 here, cleanup outside the loop */ | |
5820 | if (min_key->objectid != max_key->objectid) | |
5821 | break; | |
5822 | if (min_key->type > max_key->type) | |
5823 | break; | |
5824 | ||
d9947887 | 5825 | if (min_key->type == BTRFS_INODE_ITEM_KEY) { |
da447009 | 5826 | *need_log_inode_item = false; |
d9947887 FM |
5827 | } else if (min_key->type == BTRFS_EXTENT_DATA_KEY && |
5828 | min_key->offset >= i_size) { | |
5829 | /* | |
5830 | * Extents at and beyond eof are logged with | |
5831 | * btrfs_log_prealloc_extents(). | |
5832 | * Only regular files have BTRFS_EXTENT_DATA_KEY keys, | |
5833 | * and no keys greater than that, so bail out. | |
5834 | */ | |
5835 | break; | |
5836 | } else if ((min_key->type == BTRFS_INODE_REF_KEY || | |
5837 | min_key->type == BTRFS_INODE_EXTREF_KEY) && | |
e09d94c9 FM |
5838 | (inode->generation == trans->transid || |
5839 | ctx->logging_conflict_inodes)) { | |
da447009 FM |
5840 | u64 other_ino = 0; |
5841 | u64 other_parent = 0; | |
5842 | ||
5843 | ret = btrfs_check_ref_name_override(path->nodes[0], | |
5844 | path->slots[0], min_key, inode, | |
5845 | &other_ino, &other_parent); | |
5846 | if (ret < 0) { | |
5847 | return ret; | |
289cffcb | 5848 | } else if (ret > 0 && |
da447009 FM |
5849 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { |
5850 | if (ins_nr > 0) { | |
5851 | ins_nr++; | |
5852 | } else { | |
5853 | ins_nr = 1; | |
5854 | ins_start_slot = path->slots[0]; | |
5855 | } | |
5856 | ret = copy_items(trans, inode, dst_path, path, | |
5857 | ins_start_slot, ins_nr, | |
5858 | inode_only, logged_isize); | |
5859 | if (ret < 0) | |
5860 | return ret; | |
5861 | ins_nr = 0; | |
5862 | ||
e09d94c9 | 5863 | btrfs_release_path(path); |
5557a069 | 5864 | ret = add_conflicting_inode(trans, root, path, |
e09d94c9 FM |
5865 | other_ino, |
5866 | other_parent, ctx); | |
da447009 FM |
5867 | if (ret) |
5868 | return ret; | |
da447009 FM |
5869 | goto next_key; |
5870 | } | |
d9947887 FM |
5871 | } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) { |
5872 | /* Skip xattrs, logged later with btrfs_log_all_xattrs() */ | |
da447009 FM |
5873 | if (ins_nr == 0) |
5874 | goto next_slot; | |
5875 | ret = copy_items(trans, inode, dst_path, path, | |
5876 | ins_start_slot, | |
5877 | ins_nr, inode_only, logged_isize); | |
5878 | if (ret < 0) | |
5879 | return ret; | |
5880 | ins_nr = 0; | |
5881 | goto next_slot; | |
5882 | } | |
5883 | ||
5884 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { | |
5885 | ins_nr++; | |
5886 | goto next_slot; | |
5887 | } else if (!ins_nr) { | |
5888 | ins_start_slot = path->slots[0]; | |
5889 | ins_nr = 1; | |
5890 | goto next_slot; | |
5891 | } | |
5892 | ||
5893 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, | |
5894 | ins_nr, inode_only, logged_isize); | |
5895 | if (ret < 0) | |
5896 | return ret; | |
5897 | ins_nr = 1; | |
5898 | ins_start_slot = path->slots[0]; | |
5899 | next_slot: | |
5900 | path->slots[0]++; | |
5901 | if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) { | |
5902 | btrfs_item_key_to_cpu(path->nodes[0], min_key, | |
5903 | path->slots[0]); | |
5904 | goto again; | |
5905 | } | |
5906 | if (ins_nr) { | |
5907 | ret = copy_items(trans, inode, dst_path, path, | |
5908 | ins_start_slot, ins_nr, inode_only, | |
5909 | logged_isize); | |
5910 | if (ret < 0) | |
5911 | return ret; | |
5912 | ins_nr = 0; | |
5913 | } | |
5914 | btrfs_release_path(path); | |
5915 | next_key: | |
5916 | if (min_key->offset < (u64)-1) { | |
5917 | min_key->offset++; | |
5918 | } else if (min_key->type < max_key->type) { | |
5919 | min_key->type++; | |
5920 | min_key->offset = 0; | |
5921 | } else { | |
5922 | break; | |
5923 | } | |
96acb375 FM |
5924 | |
5925 | /* | |
5926 | * We may process many leaves full of items for our inode, so | |
5927 | * avoid monopolizing a cpu for too long by rescheduling while | |
5928 | * not holding locks on any tree. | |
5929 | */ | |
5930 | cond_resched(); | |
da447009 | 5931 | } |
d9947887 | 5932 | if (ins_nr) { |
da447009 FM |
5933 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, |
5934 | ins_nr, inode_only, logged_isize); | |
d9947887 FM |
5935 | if (ret) |
5936 | return ret; | |
5937 | } | |
5938 | ||
5939 | if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) { | |
5940 | /* | |
5941 | * Release the path because otherwise we might attempt to double | |
5942 | * lock the same leaf with btrfs_log_prealloc_extents() below. | |
5943 | */ | |
5944 | btrfs_release_path(path); | |
5945 | ret = btrfs_log_prealloc_extents(trans, inode, dst_path); | |
5946 | } | |
da447009 FM |
5947 | |
5948 | return ret; | |
5949 | } | |
5950 | ||
30b80f3c FM |
5951 | static int insert_delayed_items_batch(struct btrfs_trans_handle *trans, |
5952 | struct btrfs_root *log, | |
5953 | struct btrfs_path *path, | |
5954 | const struct btrfs_item_batch *batch, | |
5955 | const struct btrfs_delayed_item *first_item) | |
5956 | { | |
5957 | const struct btrfs_delayed_item *curr = first_item; | |
5958 | int ret; | |
5959 | ||
5960 | ret = btrfs_insert_empty_items(trans, log, path, batch); | |
5961 | if (ret) | |
5962 | return ret; | |
5963 | ||
5964 | for (int i = 0; i < batch->nr; i++) { | |
5965 | char *data_ptr; | |
5966 | ||
5967 | data_ptr = btrfs_item_ptr(path->nodes[0], path->slots[0], char); | |
5968 | write_extent_buffer(path->nodes[0], &curr->data, | |
5969 | (unsigned long)data_ptr, curr->data_len); | |
5970 | curr = list_next_entry(curr, log_list); | |
5971 | path->slots[0]++; | |
5972 | } | |
5973 | ||
5974 | btrfs_release_path(path); | |
5975 | ||
5976 | return 0; | |
5977 | } | |
5978 | ||
5979 | static int log_delayed_insertion_items(struct btrfs_trans_handle *trans, | |
5980 | struct btrfs_inode *inode, | |
5981 | struct btrfs_path *path, | |
5982 | const struct list_head *delayed_ins_list, | |
5983 | struct btrfs_log_ctx *ctx) | |
5984 | { | |
5985 | /* 195 (4095 bytes of keys and sizes) fits in a single 4K page. */ | |
5986 | const int max_batch_size = 195; | |
5987 | const int leaf_data_size = BTRFS_LEAF_DATA_SIZE(trans->fs_info); | |
5988 | const u64 ino = btrfs_ino(inode); | |
5989 | struct btrfs_root *log = inode->root->log_root; | |
5990 | struct btrfs_item_batch batch = { | |
5991 | .nr = 0, | |
5992 | .total_data_size = 0, | |
5993 | }; | |
5994 | const struct btrfs_delayed_item *first = NULL; | |
5995 | const struct btrfs_delayed_item *curr; | |
5996 | char *ins_data; | |
5997 | struct btrfs_key *ins_keys; | |
5998 | u32 *ins_sizes; | |
5999 | u64 curr_batch_size = 0; | |
6000 | int batch_idx = 0; | |
6001 | int ret; | |
6002 | ||
6003 | /* We are adding dir index items to the log tree. */ | |
6004 | lockdep_assert_held(&inode->log_mutex); | |
6005 | ||
6006 | /* | |
6007 | * We collect delayed items before copying index keys from the subvolume | |
6008 | * to the log tree. However just after we collected them, they may have | |
6009 | * been flushed (all of them or just some of them), and therefore we | |
6010 | * could have copied them from the subvolume tree to the log tree. | |
6011 | * So find the first delayed item that was not yet logged (they are | |
6012 | * sorted by index number). | |
6013 | */ | |
6014 | list_for_each_entry(curr, delayed_ins_list, log_list) { | |
6015 | if (curr->index > inode->last_dir_index_offset) { | |
6016 | first = curr; | |
6017 | break; | |
6018 | } | |
6019 | } | |
6020 | ||
6021 | /* Empty list or all delayed items were already logged. */ | |
6022 | if (!first) | |
6023 | return 0; | |
6024 | ||
6025 | ins_data = kmalloc(max_batch_size * sizeof(u32) + | |
6026 | max_batch_size * sizeof(struct btrfs_key), GFP_NOFS); | |
6027 | if (!ins_data) | |
6028 | return -ENOMEM; | |
6029 | ins_sizes = (u32 *)ins_data; | |
6030 | batch.data_sizes = ins_sizes; | |
6031 | ins_keys = (struct btrfs_key *)(ins_data + max_batch_size * sizeof(u32)); | |
6032 | batch.keys = ins_keys; | |
6033 | ||
6034 | curr = first; | |
6035 | while (!list_entry_is_head(curr, delayed_ins_list, log_list)) { | |
6036 | const u32 curr_size = curr->data_len + sizeof(struct btrfs_item); | |
6037 | ||
6038 | if (curr_batch_size + curr_size > leaf_data_size || | |
6039 | batch.nr == max_batch_size) { | |
6040 | ret = insert_delayed_items_batch(trans, log, path, | |
6041 | &batch, first); | |
6042 | if (ret) | |
6043 | goto out; | |
6044 | batch_idx = 0; | |
6045 | batch.nr = 0; | |
6046 | batch.total_data_size = 0; | |
6047 | curr_batch_size = 0; | |
6048 | first = curr; | |
6049 | } | |
6050 | ||
6051 | ins_sizes[batch_idx] = curr->data_len; | |
