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" |
e02119d5 CM |
24 | |
25 | /* magic values for the inode_only field in btrfs_log_inode: | |
26 | * | |
27 | * LOG_INODE_ALL means to log everything | |
28 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
29 | * during log replay | |
30 | */ | |
e13976cf DS |
31 | enum { |
32 | LOG_INODE_ALL, | |
33 | LOG_INODE_EXISTS, | |
34 | LOG_OTHER_INODE, | |
35 | LOG_OTHER_INODE_ALL, | |
36 | }; | |
e02119d5 | 37 | |
12fcfd22 CM |
38 | /* |
39 | * directory trouble cases | |
40 | * | |
41 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
42 | * log, we must force a full commit before doing an fsync of the directory | |
43 | * where the unlink was done. | |
44 | * ---> record transid of last unlink/rename per directory | |
45 | * | |
46 | * mkdir foo/some_dir | |
47 | * normal commit | |
48 | * rename foo/some_dir foo2/some_dir | |
49 | * mkdir foo/some_dir | |
50 | * fsync foo/some_dir/some_file | |
51 | * | |
52 | * The fsync above will unlink the original some_dir without recording | |
53 | * it in its new location (foo2). After a crash, some_dir will be gone | |
54 | * unless the fsync of some_file forces a full commit | |
55 | * | |
56 | * 2) we must log any new names for any file or dir that is in the fsync | |
57 | * log. ---> check inode while renaming/linking. | |
58 | * | |
59 | * 2a) we must log any new names for any file or dir during rename | |
60 | * when the directory they are being removed from was logged. | |
61 | * ---> check inode and old parent dir during rename | |
62 | * | |
63 | * 2a is actually the more important variant. With the extra logging | |
64 | * a crash might unlink the old name without recreating the new one | |
65 | * | |
66 | * 3) after a crash, we must go through any directories with a link count | |
67 | * of zero and redo the rm -rf | |
68 | * | |
69 | * mkdir f1/foo | |
70 | * normal commit | |
71 | * rm -rf f1/foo | |
72 | * fsync(f1) | |
73 | * | |
74 | * The directory f1 was fully removed from the FS, but fsync was never | |
75 | * called on f1, only its parent dir. After a crash the rm -rf must | |
76 | * be replayed. This must be able to recurse down the entire | |
77 | * directory tree. The inode link count fixup code takes care of the | |
78 | * ugly details. | |
79 | */ | |
80 | ||
e02119d5 CM |
81 | /* |
82 | * stages for the tree walking. The first | |
83 | * stage (0) is to only pin down the blocks we find | |
84 | * the second stage (1) is to make sure that all the inodes | |
85 | * we find in the log are created in the subvolume. | |
86 | * | |
87 | * The last stage is to deal with directories and links and extents | |
88 | * and all the other fun semantics | |
89 | */ | |
e13976cf DS |
90 | enum { |
91 | LOG_WALK_PIN_ONLY, | |
92 | LOG_WALK_REPLAY_INODES, | |
93 | LOG_WALK_REPLAY_DIR_INDEX, | |
94 | LOG_WALK_REPLAY_ALL, | |
95 | }; | |
e02119d5 | 96 | |
12fcfd22 | 97 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
90d04510 | 98 | struct btrfs_inode *inode, |
49dae1bc | 99 | int inode_only, |
8407f553 | 100 | struct btrfs_log_ctx *ctx); |
ec051c0f YZ |
101 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
102 | struct btrfs_root *root, | |
103 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
104 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
105 | struct btrfs_root *root, | |
106 | struct btrfs_root *log, | |
107 | struct btrfs_path *path, | |
108 | u64 dirid, int del_all); | |
fa1a0f42 | 109 | static void wait_log_commit(struct btrfs_root *root, int transid); |
e02119d5 CM |
110 | |
111 | /* | |
112 | * tree logging is a special write ahead log used to make sure that | |
113 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
114 | * | |
115 | * Full tree commits are expensive because they require commonly | |
116 | * modified blocks to be recowed, creating many dirty pages in the | |
117 | * extent tree an 4x-6x higher write load than ext3. | |
118 | * | |
119 | * Instead of doing a tree commit on every fsync, we use the | |
120 | * key ranges and transaction ids to find items for a given file or directory | |
121 | * that have changed in this transaction. Those items are copied into | |
122 | * a special tree (one per subvolume root), that tree is written to disk | |
123 | * and then the fsync is considered complete. | |
124 | * | |
125 | * After a crash, items are copied out of the log-tree back into the | |
126 | * subvolume tree. Any file data extents found are recorded in the extent | |
127 | * allocation tree, and the log-tree freed. | |
128 | * | |
129 | * The log tree is read three times, once to pin down all the extents it is | |
130 | * using in ram and once, once to create all the inodes logged in the tree | |
131 | * and once to do all the other items. | |
132 | */ | |
133 | ||
e02119d5 CM |
134 | /* |
135 | * start a sub transaction and setup the log tree | |
136 | * this increments the log tree writer count to make the people | |
137 | * syncing the tree wait for us to finish | |
138 | */ | |
139 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
140 | struct btrfs_root *root, |
141 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 142 | { |
0b246afa | 143 | struct btrfs_fs_info *fs_info = root->fs_info; |
47876f7c | 144 | struct btrfs_root *tree_root = fs_info->tree_root; |
fa1a0f42 | 145 | const bool zoned = btrfs_is_zoned(fs_info); |
34eb2a52 | 146 | int ret = 0; |
fa1a0f42 | 147 | bool created = false; |
7237f183 | 148 | |
47876f7c FM |
149 | /* |
150 | * First check if the log root tree was already created. If not, create | |
151 | * it before locking the root's log_mutex, just to keep lockdep happy. | |
152 | */ | |
153 | if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state)) { | |
154 | mutex_lock(&tree_root->log_mutex); | |
155 | if (!fs_info->log_root_tree) { | |
156 | ret = btrfs_init_log_root_tree(trans, fs_info); | |
fa1a0f42 | 157 | if (!ret) { |
47876f7c | 158 | set_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state); |
fa1a0f42 NA |
159 | created = true; |
160 | } | |
47876f7c FM |
161 | } |
162 | mutex_unlock(&tree_root->log_mutex); | |
163 | if (ret) | |
164 | return ret; | |
165 | } | |
166 | ||
7237f183 | 167 | mutex_lock(&root->log_mutex); |
34eb2a52 | 168 | |
fa1a0f42 | 169 | again: |
7237f183 | 170 | if (root->log_root) { |
fa1a0f42 NA |
171 | int index = (root->log_transid + 1) % 2; |
172 | ||
4884b8e8 | 173 | if (btrfs_need_log_full_commit(trans)) { |
50471a38 MX |
174 | ret = -EAGAIN; |
175 | goto out; | |
176 | } | |
34eb2a52 | 177 | |
fa1a0f42 NA |
178 | if (zoned && atomic_read(&root->log_commit[index])) { |
179 | wait_log_commit(root, root->log_transid - 1); | |
180 | goto again; | |
181 | } | |
182 | ||
ff782e0a | 183 | if (!root->log_start_pid) { |
27cdeb70 | 184 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 185 | root->log_start_pid = current->pid; |
ff782e0a | 186 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 187 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 188 | } |
34eb2a52 | 189 | } else { |
fa1a0f42 NA |
190 | /* |
191 | * This means fs_info->log_root_tree was already created | |
192 | * for some other FS trees. Do the full commit not to mix | |
193 | * nodes from multiple log transactions to do sequential | |
194 | * writing. | |
195 | */ | |
196 | if (zoned && !created) { | |
197 | ret = -EAGAIN; | |
198 | goto out; | |
199 | } | |
200 | ||
e02119d5 | 201 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 202 | if (ret) |
e87ac136 | 203 | goto out; |
34eb2a52 | 204 | |
e7a79811 | 205 | set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state); |
34eb2a52 Z |
206 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
207 | root->log_start_pid = current->pid; | |
e02119d5 | 208 | } |
34eb2a52 | 209 | |
7237f183 | 210 | atomic_inc(&root->log_writers); |
289cffcb | 211 | if (!ctx->logging_new_name) { |
34eb2a52 | 212 | int index = root->log_transid % 2; |
8b050d35 | 213 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 214 | ctx->log_transid = root->log_transid; |
8b050d35 | 215 | } |
34eb2a52 | 216 | |
e87ac136 | 217 | out: |
7237f183 | 218 | mutex_unlock(&root->log_mutex); |
e87ac136 | 219 | return ret; |
e02119d5 CM |
220 | } |
221 | ||
222 | /* | |
223 | * returns 0 if there was a log transaction running and we were able | |
224 | * to join, or returns -ENOENT if there were not transactions | |
225 | * in progress | |
226 | */ | |
227 | static int join_running_log_trans(struct btrfs_root *root) | |
228 | { | |
fa1a0f42 | 229 | const bool zoned = btrfs_is_zoned(root->fs_info); |
e02119d5 CM |
230 | int ret = -ENOENT; |
231 | ||
e7a79811 FM |
232 | if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state)) |
233 | return ret; | |
234 | ||
7237f183 | 235 | mutex_lock(&root->log_mutex); |
fa1a0f42 | 236 | again: |
e02119d5 | 237 | if (root->log_root) { |
fa1a0f42 NA |
238 | int index = (root->log_transid + 1) % 2; |
239 | ||
e02119d5 | 240 | ret = 0; |
fa1a0f42 NA |
241 | if (zoned && atomic_read(&root->log_commit[index])) { |
242 | wait_log_commit(root, root->log_transid - 1); | |
243 | goto again; | |
244 | } | |
7237f183 | 245 | atomic_inc(&root->log_writers); |
e02119d5 | 246 | } |
7237f183 | 247 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
248 | return ret; |
249 | } | |
250 | ||
12fcfd22 CM |
251 | /* |
252 | * This either makes the current running log transaction wait | |
253 | * until you call btrfs_end_log_trans() or it makes any future | |
254 | * log transactions wait until you call btrfs_end_log_trans() | |
255 | */ | |
45128b08 | 256 | void btrfs_pin_log_trans(struct btrfs_root *root) |
12fcfd22 | 257 | { |
12fcfd22 | 258 | atomic_inc(&root->log_writers); |
12fcfd22 CM |
259 | } |
260 | ||
e02119d5 CM |
261 | /* |
262 | * indicate we're done making changes to the log tree | |
263 | * and wake up anyone waiting to do a sync | |
264 | */ | |
143bede5 | 265 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 266 | { |
7237f183 | 267 | if (atomic_dec_and_test(&root->log_writers)) { |
093258e6 DS |
268 | /* atomic_dec_and_test implies a barrier */ |
269 | cond_wake_up_nomb(&root->log_writer_wait); | |
7237f183 | 270 | } |
e02119d5 CM |
271 | } |
272 | ||
247462a5 DS |
273 | static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf) |
274 | { | |
275 | filemap_fdatawait_range(buf->pages[0]->mapping, | |
276 | buf->start, buf->start + buf->len - 1); | |
277 | } | |
e02119d5 CM |
278 | |
279 | /* | |
280 | * the walk control struct is used to pass state down the chain when | |
281 | * processing the log tree. The stage field tells us which part | |
282 | * of the log tree processing we are currently doing. The others | |
283 | * are state fields used for that specific part | |
284 | */ | |
285 | struct walk_control { | |
286 | /* should we free the extent on disk when done? This is used | |
287 | * at transaction commit time while freeing a log tree | |
288 | */ | |
289 | int free; | |
290 | ||
e02119d5 CM |
291 | /* pin only walk, we record which extents on disk belong to the |
292 | * log trees | |
293 | */ | |
294 | int pin; | |
295 | ||
296 | /* what stage of the replay code we're currently in */ | |
297 | int stage; | |
298 | ||
f2d72f42 FM |
299 | /* |
300 | * Ignore any items from the inode currently being processed. Needs | |
301 | * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in | |
302 | * the LOG_WALK_REPLAY_INODES stage. | |
303 | */ | |
304 | bool ignore_cur_inode; | |
305 | ||
e02119d5 CM |
306 | /* the root we are currently replaying */ |
307 | struct btrfs_root *replay_dest; | |
308 | ||
309 | /* the trans handle for the current replay */ | |
310 | struct btrfs_trans_handle *trans; | |
311 | ||
312 | /* the function that gets used to process blocks we find in the | |
313 | * tree. Note the extent_buffer might not be up to date when it is | |
314 | * passed in, and it must be checked or read if you need the data | |
315 | * inside it | |
316 | */ | |
317 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 318 | struct walk_control *wc, u64 gen, int level); |
e02119d5 CM |
319 | }; |
320 | ||
321 | /* | |
322 | * process_func used to pin down extents, write them or wait on them | |
323 | */ | |
324 | static int process_one_buffer(struct btrfs_root *log, | |
325 | struct extent_buffer *eb, | |
581c1760 | 326 | struct walk_control *wc, u64 gen, int level) |
e02119d5 | 327 | { |
0b246afa | 328 | struct btrfs_fs_info *fs_info = log->fs_info; |
b50c6e25 JB |
329 | int ret = 0; |
330 | ||
8c2a1a30 JB |
331 | /* |
332 | * If this fs is mixed then we need to be able to process the leaves to | |
333 | * pin down any logged extents, so we have to read the block. | |
334 | */ | |
0b246afa | 335 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
581c1760 | 336 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
8c2a1a30 JB |
337 | if (ret) |
338 | return ret; | |
339 | } | |
340 | ||
c816d705 | 341 | if (wc->pin) { |
9fce5704 | 342 | ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start, |
2ff7e61e | 343 | eb->len); |
c816d705 FM |
344 | if (ret) |
345 | return ret; | |
e02119d5 | 346 | |
c816d705 FM |
347 | if (btrfs_buffer_uptodate(eb, gen, 0) && |
348 | btrfs_header_level(eb) == 0) | |
bcdc428c | 349 | ret = btrfs_exclude_logged_extents(eb); |
e02119d5 | 350 | } |
b50c6e25 | 351 | return ret; |
e02119d5 CM |
352 | } |
353 | ||
086dcbfa FM |
354 | static int do_overwrite_item(struct btrfs_trans_handle *trans, |
355 | struct btrfs_root *root, | |
356 | struct btrfs_path *path, | |
357 | struct extent_buffer *eb, int slot, | |
358 | struct btrfs_key *key) | |
e02119d5 CM |
359 | { |
360 | int ret; | |
361 | u32 item_size; | |
362 | u64 saved_i_size = 0; | |
363 | int save_old_i_size = 0; | |
364 | unsigned long src_ptr; | |
365 | unsigned long dst_ptr; | |
366 | int overwrite_root = 0; | |
4bc4bee4 | 367 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
368 | |
369 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
370 | overwrite_root = 1; | |
371 | ||
3212fa14 | 372 | item_size = btrfs_item_size(eb, slot); |
e02119d5 CM |
373 | src_ptr = btrfs_item_ptr_offset(eb, slot); |
374 | ||
086dcbfa FM |
375 | /* Our caller must have done a search for the key for us. */ |
376 | ASSERT(path->nodes[0] != NULL); | |
377 | ||
378 | /* | |
379 | * And the slot must point to the exact key or the slot where the key | |
380 | * should be at (the first item with a key greater than 'key') | |
381 | */ | |
382 | if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) { | |
383 | struct btrfs_key found_key; | |
384 | ||
385 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
386 | ret = btrfs_comp_cpu_keys(&found_key, key); | |
387 | ASSERT(ret >= 0); | |
388 | } else { | |
389 | ret = 1; | |
390 | } | |
4bc4bee4 | 391 | |
e02119d5 CM |
392 | if (ret == 0) { |
393 | char *src_copy; | |
394 | char *dst_copy; | |
3212fa14 | 395 | u32 dst_size = btrfs_item_size(path->nodes[0], |
e02119d5 CM |
396 | path->slots[0]); |
397 | if (dst_size != item_size) | |
398 | goto insert; | |
399 | ||
400 | if (item_size == 0) { | |
b3b4aa74 | 401 | btrfs_release_path(path); |
e02119d5 CM |
402 | return 0; |
403 | } | |
404 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
405 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 406 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 407 | btrfs_release_path(path); |
2a29edc6 | 408 | kfree(dst_copy); |
409 | kfree(src_copy); | |
410 | return -ENOMEM; | |
411 | } | |
e02119d5 CM |
412 | |
413 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
414 | ||
415 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
416 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
417 | item_size); | |
418 | ret = memcmp(dst_copy, src_copy, item_size); | |
419 | ||
420 | kfree(dst_copy); | |
421 | kfree(src_copy); | |
422 | /* | |
423 | * they have the same contents, just return, this saves | |
424 | * us from cowing blocks in the destination tree and doing | |
425 | * extra writes that may not have been done by a previous | |
426 | * sync | |
427 | */ | |
428 | if (ret == 0) { | |
b3b4aa74 | 429 | btrfs_release_path(path); |
e02119d5 CM |
430 | return 0; |
431 | } | |
432 | ||
4bc4bee4 JB |
433 | /* |
434 | * We need to load the old nbytes into the inode so when we | |
435 | * replay the extents we've logged we get the right nbytes. | |
436 | */ | |
437 | if (inode_item) { | |
438 | struct btrfs_inode_item *item; | |
439 | u64 nbytes; | |
d555438b | 440 | u32 mode; |
4bc4bee4 JB |
441 | |
442 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
443 | struct btrfs_inode_item); | |
444 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
445 | item = btrfs_item_ptr(eb, slot, | |
446 | struct btrfs_inode_item); | |
447 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
448 | |
449 | /* | |
450 | * If this is a directory we need to reset the i_size to | |
451 | * 0 so that we can set it up properly when replaying | |
452 | * the rest of the items in this log. | |
453 | */ | |
454 | mode = btrfs_inode_mode(eb, item); | |
455 | if (S_ISDIR(mode)) | |
456 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
457 | } |
458 | } else if (inode_item) { | |
459 | struct btrfs_inode_item *item; | |
d555438b | 460 | u32 mode; |
4bc4bee4 JB |
461 | |
462 | /* | |
463 | * New inode, set nbytes to 0 so that the nbytes comes out | |
464 | * properly when we replay the extents. | |
465 | */ | |
466 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
467 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
468 | |
469 | /* | |
470 | * If this is a directory we need to reset the i_size to 0 so | |
471 | * that we can set it up properly when replaying the rest of | |
472 | * the items in this log. | |
473 | */ | |
474 | mode = btrfs_inode_mode(eb, item); | |
475 | if (S_ISDIR(mode)) | |
476 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
477 | } |
478 | insert: | |
b3b4aa74 | 479 | btrfs_release_path(path); |
e02119d5 | 480 | /* try to insert the key into the destination tree */ |
df8d116f | 481 | path->skip_release_on_error = 1; |
e02119d5 CM |
482 | ret = btrfs_insert_empty_item(trans, root, path, |
483 | key, item_size); | |
df8d116f | 484 | path->skip_release_on_error = 0; |
e02119d5 CM |
485 | |
486 | /* make sure any existing item is the correct size */ | |
df8d116f | 487 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 | 488 | u32 found_size; |
3212fa14 | 489 | found_size = btrfs_item_size(path->nodes[0], |
e02119d5 | 490 | path->slots[0]); |
143bede5 | 491 | if (found_size > item_size) |
78ac4f9e | 492 | btrfs_truncate_item(path, item_size, 1); |
143bede5 | 493 | else if (found_size < item_size) |
c71dd880 | 494 | btrfs_extend_item(path, item_size - found_size); |
e02119d5 | 495 | } else if (ret) { |
4a500fd1 | 496 | return ret; |
e02119d5 CM |
497 | } |
498 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
499 | path->slots[0]); | |
500 | ||
501 | /* don't overwrite an existing inode if the generation number | |
502 | * was logged as zero. This is done when the tree logging code | |
503 | * is just logging an inode to make sure it exists after recovery. | |
504 | * | |
505 | * Also, don't overwrite i_size on directories during replay. | |
506 | * log replay inserts and removes directory items based on the | |
507 | * state of the tree found in the subvolume, and i_size is modified | |
508 | * as it goes | |
509 | */ | |
510 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
511 | struct btrfs_inode_item *src_item; | |
512 | struct btrfs_inode_item *dst_item; | |
513 | ||
514 | src_item = (struct btrfs_inode_item *)src_ptr; | |
515 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
516 | ||
1a4bcf47 FM |
517 | if (btrfs_inode_generation(eb, src_item) == 0) { |
518 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 519 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 520 | |
2f2ff0ee FM |
521 | /* |
522 | * For regular files an ino_size == 0 is used only when | |
523 | * logging that an inode exists, as part of a directory | |
524 | * fsync, and the inode wasn't fsynced before. In this | |
525 | * case don't set the size of the inode in the fs/subvol | |
526 | * tree, otherwise we would be throwing valid data away. | |
527 | */ | |
1a4bcf47 | 528 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee | 529 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
60d48e2e DS |
530 | ino_size != 0) |
531 | btrfs_set_inode_size(dst_eb, dst_item, ino_size); | |
e02119d5 | 532 | goto no_copy; |
1a4bcf47 | 533 | } |
e02119d5 CM |
534 | |
535 | if (overwrite_root && | |
536 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
537 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
538 | save_old_i_size = 1; | |
539 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
540 | dst_item); | |
541 | } | |
542 | } | |
543 | ||
544 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
545 | src_ptr, item_size); | |
546 | ||
547 | if (save_old_i_size) { | |
548 | struct btrfs_inode_item *dst_item; | |
549 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
550 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
551 | } | |
552 | ||
553 | /* make sure the generation is filled in */ | |
554 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
555 | struct btrfs_inode_item *dst_item; | |
556 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
557 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
558 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
559 | trans->transid); | |
560 | } | |
561 | } | |
562 | no_copy: | |
563 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 564 | btrfs_release_path(path); |
e02119d5 CM |
565 | return 0; |
566 | } | |
567 | ||
086dcbfa FM |
568 | /* |
569 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
570 | * to the src data we are copying out. | |
571 | * | |
572 | * root is the tree we are copying into, and path is a scratch | |
573 | * path for use in this function (it should be released on entry and | |
574 | * will be released on exit). | |
575 | * | |
576 | * If the key is already in the destination tree the existing item is | |
577 | * overwritten. If the existing item isn't big enough, it is extended. | |
578 | * If it is too large, it is truncated. | |
579 | * | |
580 | * If the key isn't in the destination yet, a new item is inserted. | |
581 | */ | |
582 | static int overwrite_item(struct btrfs_trans_handle *trans, | |
583 | struct btrfs_root *root, | |
584 | struct btrfs_path *path, | |
585 | struct extent_buffer *eb, int slot, | |
586 | struct btrfs_key *key) | |
587 | { | |
588 | int ret; | |
589 | ||
590 | /* Look for the key in the destination tree. */ | |
591 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
592 | if (ret < 0) | |
593 | return ret; | |
594 | ||
595 | return do_overwrite_item(trans, root, path, eb, slot, key); | |
596 | } | |
597 | ||
e02119d5 CM |
598 | /* |
599 | * simple helper to read an inode off the disk from a given root | |
600 | * This can only be called for subvolume roots and not for the log | |
601 | */ | |
602 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
603 | u64 objectid) | |
604 | { | |
605 | struct inode *inode; | |
e02119d5 | 606 | |
0202e83f | 607 | inode = btrfs_iget(root->fs_info->sb, objectid, root); |
2e19f1f9 | 608 | if (IS_ERR(inode)) |
5d4f98a2 | 609 | inode = NULL; |
e02119d5 CM |
610 | return inode; |
611 | } | |
612 | ||
613 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
614 | * subvolume 'root'. path is released on entry and should be released | |
615 | * on exit. | |
616 | * | |
617 | * extents in the log tree have not been allocated out of the extent | |
618 | * tree yet. So, this completes the allocation, taking a reference | |
619 | * as required if the extent already exists or creating a new extent | |
620 | * if it isn't in the extent allocation tree yet. | |
621 | * | |
622 | * The extent is inserted into the file, dropping any existing extents | |
623 | * from the file that overlap the new one. | |
624 | */ | |
625 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
626 | struct btrfs_root *root, | |
627 | struct btrfs_path *path, | |
628 | struct extent_buffer *eb, int slot, | |
629 | struct btrfs_key *key) | |
630 | { | |
5893dfb9 | 631 | struct btrfs_drop_extents_args drop_args = { 0 }; |
0b246afa | 632 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 633 | int found_type; |
e02119d5 | 634 | u64 extent_end; |
e02119d5 | 635 | u64 start = key->offset; |
4bc4bee4 | 636 | u64 nbytes = 0; |
e02119d5 CM |
637 | struct btrfs_file_extent_item *item; |
638 | struct inode *inode = NULL; | |
639 | unsigned long size; | |
640 | int ret = 0; | |
641 | ||
642 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
643 | found_type = btrfs_file_extent_type(eb, item); | |
644 | ||
d899e052 | 645 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
646 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
647 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
648 | extent_end = start + nbytes; | |
649 | ||
650 | /* | |
651 | * We don't add to the inodes nbytes if we are prealloc or a | |
652 | * hole. | |
653 | */ | |
654 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
655 | nbytes = 0; | |
656 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 657 | size = btrfs_file_extent_ram_bytes(eb, item); |
4bc4bee4 | 658 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
da17066c | 659 | extent_end = ALIGN(start + size, |
0b246afa | 660 | fs_info->sectorsize); |
e02119d5 CM |
661 | } else { |
662 | ret = 0; | |
663 | goto out; | |
664 | } | |
665 | ||
666 | inode = read_one_inode(root, key->objectid); | |
667 | if (!inode) { | |
668 | ret = -EIO; | |
669 | goto out; | |
670 | } | |
671 | ||
672 | /* | |
673 | * first check to see if we already have this extent in the | |
674 | * file. This must be done before the btrfs_drop_extents run | |
675 | * so we don't try to drop this extent. | |
676 | */ | |
f85b7379 DS |
677 | ret = btrfs_lookup_file_extent(trans, root, path, |
678 | btrfs_ino(BTRFS_I(inode)), start, 0); | |
e02119d5 | 679 | |
d899e052 YZ |
680 | if (ret == 0 && |
681 | (found_type == BTRFS_FILE_EXTENT_REG || | |
682 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
683 | struct btrfs_file_extent_item cmp1; |
684 | struct btrfs_file_extent_item cmp2; | |
685 | struct btrfs_file_extent_item *existing; | |
686 | struct extent_buffer *leaf; | |
687 | ||
688 | leaf = path->nodes[0]; | |
689 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
690 | struct btrfs_file_extent_item); | |
691 | ||
692 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
693 | sizeof(cmp1)); | |
694 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
695 | sizeof(cmp2)); | |
696 | ||
697 | /* | |
698 | * we already have a pointer to this exact extent, | |
699 | * we don't have to do anything | |
700 | */ | |
701 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 702 | btrfs_release_path(path); |
e02119d5 CM |
703 | goto out; |
704 | } | |
705 | } | |
b3b4aa74 | 706 | btrfs_release_path(path); |
e02119d5 CM |
707 | |
708 | /* drop any overlapping extents */ | |
5893dfb9 FM |
709 | drop_args.start = start; |
710 | drop_args.end = extent_end; | |
711 | drop_args.drop_cache = true; | |
712 | ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args); | |
3650860b JB |
713 | if (ret) |
714 | goto out; | |
e02119d5 | 715 | |
07d400a6 YZ |
716 | if (found_type == BTRFS_FILE_EXTENT_REG || |
717 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 718 | u64 offset; |
07d400a6 YZ |
719 | unsigned long dest_offset; |
720 | struct btrfs_key ins; | |
721 | ||
3168021c FM |
722 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0 && |
723 | btrfs_fs_incompat(fs_info, NO_HOLES)) | |
724 | goto update_inode; | |
725 | ||
07d400a6 YZ |
726 | ret = btrfs_insert_empty_item(trans, root, path, key, |
727 | sizeof(*item)); | |
3650860b JB |
728 | if (ret) |
729 | goto out; | |
07d400a6 YZ |
730 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
731 | path->slots[0]); | |
732 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
733 | (unsigned long)item, sizeof(*item)); | |
734 | ||
735 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
736 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
737 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 738 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 739 | |
df2c95f3 QW |
740 | /* |
741 | * Manually record dirty extent, as here we did a shallow | |
742 | * file extent item copy and skip normal backref update, | |
743 | * but modifying extent tree all by ourselves. | |
744 | * So need to manually record dirty extent for qgroup, | |
745 | * as the owner of the file extent changed from log tree | |
746 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
747 | */ | |
a95f3aaf | 748 | ret = btrfs_qgroup_trace_extent(trans, |
df2c95f3 QW |
749 | btrfs_file_extent_disk_bytenr(eb, item), |
750 | btrfs_file_extent_disk_num_bytes(eb, item), | |
751 | GFP_NOFS); | |
752 | if (ret < 0) | |
753 | goto out; | |
754 | ||
07d400a6 | 755 | if (ins.objectid > 0) { |
82fa113f | 756 | struct btrfs_ref ref = { 0 }; |
07d400a6 YZ |
757 | u64 csum_start; |
758 | u64 csum_end; | |
759 | LIST_HEAD(ordered_sums); | |
82fa113f | 760 | |
07d400a6 YZ |
761 | /* |
762 | * is this extent already allocated in the extent | |
763 | * allocation tree? If so, just add a reference | |
764 | */ | |
2ff7e61e | 765 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 | 766 | ins.offset); |
3736127a MPS |
767 | if (ret < 0) { |
768 | goto out; | |
769 | } else if (ret == 0) { | |
82fa113f QW |
770 | btrfs_init_generic_ref(&ref, |
771 | BTRFS_ADD_DELAYED_REF, | |
772 | ins.objectid, ins.offset, 0); | |
773 | btrfs_init_data_ref(&ref, | |
774 | root->root_key.objectid, | |
f42c5da6 | 775 | key->objectid, offset, 0, false); |
82fa113f | 776 | ret = btrfs_inc_extent_ref(trans, &ref); |
b50c6e25 JB |
777 | if (ret) |
778 | goto out; | |
07d400a6 YZ |
779 | } else { |
780 | /* | |
781 | * insert the extent pointer in the extent | |
782 | * allocation tree | |
783 | */ | |
5d4f98a2 | 784 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e | 785 | root->root_key.objectid, |
5d4f98a2 | 786 | key->objectid, offset, &ins); |
b50c6e25 JB |
787 | if (ret) |
788 | goto out; | |
07d400a6 | 789 | } |
b3b4aa74 | 790 | btrfs_release_path(path); |
07d400a6 YZ |
791 | |
792 | if (btrfs_file_extent_compression(eb, item)) { | |
793 | csum_start = ins.objectid; | |
794 | csum_end = csum_start + ins.offset; | |
795 | } else { | |
796 | csum_start = ins.objectid + | |
797 | btrfs_file_extent_offset(eb, item); | |
798 | csum_end = csum_start + | |
799 | btrfs_file_extent_num_bytes(eb, item); | |
800 | } | |
801 | ||
802 | ret = btrfs_lookup_csums_range(root->log_root, | |
803 | csum_start, csum_end - 1, | |
a2de733c | 804 | &ordered_sums, 0); |
3650860b JB |
805 | if (ret) |
806 | goto out; | |
b84b8390 FM |
807 | /* |
808 | * Now delete all existing cums in the csum root that | |
809 | * cover our range. We do this because we can have an | |
810 | * extent that is completely referenced by one file | |
811 | * extent item and partially referenced by another | |
812 | * file extent item (like after using the clone or | |
813 | * extent_same ioctls). In this case if we end up doing | |
814 | * the replay of the one that partially references the | |
815 | * extent first, and we do not do the csum deletion | |
816 | * below, we can get 2 csum items in the csum tree that | |
817 | * overlap each other. For example, imagine our log has | |
818 | * the two following file extent items: | |
819 | * | |
820 | * key (257 EXTENT_DATA 409600) | |
821 | * extent data disk byte 12845056 nr 102400 | |
822 | * extent data offset 20480 nr 20480 ram 102400 | |
823 | * | |
824 | * key (257 EXTENT_DATA 819200) | |
825 | * extent data disk byte 12845056 nr 102400 | |