6052 | ins_keys[batch_idx].objectid = ino; | |
6053 | ins_keys[batch_idx].type = BTRFS_DIR_INDEX_KEY; | |
6054 | ins_keys[batch_idx].offset = curr->index; | |
6055 | curr_batch_size += curr_size; | |
6056 | batch.total_data_size += curr->data_len; | |
6057 | batch.nr++; | |
6058 | batch_idx++; | |
6059 | curr = list_next_entry(curr, log_list); | |
6060 | } | |
6061 | ||
6062 | ASSERT(batch.nr >= 1); | |
6063 | ret = insert_delayed_items_batch(trans, log, path, &batch, first); | |
6064 | ||
6065 | curr = list_last_entry(delayed_ins_list, struct btrfs_delayed_item, | |
6066 | log_list); | |
6067 | inode->last_dir_index_offset = curr->index; | |
6068 | out: | |
6069 | kfree(ins_data); | |
6070 | ||
6071 | return ret; | |
6072 | } | |
6073 | ||
6074 | static int log_delayed_deletions_full(struct btrfs_trans_handle *trans, | |
6075 | struct btrfs_inode *inode, | |
6076 | struct btrfs_path *path, | |
6077 | const struct list_head *delayed_del_list, | |
6078 | struct btrfs_log_ctx *ctx) | |
6079 | { | |
6080 | const u64 ino = btrfs_ino(inode); | |
6081 | const struct btrfs_delayed_item *curr; | |
6082 | ||
6083 | curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item, | |
6084 | log_list); | |
6085 | ||
6086 | while (!list_entry_is_head(curr, delayed_del_list, log_list)) { | |
6087 | u64 first_dir_index = curr->index; | |
6088 | u64 last_dir_index; | |
6089 | const struct btrfs_delayed_item *next; | |
6090 | int ret; | |
6091 | ||
6092 | /* | |
6093 | * Find a range of consecutive dir index items to delete. Like | |
6094 | * this we log a single dir range item spanning several contiguous | |
6095 | * dir items instead of logging one range item per dir index item. | |
6096 | */ | |
6097 | next = list_next_entry(curr, log_list); | |
6098 | while (!list_entry_is_head(next, delayed_del_list, log_list)) { | |
6099 | if (next->index != curr->index + 1) | |
6100 | break; | |
6101 | curr = next; | |
6102 | next = list_next_entry(next, log_list); | |
6103 | } | |
6104 | ||
6105 | last_dir_index = curr->index; | |
6106 | ASSERT(last_dir_index >= first_dir_index); | |
6107 | ||
6108 | ret = insert_dir_log_key(trans, inode->root->log_root, path, | |
6109 | ino, first_dir_index, last_dir_index); | |
6110 | if (ret) | |
6111 | return ret; | |
6112 | curr = list_next_entry(curr, log_list); | |
6113 | } | |
6114 | ||
6115 | return 0; | |
6116 | } | |
6117 | ||
6118 | static int batch_delete_dir_index_items(struct btrfs_trans_handle *trans, | |
6119 | struct btrfs_inode *inode, | |
6120 | struct btrfs_path *path, | |
6121 | struct btrfs_log_ctx *ctx, | |
6122 | const struct list_head *delayed_del_list, | |
6123 | const struct btrfs_delayed_item *first, | |
6124 | const struct btrfs_delayed_item **last_ret) | |
6125 | { | |
6126 | const struct btrfs_delayed_item *next; | |
6127 | struct extent_buffer *leaf = path->nodes[0]; | |
6128 | const int last_slot = btrfs_header_nritems(leaf) - 1; | |
6129 | int slot = path->slots[0] + 1; | |
6130 | const u64 ino = btrfs_ino(inode); | |
6131 | ||
6132 | next = list_next_entry(first, log_list); | |
6133 | ||
6134 | while (slot < last_slot && | |
6135 | !list_entry_is_head(next, delayed_del_list, log_list)) { | |
6136 | struct btrfs_key key; | |
6137 | ||
6138 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
6139 | if (key.objectid != ino || | |
6140 | key.type != BTRFS_DIR_INDEX_KEY || | |
6141 | key.offset != next->index) | |
6142 | break; | |
6143 | ||
6144 | slot++; | |
6145 | *last_ret = next; | |
6146 | next = list_next_entry(next, log_list); | |
6147 | } | |
6148 | ||
6149 | return btrfs_del_items(trans, inode->root->log_root, path, | |
6150 | path->slots[0], slot - path->slots[0]); | |
6151 | } | |
6152 | ||
6153 | static int log_delayed_deletions_incremental(struct btrfs_trans_handle *trans, | |
6154 | struct btrfs_inode *inode, | |
6155 | struct btrfs_path *path, | |
6156 | const struct list_head *delayed_del_list, | |
6157 | struct btrfs_log_ctx *ctx) | |
6158 | { | |
6159 | struct btrfs_root *log = inode->root->log_root; | |
6160 | const struct btrfs_delayed_item *curr; | |
6161 | u64 last_range_start; | |
6162 | u64 last_range_end = 0; | |
6163 | struct btrfs_key key; | |
6164 | ||
6165 | key.objectid = btrfs_ino(inode); | |
6166 | key.type = BTRFS_DIR_INDEX_KEY; | |
6167 | curr = list_first_entry(delayed_del_list, struct btrfs_delayed_item, | |
6168 | log_list); | |
6169 | ||
6170 | while (!list_entry_is_head(curr, delayed_del_list, log_list)) { | |
6171 | const struct btrfs_delayed_item *last = curr; | |
6172 | u64 first_dir_index = curr->index; | |
6173 | u64 last_dir_index; | |
6174 | bool deleted_items = false; | |
6175 | int ret; | |
6176 | ||
6177 | key.offset = curr->index; | |
6178 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); | |
6179 | if (ret < 0) { | |
6180 | return ret; | |
6181 | } else if (ret == 0) { | |
6182 | ret = batch_delete_dir_index_items(trans, inode, path, ctx, | |
6183 | delayed_del_list, curr, | |
6184 | &last); | |
6185 | if (ret) | |
6186 | return ret; | |
6187 | deleted_items = true; | |
6188 | } | |
6189 | ||
6190 | btrfs_release_path(path); | |
6191 | ||
6192 | /* | |
6193 | * If we deleted items from the leaf, it means we have a range | |
6194 | * item logging their range, so no need to add one or update an | |
6195 | * existing one. Otherwise we have to log a dir range item. | |
6196 | */ | |
6197 | if (deleted_items) | |
6198 | goto next_batch; | |
6199 | ||
6200 | last_dir_index = last->index; | |
6201 | ASSERT(last_dir_index >= first_dir_index); | |
6202 | /* | |
6203 | * If this range starts right after where the previous one ends, | |
6204 | * then we want to reuse the previous range item and change its | |
6205 | * end offset to the end of this range. This is just to minimize | |
6206 | * leaf space usage, by avoiding adding a new range item. | |
6207 | */ | |
6208 | if (last_range_end != 0 && first_dir_index == last_range_end + 1) | |
6209 | first_dir_index = last_range_start; | |
6210 | ||
6211 | ret = insert_dir_log_key(trans, log, path, key.objectid, | |
6212 | first_dir_index, last_dir_index); | |
6213 | if (ret) | |
6214 | return ret; | |
6215 | ||
6216 | last_range_start = first_dir_index; | |
6217 | last_range_end = last_dir_index; | |
6218 | next_batch: | |
6219 | curr = list_next_entry(last, log_list); | |
6220 | } | |
6221 | ||
6222 | return 0; | |
6223 | } | |
6224 | ||
6225 | static int log_delayed_deletion_items(struct btrfs_trans_handle *trans, | |
6226 | struct btrfs_inode *inode, | |
6227 | struct btrfs_path *path, | |
6228 | const struct list_head *delayed_del_list, | |
6229 | struct btrfs_log_ctx *ctx) | |
6230 | { | |
6231 | /* | |
6232 | * We are deleting dir index items from the log tree or adding range | |
6233 | * items to it. | |
6234 | */ | |
6235 | lockdep_assert_held(&inode->log_mutex); | |
6236 | ||
6237 | if (list_empty(delayed_del_list)) | |
6238 | return 0; | |
6239 | ||
6240 | if (ctx->logged_before) | |
6241 | return log_delayed_deletions_incremental(trans, inode, path, | |
6242 | delayed_del_list, ctx); | |
6243 | ||
6244 | return log_delayed_deletions_full(trans, inode, path, delayed_del_list, | |
6245 | ctx); | |
6246 | } | |
6247 | ||
6248 | /* | |
6249 | * Similar logic as for log_new_dir_dentries(), but it iterates over the delayed | |
6250 | * items instead of the subvolume tree. | |
6251 | */ | |
6252 | static int log_new_delayed_dentries(struct btrfs_trans_handle *trans, | |
6253 | struct btrfs_inode *inode, | |
6254 | const struct list_head *delayed_ins_list, | |
6255 | struct btrfs_log_ctx *ctx) | |
6256 | { | |
6257 | const bool orig_log_new_dentries = ctx->log_new_dentries; | |
6258 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
6259 | struct btrfs_delayed_item *item; | |
6260 | int ret = 0; | |
6261 | ||
6262 | /* | |
6263 | * No need for the log mutex, plus to avoid potential deadlocks or | |
6264 | * lockdep annotations due to nesting of delayed inode mutexes and log | |
6265 | * mutexes. | |
6266 | */ | |
6267 | lockdep_assert_not_held(&inode->log_mutex); | |
6268 | ||
6269 | ASSERT(!ctx->logging_new_delayed_dentries); | |
6270 | ctx->logging_new_delayed_dentries = true; | |
6271 | ||
6272 | list_for_each_entry(item, delayed_ins_list, log_list) { | |
6273 | struct btrfs_dir_item *dir_item; | |
6274 | struct inode *di_inode; | |
6275 | struct btrfs_key key; | |
6276 | int log_mode = LOG_INODE_EXISTS; | |
6277 | ||
6278 | dir_item = (struct btrfs_dir_item *)item->data; | |
6279 | btrfs_disk_key_to_cpu(&key, &dir_item->location); | |
6280 | ||
6281 | if (key.type == BTRFS_ROOT_ITEM_KEY) | |
6282 | continue; | |
6283 | ||
6284 | di_inode = btrfs_iget(fs_info->sb, key.objectid, inode->root); | |
6285 | if (IS_ERR(di_inode)) { | |
6286 | ret = PTR_ERR(di_inode); | |
6287 | break; | |
6288 | } | |
6289 | ||
6290 | if (!need_log_inode(trans, BTRFS_I(di_inode))) { | |
6291 | btrfs_add_delayed_iput(di_inode); | |
6292 | continue; | |
6293 | } | |
6294 | ||
94a48aef | 6295 | if (btrfs_stack_dir_ftype(dir_item) == BTRFS_FT_DIR) |
30b80f3c FM |
6296 | log_mode = LOG_INODE_ALL; |
6297 | ||
6298 | ctx->log_new_dentries = false; | |
6299 | ret = btrfs_log_inode(trans, BTRFS_I(di_inode), log_mode, ctx); | |
6300 | ||
6301 | if (!ret && ctx->log_new_dentries) | |
6302 | ret = log_new_dir_dentries(trans, BTRFS_I(di_inode), ctx); | |
6303 | ||
6304 | btrfs_add_delayed_iput(di_inode); | |
6305 | ||
6306 | if (ret) | |
6307 | break; | |
6308 | } | |
6309 | ||
6310 | ctx->log_new_dentries = orig_log_new_dentries; | |
6311 | ctx->logging_new_delayed_dentries = false; | |
6312 | ||
6313 | return ret; | |
6314 | } | |
6315 | ||
e02119d5 CM |
6316 | /* log a single inode in the tree log. |
6317 | * At least one parent directory for this inode must exist in the tree | |
6318 | * or be logged already. | |
6319 | * | |
6320 | * Any items from this inode changed by the current transaction are copied | |
6321 | * to the log tree. An extra reference is taken on any extents in this | |
6322 | * file, allowing us to avoid a whole pile of corner cases around logging | |
6323 | * blocks that have been removed from the tree. | |
6324 | * | |
6325 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
6326 | * does. | |
6327 | * | |
6328 | * This handles both files and directories. | |
6329 | */ | |
12fcfd22 | 6330 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
90d04510 | 6331 | struct btrfs_inode *inode, |
49dae1bc | 6332 | int inode_only, |
8407f553 | 6333 | struct btrfs_log_ctx *ctx) |
e02119d5 CM |
6334 | { |
6335 | struct btrfs_path *path; | |
6336 | struct btrfs_path *dst_path; | |
6337 | struct btrfs_key min_key; | |
6338 | struct btrfs_key max_key; | |
90d04510 | 6339 | struct btrfs_root *log = inode->root->log_root; |
65faced5 | 6340 | int ret; |
5dc562c5 | 6341 | bool fast_search = false; |
a59108a7 NB |
6342 | u64 ino = btrfs_ino(inode); |
6343 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
1a4bcf47 | 6344 | u64 logged_isize = 0; |
e4545de5 | 6345 | bool need_log_inode_item = true; |
9a8fca62 | 6346 | bool xattrs_logged = false; |
2ac691d8 | 6347 | bool inode_item_dropped = true; |
30b80f3c FM |
6348 | bool full_dir_logging = false; |
6349 | LIST_HEAD(delayed_ins_list); | |
6350 | LIST_HEAD(delayed_del_list); | |
e02119d5 | 6351 | |
e02119d5 | 6352 | path = btrfs_alloc_path(); |
5df67083 TI |
6353 | if (!path) |
6354 | return -ENOMEM; | |
e02119d5 | 6355 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
6356 | if (!dst_path) { |
6357 | btrfs_free_path(path); | |
6358 | return -ENOMEM; | |
6359 | } | |
e02119d5 | 6360 | |
33345d01 | 6361 | min_key.objectid = ino; |
e02119d5 CM |
6362 | min_key.type = BTRFS_INODE_ITEM_KEY; |
6363 | min_key.offset = 0; | |
6364 | ||
33345d01 | 6365 | max_key.objectid = ino; |
12fcfd22 | 6366 | |
12fcfd22 | 6367 | |
5dc562c5 | 6368 | /* today the code can only do partial logging of directories */ |
a59108a7 | 6369 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5269b67e | 6370 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 6371 | &inode->runtime_flags) && |
781feef7 | 6372 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
6373 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
6374 | else | |
6375 | max_key.type = (u8)-1; | |
6376 | max_key.offset = (u64)-1; | |
6377 | ||
30b80f3c FM |
6378 | if (S_ISDIR(inode->vfs_inode.i_mode) && inode_only == LOG_INODE_ALL) |
6379 | full_dir_logging = true; | |
6380 | ||
2c2c452b | 6381 | /* |
30b80f3c FM |
6382 | * If we are logging a directory while we are logging dentries of the |
6383 | * delayed items of some other inode, then we need to flush the delayed | |
6384 | * items of this directory and not log the delayed items directly. This | |
6385 | * is to prevent more than one level of recursion into btrfs_log_inode() | |
6386 | * by having something like this: | |
6387 | * | |
6388 | * $ mkdir -p a/b/c/d/e/f/g/h/... | |
6389 | * $ xfs_io -c "fsync" a | |
6390 | * | |
6391 | * Where all directories in the path did not exist before and are | |
6392 | * created in the current transaction. | |
6393 | * So in such a case we directly log the delayed items of the main | |
6394 | * directory ("a") without flushing them first, while for each of its | |
6395 | * subdirectories we flush their delayed items before logging them. | |
6396 | * This prevents a potential unbounded recursion like this: | |
6397 | * | |
6398 | * btrfs_log_inode() | |
6399 | * log_new_delayed_dentries() | |
6400 | * btrfs_log_inode() | |
6401 | * log_new_delayed_dentries() | |
6402 | * btrfs_log_inode() | |
6403 | * log_new_delayed_dentries() | |
6404 | * (...) | |
6405 | * | |
6406 | * We have thresholds for the maximum number of delayed items to have in | |
6407 | * memory, and once they are hit, the items are flushed asynchronously. | |
6408 | * However the limit is quite high, so lets prevent deep levels of | |
6409 | * recursion to happen by limiting the maximum depth to be 1. | |
2c2c452b | 6410 | */ |
30b80f3c | 6411 | if (full_dir_logging && ctx->logging_new_delayed_dentries) { |
65faced5 FM |
6412 | ret = btrfs_commit_inode_delayed_items(trans, inode); |
6413 | if (ret) | |
f6df27dd | 6414 | goto out; |
16cdcec7 MX |
6415 | } |
6416 | ||
e09d94c9 | 6417 | mutex_lock(&inode->log_mutex); |
e02119d5 | 6418 | |
d0e64a98 FM |
6419 | /* |
6420 | * For symlinks, we must always log their content, which is stored in an | |
6421 | * inline extent, otherwise we could end up with an empty symlink after | |
6422 | * log replay, which is invalid on linux (symlink(2) returns -ENOENT if | |
6423 | * one attempts to create an empty symlink). | |
6424 | * We don't need to worry about flushing delalloc, because when we create | |
6425 | * the inline extent when the symlink is created (we never have delalloc | |
6426 | * for symlinks). | |
6427 | */ | |
6428 | if (S_ISLNK(inode->vfs_inode.i_mode)) | |
6429 | inode_only = LOG_INODE_ALL; | |
6430 | ||
0f8ce498 FM |
6431 | /* |
6432 | * Before logging the inode item, cache the value returned by | |
6433 | * inode_logged(), because after that we have the need to figure out if | |
6434 | * the inode was previously logged in this transaction. | |
6435 | */ | |
6436 | ret = inode_logged(trans, inode, path); | |
65faced5 | 6437 | if (ret < 0) |
0f8ce498 | 6438 | goto out_unlock; |
0f8ce498 | 6439 | ctx->logged_before = (ret == 1); |
65faced5 | 6440 | ret = 0; |
0f8ce498 | 6441 | |
64d6b281 FM |
6442 | /* |
6443 | * This is for cases where logging a directory could result in losing a | |
6444 | * a file after replaying the log. For example, if we move a file from a | |
6445 | * directory A to a directory B, then fsync directory A, we have no way | |
6446 | * to known the file was moved from A to B, so logging just A would | |
6447 | * result in losing the file after a log replay. | |
6448 | */ | |
30b80f3c | 6449 | if (full_dir_logging && inode->last_unlink_trans >= trans->transid) { |
64d6b281 | 6450 | btrfs_set_log_full_commit(trans); |
f31f09f6 | 6451 | ret = BTRFS_LOG_FORCE_COMMIT; |
64d6b281 FM |
6452 | goto out_unlock; |
6453 | } | |
6454 | ||
e02119d5 CM |
6455 | /* |
6456 | * a brute force approach to making sure we get the most uptodate | |
6457 | * copies of everything. | |
6458 | */ | |
a59108a7 | 6459 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
ab12313a | 6460 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags); |
0f8ce498 FM |
6461 | if (ctx->logged_before) |
6462 | ret = drop_inode_items(trans, log, path, inode, | |
04fc7d51 | 6463 | BTRFS_XATTR_ITEM_KEY); |
e02119d5 | 6464 | } else { |
0f8ce498 | 6465 | if (inode_only == LOG_INODE_EXISTS && ctx->logged_before) { |
1a4bcf47 FM |
6466 | /* |
6467 | * Make sure the new inode item we write to the log has | |
6468 | * the same isize as the current one (if it exists). | |
6469 | * This is necessary to prevent data loss after log | |
6470 | * replay, and also to prevent doing a wrong expanding | |
6471 | * truncate - for e.g. create file, write 4K into offset | |