826 | * extent data offset 0 nr 102400 ram 102400 | |
827 | * | |
828 | * Where the second one fully references the 100K extent | |
829 | * that starts at disk byte 12845056, and the log tree | |
830 | * has a single csum item that covers the entire range | |
831 | * of the extent: | |
832 | * | |
833 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
834 | * | |
835 | * After the first file extent item is replayed, the | |
836 | * csum tree gets the following csum item: | |
837 | * | |
838 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
839 | * | |
840 | * Which covers the 20K sub-range starting at offset 20K | |
841 | * of our extent. Now when we replay the second file | |
842 | * extent item, if we do not delete existing csum items | |
843 | * that cover any of its blocks, we end up getting two | |
844 | * csum items in our csum tree that overlap each other: | |
845 | * | |
846 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
847 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
848 | * | |
849 | * Which is a problem, because after this anyone trying | |
850 | * to lookup up for the checksum of any block of our | |
851 | * extent starting at an offset of 40K or higher, will | |
852 | * end up looking at the second csum item only, which | |
853 | * does not contain the checksum for any block starting | |
854 | * at offset 40K or higher of our extent. | |
855 | */ | |
07d400a6 YZ |
856 | while (!list_empty(&ordered_sums)) { |
857 | struct btrfs_ordered_sum *sums; | |
fc28b25e JB |
858 | struct btrfs_root *csum_root; |
859 | ||
07d400a6 YZ |
860 | sums = list_entry(ordered_sums.next, |
861 | struct btrfs_ordered_sum, | |
862 | list); | |
fc28b25e JB |
863 | csum_root = btrfs_csum_root(fs_info, |
864 | sums->bytenr); | |
b84b8390 | 865 | if (!ret) |
fc28b25e | 866 | ret = btrfs_del_csums(trans, csum_root, |
5b4aacef JM |
867 | sums->bytenr, |
868 | sums->len); | |
3650860b JB |
869 | if (!ret) |
870 | ret = btrfs_csum_file_blocks(trans, | |
fc28b25e JB |
871 | csum_root, |
872 | sums); | |
07d400a6 YZ |
873 | list_del(&sums->list); |
874 | kfree(sums); | |
875 | } | |
3650860b JB |
876 | if (ret) |
877 | goto out; | |
07d400a6 | 878 | } else { |
b3b4aa74 | 879 | btrfs_release_path(path); |
07d400a6 YZ |
880 | } |
881 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
882 | /* inline extents are easy, we just overwrite them */ | |
883 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
884 | if (ret) |
885 | goto out; | |
07d400a6 | 886 | } |
e02119d5 | 887 | |
9ddc959e JB |
888 | ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, |
889 | extent_end - start); | |
890 | if (ret) | |
891 | goto out; | |
892 | ||
3168021c | 893 | update_inode: |
2766ff61 | 894 | btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found); |
9a56fcd1 | 895 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e02119d5 CM |
896 | out: |
897 | if (inode) | |
898 | iput(inode); | |
899 | return ret; | |
900 | } | |
901 | ||
902 | /* | |
903 | * when cleaning up conflicts between the directory names in the | |
904 | * subvolume, directory names in the log and directory names in the | |
905 | * inode back references, we may have to unlink inodes from directories. | |
906 | * | |
907 | * This is a helper function to do the unlink of a specific directory | |
908 | * item | |
909 | */ | |
910 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
e02119d5 | 911 | struct btrfs_path *path, |
207e7d92 | 912 | struct btrfs_inode *dir, |
e02119d5 CM |
913 | struct btrfs_dir_item *di) |
914 | { | |
9798ba24 | 915 | struct btrfs_root *root = dir->root; |
e02119d5 CM |
916 | struct inode *inode; |
917 | char *name; | |
918 | int name_len; | |
919 | struct extent_buffer *leaf; | |
920 | struct btrfs_key location; | |
921 | int ret; | |
922 | ||
923 | leaf = path->nodes[0]; | |
924 | ||
925 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
926 | name_len = btrfs_dir_name_len(leaf, di); | |
927 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 928 | if (!name) |
929 | return -ENOMEM; | |
930 | ||
e02119d5 | 931 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 932 | btrfs_release_path(path); |
e02119d5 CM |
933 | |
934 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 935 | if (!inode) { |
3650860b JB |
936 | ret = -EIO; |
937 | goto out; | |
c00e9493 | 938 | } |
e02119d5 | 939 | |
ec051c0f | 940 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
941 | if (ret) |
942 | goto out; | |
12fcfd22 | 943 | |
4467af88 | 944 | ret = btrfs_unlink_inode(trans, dir, BTRFS_I(inode), name, |
207e7d92 | 945 | name_len); |
3650860b JB |
946 | if (ret) |
947 | goto out; | |
ada9af21 | 948 | else |
e5c304e6 | 949 | ret = btrfs_run_delayed_items(trans); |
3650860b | 950 | out: |
e02119d5 | 951 | kfree(name); |
e02119d5 CM |
952 | iput(inode); |
953 | return ret; | |
954 | } | |
955 | ||
956 | /* | |
77a5b9e3 FM |
957 | * See if a given name and sequence number found in an inode back reference are |
958 | * already in a directory and correctly point to this inode. | |
959 | * | |
960 | * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it | |
961 | * exists. | |
e02119d5 CM |
962 | */ |
963 | static noinline int inode_in_dir(struct btrfs_root *root, | |
964 | struct btrfs_path *path, | |
965 | u64 dirid, u64 objectid, u64 index, | |
966 | const char *name, int name_len) | |
967 | { | |
968 | struct btrfs_dir_item *di; | |
969 | struct btrfs_key location; | |
77a5b9e3 | 970 | int ret = 0; |
e02119d5 CM |
971 | |
972 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
973 | index, name, name_len, 0); | |
77a5b9e3 | 974 | if (IS_ERR(di)) { |
8dcbc261 | 975 | ret = PTR_ERR(di); |
77a5b9e3 FM |
976 | goto out; |
977 | } else if (di) { | |
e02119d5 CM |
978 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); |
979 | if (location.objectid != objectid) | |
980 | goto out; | |
77a5b9e3 | 981 | } else { |
e02119d5 | 982 | goto out; |
77a5b9e3 | 983 | } |
e02119d5 | 984 | |
77a5b9e3 | 985 | btrfs_release_path(path); |
e02119d5 | 986 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); |
77a5b9e3 FM |
987 | if (IS_ERR(di)) { |
988 | ret = PTR_ERR(di); | |
e02119d5 | 989 | goto out; |
77a5b9e3 FM |
990 | } else if (di) { |
991 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
992 | if (location.objectid == objectid) | |
993 | ret = 1; | |
994 | } | |
e02119d5 | 995 | out: |
b3b4aa74 | 996 | btrfs_release_path(path); |
77a5b9e3 | 997 | return ret; |
e02119d5 CM |
998 | } |
999 | ||
1000 | /* | |
1001 | * helper function to check a log tree for a named back reference in | |
1002 | * an inode. This is used to decide if a back reference that is | |
1003 | * found in the subvolume conflicts with what we find in the log. | |
1004 | * | |
1005 | * inode backreferences may have multiple refs in a single item, | |
1006 | * during replay we process one reference at a time, and we don't | |
1007 | * want to delete valid links to a file from the subvolume if that | |
1008 | * link is also in the log. | |
1009 | */ | |
1010 | static noinline int backref_in_log(struct btrfs_root *log, | |
1011 | struct btrfs_key *key, | |
f186373f | 1012 | u64 ref_objectid, |
df8d116f | 1013 | const char *name, int namelen) |
e02119d5 CM |
1014 | { |
1015 | struct btrfs_path *path; | |
e02119d5 | 1016 | int ret; |
e02119d5 CM |
1017 | |
1018 | path = btrfs_alloc_path(); | |
2a29edc6 | 1019 | if (!path) |
1020 | return -ENOMEM; | |
1021 | ||
e02119d5 | 1022 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
d3316c82 NB |
1023 | if (ret < 0) { |
1024 | goto out; | |
1025 | } else if (ret == 1) { | |
89cbf5f6 | 1026 | ret = 0; |
f186373f MF |
1027 | goto out; |
1028 | } | |
1029 | ||
89cbf5f6 NB |
1030 | if (key->type == BTRFS_INODE_EXTREF_KEY) |
1031 | ret = !!btrfs_find_name_in_ext_backref(path->nodes[0], | |
1032 | path->slots[0], | |
1033 | ref_objectid, | |
1034 | name, namelen); | |
1035 | else | |
1036 | ret = !!btrfs_find_name_in_backref(path->nodes[0], | |
1037 | path->slots[0], | |
1038 | name, namelen); | |
e02119d5 CM |
1039 | out: |
1040 | btrfs_free_path(path); | |
89cbf5f6 | 1041 | return ret; |
e02119d5 CM |
1042 | } |
1043 | ||
5a1d7843 | 1044 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 1045 | struct btrfs_root *root, |
e02119d5 | 1046 | struct btrfs_path *path, |
5a1d7843 | 1047 | struct btrfs_root *log_root, |
94c91a1f NB |
1048 | struct btrfs_inode *dir, |
1049 | struct btrfs_inode *inode, | |
f186373f MF |
1050 | u64 inode_objectid, u64 parent_objectid, |
1051 | u64 ref_index, char *name, int namelen, | |
1052 | int *search_done) | |
e02119d5 | 1053 | { |
34f3e4f2 | 1054 | int ret; |
f186373f MF |
1055 | char *victim_name; |
1056 | int victim_name_len; | |
1057 | struct extent_buffer *leaf; | |
5a1d7843 | 1058 | struct btrfs_dir_item *di; |
f186373f MF |
1059 | struct btrfs_key search_key; |
1060 | struct btrfs_inode_extref *extref; | |
c622ae60 | 1061 | |
f186373f MF |
1062 | again: |
1063 | /* Search old style refs */ | |
1064 | search_key.objectid = inode_objectid; | |
1065 | search_key.type = BTRFS_INODE_REF_KEY; | |
1066 | search_key.offset = parent_objectid; | |
1067 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 1068 | if (ret == 0) { |
e02119d5 CM |
1069 | struct btrfs_inode_ref *victim_ref; |
1070 | unsigned long ptr; | |
1071 | unsigned long ptr_end; | |
f186373f MF |
1072 | |
1073 | leaf = path->nodes[0]; | |
e02119d5 CM |
1074 | |
1075 | /* are we trying to overwrite a back ref for the root directory | |
1076 | * if so, just jump out, we're done | |
1077 | */ | |
f186373f | 1078 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1079 | return 1; |
e02119d5 CM |
1080 | |
1081 | /* check all the names in this back reference to see | |
1082 | * if they are in the log. if so, we allow them to stay | |
1083 | * otherwise they must be unlinked as a conflict | |
1084 | */ | |
1085 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3212fa14 | 1086 | ptr_end = ptr + btrfs_item_size(leaf, path->slots[0]); |
d397712b | 1087 | while (ptr < ptr_end) { |
e02119d5 CM |
1088 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1089 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1090 | victim_ref); | |
1091 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1092 | if (!victim_name) |
1093 | return -ENOMEM; | |
e02119d5 CM |
1094 | |
1095 | read_extent_buffer(leaf, victim_name, | |
1096 | (unsigned long)(victim_ref + 1), | |
1097 | victim_name_len); | |
1098 | ||
d3316c82 NB |
1099 | ret = backref_in_log(log_root, &search_key, |
1100 | parent_objectid, victim_name, | |
1101 | victim_name_len); | |
1102 | if (ret < 0) { | |
1103 | kfree(victim_name); | |
1104 | return ret; | |
1105 | } else if (!ret) { | |
94c91a1f | 1106 | inc_nlink(&inode->vfs_inode); |
b3b4aa74 | 1107 | btrfs_release_path(path); |
12fcfd22 | 1108 | |
4467af88 | 1109 | ret = btrfs_unlink_inode(trans, dir, inode, |
4ec5934e | 1110 | victim_name, victim_name_len); |
f186373f | 1111 | kfree(victim_name); |
3650860b JB |
1112 | if (ret) |
1113 | return ret; | |
e5c304e6 | 1114 | ret = btrfs_run_delayed_items(trans); |
ada9af21 FDBM |
1115 | if (ret) |
1116 | return ret; | |
f186373f MF |
1117 | *search_done = 1; |
1118 | goto again; | |
e02119d5 CM |
1119 | } |
1120 | kfree(victim_name); | |
f186373f | 1121 | |
e02119d5 CM |
1122 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1123 | } | |
e02119d5 | 1124 | |
c622ae60 | 1125 | /* |
1126 | * NOTE: we have searched root tree and checked the | |
bb7ab3b9 | 1127 | * corresponding ref, it does not need to check again. |
c622ae60 | 1128 | */ |
5a1d7843 | 1129 | *search_done = 1; |
e02119d5 | 1130 | } |
b3b4aa74 | 1131 | btrfs_release_path(path); |
e02119d5 | 1132 | |
f186373f MF |
1133 | /* Same search but for extended refs */ |
1134 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1135 | inode_objectid, parent_objectid, 0, | |
1136 | 0); | |
1137 | if (!IS_ERR_OR_NULL(extref)) { | |
1138 | u32 item_size; | |
1139 | u32 cur_offset = 0; | |
1140 | unsigned long base; | |
1141 | struct inode *victim_parent; | |
1142 | ||
1143 | leaf = path->nodes[0]; | |
1144 | ||
3212fa14 | 1145 | item_size = btrfs_item_size(leaf, path->slots[0]); |
f186373f MF |
1146 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); |
1147 | ||
1148 | while (cur_offset < item_size) { | |
dd9ef135 | 1149 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1150 | |
1151 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1152 | ||
1153 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1154 | goto next; | |
1155 | ||
1156 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1157 | if (!victim_name) |
1158 | return -ENOMEM; | |
f186373f MF |
1159 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1160 | victim_name_len); | |
1161 | ||
1162 | search_key.objectid = inode_objectid; | |
1163 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1164 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1165 | victim_name, | |
1166 | victim_name_len); | |
d3316c82 NB |
1167 | ret = backref_in_log(log_root, &search_key, |
1168 | parent_objectid, victim_name, | |
1169 | victim_name_len); | |
1170 | if (ret < 0) { | |
f35838a6 | 1171 | kfree(victim_name); |
d3316c82 NB |
1172 | return ret; |
1173 | } else if (!ret) { | |
f186373f MF |
1174 | ret = -ENOENT; |
1175 | victim_parent = read_one_inode(root, | |
94c91a1f | 1176 | parent_objectid); |
f186373f | 1177 | if (victim_parent) { |
94c91a1f | 1178 | inc_nlink(&inode->vfs_inode); |
f186373f MF |
1179 | btrfs_release_path(path); |
1180 | ||
4467af88 | 1181 | ret = btrfs_unlink_inode(trans, |
4ec5934e | 1182 | BTRFS_I(victim_parent), |
94c91a1f | 1183 | inode, |
4ec5934e NB |
1184 | victim_name, |
1185 | victim_name_len); | |
ada9af21 FDBM |
1186 | if (!ret) |
1187 | ret = btrfs_run_delayed_items( | |
e5c304e6 | 1188 | trans); |
f186373f | 1189 | } |
f186373f MF |
1190 | iput(victim_parent); |
1191 | kfree(victim_name); | |
3650860b JB |
1192 | if (ret) |
1193 | return ret; | |
f186373f MF |
1194 | *search_done = 1; |
1195 | goto again; | |
1196 | } | |
1197 | kfree(victim_name); | |
f186373f MF |
1198 | next: |
1199 | cur_offset += victim_name_len + sizeof(*extref); | |
1200 | } | |
1201 | *search_done = 1; | |
1202 | } | |
1203 | btrfs_release_path(path); | |
1204 | ||
34f3e4f2 | 1205 | /* look for a conflicting sequence number */ |
94c91a1f | 1206 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
f186373f | 1207 | ref_index, name, namelen, 0); |
52db7779 | 1208 | if (IS_ERR(di)) { |
8dcbc261 | 1209 | return PTR_ERR(di); |
52db7779 | 1210 | } else if (di) { |
9798ba24 | 1211 | ret = drop_one_dir_item(trans, path, dir, di); |
3650860b JB |
1212 | if (ret) |
1213 | return ret; | |
34f3e4f2 | 1214 | } |
1215 | btrfs_release_path(path); | |
1216 | ||
52042d8e | 1217 | /* look for a conflicting name */ |
94c91a1f | 1218 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), |
34f3e4f2 | 1219 | name, namelen, 0); |
52db7779 FM |
1220 | if (IS_ERR(di)) { |
1221 | return PTR_ERR(di); | |
1222 | } else if (di) { | |
9798ba24 | 1223 | ret = drop_one_dir_item(trans, path, dir, di); |
3650860b JB |
1224 | if (ret) |
1225 | return ret; | |
34f3e4f2 | 1226 | } |
1227 | btrfs_release_path(path); | |
1228 | ||
5a1d7843 JS |
1229 | return 0; |
1230 | } | |
e02119d5 | 1231 | |
bae15d95 QW |
1232 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1233 | u32 *namelen, char **name, u64 *index, | |
1234 | u64 *parent_objectid) | |
f186373f MF |
1235 | { |
1236 | struct btrfs_inode_extref *extref; | |
1237 | ||
1238 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1239 | ||
1240 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1241 | *name = kmalloc(*namelen, GFP_NOFS); | |
1242 | if (*name == NULL) | |
1243 | return -ENOMEM; | |
1244 | ||
1245 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1246 | *namelen); | |
1247 | ||
1f250e92 FM |
1248 | if (index) |
1249 | *index = btrfs_inode_extref_index(eb, extref); | |
f186373f MF |
1250 | if (parent_objectid) |
1251 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1252 | ||
1253 | return 0; | |
1254 | } | |
1255 | ||
bae15d95 QW |
1256 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1257 | u32 *namelen, char **name, u64 *index) | |
f186373f MF |
1258 | { |
1259 | struct btrfs_inode_ref *ref; | |
1260 | ||
1261 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1262 | ||
1263 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1264 | *name = kmalloc(*namelen, GFP_NOFS); | |
1265 | if (*name == NULL) | |
1266 | return -ENOMEM; | |
1267 | ||
1268 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1269 | ||
1f250e92 FM |
1270 | if (index) |
1271 | *index = btrfs_inode_ref_index(eb, ref); | |
f186373f MF |
1272 | |
1273 | return 0; | |
1274 | } | |
1275 | ||
1f250e92 FM |
1276 | /* |
1277 | * Take an inode reference item from the log tree and iterate all names from the | |
1278 | * inode reference item in the subvolume tree with the same key (if it exists). | |
1279 | * For any name that is not in the inode reference item from the log tree, do a | |
1280 | * proper unlink of that name (that is, remove its entry from the inode | |
1281 | * reference item and both dir index keys). | |
1282 | */ | |
1283 | static int unlink_old_inode_refs(struct btrfs_trans_handle *trans, | |
1284 | struct btrfs_root *root, | |
1285 | struct btrfs_path *path, | |
1286 | struct btrfs_inode *inode, | |
1287 | struct extent_buffer *log_eb, | |
1288 | int log_slot, | |
1289 | struct btrfs_key *key) | |
1290 | { | |
1291 | int ret; | |
1292 | unsigned long ref_ptr; | |
1293 | unsigned long ref_end; | |
1294 | struct extent_buffer *eb; | |
1295 | ||
1296 | again: | |
1297 | btrfs_release_path(path); | |
1298 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1299 | if (ret > 0) { | |
1300 | ret = 0; | |
1301 | goto out; | |
1302 | } | |
1303 | if (ret < 0) | |
1304 | goto out; | |
1305 | ||
1306 | eb = path->nodes[0]; | |
1307 | ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]); | |
3212fa14 | 1308 | ref_end = ref_ptr + btrfs_item_size(eb, path->slots[0]); |
1f250e92 FM |
1309 | while (ref_ptr < ref_end) { |
1310 | char *name = NULL; | |
1311 | int namelen; | |
1312 | u64 parent_id; | |
1313 | ||
1314 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1315 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1316 | NULL, &parent_id); | |
1317 | } else { | |
1318 | parent_id = key->offset; | |
1319 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1320 | NULL); | |
1321 | } | |
1322 | if (ret) | |
1323 | goto out; | |
1324 | ||
1325 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
6ff49c6a NB |
1326 | ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot, |
1327 | parent_id, name, | |
1328 | namelen); | |
1f250e92 | 1329 | else |
9bb8407f NB |
1330 | ret = !!btrfs_find_name_in_backref(log_eb, log_slot, |
1331 | name, namelen); | |
1f250e92 FM |
1332 | |
1333 | if (!ret) { | |
1334 | struct inode *dir; | |
1335 | ||
1336 | btrfs_release_path(path); | |
1337 | dir = read_one_inode(root, parent_id); | |
1338 | if (!dir) { | |
1339 | ret = -ENOENT; | |
1340 | kfree(name); | |
1341 | goto out; | |
1342 | } | |
4467af88 | 1343 | ret = btrfs_unlink_inode(trans, BTRFS_I(dir), |
1f250e92 FM |
1344 | inode, name, namelen); |
1345 | kfree(name); | |
1346 | iput(dir); | |
4751dc99 FM |
1347 | /* |
1348 | * Whenever we need to check if a name exists or not, we | |
1349 | * check the subvolume tree. So after an unlink we must | |
1350 | * run delayed items, so that future checks for a name | |
1351 | * during log replay see that the name does not exists | |
1352 | * anymore. | |
1353 | */ | |
1354 | if (!ret) | |
1355 | ret = btrfs_run_delayed_items(trans); | |
1f250e92 FM |
1356 | if (ret) |
1357 | goto out; | |
1358 | goto again; | |
1359 | } | |
1360 | ||
1361 | kfree(name); | |
1362 | ref_ptr += namelen; | |
1363 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1364 | ref_ptr += sizeof(struct btrfs_inode_extref); | |
1365 | else | |
1366 | ref_ptr += sizeof(struct btrfs_inode_ref); | |
1367 | } | |
1368 | ret = 0; | |
1369 | out: | |
1370 | btrfs_release_path(path); | |
1371 | return ret; | |
1372 | } | |
1373 | ||
0d836392 FM |
1374 | static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir, |
1375 | const u8 ref_type, const char *name, | |
1376 | const int namelen) | |
1377 | { | |
1378 | struct btrfs_key key; | |
1379 | struct btrfs_path *path; | |
1380 | const u64 parent_id = btrfs_ino(BTRFS_I(dir)); | |
1381 | int ret; | |
1382 | ||
1383 | path = btrfs_alloc_path(); | |
1384 | if (!path) | |
1385 | return -ENOMEM; | |
1386 | ||
1387 | key.objectid = btrfs_ino(BTRFS_I(inode)); | |
1388 | key.type = ref_type; | |
1389 | if (key.type == BTRFS_INODE_REF_KEY) | |
1390 | key.offset = parent_id; | |
1391 | else | |
1392 | key.offset = btrfs_extref_hash(parent_id, name, namelen); | |
1393 | ||
1394 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0); | |
1395 | if (ret < 0) | |
1396 | goto out; | |
1397 | if (ret > 0) { | |
1398 | ret = 0; | |
1399 | goto out; | |
1400 | } | |
1401 | if (key.type == BTRFS_INODE_EXTREF_KEY) | |
6ff49c6a NB |
1402 | ret = !!btrfs_find_name_in_ext_backref(path->nodes[0], |
1403 | path->slots[0], parent_id, name, namelen); | |
0d836392 | 1404 | else |
9bb8407f NB |
1405 | ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0], |
1406 | name, namelen); | |
0d836392 FM |
1407 | |
1408 | out: | |
1409 | btrfs_free_path(path); | |
1410 | return ret; | |
1411 | } | |
1412 | ||
6d9cc072 | 1413 | static int add_link(struct btrfs_trans_handle *trans, |
6b5fc433 FM |
1414 | struct inode *dir, struct inode *inode, const char *name, |
1415 | int namelen, u64 ref_index) | |
1416 | { | |
6d9cc072 | 1417 | struct btrfs_root *root = BTRFS_I(dir)->root; |
6b5fc433 FM |
1418 | struct btrfs_dir_item *dir_item; |
1419 | struct btrfs_key key; | |
1420 | struct btrfs_path *path; | |
1421 | struct inode *other_inode = NULL; | |
1422 | int ret; | |
1423 | ||
1424 | path = btrfs_alloc_path(); | |
1425 | if (!path) | |
1426 | return -ENOMEM; | |
1427 | ||
1428 | dir_item = btrfs_lookup_dir_item(NULL, root, path, | |
1429 | btrfs_ino(BTRFS_I(dir)), | |
1430 | name, namelen, 0); | |
1431 | if (!dir_item) { | |
1432 | btrfs_release_path(path); | |
1433 | goto add_link; | |
1434 | } else if (IS_ERR(dir_item)) { | |
1435 | ret = PTR_ERR(dir_item); | |
1436 | goto out; | |
1437 | } | |
1438 | ||
1439 | /* | |
1440 | * Our inode's dentry collides with the dentry of another inode which is | |
1441 | * in the log but not yet processed since it has a higher inode number. | |
1442 | * So delete that other dentry. | |
1443 | */ | |
1444 | btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key); | |
1445 | btrfs_release_path(path); | |
1446 | other_inode = read_one_inode(root, key.objectid); | |
1447 | if (!other_inode) { | |
1448 | ret = -ENOENT; | |
1449 | goto out; | |
1450 | } | |
4467af88 | 1451 | ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(other_inode), |
6b5fc433 FM |
1452 | name, namelen); |
1453 | if (ret) | |
1454 | goto out; | |
1455 | /* | |
1456 | * If we dropped the link count to 0, bump it so that later the iput() | |
1457 | * on the inode will not free it. We will fixup the link count later. | |
1458 | */ | |
1459 | if (other_inode->i_nlink == 0) | |
1460 | inc_nlink(other_inode); | |
1461 | ||
1462 | ret = btrfs_run_delayed_items(trans); | |
1463 | if (ret) | |
1464 | goto out; | |
1465 | add_link: | |
1466 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), | |
1467 | name, namelen, 0, ref_index); | |
1468 | out: | |
1469 | iput(other_inode); | |
1470 | btrfs_free_path(path); | |
1471 | ||
1472 | return ret; | |
1473 | } | |
1474 | ||
5a1d7843 JS |
1475 | /* |
1476 | * replay one inode back reference item found in the log tree. | |
1477 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1478 | * root is the destination we are replaying into, and path is for temp | |
1479 | * use by this function. (it should be released on return). | |
1480 | */ | |
1481 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1482 | struct btrfs_root *root, | |
1483 | struct btrfs_root *log, | |
1484 | struct btrfs_path *path, | |
1485 | struct extent_buffer *eb, int slot, | |
1486 | struct btrfs_key *key) | |
1487 | { | |
03b2f08b GB |
1488 | struct inode *dir = NULL; |
1489 | struct inode *inode = NULL; | |
5a1d7843 JS |
1490 | unsigned long ref_ptr; |
1491 | unsigned long ref_end; | |
03b2f08b | 1492 | char *name = NULL; |
5a1d7843 JS |
1493 | int namelen; |
1494 | int ret; | |
1495 | int search_done = 0; | |
f186373f MF |
1496 | int log_ref_ver = 0; |
1497 | u64 parent_objectid; | |
1498 | u64 inode_objectid; | |
f46dbe3d | 1499 | u64 ref_index = 0; |
f186373f MF |
1500 | int ref_struct_size; |
1501 | ||
1502 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
3212fa14 | 1503 | ref_end = ref_ptr + btrfs_item_size(eb, slot); |
f186373f MF |
1504 | |
1505 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1506 | struct btrfs_inode_extref *r; | |
1507 | ||
1508 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1509 | log_ref_ver = 1; | |
1510 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1511 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1512 | } else { | |
1513 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1514 | parent_objectid = key->offset; | |
1515 | } | |
1516 | inode_objectid = key->objectid; | |
e02119d5 | 1517 | |
5a1d7843 JS |
1518 | /* |
1519 | * it is possible that we didn't log all the parent directories | |
1520 | * for a given inode. If we don't find the dir, just don't | |
1521 | * copy the back ref in. The link count fixup code will take | |
1522 | * care of the rest | |
1523 | */ | |
f186373f | 1524 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1525 | if (!dir) { |
1526 | ret = -ENOENT; | |
1527 | goto out; | |
1528 | } | |
5a1d7843 | 1529 | |
f186373f | 1530 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1531 | if (!inode) { |
03b2f08b GB |
1532 | ret = -EIO; |
1533 | goto out; | |
5a1d7843 JS |
1534 | } |
1535 | ||
5a1d7843 | 1536 | while (ref_ptr < ref_end) { |
f186373f | 1537 | if (log_ref_ver) { |
bae15d95 QW |
1538 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, |
1539 | &ref_index, &parent_objectid); | |
f186373f MF |
1540 | /* |
1541 | * parent object can change from one array | |
1542 | * item to another. | |
1543 | */ | |
1544 | if (!dir) | |
1545 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1546 | if (!dir) { |
1547 | ret = -ENOENT; | |
1548 | goto out; | |
1549 | } | |
f186373f | 1550 | } else { |
bae15d95 QW |
1551 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, |
1552 | &ref_index); | |
f186373f MF |
1553 | } |
1554 | if (ret) | |
03b2f08b | 1555 | goto out; |
5a1d7843 | 1556 | |
77a5b9e3 FM |
1557 | ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), |
1558 | btrfs_ino(BTRFS_I(inode)), ref_index, | |
1559 | name, namelen); | |
1560 | if (ret < 0) { | |
1561 | goto out; | |
1562 | } else if (ret == 0) { | |
5a1d7843 JS |
1563 | /* |
1564 | * look for a conflicting back reference in the | |
1565 | * metadata. if we find one we have to unlink that name | |
1566 | * of the file before we add our new link. Later on, we | |
1567 | * overwrite any existing back reference, and we don't | |
1568 | * want to create dangling pointers in the directory. | |
1569 | */ | |
1570 | ||
1571 | if (!search_done) { | |
1572 | ret = __add_inode_ref(trans, root, path, log, | |
94c91a1f | 1573 | BTRFS_I(dir), |
d75eefdf | 1574 | BTRFS_I(inode), |
f186373f MF |
1575 | inode_objectid, |
1576 | parent_objectid, | |
1577 | ref_index, name, namelen, | |
5a1d7843 | 1578 | &search_done); |
03b2f08b GB |
1579 | if (ret) { |
1580 | if (ret == 1) | |
1581 | ret = 0; | |
3650860b JB |
1582 | goto out; |
1583 | } | |
5a1d7843 JS |
1584 | } |
1585 | ||
0d836392 FM |
1586 | /* |
1587 | * If a reference item already exists for this inode | |
1588 | * with the same parent and name, but different index, | |
1589 | * drop it and the corresponding directory index entries | |
1590 | * from the parent before adding the new reference item | |
1591 | * and dir index entries, otherwise we would fail with | |
1592 | * -EEXIST returned from btrfs_add_link() below. | |
1593 | */ | |
1594 | ret = btrfs_inode_ref_exists(inode, dir, key->type, | |
1595 | name, namelen); | |
1596 | if (ret > 0) { | |
4467af88 | 1597 | ret = btrfs_unlink_inode(trans, |
0d836392 FM |
1598 | BTRFS_I(dir), |
1599 | BTRFS_I(inode), | |
1600 | name, namelen); | |
1601 | /* | |
1602 | * If we dropped the link count to 0, bump it so | |
1603 | * that later the iput() on the inode will not | |
1604 | * free it. We will fixup the link count later. | |
1605 | */ | |
1606 | if (!ret && inode->i_nlink == 0) | |
1607 | inc_nlink(inode); | |
4751dc99 FM |
1608 | /* |
1609 | * Whenever we need to check if a name exists or | |
1610 | * not, we check the subvolume tree. So after an | |
1611 | * unlink we must run delayed items, so that future | |
1612 | * checks for a name during log replay see that the | |
1613 | * name does not exists anymore. | |
1614 | */ | |
1615 | if (!ret) | |
1616 | ret = btrfs_run_delayed_items(trans); | |
0d836392 FM |
1617 | } |
1618 | if (ret < 0) | |
1619 | goto out; | |
1620 | ||
5a1d7843 | 1621 | /* insert our name */ |
6d9cc072 | 1622 | ret = add_link(trans, dir, inode, name, namelen, |
6b5fc433 | 1623 | ref_index); |
3650860b JB |
1624 | if (ret) |
1625 | goto out; | |
5a1d7843 | 1626 | |
f96d4474 JB |
1627 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
1628 | if (ret) | |
1629 | goto out; | |
5a1d7843 | 1630 | } |
77a5b9e3 | 1631 | /* Else, ret == 1, we already have a perfect match, we're done. */ |
5a1d7843 | 1632 | |
f186373f | 1633 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1634 | kfree(name); |
03b2f08b | 1635 | name = NULL; |
f186373f MF |
1636 | if (log_ref_ver) { |
1637 | iput(dir); | |
1638 | dir = NULL; | |
1639 | } | |
5a1d7843 | 1640 | } |
e02119d5 | 1641 | |
1f250e92 FM |
1642 | /* |
1643 | * Before we overwrite the inode reference item in the subvolume tree | |
1644 | * with the item from the log tree, we must unlink all names from the | |
1645 | * parent directory that are in the subvolume's tree inode reference | |
1646 | * item, otherwise we end up with an inconsistent subvolume tree where | |
1647 | * dir index entries exist for a name but there is no inode reference | |
1648 | * item with the same name. | |
1649 | */ | |
1650 | ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot, | |
1651 | key); | |
1652 | if (ret) | |
1653 | goto out; | |
1654 | ||
e02119d5 CM |
1655 | /* finally write the back reference in the inode */ |
1656 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1657 | out: |
b3b4aa74 | 1658 | btrfs_release_path(path); |
03b2f08b | 1659 | kfree(name); |
e02119d5 CM |
1660 | iput(dir); |
1661 | iput(inode); | |
3650860b | 1662 | return ret; |
e02119d5 CM |
1663 | } |
1664 | ||
f186373f | 1665 | static int count_inode_extrefs(struct btrfs_root *root, |
36283658 | 1666 | struct btrfs_inode *inode, struct btrfs_path *path) |
f186373f MF |
1667 | { |
1668 | int ret = 0; | |
1669 | int name_len; | |
1670 | unsigned int nlink = 0; | |
1671 | u32 item_size; | |
1672 | u32 cur_offset = 0; | |
36283658 | 1673 | u64 inode_objectid = btrfs_ino(inode); |
f186373f MF |
1674 | u64 offset = 0; |
1675 | unsigned long ptr; | |
1676 | struct btrfs_inode_extref *extref; | |
1677 | struct extent_buffer *leaf; | |
1678 | ||
1679 | while (1) { | |
1680 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1681 | &extref, &offset); | |
1682 | if (ret) | |
1683 | break; | |
c71bf099 | 1684 | |
f186373f | 1685 | leaf = path->nodes[0]; |
3212fa14 | 1686 | item_size = btrfs_item_size(leaf, path->slots[0]); |
f186373f | 1687 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); |
2c2c452b | 1688 | cur_offset = 0; |
f186373f MF |
1689 | |
1690 | while (cur_offset < item_size) { | |
1691 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1692 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1693 | ||
1694 | nlink++; | |
1695 | ||
1696 | cur_offset += name_len + sizeof(*extref); | |
1697 | } | |
1698 | ||
1699 | offset++; | |
1700 | btrfs_release_path(path); | |
1701 | } | |
1702 | btrfs_release_path(path); | |
1703 | ||
2c2c452b | 1704 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1705 | return ret; |
1706 | return nlink; | |
1707 | } | |
1708 | ||
1709 | static int count_inode_refs(struct btrfs_root *root, | |
f329e319 | 1710 | struct btrfs_inode *inode, struct btrfs_path *path) |
e02119d5 | 1711 | { |
e02119d5 CM |
1712 | int ret; |
1713 | struct btrfs_key key; | |
f186373f | 1714 | unsigned int nlink = 0; |
e02119d5 CM |
1715 | unsigned long ptr; |
1716 | unsigned long ptr_end; | |
1717 | int name_len; | |
f329e319 | 1718 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1719 | |
33345d01 | 1720 | key.objectid = ino; |
e02119d5 CM |
1721 | key.type = BTRFS_INODE_REF_KEY; |
1722 | key.offset = (u64)-1; | |
1723 | ||
d397712b | 1724 | while (1) { |
e02119d5 CM |
1725 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1726 | if (ret < 0) | |
1727 | break; | |
1728 | if (ret > 0) { | |
1729 | if (path->slots[0] == 0) | |
1730 | break; | |
1731 | path->slots[0]--; | |
1732 | } | |
e93ae26f | 1733 | process_slot: |
e02119d5 CM |
1734 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1735 | path->slots[0]); | |
33345d01 | 1736 | if (key.objectid != ino || |
e02119d5 CM |
1737 | key.type != BTRFS_INODE_REF_KEY) |
1738 | break; | |
1739 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
3212fa14 | 1740 | ptr_end = ptr + btrfs_item_size(path->nodes[0], |
e02119d5 | 1741 | path->slots[0]); |
d397712b | 1742 | while (ptr < ptr_end) { |
e02119d5 CM |
1743 | struct btrfs_inode_ref *ref; |
1744 | ||
1745 | ref = (struct btrfs_inode_ref *)ptr; | |
1746 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1747 | ref); | |
1748 | ptr = (unsigned long)(ref + 1) + name_len; | |
1749 | nlink++; | |
1750 | } | |
1751 | ||
1752 | if (key.offset == 0) | |
1753 | break; | |
e93ae26f FDBM |
1754 | if (path->slots[0] > 0) { |
1755 | path->slots[0]--; | |
1756 | goto process_slot; | |
1757 | } | |
e02119d5 | 1758 | key.offset--; |
b3b4aa74 | 1759 | btrfs_release_path(path); |
e02119d5 | 1760 | } |
b3b4aa74 | 1761 | btrfs_release_path(path); |
f186373f MF |
1762 | |
1763 | return nlink; | |
1764 | } | |
1765 | ||
1766 | /* | |
1767 | * There are a few corners where the link count of the file can't | |
1768 | * be properly maintained during replay. So, instead of adding | |
1769 | * lots of complexity to the log code, we just scan the backrefs | |
1770 | * for any file that has been through replay. | |
1771 | * | |
1772 | * The scan will update the link count on the inode to reflect the | |
1773 | * number of back refs found. If it goes down to zero, the iput | |
1774 | * will free the inode. | |
1775 | */ | |
1776 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1777 | struct btrfs_root *root, | |
1778 | struct inode *inode) | |
1779 | { | |
1780 | struct btrfs_path *path; | |
1781 | int ret; | |
1782 | u64 nlink = 0; | |
4a0cc7ca | 1783 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1784 | |
1785 | path = btrfs_alloc_path(); | |
1786 | if (!path) | |
1787 | return -ENOMEM; | |
1788 | ||
f329e319 | 1789 | ret = count_inode_refs(root, BTRFS_I(inode), path); |
f186373f MF |
1790 | if (ret < 0) |
1791 | goto out; | |
1792 | ||
1793 | nlink = ret; | |
1794 | ||
36283658 | 1795 | ret = count_inode_extrefs(root, BTRFS_I(inode), path); |
f186373f MF |
1796 | if (ret < 0) |
1797 | goto out; | |
1798 | ||
1799 | nlink += ret; | |
1800 | ||
1801 | ret = 0; | |
1802 | ||
e02119d5 | 1803 | if (nlink != inode->i_nlink) { |
bfe86848 | 1804 | set_nlink(inode, nlink); |
f96d4474 JB |
1805 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
1806 | if (ret) | |
1807 | goto out; | |
e02119d5 | 1808 | } |
8d5bf1cb | 1809 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1810 | |
c71bf099 YZ |
1811 | if (inode->i_nlink == 0) { |
1812 | if (S_ISDIR(inode->i_mode)) { | |
1813 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1814 | ino, 1); |
3650860b JB |
1815 | if (ret) |
1816 | goto out; | |
c71bf099 | 1817 | } |
ecdcf3c2 NB |
1818 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1819 | if (ret == -EEXIST) | |
1820 | ret = 0; | |
12fcfd22 | 1821 | } |
12fcfd22 | 1822 | |
f186373f MF |
1823 | out: |
1824 | btrfs_free_path(path); | |
1825 | return ret; | |
e02119d5 CM |
1826 | } |
1827 | ||
1828 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1829 | struct btrfs_root *root, | |
1830 | struct btrfs_path *path) | |
1831 | { | |
1832 | int ret; | |
1833 | struct btrfs_key key; | |
1834 | struct inode *inode; | |
1835 | ||
1836 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1837 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1838 | key.offset = (u64)-1; | |
d397712b | 1839 | while (1) { |
e02119d5 CM |
1840 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1841 | if (ret < 0) | |
1842 | break; | |
1843 | ||
1844 | if (ret == 1) { | |
011b28ac | 1845 | ret = 0; |
e02119d5 CM |
1846 | if (path->slots[0] == 0) |
1847 | break; | |
1848 | path->slots[0]--; | |
1849 | } | |
1850 | ||
1851 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1852 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1853 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1854 | break; | |
1855 | ||
1856 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 | 1857 | if (ret) |
011b28ac | 1858 | break; |
e02119d5 | 1859 | |
b3b4aa74 | 1860 | btrfs_release_path(path); |
e02119d5 | 1861 | inode = read_one_inode(root, key.offset); |
011b28ac JB |
1862 | if (!inode) { |
1863 | ret = -EIO; | |
1864 | break; | |
1865 | } | |
e02119d5 CM |
1866 | |
1867 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1868 | iput(inode); |
3650860b | 1869 | if (ret) |
011b28ac | 1870 | break; |
e02119d5 | 1871 | |
12fcfd22 CM |
1872 | /* |
1873 | * fixup on a directory may create new entries, | |
1874 | * make sure we always look for the highset possible | |
1875 | * offset | |
1876 | */ | |
1877 | key.offset = (u64)-1; | |
e02119d5 | 1878 | } |
b3b4aa74 | 1879 | btrfs_release_path(path); |
65a246c5 | 1880 | return ret; |
e02119d5 CM |
1881 | } |
1882 | ||
1883 | ||
1884 | /* | |
1885 | * record a given inode in the fixup dir so we can check its link | |
1886 | * count when replay is done. The link count is incremented here | |
1887 | * so the inode won't go away until we check it | |
1888 | */ | |
1889 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1890 | struct btrfs_root *root, | |
1891 | struct btrfs_path *path, | |
1892 | u64 objectid) | |
1893 | { | |
1894 | struct btrfs_key key; | |
1895 | int ret = 0; | |
1896 | struct inode *inode; | |
1897 | ||
1898 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1899 | if (!inode) |
1900 | return -EIO; | |
e02119d5 CM |
1901 | |
1902 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1903 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1904 | key.offset = objectid; |
1905 | ||
1906 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1907 | ||
b3b4aa74 | 1908 | btrfs_release_path(path); |
e02119d5 | 1909 | if (ret == 0) { |
9bf7a489 JB |
1910 | if (!inode->i_nlink) |
1911 | set_nlink(inode, 1); | |
1912 | else | |
8b558c5f | 1913 | inc_nlink(inode); |
9a56fcd1 | 1914 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e02119d5 CM |
1915 | } else if (ret == -EEXIST) { |
1916 | ret = 0; | |
e02119d5 CM |
1917 | } |
1918 | iput(inode); | |
1919 | ||
1920 | return ret; | |
1921 | } | |
1922 | ||
1923 | /* | |
1924 | * when replaying the log for a directory, we only insert names | |
1925 | * for inodes that actually exist. This means an fsync on a directory | |
1926 | * does not implicitly fsync all the new files in it | |
1927 | */ | |
1928 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1929 | struct btrfs_root *root, | |
e02119d5 | 1930 | u64 dirid, u64 index, |
60d53eb3 | 1931 | char *name, int name_len, |
e02119d5 CM |
1932 | struct btrfs_key *location) |
1933 | { | |
1934 | struct inode *inode; | |
1935 | struct inode *dir; | |
1936 | int ret; | |
1937 | ||
1938 | inode = read_one_inode(root, location->objectid); | |
1939 | if (!inode) | |
1940 | return -ENOENT; | |
1941 | ||
1942 | dir = read_one_inode(root, dirid); | |
1943 | if (!dir) { | |
1944 | iput(inode); | |
1945 | return -EIO; | |
1946 | } | |
d555438b | 1947 | |
db0a669f NB |
1948 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, |
1949 | name_len, 1, index); | |
e02119d5 CM |
1950 | |
1951 | /* FIXME, put inode into FIXUP list */ | |
1952 | ||
1953 | iput(inode); | |
1954 | iput(dir); | |
1955 | return ret; | |
1956 | } | |
1957 | ||
339d0354 FM |
1958 | static int delete_conflicting_dir_entry(struct btrfs_trans_handle *trans, |
1959 | struct btrfs_inode *dir, | |
1960 | struct btrfs_path *path, | |
1961 | struct btrfs_dir_item *dst_di, | |
1962 | const struct btrfs_key *log_key, | |
1963 | u8 log_type, | |
1964 | bool exists) | |
1965 | { | |
1966 | struct btrfs_key found_key; | |
1967 | ||
1968 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1969 | /* The existing dentry points to the same inode, don't delete it. */ | |
1970 | if (found_key.objectid == log_key->objectid && | |
1971 | found_key.type == log_key->type && | |
1972 | found_key.offset == log_key->offset && | |
1973 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) | |
1974 | return 1; | |
1975 | ||
1976 | /* | |
1977 | * Don't drop the conflicting directory entry if the inode for the new | |
1978 | * entry doesn't exist. | |
1979 | */ | |
1980 | if (!exists) | |
1981 | return 0; | |
1982 | ||
1983 | return drop_one_dir_item(trans, path, dir, dst_di); | |
1984 | } | |
1985 | ||
e02119d5 CM |
1986 | /* |
1987 | * take a single entry in a log directory item and replay it into | |
1988 | * the subvolume. | |
1989 | * | |
1990 | * if a conflicting item exists in the subdirectory already, | |
1991 | * the inode it points to is unlinked and put into the link count | |
1992 | * fix up tree. | |
1993 | * | |
1994 | * If a name from the log points to a file or directory that does | |
1995 | * not exist in the FS, it is skipped. fsyncs on directories | |
1996 | * do not force down inodes inside that directory, just changes to the | |
1997 | * names or unlinks in a directory. | |
bb53eda9 FM |
1998 | * |
1999 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
2000 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
2001 | */ |
2002 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
2003 | struct btrfs_root *root, | |
2004 | struct btrfs_path *path, | |
2005 | struct extent_buffer *eb, | |
2006 | struct btrfs_dir_item *di, | |
2007 | struct btrfs_key *key) | |
2008 | { | |
2009 | char *name; | |
2010 | int name_len; | |
339d0354 FM |
2011 | struct btrfs_dir_item *dir_dst_di; |
2012 | struct btrfs_dir_item *index_dst_di; | |
2013 | bool dir_dst_matches = false; | |
2014 | bool index_dst_matches = false; | |
e02119d5 | 2015 | struct btrfs_key log_key; |
339d0354 | 2016 | struct btrfs_key search_key; |
e02119d5 | 2017 | struct inode *dir; |
e02119d5 | 2018 | u8 log_type; |
cfd31269 FM |
2019 | bool exists; |
2020 | int ret; | |
339d0354 | 2021 | bool update_size = true; |
bb53eda9 | 2022 | bool name_added = false; |
e02119d5 CM |
2023 | |
2024 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
2025 | if (!dir) |
2026 | return -EIO; | |
e02119d5 CM |
2027 | |
2028 | name_len = btrfs_dir_name_len(eb, di); | |
2029 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
2030 | if (!name) { |
2031 | ret = -ENOMEM; | |
2032 | goto out; | |
2033 | } | |
2a29edc6 | 2034 | |
e02119d5 CM |
2035 | log_type = btrfs_dir_type(eb, di); |
2036 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2037 | name_len); | |
2038 | ||
2039 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
cfd31269 | 2040 | ret = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
b3b4aa74 | 2041 | btrfs_release_path(path); |
cfd31269 FM |
2042 | if (ret < 0) |
2043 | goto out; | |
2044 | exists = (ret == 0); | |
2045 | ret = 0; | |
4bef0848 | 2046 | |
339d0354 FM |
2047 | dir_dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, |
2048 | name, name_len, 1); | |
2049 | if (IS_ERR(dir_dst_di)) { | |
2050 | ret = PTR_ERR(dir_dst_di); | |
3650860b | 2051 | goto out; |
339d0354 FM |
2052 | } else if (dir_dst_di) { |
2053 | ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path, | |
2054 | dir_dst_di, &log_key, log_type, | |
2055 | exists); | |
2056 | if (ret < 0) | |
2057 | goto out; | |
2058 | dir_dst_matches = (ret == 1); | |
e02119d5 | 2059 | } |
e15ac641 | 2060 | |
339d0354 FM |
2061 | btrfs_release_path(path); |
2062 | ||
2063 | index_dst_di = btrfs_lookup_dir_index_item(trans, root, path, | |
2064 | key->objectid, key->offset, | |
2065 | name, name_len, 1); | |
2066 | if (IS_ERR(index_dst_di)) { | |
2067 | ret = PTR_ERR(index_dst_di); | |
e15ac641 | 2068 | goto out; |
339d0354 FM |
2069 | } else if (index_dst_di) { |
2070 | ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path, | |
2071 | index_dst_di, &log_key, | |
2072 | log_type, exists); | |
2073 | if (ret < 0) | |
e02119d5 | 2074 | goto out; |
339d0354 | 2075 | index_dst_matches = (ret == 1); |
e02119d5 CM |
2076 | } |
2077 | ||
339d0354 FM |
2078 | btrfs_release_path(path); |
2079 | ||
2080 | if (dir_dst_matches && index_dst_matches) { | |
2081 | ret = 0; | |
a2cc11db | 2082 | update_size = false; |
e02119d5 CM |
2083 | goto out; |
2084 | } | |
2085 | ||
725af92a NB |
2086 | /* |
2087 | * Check if the inode reference exists in the log for the given name, | |
2088 | * inode and parent inode | |
2089 | */ | |
339d0354 FM |
2090 | search_key.objectid = log_key.objectid; |
2091 | search_key.type = BTRFS_INODE_REF_KEY; | |
2092 | search_key.offset = key->objectid; | |
2093 | ret = backref_in_log(root->log_root, &search_key, 0, name, name_len); | |
725af92a NB |
2094 | if (ret < 0) { |
2095 | goto out; | |
2096 | } else if (ret) { | |
2097 | /* The dentry will be added later. */ | |
2098 | ret = 0; | |
2099 | update_size = false; | |
2100 | goto out; | |
2101 | } | |
2102 | ||
339d0354 FM |
2103 | search_key.objectid = log_key.objectid; |
2104 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
2105 | search_key.offset = key->objectid; | |
2106 | ret = backref_in_log(root->log_root, &search_key, key->objectid, name, | |
725af92a NB |
2107 | name_len); |
2108 | if (ret < 0) { | |
2109 | goto out; | |
2110 | } else if (ret) { | |
df8d116f FM |
2111 | /* The dentry will be added later. */ |
2112 | ret = 0; | |
2113 | update_size = false; | |
2114 | goto out; | |
2115 | } | |
b3b4aa74 | 2116 | btrfs_release_path(path); |
60d53eb3 Z |
2117 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
2118 | name, name_len, &log_key); | |
df8d116f | 2119 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 2120 | goto out; |
bb53eda9 FM |
2121 | if (!ret) |
2122 | name_added = true; | |
d555438b | 2123 | update_size = false; |
3650860b | 2124 | ret = 0; |
339d0354 FM |
2125 | |
2126 | out: | |
2127 | if (!ret && update_size) { | |
2128 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2); | |
2129 | ret = btrfs_update_inode(trans, root, BTRFS_I(dir)); | |
2130 | } | |
2131 | kfree(name); | |
2132 | iput(dir); | |
2133 | if (!ret && name_added) | |
2134 | ret = 1; | |
2135 | return ret; | |
e02119d5 CM |
2136 | } |
2137 | ||
339d0354 | 2138 | /* Replay one dir item from a BTRFS_DIR_INDEX_KEY key. */ |
e02119d5 CM |
2139 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, |
2140 | struct btrfs_root *root, | |
2141 | struct btrfs_path *path, | |
2142 | struct extent_buffer *eb, int slot, | |
2143 | struct btrfs_key *key) | |
2144 | { | |
339d0354 | 2145 | int ret; |
e02119d5 | 2146 | struct btrfs_dir_item *di; |
e02119d5 | 2147 | |
339d0354 FM |
2148 | /* We only log dir index keys, which only contain a single dir item. */ |
2149 | ASSERT(key->type == BTRFS_DIR_INDEX_KEY); | |
bb53eda9 | 2150 | |
339d0354 FM |
2151 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); |
2152 | ret = replay_one_name(trans, root, path, eb, di, key); | |
2153 | if (ret < 0) | |
2154 | return ret; | |
bb53eda9 | 2155 | |
339d0354 FM |
2156 | /* |
2157 | * If this entry refers to a non-directory (directories can not have a | |
2158 | * link count > 1) and it was added in the transaction that was not | |
2159 | * committed, make sure we fixup the link count of the inode the entry | |
2160 | * points to. Otherwise something like the following would result in a | |
2161 | * directory pointing to an inode with a wrong link that does not account | |
2162 | * for this dir entry: | |
2163 | * | |
2164 | * mkdir testdir | |
2165 | * touch testdir/foo | |
2166 | * touch testdir/bar | |
2167 | * sync | |
2168 | * | |
2169 | * ln testdir/bar testdir/bar_link | |
2170 | * ln testdir/foo testdir/foo_link | |
2171 | * xfs_io -c "fsync" testdir/bar | |
2172 | * | |
2173 | * <power failure> | |
2174 | * | |
2175 | * mount fs, log replay happens | |
2176 | * | |
2177 | * File foo would remain with a link count of 1 when it has two entries | |
2178 | * pointing to it in the directory testdir. This would make it impossible | |
2179 | * to ever delete the parent directory has it would result in stale | |
2180 | * dentries that can never be deleted. | |
2181 | */ | |
2182 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
2183 | struct btrfs_path *fixup_path; | |
2184 | struct btrfs_key di_key; | |
bb53eda9 | 2185 | |
339d0354 FM |
2186 | fixup_path = btrfs_alloc_path(); |
2187 | if (!fixup_path) | |
2188 | return -ENOMEM; | |
2189 | ||
2190 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2191 | ret = link_to_fixup_dir(trans, root, fixup_path, di_key.objectid); | |
2192 | btrfs_free_path(fixup_path); | |
e02119d5 | 2193 | } |
339d0354 | 2194 | |
bb53eda9 | 2195 | return ret; |
e02119d5 CM |
2196 | } |
2197 | ||
2198 | /* | |
2199 | * directory replay has two parts. There are the standard directory | |
2200 | * items in the log copied from the subvolume, and range items | |
2201 | * created in the log while the subvolume was logged. | |
2202 | * | |
2203 | * The range items tell us which parts of the key space the log | |
2204 | * is authoritative for. During replay, if a key in the subvolume | |
2205 | * directory is in a logged range item, but not actually in the log | |
2206 | * that means it was deleted from the directory before the fsync | |
2207 | * and should be removed. | |
2208 | */ | |
2209 | static noinline int find_dir_range(struct btrfs_root *root, | |
2210 | struct btrfs_path *path, | |
ccae4a19 | 2211 | u64 dirid, |
e02119d5 CM |
2212 | u64 *start_ret, u64 *end_ret) |
2213 | { | |
2214 | struct btrfs_key key; | |
2215 | u64 found_end; | |
2216 | struct btrfs_dir_log_item *item; | |
2217 | int ret; | |
2218 | int nritems; | |
2219 | ||
2220 | if (*start_ret == (u64)-1) | |
2221 | return 1; | |
2222 | ||
2223 | key.objectid = dirid; | |
ccae4a19 | 2224 | key.type = BTRFS_DIR_LOG_INDEX_KEY; |
e02119d5 CM |
2225 | key.offset = *start_ret; |
2226 | ||
2227 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2228 | if (ret < 0) | |
2229 | goto out; | |
2230 | if (ret > 0) { | |
2231 | if (path->slots[0] == 0) | |
2232 | goto out; | |
2233 | path->slots[0]--; | |
2234 | } | |
2235 | if (ret != 0) | |
2236 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2237 | ||
ccae4a19 | 2238 | if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) { |
e02119d5 CM |
2239 | ret = 1; |
2240 | goto next; | |
2241 | } | |
2242 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2243 | struct btrfs_dir_log_item); | |
2244 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2245 | ||
2246 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
2247 | ret = 0; | |
2248 | *start_ret = key.offset; | |
2249 | *end_ret = found_end; | |
2250 | goto out; | |
2251 | } | |
2252 | ret = 1; | |
2253 | next: | |
2254 | /* check the next slot in the tree to see if it is a valid item */ | |
2255 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 2256 | path->slots[0]++; |
e02119d5 CM |
2257 | if (path->slots[0] >= nritems) { |
2258 | ret = btrfs_next_leaf(root, path); | |
2259 | if (ret) | |
2260 | goto out; | |
e02119d5 CM |
2261 | } |
2262 | ||
2263 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2264 | ||
ccae4a19 | 2265 | if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) { |
e02119d5 CM |
2266 | ret = 1; |
2267 | goto out; | |
2268 | } | |
2269 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2270 | struct btrfs_dir_log_item); | |
2271 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2272 | *start_ret = key.offset; | |
2273 | *end_ret = found_end; | |
2274 | ret = 0; | |
2275 | out: | |
b3b4aa74 | 2276 | btrfs_release_path(path); |
e02119d5 CM |
2277 | return ret; |
2278 | } | |
2279 | ||
2280 | /* | |
2281 | * this looks for a given directory item in the log. If the directory | |
2282 | * item is not in the log, the item is removed and the inode it points | |
2283 | * to is unlinked | |
2284 | */ | |
2285 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
e02119d5 CM |
2286 | struct btrfs_root *log, |
2287 | struct btrfs_path *path, | |
2288 | struct btrfs_path *log_path, | |
2289 | struct inode *dir, | |
2290 | struct btrfs_key *dir_key) | |
2291 | { | |
d1ed82f3 | 2292 | struct btrfs_root *root = BTRFS_I(dir)->root; |
e02119d5 CM |
2293 | int ret; |
2294 | struct extent_buffer *eb; | |
2295 | int slot; | |
e02119d5 | 2296 | struct btrfs_dir_item *di; |
e02119d5 | 2297 | int name_len; |
e02119d5 | 2298 | char *name; |
ccae4a19 | 2299 | struct inode *inode = NULL; |
e02119d5 CM |
2300 | struct btrfs_key location; |
2301 | ||
ccae4a19 FM |
2302 | /* |
2303 | * Currenly we only log dir index keys. Even if we replay a log created | |
2304 | * by an older kernel that logged both dir index and dir item keys, all | |
2305 | * we need to do is process the dir index keys, we (and our caller) can | |
2306 | * safely ignore dir item keys (key type BTRFS_DIR_ITEM_KEY). | |
2307 | */ | |
2308 | ASSERT(dir_key->type == BTRFS_DIR_INDEX_KEY); | |
2309 | ||
e02119d5 CM |
2310 | eb = path->nodes[0]; |
2311 | slot = path->slots[0]; | |
ccae4a19 FM |
2312 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); |
2313 | name_len = btrfs_dir_name_len(eb, di); | |
2314 | name = kmalloc(name_len, GFP_NOFS); | |
2315 | if (!name) { | |
2316 | ret = -ENOMEM; | |
2317 | goto out; | |
2318 | } | |
e02119d5 | 2319 | |
ccae4a19 | 2320 | read_extent_buffer(eb, name, (unsigned long)(di + 1), name_len); |
3650860b | 2321 | |
ccae4a19 FM |
2322 | if (log) { |
2323 | struct btrfs_dir_item *log_di; | |
e02119d5 | 2324 | |
ccae4a19 FM |
2325 | log_di = btrfs_lookup_dir_index_item(trans, log, log_path, |
2326 | dir_key->objectid, | |
2327 | dir_key->offset, | |
2328 | name, name_len, 0); | |
2329 | if (IS_ERR(log_di)) { | |
2330 | ret = PTR_ERR(log_di); | |
2331 | goto out; | |
2332 | } else if (log_di) { | |
2333 | /* The dentry exists in the log, we have nothing to do. */ | |
e02119d5 CM |
2334 | ret = 0; |
2335 | goto out; | |
2336 | } | |
ccae4a19 | 2337 | } |
e02119d5 | 2338 | |
ccae4a19 FM |
2339 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
2340 | btrfs_release_path(path); | |
2341 | btrfs_release_path(log_path); | |
2342 | inode = read_one_inode(root, location.objectid); | |
2343 | if (!inode) { | |
2344 | ret = -EIO; | |
2345 | goto out; | |
e02119d5 | 2346 | } |
ccae4a19 FM |
2347 | |
2348 | ret = link_to_fixup_dir(trans, root, path, location.objectid); | |
2349 | if (ret) | |
2350 | goto out; | |
2351 | ||
2352 | inc_nlink(inode); | |
2353 | ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(inode), name, | |
2354 | name_len); | |
2355 | if (ret) | |
2356 | goto out; | |
2357 | ||
2358 | ret = btrfs_run_delayed_items(trans); | |
2359 | if (ret) | |
2360 | goto out; | |
2361 | ||
2362 | /* | |
2363 | * Unlike dir item keys, dir index keys can only have one name (entry) in | |
2364 | * them, as there are no key collisions since each key has a unique offset | |
2365 | * (an index number), so we're done. | |
2366 | */ | |
e02119d5 | 2367 | out: |
b3b4aa74 DS |
2368 | btrfs_release_path(path); |
2369 | btrfs_release_path(log_path); | |
ccae4a19 FM |
2370 | kfree(name); |
2371 | iput(inode); | |
e02119d5 CM |
2372 | return ret; |
2373 | } | |
2374 | ||
4f764e51 FM |
2375 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2376 | struct btrfs_root *root, | |
2377 | struct btrfs_root *log, | |
2378 | struct btrfs_path *path, | |
2379 | const u64 ino) | |
2380 | { | |
2381 | struct btrfs_key search_key; | |
2382 | struct btrfs_path *log_path; | |
2383 | int i; | |
2384 | int nritems; | |
2385 | int ret; | |
2386 | ||
2387 | log_path = btrfs_alloc_path(); | |
2388 | if (!log_path) | |
2389 | return -ENOMEM; | |
2390 | ||
2391 | search_key.objectid = ino; | |
2392 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2393 | search_key.offset = 0; | |
2394 | again: | |
2395 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2396 | if (ret < 0) | |
2397 | goto out; | |
2398 | process_leaf: | |
2399 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2400 | for (i = path->slots[0]; i < nritems; i++) { | |
2401 | struct btrfs_key key; | |
2402 | struct btrfs_dir_item *di; | |
2403 | struct btrfs_dir_item *log_di; | |
2404 | u32 total_size; | |
2405 | u32 cur; | |
2406 | ||
2407 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2408 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2409 | ret = 0; | |
2410 | goto out; | |
2411 | } | |
2412 | ||
2413 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
3212fa14 | 2414 | total_size = btrfs_item_size(path->nodes[0], i); |
4f764e51 FM |
2415 | cur = 0; |
2416 | while (cur < total_size) { | |
2417 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2418 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2419 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2420 | char *name; | |
2421 | ||
2422 | name = kmalloc(name_len, GFP_NOFS); | |
2423 | if (!name) { | |
2424 | ret = -ENOMEM; | |
2425 | goto out; | |
2426 | } | |
2427 | read_extent_buffer(path->nodes[0], name, | |
2428 | (unsigned long)(di + 1), name_len); | |
2429 | ||
2430 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2431 | name, name_len, 0); | |
2432 | btrfs_release_path(log_path); | |
2433 | if (!log_di) { | |
2434 | /* Doesn't exist in log tree, so delete it. */ | |
2435 | btrfs_release_path(path); | |
2436 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2437 | name, name_len, -1); | |
2438 | kfree(name); | |
2439 | if (IS_ERR(di)) { | |
2440 | ret = PTR_ERR(di); | |
2441 | goto out; | |
2442 | } | |
2443 | ASSERT(di); | |
2444 | ret = btrfs_delete_one_dir_name(trans, root, | |
2445 | path, di); | |
2446 | if (ret) | |
2447 | goto out; | |
2448 | btrfs_release_path(path); | |
2449 | search_key = key; | |
2450 | goto again; | |
2451 | } | |
2452 | kfree(name); | |
2453 | if (IS_ERR(log_di)) { | |
2454 | ret = PTR_ERR(log_di); | |
2455 | goto out; | |
2456 | } | |
2457 | cur += this_len; | |
2458 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2459 | } | |
2460 | } | |
2461 | ret = btrfs_next_leaf(root, path); | |
2462 | if (ret > 0) | |
2463 | ret = 0; | |
2464 | else if (ret == 0) | |
2465 | goto process_leaf; | |
2466 | out: | |
2467 | btrfs_free_path(log_path); | |
2468 | btrfs_release_path(path); | |
2469 | return ret; | |
2470 | } | |
2471 | ||
2472 | ||
e02119d5 CM |
2473 | /* |
2474 | * deletion replay happens before we copy any new directory items | |
2475 | * out of the log or out of backreferences from inodes. It | |
2476 | * scans the log to find ranges of keys that log is authoritative for, | |
2477 | * and then scans the directory to find items in those ranges that are | |
2478 | * not present in the log. | |
2479 | * | |
2480 | * Anything we don't find in the log is unlinked and removed from the | |
2481 | * directory. | |
2482 | */ | |
2483 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2484 | struct btrfs_root *root, | |
2485 | struct btrfs_root *log, | |
2486 | struct btrfs_path *path, | |
12fcfd22 | 2487 | u64 dirid, int del_all) |
e02119d5 CM |
2488 | { |
2489 | u64 range_start; | |
2490 | u64 range_end; | |
e02119d5 CM |
2491 | int ret = 0; |
2492 | struct btrfs_key dir_key; | |
2493 | struct btrfs_key found_key; | |
2494 | struct btrfs_path *log_path; | |
2495 | struct inode *dir; | |
2496 | ||
2497 | dir_key.objectid = dirid; | |
ccae4a19 | 2498 | dir_key.type = BTRFS_DIR_INDEX_KEY; |
e02119d5 CM |
2499 | log_path = btrfs_alloc_path(); |
2500 | if (!log_path) | |
2501 | return -ENOMEM; | |
2502 | ||
2503 | dir = read_one_inode(root, dirid); | |
2504 | /* it isn't an error if the inode isn't there, that can happen | |
2505 | * because we replay the deletes before we copy in the inode item | |
2506 | * from the log | |
2507 | */ | |
2508 | if (!dir) { | |
2509 | btrfs_free_path(log_path); | |
2510 | return 0; | |
2511 | } | |
ccae4a19 | 2512 | |
e02119d5 CM |
2513 | range_start = 0; |
2514 | range_end = 0; | |
d397712b | 2515 | while (1) { |
12fcfd22 CM |
2516 | if (del_all) |
2517 | range_end = (u64)-1; | |
2518 | else { | |
ccae4a19 | 2519 | ret = find_dir_range(log, path, dirid, |
12fcfd22 | 2520 | &range_start, &range_end); |
10adb115 FM |
2521 | if (ret < 0) |
2522 | goto out; | |
2523 | else if (ret > 0) | |
12fcfd22 CM |
2524 | break; |
2525 | } | |
e02119d5 CM |
2526 | |
2527 | dir_key.offset = range_start; | |
d397712b | 2528 | while (1) { |
e02119d5 CM |
2529 | int nritems; |
2530 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2531 | 0, 0); | |
2532 | if (ret < 0) | |
2533 | goto out; | |
2534 | ||
2535 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2536 | if (path->slots[0] >= nritems) { | |
2537 | ret = btrfs_next_leaf(root, path); | |
b98def7c | 2538 | if (ret == 1) |
e02119d5 | 2539 | break; |
b98def7c LB |
2540 | else if (ret < 0) |
2541 | goto out; | |
e02119d5 CM |
2542 | } |
2543 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2544 | path->slots[0]); | |
2545 | if (found_key.objectid != dirid || | |
ccae4a19 FM |
2546 | found_key.type != dir_key.type) { |
2547 | ret = 0; | |
2548 | goto out; | |
2549 | } | |
e02119d5 CM |
2550 | |
2551 | if (found_key.offset > range_end) | |
2552 | break; | |
2553 | ||
d1ed82f3 | 2554 | ret = check_item_in_log(trans, log, path, |
12fcfd22 CM |
2555 | log_path, dir, |
2556 | &found_key); | |
3650860b JB |
2557 | if (ret) |
2558 | goto out; | |
e02119d5 CM |
2559 | if (found_key.offset == (u64)-1) |
2560 | break; | |
2561 | dir_key.offset = found_key.offset + 1; | |
2562 | } | |
b3b4aa74 | 2563 | btrfs_release_path(path); |
e02119d5 CM |
2564 | if (range_end == (u64)-1) |
2565 | break; | |
2566 | range_start = range_end + 1; | |
2567 | } | |
e02119d5 | 2568 | ret = 0; |
e02119d5 | 2569 | out: |
b3b4aa74 | 2570 | btrfs_release_path(path); |
e02119d5 CM |
2571 | btrfs_free_path(log_path); |
2572 | iput(dir); | |
2573 | return ret; | |
2574 | } | |
2575 | ||
2576 | /* | |
2577 | * the process_func used to replay items from the log tree. This | |
2578 | * gets called in two different stages. The first stage just looks | |
2579 | * for inodes and makes sure they are all copied into the subvolume. | |
2580 | * | |
2581 | * The second stage copies all the other item types from the log into | |
2582 | * the subvolume. The two stage approach is slower, but gets rid of | |
2583 | * lots of complexity around inodes referencing other inodes that exist | |
2584 | * only in the log (references come from either directory items or inode | |
2585 | * back refs). | |
2586 | */ | |
2587 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 2588 | struct walk_control *wc, u64 gen, int level) |