6472 | * 0, fsync, write 4K into offset 4096, add hard link, | |
6473 | * fsync some other file (to sync log), power fail - if | |
6474 | * we use the inode's current i_size, after log replay | |
6475 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
6476 | * (zeroes), as if an expanding truncate happened, | |
6477 | * instead of getting a file of 4Kb only. | |
6478 | */ | |
65faced5 FM |
6479 | ret = logged_inode_size(log, inode, path, &logged_isize); |
6480 | if (ret) | |
1a4bcf47 FM |
6481 | goto out_unlock; |
6482 | } | |
a742994a | 6483 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 6484 | &inode->runtime_flags)) { |
a742994a | 6485 | if (inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 6486 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
0f8ce498 FM |
6487 | if (ctx->logged_before) |
6488 | ret = drop_inode_items(trans, log, path, | |
6489 | inode, max_key.type); | |
a742994a FM |
6490 | } else { |
6491 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a59108a7 | 6492 | &inode->runtime_flags); |
a742994a | 6493 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 6494 | &inode->runtime_flags); |
0f8ce498 | 6495 | if (ctx->logged_before) |
4934a815 FM |
6496 | ret = truncate_inode_items(trans, log, |
6497 | inode, 0, 0); | |
a742994a | 6498 | } |
4f764e51 | 6499 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 6500 | &inode->runtime_flags) || |
6cfab851 | 6501 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 6502 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 6503 | fast_search = true; |
4f764e51 | 6504 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
0f8ce498 FM |
6505 | if (ctx->logged_before) |
6506 | ret = drop_inode_items(trans, log, path, inode, | |
6507 | max_key.type); | |
a95249b3 JB |
6508 | } else { |
6509 | if (inode_only == LOG_INODE_ALL) | |
6510 | fast_search = true; | |
2ac691d8 | 6511 | inode_item_dropped = false; |
a95249b3 | 6512 | goto log_extents; |
5dc562c5 | 6513 | } |
a95249b3 | 6514 | |
e02119d5 | 6515 | } |
65faced5 | 6516 | if (ret) |
4a500fd1 | 6517 | goto out_unlock; |
e02119d5 | 6518 | |
30b80f3c FM |
6519 | /* |
6520 | * If we are logging a directory in full mode, collect the delayed items | |
6521 | * before iterating the subvolume tree, so that we don't miss any new | |
6522 | * dir index items in case they get flushed while or right after we are | |
6523 | * iterating the subvolume tree. | |
6524 | */ | |
6525 | if (full_dir_logging && !ctx->logging_new_delayed_dentries) | |
6526 | btrfs_log_get_delayed_items(inode, &delayed_ins_list, | |
6527 | &delayed_del_list); | |
6528 | ||
65faced5 | 6529 | ret = copy_inode_items_to_log(trans, inode, &min_key, &max_key, |
da447009 | 6530 | path, dst_path, logged_isize, |
e09d94c9 | 6531 | inode_only, ctx, |
7af59743 | 6532 | &need_log_inode_item); |
65faced5 | 6533 | if (ret) |
da447009 | 6534 | goto out_unlock; |
5dc562c5 | 6535 | |
36283bf7 FM |
6536 | btrfs_release_path(path); |
6537 | btrfs_release_path(dst_path); | |
65faced5 FM |
6538 | ret = btrfs_log_all_xattrs(trans, inode, path, dst_path); |
6539 | if (ret) | |
36283bf7 | 6540 | goto out_unlock; |
9a8fca62 | 6541 | xattrs_logged = true; |
a89ca6f2 FM |
6542 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
6543 | btrfs_release_path(path); | |
6544 | btrfs_release_path(dst_path); | |
65faced5 FM |
6545 | ret = btrfs_log_holes(trans, inode, path); |
6546 | if (ret) | |
a89ca6f2 FM |
6547 | goto out_unlock; |
6548 | } | |
a95249b3 | 6549 | log_extents: |
f3b15ccd JB |
6550 | btrfs_release_path(path); |
6551 | btrfs_release_path(dst_path); | |
e4545de5 | 6552 | if (need_log_inode_item) { |
65faced5 FM |
6553 | ret = log_inode_item(trans, log, dst_path, inode, inode_item_dropped); |
6554 | if (ret) | |
b590b839 FM |
6555 | goto out_unlock; |
6556 | /* | |
6557 | * If we are doing a fast fsync and the inode was logged before | |
6558 | * in this transaction, we don't need to log the xattrs because | |
6559 | * they were logged before. If xattrs were added, changed or | |
6560 | * deleted since the last time we logged the inode, then we have | |
6561 | * already logged them because the inode had the runtime flag | |
6562 | * BTRFS_INODE_COPY_EVERYTHING set. | |
6563 | */ | |
6564 | if (!xattrs_logged && inode->logged_trans < trans->transid) { | |
65faced5 FM |
6565 | ret = btrfs_log_all_xattrs(trans, inode, path, dst_path); |
6566 | if (ret) | |
b590b839 | 6567 | goto out_unlock; |
9a8fca62 FM |
6568 | btrfs_release_path(path); |
6569 | } | |
e4545de5 | 6570 | } |
5dc562c5 | 6571 | if (fast_search) { |
90d04510 | 6572 | ret = btrfs_log_changed_extents(trans, inode, dst_path, ctx); |
65faced5 | 6573 | if (ret) |
5dc562c5 | 6574 | goto out_unlock; |
d006a048 | 6575 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
6576 | struct extent_map *em, *n; |
6577 | ||
49dae1bc | 6578 | write_lock(&em_tree->lock); |
48778179 FM |
6579 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, list) |
6580 | list_del_init(&em->list); | |
49dae1bc | 6581 | write_unlock(&em_tree->lock); |
5dc562c5 JB |
6582 | } |
6583 | ||
30b80f3c | 6584 | if (full_dir_logging) { |
90d04510 | 6585 | ret = log_directory_changes(trans, inode, path, dst_path, ctx); |
65faced5 | 6586 | if (ret) |
4a500fd1 | 6587 | goto out_unlock; |
30b80f3c FM |
6588 | ret = log_delayed_insertion_items(trans, inode, path, |
6589 | &delayed_ins_list, ctx); | |
6590 | if (ret) | |
6591 | goto out_unlock; | |
6592 | ret = log_delayed_deletion_items(trans, inode, path, | |
6593 | &delayed_del_list, ctx); | |
6594 | if (ret) | |
6595 | goto out_unlock; | |
e02119d5 | 6596 | } |
49dae1bc | 6597 | |
130341be FM |
6598 | spin_lock(&inode->lock); |
6599 | inode->logged_trans = trans->transid; | |
d1d832a0 | 6600 | /* |
130341be FM |
6601 | * Don't update last_log_commit if we logged that an inode exists. |
6602 | * We do this for three reasons: | |
6603 | * | |
6604 | * 1) We might have had buffered writes to this inode that were | |
6605 | * flushed and had their ordered extents completed in this | |
6606 | * transaction, but we did not previously log the inode with | |
6607 | * LOG_INODE_ALL. Later the inode was evicted and after that | |
6608 | * it was loaded again and this LOG_INODE_EXISTS log operation | |
6609 | * happened. We must make sure that if an explicit fsync against | |
6610 | * the inode is performed later, it logs the new extents, an | |
6611 | * updated inode item, etc, and syncs the log. The same logic | |
6612 | * applies to direct IO writes instead of buffered writes. | |
6613 | * | |
6614 | * 2) When we log the inode with LOG_INODE_EXISTS, its inode item | |
6615 | * is logged with an i_size of 0 or whatever value was logged | |
6616 | * before. If later the i_size of the inode is increased by a | |
6617 | * truncate operation, the log is synced through an fsync of | |
6618 | * some other inode and then finally an explicit fsync against | |
6619 | * this inode is made, we must make sure this fsync logs the | |
6620 | * inode with the new i_size, the hole between old i_size and | |
6621 | * the new i_size, and syncs the log. | |
6622 | * | |
6623 | * 3) If we are logging that an ancestor inode exists as part of | |
6624 | * logging a new name from a link or rename operation, don't update | |
6625 | * its last_log_commit - otherwise if an explicit fsync is made | |
6626 | * against an ancestor, the fsync considers the inode in the log | |
6627 | * and doesn't sync the log, resulting in the ancestor missing after | |
6628 | * a power failure unless the log was synced as part of an fsync | |
6629 | * against any other unrelated inode. | |
d1d832a0 | 6630 | */ |
130341be FM |
6631 | if (inode_only != LOG_INODE_EXISTS) |
6632 | inode->last_log_commit = inode->last_sub_trans; | |
6633 | spin_unlock(&inode->lock); | |
23e3337f FM |
6634 | |
6635 | /* | |
6636 | * Reset the last_reflink_trans so that the next fsync does not need to | |
6637 | * go through the slower path when logging extents and their checksums. | |
6638 | */ | |
6639 | if (inode_only == LOG_INODE_ALL) | |
6640 | inode->last_reflink_trans = 0; | |
6641 | ||
4a500fd1 | 6642 | out_unlock: |
a59108a7 | 6643 | mutex_unlock(&inode->log_mutex); |
f6df27dd | 6644 | out: |
e02119d5 CM |
6645 | btrfs_free_path(path); |
6646 | btrfs_free_path(dst_path); | |
0f8ce498 | 6647 | |
e09d94c9 FM |
6648 | if (ret) |
6649 | free_conflicting_inodes(ctx); | |
6650 | else | |
6651 | ret = log_conflicting_inodes(trans, inode->root, ctx); | |
0f8ce498 | 6652 | |
30b80f3c FM |