e02119d5 CM |
2589 | { |
2590 | int nritems; | |
2591 | struct btrfs_path *path; | |
2592 | struct btrfs_root *root = wc->replay_dest; | |
2593 | struct btrfs_key key; | |
e02119d5 CM |
2594 | int i; |
2595 | int ret; | |
2596 | ||
581c1760 | 2597 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
018642a1 TI |
2598 | if (ret) |
2599 | return ret; | |
e02119d5 CM |
2600 | |
2601 | level = btrfs_header_level(eb); | |
2602 | ||
2603 | if (level != 0) | |
2604 | return 0; | |
2605 | ||
2606 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2607 | if (!path) |
2608 | return -ENOMEM; | |
e02119d5 CM |
2609 | |
2610 | nritems = btrfs_header_nritems(eb); | |
2611 | for (i = 0; i < nritems; i++) { | |
2612 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2613 | |
2614 | /* inode keys are done during the first stage */ | |
2615 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2616 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2617 | struct btrfs_inode_item *inode_item; |
2618 | u32 mode; | |
2619 | ||
2620 | inode_item = btrfs_item_ptr(eb, i, | |
2621 | struct btrfs_inode_item); | |
f2d72f42 FM |
2622 | /* |
2623 | * If we have a tmpfile (O_TMPFILE) that got fsync'ed | |
2624 | * and never got linked before the fsync, skip it, as | |
2625 | * replaying it is pointless since it would be deleted | |
2626 | * later. We skip logging tmpfiles, but it's always | |
2627 | * possible we are replaying a log created with a kernel | |
2628 | * that used to log tmpfiles. | |
2629 | */ | |
2630 | if (btrfs_inode_nlink(eb, inode_item) == 0) { | |
2631 | wc->ignore_cur_inode = true; | |
2632 | continue; | |
2633 | } else { | |
2634 | wc->ignore_cur_inode = false; | |
2635 | } | |
4f764e51 FM |
2636 | ret = replay_xattr_deletes(wc->trans, root, log, |
2637 | path, key.objectid); | |
2638 | if (ret) | |
2639 | break; | |
e02119d5 CM |
2640 | mode = btrfs_inode_mode(eb, inode_item); |
2641 | if (S_ISDIR(mode)) { | |
2642 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2643 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2644 | if (ret) |
2645 | break; | |
e02119d5 CM |
2646 | } |
2647 | ret = overwrite_item(wc->trans, root, path, | |
2648 | eb, i, &key); | |
b50c6e25 JB |
2649 | if (ret) |
2650 | break; | |
e02119d5 | 2651 | |
471d557a FM |
2652 | /* |
2653 | * Before replaying extents, truncate the inode to its | |
2654 | * size. We need to do it now and not after log replay | |
2655 | * because before an fsync we can have prealloc extents | |
2656 | * added beyond the inode's i_size. If we did it after, | |
2657 | * through orphan cleanup for example, we would drop | |
2658 | * those prealloc extents just after replaying them. | |
e02119d5 CM |
2659 | */ |
2660 | if (S_ISREG(mode)) { | |
5893dfb9 | 2661 | struct btrfs_drop_extents_args drop_args = { 0 }; |
471d557a FM |
2662 | struct inode *inode; |
2663 | u64 from; | |
2664 | ||
2665 | inode = read_one_inode(root, key.objectid); | |
2666 | if (!inode) { | |
2667 | ret = -EIO; | |
2668 | break; | |
2669 | } | |
2670 | from = ALIGN(i_size_read(inode), | |
2671 | root->fs_info->sectorsize); | |
5893dfb9 FM |
2672 | drop_args.start = from; |
2673 | drop_args.end = (u64)-1; | |
2674 | drop_args.drop_cache = true; | |
2675 | ret = btrfs_drop_extents(wc->trans, root, | |
2676 | BTRFS_I(inode), | |
2677 | &drop_args); | |
471d557a | 2678 | if (!ret) { |
2766ff61 FM |
2679 | inode_sub_bytes(inode, |
2680 | drop_args.bytes_found); | |
f2d72f42 | 2681 | /* Update the inode's nbytes. */ |
471d557a | 2682 | ret = btrfs_update_inode(wc->trans, |
9a56fcd1 | 2683 | root, BTRFS_I(inode)); |
471d557a FM |
2684 | } |
2685 | iput(inode); | |
b50c6e25 JB |
2686 | if (ret) |
2687 | break; | |
e02119d5 | 2688 | } |
c71bf099 | 2689 | |
e02119d5 CM |
2690 | ret = link_to_fixup_dir(wc->trans, root, |
2691 | path, key.objectid); | |
b50c6e25 JB |
2692 | if (ret) |
2693 | break; | |
e02119d5 | 2694 | } |
dd8e7217 | 2695 | |
f2d72f42 FM |
2696 | if (wc->ignore_cur_inode) |
2697 | continue; | |
2698 | ||
dd8e7217 JB |
2699 | if (key.type == BTRFS_DIR_INDEX_KEY && |
2700 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2701 | ret = replay_one_dir_item(wc->trans, root, path, | |
2702 | eb, i, &key); | |
2703 | if (ret) | |
2704 | break; | |
2705 | } | |
2706 | ||
e02119d5 CM |
2707 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2708 | continue; | |
2709 | ||
2710 | /* these keys are simply copied */ | |
2711 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2712 | ret = overwrite_item(wc->trans, root, path, | |
2713 | eb, i, &key); | |
b50c6e25 JB |
2714 | if (ret) |
2715 | break; | |
2da1c669 LB |
2716 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2717 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2718 | ret = add_inode_ref(wc->trans, root, log, path, |
2719 | eb, i, &key); | |
b50c6e25 JB |
2720 | if (ret && ret != -ENOENT) |
2721 | break; | |
2722 | ret = 0; | |
e02119d5 CM |
2723 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2724 | ret = replay_one_extent(wc->trans, root, path, | |
2725 | eb, i, &key); | |
b50c6e25 JB |
2726 | if (ret) |
2727 | break; | |
e02119d5 | 2728 | } |
339d0354 FM |
2729 | /* |
2730 | * We don't log BTRFS_DIR_ITEM_KEY keys anymore, only the | |
2731 | * BTRFS_DIR_INDEX_KEY items which we use to derive the | |
2732 | * BTRFS_DIR_ITEM_KEY items. If we are replaying a log from an | |
2733 | * older kernel with such keys, ignore them. | |
2734 | */ | |
e02119d5 CM |
2735 | } |
2736 | btrfs_free_path(path); | |
b50c6e25 | 2737 | return ret; |
e02119d5 CM |
2738 | } |
2739 | ||
6787bb9f NB |
2740 | /* |
2741 | * Correctly adjust the reserved bytes occupied by a log tree extent buffer | |
2742 | */ | |
2743 | static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start) | |
2744 | { | |
2745 | struct btrfs_block_group *cache; | |
2746 | ||
2747 | cache = btrfs_lookup_block_group(fs_info, start); | |
2748 | if (!cache) { | |
2749 | btrfs_err(fs_info, "unable to find block group for %llu", start); | |
2750 | return; | |
2751 | } | |
2752 | ||
2753 | spin_lock(&cache->space_info->lock); | |
2754 | spin_lock(&cache->lock); | |
2755 | cache->reserved -= fs_info->nodesize; | |
2756 | cache->space_info->bytes_reserved -= fs_info->nodesize; | |
2757 | spin_unlock(&cache->lock); | |
2758 | spin_unlock(&cache->space_info->lock); | |
2759 | ||
2760 | btrfs_put_block_group(cache); | |
2761 | } | |
2762 | ||
d397712b | 2763 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2764 | struct btrfs_root *root, |
2765 | struct btrfs_path *path, int *level, | |
2766 | struct walk_control *wc) | |
2767 | { | |
0b246afa | 2768 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2769 | u64 bytenr; |
2770 | u64 ptr_gen; | |
2771 | struct extent_buffer *next; | |
2772 | struct extent_buffer *cur; | |
e02119d5 CM |
2773 | u32 blocksize; |
2774 | int ret = 0; | |
2775 | ||
d397712b | 2776 | while (*level > 0) { |
581c1760 QW |
2777 | struct btrfs_key first_key; |
2778 | ||
e02119d5 CM |
2779 | cur = path->nodes[*level]; |
2780 | ||
fae7f21c | 2781 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2782 | |
2783 | if (path->slots[*level] >= | |
2784 | btrfs_header_nritems(cur)) | |
2785 | break; | |
2786 | ||
2787 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2788 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
581c1760 | 2789 | btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]); |
0b246afa | 2790 | blocksize = fs_info->nodesize; |
e02119d5 | 2791 | |
3fbaf258 JB |
2792 | next = btrfs_find_create_tree_block(fs_info, bytenr, |
2793 | btrfs_header_owner(cur), | |
2794 | *level - 1); | |
c871b0f2 LB |
2795 | if (IS_ERR(next)) |
2796 | return PTR_ERR(next); | |
e02119d5 | 2797 | |
e02119d5 | 2798 | if (*level == 1) { |
581c1760 QW |
2799 | ret = wc->process_func(root, next, wc, ptr_gen, |
2800 | *level - 1); | |
b50c6e25 JB |
2801 | if (ret) { |
2802 | free_extent_buffer(next); | |
1e5063d0 | 2803 | return ret; |
b50c6e25 | 2804 | } |
4a500fd1 | 2805 | |
e02119d5 CM |
2806 | path->slots[*level]++; |
2807 | if (wc->free) { | |
581c1760 QW |
2808 | ret = btrfs_read_buffer(next, ptr_gen, |
2809 | *level - 1, &first_key); | |
018642a1 TI |
2810 | if (ret) { |
2811 | free_extent_buffer(next); | |
2812 | return ret; | |
2813 | } | |
e02119d5 | 2814 | |
681ae509 JB |
2815 | if (trans) { |
2816 | btrfs_tree_lock(next); | |
6a884d7d | 2817 | btrfs_clean_tree_block(next); |
681ae509 JB |
2818 | btrfs_wait_tree_block_writeback(next); |
2819 | btrfs_tree_unlock(next); | |
7bfc1007 | 2820 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2821 | bytenr, blocksize); |
2822 | if (ret) { | |
2823 | free_extent_buffer(next); | |
2824 | return ret; | |
2825 | } | |
d3575156 NA |
2826 | btrfs_redirty_list_add( |
2827 | trans->transaction, next); | |
1846430c LB |
2828 | } else { |
2829 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2830 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2831 | unaccount_log_buffer(fs_info, bytenr); |
3650860b | 2832 | } |
e02119d5 CM |
2833 | } |
2834 | free_extent_buffer(next); | |
2835 | continue; | |
2836 | } | |
581c1760 | 2837 | ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key); |
018642a1 TI |
2838 | if (ret) { |
2839 | free_extent_buffer(next); | |
2840 | return ret; | |
2841 | } | |
e02119d5 | 2842 | |
e02119d5 CM |
2843 | if (path->nodes[*level-1]) |
2844 | free_extent_buffer(path->nodes[*level-1]); | |
2845 | path->nodes[*level-1] = next; | |
2846 | *level = btrfs_header_level(next); | |
2847 | path->slots[*level] = 0; | |
2848 | cond_resched(); | |
2849 | } | |
4a500fd1 | 2850 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2851 | |
2852 | cond_resched(); | |
2853 | return 0; | |
2854 | } | |
2855 | ||
d397712b | 2856 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2857 | struct btrfs_root *root, |
2858 | struct btrfs_path *path, int *level, | |
2859 | struct walk_control *wc) | |
2860 | { | |
0b246afa | 2861 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2862 | int i; |
2863 | int slot; | |
2864 | int ret; | |
2865 | ||
d397712b | 2866 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2867 | slot = path->slots[i]; |
4a500fd1 | 2868 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2869 | path->slots[i]++; |
2870 | *level = i; | |
2871 | WARN_ON(*level == 0); | |
2872 | return 0; | |
2873 | } else { | |
1e5063d0 | 2874 | ret = wc->process_func(root, path->nodes[*level], wc, |
581c1760 QW |
2875 | btrfs_header_generation(path->nodes[*level]), |
2876 | *level); | |
1e5063d0 MF |
2877 | if (ret) |
2878 | return ret; | |
2879 | ||
e02119d5 CM |
2880 | if (wc->free) { |
2881 | struct extent_buffer *next; | |
2882 | ||
2883 | next = path->nodes[*level]; | |
2884 | ||
681ae509 JB |
2885 | if (trans) { |
2886 | btrfs_tree_lock(next); | |
6a884d7d | 2887 | btrfs_clean_tree_block(next); |
681ae509 JB |
2888 | btrfs_wait_tree_block_writeback(next); |
2889 | btrfs_tree_unlock(next); | |
7bfc1007 | 2890 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2891 | path->nodes[*level]->start, |
2892 | path->nodes[*level]->len); | |
2893 | if (ret) | |
2894 | return ret; | |
84c25448 NA |
2895 | btrfs_redirty_list_add(trans->transaction, |
2896 | next); | |
1846430c LB |
2897 | } else { |
2898 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2899 | clear_extent_buffer_dirty(next); | |
e02119d5 | 2900 | |
10e958d5 NB |
2901 | unaccount_log_buffer(fs_info, |
2902 | path->nodes[*level]->start); | |
2903 | } | |
e02119d5 CM |
2904 | } |
2905 | free_extent_buffer(path->nodes[*level]); | |
2906 | path->nodes[*level] = NULL; | |
2907 | *level = i + 1; | |
2908 | } | |
2909 | } | |
2910 | return 1; | |
2911 | } | |
2912 | ||
2913 | /* | |
2914 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2915 | * the tree freeing any blocks that have a ref count of zero after being | |
2916 | * decremented. | |
2917 | */ | |
2918 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2919 | struct btrfs_root *log, struct walk_control *wc) | |
2920 | { | |
2ff7e61e | 2921 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2922 | int ret = 0; |
2923 | int wret; | |
2924 | int level; | |
2925 | struct btrfs_path *path; | |
e02119d5 CM |
2926 | int orig_level; |
2927 | ||
2928 | path = btrfs_alloc_path(); | |
db5b493a TI |
2929 | if (!path) |
2930 | return -ENOMEM; | |
e02119d5 CM |
2931 | |
2932 | level = btrfs_header_level(log->node); | |
2933 | orig_level = level; | |
2934 | path->nodes[level] = log->node; | |
67439dad | 2935 | atomic_inc(&log->node->refs); |
e02119d5 CM |
2936 | path->slots[level] = 0; |
2937 | ||
d397712b | 2938 | while (1) { |
e02119d5 CM |
2939 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2940 | if (wret > 0) | |
2941 | break; | |
79787eaa | 2942 | if (wret < 0) { |
e02119d5 | 2943 | ret = wret; |
79787eaa JM |
2944 | goto out; |
2945 | } | |
e02119d5 CM |
2946 | |
2947 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2948 | if (wret > 0) | |
2949 | break; | |
79787eaa | 2950 | if (wret < 0) { |
e02119d5 | 2951 | ret = wret; |
79787eaa JM |
2952 | goto out; |
2953 | } | |
e02119d5 CM |
2954 | } |
2955 | ||
2956 | /* was the root node processed? if not, catch it here */ | |
2957 | if (path->nodes[orig_level]) { | |
79787eaa | 2958 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
581c1760 QW |
2959 | btrfs_header_generation(path->nodes[orig_level]), |
2960 | orig_level); | |
79787eaa JM |
2961 | if (ret) |
2962 | goto out; | |
e02119d5 CM |
2963 | if (wc->free) { |
2964 | struct extent_buffer *next; | |
2965 | ||
2966 | next = path->nodes[orig_level]; | |
2967 | ||
681ae509 JB |
2968 | if (trans) { |
2969 | btrfs_tree_lock(next); | |
6a884d7d | 2970 | btrfs_clean_tree_block(next); |
681ae509 JB |
2971 | btrfs_wait_tree_block_writeback(next); |
2972 | btrfs_tree_unlock(next); | |
7bfc1007 | 2973 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2974 | next->start, next->len); |
2975 | if (ret) | |
2976 | goto out; | |
84c25448 | 2977 | btrfs_redirty_list_add(trans->transaction, next); |
1846430c LB |
2978 | } else { |
2979 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2980 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2981 | unaccount_log_buffer(fs_info, next->start); |
681ae509 | 2982 | } |
e02119d5 CM |
2983 | } |
2984 | } | |
2985 | ||
79787eaa | 2986 | out: |
e02119d5 | 2987 | btrfs_free_path(path); |
e02119d5 CM |
2988 | return ret; |
2989 | } | |
2990 | ||
7237f183 YZ |
2991 | /* |
2992 | * helper function to update the item for a given subvolumes log root | |
2993 | * in the tree of log roots | |
2994 | */ | |
2995 | static int update_log_root(struct btrfs_trans_handle *trans, | |
4203e968 JB |
2996 | struct btrfs_root *log, |
2997 | struct btrfs_root_item *root_item) | |
7237f183 | 2998 | { |
0b246afa | 2999 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
3000 | int ret; |
3001 | ||
3002 | if (log->log_transid == 1) { | |
3003 | /* insert root item on the first sync */ | |
0b246afa | 3004 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
4203e968 | 3005 | &log->root_key, root_item); |
7237f183 | 3006 | } else { |
0b246afa | 3007 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
4203e968 | 3008 | &log->root_key, root_item); |
7237f183 YZ |
3009 | } |
3010 | return ret; | |
3011 | } | |
3012 | ||
60d53eb3 | 3013 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
3014 | { |
3015 | DEFINE_WAIT(wait); | |
7237f183 | 3016 | int index = transid % 2; |
e02119d5 | 3017 | |
7237f183 YZ |
3018 | /* |
3019 | * we only allow two pending log transactions at a time, | |
3020 | * so we know that if ours is more than 2 older than the | |
3021 | * current transaction, we're done | |
3022 | */ | |
49e83f57 | 3023 | for (;;) { |
7237f183 YZ |
3024 | prepare_to_wait(&root->log_commit_wait[index], |
3025 | &wait, TASK_UNINTERRUPTIBLE); | |
12fcfd22 | 3026 | |
49e83f57 LB |
3027 | if (!(root->log_transid_committed < transid && |
3028 | atomic_read(&root->log_commit[index]))) | |
3029 | break; | |
12fcfd22 | 3030 | |
49e83f57 LB |
3031 | mutex_unlock(&root->log_mutex); |
3032 | schedule(); | |
7237f183 | 3033 | mutex_lock(&root->log_mutex); |
49e83f57 LB |
3034 | } |
3035 | finish_wait(&root->log_commit_wait[index], &wait); | |
7237f183 YZ |
3036 | } |
3037 | ||
60d53eb3 | 3038 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
3039 | { |
3040 | DEFINE_WAIT(wait); | |
8b050d35 | 3041 | |
49e83f57 LB |
3042 | for (;;) { |
3043 | prepare_to_wait(&root->log_writer_wait, &wait, | |
3044 | TASK_UNINTERRUPTIBLE); | |
3045 | if (!atomic_read(&root->log_writers)) | |
3046 | break; | |
3047 | ||
7237f183 | 3048 | mutex_unlock(&root->log_mutex); |
49e83f57 | 3049 | schedule(); |
575849ec | 3050 | mutex_lock(&root->log_mutex); |
7237f183 | 3051 | } |
49e83f57 | 3052 | finish_wait(&root->log_writer_wait, &wait); |
e02119d5 CM |
3053 | } |
3054 | ||
8b050d35 MX |
3055 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
3056 | struct btrfs_log_ctx *ctx) | |
3057 | { | |
8b050d35 MX |
3058 | mutex_lock(&root->log_mutex); |
3059 | list_del_init(&ctx->list); | |
3060 | mutex_unlock(&root->log_mutex); | |
3061 | } | |
3062 | ||
3063 | /* | |
3064 | * Invoked in log mutex context, or be sure there is no other task which | |
3065 | * can access the list. | |
3066 | */ | |
3067 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
3068 | int index, int error) | |
3069 | { | |
3070 | struct btrfs_log_ctx *ctx; | |
570dd450 | 3071 | struct btrfs_log_ctx *safe; |
8b050d35 | 3072 | |
570dd450 CM |
3073 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
3074 | list_del_init(&ctx->list); | |
8b050d35 | 3075 | ctx->log_ret = error; |
570dd450 | 3076 | } |
8b050d35 MX |
3077 | } |
3078 | ||
e02119d5 CM |
3079 | /* |
3080 | * btrfs_sync_log does sends a given tree log down to the disk and | |
3081 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
3082 | * you know that any inodes previously logged are safely on disk only |
3083 | * if it returns 0. | |
3084 | * | |
3085 | * Any other return value means you need to call btrfs_commit_transaction. | |
3086 | * Some of the edge cases for fsyncing directories that have had unlinks | |
3087 | * or renames done in the past mean that sometimes the only safe | |
3088 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
3089 | * that has happened. | |
e02119d5 CM |
3090 | */ |
3091 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 3092 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 3093 | { |
7237f183 YZ |
3094 | int index1; |
3095 | int index2; | |
8cef4e16 | 3096 | int mark; |
e02119d5 | 3097 | int ret; |
0b246afa | 3098 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 3099 | struct btrfs_root *log = root->log_root; |
0b246afa | 3100 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
4203e968 | 3101 | struct btrfs_root_item new_root_item; |
bb14a59b | 3102 | int log_transid = 0; |
8b050d35 | 3103 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 3104 | struct blk_plug plug; |
47876f7c FM |
3105 | u64 log_root_start; |
3106 | u64 log_root_level; | |
e02119d5 | 3107 | |
7237f183 | 3108 | mutex_lock(&root->log_mutex); |
d1433deb MX |
3109 | log_transid = ctx->log_transid; |
3110 | if (root->log_transid_committed >= log_transid) { | |
3111 | mutex_unlock(&root->log_mutex); | |
3112 | return ctx->log_ret; | |
3113 | } | |
3114 | ||
3115 | index1 = log_transid % 2; | |
7237f183 | 3116 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 3117 | wait_log_commit(root, log_transid); |
7237f183 | 3118 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3119 | return ctx->log_ret; |
e02119d5 | 3120 | } |
d1433deb | 3121 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
3122 | atomic_set(&root->log_commit[index1], 1); |
3123 | ||
3124 | /* wait for previous tree log sync to complete */ | |
3125 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 3126 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 3127 | |
86df7eb9 | 3128 | while (1) { |
2ecb7923 | 3129 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 3130 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 3131 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 3132 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
3133 | mutex_unlock(&root->log_mutex); |
3134 | schedule_timeout_uninterruptible(1); | |
3135 | mutex_lock(&root->log_mutex); | |
3136 | } | |
60d53eb3 | 3137 | wait_for_writer(root); |
2ecb7923 | 3138 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
3139 | break; |
3140 | } | |
e02119d5 | 3141 | |
12fcfd22 | 3142 | /* bail out if we need to do a full commit */ |
4884b8e8 | 3143 | if (btrfs_need_log_full_commit(trans)) { |
12fcfd22 CM |
3144 | ret = -EAGAIN; |
3145 | mutex_unlock(&root->log_mutex); | |
3146 | goto out; | |
3147 | } | |
3148 | ||
8cef4e16 YZ |
3149 | if (log_transid % 2 == 0) |
3150 | mark = EXTENT_DIRTY; | |
3151 | else | |
3152 | mark = EXTENT_NEW; | |
3153 | ||
690587d1 CM |
3154 | /* we start IO on all the marked extents here, but we don't actually |
3155 | * wait for them until later. | |
3156 | */ | |
c6adc9cc | 3157 | blk_start_plug(&plug); |
2ff7e61e | 3158 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
b528f467 NA |
3159 | /* |
3160 | * -EAGAIN happens when someone, e.g., a concurrent transaction | |
3161 | * commit, writes a dirty extent in this tree-log commit. This | |
3162 | * concurrent write will create a hole writing out the extents, | |
3163 | * and we cannot proceed on a zoned filesystem, requiring | |
3164 | * sequential writing. While we can bail out to a full commit | |
3165 | * here, but we can continue hoping the concurrent writing fills | |
3166 | * the hole. | |
3167 | */ | |
3168 | if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) | |
3169 | ret = 0; | |
79787eaa | 3170 | if (ret) { |
c6adc9cc | 3171 | blk_finish_plug(&plug); |
66642832 | 3172 | btrfs_abort_transaction(trans, ret); |
90787766 | 3173 | btrfs_set_log_full_commit(trans); |
79787eaa JM |
3174 | mutex_unlock(&root->log_mutex); |
3175 | goto out; | |
3176 | } | |
7237f183 | 3177 | |
4203e968 JB |
3178 | /* |
3179 | * We _must_ update under the root->log_mutex in order to make sure we | |
3180 | * have a consistent view of the log root we are trying to commit at | |
3181 | * this moment. | |
3182 | * | |
3183 | * We _must_ copy this into a local copy, because we are not holding the | |
3184 | * log_root_tree->log_mutex yet. This is important because when we | |
3185 | * commit the log_root_tree we must have a consistent view of the | |
3186 | * log_root_tree when we update the super block to point at the | |
3187 | * log_root_tree bytenr. If we update the log_root_tree here we'll race | |
3188 | * with the commit and possibly point at the new block which we may not | |
3189 | * have written out. | |
3190 | */ | |
5d4f98a2 | 3191 | btrfs_set_root_node(&log->root_item, log->node); |
4203e968 | 3192 | memcpy(&new_root_item, &log->root_item, sizeof(new_root_item)); |
7237f183 | 3193 | |
7237f183 YZ |
3194 | root->log_transid++; |
3195 | log->log_transid = root->log_transid; | |
ff782e0a | 3196 | root->log_start_pid = 0; |
7237f183 | 3197 | /* |
8cef4e16 YZ |
3198 | * IO has been started, blocks of the log tree have WRITTEN flag set |
3199 | * in their headers. new modifications of the log will be written to | |
3200 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
3201 | */ |
3202 | mutex_unlock(&root->log_mutex); | |
3203 | ||
3ddebf27 | 3204 | if (btrfs_is_zoned(fs_info)) { |
e75f9fd1 | 3205 | mutex_lock(&fs_info->tree_root->log_mutex); |
3ddebf27 NA |
3206 | if (!log_root_tree->node) { |
3207 | ret = btrfs_alloc_log_tree_node(trans, log_root_tree); | |
3208 | if (ret) { | |
ea32af47 | 3209 | mutex_unlock(&fs_info->tree_root->log_mutex); |
3ddebf27 NA |
3210 | goto out; |
3211 | } | |
3212 | } | |
e75f9fd1 | 3213 | mutex_unlock(&fs_info->tree_root->log_mutex); |
3ddebf27 NA |
3214 | } |
3215 | ||
e75f9fd1 NA |
3216 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
3217 | ||
3218 | mutex_lock(&log_root_tree->log_mutex); | |
3219 | ||
e3d3b415 FM |
3220 | index2 = log_root_tree->log_transid % 2; |
3221 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
3222 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
3223 | ||
4203e968 JB |
3224 | /* |
3225 | * Now we are safe to update the log_root_tree because we're under the | |
3226 | * log_mutex, and we're a current writer so we're holding the commit | |
3227 | * open until we drop the log_mutex. | |
3228 | */ | |
3229 | ret = update_log_root(trans, log, &new_root_item); | |
4a500fd1 | 3230 | if (ret) { |
d1433deb MX |
3231 | if (!list_empty(&root_log_ctx.list)) |
3232 | list_del_init(&root_log_ctx.list); | |
3233 | ||
c6adc9cc | 3234 | blk_finish_plug(&plug); |
90787766 | 3235 | btrfs_set_log_full_commit(trans); |
995946dd | 3236 | |
79787eaa | 3237 | if (ret != -ENOSPC) { |
66642832 | 3238 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3239 | mutex_unlock(&log_root_tree->log_mutex); |
3240 | goto out; | |
3241 | } | |
bf89d38f | 3242 | btrfs_wait_tree_log_extents(log, mark); |
4a500fd1 YZ |
3243 | mutex_unlock(&log_root_tree->log_mutex); |
3244 | ret = -EAGAIN; | |
3245 | goto out; | |
3246 | } | |
3247 | ||
d1433deb | 3248 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 3249 | blk_finish_plug(&plug); |
cbd60aa7 | 3250 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
3251 | mutex_unlock(&log_root_tree->log_mutex); |
3252 | ret = root_log_ctx.log_ret; | |
3253 | goto out; | |
3254 | } | |
8b050d35 | 3255 | |
d1433deb | 3256 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 3257 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 3258 | blk_finish_plug(&plug); |
bf89d38f | 3259 | ret = btrfs_wait_tree_log_extents(log, mark); |
60d53eb3 | 3260 | wait_log_commit(log_root_tree, |
d1433deb | 3261 | root_log_ctx.log_transid); |
7237f183 | 3262 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
3263 | if (!ret) |
3264 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
3265 | goto out; |
3266 | } | |
d1433deb | 3267 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
3268 | atomic_set(&log_root_tree->log_commit[index2], 1); |
3269 | ||
12fcfd22 | 3270 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 3271 | wait_log_commit(log_root_tree, |
d1433deb | 3272 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
3273 | } |
3274 | ||
12fcfd22 CM |
3275 | /* |
3276 | * now that we've moved on to the tree of log tree roots, | |
3277 | * check the full commit flag again | |
3278 | */ | |
4884b8e8 | 3279 | if (btrfs_need_log_full_commit(trans)) { |
c6adc9cc | 3280 | blk_finish_plug(&plug); |
bf89d38f | 3281 | btrfs_wait_tree_log_extents(log, mark); |
12fcfd22 CM |
3282 | mutex_unlock(&log_root_tree->log_mutex); |
3283 | ret = -EAGAIN; | |
3284 | goto out_wake_log_root; | |
3285 | } | |
7237f183 | 3286 | |
2ff7e61e | 3287 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
3288 | &log_root_tree->dirty_log_pages, |
3289 | EXTENT_DIRTY | EXTENT_NEW); | |
3290 | blk_finish_plug(&plug); | |
b528f467 NA |
3291 | /* |
3292 | * As described above, -EAGAIN indicates a hole in the extents. We | |
3293 | * cannot wait for these write outs since the waiting cause a | |
3294 | * deadlock. Bail out to the full commit instead. | |
3295 | */ | |
3296 | if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) { | |
3297 | btrfs_set_log_full_commit(trans); | |
3298 | btrfs_wait_tree_log_extents(log, mark); | |
3299 | mutex_unlock(&log_root_tree->log_mutex); | |
3300 | goto out_wake_log_root; | |
3301 | } else if (ret) { | |
90787766 | 3302 | btrfs_set_log_full_commit(trans); |
66642832 | 3303 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3304 | mutex_unlock(&log_root_tree->log_mutex); |
3305 | goto out_wake_log_root; | |
3306 | } | |
bf89d38f | 3307 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 3308 | if (!ret) |
bf89d38f JM |
3309 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
3310 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 3311 | if (ret) { |
90787766 | 3312 | btrfs_set_log_full_commit(trans); |
5ab5e44a FM |
3313 | mutex_unlock(&log_root_tree->log_mutex); |
3314 | goto out_wake_log_root; | |
3315 | } | |
e02119d5 | 3316 | |
47876f7c FM |
3317 | log_root_start = log_root_tree->node->start; |
3318 | log_root_level = btrfs_header_level(log_root_tree->node); | |
7237f183 | 3319 | log_root_tree->log_transid++; |
7237f183 YZ |
3320 | mutex_unlock(&log_root_tree->log_mutex); |
3321 | ||
3322 | /* | |
47876f7c FM |
3323 | * Here we are guaranteed that nobody is going to write the superblock |
3324 | * for the current transaction before us and that neither we do write | |
3325 | * our superblock before the previous transaction finishes its commit | |
3326 | * and writes its superblock, because: | |
3327 | * | |
3328 | * 1) We are holding a handle on the current transaction, so no body | |
3329 | * can commit it until we release the handle; | |
3330 | * | |
3331 | * 2) Before writing our superblock we acquire the tree_log_mutex, so | |
3332 | * if the previous transaction is still committing, and hasn't yet | |
3333 | * written its superblock, we wait for it to do it, because a | |
3334 | * transaction commit acquires the tree_log_mutex when the commit | |
3335 | * begins and releases it only after writing its superblock. | |
7237f183 | 3336 | */ |
47876f7c | 3337 | mutex_lock(&fs_info->tree_log_mutex); |
165ea85f JB |
3338 | |
3339 | /* | |
3340 | * The previous transaction writeout phase could have failed, and thus | |
3341 | * marked the fs in an error state. We must not commit here, as we | |
3342 | * could have updated our generation in the super_for_commit and | |
3343 | * writing the super here would result in transid mismatches. If there | |
3344 | * is an error here just bail. | |
3345 | */ | |
84961539 | 3346 | if (BTRFS_FS_ERROR(fs_info)) { |
165ea85f JB |
3347 | ret = -EIO; |
3348 | btrfs_set_log_full_commit(trans); | |
3349 | btrfs_abort_transaction(trans, ret); | |
3350 | mutex_unlock(&fs_info->tree_log_mutex); | |
3351 | goto out_wake_log_root; | |
3352 | } | |
3353 | ||
47876f7c FM |
3354 | btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start); |
3355 | btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level); | |
eece6a9c | 3356 | ret = write_all_supers(fs_info, 1); |
47876f7c | 3357 | mutex_unlock(&fs_info->tree_log_mutex); |
5af3e8cc | 3358 | if (ret) { |
90787766 | 3359 | btrfs_set_log_full_commit(trans); |
66642832 | 3360 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
3361 | goto out_wake_log_root; |
3362 | } | |
7237f183 | 3363 | |
e1a6d264 FM |
3364 | /* |
3365 | * We know there can only be one task here, since we have not yet set | |
3366 | * root->log_commit[index1] to 0 and any task attempting to sync the | |
3367 | * log must wait for the previous log transaction to commit if it's | |
3368 | * still in progress or wait for the current log transaction commit if | |
3369 | * someone else already started it. We use <= and not < because the | |
3370 | * first log transaction has an ID of 0. | |
3371 | */ | |
3372 | ASSERT(root->last_log_commit <= log_transid); | |
3373 | root->last_log_commit = log_transid; | |
257c62e1 | 3374 | |
12fcfd22 | 3375 | out_wake_log_root: |
570dd450 | 3376 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
3377 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
3378 | ||
d1433deb | 3379 | log_root_tree->log_transid_committed++; |
7237f183 | 3380 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
3381 | mutex_unlock(&log_root_tree->log_mutex); |
3382 | ||
33a9eca7 | 3383 | /* |
093258e6 DS |
3384 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3385 | * all the updates above are seen by the woken threads. It might not be | |
3386 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3387 | */ |
093258e6 | 3388 | cond_wake_up(&log_root_tree->log_commit_wait[index2]); |
e02119d5 | 3389 | out: |
d1433deb | 3390 | mutex_lock(&root->log_mutex); |
570dd450 | 3391 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 3392 | root->log_transid_committed++; |
7237f183 | 3393 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 3394 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3395 | |
33a9eca7 | 3396 | /* |
093258e6 DS |
3397 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3398 | * all the updates above are seen by the woken threads. It might not be | |
3399 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3400 | */ |
093258e6 | 3401 | cond_wake_up(&root->log_commit_wait[index1]); |
b31eabd8 | 3402 | return ret; |
e02119d5 CM |
3403 | } |
3404 | ||
4a500fd1 YZ |
3405 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3406 | struct btrfs_root *log) | |
e02119d5 CM |
3407 | { |
3408 | int ret; | |
e02119d5 CM |
3409 | struct walk_control wc = { |
3410 | .free = 1, | |
3411 | .process_func = process_one_buffer | |
3412 | }; | |
3413 | ||
3ddebf27 NA |
3414 | if (log->node) { |
3415 | ret = walk_log_tree(trans, log, &wc); | |
3416 | if (ret) { | |
40cdc509 FM |
3417 | /* |
3418 | * We weren't able to traverse the entire log tree, the | |
3419 | * typical scenario is getting an -EIO when reading an | |
3420 | * extent buffer of the tree, due to a previous writeback | |
3421 | * failure of it. | |
3422 | */ | |
3423 | set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR, | |
3424 | &log->fs_info->fs_state); | |
3425 | ||
3426 | /* | |
3427 | * Some extent buffers of the log tree may still be dirty | |
3428 | * and not yet written back to storage, because we may | |
3429 | * have updates to a log tree without syncing a log tree, | |
3430 | * such as during rename and link operations. So flush | |
3431 | * them out and wait for their writeback to complete, so | |
3432 | * that we properly cleanup their state and pages. | |
3433 | */ | |
3434 | btrfs_write_marked_extents(log->fs_info, | |
3435 | &log->dirty_log_pages, | |
3436 | EXTENT_DIRTY | EXTENT_NEW); | |
3437 | btrfs_wait_tree_log_extents(log, | |
3438 | EXTENT_DIRTY | EXTENT_NEW); | |
3439 | ||
3ddebf27 NA |
3440 | if (trans) |
3441 | btrfs_abort_transaction(trans, ret); | |
3442 | else | |
3443 | btrfs_handle_fs_error(log->fs_info, ret, NULL); | |
3444 | } | |
374b0e2d | 3445 | } |
e02119d5 | 3446 | |
59b0713a FM |
3447 | clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1, |
3448 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT); | |
e289f03e | 3449 | extent_io_tree_release(&log->log_csum_range); |
d3575156 | 3450 | |
00246528 | 3451 | btrfs_put_root(log); |
4a500fd1 YZ |
3452 | } |
3453 | ||
3454 | /* | |
3455 | * free all the extents used by the tree log. This should be called | |
3456 | * at commit time of the full transaction | |
3457 | */ | |
3458 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3459 | { | |
3460 | if (root->log_root) { | |
3461 | free_log_tree(trans, root->log_root); | |
3462 | root->log_root = NULL; | |
e7a79811 | 3463 | clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state); |
4a500fd1 YZ |
3464 | } |
3465 | return 0; | |
3466 | } | |
3467 | ||
3468 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3469 | struct btrfs_fs_info *fs_info) | |
3470 | { | |
3471 | if (fs_info->log_root_tree) { | |
3472 | free_log_tree(trans, fs_info->log_root_tree); | |
3473 | fs_info->log_root_tree = NULL; | |
47876f7c | 3474 | clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state); |
4a500fd1 | 3475 | } |
e02119d5 CM |
3476 | return 0; |
3477 | } | |
3478 | ||
803f0f64 | 3479 | /* |
0f8ce498 FM |
3480 | * Check if an inode was logged in the current transaction. This correctly deals |
3481 | * with the case where the inode was logged but has a logged_trans of 0, which | |
3482 | * happens if the inode is evicted and loaded again, as logged_trans is an in | |
3483 | * memory only field (not persisted). | |
3484 | * | |
3485 | * Returns 1 if the inode was logged before in the transaction, 0 if it was not, | |
3486 | * and < 0 on error. | |
803f0f64 | 3487 | */ |
0f8ce498 FM |
3488 | static int inode_logged(struct btrfs_trans_handle *trans, |
3489 | struct btrfs_inode *inode, | |
3490 | struct btrfs_path *path_in) | |
803f0f64 | 3491 | { |
0f8ce498 FM |
3492 | struct btrfs_path *path = path_in; |
3493 | struct btrfs_key key; | |
3494 | int ret; | |
3495 | ||
803f0f64 | 3496 | if (inode->logged_trans == trans->transid) |
0f8ce498 | 3497 | return 1; |
803f0f64 | 3498 | |
0f8ce498 FM |
3499 | /* |
3500 | * If logged_trans is not 0, then we know the inode logged was not logged | |
3501 | * in this transaction, so we can return false right away. | |
3502 | */ | |
3503 | if (inode->logged_trans > 0) | |
3504 | return 0; | |
3505 | ||
3506 | /* | |
3507 | * If no log tree was created for this root in this transaction, then | |
3508 | * the inode can not have been logged in this transaction. In that case | |
3509 | * set logged_trans to anything greater than 0 and less than the current | |
3510 | * transaction's ID, to avoid the search below in a future call in case | |
3511 | * a log tree gets created after this. | |
3512 | */ | |
3513 | if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &inode->root->state)) { | |
3514 | inode->logged_trans = trans->transid - 1; | |
3515 | return 0; | |
3516 | } | |
3517 | ||
3518 | /* | |
3519 | * We have a log tree and the inode's logged_trans is 0. We can't tell | |
3520 | * for sure if the inode was logged before in this transaction by looking | |
3521 | * only at logged_trans. We could be pessimistic and assume it was, but | |
3522 | * that can lead to unnecessarily logging an inode during rename and link | |
3523 | * operations, and then further updating the log in followup rename and | |
3524 | * link operations, specially if it's a directory, which adds latency | |
3525 | * visible to applications doing a series of rename or link operations. | |
3526 | * | |
3527 | * A logged_trans of 0 here can mean several things: | |
3528 | * | |
3529 | * 1) The inode was never logged since the filesystem was mounted, and may | |
3530 | * or may have not been evicted and loaded again; | |
3531 | * | |
3532 | * 2) The inode was logged in a previous transaction, then evicted and | |
3533 | * then loaded again; | |
3534 | * | |
3535 | * 3) The inode was logged in the current transaction, then evicted and | |
3536 | * then loaded again. | |
3537 | * | |
3538 | * For cases 1) and 2) we don't want to return true, but we need to detect | |
3539 | * case 3) and return true. So we do a search in the log root for the inode | |
3540 | * item. | |
3541 | */ | |
3542 | key.objectid = btrfs_ino(inode); | |
3543 | key.type = BTRFS_INODE_ITEM_KEY; | |
3544 | key.offset = 0; | |
3545 | ||
3546 | if (!path) { | |
3547 | path = btrfs_alloc_path(); | |
3548 | if (!path) | |
3549 | return -ENOMEM; | |
3550 | } | |
3551 | ||
3552 | ret = btrfs_search_slot(NULL, inode->root->log_root, &key, path, 0, 0); | |
3553 | ||
3554 | if (path_in) | |
3555 | btrfs_release_path(path); | |
3556 | else | |
3557 | btrfs_free_path(path); | |
1e0860f3 | 3558 | |
6e8e777d | 3559 | /* |
0f8ce498 FM |
3560 | * Logging an inode always results in logging its inode item. So if we |
3561 | * did not find the item we know the inode was not logged for sure. | |
6e8e777d | 3562 | */ |
0f8ce498 FM |
3563 | if (ret < 0) { |
3564 | return ret; | |
3565 | } else if (ret > 0) { | |
3566 | /* | |
3567 | * Set logged_trans to a value greater than 0 and less then the | |
3568 | * current transaction to avoid doing the search in future calls. | |
3569 | */ | |
3570 | inode->logged_trans = trans->transid - 1; | |
3571 | return 0; | |
3572 | } | |
3573 | ||
3574 | /* | |
3575 | * The inode was previously logged and then evicted, set logged_trans to | |
3576 | * the current transacion's ID, to avoid future tree searches as long as | |
3577 | * the inode is not evicted again. | |
3578 | */ | |
3579 | inode->logged_trans = trans->transid; | |
3580 | ||
3581 | /* | |
3582 | * If it's a directory, then we must set last_dir_index_offset to the | |
3583 | * maximum possible value, so that the next attempt to log the inode does | |
3584 | * not skip checking if dir index keys found in modified subvolume tree | |
3585 | * leaves have been logged before, otherwise it would result in attempts | |
3586 | * to insert duplicate dir index keys in the log tree. This must be done | |
3587 | * because last_dir_index_offset is an in-memory only field, not persisted | |
3588 | * in the inode item or any other on-disk structure, so its value is lost | |
3589 | * once the inode is evicted. | |
3590 | */ | |
3591 | if (S_ISDIR(inode->vfs_inode.i_mode)) | |
3592 | inode->last_dir_index_offset = (u64)-1; | |
803f0f64 | 3593 | |
0f8ce498 | 3594 | return 1; |
803f0f64 FM |
3595 | } |
3596 | ||
839061fe FM |
3597 | /* |
3598 | * Delete a directory entry from the log if it exists. | |
3599 | * | |
3600 | * Returns < 0 on error | |
3601 | * 1 if the entry does not exists | |
3602 | * 0 if the entry existed and was successfully deleted | |
3603 | */ | |
3604 | static int del_logged_dentry(struct btrfs_trans_handle *trans, | |
3605 | struct btrfs_root *log, | |
3606 | struct btrfs_path *path, | |
3607 | u64 dir_ino, | |
3608 | const char *name, int name_len, | |
3609 | u64 index) | |
3610 | { | |
3611 | struct btrfs_dir_item *di; | |
3612 | ||
3613 | /* | |
3614 | * We only log dir index items of a directory, so we don't need to look | |
3615 | * for dir item keys. | |
3616 | */ | |
3617 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, | |
3618 | index, name, name_len, -1); | |
3619 | if (IS_ERR(di)) | |
3620 | return PTR_ERR(di); | |
3621 | else if (!di) | |
3622 | return 1; | |
3623 | ||
3624 | /* | |
3625 | * We do not need to update the size field of the directory's | |
3626 | * inode item because on log replay we update the field to reflect | |
3627 | * all existing entries in the directory (see overwrite_item()). | |
3628 | */ | |
3629 | return btrfs_delete_one_dir_name(trans, log, path, di); | |
3630 | } | |
3631 | ||
e02119d5 CM |
3632 | /* |
3633 | * If both a file and directory are logged, and unlinks or renames are | |
3634 | * mixed in, we have a few interesting corners: | |
3635 | * | |
3636 | * create file X in dir Y | |
3637 | * link file X to X.link in dir Y | |
3638 | * fsync file X | |
3639 | * unlink file X but leave X.link | |
3640 | * fsync dir Y | |
3641 | * | |
3642 | * After a crash we would expect only X.link to exist. But file X | |
3643 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3644 | * | |
3645 | * We solve this by removing directory entries and inode backrefs from the | |
3646 | * log when a file that was logged in the current transaction is | |
3647 | * unlinked. Any later fsync will include the updated log entries, and | |
3648 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3649 | * | |
3650 | * This optimizations allows us to avoid relogging the entire inode | |
3651 | * or the entire directory. | |
3652 | */ | |
9a35fc95 JB |
3653 | void btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, |
3654 | struct btrfs_root *root, | |
3655 | const char *name, int name_len, | |
3656 | struct btrfs_inode *dir, u64 index) | |
e02119d5 | 3657 | { |
e02119d5 CM |
3658 | struct btrfs_path *path; |
3659 | int ret; | |
e02119d5 | 3660 | |
0f8ce498 FM |
3661 | ret = inode_logged(trans, dir, NULL); |
3662 | if (ret == 0) | |
3663 | return; | |
3664 | else if (ret < 0) { | |
3665 | btrfs_set_log_full_commit(trans); | |
9a35fc95 | 3666 | return; |
0f8ce498 | 3667 | } |
3a5f1d45 | 3668 | |
e02119d5 CM |
3669 | ret = join_running_log_trans(root); |
3670 | if (ret) | |
9a35fc95 | 3671 | return; |
e02119d5 | 3672 | |
49f34d1f | 3673 | mutex_lock(&dir->log_mutex); |
e02119d5 | 3674 | |
e02119d5 | 3675 | path = btrfs_alloc_path(); |
a62f44a5 | 3676 | if (!path) { |
839061fe | 3677 | ret = -ENOMEM; |
a62f44a5 TI |
3678 | goto out_unlock; |
3679 | } | |
2a29edc6 | 3680 | |
839061fe FM |
3681 | ret = del_logged_dentry(trans, root->log_root, path, btrfs_ino(dir), |
3682 | name, name_len, index); | |
e02119d5 | 3683 | btrfs_free_path(path); |
a62f44a5 | 3684 | out_unlock: |
49f34d1f | 3685 | mutex_unlock(&dir->log_mutex); |
839061fe | 3686 | if (ret < 0) |
90787766 | 3687 | btrfs_set_log_full_commit(trans); |
12fcfd22 | 3688 | btrfs_end_log_trans(root); |
e02119d5 CM |
3689 | } |
3690 | ||
3691 | /* see comments for btrfs_del_dir_entries_in_log */ | |
9a35fc95 JB |
3692 | void btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, |
3693 | struct btrfs_root *root, | |
3694 | const char *name, int name_len, | |
3695 | struct btrfs_inode *inode, u64 dirid) | |
e02119d5 CM |
3696 | { |
3697 | struct btrfs_root *log; | |
3698 | u64 index; | |
3699 | int ret; | |
3700 | ||
0f8ce498 FM |
3701 | ret = inode_logged(trans, inode, NULL); |
3702 | if (ret == 0) | |
9a35fc95 | 3703 | return; |
0f8ce498 FM |
3704 | else if (ret < 0) { |
3705 | btrfs_set_log_full_commit(trans); | |
3706 | return; | |
3707 | } | |
3a5f1d45 | 3708 | |
e02119d5 CM |
3709 | ret = join_running_log_trans(root); |
3710 | if (ret) | |
9a35fc95 | 3711 | return; |
e02119d5 | 3712 | log = root->log_root; |
a491abb2 | 3713 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3714 | |
a491abb2 | 3715 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 | 3716 | dirid, &index); |
a491abb2 | 3717 | mutex_unlock(&inode->log_mutex); |
9a35fc95 | 3718 | if (ret < 0 && ret != -ENOENT) |
90787766 | 3719 | btrfs_set_log_full_commit(trans); |
12fcfd22 | 3720 | btrfs_end_log_trans(root); |
e02119d5 CM |
3721 | } |
3722 | ||
3723 | /* | |
3724 | * creates a range item in the log for 'dirid'. first_offset and | |
3725 | * last_offset tell us which parts of the key space the log should | |
3726 | * be considered authoritative for. | |
3727 | */ | |
3728 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3729 | struct btrfs_root *log, | |
3730 | struct btrfs_path *path, | |
339d0354 | 3731 | u64 dirid, |
e02119d5 CM |
3732 | u64 first_offset, u64 last_offset) |
3733 | { | |
3734 | int ret; | |
3735 | struct btrfs_key key; | |
3736 | struct btrfs_dir_log_item *item; | |
3737 | ||
3738 | key.objectid = dirid; | |
3739 | key.offset = first_offset; | |
339d0354 | 3740 | key.type = BTRFS_DIR_LOG_INDEX_KEY; |
e02119d5 | 3741 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); |
4a500fd1 YZ |
3742 | if (ret) |
3743 | return ret; | |
e02119d5 CM |
3744 | |
3745 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3746 | struct btrfs_dir_log_item); | |
3747 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3748 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3749 | btrfs_release_path(path); |
e02119d5 CM |
3750 | return 0; |
3751 | } | |
3752 | ||
086dcbfa FM |
3753 | static int flush_dir_items_batch(struct btrfs_trans_handle *trans, |
3754 | struct btrfs_root *log, | |
3755 | struct extent_buffer *src, | |
3756 | struct btrfs_path *dst_path, | |
3757 | int start_slot, | |
3758 | int count) | |
3759 | { | |
3760 | char *ins_data = NULL; | |
b7ef5f3a | 3761 | struct btrfs_item_batch batch; |
086dcbfa | 3762 | struct extent_buffer *dst; |
da1b811f FM |
3763 | unsigned long src_offset; |
3764 | unsigned long dst_offset; | |
086dcbfa FM |
3765 | struct btrfs_key key; |
3766 | u32 item_size; | |
3767 | int ret; | |
3768 | int i; | |
3769 | ||
3770 | ASSERT(count > 0); | |
b7ef5f3a | 3771 | batch.nr = count; |
086dcbfa FM |
3772 | |
3773 | if (count == 1) { | |
3774 | btrfs_item_key_to_cpu(src, &key, start_slot); | |
3212fa14 | 3775 | item_size = btrfs_item_size(src, start_slot); |
b7ef5f3a FM |
3776 | batch.keys = &key; |
3777 | batch.data_sizes = &item_size; | |
3778 | batch.total_data_size = item_size; | |
086dcbfa | 3779 | } else { |
b7ef5f3a FM |
3780 | struct btrfs_key *ins_keys; |
3781 | u32 *ins_sizes; | |
3782 | ||
086dcbfa FM |
3783 | ins_data = kmalloc(count * sizeof(u32) + |
3784 | count * sizeof(struct btrfs_key), GFP_NOFS); | |
3785 | if (!ins_data) | |
3786 | return -ENOMEM; | |
3787 | ||
3788 | ins_sizes = (u32 *)ins_data; | |
3789 | ins_keys = (struct btrfs_key *)(ins_data + count * sizeof(u32)); | |
b7ef5f3a FM |
3790 | batch.keys = ins_keys; |
3791 | batch.data_sizes = ins_sizes; | |
3792 | batch.total_data_size = 0; | |
086dcbfa FM |
3793 | |
3794 | for (i = 0; i < count; i++) { | |
3795 | const int slot = start_slot + i; | |
3796 | ||
3797 | btrfs_item_key_to_cpu(src, &ins_keys[i], slot); | |
3212fa14 | 3798 | ins_sizes[i] = btrfs_item_size(src, slot); |
b7ef5f3a | 3799 | batch.total_data_size += ins_sizes[i]; |
086dcbfa FM |
3800 | } |
3801 | } | |
3802 | ||
b7ef5f3a | 3803 | ret = btrfs_insert_empty_items(trans, log, dst_path, &batch); |
086dcbfa FM |
3804 | if (ret) |
3805 | goto out; | |
3806 | ||
3807 | dst = dst_path->nodes[0]; | |
da1b811f FM |
3808 | /* |
3809 | * Copy all the items in bulk, in a single copy operation. Item data is | |
3810 | * organized such that it's placed at the end of a leaf and from right | |
3811 | * to left. For example, the data for the second item ends at an offset | |
3812 | * that matches the offset where the data for the first item starts, the | |
3813 | * data for the third item ends at an offset that matches the offset | |
3814 | * where the data of the second items starts, and so on. | |
3815 | * Therefore our source and destination start offsets for copy match the | |
3816 | * offsets of the last items (highest slots). | |
3817 | */ | |
3818 | dst_offset = btrfs_item_ptr_offset(dst, dst_path->slots[0] + count - 1); | |
3819 | src_offset = btrfs_item_ptr_offset(src, start_slot + count - 1); | |
3820 | copy_extent_buffer(dst, src, dst_offset, src_offset, batch.total_data_size); | |
086dcbfa FM |
3821 | btrfs_release_path(dst_path); |
3822 | out: | |
3823 | kfree(ins_data); | |
3824 | ||
3825 | return ret; | |
3826 | } | |
3827 | ||
eb10d85e FM |
3828 | static int process_dir_items_leaf(struct btrfs_trans_handle *trans, |
3829 | struct btrfs_inode *inode, | |
3830 | struct btrfs_path *path, | |
3831 | struct btrfs_path *dst_path, | |
732d591a FM |
3832 | struct btrfs_log_ctx *ctx, |
3833 | u64 *last_old_dentry_offset) | |
eb10d85e FM |
3834 | { |
3835 | struct btrfs_root *log = inode->root->log_root; | |
3836 | struct extent_buffer *src = path->nodes[0]; | |
3837 | const int nritems = btrfs_header_nritems(src); | |
3838 | const u64 ino = btrfs_ino(inode); | |
086dcbfa FM |
3839 | bool last_found = false; |
3840 | int batch_start = 0; | |
3841 | int batch_size = 0; | |
eb10d85e FM |
3842 | int i; |
3843 | ||
3844 | for (i = path->slots[0]; i < nritems; i++) { | |
732d591a | 3845 | struct btrfs_dir_item *di; |
eb10d85e | 3846 | struct btrfs_key key; |
eb10d85e FM |
3847 | int ret; |
3848 | ||
3849 | btrfs_item_key_to_cpu(src, &key, i); | |
3850 | ||
339d0354 | 3851 | if (key.objectid != ino || key.type != BTRFS_DIR_INDEX_KEY) { |
086dcbfa FM |
3852 | last_found = true; |
3853 | break; | |
3854 | } | |
eb10d85e | 3855 | |
732d591a | 3856 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); |
dc287224 | 3857 | ctx->last_dir_item_offset = key.offset; |
732d591a FM |
3858 | |
3859 | /* | |
3860 | * Skip ranges of items that consist only of dir item keys created | |
3861 | * in past transactions. However if we find a gap, we must log a | |
3862 | * dir index range item for that gap, so that index keys in that | |
3863 | * gap are deleted during log replay. | |
3864 | */ | |
3865 | if (btrfs_dir_transid(src, di) < trans->transid) { | |
3866 | if (key.offset > *last_old_dentry_offset + 1) { | |
3867 | ret = insert_dir_log_key(trans, log, dst_path, | |
3868 | ino, *last_old_dentry_offset + 1, | |
3869 | key.offset - 1); | |
3870 | /* | |
3871 | * -EEXIST should never happen because when we | |
3872 | * log a directory in full mode (LOG_INODE_ALL) | |
3873 | * we drop all BTRFS_DIR_LOG_INDEX_KEY keys from | |
3874 | * the log tree. | |
3875 | */ | |
3876 | ASSERT(ret != -EEXIST); | |
3877 | if (ret < 0) | |
3878 | return ret; | |
3879 | } | |
3880 | ||
3881 | *last_old_dentry_offset = key.offset; | |
3882 | continue; | |
3883 | } | |
eb10d85e FM |
3884 | /* |
3885 | * We must make sure that when we log a directory entry, the | |
3886 | * corresponding inode, after log replay, has a matching link | |
3887 | * count. For example: | |
3888 | * | |
3889 | * touch foo | |
3890 | * mkdir mydir | |
3891 | * sync | |
3892 | * ln foo mydir/bar | |
3893 | * xfs_io -c "fsync" mydir | |
3894 | * <crash> | |
3895 | * <mount fs and log replay> | |
3896 | * | |
3897 | * Would result in a fsync log that when replayed, our file inode | |
3898 | * would have a link count of 1, but we get two directory entries | |
3899 | * pointing to the same inode. After removing one of the names, | |
3900 | * it would not be possible to remove the other name, which | |
3901 | * resulted always in stale file handle errors, and would not be | |
3902 | * possible to rmdir the parent directory, since its i_size could | |
3903 | * never be decremented to the value BTRFS_EMPTY_DIR_SIZE, | |
3904 | * resulting in -ENOTEMPTY errors. | |
3905 | */ | |
086dcbfa | 3906 | if (!ctx->log_new_dentries) { |
086dcbfa FM |
3907 | struct btrfs_key di_key; |
3908 | ||
086dcbfa | 3909 | btrfs_dir_item_key_to_cpu(src, di, &di_key); |
732d591a | 3910 | if (di_key.type != BTRFS_ROOT_ITEM_KEY) |
086dcbfa FM |
3911 | ctx->log_new_dentries = true; |
3912 | } | |
3913 | ||
0f8ce498 | 3914 | if (!ctx->logged_before) |
086dcbfa | 3915 | goto add_to_batch; |
dc287224 FM |
3916 | |
3917 | /* | |
3918 | * If we were logged before and have logged dir items, we can skip | |
3919 | * checking if any item with a key offset larger than the last one | |
3920 | * we logged is in the log tree, saving time and avoiding adding | |
0f8ce498 FM |
3921 | * contention on the log tree. We can only rely on the value of |
3922 | * last_dir_index_offset when we know for sure that the inode was | |
3923 | * previously logged in the current transaction. | |
dc287224 | 3924 | */ |
339d0354 | 3925 | if (key.offset > inode->last_dir_index_offset) |
dc287224 | 3926 | goto add_to_batch; |
086dcbfa FM |
3927 | /* |
3928 | * Check if the key was already logged before. If not we can add | |
3929 | * it to a batch for bulk insertion. | |
3930 | */ | |
3931 | ret = btrfs_search_slot(NULL, log, &key, dst_path, 0, 0); | |
3932 | if (ret < 0) { | |
3933 | return ret; | |
3934 | } else if (ret > 0) { | |
3935 | btrfs_release_path(dst_path); | |
3936 | goto add_to_batch; | |
3937 | } | |
3938 | ||
3939 | /* | |
3940 | * Item exists in the log. Overwrite the item in the log if it | |
3941 | * has different content or do nothing if it has exactly the same | |
3942 | * content. And then flush the current batch if any - do it after | |
3943 | * overwriting the current item, or we would deadlock otherwise, | |
3944 | * since we are holding a path for the existing item. | |
3945 | */ | |
3946 | ret = do_overwrite_item(trans, log, dst_path, src, i, &key); | |
3947 | if (ret < 0) | |
3948 | return ret; | |
3949 | ||
3950 | if (batch_size > 0) { | |
3951 | ret = flush_dir_items_batch(trans, log, src, dst_path, | |
3952 | batch_start, batch_size); | |
3953 | if (ret < 0) | |
3954 | return ret; | |
3955 | batch_size = 0; | |
3956 | } | |
3957 | continue; | |
3958 | add_to_batch: | |
3959 | if (batch_size == 0) | |
3960 | batch_start = i; | |
3961 | batch_size++; | |
eb10d85e FM |
3962 | } |
3963 | ||
086dcbfa FM |
3964 | if (batch_size > 0) { |
3965 | int ret; | |
3966 | ||
3967 | ret = flush_dir_items_batch(trans, log, src, dst_path, | |
3968 | batch_start, batch_size); | |
3969 | if (ret < 0) | |
3970 | return ret; | |
3971 | } | |
3972 | ||
3973 | return last_found ? 1 : 0; | |
eb10d85e FM |
3974 | } |
3975 | ||
e02119d5 CM |
3976 | /* |
3977 | * log all the items included in the current transaction for a given | |
3978 | * directory. This also creates the range items in the log tree required | |
3979 | * to replay anything deleted before the fsync | |
3980 | */ | |
3981 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
90d04510 | 3982 | struct btrfs_inode *inode, |
e02119d5 | 3983 | struct btrfs_path *path, |
339d0354 | 3984 | struct btrfs_path *dst_path, |
2f2ff0ee | 3985 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3986 | u64 min_offset, u64 *last_offset_ret) |
3987 | { | |
3988 | struct btrfs_key min_key; | |
90d04510 | 3989 | struct btrfs_root *root = inode->root; |
e02119d5 | 3990 | struct btrfs_root *log = root->log_root; |
4a500fd1 | 3991 | int err = 0; |
e02119d5 | 3992 | int ret; |
732d591a | 3993 | u64 last_old_dentry_offset = min_offset - 1; |
e02119d5 | 3994 | u64 last_offset = (u64)-1; |
684a5773 | 3995 | u64 ino = btrfs_ino(inode); |
e02119d5 | 3996 | |
33345d01 | 3997 | min_key.objectid = ino; |
339d0354 | 3998 | min_key.type = BTRFS_DIR_INDEX_KEY; |
e02119d5 CM |
3999 | min_key.offset = min_offset; |
4000 | ||
6174d3cb | 4001 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
4002 | |
4003 | /* | |
4004 | * we didn't find anything from this transaction, see if there | |
4005 | * is anything at all | |
4006 | */ | |
339d0354 FM |
4007 | if (ret != 0 || min_key.objectid != ino || |
4008 | min_key.type != BTRFS_DIR_INDEX_KEY) { | |
33345d01 | 4009 | min_key.objectid = ino; |
339d0354 | 4010 | min_key.type = BTRFS_DIR_INDEX_KEY; |
e02119d5 | 4011 | min_key.offset = (u64)-1; |
b3b4aa74 | 4012 | btrfs_release_path(path); |
e02119d5 CM |
4013 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
4014 | if (ret < 0) { | |
b3b4aa74 | 4015 | btrfs_release_path(path); |
e02119d5 CM |
4016 | return ret; |
4017 | } | |
339d0354 | 4018 | ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY); |
e02119d5 CM |
4019 | |
4020 | /* if ret == 0 there are items for this type, | |
4021 | * create a range to tell us the last key of this type. | |
4022 | * otherwise, there are no items in this directory after | |
4023 | * *min_offset, and we create a range to indicate that. | |
4024 | */ | |
4025 | if (ret == 0) { | |
4026 | struct btrfs_key tmp; | |
732d591a | 4027 | |
e02119d5 CM |
4028 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, |
4029 | path->slots[0]); | |
339d0354 | 4030 | if (tmp.type == BTRFS_DIR_INDEX_KEY) |
732d591a | 4031 | last_old_dentry_offset = tmp.offset; |
e02119d5 CM |
4032 | } |
4033 | goto done; | |
4034 | } | |
4035 | ||
4036 | /* go backward to find any previous key */ | |
339d0354 | 4037 | ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY); |
e02119d5 CM |
4038 | if (ret == 0) { |
4039 | struct btrfs_key tmp; | |
a450a4af | 4040 | |
e02119d5 | 4041 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); |
a450a4af FM |
4042 | /* |
4043 | * The dir index key before the first one we found that needs to | |
4044 | * be logged might be in a previous leaf, and there might be a | |
4045 | * gap between these keys, meaning that we had deletions that | |
4046 | * happened. So the key range item we log (key type | |
4047 | * BTRFS_DIR_LOG_INDEX_KEY) must cover a range that starts at the | |
4048 | * previous key's offset plus 1, so that those deletes are replayed. | |
4049 | */ | |
4050 | if (tmp.type == BTRFS_DIR_INDEX_KEY) | |
732d591a | 4051 | last_old_dentry_offset = tmp.offset; |
e02119d5 | 4052 | } |
b3b4aa74 | 4053 | btrfs_release_path(path); |
e02119d5 | 4054 | |
2cc83342 JB |
4055 | /* |
4056 | * Find the first key from this transaction again. See the note for | |
4057 | * log_new_dir_dentries, if we're logging a directory recursively we | |
4058 | * won't be holding its i_mutex, which means we can modify the directory | |
4059 | * while we're logging it. If we remove an entry between our first | |
4060 | * search and this search we'll not find the key again and can just | |
4061 | * bail. | |
4062 | */ | |
bb56f02f | 4063 | search: |
e02119d5 | 4064 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
2cc83342 | 4065 | if (ret != 0) |
e02119d5 | 4066 | goto done; |
e02119d5 CM |
4067 | |
4068 | /* | |
4069 | * we have a block from this transaction, log every item in it | |
4070 | * from our directory | |
4071 | */ | |
d397712b | 4072 | while (1) { |
732d591a FM |
4073 | ret = process_dir_items_leaf(trans, inode, path, dst_path, ctx, |
4074 | &last_old_dentry_offset); | |
eb10d85e FM |
4075 | if (ret != 0) { |
4076 | if (ret < 0) | |
4a500fd1 | 4077 | err = ret; |
eb10d85e | 4078 | goto done; |
e02119d5 | 4079 | } |
eb10d85e | 4080 | path->slots[0] = btrfs_header_nritems(path->nodes[0]); |
e02119d5 CM |
4081 | |
4082 | /* | |
4083 | * look ahead to the next item and see if it is also | |
4084 | * from this directory and from this transaction | |
4085 | */ | |
4086 | ret = btrfs_next_leaf(root, path); | |
80c0b421 LB |
4087 | if (ret) { |
4088 | if (ret == 1) | |
4089 | last_offset = (u64)-1; | |
4090 | else | |
4091 | err = ret; | |
e02119d5 CM |
4092 | goto done; |
4093 | } | |
eb10d85e | 4094 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, path->slots[0]); |
339d0354 | 4095 | if (min_key.objectid != ino || min_key.type != BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
4096 | last_offset = (u64)-1; |
4097 | goto done; | |
4098 | } | |
4099 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
a450a4af FM |
4100 | /* |
4101 | * The next leaf was not changed in the current transaction | |
4102 | * and has at least one dir index key. | |
4103 | * We check for the next key because there might have been | |
4104 | * one or more deletions between the last key we logged and | |
4105 | * that next key. So the key range item we log (key type | |
4106 | * BTRFS_DIR_LOG_INDEX_KEY) must end at the next key's | |
4107 | * offset minus 1, so that those deletes are replayed. | |
4108 | */ | |
4109 | last_offset = min_key.offset - 1; | |
e02119d5 CM |
4110 | goto done; |
4111 | } | |
eb10d85e FM |
4112 | if (need_resched()) { |
4113 | btrfs_release_path(path); | |
4114 | cond_resched(); | |
4115 | goto search; | |
4116 | } | |
e02119d5 CM |
4117 | } |
4118 | done: | |
b3b4aa74 DS |
4119 | btrfs_release_path(path); |
4120 | btrfs_release_path(dst_path); | |
e02119d5 | 4121 | |
4a500fd1 YZ |
4122 | if (err == 0) { |
4123 | *last_offset_ret = last_offset; | |
4124 | /* | |
732d591a FM |
4125 | * In case the leaf was changed in the current transaction but |
4126 | * all its dir items are from a past transaction, the last item | |
4127 | * in the leaf is a dir item and there's no gap between that last | |
4128 | * dir item and the first one on the next leaf (which did not | |
4129 | * change in the current transaction), then we don't need to log | |
4130 | * a range, last_old_dentry_offset is == to last_offset. | |
4a500fd1 | 4131 | */ |
732d591a FM |
4132 | ASSERT(last_old_dentry_offset <= last_offset); |
4133 | if (last_old_dentry_offset < last_offset) { | |
4134 | ret = insert_dir_log_key(trans, log, path, ino, | |
4135 | last_old_dentry_offset + 1, | |
4136 | last_offset); | |
4137 | if (ret) | |
4138 | err = ret; | |
4139 | } | |
4a500fd1 YZ |
4140 | } |
4141 | return err; | |
e02119d5 CM |
4142 | } |
4143 | ||
4144 | /* | |
4145 | * logging directories is very similar to logging inodes, We find all the items | |
4146 | * from the current transaction and write them to the log. | |
4147 | * | |
4148 | * The recovery code scans the directory in the subvolume, and if it finds a | |
4149 | * key in the range logged that is not present in the log tree, then it means | |
4150 | * that dir entry was unlinked during the transaction. | |
4151 | * | |
4152 | * In order for that scan to work, we must include one key smaller than | |
4153 | * the smallest logged by this transaction and one key larger than the largest | |
4154 | * key logged by this transaction. | |
4155 | */ | |
4156 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
90d04510 | 4157 | struct btrfs_inode *inode, |
e02119d5 | 4158 | struct btrfs_path *path, |
2f2ff0ee FM |
4159 | struct btrfs_path *dst_path, |