6653 | if (full_dir_logging && !ctx->logging_new_delayed_dentries) { |
6654 | if (!ret) | |
6655 | ret = log_new_delayed_dentries(trans, inode, | |
6656 | &delayed_ins_list, ctx); | |
6657 | ||
6658 | btrfs_log_put_delayed_items(inode, &delayed_ins_list, | |
6659 | &delayed_del_list); | |
6660 | } | |
6661 | ||
65faced5 | 6662 | return ret; |
e02119d5 CM |
6663 | } |
6664 | ||
18aa0922 | 6665 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
d0a0b78d | 6666 | struct btrfs_inode *inode, |
18aa0922 FM |
6667 | struct btrfs_log_ctx *ctx) |
6668 | { | |
3ffbd68c | 6669 | struct btrfs_fs_info *fs_info = trans->fs_info; |
18aa0922 FM |
6670 | int ret; |
6671 | struct btrfs_path *path; | |
6672 | struct btrfs_key key; | |
d0a0b78d NB |
6673 | struct btrfs_root *root = inode->root; |
6674 | const u64 ino = btrfs_ino(inode); | |
18aa0922 FM |
6675 | |
6676 | path = btrfs_alloc_path(); | |
6677 | if (!path) | |
6678 | return -ENOMEM; | |
6679 | path->skip_locking = 1; | |
6680 | path->search_commit_root = 1; | |
6681 | ||
6682 | key.objectid = ino; | |
6683 | key.type = BTRFS_INODE_REF_KEY; | |
6684 | key.offset = 0; | |
6685 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6686 | if (ret < 0) | |
6687 | goto out; | |
6688 | ||
6689 | while (true) { | |
6690 | struct extent_buffer *leaf = path->nodes[0]; | |
6691 | int slot = path->slots[0]; | |
6692 | u32 cur_offset = 0; | |
6693 | u32 item_size; | |
6694 | unsigned long ptr; | |
6695 | ||
6696 | if (slot >= btrfs_header_nritems(leaf)) { | |
6697 | ret = btrfs_next_leaf(root, path); | |
6698 | if (ret < 0) | |
6699 | goto out; | |
6700 | else if (ret > 0) | |
6701 | break; | |
6702 | continue; | |
6703 | } | |
6704 | ||
6705 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
6706 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
6707 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
6708 | break; | |
6709 | ||
3212fa14 | 6710 | item_size = btrfs_item_size(leaf, slot); |
18aa0922 FM |
6711 | ptr = btrfs_item_ptr_offset(leaf, slot); |
6712 | while (cur_offset < item_size) { | |
6713 | struct btrfs_key inode_key; | |
6714 | struct inode *dir_inode; | |
6715 | ||
6716 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
6717 | inode_key.offset = 0; | |
6718 | ||
6719 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
6720 | struct btrfs_inode_extref *extref; | |
6721 | ||
6722 | extref = (struct btrfs_inode_extref *) | |
6723 | (ptr + cur_offset); | |
6724 | inode_key.objectid = btrfs_inode_extref_parent( | |
6725 | leaf, extref); | |
6726 | cur_offset += sizeof(*extref); | |
6727 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
6728 | extref); | |
6729 | } else { | |
6730 | inode_key.objectid = key.offset; | |
6731 | cur_offset = item_size; | |
6732 | } | |
6733 | ||
0202e83f DS |
6734 | dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid, |
6735 | root); | |
0f375eed FM |
6736 | /* |
6737 | * If the parent inode was deleted, return an error to | |
6738 | * fallback to a transaction commit. This is to prevent | |
6739 | * getting an inode that was moved from one parent A to | |
6740 | * a parent B, got its former parent A deleted and then | |
6741 | * it got fsync'ed, from existing at both parents after | |
6742 | * a log replay (and the old parent still existing). | |
6743 | * Example: | |
6744 | * | |
6745 | * mkdir /mnt/A | |
6746 | * mkdir /mnt/B | |
6747 | * touch /mnt/B/bar | |
6748 | * sync | |
6749 | * mv /mnt/B/bar /mnt/A/bar | |
6750 | * mv -T /mnt/A /mnt/B | |
6751 | * fsync /mnt/B/bar | |
6752 | * <power fail> | |
6753 | * | |
6754 | * If we ignore the old parent B which got deleted, | |
6755 | * after a log replay we would have file bar linked | |
6756 | * at both parents and the old parent B would still | |
6757 | * exist. | |
6758 | */ | |
6759 | if (IS_ERR(dir_inode)) { | |
6760 | ret = PTR_ERR(dir_inode); | |
6761 | goto out; | |
6762 | } | |
18aa0922 | 6763 | |
3e6a86a1 FM |
6764 | if (!need_log_inode(trans, BTRFS_I(dir_inode))) { |
6765 | btrfs_add_delayed_iput(dir_inode); | |
6766 | continue; | |
6767 | } | |
6768 | ||
289cffcb | 6769 | ctx->log_new_dentries = false; |
90d04510 | 6770 | ret = btrfs_log_inode(trans, BTRFS_I(dir_inode), |
48778179 | 6771 | LOG_INODE_ALL, ctx); |
289cffcb | 6772 | if (!ret && ctx->log_new_dentries) |
8786a6d7 | 6773 | ret = log_new_dir_dentries(trans, |
f85b7379 | 6774 | BTRFS_I(dir_inode), ctx); |
410f954c | 6775 | btrfs_add_delayed_iput(dir_inode); |
18aa0922 FM |
6776 | if (ret) |
6777 | goto out; | |
6778 | } | |
6779 | path->slots[0]++; | |
6780 | } | |
6781 | ret = 0; | |
6782 | out: | |
6783 | btrfs_free_path(path); | |
6784 | return ret; | |
6785 | } | |
6786 | ||
b8aa330d FM |
6787 | static int log_new_ancestors(struct btrfs_trans_handle *trans, |
6788 | struct btrfs_root *root, | |
6789 | struct btrfs_path *path, | |
6790 | struct btrfs_log_ctx *ctx) | |
6791 | { | |
6792 | struct btrfs_key found_key; | |
6793 | ||
6794 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
6795 | ||
6796 | while (true) { | |
6797 | struct btrfs_fs_info *fs_info = root->fs_info; | |
b8aa330d FM |
6798 | struct extent_buffer *leaf = path->nodes[0]; |
6799 | int slot = path->slots[0]; | |
6800 | struct btrfs_key search_key; | |
6801 | struct inode *inode; | |
0202e83f | 6802 | u64 ino; |
b8aa330d FM |
6803 | int ret = 0; |
6804 | ||
6805 | btrfs_release_path(path); | |
6806 | ||
0202e83f DS |
6807 | ino = found_key.offset; |
6808 | ||
b8aa330d FM |
6809 | search_key.objectid = found_key.offset; |
6810 | search_key.type = BTRFS_INODE_ITEM_KEY; | |
6811 | search_key.offset = 0; | |
0202e83f | 6812 | inode = btrfs_iget(fs_info->sb, ino, root); |
b8aa330d FM |
6813 | if (IS_ERR(inode)) |
6814 | return PTR_ERR(inode); | |
6815 | ||
ab12313a FM |
6816 | if (BTRFS_I(inode)->generation >= trans->transid && |
6817 | need_log_inode(trans, BTRFS_I(inode))) | |
90d04510 | 6818 | ret = btrfs_log_inode(trans, BTRFS_I(inode), |
48778179 | 6819 | LOG_INODE_EXISTS, ctx); |
410f954c | 6820 | btrfs_add_delayed_iput(inode); |
b8aa330d FM |
6821 | if (ret) |
6822 | return ret; | |
6823 | ||
6824 | if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID) | |
6825 | break; | |
6826 | ||
6827 | search_key.type = BTRFS_INODE_REF_KEY; | |
6828 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
6829 | if (ret < 0) | |
6830 | return ret; | |
6831 | ||
6832 | leaf = path->nodes[0]; | |
6833 | slot = path->slots[0]; | |
6834 | if (slot >= btrfs_header_nritems(leaf)) { | |
6835 | ret = btrfs_next_leaf(root, path); | |
6836 | if (ret < 0) | |
6837 | return ret; | |
6838 | else if (ret > 0) | |
6839 | return -ENOENT; | |
6840 | leaf = path->nodes[0]; | |
6841 | slot = path->slots[0]; | |
6842 | } | |
6843 | ||
6844 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
6845 | if (found_key.objectid != search_key.objectid || | |
6846 | found_key.type != BTRFS_INODE_REF_KEY) | |
6847 | return -ENOENT; | |
6848 | } | |
6849 | return 0; | |
6850 | } | |
6851 | ||
6852 | static int log_new_ancestors_fast(struct btrfs_trans_handle *trans, | |
6853 | struct btrfs_inode *inode, | |
6854 | struct dentry *parent, | |
6855 | struct btrfs_log_ctx *ctx) | |
6856 | { | |
6857 | struct btrfs_root *root = inode->root; | |
b8aa330d FM |
6858 | struct dentry *old_parent = NULL; |
6859 | struct super_block *sb = inode->vfs_inode.i_sb; | |
6860 | int ret = 0; | |
6861 | ||
6862 | while (true) { | |
6863 | if (!parent || d_really_is_negative(parent) || | |
6864 | sb != parent->d_sb) | |
6865 | break; | |
6866 | ||
6867 | inode = BTRFS_I(d_inode(parent)); | |
6868 | if (root != inode->root) | |
6869 | break; | |
6870 | ||
ab12313a FM |
6871 | if (inode->generation >= trans->transid && |
6872 | need_log_inode(trans, inode)) { | |
90d04510 | 6873 | ret = btrfs_log_inode(trans, inode, |
48778179 | 6874 | LOG_INODE_EXISTS, ctx); |
b8aa330d FM |
6875 | if (ret) |
6876 | break; | |
6877 | } | |
6878 | if (IS_ROOT(parent)) | |
6879 | break; | |
6880 | ||
6881 | parent = dget_parent(parent); | |
6882 | dput(old_parent); | |
6883 | old_parent = parent; | |
6884 | } | |
6885 | dput(old_parent); | |
6886 | ||
6887 | return ret; | |
6888 | } | |
6889 | ||
6890 | static int log_all_new_ancestors(struct btrfs_trans_handle *trans, | |
6891 | struct btrfs_inode *inode, | |
6892 | struct dentry *parent, | |
6893 | struct btrfs_log_ctx *ctx) | |
6894 | { | |
6895 | struct btrfs_root *root = inode->root; | |
6896 | const u64 ino = btrfs_ino(inode); | |
6897 | struct btrfs_path *path; | |
6898 | struct btrfs_key search_key; | |