4160 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
4161 | { |
4162 | u64 min_key; | |
4163 | u64 max_key; | |
4164 | int ret; | |
e02119d5 | 4165 | |
732d591a | 4166 | min_key = BTRFS_DIR_START_INDEX; |
e02119d5 | 4167 | max_key = 0; |
339d0354 | 4168 | ctx->last_dir_item_offset = inode->last_dir_index_offset; |
dc287224 | 4169 | |
d397712b | 4170 | while (1) { |
339d0354 | 4171 | ret = log_dir_items(trans, inode, path, dst_path, |
dbf39ea4 | 4172 | ctx, min_key, &max_key); |
4a500fd1 YZ |
4173 | if (ret) |
4174 | return ret; | |
e02119d5 CM |
4175 | if (max_key == (u64)-1) |
4176 | break; | |
4177 | min_key = max_key + 1; | |
4178 | } | |
4179 | ||
339d0354 FM |
4180 | inode->last_dir_index_offset = ctx->last_dir_item_offset; |
4181 | ||
e02119d5 CM |
4182 | return 0; |
4183 | } | |
4184 | ||
4185 | /* | |
4186 | * a helper function to drop items from the log before we relog an | |
4187 | * inode. max_key_type indicates the highest item type to remove. | |
4188 | * This cannot be run for file data extents because it does not | |
4189 | * free the extents they point to. | |
4190 | */ | |
88e221cd | 4191 | static int drop_inode_items(struct btrfs_trans_handle *trans, |
e02119d5 CM |
4192 | struct btrfs_root *log, |
4193 | struct btrfs_path *path, | |
88e221cd FM |
4194 | struct btrfs_inode *inode, |
4195 | int max_key_type) | |
e02119d5 CM |
4196 | { |
4197 | int ret; | |
4198 | struct btrfs_key key; | |
4199 | struct btrfs_key found_key; | |
18ec90d6 | 4200 | int start_slot; |
e02119d5 | 4201 | |
88e221cd | 4202 | key.objectid = btrfs_ino(inode); |
e02119d5 CM |
4203 | key.type = max_key_type; |
4204 | key.offset = (u64)-1; | |
4205 | ||
d397712b | 4206 | while (1) { |
e02119d5 | 4207 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 4208 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 4209 | if (ret < 0) |
e02119d5 CM |
4210 | break; |
4211 | ||
4212 | if (path->slots[0] == 0) | |
4213 | break; | |
4214 | ||
4215 | path->slots[0]--; | |
4216 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
4217 | path->slots[0]); | |
4218 | ||
88e221cd | 4219 | if (found_key.objectid != key.objectid) |
e02119d5 CM |
4220 | break; |
4221 | ||
18ec90d6 JB |
4222 | found_key.offset = 0; |
4223 | found_key.type = 0; | |
e3b83361 | 4224 | ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot); |
cbca7d59 FM |
4225 | if (ret < 0) |
4226 | break; | |
18ec90d6 JB |
4227 | |
4228 | ret = btrfs_del_items(trans, log, path, start_slot, | |
4229 | path->slots[0] - start_slot + 1); | |
4230 | /* | |
4231 | * If start slot isn't 0 then we don't need to re-search, we've | |
4232 | * found the last guy with the objectid in this tree. | |
4233 | */ | |
4234 | if (ret || start_slot != 0) | |
65a246c5 | 4235 | break; |
b3b4aa74 | 4236 | btrfs_release_path(path); |
e02119d5 | 4237 | } |
b3b4aa74 | 4238 | btrfs_release_path(path); |
5bdbeb21 JB |
4239 | if (ret > 0) |
4240 | ret = 0; | |
4a500fd1 | 4241 | return ret; |
e02119d5 CM |
4242 | } |
4243 | ||
8a2b3da1 FM |
4244 | static int truncate_inode_items(struct btrfs_trans_handle *trans, |
4245 | struct btrfs_root *log_root, | |
4246 | struct btrfs_inode *inode, | |
4247 | u64 new_size, u32 min_type) | |
4248 | { | |
d9ac19c3 JB |
4249 | struct btrfs_truncate_control control = { |
4250 | .new_size = new_size, | |
487e81d2 | 4251 | .ino = btrfs_ino(inode), |
d9ac19c3 | 4252 | .min_type = min_type, |
5caa490e | 4253 | .skip_ref_updates = true, |
d9ac19c3 | 4254 | }; |
8a2b3da1 | 4255 | |
8697b8f8 | 4256 | return btrfs_truncate_inode_items(trans, log_root, &control); |
8a2b3da1 FM |
4257 | } |
4258 | ||
94edf4ae JB |
4259 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
4260 | struct extent_buffer *leaf, | |
4261 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
4262 | struct inode *inode, int log_inode_only, |
4263 | u64 logged_isize) | |
94edf4ae | 4264 | { |
0b1c6cca | 4265 | struct btrfs_map_token token; |
77eea05e | 4266 | u64 flags; |
0b1c6cca | 4267 | |
c82f823c | 4268 | btrfs_init_map_token(&token, leaf); |
94edf4ae JB |
4269 | |
4270 | if (log_inode_only) { | |
4271 | /* set the generation to zero so the recover code | |
4272 | * can tell the difference between an logging | |
4273 | * just to say 'this inode exists' and a logging | |
4274 | * to say 'update this inode with these values' | |
4275 | */ | |
cc4c13d5 DS |
4276 | btrfs_set_token_inode_generation(&token, item, 0); |
4277 | btrfs_set_token_inode_size(&token, item, logged_isize); | |
94edf4ae | 4278 | } else { |
cc4c13d5 DS |
4279 | btrfs_set_token_inode_generation(&token, item, |
4280 | BTRFS_I(inode)->generation); | |
4281 | btrfs_set_token_inode_size(&token, item, inode->i_size); | |
0b1c6cca JB |
4282 | } |
4283 | ||
cc4c13d5 DS |
4284 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
4285 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
4286 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
4287 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
4288 | ||
4289 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
4290 | inode->i_atime.tv_sec); | |
4291 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
4292 | inode->i_atime.tv_nsec); | |
4293 | ||
4294 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
4295 | inode->i_mtime.tv_sec); | |
4296 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
4297 | inode->i_mtime.tv_nsec); | |
4298 | ||
4299 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
4300 | inode->i_ctime.tv_sec); | |
4301 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
4302 | inode->i_ctime.tv_nsec); | |
4303 | ||
e593e54e FM |
4304 | /* |
4305 | * We do not need to set the nbytes field, in fact during a fast fsync | |
4306 | * its value may not even be correct, since a fast fsync does not wait | |
4307 | * for ordered extent completion, which is where we update nbytes, it | |
4308 | * only waits for writeback to complete. During log replay as we find | |
4309 | * file extent items and replay them, we adjust the nbytes field of the | |
4310 | * inode item in subvolume tree as needed (see overwrite_item()). | |
4311 | */ | |
cc4c13d5 DS |
4312 | |
4313 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
4314 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
4315 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
77eea05e BB |
4316 | flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags, |
4317 | BTRFS_I(inode)->ro_flags); | |
4318 | btrfs_set_token_inode_flags(&token, item, flags); | |
cc4c13d5 | 4319 | btrfs_set_token_inode_block_group(&token, item, 0); |
94edf4ae JB |
4320 | } |
4321 | ||
a95249b3 JB |
4322 | static int log_inode_item(struct btrfs_trans_handle *trans, |
4323 | struct btrfs_root *log, struct btrfs_path *path, | |
2ac691d8 | 4324 | struct btrfs_inode *inode, bool inode_item_dropped) |
a95249b3 JB |
4325 | { |
4326 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
4327 | int ret; |
4328 | ||
2ac691d8 FM |
4329 | /* |
4330 | * If we are doing a fast fsync and the inode was logged before in the | |
4331 | * current transaction, then we know the inode was previously logged and | |
4332 | * it exists in the log tree. For performance reasons, in this case use | |
4333 | * btrfs_search_slot() directly with ins_len set to 0 so that we never | |
4334 | * attempt a write lock on the leaf's parent, which adds unnecessary lock | |
4335 | * contention in case there are concurrent fsyncs for other inodes of the | |
4336 | * same subvolume. Using btrfs_insert_empty_item() when the inode item | |
4337 | * already exists can also result in unnecessarily splitting a leaf. | |
4338 | */ | |
4339 | if (!inode_item_dropped && inode->logged_trans == trans->transid) { | |
4340 | ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1); | |
4341 | ASSERT(ret <= 0); | |
4342 | if (ret > 0) | |
4343 | ret = -ENOENT; | |
4344 | } else { | |
4345 | /* | |
4346 | * This means it is the first fsync in the current transaction, | |
4347 | * so the inode item is not in the log and we need to insert it. | |
4348 | * We can never get -EEXIST because we are only called for a fast | |
4349 | * fsync and in case an inode eviction happens after the inode was | |
4350 | * logged before in the current transaction, when we load again | |
4351 | * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime | |
4352 | * flags and set ->logged_trans to 0. | |
4353 | */ | |
4354 | ret = btrfs_insert_empty_item(trans, log, path, &inode->location, | |
4355 | sizeof(*inode_item)); | |
4356 | ASSERT(ret != -EEXIST); | |
4357 | } | |
4358 | if (ret) | |
a95249b3 JB |
4359 | return ret; |
4360 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4361 | struct btrfs_inode_item); | |
6d889a3b NB |
4362 | fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode, |
4363 | 0, 0); | |
a95249b3 JB |
4364 | btrfs_release_path(path); |
4365 | return 0; | |
4366 | } | |
4367 | ||
40e046ac | 4368 | static int log_csums(struct btrfs_trans_handle *trans, |
3ebac17c | 4369 | struct btrfs_inode *inode, |
40e046ac FM |
4370 | struct btrfs_root *log_root, |
4371 | struct btrfs_ordered_sum *sums) | |
4372 | { | |
e289f03e FM |
4373 | const u64 lock_end = sums->bytenr + sums->len - 1; |
4374 | struct extent_state *cached_state = NULL; | |
40e046ac FM |
4375 | int ret; |
4376 | ||
3ebac17c FM |
4377 | /* |
4378 | * If this inode was not used for reflink operations in the current | |
4379 | * transaction with new extents, then do the fast path, no need to | |
4380 | * worry about logging checksum items with overlapping ranges. | |
4381 | */ | |
4382 | if (inode->last_reflink_trans < trans->transid) | |
4383 | return btrfs_csum_file_blocks(trans, log_root, sums); | |
4384 | ||
e289f03e FM |
4385 | /* |
4386 | * Serialize logging for checksums. This is to avoid racing with the | |
4387 | * same checksum being logged by another task that is logging another | |
4388 | * file which happens to refer to the same extent as well. Such races | |
4389 | * can leave checksum items in the log with overlapping ranges. | |
4390 | */ | |
4391 | ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr, | |
4392 | lock_end, &cached_state); | |
4393 | if (ret) | |
4394 | return ret; | |
40e046ac FM |
4395 | /* |
4396 | * Due to extent cloning, we might have logged a csum item that covers a | |
4397 | * subrange of a cloned extent, and later we can end up logging a csum | |
4398 | * item for a larger subrange of the same extent or the entire range. | |
4399 | * This would leave csum items in the log tree that cover the same range | |
4400 | * and break the searches for checksums in the log tree, resulting in | |
4401 | * some checksums missing in the fs/subvolume tree. So just delete (or | |
4402 | * trim and adjust) any existing csum items in the log for this range. | |
4403 | */ | |
4404 | ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len); | |
e289f03e FM |
4405 | if (!ret) |
4406 | ret = btrfs_csum_file_blocks(trans, log_root, sums); | |
40e046ac | 4407 | |
e289f03e FM |
4408 | unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end, |
4409 | &cached_state); | |
4410 | ||
4411 | return ret; | |
40e046ac FM |
4412 | } |
4413 | ||
31ff1cd2 | 4414 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 4415 | struct btrfs_inode *inode, |
31ff1cd2 | 4416 | struct btrfs_path *dst_path, |
0e56315c | 4417 | struct btrfs_path *src_path, |
1a4bcf47 FM |
4418 | int start_slot, int nr, int inode_only, |
4419 | u64 logged_isize) | |
31ff1cd2 | 4420 | { |
44d70e19 | 4421 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 | 4422 | struct btrfs_file_extent_item *extent; |
16e7549f | 4423 | struct extent_buffer *src = src_path->nodes[0]; |
7f30c072 | 4424 | int ret = 0; |
31ff1cd2 CM |
4425 | struct btrfs_key *ins_keys; |
4426 | u32 *ins_sizes; | |
b7ef5f3a | 4427 | struct btrfs_item_batch batch; |
31ff1cd2 CM |
4428 | char *ins_data; |
4429 | int i; | |
7f30c072 | 4430 | int dst_index; |
7f30c072 FM |
4431 | const bool skip_csum = (inode->flags & BTRFS_INODE_NODATASUM); |
4432 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
d20f7043 | 4433 | |
31ff1cd2 CM |
4434 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + |
4435 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 4436 | if (!ins_data) |
4437 | return -ENOMEM; | |
4438 | ||
31ff1cd2 CM |
4439 | ins_sizes = (u32 *)ins_data; |
4440 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
b7ef5f3a FM |
4441 | batch.keys = ins_keys; |
4442 | batch.data_sizes = ins_sizes; | |
4443 | batch.total_data_size = 0; | |
7f30c072 | 4444 | batch.nr = 0; |
31ff1cd2 | 4445 | |
7f30c072 | 4446 | dst_index = 0; |
31ff1cd2 | 4447 | for (i = 0; i < nr; i++) { |
7f30c072 FM |
4448 | const int src_slot = start_slot + i; |
4449 | struct btrfs_root *csum_root; | |
5b7ce5e2 FM |
4450 | struct btrfs_ordered_sum *sums; |
4451 | struct btrfs_ordered_sum *sums_next; | |
4452 | LIST_HEAD(ordered_sums); | |
7f30c072 FM |
4453 | u64 disk_bytenr; |
4454 | u64 disk_num_bytes; | |
4455 | u64 extent_offset; | |
4456 | u64 extent_num_bytes; | |
4457 | bool is_old_extent; | |
4458 | ||
4459 | btrfs_item_key_to_cpu(src, &ins_keys[dst_index], src_slot); | |
4460 | ||
4461 | if (ins_keys[dst_index].type != BTRFS_EXTENT_DATA_KEY) | |
4462 | goto add_to_batch; | |
4463 | ||
4464 | extent = btrfs_item_ptr(src, src_slot, | |
4465 | struct btrfs_file_extent_item); | |
4466 | ||
4467 | is_old_extent = (btrfs_file_extent_generation(src, extent) < | |
4468 | trans->transid); | |
4469 | ||
4470 | /* | |
4471 | * Don't copy extents from past generations. That would make us | |
4472 | * log a lot more metadata for common cases like doing only a | |
4473 | * few random writes into a file and then fsync it for the first | |
4474 | * time or after the full sync flag is set on the inode. We can | |
4475 | * get leaves full of extent items, most of which are from past | |
4476 | * generations, so we can skip them - as long as the inode has | |
4477 | * not been the target of a reflink operation in this transaction, | |
4478 | * as in that case it might have had file extent items with old | |
4479 | * generations copied into it. We also must always log prealloc | |
4480 | * extents that start at or beyond eof, otherwise we would lose | |
4481 | * them on log replay. | |
4482 | */ | |
4483 | if (is_old_extent && | |
4484 | ins_keys[dst_index].offset < i_size && | |
4485 | inode->last_reflink_trans < trans->transid) | |
4486 | continue; | |
4487 | ||
4488 | if (skip_csum) | |
4489 | goto add_to_batch; | |
4490 | ||
4491 | /* Only regular extents have checksums. */ | |
4492 | if (btrfs_file_extent_type(src, extent) != BTRFS_FILE_EXTENT_REG) | |
4493 | goto add_to_batch; | |
4494 | ||
4495 | /* | |
4496 | * If it's an extent created in a past transaction, then its | |
4497 | * checksums are already accessible from the committed csum tree, | |
4498 | * no need to log them. | |
4499 | */ | |
4500 | if (is_old_extent) | |
4501 | goto add_to_batch; | |
4502 | ||
4503 | disk_bytenr = btrfs_file_extent_disk_bytenr(src, extent); | |
4504 | /* If it's an explicit hole, there are no checksums. */ | |
4505 | if (disk_bytenr == 0) | |
4506 | goto add_to_batch; | |
4507 | ||
4508 | disk_num_bytes = btrfs_file_extent_disk_num_bytes(src, extent); | |
4509 | ||
4510 | if (btrfs_file_extent_compression(src, extent)) { | |
4511 | extent_offset = 0; | |
4512 | extent_num_bytes = disk_num_bytes; | |
4513 | } else { | |
4514 | extent_offset = btrfs_file_extent_offset(src, extent); | |
4515 | extent_num_bytes = btrfs_file_extent_num_bytes(src, extent); | |
4516 | } | |
4517 | ||
4518 | csum_root = btrfs_csum_root(trans->fs_info, disk_bytenr); | |
4519 | disk_bytenr += extent_offset; | |
4520 | ret = btrfs_lookup_csums_range(csum_root, disk_bytenr, | |
4521 | disk_bytenr + extent_num_bytes - 1, | |
4522 | &ordered_sums, 0); | |
4523 | if (ret) | |
4524 | goto out; | |
4525 | ||
5b7ce5e2 FM |
4526 | list_for_each_entry_safe(sums, sums_next, &ordered_sums, list) { |
4527 | if (!ret) | |
4528 | ret = log_csums(trans, inode, log, sums); | |
4529 | list_del(&sums->list); | |
4530 | kfree(sums); | |
4531 | } | |
4532 | if (ret) | |
4533 | goto out; | |
4534 | ||
7f30c072 FM |
4535 | add_to_batch: |
4536 | ins_sizes[dst_index] = btrfs_item_size(src, src_slot); | |
4537 | batch.total_data_size += ins_sizes[dst_index]; | |
4538 | batch.nr++; | |
4539 | dst_index++; | |
31ff1cd2 | 4540 | } |
7f30c072 FM |
4541 | |
4542 | /* | |
4543 | * We have a leaf full of old extent items that don't need to be logged, | |
4544 | * so we don't need to do anything. | |
4545 | */ | |
4546 | if (batch.nr == 0) | |
4547 | goto out; | |
4548 | ||
b7ef5f3a | 4549 | ret = btrfs_insert_empty_items(trans, log, dst_path, &batch); |
7f30c072 FM |
4550 | if (ret) |
4551 | goto out; | |
4552 | ||
4553 | dst_index = 0; | |
4554 | for (i = 0; i < nr; i++) { | |
4555 | const int src_slot = start_slot + i; | |
4556 | const int dst_slot = dst_path->slots[0] + dst_index; | |
4557 | struct btrfs_key key; | |
4558 | unsigned long src_offset; | |
4559 | unsigned long dst_offset; | |
4560 | ||
4561 | /* | |
4562 | * We're done, all the remaining items in the source leaf | |
4563 | * correspond to old file extent items. | |
4564 | */ | |
4565 | if (dst_index >= batch.nr) | |
4566 | break; | |
4567 | ||
4568 | btrfs_item_key_to_cpu(src, &key, src_slot); | |
4569 | ||
4570 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
4571 | goto copy_item; | |
31ff1cd2 | 4572 | |
7f30c072 FM |
4573 | extent = btrfs_item_ptr(src, src_slot, |
4574 | struct btrfs_file_extent_item); | |
31ff1cd2 | 4575 | |
7f30c072 FM |
4576 | /* See the comment in the previous loop, same logic. */ |
4577 | if (btrfs_file_extent_generation(src, extent) < trans->transid && | |
4578 | key.offset < i_size && | |
4579 | inode->last_reflink_trans < trans->transid) | |
4580 | continue; | |
4581 | ||
4582 | copy_item: | |
4583 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], dst_slot); | |
4584 | src_offset = btrfs_item_ptr_offset(src, src_slot); | |
31ff1cd2 | 4585 | |
7f30c072 FM |
4586 | if (key.type == BTRFS_INODE_ITEM_KEY) { |
4587 | struct btrfs_inode_item *inode_item; | |
4588 | ||
4589 | inode_item = btrfs_item_ptr(dst_path->nodes[0], dst_slot, | |
31ff1cd2 | 4590 | struct btrfs_inode_item); |
94edf4ae | 4591 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
f85b7379 DS |
4592 | &inode->vfs_inode, |
4593 | inode_only == LOG_INODE_EXISTS, | |
1a4bcf47 | 4594 | logged_isize); |
94edf4ae JB |
4595 | } else { |
4596 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
7f30c072 | 4597 | src_offset, ins_sizes[dst_index]); |
31ff1cd2 | 4598 | } |
94edf4ae | 4599 | |
7f30c072 | 4600 | dst_index++; |
31ff1cd2 CM |
4601 | } |
4602 | ||
4603 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 4604 | btrfs_release_path(dst_path); |
7f30c072 | 4605 | out: |
31ff1cd2 | 4606 | kfree(ins_data); |
d20f7043 | 4607 | |
4a500fd1 | 4608 | return ret; |
31ff1cd2 CM |
4609 | } |
4610 | ||
4f0f586b ST |
4611 | static int extent_cmp(void *priv, const struct list_head *a, |
4612 | const struct list_head *b) | |
5dc562c5 | 4613 | { |
214cc184 | 4614 | const struct extent_map *em1, *em2; |
5dc562c5 JB |
4615 | |
4616 | em1 = list_entry(a, struct extent_map, list); | |
4617 | em2 = list_entry(b, struct extent_map, list); | |
4618 | ||
4619 | if (em1->start < em2->start) | |
4620 | return -1; | |
4621 | else if (em1->start > em2->start) | |
4622 | return 1; | |
4623 | return 0; | |
4624 | } | |
4625 | ||
e7175a69 JB |
4626 | static int log_extent_csums(struct btrfs_trans_handle *trans, |
4627 | struct btrfs_inode *inode, | |
a9ecb653 | 4628 | struct btrfs_root *log_root, |
48778179 FM |
4629 | const struct extent_map *em, |
4630 | struct btrfs_log_ctx *ctx) | |
5dc562c5 | 4631 | { |
48778179 | 4632 | struct btrfs_ordered_extent *ordered; |
fc28b25e | 4633 | struct btrfs_root *csum_root; |
2ab28f32 JB |
4634 | u64 csum_offset; |
4635 | u64 csum_len; | |
48778179 FM |
4636 | u64 mod_start = em->mod_start; |
4637 | u64 mod_len = em->mod_len; | |
8407f553 FM |
4638 | LIST_HEAD(ordered_sums); |
4639 | int ret = 0; | |
0aa4a17d | 4640 | |
e7175a69 JB |
4641 | if (inode->flags & BTRFS_INODE_NODATASUM || |
4642 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
8407f553 | 4643 | em->block_start == EXTENT_MAP_HOLE) |
70c8a91c | 4644 | return 0; |
5dc562c5 | 4645 | |
48778179 FM |
4646 | list_for_each_entry(ordered, &ctx->ordered_extents, log_list) { |
4647 | const u64 ordered_end = ordered->file_offset + ordered->num_bytes; | |
4648 | const u64 mod_end = mod_start + mod_len; | |
4649 | struct btrfs_ordered_sum *sums; | |
4650 | ||
4651 | if (mod_len == 0) | |
4652 | break; | |
4653 | ||
4654 | if (ordered_end <= mod_start) | |
4655 | continue; | |
4656 | if (mod_end <= ordered->file_offset) | |
4657 | break; | |
4658 | ||
4659 | /* | |
4660 | * We are going to copy all the csums on this ordered extent, so | |
4661 | * go ahead and adjust mod_start and mod_len in case this ordered | |
4662 | * extent has already been logged. | |
4663 | */ | |
4664 | if (ordered->file_offset > mod_start) { | |
4665 | if (ordered_end >= mod_end) | |
4666 | mod_len = ordered->file_offset - mod_start; | |
4667 | /* | |
4668 | * If we have this case | |
4669 | * | |
4670 | * |--------- logged extent ---------| | |
4671 | * |----- ordered extent ----| | |
4672 | * | |
4673 | * Just don't mess with mod_start and mod_len, we'll | |
4674 | * just end up logging more csums than we need and it | |
4675 | * will be ok. | |
4676 | */ | |
4677 | } else { | |
4678 | if (ordered_end < mod_end) { | |
4679 | mod_len = mod_end - ordered_end; | |
4680 | mod_start = ordered_end; | |
4681 | } else { | |
4682 | mod_len = 0; | |
4683 | } | |
4684 | } | |
4685 | ||
4686 | /* | |
4687 | * To keep us from looping for the above case of an ordered | |
4688 | * extent that falls inside of the logged extent. | |
4689 | */ | |
4690 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags)) | |
4691 | continue; | |
4692 | ||
4693 | list_for_each_entry(sums, &ordered->list, list) { | |
4694 | ret = log_csums(trans, inode, log_root, sums); | |
4695 | if (ret) | |
4696 | return ret; | |
4697 | } | |
4698 | } | |
4699 | ||
4700 | /* We're done, found all csums in the ordered extents. */ | |
4701 | if (mod_len == 0) | |
4702 | return 0; | |
4703 | ||
e7175a69 | 4704 | /* If we're compressed we have to save the entire range of csums. */ |
488111aa FDBM |
4705 | if (em->compress_type) { |
4706 | csum_offset = 0; | |
8407f553 | 4707 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa | 4708 | } else { |
48778179 FM |
4709 | csum_offset = mod_start - em->start; |
4710 | csum_len = mod_len; | |
488111aa | 4711 | } |
2ab28f32 | 4712 | |
70c8a91c | 4713 | /* block start is already adjusted for the file extent offset. */ |
fc28b25e JB |
4714 | csum_root = btrfs_csum_root(trans->fs_info, em->block_start); |
4715 | ret = btrfs_lookup_csums_range(csum_root, | |
70c8a91c JB |
4716 | em->block_start + csum_offset, |
4717 | em->block_start + csum_offset + | |
4718 | csum_len - 1, &ordered_sums, 0); | |
4719 | if (ret) | |
4720 | return ret; | |
5dc562c5 | 4721 | |
70c8a91c JB |
4722 | while (!list_empty(&ordered_sums)) { |
4723 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4724 | struct btrfs_ordered_sum, | |
4725 | list); | |
4726 | if (!ret) | |
3ebac17c | 4727 | ret = log_csums(trans, inode, log_root, sums); |
70c8a91c JB |
4728 | list_del(&sums->list); |
4729 | kfree(sums); | |
5dc562c5 JB |
4730 | } |
4731 | ||
70c8a91c | 4732 | return ret; |
5dc562c5 JB |
4733 | } |
4734 | ||
8407f553 | 4735 | static int log_one_extent(struct btrfs_trans_handle *trans, |
90d04510 | 4736 | struct btrfs_inode *inode, |
8407f553 FM |
4737 | const struct extent_map *em, |
4738 | struct btrfs_path *path, | |
8407f553 FM |
4739 | struct btrfs_log_ctx *ctx) |
4740 | { | |
5893dfb9 | 4741 | struct btrfs_drop_extents_args drop_args = { 0 }; |
90d04510 | 4742 | struct btrfs_root *log = inode->root->log_root; |
e1f53ed8 | 4743 | struct btrfs_file_extent_item fi = { 0 }; |
8407f553 | 4744 | struct extent_buffer *leaf; |
8407f553 FM |
4745 | struct btrfs_key key; |
4746 | u64 extent_offset = em->start - em->orig_start; | |
4747 | u64 block_len; | |
4748 | int ret; | |
8407f553 | 4749 | |
e1f53ed8 FM |
4750 | btrfs_set_stack_file_extent_generation(&fi, trans->transid); |
4751 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4752 | btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_PREALLOC); | |
4753 | else | |
4754 | btrfs_set_stack_file_extent_type(&fi, BTRFS_FILE_EXTENT_REG); | |
4755 | ||
4756 | block_len = max(em->block_len, em->orig_block_len); | |
4757 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
4758 | btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start); | |
4759 | btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len); | |
4760 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
4761 | btrfs_set_stack_file_extent_disk_bytenr(&fi, em->block_start - | |
4762 | extent_offset); | |
4763 | btrfs_set_stack_file_extent_disk_num_bytes(&fi, block_len); | |
4764 | } | |
4765 | ||
4766 | btrfs_set_stack_file_extent_offset(&fi, extent_offset); | |
4767 | btrfs_set_stack_file_extent_num_bytes(&fi, em->len); | |
4768 | btrfs_set_stack_file_extent_ram_bytes(&fi, em->ram_bytes); | |
4769 | btrfs_set_stack_file_extent_compression(&fi, em->compress_type); | |
4770 | ||
48778179 | 4771 | ret = log_extent_csums(trans, inode, log, em, ctx); |
8407f553 FM |
4772 | if (ret) |
4773 | return ret; | |
4774 | ||
5328b2a7 FM |
4775 | /* |
4776 | * If this is the first time we are logging the inode in the current | |
4777 | * transaction, we can avoid btrfs_drop_extents(), which is expensive | |
4778 | * because it does a deletion search, which always acquires write locks | |
4779 | * for extent buffers at levels 2, 1 and 0. This not only wastes time | |
4780 | * but also adds significant contention in a log tree, since log trees | |
4781 | * are small, with a root at level 2 or 3 at most, due to their short | |
4782 | * life span. | |
4783 | */ | |
0f8ce498 | 4784 | if (ctx->logged_before) { |
5328b2a7 FM |
4785 | drop_args.path = path; |
4786 | drop_args.start = em->start; | |
4787 | drop_args.end = em->start + em->len; | |
4788 | drop_args.replace_extent = true; | |
e1f53ed8 | 4789 | drop_args.extent_item_size = sizeof(fi); |
5328b2a7 FM |
4790 | ret = btrfs_drop_extents(trans, log, inode, &drop_args); |
4791 | if (ret) | |
4792 | return ret; | |
4793 | } | |
8407f553 | 4794 | |
5893dfb9 | 4795 | if (!drop_args.extent_inserted) { |
9d122629 | 4796 | key.objectid = btrfs_ino(inode); |
8407f553 FM |
4797 | key.type = BTRFS_EXTENT_DATA_KEY; |
4798 | key.offset = em->start; | |
4799 | ||
4800 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
e1f53ed8 | 4801 | sizeof(fi)); |
8407f553 FM |
4802 | if (ret) |
4803 | return ret; | |
4804 | } | |
4805 | leaf = path->nodes[0]; | |
e1f53ed8 FM |
4806 | write_extent_buffer(leaf, &fi, |
4807 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4808 | sizeof(fi)); | |
8407f553 FM |
4809 | btrfs_mark_buffer_dirty(leaf); |
4810 | ||
4811 | btrfs_release_path(path); | |
4812 | ||
4813 | return ret; | |
4814 | } | |
4815 | ||
31d11b83 FM |
4816 | /* |
4817 | * Log all prealloc extents beyond the inode's i_size to make sure we do not | |
d9947887 | 4818 | * lose them after doing a full/fast fsync and replaying the log. We scan the |
31d11b83 FM |
4819 | * subvolume's root instead of iterating the inode's extent map tree because |
4820 | * otherwise we can log incorrect extent items based on extent map conversion. | |
4821 | * That can happen due to the fact that extent maps are merged when they | |
4822 | * are not in the extent map tree's list of modified extents. | |
4823 | */ | |
4824 | static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans, | |
4825 | struct btrfs_inode *inode, | |
4826 | struct btrfs_path *path) | |
4827 | { | |
4828 | struct btrfs_root *root = inode->root; | |
4829 | struct btrfs_key key; | |
4830 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
4831 | const u64 ino = btrfs_ino(inode); | |
4832 | struct btrfs_path *dst_path = NULL; | |
0e56315c | 4833 | bool dropped_extents = false; |
f135cea3 FM |
4834 | u64 truncate_offset = i_size; |
4835 | struct extent_buffer *leaf; | |
4836 | int slot; | |
31d11b83 FM |
4837 | int ins_nr = 0; |
4838 | int start_slot; | |
4839 | int ret; | |
4840 | ||
4841 | if (!(inode->flags & BTRFS_INODE_PREALLOC)) | |
4842 | return 0; | |
4843 | ||
4844 | key.objectid = ino; | |
4845 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4846 | key.offset = i_size; | |
4847 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4848 | if (ret < 0) | |
4849 | goto out; | |
4850 | ||
f135cea3 FM |
4851 | /* |
4852 | * We must check if there is a prealloc extent that starts before the | |
4853 | * i_size and crosses the i_size boundary. This is to ensure later we | |
4854 | * truncate down to the end of that extent and not to the i_size, as | |
4855 | * otherwise we end up losing part of the prealloc extent after a log | |
4856 | * replay and with an implicit hole if there is another prealloc extent | |
4857 | * that starts at an offset beyond i_size. | |
4858 | */ | |
4859 | ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY); | |
4860 | if (ret < 0) | |
4861 | goto out; | |
4862 | ||
4863 | if (ret == 0) { | |
4864 | struct btrfs_file_extent_item *ei; | |
4865 | ||
4866 | leaf = path->nodes[0]; | |
4867 | slot = path->slots[0]; | |
4868 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4869 | ||
4870 | if (btrfs_file_extent_type(leaf, ei) == | |
4871 | BTRFS_FILE_EXTENT_PREALLOC) { | |
4872 | u64 extent_end; | |
4873 | ||
4874 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4875 | extent_end = key.offset + | |
4876 | btrfs_file_extent_num_bytes(leaf, ei); | |
4877 | ||
4878 | if (extent_end > i_size) | |
4879 | truncate_offset = extent_end; | |
4880 | } | |
4881 | } else { | |
4882 | ret = 0; | |
4883 | } | |
4884 | ||
31d11b83 | 4885 | while (true) { |
f135cea3 FM |
4886 | leaf = path->nodes[0]; |
4887 | slot = path->slots[0]; | |
31d11b83 FM |
4888 | |
4889 | if (slot >= btrfs_header_nritems(leaf)) { | |
4890 | if (ins_nr > 0) { | |
4891 | ret = copy_items(trans, inode, dst_path, path, | |
0e56315c | 4892 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4893 | if (ret < 0) |
4894 | goto out; | |
4895 | ins_nr = 0; | |
4896 | } | |
4897 | ret = btrfs_next_leaf(root, path); | |
4898 | if (ret < 0) | |
4899 | goto out; | |
4900 | if (ret > 0) { | |
4901 | ret = 0; | |
4902 | break; | |
4903 | } | |
4904 | continue; | |
4905 | } | |
4906 | ||
4907 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4908 | if (key.objectid > ino) | |
4909 | break; | |
4910 | if (WARN_ON_ONCE(key.objectid < ino) || | |
4911 | key.type < BTRFS_EXTENT_DATA_KEY || | |
4912 | key.offset < i_size) { | |
4913 | path->slots[0]++; | |
4914 | continue; | |
4915 | } | |
0e56315c | 4916 | if (!dropped_extents) { |
31d11b83 FM |
4917 | /* |
4918 | * Avoid logging extent items logged in past fsync calls | |
4919 | * and leading to duplicate keys in the log tree. | |
4920 | */ | |
8a2b3da1 FM |
4921 | ret = truncate_inode_items(trans, root->log_root, inode, |
4922 | truncate_offset, | |
4923 | BTRFS_EXTENT_DATA_KEY); | |
31d11b83 FM |
4924 | if (ret) |
4925 | goto out; | |
0e56315c | 4926 | dropped_extents = true; |
31d11b83 FM |
4927 | } |
4928 | if (ins_nr == 0) | |
4929 | start_slot = slot; | |
4930 | ins_nr++; | |
4931 | path->slots[0]++; | |
4932 | if (!dst_path) { | |
4933 | dst_path = btrfs_alloc_path(); | |
4934 | if (!dst_path) { | |
4935 | ret = -ENOMEM; | |
4936 | goto out; | |
4937 | } | |
4938 | } | |
4939 | } | |
0bc2d3c0 | 4940 | if (ins_nr > 0) |
0e56315c | 4941 | ret = copy_items(trans, inode, dst_path, path, |
31d11b83 | 4942 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4943 | out: |
4944 | btrfs_release_path(path); | |
4945 | btrfs_free_path(dst_path); | |
4946 | return ret; | |
4947 | } | |
4948 | ||
5dc562c5 | 4949 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
9d122629 | 4950 | struct btrfs_inode *inode, |
827463c4 | 4951 | struct btrfs_path *path, |
48778179 | 4952 | struct btrfs_log_ctx *ctx) |
5dc562c5 | 4953 | { |
48778179 FM |
4954 | struct btrfs_ordered_extent *ordered; |
4955 | struct btrfs_ordered_extent *tmp; | |
5dc562c5 JB |
4956 | struct extent_map *em, *n; |
4957 | struct list_head extents; | |
9d122629 | 4958 | struct extent_map_tree *tree = &inode->extent_tree; |
5dc562c5 | 4959 | int ret = 0; |
2ab28f32 | 4960 | int num = 0; |
5dc562c5 JB |
4961 | |
4962 | INIT_LIST_HEAD(&extents); | |
4963 | ||
5dc562c5 | 4964 | write_lock(&tree->lock); |
5dc562c5 JB |
4965 | |
4966 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
4967 | list_del_init(&em->list); | |
2ab28f32 JB |
4968 | /* |
4969 | * Just an arbitrary number, this can be really CPU intensive | |
4970 | * once we start getting a lot of extents, and really once we | |
4971 | * have a bunch of extents we just want to commit since it will | |
4972 | * be faster. | |
4973 | */ | |
4974 | if (++num > 32768) { | |
4975 | list_del_init(&tree->modified_extents); | |
4976 | ret = -EFBIG; | |
4977 | goto process; | |
4978 | } | |
4979 | ||
5f96bfb7 | 4980 | if (em->generation < trans->transid) |
5dc562c5 | 4981 | continue; |
8c6c5928 | 4982 | |
31d11b83 FM |
4983 | /* We log prealloc extents beyond eof later. */ |
4984 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && | |
4985 | em->start >= i_size_read(&inode->vfs_inode)) | |
4986 | continue; | |
4987 | ||
ff44c6e3 | 4988 | /* Need a ref to keep it from getting evicted from cache */ |
490b54d6 | 4989 | refcount_inc(&em->refs); |
ff44c6e3 | 4990 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); |
5dc562c5 | 4991 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4992 | num++; |
5dc562c5 JB |
4993 | } |
4994 | ||
4995 | list_sort(NULL, &extents, extent_cmp); | |
2ab28f32 | 4996 | process: |
5dc562c5 JB |
4997 | while (!list_empty(&extents)) { |
4998 | em = list_entry(extents.next, struct extent_map, list); | |
4999 | ||
5000 | list_del_init(&em->list); | |
5001 | ||
5002 | /* | |
5003 | * If we had an error we just need to delete everybody from our | |
5004 | * private list. | |
5005 | */ | |
ff44c6e3 | 5006 | if (ret) { |
201a9038 | 5007 | clear_em_logging(tree, em); |
ff44c6e3 | 5008 | free_extent_map(em); |
5dc562c5 | 5009 | continue; |
ff44c6e3 JB |
5010 | } |
5011 | ||
5012 | write_unlock(&tree->lock); | |
5dc562c5 | 5013 | |
90d04510 | 5014 | ret = log_one_extent(trans, inode, em, path, ctx); |
ff44c6e3 | 5015 | write_lock(&tree->lock); |
201a9038 JB |
5016 | clear_em_logging(tree, em); |
5017 | free_extent_map(em); | |
5dc562c5 | 5018 | } |
ff44c6e3 JB |
5019 | WARN_ON(!list_empty(&extents)); |
5020 | write_unlock(&tree->lock); | |
5dc562c5 | 5021 | |
31d11b83 FM |
5022 | if (!ret) |
5023 | ret = btrfs_log_prealloc_extents(trans, inode, path); | |
48778179 FM |
5024 | if (ret) |
5025 | return ret; | |
31d11b83 | 5026 | |
48778179 FM |
5027 | /* |
5028 | * We have logged all extents successfully, now make sure the commit of | |
5029 | * the current transaction waits for the ordered extents to complete | |
5030 | * before it commits and wipes out the log trees, otherwise we would | |
5031 | * lose data if an ordered extents completes after the transaction | |
5032 | * commits and a power failure happens after the transaction commit. | |
5033 | */ | |
5034 | list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) { | |
5035 | list_del_init(&ordered->log_list); | |
5036 | set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags); | |
5037 | ||
5038 | if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) { | |
5039 | spin_lock_irq(&inode->ordered_tree.lock); | |
5040 | if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) { | |
5041 | set_bit(BTRFS_ORDERED_PENDING, &ordered->flags); | |
5042 | atomic_inc(&trans->transaction->pending_ordered); | |
5043 | } | |
5044 | spin_unlock_irq(&inode->ordered_tree.lock); | |
5045 | } | |
5046 | btrfs_put_ordered_extent(ordered); | |
5047 | } | |
5048 | ||
5049 | return 0; | |
5dc562c5 JB |
5050 | } |
5051 | ||
481b01c0 | 5052 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
5053 | struct btrfs_path *path, u64 *size_ret) |
5054 | { | |
5055 | struct btrfs_key key; | |
5056 | int ret; | |
5057 | ||
481b01c0 | 5058 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
5059 | key.type = BTRFS_INODE_ITEM_KEY; |
5060 | key.offset = 0; | |
5061 | ||
5062 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
5063 | if (ret < 0) { | |
5064 | return ret; | |
5065 | } else if (ret > 0) { | |
2f2ff0ee | 5066 | *size_ret = 0; |
1a4bcf47 FM |
5067 | } else { |
5068 | struct btrfs_inode_item *item; | |
5069 | ||
5070 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5071 | struct btrfs_inode_item); | |
5072 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
bf504110 FM |
5073 | /* |
5074 | * If the in-memory inode's i_size is smaller then the inode | |
5075 | * size stored in the btree, return the inode's i_size, so | |
5076 | * that we get a correct inode size after replaying the log | |
5077 | * when before a power failure we had a shrinking truncate | |
5078 | * followed by addition of a new name (rename / new hard link). | |
5079 | * Otherwise return the inode size from the btree, to avoid | |
5080 | * data loss when replaying a log due to previously doing a | |
5081 | * write that expands the inode's size and logging a new name | |
5082 | * immediately after. | |
5083 | */ | |
5084 | if (*size_ret > inode->vfs_inode.i_size) | |
5085 | *size_ret = inode->vfs_inode.i_size; | |
1a4bcf47 FM |
5086 | } |
5087 | ||
5088 | btrfs_release_path(path); | |
5089 | return 0; | |
5090 | } | |
5091 | ||
36283bf7 FM |
5092 | /* |
5093 | * At the moment we always log all xattrs. This is to figure out at log replay | |
5094 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
5095 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
5096 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
5097 | * happens, causing the log to be replayed the next time the fs is mounted, | |
5098 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
5099 | * with a journal, ext3/4, xfs, f2fs, etc). | |
5100 | */ | |
5101 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
1a93c36a | 5102 | struct btrfs_inode *inode, |
36283bf7 FM |
5103 | struct btrfs_path *path, |
5104 | struct btrfs_path *dst_path) | |
5105 | { | |
90d04510 | 5106 | struct btrfs_root *root = inode->root; |
36283bf7 FM |
5107 | int ret; |
5108 | struct btrfs_key key; | |
1a93c36a | 5109 | const u64 ino = btrfs_ino(inode); |
36283bf7 FM |
5110 | int ins_nr = 0; |
5111 | int start_slot = 0; | |
f2f121ab FM |
5112 | bool found_xattrs = false; |
5113 | ||
5114 | if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags)) | |
5115 | return 0; | |
36283bf7 FM |
5116 | |
5117 | key.objectid = ino; | |
5118 | key.type = BTRFS_XATTR_ITEM_KEY; | |
5119 | key.offset = 0; | |
5120 | ||
5121 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5122 | if (ret < 0) | |
5123 | return ret; | |
5124 | ||
5125 | while (true) { | |
5126 | int slot = path->slots[0]; | |
5127 | struct extent_buffer *leaf = path->nodes[0]; | |
5128 | int nritems = btrfs_header_nritems(leaf); | |
5129 | ||
5130 | if (slot >= nritems) { | |
5131 | if (ins_nr > 0) { | |
1a93c36a | 5132 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 5133 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
5134 | if (ret < 0) |
5135 | return ret; | |
5136 | ins_nr = 0; | |
5137 | } | |
5138 | ret = btrfs_next_leaf(root, path); | |
5139 | if (ret < 0) | |
5140 | return ret; | |
5141 | else if (ret > 0) | |
5142 | break; | |
5143 | continue; | |
5144 | } | |
5145 | ||
5146 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5147 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
5148 | break; | |
5149 | ||
5150 | if (ins_nr == 0) | |
5151 | start_slot = slot; | |
5152 | ins_nr++; | |
5153 | path->slots[0]++; | |
f2f121ab | 5154 | found_xattrs = true; |
36283bf7 FM |
5155 | cond_resched(); |
5156 | } | |
5157 | if (ins_nr > 0) { | |
1a93c36a | 5158 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 5159 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
5160 | if (ret < 0) |
5161 | return ret; | |
5162 | } | |
5163 | ||
f2f121ab FM |
5164 | if (!found_xattrs) |
5165 | set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags); | |
5166 | ||
36283bf7 FM |
5167 | return 0; |
5168 | } | |
5169 | ||
a89ca6f2 | 5170 | /* |
0e56315c FM |
5171 | * When using the NO_HOLES feature if we punched a hole that causes the |
5172 | * deletion of entire leafs or all the extent items of the first leaf (the one | |
5173 | * that contains the inode item and references) we may end up not processing | |
5174 | * any extents, because there are no leafs with a generation matching the | |
5175 | * current transaction that have extent items for our inode. So we need to find | |
5176 | * if any holes exist and then log them. We also need to log holes after any | |
5177 | * truncate operation that changes the inode's size. | |
a89ca6f2 | 5178 | */ |
0e56315c | 5179 | static int btrfs_log_holes(struct btrfs_trans_handle *trans, |
0e56315c | 5180 | struct btrfs_inode *inode, |
7af59743 | 5181 | struct btrfs_path *path) |
a89ca6f2 | 5182 | { |
90d04510 | 5183 | struct btrfs_root *root = inode->root; |
0b246afa | 5184 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 | 5185 | struct btrfs_key key; |
a0308dd7 NB |
5186 | const u64 ino = btrfs_ino(inode); |
5187 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
7af59743 | 5188 | u64 prev_extent_end = 0; |
0e56315c | 5189 | int ret; |
a89ca6f2 | 5190 | |
0e56315c | 5191 | if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0) |
a89ca6f2 FM |
5192 | return 0; |
5193 | ||
5194 | key.objectid = ino; | |
5195 | key.type = BTRFS_EXTENT_DATA_KEY; | |
7af59743 | 5196 | key.offset = 0; |
a89ca6f2 FM |
5197 | |
5198 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
a89ca6f2 FM |
5199 | if (ret < 0) |
5200 | return ret; | |
5201 | ||
0e56315c | 5202 | while (true) { |
0e56315c | 5203 | struct extent_buffer *leaf = path->nodes[0]; |
a89ca6f2 | 5204 | |
0e56315c FM |
5205 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
5206 | ret = btrfs_next_leaf(root, path); | |
5207 | if (ret < 0) | |
5208 | return ret; | |
5209 | if (ret > 0) { | |
5210 | ret = 0; | |
5211 | break; | |
5212 | } | |
5213 | leaf = path->nodes[0]; | |
5214 | } | |
5215 | ||
5216 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
5217 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
5218 | break; | |
5219 | ||
5220 | /* We have a hole, log it. */ | |
5221 | if (prev_extent_end < key.offset) { | |
7af59743 | 5222 | const u64 hole_len = key.offset - prev_extent_end; |
0e56315c FM |
5223 | |
5224 | /* | |
5225 | * Release the path to avoid deadlocks with other code | |
5226 | * paths that search the root while holding locks on | |
5227 | * leafs from the log root. | |
5228 | */ | |
5229 | btrfs_release_path(path); | |
5230 | ret = btrfs_insert_file_extent(trans, root->log_root, | |
5231 | ino, prev_extent_end, 0, | |
5232 | 0, hole_len, 0, hole_len, | |
5233 | 0, 0, 0); | |
5234 | if (ret < 0) | |
5235 | return ret; | |
5236 | ||
5237 | /* | |
5238 | * Search for the same key again in the root. Since it's | |
5239 | * an extent item and we are holding the inode lock, the | |
5240 | * key must still exist. If it doesn't just emit warning | |
5241 | * and return an error to fall back to a transaction | |
5242 | * commit. | |
5243 | */ | |
5244 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5245 | if (ret < 0) | |
5246 | return ret; | |
5247 | if (WARN_ON(ret > 0)) | |
5248 | return -ENOENT; | |
5249 | leaf = path->nodes[0]; | |
5250 | } | |
a89ca6f2 | 5251 | |
7af59743 | 5252 | prev_extent_end = btrfs_file_extent_end(path); |
0e56315c FM |
5253 | path->slots[0]++; |
5254 | cond_resched(); | |
a89ca6f2 | 5255 | } |
a89ca6f2 | 5256 | |
7af59743 | 5257 | if (prev_extent_end < i_size) { |
0e56315c | 5258 | u64 hole_len; |
a89ca6f2 | 5259 | |
0e56315c | 5260 | btrfs_release_path(path); |
7af59743 | 5261 | hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize); |
0e56315c FM |
5262 | ret = btrfs_insert_file_extent(trans, root->log_root, |
5263 | ino, prev_extent_end, 0, 0, | |
5264 | hole_len, 0, hole_len, | |
5265 | 0, 0, 0); | |
5266 | if (ret < 0) | |
5267 | return ret; | |
5268 | } | |
5269 | ||
5270 | return 0; | |
a89ca6f2 FM |
5271 | } |
5272 | ||
56f23fdb FM |
5273 | /* |
5274 | * When we are logging a new inode X, check if it doesn't have a reference that | |
5275 | * matches the reference from some other inode Y created in a past transaction | |
5276 | * and that was renamed in the current transaction. If we don't do this, then at | |
5277 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
5278 | * | |
5279 | * mkdir /mnt/x | |
5280 | * echo "hello world" > /mnt/x/foobar | |
5281 | * sync | |
5282 | * mv /mnt/x /mnt/y | |
5283 | * mkdir /mnt/x # or touch /mnt/x | |
5284 | * xfs_io -c fsync /mnt/x | |
5285 | * <power fail> | |
5286 | * mount fs, trigger log replay | |
5287 | * | |
5288 | * After the log replay procedure, we would lose the first directory and all its | |
5289 | * files (file foobar). | |
5290 | * For the case where inode Y is not a directory we simply end up losing it: | |
5291 | * | |
5292 | * echo "123" > /mnt/foo | |
5293 | * sync | |
5294 | * mv /mnt/foo /mnt/bar | |
5295 | * echo "abc" > /mnt/foo | |
5296 | * xfs_io -c fsync /mnt/foo | |
5297 | * <power fail> | |
5298 | * | |
5299 | * We also need this for cases where a snapshot entry is replaced by some other | |
5300 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
5301 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
5302 | * if it were a regular entry: | |
5303 | * | |
5304 | * mkdir /mnt/x | |
5305 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
5306 | * btrfs subvolume delete /mnt/x/snap | |
5307 | * rmdir /mnt/x | |
5308 | * mkdir /mnt/x | |
5309 | * fsync /mnt/x or fsync some new file inside it | |
5310 | * <power fail> | |
5311 | * | |
5312 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
5313 | * the same transaction. | |
5314 | */ | |
5315 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
5316 | const int slot, | |
5317 | const struct btrfs_key *key, | |
4791c8f1 | 5318 | struct btrfs_inode *inode, |
a3baaf0d | 5319 | u64 *other_ino, u64 *other_parent) |
56f23fdb FM |
5320 | { |
5321 | int ret; | |
5322 | struct btrfs_path *search_path; | |
5323 | char *name = NULL; | |
5324 | u32 name_len = 0; | |
3212fa14 | 5325 | u32 item_size = btrfs_item_size(eb, slot); |
56f23fdb FM |
5326 | u32 cur_offset = 0; |
5327 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
5328 | ||
5329 | search_path = btrfs_alloc_path(); | |
5330 | if (!search_path) | |
5331 | return -ENOMEM; | |
5332 | search_path->search_commit_root = 1; | |
5333 | search_path->skip_locking = 1; | |
5334 | ||
5335 | while (cur_offset < item_size) { | |
5336 | u64 parent; | |
5337 | u32 this_name_len; | |
5338 | u32 this_len; | |
5339 | unsigned long name_ptr; | |
5340 | struct btrfs_dir_item *di; | |
5341 | ||
5342 | if (key->type == BTRFS_INODE_REF_KEY) { | |
5343 | struct btrfs_inode_ref *iref; | |
5344 | ||
5345 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
5346 | parent = key->offset; | |
5347 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
5348 | name_ptr = (unsigned long)(iref + 1); | |
5349 | this_len = sizeof(*iref) + this_name_len; | |
5350 | } else { | |
5351 | struct btrfs_inode_extref *extref; | |
5352 | ||
5353 | extref = (struct btrfs_inode_extref *)(ptr + | |
5354 | cur_offset); | |
5355 | parent = btrfs_inode_extref_parent(eb, extref); | |
5356 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
5357 | name_ptr = (unsigned long)&extref->name; | |
5358 | this_len = sizeof(*extref) + this_name_len; | |
5359 | } | |
5360 | ||
5361 | if (this_name_len > name_len) { | |
5362 | char *new_name; | |
5363 | ||
5364 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
5365 | if (!new_name) { | |
5366 | ret = -ENOMEM; | |
5367 | goto out; | |
5368 | } | |
5369 | name_len = this_name_len; | |
5370 | name = new_name; | |
5371 | } | |
5372 | ||
5373 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4791c8f1 NB |
5374 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
5375 | parent, name, this_name_len, 0); | |
56f23fdb | 5376 | if (di && !IS_ERR(di)) { |
44f714da FM |
5377 | struct btrfs_key di_key; |
5378 | ||
5379 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
5380 | di, &di_key); | |
5381 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
6b5fc433 FM |
5382 | if (di_key.objectid != key->objectid) { |
5383 | ret = 1; | |
5384 | *other_ino = di_key.objectid; | |
a3baaf0d | 5385 | *other_parent = parent; |
6b5fc433 FM |
5386 | } else { |
5387 | ret = 0; | |
5388 | } | |
44f714da FM |
5389 | } else { |
5390 | ret = -EAGAIN; | |
5391 | } | |
56f23fdb FM |
5392 | goto out; |
5393 | } else if (IS_ERR(di)) { | |
5394 | ret = PTR_ERR(di); | |
5395 | goto out; | |
5396 | } | |
5397 | btrfs_release_path(search_path); | |
5398 | ||
5399 | cur_offset += this_len; | |
5400 | } | |
5401 | ret = 0; | |
5402 | out: | |
5403 | btrfs_free_path(search_path); | |
5404 | kfree(name); | |
5405 | return ret; | |
5406 | } | |
5407 | ||
6b5fc433 FM |
5408 | struct btrfs_ino_list { |
5409 | u64 ino; | |
a3baaf0d | 5410 | u64 parent; |
6b5fc433 FM |
5411 | struct list_head list; |
5412 | }; | |
5413 | ||
5414 | static int log_conflicting_inodes(struct btrfs_trans_handle *trans, | |
5415 | struct btrfs_root *root, | |
5416 | struct btrfs_path *path, | |
5417 | struct btrfs_log_ctx *ctx, | |
a3baaf0d | 5418 | u64 ino, u64 parent) |
6b5fc433 FM |
5419 | { |
5420 | struct btrfs_ino_list *ino_elem; | |
5421 | LIST_HEAD(inode_list); | |
5422 | int ret = 0; | |
5423 | ||
5424 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
5425 | if (!ino_elem) | |
5426 | return -ENOMEM; | |
5427 | ino_elem->ino = ino; | |
a3baaf0d | 5428 | ino_elem->parent = parent; |
6b5fc433 FM |
5429 | list_add_tail(&ino_elem->list, &inode_list); |
5430 | ||
5431 | while (!list_empty(&inode_list)) { | |
5432 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5433 | struct btrfs_key key; | |
5434 | struct inode *inode; | |
5435 | ||
5436 | ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list, | |
5437 | list); | |
5438 | ino = ino_elem->ino; | |
a3baaf0d | 5439 | parent = ino_elem->parent; |
6b5fc433 FM |
5440 | list_del(&ino_elem->list); |
5441 | kfree(ino_elem); | |
5442 | if (ret) | |
5443 | continue; | |
5444 | ||
5445 | btrfs_release_path(path); | |
5446 | ||
0202e83f | 5447 | inode = btrfs_iget(fs_info->sb, ino, root); |
6b5fc433 FM |
5448 | /* |
5449 | * If the other inode that had a conflicting dir entry was | |
a3baaf0d FM |
5450 | * deleted in the current transaction, we need to log its parent |
5451 | * directory. | |
6b5fc433 FM |
5452 | */ |
5453 | if (IS_ERR(inode)) { | |
5454 | ret = PTR_ERR(inode); | |
a3baaf0d | 5455 | if (ret == -ENOENT) { |
0202e83f | 5456 | inode = btrfs_iget(fs_info->sb, parent, root); |
a3baaf0d FM |
5457 | if (IS_ERR(inode)) { |
5458 | ret = PTR_ERR(inode); | |
5459 | } else { | |
90d04510 | 5460 | ret = btrfs_log_inode(trans, |
a3baaf0d FM |
5461 | BTRFS_I(inode), |
5462 | LOG_OTHER_INODE_ALL, | |
48778179 | 5463 | ctx); |
410f954c | 5464 | btrfs_add_delayed_iput(inode); |
a3baaf0d FM |
5465 | } |
5466 | } | |
6b5fc433 FM |
5467 | continue; |
5468 | } | |
b5e4ff9d FM |
5469 | /* |
5470 | * If the inode was already logged skip it - otherwise we can | |
5471 | * hit an infinite loop. Example: | |
5472 | * | |
5473 | * From the commit root (previous transaction) we have the | |
5474 | * following inodes: | |
5475 | * | |
5476 | * inode 257 a directory | |
5477 | * inode 258 with references "zz" and "zz_link" on inode 257 | |
5478 | * inode 259 with reference "a" on inode 257 | |
5479 | * | |
5480 | * And in the current (uncommitted) transaction we have: | |
5481 | * | |
5482 | * inode 257 a directory, unchanged | |
5483 | * inode 258 with references "a" and "a2" on inode 257 | |
5484 | * inode 259 with reference "zz_link" on inode 257 | |
5485 | * inode 261 with reference "zz" on inode 257 | |
5486 | * | |
5487 | * When logging inode 261 the following infinite loop could | |
5488 | * happen if we don't skip already logged inodes: | |
5489 | * | |
5490 | * - we detect inode 258 as a conflicting inode, with inode 261 | |
5491 | * on reference "zz", and log it; | |
5492 | * | |
5493 | * - we detect inode 259 as a conflicting inode, with inode 258 | |
5494 | * on reference "a", and log it; | |
5495 | * | |
5496 | * - we detect inode 258 as a conflicting inode, with inode 259 | |
5497 | * on reference "zz_link", and log it - again! After this we | |
5498 | * repeat the above steps forever. | |
5499 | */ | |
5500 | spin_lock(&BTRFS_I(inode)->lock); | |
5501 | /* | |
5502 | * Check the inode's logged_trans only instead of | |
5503 | * btrfs_inode_in_log(). This is because the last_log_commit of | |
1f295373 FM |
5504 | * the inode is not updated when we only log that it exists (see |
5505 | * btrfs_log_inode()). | |
b5e4ff9d FM |
5506 | */ |
5507 | if (BTRFS_I(inode)->logged_trans == trans->transid) { | |
5508 | spin_unlock(&BTRFS_I(inode)->lock); | |
5509 | btrfs_add_delayed_iput(inode); | |
5510 | continue; | |
5511 | } | |
5512 | spin_unlock(&BTRFS_I(inode)->lock); | |
6b5fc433 FM |
5513 | /* |
5514 | * We are safe logging the other inode without acquiring its | |
5515 | * lock as long as we log with the LOG_INODE_EXISTS mode. We | |
5516 | * are safe against concurrent renames of the other inode as | |
5517 | * well because during a rename we pin the log and update the | |
5518 | * log with the new name before we unpin it. | |
5519 | */ | |
90d04510 | 5520 | ret = btrfs_log_inode(trans, BTRFS_I(inode), LOG_OTHER_INODE, ctx); |
6b5fc433 | 5521 | if (ret) { |
410f954c | 5522 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
5523 | continue; |
5524 | } | |
5525 | ||
5526 | key.objectid = ino; | |
5527 | key.type = BTRFS_INODE_REF_KEY; | |
5528 | key.offset = 0; | |
5529 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5530 | if (ret < 0) { | |
410f954c | 5531 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
5532 | continue; |
5533 | } | |
5534 | ||
5535 | while (true) { | |
5536 | struct extent_buffer *leaf = path->nodes[0]; | |
5537 | int slot = path->slots[0]; | |
5538 | u64 other_ino = 0; | |
a3baaf0d | 5539 | u64 other_parent = 0; |
6b5fc433 FM |
5540 | |
5541 | if (slot >= btrfs_header_nritems(leaf)) { | |
5542 | ret = btrfs_next_leaf(root, path); | |
5543 | if (ret < 0) { | |
5544 | break; | |
5545 | } else if (ret > 0) { | |
5546 | ret = 0; | |
5547 | break; | |
5548 | } | |
5549 | continue; | |
5550 | } | |
5551 | ||
5552 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5553 | if (key.objectid != ino || | |
5554 | (key.type != BTRFS_INODE_REF_KEY && | |
5555 | key.type != BTRFS_INODE_EXTREF_KEY)) { | |
5556 | ret = 0; | |
5557 | break; | |
5558 | } | |
5559 | ||
5560 | ret = btrfs_check_ref_name_override(leaf, slot, &key, | |
a3baaf0d FM |
5561 | BTRFS_I(inode), &other_ino, |
5562 | &other_parent); | |
6b5fc433 FM |
5563 | if (ret < 0) |
5564 | break; | |
5565 | if (ret > 0) { | |
5566 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
5567 | if (!ino_elem) { | |
5568 | ret = -ENOMEM; | |
5569 | break; | |
5570 | } | |
5571 | ino_elem->ino = other_ino; | |
a3baaf0d | 5572 | ino_elem->parent = other_parent; |
6b5fc433 FM |
5573 | list_add_tail(&ino_elem->list, &inode_list); |
5574 | ret = 0; | |
5575 | } | |
5576 | path->slots[0]++; | |
5577 | } | |
410f954c | 5578 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
5579 | } |
5580 | ||
5581 | return ret; | |
5582 | } | |
5583 | ||
da447009 FM |
5584 | static int copy_inode_items_to_log(struct btrfs_trans_handle *trans, |
5585 | struct btrfs_inode *inode, | |
5586 | struct btrfs_key *min_key, | |
5587 | const struct btrfs_key *max_key, | |
5588 | struct btrfs_path *path, | |
5589 | struct btrfs_path *dst_path, | |
5590 | const u64 logged_isize, | |
5591 | const bool recursive_logging, | |
5592 | const int inode_only, | |
5593 | struct btrfs_log_ctx *ctx, | |
5594 | bool *need_log_inode_item) | |
5595 | { | |
d9947887 | 5596 | const u64 i_size = i_size_read(&inode->vfs_inode); |
da447009 FM |
5597 | struct btrfs_root *root = inode->root; |
5598 | int ins_start_slot = 0; | |
5599 | int ins_nr = 0; | |
5600 | int ret; | |
5601 | ||
5602 | while (1) { | |
5603 | ret = btrfs_search_forward(root, min_key, path, trans->transid); | |
5604 | if (ret < 0) | |
5605 | return ret; | |
5606 | if (ret > 0) { | |
5607 | ret = 0; | |
5608 | break; | |
5609 | } | |
5610 | again: | |
5611 | /* Note, ins_nr might be > 0 here, cleanup outside the loop */ | |
5612 | if (min_key->objectid != max_key->objectid) | |
5613 | break; | |
5614 | if (min_key->type > max_key->type) | |
5615 | break; | |
5616 | ||
d9947887 | 5617 | if (min_key->type == BTRFS_INODE_ITEM_KEY) { |
da447009 | 5618 | *need_log_inode_item = false; |
d9947887 FM |
5619 | } else if (min_key->type == BTRFS_EXTENT_DATA_KEY && |
5620 | min_key->offset >= i_size) { | |
5621 | /* | |
5622 | * Extents at and beyond eof are logged with | |
5623 | * btrfs_log_prealloc_extents(). | |
5624 | * Only regular files have BTRFS_EXTENT_DATA_KEY keys, | |
5625 | * and no keys greater than that, so bail out. | |
5626 | */ | |
5627 | break; | |
5628 | } else if ((min_key->type == BTRFS_INODE_REF_KEY || | |
5629 | min_key->type == BTRFS_INODE_EXTREF_KEY) && | |
5630 | inode->generation == trans->transid && | |
5631 | !recursive_logging) { | |
da447009 FM |
5632 | u64 other_ino = 0; |
5633 | u64 other_parent = 0; | |
5634 | ||
5635 | ret = btrfs_check_ref_name_override(path->nodes[0], | |
5636 | path->slots[0], min_key, inode, | |
5637 | &other_ino, &other_parent); | |
5638 | if (ret < 0) { | |
5639 | return ret; | |
289cffcb | 5640 | } else if (ret > 0 && |
da447009 FM |
5641 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { |
5642 | if (ins_nr > 0) { | |
5643 | ins_nr++; | |
5644 | } else { | |
5645 | ins_nr = 1; | |
5646 | ins_start_slot = path->slots[0]; | |
5647 | } | |
5648 | ret = copy_items(trans, inode, dst_path, path, | |
5649 | ins_start_slot, ins_nr, | |
5650 | inode_only, logged_isize); | |
5651 | if (ret < 0) | |
5652 | return ret; | |
5653 | ins_nr = 0; | |
5654 | ||
5655 | ret = log_conflicting_inodes(trans, root, path, | |
5656 | ctx, other_ino, other_parent); | |
5657 | if (ret) | |
5658 | return ret; | |
5659 | btrfs_release_path(path); | |
5660 | goto next_key; | |
5661 | } | |
d9947887 FM |
5662 | } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) { |
5663 | /* Skip xattrs, logged later with btrfs_log_all_xattrs() */ | |
da447009 FM |
5664 | if (ins_nr == 0) |
5665 | goto next_slot; | |
5666 | ret = copy_items(trans, inode, dst_path, path, | |
5667 | ins_start_slot, | |
5668 | ins_nr, inode_only, logged_isize); | |
5669 | if (ret < 0) | |
5670 | return ret; | |
5671 | ins_nr = 0; | |
5672 | goto next_slot; | |
5673 | } | |
5674 | ||
5675 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { | |
5676 | ins_nr++; | |
5677 | goto next_slot; | |
5678 | } else if (!ins_nr) { | |
5679 | ins_start_slot = path->slots[0]; | |
5680 | ins_nr = 1; | |
5681 | goto next_slot; | |
5682 | } | |
5683 | ||
5684 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, | |
5685 | ins_nr, inode_only, logged_isize); | |
5686 | if (ret < 0) | |
5687 | return ret; | |
5688 | ins_nr = 1; | |
5689 | ins_start_slot = path->slots[0]; | |
5690 | next_slot: | |
5691 | path->slots[0]++; | |
5692 | if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) { | |
5693 | btrfs_item_key_to_cpu(path->nodes[0], min_key, | |
5694 | path->slots[0]); | |
5695 | goto again; | |
5696 | } | |
5697 | if (ins_nr) { | |
5698 | ret = copy_items(trans, inode, dst_path, path, | |
5699 | ins_start_slot, ins_nr, inode_only, | |
5700 | logged_isize); | |
5701 | if (ret < 0) | |
5702 | return ret; | |
5703 | ins_nr = 0; | |
5704 | } | |
5705 | btrfs_release_path(path); | |
5706 | next_key: | |
5707 | if (min_key->offset < (u64)-1) { | |
5708 | min_key->offset++; | |
5709 | } else if (min_key->type < max_key->type) { | |
5710 | min_key->type++; | |
5711 | min_key->offset = 0; | |
5712 | } else { | |
5713 | break; | |
5714 | } | |
96acb375 FM |
5715 | |
5716 | /* | |
5717 | * We may process many leaves full of items for our inode, so | |
5718 | * avoid monopolizing a cpu for too long by rescheduling while | |
5719 | * not holding locks on any tree. | |
5720 | */ | |
5721 | cond_resched(); | |
da447009 | 5722 | } |
d9947887 | 5723 | if (ins_nr) { |
da447009 FM |
5724 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, |
5725 | ins_nr, inode_only, logged_isize); | |
d9947887 FM |
5726 | if (ret) |
5727 | return ret; | |
5728 | } | |
5729 | ||
5730 | if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) { | |
5731 | /* | |
5732 | * Release the path because otherwise we might attempt to double | |
5733 | * lock the same leaf with btrfs_log_prealloc_extents() below. | |
5734 | */ | |
5735 | btrfs_release_path(path); | |
5736 | ret = btrfs_log_prealloc_extents(trans, inode, dst_path); | |
5737 | } | |
da447009 FM |
5738 | |
5739 | return ret; | |
5740 | } | |
5741 | ||
e02119d5 CM |
5742 | /* log a single inode in the tree log. |
5743 | * At least one parent directory for this inode must exist in the tree | |
5744 | * or be logged already. | |
5745 | * | |
5746 | * Any items from this inode changed by the current transaction are copied | |
5747 | * to the log tree. An extra reference is taken on any extents in this | |
5748 | * file, allowing us to avoid a whole pile of corner cases around logging | |
5749 | * blocks that have been removed from the tree. | |
5750 | * | |
5751 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
5752 | * does. | |
5753 | * | |
5754 | * This handles both files and directories. | |
5755 | */ | |
12fcfd22 | 5756 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
90d04510 | 5757 | struct btrfs_inode *inode, |
49dae1bc | 5758 | int inode_only, |
8407f553 | 5759 | struct btrfs_log_ctx *ctx) |
e02119d5 CM |
5760 | { |
5761 | struct btrfs_path *path; | |
5762 | struct btrfs_path *dst_path; | |
5763 | struct btrfs_key min_key; | |
5764 | struct btrfs_key max_key; | |
90d04510 | 5765 | struct btrfs_root *log = inode->root->log_root; |
65faced5 | 5766 | int ret; |
5dc562c5 | 5767 | bool fast_search = false; |
a59108a7 NB |
5768 | u64 ino = btrfs_ino(inode); |
5769 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
1a4bcf47 | 5770 | u64 logged_isize = 0; |
e4545de5 | 5771 | bool need_log_inode_item = true; |
9a8fca62 | 5772 | bool xattrs_logged = false; |
a3baaf0d | 5773 | bool recursive_logging = false; |
2ac691d8 | 5774 | bool inode_item_dropped = true; |
0f8ce498 | 5775 | const bool orig_logged_before = ctx->logged_before; |
e02119d5 | 5776 | |
e02119d5 | 5777 | path = btrfs_alloc_path(); |
5df67083 TI |
5778 | if (!path) |
5779 | return -ENOMEM; | |
e02119d5 | 5780 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
5781 | if (!dst_path) { |
5782 | btrfs_free_path(path); | |
5783 | return -ENOMEM; | |
5784 | } | |
e02119d5 | 5785 | |
33345d01 | 5786 | min_key.objectid = ino; |
e02119d5 CM |
5787 | min_key.type = BTRFS_INODE_ITEM_KEY; |
5788 | min_key.offset = 0; | |
5789 | ||
33345d01 | 5790 | max_key.objectid = ino; |
12fcfd22 | 5791 | |
12fcfd22 | 5792 | |
5dc562c5 | 5793 | /* today the code can only do partial logging of directories */ |