6899 | int ret; | |
6900 | ||
6901 | /* | |
6902 | * For a single hard link case, go through a fast path that does not | |
6903 | * need to iterate the fs/subvolume tree. | |
6904 | */ | |
6905 | if (inode->vfs_inode.i_nlink < 2) | |
6906 | return log_new_ancestors_fast(trans, inode, parent, ctx); | |
6907 | ||
6908 | path = btrfs_alloc_path(); | |
6909 | if (!path) | |
6910 | return -ENOMEM; | |
6911 | ||
6912 | search_key.objectid = ino; | |
6913 | search_key.type = BTRFS_INODE_REF_KEY; | |
6914 | search_key.offset = 0; | |
6915 | again: | |
6916 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
6917 | if (ret < 0) | |
6918 | goto out; | |
6919 | if (ret == 0) | |
6920 | path->slots[0]++; | |
6921 | ||
6922 | while (true) { | |
6923 | struct extent_buffer *leaf = path->nodes[0]; | |
6924 | int slot = path->slots[0]; | |
6925 | struct btrfs_key found_key; | |
6926 | ||
6927 | if (slot >= btrfs_header_nritems(leaf)) { | |
6928 | ret = btrfs_next_leaf(root, path); | |
6929 | if (ret < 0) | |
6930 | goto out; | |
6931 | else if (ret > 0) | |
6932 | break; | |
6933 | continue; | |
6934 | } | |
6935 | ||
6936 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
6937 | if (found_key.objectid != ino || | |
6938 | found_key.type > BTRFS_INODE_EXTREF_KEY) | |
6939 | break; | |
6940 | ||
6941 | /* | |
6942 | * Don't deal with extended references because they are rare | |
6943 | * cases and too complex to deal with (we would need to keep | |
6944 | * track of which subitem we are processing for each item in | |
6945 | * this loop, etc). So just return some error to fallback to | |
6946 | * a transaction commit. | |
6947 | */ | |
6948 | if (found_key.type == BTRFS_INODE_EXTREF_KEY) { | |
6949 | ret = -EMLINK; | |
6950 | goto out; | |
6951 | } | |
6952 | ||
6953 | /* | |
6954 | * Logging ancestors needs to do more searches on the fs/subvol | |
6955 | * tree, so it releases the path as needed to avoid deadlocks. | |
6956 | * Keep track of the last inode ref key and resume from that key | |
6957 | * after logging all new ancestors for the current hard link. | |
6958 | */ | |
6959 | memcpy(&search_key, &found_key, sizeof(search_key)); | |
6960 | ||
6961 | ret = log_new_ancestors(trans, root, path, ctx); | |
6962 | if (ret) | |
6963 | goto out; | |
6964 | btrfs_release_path(path); | |
6965 | goto again; | |
6966 | } | |
6967 | ret = 0; | |
6968 | out: | |
6969 | btrfs_free_path(path); | |
6970 | return ret; | |
6971 | } | |
6972 | ||
e02119d5 CM |
6973 | /* |
6974 | * helper function around btrfs_log_inode to make sure newly created | |
6975 | * parent directories also end up in the log. A minimal inode and backref | |
6976 | * only logging is done of any parent directories that are older than | |
6977 | * the last committed transaction | |
6978 | */ | |
48a3b636 | 6979 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
19df27a9 | 6980 | struct btrfs_inode *inode, |
49dae1bc | 6981 | struct dentry *parent, |
41a1eada | 6982 | int inode_only, |
8b050d35 | 6983 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6984 | { |
f882274b | 6985 | struct btrfs_root *root = inode->root; |
0b246afa | 6986 | struct btrfs_fs_info *fs_info = root->fs_info; |
12fcfd22 | 6987 | int ret = 0; |
2f2ff0ee | 6988 | bool log_dentries = false; |
12fcfd22 | 6989 | |
0b246afa | 6990 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
f31f09f6 | 6991 | ret = BTRFS_LOG_FORCE_COMMIT; |
3a5e1404 SW |
6992 | goto end_no_trans; |
6993 | } | |
6994 | ||
f882274b | 6995 | if (btrfs_root_refs(&root->root_item) == 0) { |
f31f09f6 | 6996 | ret = BTRFS_LOG_FORCE_COMMIT; |
76dda93c YZ |
6997 | goto end_no_trans; |
6998 | } | |
6999 | ||
f2d72f42 FM |
7000 | /* |
7001 | * Skip already logged inodes or inodes corresponding to tmpfiles | |
7002 | * (since logging them is pointless, a link count of 0 means they | |
7003 | * will never be accessible). | |
7004 | */ | |
626e9f41 FM |
7005 | if ((btrfs_inode_in_log(inode, trans->transid) && |
7006 | list_empty(&ctx->ordered_extents)) || | |
f2d72f42 | 7007 | inode->vfs_inode.i_nlink == 0) { |
257c62e1 CM |
7008 | ret = BTRFS_NO_LOG_SYNC; |
7009 | goto end_no_trans; | |
7010 | } | |
7011 | ||
8b050d35 | 7012 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 7013 | if (ret) |
e87ac136 | 7014 | goto end_no_trans; |
e02119d5 | 7015 | |
90d04510 | 7016 | ret = btrfs_log_inode(trans, inode, inode_only, ctx); |
4a500fd1 YZ |
7017 | if (ret) |
7018 | goto end_trans; | |
12fcfd22 | 7019 | |
af4176b4 CM |
7020 | /* |
7021 | * for regular files, if its inode is already on disk, we don't | |
7022 | * have to worry about the parents at all. This is because | |
7023 | * we can use the last_unlink_trans field to record renames | |
7024 | * and other fun in this file. | |
7025 | */ | |
19df27a9 | 7026 | if (S_ISREG(inode->vfs_inode.i_mode) && |
47d3db41 FM |
7027 | inode->generation < trans->transid && |
7028 | inode->last_unlink_trans < trans->transid) { | |
4a500fd1 YZ |
7029 | ret = 0; |
7030 | goto end_trans; | |
7031 | } | |
af4176b4 | 7032 | |
289cffcb | 7033 | if (S_ISDIR(inode->vfs_inode.i_mode) && ctx->log_new_dentries) |
2f2ff0ee FM |
7034 | log_dentries = true; |
7035 | ||
18aa0922 | 7036 | /* |
01327610 | 7037 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
7038 | * inodes are fully logged. This is to prevent leaving dangling |
7039 | * directory index entries in directories that were our parents but are | |
7040 | * not anymore. Not doing this results in old parent directory being | |
7041 | * impossible to delete after log replay (rmdir will always fail with | |
7042 | * error -ENOTEMPTY). | |
7043 | * | |
7044 | * Example 1: | |
7045 | * | |
7046 | * mkdir testdir | |
7047 | * touch testdir/foo | |
7048 | * ln testdir/foo testdir/bar | |
7049 | * sync | |
7050 | * unlink testdir/bar | |
7051 | * xfs_io -c fsync testdir/foo | |
7052 | * <power failure> | |
7053 | * mount fs, triggers log replay | |
7054 | * | |
7055 | * If we don't log the parent directory (testdir), after log replay the | |
7056 | * directory still has an entry pointing to the file inode using the bar | |
7057 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
7058 | * the file inode has a link count of 1. | |
7059 | * | |
7060 | * Example 2: | |
7061 | * | |
7062 | * mkdir testdir | |
7063 | * touch foo | |
7064 | * ln foo testdir/foo2 | |
7065 | * ln foo testdir/foo3 | |
7066 | * sync | |
7067 | * unlink testdir/foo3 | |
7068 | * xfs_io -c fsync foo | |
7069 | * <power failure> | |
7070 | * mount fs, triggers log replay | |
7071 | * | |
7072 | * Similar as the first example, after log replay the parent directory | |
7073 | * testdir still has an entry pointing to the inode file with name foo3 | |
7074 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
7075 | * and has a link count of 2. | |
7076 | */ | |
47d3db41 | 7077 | if (inode->last_unlink_trans >= trans->transid) { |
b8aa330d | 7078 | ret = btrfs_log_all_parents(trans, inode, ctx); |
18aa0922 FM |
7079 | if (ret) |
7080 | goto end_trans; | |
7081 | } | |
7082 | ||
b8aa330d FM |
7083 | ret = log_all_new_ancestors(trans, inode, parent, ctx); |
7084 | if (ret) | |
41bd6067 | 7085 | goto end_trans; |
76dda93c | 7086 | |
2f2ff0ee | 7087 | if (log_dentries) |
8786a6d7 | 7088 | ret = log_new_dir_dentries(trans, inode, ctx); |
2f2ff0ee FM |
7089 | else |
7090 | ret = 0; | |
4a500fd1 YZ |
7091 | end_trans: |
7092 | if (ret < 0) { | |
90787766 | 7093 | btrfs_set_log_full_commit(trans); |
f31f09f6 | 7094 | ret = BTRFS_LOG_FORCE_COMMIT; |
4a500fd1 | 7095 | } |
8b050d35 MX |
7096 | |
7097 | if (ret) | |
7098 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
7099 | btrfs_end_log_trans(root); |
7100 | end_no_trans: | |
7101 | return ret; | |
e02119d5 CM |
7102 | } |
7103 | ||
7104 | /* | |
7105 | * it is not safe to log dentry if the chunk root has added new | |
7106 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
7107 | * If this returns 1, you must commit the transaction to safely get your | |
7108 | * data on disk. | |
7109 | */ | |
7110 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
e5b84f7a | 7111 | struct dentry *dentry, |
8b050d35 | 7112 | struct btrfs_log_ctx *ctx) |
e02119d5 | 7113 | { |
6a912213 JB |
7114 | struct dentry *parent = dget_parent(dentry); |
7115 | int ret; | |
7116 | ||
f882274b | 7117 | ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent, |
48778179 | 7118 | LOG_INODE_ALL, ctx); |
6a912213 JB |
7119 | dput(parent); |
7120 | ||
7121 | return ret; | |
e02119d5 CM |
7122 | } |
7123 | ||
7124 | /* | |
7125 | * should be called during mount to recover any replay any log trees | |
7126 | * from the FS | |
7127 | */ | |
7128 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
7129 | { | |
7130 | int ret; | |
7131 | struct btrfs_path *path; | |
7132 | struct btrfs_trans_handle *trans; | |
7133 | struct btrfs_key key; | |
7134 | struct btrfs_key found_key; | |
e02119d5 CM |
7135 | struct btrfs_root *log; |
7136 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
7137 | struct walk_control wc = { | |
7138 | .process_func = process_one_buffer, | |
430a6626 | 7139 | .stage = LOG_WALK_PIN_ONLY, |
e02119d5 CM |
7140 | }; |
7141 | ||
e02119d5 | 7142 | path = btrfs_alloc_path(); |
db5b493a TI |
7143 | if (!path) |
7144 | return -ENOMEM; | |
7145 | ||
afcdd129 | 7146 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 7147 | |
4a500fd1 | 7148 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
7149 | if (IS_ERR(trans)) { |
7150 | ret = PTR_ERR(trans); | |
7151 | goto error; | |
7152 | } | |
e02119d5 CM |
7153 | |
7154 | wc.trans = trans; | |
7155 | wc.pin = 1; | |
7156 | ||
db5b493a | 7157 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 7158 | if (ret) { |
ba51e2a1 | 7159 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
7160 | goto error; |
7161 | } | |
e02119d5 CM |
7162 | |
7163 | again: | |
7164 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
7165 | key.offset = (u64)-1; | |
962a298f | 7166 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 7167 | |
d397712b | 7168 | while (1) { |
e02119d5 | 7169 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
7170 | |
7171 | if (ret < 0) { | |
ba51e2a1 | 7172 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
7173 | goto error; |
7174 | } | |
e02119d5 CM |
7175 | if (ret > 0) { |
7176 | if (path->slots[0] == 0) | |
7177 | break; | |
7178 | path->slots[0]--; | |
7179 | } | |
7180 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
7181 | path->slots[0]); | |
b3b4aa74 | 7182 | btrfs_release_path(path); |
e02119d5 CM |
7183 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
7184 | break; | |
7185 | ||
62a2c73e | 7186 | log = btrfs_read_tree_root(log_root_tree, &found_key); |
79787eaa JM |
7187 | if (IS_ERR(log)) { |
7188 | ret = PTR_ERR(log); | |
ba51e2a1 | 7189 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
7190 | goto error; |
7191 | } | |
e02119d5 | 7192 | |
56e9357a DS |
7193 | wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset, |
7194 | true); | |
79787eaa JM |
7195 | if (IS_ERR(wc.replay_dest)) { |
7196 | ret = PTR_ERR(wc.replay_dest); | |
9bc574de JB |
7197 | |
7198 | /* | |
7199 | * We didn't find the subvol, likely because it was | |
7200 | * deleted. This is ok, simply skip this log and go to | |
7201 | * the next one. | |
7202 | * | |
7203 | * We need to exclude the root because we can't have | |
7204 | * other log replays overwriting this log as we'll read | |
7205 | * it back in a few more times. This will keep our | |
7206 | * block from being modified, and we'll just bail for | |
7207 | * each subsequent pass. | |
7208 | */ | |
7209 | if (ret == -ENOENT) | |
9fce5704 | 7210 | ret = btrfs_pin_extent_for_log_replay(trans, |
9bc574de JB |
7211 | log->node->start, |
7212 | log->node->len); | |
00246528 | 7213 | btrfs_put_root(log); |
9bc574de JB |
7214 | |
7215 | if (!ret) | |
7216 | goto next; | |
ba51e2a1 | 7217 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
7218 | goto error; |
7219 | } | |
e02119d5 | 7220 | |
07d400a6 | 7221 | wc.replay_dest->log_root = log; |
2002ae11 JB |
7222 | ret = btrfs_record_root_in_trans(trans, wc.replay_dest); |
7223 | if (ret) | |
7224 | /* The loop needs to continue due to the root refs */ | |
ba51e2a1 | 7225 | btrfs_abort_transaction(trans, ret); |
2002ae11 JB |
7226 | else |
7227 | ret = walk_log_tree(trans, log, &wc); | |
e02119d5 | 7228 | |
b50c6e25 | 7229 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
7230 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
7231 | path); | |
ba51e2a1 JB |
7232 | if (ret) |
7233 | btrfs_abort_transaction(trans, ret); | |
e02119d5 CM |
7234 | } |
7235 | ||
900c9981 LB |
7236 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
7237 | struct btrfs_root *root = wc.replay_dest; | |
7238 | ||
7239 | btrfs_release_path(path); | |
7240 | ||
7241 | /* | |
7242 | * We have just replayed everything, and the highest | |
7243 | * objectid of fs roots probably has changed in case | |
7244 | * some inode_item's got replayed. | |
7245 | * | |
7246 | * root->objectid_mutex is not acquired as log replay | |
7247 | * could only happen during mount. | |
7248 | */ | |
453e4873 | 7249 | ret = btrfs_init_root_free_objectid(root); |
ba51e2a1 JB |
7250 | if (ret) |
7251 | btrfs_abort_transaction(trans, ret); | |
900c9981 LB |
7252 | } |
7253 | ||
07d400a6 | 7254 | wc.replay_dest->log_root = NULL; |
00246528 | 7255 | btrfs_put_root(wc.replay_dest); |
00246528 | 7256 | btrfs_put_root(log); |
e02119d5 | 7257 | |
b50c6e25 JB |
7258 | if (ret) |
7259 | goto error; | |
9bc574de | 7260 | next: |
e02119d5 CM |
7261 | if (found_key.offset == 0) |
7262 | break; | |
9bc574de | 7263 | key.offset = found_key.offset - 1; |
e02119d5 | 7264 | } |
b3b4aa74 | 7265 | btrfs_release_path(path); |
e02119d5 CM |
7266 | |
7267 | /* step one is to pin it all, step two is to replay just inodes */ | |
7268 | if (wc.pin) { | |
7269 | wc.pin = 0; | |
7270 | wc.process_func = replay_one_buffer; | |
7271 | wc.stage = LOG_WALK_REPLAY_INODES; | |
7272 | goto again; | |
7273 | } | |
7274 | /* step three is to replay everything */ | |
7275 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
7276 | wc.stage++; | |
7277 | goto again; | |
7278 | } | |
7279 | ||
7280 | btrfs_free_path(path); | |
7281 | ||
abefa55a | 7282 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 7283 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
7284 | if (ret) |
7285 | return ret; | |
7286 | ||
e02119d5 | 7287 | log_root_tree->log_root = NULL; |
afcdd129 | 7288 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
00246528 | 7289 | btrfs_put_root(log_root_tree); |
79787eaa | 7290 | |
abefa55a | 7291 | return 0; |
79787eaa | 7292 | error: |
b50c6e25 | 7293 | if (wc.trans) |
3a45bb20 | 7294 | btrfs_end_transaction(wc.trans); |
1aeb6b56 | 7295 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
79787eaa JM |
7296 | btrfs_free_path(path); |
7297 | return ret; | |
e02119d5 | 7298 | } |
12fcfd22 CM |
7299 | |
7300 | /* | |
7301 | * there are some corner cases where we want to force a full | |
7302 | * commit instead of allowing a directory to be logged. | |
7303 | * | |
7304 | * They revolve around files there were unlinked from the directory, and | |
7305 | * this function updates the parent directory so that a full commit is | |
7306 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
7307 | * |
7308 | * Must be called before the unlink operations (updates to the subvolume tree, | |
7309 | * inodes, etc) are done. | |
12fcfd22 CM |
7310 | */ |
7311 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 7312 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
7313 | int for_rename) |
7314 | { | |
af4176b4 CM |
7315 | /* |
7316 | * when we're logging a file, if it hasn't been renamed | |
7317 | * or unlinked, and its inode is fully committed on disk, | |
7318 | * we don't have to worry about walking up the directory chain | |
7319 | * to log its parents. | |
7320 | * | |
7321 | * So, we use the last_unlink_trans field to put this transid | |
7322 | * into the file. When the file is logged we check it and | |
7323 | * don't log the parents if the file is fully on disk. | |
7324 | */ | |
4176bdbf NB |
7325 | mutex_lock(&inode->log_mutex); |
7326 | inode->last_unlink_trans = trans->transid; | |
7327 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 7328 | |
12fcfd22 CM |
7329 | /* |
7330 | * if this directory was already logged any new | |
7331 | * names for this file/dir will get recorded | |
7332 | */ | |
4176bdbf | 7333 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
7334 | return; |
7335 | ||
7336 | /* | |
7337 | * if the inode we're about to unlink was logged, | |
7338 | * the log will be properly updated for any new names | |
7339 | */ | |
4176bdbf | 7340 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
7341 | return; |
7342 | ||
7343 | /* | |
7344 | * when renaming files across directories, if the directory | |
7345 | * there we're unlinking from gets fsync'd later on, there's | |
7346 | * no way to find the destination directory later and fsync it | |
7347 | * properly. So, we have to be conservative and force commits | |
7348 | * so the new name gets discovered. | |
7349 | */ | |
7350 | if (for_rename) | |
7351 | goto record; | |
7352 | ||
7353 | /* we can safely do the unlink without any special recording */ | |
7354 | return; | |
7355 | ||
7356 | record: | |
4176bdbf NB |
7357 | mutex_lock(&dir->log_mutex); |
7358 | dir->last_unlink_trans = trans->transid; | |
7359 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
7360 | } |
7361 | ||
7362 | /* | |
7363 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
7364 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
7365 | * that after replaying the log tree of the parent directory's root we will not | |
7366 | * see the snapshot anymore and at log replay time we will not see any log tree | |
7367 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
7368 | * it after replaying the log tree of the parent directory (which would replay | |
7369 | * the snapshot delete operation). | |
2be63d5c FM |
7370 | * |
7371 | * Must be called before the actual snapshot destroy operation (updates to the | |
7372 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
7373 | */ |
7374 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 7375 | struct btrfs_inode *dir) |
1ec9a1ae | 7376 | { |
43663557 NB |
7377 | mutex_lock(&dir->log_mutex); |
7378 | dir->last_unlink_trans = trans->transid; | |
7379 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
7380 | } |
7381 | ||
d5f5bd54 FM |
7382 | /** |
7383 | * Update the log after adding a new name for an inode. | |
7384 | * | |
7385 | * @trans: Transaction handle. | |
7386 | * @old_dentry: The dentry associated with the old name and the old | |
7387 | * parent directory. | |
7388 | * @old_dir: The inode of the previous parent directory for the case | |
7389 | * of a rename. For a link operation, it must be NULL. | |
88d2beec FM |
7390 | * @old_dir_index: The index number associated with the old name, meaningful |
7391 | * only for rename operations (when @old_dir is not NULL). | |
7392 | * Ignored for link operations. | |
d5f5bd54 FM |
7393 | * @parent: The dentry associated with the directory under which the |
7394 | * new name is located. | |
7395 | * | |
7396 | * Call this after adding a new name for an inode, as a result of a link or | |
7397 | * rename operation, and it will properly update the log to reflect the new name. | |
12fcfd22 | 7398 | */ |
75b463d2 | 7399 | void btrfs_log_new_name(struct btrfs_trans_handle *trans, |
d5f5bd54 | 7400 | struct dentry *old_dentry, struct btrfs_inode *old_dir, |
88d2beec | 7401 | u64 old_dir_index, struct dentry *parent) |
12fcfd22 | 7402 | { |
d5f5bd54 | 7403 | struct btrfs_inode *inode = BTRFS_I(d_inode(old_dentry)); |
259c4b96 | 7404 | struct btrfs_root *root = inode->root; |
75b463d2 | 7405 | struct btrfs_log_ctx ctx; |
259c4b96 | 7406 | bool log_pinned = false; |
0f8ce498 | 7407 | int ret; |
12fcfd22 | 7408 | |
af4176b4 CM |
7409 | /* |
7410 | * this will force the logging code to walk the dentry chain | |
7411 | * up for the file | |
7412 | */ | |
9a6509c4 | 7413 | if (!S_ISDIR(inode->vfs_inode.i_mode)) |
9ca5fbfb | 7414 | inode->last_unlink_trans = trans->transid; |
af4176b4 | 7415 | |
12fcfd22 CM |
7416 | /* |
7417 | * if this inode hasn't been logged and directory we're renaming it | |
7418 | * from hasn't been logged, we don't need to log it | |
7419 | */ | |
0f8ce498 FM |
7420 | ret = inode_logged(trans, inode, NULL); |
7421 | if (ret < 0) { | |
7422 | goto out; | |
7423 | } else if (ret == 0) { | |
7424 | if (!old_dir) | |
7425 | return; | |
7426 | /* | |
7427 | * If the inode was not logged and we are doing a rename (old_dir is not | |
7428 | * NULL), check if old_dir was logged - if it was not we can return and | |
7429 | * do nothing. | |
7430 | */ | |
7431 | ret = inode_logged(trans, old_dir, NULL); | |
7432 | if (ret < 0) | |
7433 | goto out; | |
7434 | else if (ret == 0) | |
7435 | return; | |
7436 | } | |
7437 | ret = 0; | |
12fcfd22 | 7438 | |
54a40fc3 FM |
7439 | /* |
7440 | * If we are doing a rename (old_dir is not NULL) from a directory that | |
88d2beec FM |
7441 | * was previously logged, make sure that on log replay we get the old |
7442 | * dir entry deleted. This is needed because we will also log the new | |
7443 | * name of the renamed inode, so we need to make sure that after log | |
7444 | * replay we don't end up with both the new and old dir entries existing. | |
54a40fc3 | 7445 | */ |
88d2beec FM |
7446 | if (old_dir && old_dir->logged_trans == trans->transid) { |
7447 | struct btrfs_root *log = old_dir->root->log_root; | |
7448 | struct btrfs_path *path; | |
ab3c5c18 | 7449 | struct fscrypt_name fname; |
88d2beec FM |
7450 | |
7451 | ASSERT(old_dir_index >= BTRFS_DIR_START_INDEX); | |
7452 | ||
ab3c5c18 STD |
7453 | ret = fscrypt_setup_filename(&old_dir->vfs_inode, |
7454 | &old_dentry->d_name, 0, &fname); | |
7455 | if (ret) | |
7456 | goto out; | |
259c4b96 FM |
7457 | /* |
7458 | * We have two inodes to update in the log, the old directory and | |
7459 | * the inode that got renamed, so we must pin the log to prevent | |
7460 | * anyone from syncing the log until we have updated both inodes | |
7461 | * in the log. | |
7462 | */ | |
723df2bc FM |
7463 | ret = join_running_log_trans(root); |
7464 | /* | |
7465 | * At least one of the inodes was logged before, so this should | |
7466 | * not fail, but if it does, it's not serious, just bail out and | |
7467 | * mark the log for a full commit. | |
7468 | */ | |
7469 | if (WARN_ON_ONCE(ret < 0)) | |
7470 | goto out; | |
259c4b96 | 7471 | log_pinned = true; |
259c4b96 | 7472 | |
88d2beec FM |
7473 | path = btrfs_alloc_path(); |
7474 | if (!path) { | |
259c4b96 | 7475 | ret = -ENOMEM; |
ab3c5c18 | 7476 | fscrypt_free_filename(&fname); |
259c4b96 | 7477 | goto out; |
88d2beec FM |
7478 | } |
7479 | ||
7480 | /* | |
7481 | * Other concurrent task might be logging the old directory, | |
7482 | * as it can be triggered when logging other inode that had or | |
750ee454 FM |
7483 | * still has a dentry in the old directory. We lock the old |
7484 | * directory's log_mutex to ensure the deletion of the old | |
7485 | * name is persisted, because during directory logging we | |
7486 | * delete all BTRFS_DIR_LOG_INDEX_KEY keys and the deletion of | |
7487 | * the old name's dir index item is in the delayed items, so | |
7488 | * it could be missed by an in progress directory logging. | |
88d2beec FM |
7489 | */ |
7490 | mutex_lock(&old_dir->log_mutex); | |
7491 | ret = del_logged_dentry(trans, log, path, btrfs_ino(old_dir), | |
6db75318 | 7492 | &fname.disk_name, old_dir_index); |
88d2beec FM |
7493 | if (ret > 0) { |
7494 | /* | |
7495 | * The dentry does not exist in the log, so record its | |
7496 | * deletion. | |
7497 | */ | |
7498 | btrfs_release_path(path); | |
7499 | ret = insert_dir_log_key(trans, log, path, | |
7500 | btrfs_ino(old_dir), | |
7501 | old_dir_index, old_dir_index); | |
7502 | } | |
7503 | mutex_unlock(&old_dir->log_mutex); | |
7504 | ||
7505 | btrfs_free_path(path); | |
ab3c5c18 | 7506 | fscrypt_free_filename(&fname); |
259c4b96 FM |
7507 | if (ret < 0) |
7508 | goto out; | |
88d2beec | 7509 | } |
54a40fc3 | 7510 | |
75b463d2 FM |
7511 | btrfs_init_log_ctx(&ctx, &inode->vfs_inode); |
7512 | ctx.logging_new_name = true; | |
7513 | /* | |
7514 | * We don't care about the return value. If we fail to log the new name | |
7515 | * then we know the next attempt to sync the log will fallback to a full | |
7516 | * transaction commit (due to a call to btrfs_set_log_full_commit()), so | |
7517 | * we don't need to worry about getting a log committed that has an | |
7518 | * inconsistent state after a rename operation. | |
7519 | */ | |
48778179 | 7520 | btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx); |
e09d94c9 | 7521 | ASSERT(list_empty(&ctx.conflict_inodes)); |
259c4b96 | 7522 | out: |
0f8ce498 FM |
7523 | /* |
7524 | * If an error happened mark the log for a full commit because it's not | |
7525 | * consistent and up to date or we couldn't find out if one of the | |
7526 | * inodes was logged before in this transaction. Do it before unpinning | |
7527 | * the log, to avoid any races with someone else trying to commit it. | |
7528 | */ | |
7529 | if (ret < 0) | |
7530 | btrfs_set_log_full_commit(trans); | |
7531 | if (log_pinned) | |
259c4b96 | 7532 | btrfs_end_log_trans(root); |
12fcfd22 CM |
7533 | } |
7534 |