a59108a7 | 5794 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5269b67e | 5795 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5796 | &inode->runtime_flags) && |
781feef7 | 5797 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
5798 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5799 | else | |
5800 | max_key.type = (u8)-1; | |
5801 | max_key.offset = (u64)-1; | |
5802 | ||
2c2c452b | 5803 | /* |
5aa7d1a7 FM |
5804 | * Only run delayed items if we are a directory. We want to make sure |
5805 | * all directory indexes hit the fs/subvolume tree so we can find them | |
5806 | * and figure out which index ranges have to be logged. | |
2c2c452b | 5807 | */ |
f6df27dd | 5808 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
65faced5 FM |
5809 | ret = btrfs_commit_inode_delayed_items(trans, inode); |
5810 | if (ret) | |
f6df27dd | 5811 | goto out; |
16cdcec7 MX |
5812 | } |
5813 | ||
a3baaf0d FM |
5814 | if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) { |
5815 | recursive_logging = true; | |
5816 | if (inode_only == LOG_OTHER_INODE) | |
5817 | inode_only = LOG_INODE_EXISTS; | |
5818 | else | |
5819 | inode_only = LOG_INODE_ALL; | |
a59108a7 | 5820 | mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING); |
781feef7 | 5821 | } else { |
a59108a7 | 5822 | mutex_lock(&inode->log_mutex); |
781feef7 | 5823 | } |
e02119d5 | 5824 | |
0f8ce498 FM |
5825 | /* |
5826 | * Before logging the inode item, cache the value returned by | |
5827 | * inode_logged(), because after that we have the need to figure out if | |
5828 | * the inode was previously logged in this transaction. | |
5829 | */ | |
5830 | ret = inode_logged(trans, inode, path); | |
65faced5 | 5831 | if (ret < 0) |
0f8ce498 | 5832 | goto out_unlock; |
0f8ce498 | 5833 | ctx->logged_before = (ret == 1); |
65faced5 | 5834 | ret = 0; |
0f8ce498 | 5835 | |
64d6b281 FM |
5836 | /* |
5837 | * This is for cases where logging a directory could result in losing a | |
5838 | * a file after replaying the log. For example, if we move a file from a | |
5839 | * directory A to a directory B, then fsync directory A, we have no way | |
5840 | * to known the file was moved from A to B, so logging just A would | |
5841 | * result in losing the file after a log replay. | |
5842 | */ | |
5843 | if (S_ISDIR(inode->vfs_inode.i_mode) && | |
5844 | inode_only == LOG_INODE_ALL && | |
5845 | inode->last_unlink_trans >= trans->transid) { | |
5846 | btrfs_set_log_full_commit(trans); | |
65faced5 | 5847 | ret = 1; |
64d6b281 FM |
5848 | goto out_unlock; |
5849 | } | |
5850 | ||
e02119d5 CM |
5851 | /* |
5852 | * a brute force approach to making sure we get the most uptodate | |
5853 | * copies of everything. | |
5854 | */ | |
a59108a7 | 5855 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
e02119d5 CM |
5856 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; |
5857 | ||
ab12313a | 5858 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags); |
4f764e51 FM |
5859 | if (inode_only == LOG_INODE_EXISTS) |
5860 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
0f8ce498 FM |
5861 | if (ctx->logged_before) |
5862 | ret = drop_inode_items(trans, log, path, inode, | |
5863 | max_key_type); | |
e02119d5 | 5864 | } else { |
0f8ce498 | 5865 | if (inode_only == LOG_INODE_EXISTS && ctx->logged_before) { |
1a4bcf47 FM |
5866 | /* |
5867 | * Make sure the new inode item we write to the log has | |
5868 | * the same isize as the current one (if it exists). | |
5869 | * This is necessary to prevent data loss after log | |
5870 | * replay, and also to prevent doing a wrong expanding | |
5871 | * truncate - for e.g. create file, write 4K into offset | |
5872 | * 0, fsync, write 4K into offset 4096, add hard link, | |
5873 | * fsync some other file (to sync log), power fail - if | |
5874 | * we use the inode's current i_size, after log replay | |
5875 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
5876 | * (zeroes), as if an expanding truncate happened, | |
5877 | * instead of getting a file of 4Kb only. | |
5878 | */ | |
65faced5 FM |
5879 | ret = logged_inode_size(log, inode, path, &logged_isize); |
5880 | if (ret) | |
1a4bcf47 FM |
5881 | goto out_unlock; |
5882 | } | |
a742994a | 5883 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5884 | &inode->runtime_flags)) { |
a742994a | 5885 | if (inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5886 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
0f8ce498 FM |
5887 | if (ctx->logged_before) |
5888 | ret = drop_inode_items(trans, log, path, | |
5889 | inode, max_key.type); | |
a742994a FM |
5890 | } else { |
5891 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a59108a7 | 5892 | &inode->runtime_flags); |
a742994a | 5893 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5894 | &inode->runtime_flags); |
0f8ce498 | 5895 | if (ctx->logged_before) |
4934a815 FM |
5896 | ret = truncate_inode_items(trans, log, |
5897 | inode, 0, 0); | |
a742994a | 5898 | } |
4f764e51 | 5899 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5900 | &inode->runtime_flags) || |
6cfab851 | 5901 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5902 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 5903 | fast_search = true; |
4f764e51 | 5904 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
0f8ce498 FM |
5905 | if (ctx->logged_before) |
5906 | ret = drop_inode_items(trans, log, path, inode, | |
5907 | max_key.type); | |
a95249b3 JB |
5908 | } else { |
5909 | if (inode_only == LOG_INODE_ALL) | |
5910 | fast_search = true; | |
2ac691d8 | 5911 | inode_item_dropped = false; |
a95249b3 | 5912 | goto log_extents; |
5dc562c5 | 5913 | } |
a95249b3 | 5914 | |
e02119d5 | 5915 | } |
65faced5 | 5916 | if (ret) |
4a500fd1 | 5917 | goto out_unlock; |
e02119d5 | 5918 | |
65faced5 | 5919 | ret = copy_inode_items_to_log(trans, inode, &min_key, &max_key, |
da447009 | 5920 | path, dst_path, logged_isize, |
7af59743 FM |
5921 | recursive_logging, inode_only, ctx, |
5922 | &need_log_inode_item); | |
65faced5 | 5923 | if (ret) |
da447009 | 5924 | goto out_unlock; |
5dc562c5 | 5925 | |
36283bf7 FM |
5926 | btrfs_release_path(path); |
5927 | btrfs_release_path(dst_path); | |
65faced5 FM |
5928 | ret = btrfs_log_all_xattrs(trans, inode, path, dst_path); |
5929 | if (ret) | |
36283bf7 | 5930 | goto out_unlock; |
9a8fca62 | 5931 | xattrs_logged = true; |
a89ca6f2 FM |
5932 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
5933 | btrfs_release_path(path); | |
5934 | btrfs_release_path(dst_path); | |
65faced5 FM |
5935 | ret = btrfs_log_holes(trans, inode, path); |
5936 | if (ret) | |
a89ca6f2 FM |
5937 | goto out_unlock; |
5938 | } | |
a95249b3 | 5939 | log_extents: |
f3b15ccd JB |
5940 | btrfs_release_path(path); |
5941 | btrfs_release_path(dst_path); | |
e4545de5 | 5942 | if (need_log_inode_item) { |
65faced5 FM |
5943 | ret = log_inode_item(trans, log, dst_path, inode, inode_item_dropped); |
5944 | if (ret) | |
b590b839 FM |
5945 | goto out_unlock; |
5946 | /* | |
5947 | * If we are doing a fast fsync and the inode was logged before | |
5948 | * in this transaction, we don't need to log the xattrs because | |
5949 | * they were logged before. If xattrs were added, changed or | |
5950 | * deleted since the last time we logged the inode, then we have | |
5951 | * already logged them because the inode had the runtime flag | |
5952 | * BTRFS_INODE_COPY_EVERYTHING set. | |
5953 | */ | |
5954 | if (!xattrs_logged && inode->logged_trans < trans->transid) { | |
65faced5 FM |
5955 | ret = btrfs_log_all_xattrs(trans, inode, path, dst_path); |
5956 | if (ret) | |
b590b839 | 5957 | goto out_unlock; |
9a8fca62 FM |
5958 | btrfs_release_path(path); |
5959 | } | |
e4545de5 | 5960 | } |
5dc562c5 | 5961 | if (fast_search) { |
90d04510 | 5962 | ret = btrfs_log_changed_extents(trans, inode, dst_path, ctx); |
65faced5 | 5963 | if (ret) |
5dc562c5 | 5964 | goto out_unlock; |
d006a048 | 5965 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
5966 | struct extent_map *em, *n; |
5967 | ||
49dae1bc | 5968 | write_lock(&em_tree->lock); |
48778179 FM |
5969 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, list) |
5970 | list_del_init(&em->list); | |
49dae1bc | 5971 | write_unlock(&em_tree->lock); |
5dc562c5 JB |
5972 | } |
5973 | ||
a59108a7 | 5974 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) { |
90d04510 | 5975 | ret = log_directory_changes(trans, inode, path, dst_path, ctx); |
65faced5 | 5976 | if (ret) |
4a500fd1 | 5977 | goto out_unlock; |
e02119d5 | 5978 | } |
49dae1bc | 5979 | |
130341be FM |
5980 | spin_lock(&inode->lock); |
5981 | inode->logged_trans = trans->transid; | |
d1d832a0 | 5982 | /* |
130341be FM |
5983 | * Don't update last_log_commit if we logged that an inode exists. |
5984 | * We do this for three reasons: | |
5985 | * | |
5986 | * 1) We might have had buffered writes to this inode that were | |
5987 | * flushed and had their ordered extents completed in this | |
5988 | * transaction, but we did not previously log the inode with | |
5989 | * LOG_INODE_ALL. Later the inode was evicted and after that | |
5990 | * it was loaded again and this LOG_INODE_EXISTS log operation | |
5991 | * happened. We must make sure that if an explicit fsync against | |
5992 | * the inode is performed later, it logs the new extents, an | |
5993 | * updated inode item, etc, and syncs the log. The same logic | |
5994 | * applies to direct IO writes instead of buffered writes. | |
5995 | * | |
5996 | * 2) When we log the inode with LOG_INODE_EXISTS, its inode item | |
5997 | * is logged with an i_size of 0 or whatever value was logged | |
5998 | * before. If later the i_size of the inode is increased by a | |
5999 | * truncate operation, the log is synced through an fsync of | |
6000 | * some other inode and then finally an explicit fsync against | |
6001 | * this inode is made, we must make sure this fsync logs the | |
6002 | * inode with the new i_size, the hole between old i_size and | |
6003 | * the new i_size, and syncs the log. | |
6004 | * | |
6005 | * 3) If we are logging that an ancestor inode exists as part of | |
6006 | * logging a new name from a link or rename operation, don't update | |
6007 | * its last_log_commit - otherwise if an explicit fsync is made | |
6008 | * against an ancestor, the fsync considers the inode in the log | |
6009 | * and doesn't sync the log, resulting in the ancestor missing after | |
6010 | * a power failure unless the log was synced as part of an fsync | |
6011 | * against any other unrelated inode. | |
d1d832a0 | 6012 | */ |
130341be FM |
6013 | if (inode_only != LOG_INODE_EXISTS) |
6014 | inode->last_log_commit = inode->last_sub_trans; | |
6015 | spin_unlock(&inode->lock); | |
4a500fd1 | 6016 | out_unlock: |
a59108a7 | 6017 | mutex_unlock(&inode->log_mutex); |
f6df27dd | 6018 | out: |
e02119d5 CM |
6019 | btrfs_free_path(path); |
6020 | btrfs_free_path(dst_path); | |
0f8ce498 FM |
6021 | |
6022 | if (recursive_logging) | |
6023 | ctx->logged_before = orig_logged_before; | |
6024 | ||
65faced5 | 6025 | return ret; |
e02119d5 CM |
6026 | } |
6027 | ||
ab12313a FM |
6028 | /* |
6029 | * Check if we need to log an inode. This is used in contexts where while | |
6030 | * logging an inode we need to log another inode (either that it exists or in | |
6031 | * full mode). This is used instead of btrfs_inode_in_log() because the later | |
6032 | * requires the inode to be in the log and have the log transaction committed, | |
6033 | * while here we do not care if the log transaction was already committed - our | |
6034 | * caller will commit the log later - and we want to avoid logging an inode | |
6035 | * multiple times when multiple tasks have joined the same log transaction. | |
6036 | */ | |
6037 | static bool need_log_inode(struct btrfs_trans_handle *trans, | |
6038 | struct btrfs_inode *inode) | |
6039 | { | |
8be2ba2e FM |
6040 | /* |
6041 | * If a directory was not modified, no dentries added or removed, we can | |
6042 | * and should avoid logging it. | |
6043 | */ | |
6044 | if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid) | |
6045 | return false; | |
6046 | ||
ab12313a FM |
6047 | /* |
6048 | * If this inode does not have new/updated/deleted xattrs since the last | |
6049 | * time it was logged and is flagged as logged in the current transaction, | |
6050 | * we can skip logging it. As for new/deleted names, those are updated in | |
6051 | * the log by link/unlink/rename operations. | |
6052 | * In case the inode was logged and then evicted and reloaded, its | |
6053 | * logged_trans will be 0, in which case we have to fully log it since | |
6054 | * logged_trans is a transient field, not persisted. | |
6055 | */ | |
6056 | if (inode->logged_trans == trans->transid && | |
6057 | !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags)) | |
6058 | return false; | |
6059 | ||
6060 | return true; | |
6061 | } | |
6062 | ||
2f2ff0ee FM |
6063 | struct btrfs_dir_list { |
6064 | u64 ino; | |
6065 | struct list_head list; | |
6066 | }; | |
6067 | ||
6068 | /* | |
6069 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
6070 | * details about the why it is needed. | |
6071 | * This is a recursive operation - if an existing dentry corresponds to a | |
6072 | * directory, that directory's new entries are logged too (same behaviour as | |
6073 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
6074 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
6075 | * complains about the following circular lock dependency / possible deadlock: | |
6076 | * | |
6077 | * CPU0 CPU1 | |
6078 | * ---- ---- | |
6079 | * lock(&type->i_mutex_dir_key#3/2); | |
6080 | * lock(sb_internal#2); | |
6081 | * lock(&type->i_mutex_dir_key#3/2); | |
6082 | * lock(&sb->s_type->i_mutex_key#14); | |
6083 | * | |
6084 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
6085 | * sb_start_intwrite() in btrfs_start_transaction(). | |
6086 | * Not locking i_mutex of the inodes is still safe because: | |
6087 | * | |
6088 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
6089 | * that while logging the inode new references (names) are added or removed | |
6090 | * from the inode, leaving the logged inode item with a link count that does | |
6091 | * not match the number of logged inode reference items. This is fine because | |
6092 | * at log replay time we compute the real number of links and correct the | |
6093 | * link count in the inode item (see replay_one_buffer() and | |
6094 | * link_to_fixup_dir()); | |
6095 | * | |
6096 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
339d0354 FM |
6097 | * while logging the inode's items new index items (key type |
6098 | * BTRFS_DIR_INDEX_KEY) are added to fs/subvol tree and the logged inode item | |
2f2ff0ee | 6099 | * has a size that doesn't match the sum of the lengths of all the logged |
339d0354 FM |
6100 | * names - this is ok, not a problem, because at log replay time we set the |
6101 | * directory's i_size to the correct value (see replay_one_name() and | |
6102 | * do_overwrite_item()). | |
2f2ff0ee FM |
6103 | */ |
6104 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
6105 | struct btrfs_root *root, | |
51cc0d32 | 6106 | struct btrfs_inode *start_inode, |
2f2ff0ee FM |
6107 | struct btrfs_log_ctx *ctx) |
6108 | { | |
0b246afa | 6109 | struct btrfs_fs_info *fs_info = root->fs_info; |
2f2ff0ee FM |
6110 | struct btrfs_path *path; |
6111 | LIST_HEAD(dir_list); | |
6112 | struct btrfs_dir_list *dir_elem; | |
6113 | int ret = 0; | |
6114 | ||
c48792c6 FM |
6115 | /* |
6116 | * If we are logging a new name, as part of a link or rename operation, | |
6117 | * don't bother logging new dentries, as we just want to log the names | |
6118 | * of an inode and that any new parents exist. | |
6119 | */ | |
6120 | if (ctx->logging_new_name) | |
6121 | return 0; | |
6122 | ||
2f2ff0ee FM |
6123 | path = btrfs_alloc_path(); |
6124 | if (!path) | |
6125 | return -ENOMEM; | |
6126 | ||
6127 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
6128 | if (!dir_elem) { | |
6129 | btrfs_free_path(path); | |
6130 | return -ENOMEM; | |
6131 | } | |
51cc0d32 | 6132 | dir_elem->ino = btrfs_ino(start_inode); |
2f2ff0ee FM |
6133 | list_add_tail(&dir_elem->list, &dir_list); |
6134 | ||
6135 | while (!list_empty(&dir_list)) { | |
6136 | struct extent_buffer *leaf; | |
6137 | struct btrfs_key min_key; | |
6138 | int nritems; | |
6139 | int i; | |
6140 | ||
6141 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
6142 | list); | |
6143 | if (ret) | |
6144 | goto next_dir_inode; | |
6145 | ||
6146 | min_key.objectid = dir_elem->ino; | |
339d0354 | 6147 | min_key.type = BTRFS_DIR_INDEX_KEY; |
2f2ff0ee FM |
6148 | min_key.offset = 0; |
6149 | again: | |
6150 | btrfs_release_path(path); | |
732d591a | 6151 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
2f2ff0ee FM |
6152 | if (ret < 0) { |
6153 | goto next_dir_inode; | |
6154 | } else if (ret > 0) { | |
6155 | ret = 0; | |
6156 | goto next_dir_inode; | |
6157 | } | |
6158 | ||
2f2ff0ee FM |
6159 | leaf = path->nodes[0]; |
6160 | nritems = btrfs_header_nritems(leaf); | |
6161 | for (i = path->slots[0]; i < nritems; i++) { | |
6162 | struct btrfs_dir_item *di; | |
6163 | struct btrfs_key di_key; | |
6164 | struct inode *di_inode; | |
6165 | struct btrfs_dir_list *new_dir_elem; | |
6166 | int log_mode = LOG_INODE_EXISTS; | |
6167 | int type; | |
6168 | ||
6169 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
6170 | if (min_key.objectid != dir_elem->ino || | |
339d0354 | 6171 | min_key.type != BTRFS_DIR_INDEX_KEY) |
2f2ff0ee FM |
6172 | goto next_dir_inode; |
6173 | ||
6174 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
6175 | type = btrfs_dir_type(leaf, di); | |
de6bc7f5 | 6176 | if (btrfs_dir_transid(leaf, di) < trans->transid) |
2f2ff0ee FM |
6177 | continue; |
6178 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
6179 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
6180 | continue; | |
6181 | ||
ec125cfb | 6182 | btrfs_release_path(path); |
0202e83f | 6183 | di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root); |
2f2ff0ee FM |
6184 | if (IS_ERR(di_inode)) { |
6185 | ret = PTR_ERR(di_inode); | |
6186 | goto next_dir_inode; | |
6187 | } | |
6188 | ||
0e44cb3f | 6189 | if (!need_log_inode(trans, BTRFS_I(di_inode))) { |
410f954c | 6190 | btrfs_add_delayed_iput(di_inode); |
ec125cfb | 6191 | break; |
2f2ff0ee FM |
6192 | } |
6193 | ||
6194 | ctx->log_new_dentries = false; | |
3f9749f6 | 6195 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) |
2f2ff0ee | 6196 | log_mode = LOG_INODE_ALL; |
90d04510 | 6197 | ret = btrfs_log_inode(trans, BTRFS_I(di_inode), |
48778179 | 6198 | log_mode, ctx); |
410f954c | 6199 | btrfs_add_delayed_iput(di_inode); |
2f2ff0ee FM |
6200 | if (ret) |
6201 | goto next_dir_inode; | |
6202 | if (ctx->log_new_dentries) { | |
6203 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
6204 | GFP_NOFS); | |
6205 | if (!new_dir_elem) { | |
6206 | ret = -ENOMEM; | |
6207 | goto next_dir_inode; | |
6208 | } | |
6209 | new_dir_elem->ino = di_key.objectid; | |
6210 | list_add_tail(&new_dir_elem->list, &dir_list); | |
6211 | } | |
6212 | break; | |
6213 | } | |
2f2ff0ee FM |
6214 | if (min_key.offset < (u64)-1) { |
6215 | min_key.offset++; | |
6216 | goto again; | |
6217 | } | |
6218 | next_dir_inode: | |
6219 | list_del(&dir_elem->list); | |
6220 | kfree(dir_elem); | |
6221 | } | |
6222 | ||
6223 | btrfs_free_path(path); | |
6224 | return ret; | |
6225 | } | |
6226 | ||
18aa0922 | 6227 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
d0a0b78d | 6228 | struct btrfs_inode *inode, |
18aa0922 FM |
6229 | struct btrfs_log_ctx *ctx) |
6230 | { | |
3ffbd68c | 6231 | struct btrfs_fs_info *fs_info = trans->fs_info; |
18aa0922 FM |
6232 | int ret; |
6233 | struct btrfs_path *path; | |
6234 | struct btrfs_key key; | |
d0a0b78d NB |
6235 | struct btrfs_root *root = inode->root; |
6236 | const u64 ino = btrfs_ino(inode); | |
18aa0922 FM |
6237 | |
6238 | path = btrfs_alloc_path(); | |
6239 | if (!path) | |
6240 | return -ENOMEM; | |
6241 | path->skip_locking = 1; | |
6242 | path->search_commit_root = 1; | |
6243 | ||
6244 | key.objectid = ino; | |
6245 | key.type = BTRFS_INODE_REF_KEY; | |
6246 | key.offset = 0; | |
6247 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6248 | if (ret < 0) | |
6249 | goto out; | |
6250 | ||
6251 | while (true) { | |
6252 | struct extent_buffer *leaf = path->nodes[0]; | |
6253 | int slot = path->slots[0]; | |
6254 | u32 cur_offset = 0; | |
6255 | u32 item_size; | |
6256 | unsigned long ptr; | |
6257 | ||
6258 | if (slot >= btrfs_header_nritems(leaf)) { | |
6259 | ret = btrfs_next_leaf(root, path); | |
6260 | if (ret < 0) | |
6261 | goto out; | |
6262 | else if (ret > 0) | |
6263 | break; | |
6264 | continue; | |
6265 | } | |
6266 | ||
6267 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
6268 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
6269 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
6270 | break; | |
6271 | ||
3212fa14 | 6272 | item_size = btrfs_item_size(leaf, slot); |
18aa0922 FM |
6273 | ptr = btrfs_item_ptr_offset(leaf, slot); |
6274 | while (cur_offset < item_size) { | |
6275 | struct btrfs_key inode_key; | |
6276 | struct inode *dir_inode; | |
6277 | ||
6278 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
6279 | inode_key.offset = 0; | |
6280 | ||
6281 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
6282 | struct btrfs_inode_extref *extref; | |
6283 | ||
6284 | extref = (struct btrfs_inode_extref *) | |
6285 | (ptr + cur_offset); | |
6286 | inode_key.objectid = btrfs_inode_extref_parent( | |
6287 | leaf, extref); | |
6288 | cur_offset += sizeof(*extref); | |
6289 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
6290 | extref); | |
6291 | } else { | |
6292 | inode_key.objectid = key.offset; | |
6293 | cur_offset = item_size; | |
6294 | } | |
6295 | ||
0202e83f DS |
6296 | dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid, |
6297 | root); | |
0f375eed FM |
6298 | /* |
6299 | * If the parent inode was deleted, return an error to | |
6300 | * fallback to a transaction commit. This is to prevent | |
6301 | * getting an inode that was moved from one parent A to | |
6302 | * a parent B, got its former parent A deleted and then | |
6303 | * it got fsync'ed, from existing at both parents after | |
6304 | * a log replay (and the old parent still existing). | |
6305 | * Example: | |
6306 | * | |
6307 | * mkdir /mnt/A | |
6308 | * mkdir /mnt/B | |
6309 | * touch /mnt/B/bar | |
6310 | * sync | |
6311 | * mv /mnt/B/bar /mnt/A/bar | |
6312 | * mv -T /mnt/A /mnt/B | |
6313 | * fsync /mnt/B/bar | |
6314 | * <power fail> | |
6315 | * | |
6316 | * If we ignore the old parent B which got deleted, | |
6317 | * after a log replay we would have file bar linked | |
6318 | * at both parents and the old parent B would still | |
6319 | * exist. | |
6320 | */ | |
6321 | if (IS_ERR(dir_inode)) { | |
6322 | ret = PTR_ERR(dir_inode); | |
6323 | goto out; | |
6324 | } | |
18aa0922 | 6325 | |
3e6a86a1 FM |
6326 | if (!need_log_inode(trans, BTRFS_I(dir_inode))) { |
6327 | btrfs_add_delayed_iput(dir_inode); | |
6328 | continue; | |
6329 | } | |
6330 | ||
289cffcb | 6331 | ctx->log_new_dentries = false; |
90d04510 | 6332 | ret = btrfs_log_inode(trans, BTRFS_I(dir_inode), |
48778179 | 6333 | LOG_INODE_ALL, ctx); |
289cffcb | 6334 | if (!ret && ctx->log_new_dentries) |
657ed1aa | 6335 | ret = log_new_dir_dentries(trans, root, |
f85b7379 | 6336 | BTRFS_I(dir_inode), ctx); |
410f954c | 6337 | btrfs_add_delayed_iput(dir_inode); |
18aa0922 FM |
6338 | if (ret) |
6339 | goto out; | |
6340 | } | |
6341 | path->slots[0]++; | |
6342 | } | |
6343 | ret = 0; | |
6344 | out: | |
6345 | btrfs_free_path(path); | |
6346 | return ret; | |
6347 | } | |
6348 | ||
b8aa330d FM |
6349 | static int log_new_ancestors(struct btrfs_trans_handle *trans, |
6350 | struct btrfs_root *root, | |
6351 | struct btrfs_path *path, | |
6352 | struct btrfs_log_ctx *ctx) | |
6353 | { | |
6354 | struct btrfs_key found_key; | |
6355 | ||
6356 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
6357 | ||
6358 | while (true) { | |
6359 | struct btrfs_fs_info *fs_info = root->fs_info; | |
b8aa330d FM |
6360 | struct extent_buffer *leaf = path->nodes[0]; |
6361 | int slot = path->slots[0]; | |
6362 | struct btrfs_key search_key; | |
6363 | struct inode *inode; | |
0202e83f | 6364 | u64 ino; |
b8aa330d FM |
6365 | int ret = 0; |
6366 | ||
6367 | btrfs_release_path(path); | |
6368 | ||
0202e83f DS |
6369 | ino = found_key.offset; |
6370 | ||
b8aa330d FM |
6371 | search_key.objectid = found_key.offset; |
6372 | search_key.type = BTRFS_INODE_ITEM_KEY; | |
6373 | search_key.offset = 0; | |
0202e83f | 6374 | inode = btrfs_iget(fs_info->sb, ino, root); |
b8aa330d FM |
6375 | if (IS_ERR(inode)) |
6376 | return PTR_ERR(inode); | |
6377 | ||
ab12313a FM |
6378 | if (BTRFS_I(inode)->generation >= trans->transid && |
6379 | need_log_inode(trans, BTRFS_I(inode))) | |
90d04510 | 6380 | ret = btrfs_log_inode(trans, BTRFS_I(inode), |
48778179 | 6381 | LOG_INODE_EXISTS, ctx); |
410f954c | 6382 | btrfs_add_delayed_iput(inode); |
b8aa330d FM |
6383 | if (ret) |
6384 | return ret; | |
6385 | ||
6386 | if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID) | |
6387 | break; | |
6388 | ||
6389 | search_key.type = BTRFS_INODE_REF_KEY; | |
6390 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
6391 | if (ret < 0) | |
6392 | return ret; | |
6393 | ||
6394 | leaf = path->nodes[0]; | |
6395 | slot = path->slots[0]; | |
6396 | if (slot >= btrfs_header_nritems(leaf)) { | |
6397 | ret = btrfs_next_leaf(root, path); | |
6398 | if (ret < 0) | |
6399 | return ret; | |
6400 | else if (ret > 0) | |
6401 | return -ENOENT; | |
6402 | leaf = path->nodes[0]; | |
6403 | slot = path->slots[0]; | |
6404 | } | |
6405 | ||
6406 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
6407 | if (found_key.objectid != search_key.objectid || | |
6408 | found_key.type != BTRFS_INODE_REF_KEY) | |
6409 | return -ENOENT; | |
6410 | } | |
6411 | return 0; | |
6412 | } | |
6413 | ||
6414 | static int log_new_ancestors_fast(struct btrfs_trans_handle *trans, | |
6415 | struct btrfs_inode *inode, | |
6416 | struct dentry *parent, | |
6417 | struct btrfs_log_ctx *ctx) | |
6418 | { | |
6419 | struct btrfs_root *root = inode->root; | |
b8aa330d FM |
6420 | struct dentry *old_parent = NULL; |
6421 | struct super_block *sb = inode->vfs_inode.i_sb; | |
6422 | int ret = 0; | |
6423 | ||
6424 | while (true) { | |
6425 | if (!parent || d_really_is_negative(parent) || | |
6426 | sb != parent->d_sb) | |
6427 | break; | |
6428 | ||
6429 | inode = BTRFS_I(d_inode(parent)); | |
6430 | if (root != inode->root) | |
6431 | break; | |
6432 | ||
ab12313a FM |
6433 | if (inode->generation >= trans->transid && |
6434 | need_log_inode(trans, inode)) { | |
90d04510 | 6435 | ret = btrfs_log_inode(trans, inode, |
48778179 | 6436 | LOG_INODE_EXISTS, ctx); |
b8aa330d FM |
6437 | if (ret) |
6438 | break; | |
6439 | } | |
6440 | if (IS_ROOT(parent)) | |
6441 | break; | |
6442 | ||
6443 | parent = dget_parent(parent); | |
6444 | dput(old_parent); | |
6445 | old_parent = parent; | |
6446 | } | |
6447 | dput(old_parent); | |
6448 | ||
6449 | return ret; | |
6450 | } | |
6451 | ||
6452 | static int log_all_new_ancestors(struct btrfs_trans_handle *trans, | |
6453 | struct btrfs_inode *inode, | |
6454 | struct dentry *parent, | |
6455 | struct btrfs_log_ctx *ctx) | |
6456 | { | |
6457 | struct btrfs_root *root = inode->root; | |
6458 | const u64 ino = btrfs_ino(inode); | |
6459 | struct btrfs_path *path; | |
6460 | struct btrfs_key search_key; | |
6461 | int ret; | |
6462 | ||
6463 | /* | |
6464 | * For a single hard link case, go through a fast path that does not | |
6465 | * need to iterate the fs/subvolume tree. | |
6466 | */ | |
6467 | if (inode->vfs_inode.i_nlink < 2) | |
6468 | return log_new_ancestors_fast(trans, inode, parent, ctx); | |
6469 | ||
6470 | path = btrfs_alloc_path(); | |
6471 | if (!path) | |
6472 | return -ENOMEM; | |
6473 | ||
6474 | search_key.objectid = ino; | |
6475 | search_key.type = BTRFS_INODE_REF_KEY; | |
6476 | search_key.offset = 0; | |
6477 | again: | |
6478 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
6479 | if (ret < 0) | |
6480 | goto out; | |
6481 | if (ret == 0) | |
6482 | path->slots[0]++; | |
6483 | ||
6484 | while (true) { | |
6485 | struct extent_buffer *leaf = path->nodes[0]; | |
6486 | int slot = path->slots[0]; | |
6487 | struct btrfs_key found_key; | |
6488 | ||
6489 | if (slot >= btrfs_header_nritems(leaf)) { | |
6490 | ret = btrfs_next_leaf(root, path); | |
6491 | if (ret < 0) | |
6492 | goto out; | |
6493 | else if (ret > 0) | |
6494 | break; | |
6495 | continue; | |
6496 | } | |
6497 | ||
6498 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
6499 | if (found_key.objectid != ino || | |
6500 | found_key.type > BTRFS_INODE_EXTREF_KEY) | |
6501 | break; | |
6502 | ||
6503 | /* | |
6504 | * Don't deal with extended references because they are rare | |
6505 | * cases and too complex to deal with (we would need to keep | |
6506 | * track of which subitem we are processing for each item in | |
6507 | * this loop, etc). So just return some error to fallback to | |
6508 | * a transaction commit. | |
6509 | */ | |
6510 | if (found_key.type == BTRFS_INODE_EXTREF_KEY) { | |
6511 | ret = -EMLINK; | |
6512 | goto out; | |
6513 | } | |
6514 | ||
6515 | /* | |
6516 | * Logging ancestors needs to do more searches on the fs/subvol | |
6517 | * tree, so it releases the path as needed to avoid deadlocks. | |
6518 | * Keep track of the last inode ref key and resume from that key | |
6519 | * after logging all new ancestors for the current hard link. | |
6520 | */ | |
6521 | memcpy(&search_key, &found_key, sizeof(search_key)); | |
6522 | ||
6523 | ret = log_new_ancestors(trans, root, path, ctx); | |
6524 | if (ret) | |
6525 | goto out; | |
6526 | btrfs_release_path(path); | |
6527 | goto again; | |
6528 | } | |
6529 | ret = 0; | |
6530 | out: | |
6531 | btrfs_free_path(path); | |
6532 | return ret; | |
6533 | } | |
6534 | ||
e02119d5 CM |
6535 | /* |
6536 | * helper function around btrfs_log_inode to make sure newly created | |
6537 | * parent directories also end up in the log. A minimal inode and backref | |
6538 | * only logging is done of any parent directories that are older than | |
6539 | * the last committed transaction | |
6540 | */ | |
48a3b636 | 6541 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
19df27a9 | 6542 | struct btrfs_inode *inode, |
49dae1bc | 6543 | struct dentry *parent, |
41a1eada | 6544 | int inode_only, |
8b050d35 | 6545 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6546 | { |
f882274b | 6547 | struct btrfs_root *root = inode->root; |
0b246afa | 6548 | struct btrfs_fs_info *fs_info = root->fs_info; |
12fcfd22 | 6549 | int ret = 0; |
2f2ff0ee | 6550 | bool log_dentries = false; |
12fcfd22 | 6551 | |
0b246afa | 6552 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
3a5e1404 SW |
6553 | ret = 1; |
6554 | goto end_no_trans; | |
6555 | } | |
6556 | ||
f882274b | 6557 | if (btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
6558 | ret = 1; |
6559 | goto end_no_trans; | |
6560 | } | |
6561 | ||
f2d72f42 FM |
6562 | /* |
6563 | * Skip already logged inodes or inodes corresponding to tmpfiles | |
6564 | * (since logging them is pointless, a link count of 0 means they | |
6565 | * will never be accessible). | |
6566 | */ | |
626e9f41 FM |
6567 | if ((btrfs_inode_in_log(inode, trans->transid) && |
6568 | list_empty(&ctx->ordered_extents)) || | |
f2d72f42 | 6569 | inode->vfs_inode.i_nlink == 0) { |
257c62e1 CM |
6570 | ret = BTRFS_NO_LOG_SYNC; |
6571 | goto end_no_trans; | |
6572 | } | |
6573 | ||
8b050d35 | 6574 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 6575 | if (ret) |
e87ac136 | 6576 | goto end_no_trans; |
e02119d5 | 6577 | |
90d04510 | 6578 | ret = btrfs_log_inode(trans, inode, inode_only, ctx); |
4a500fd1 YZ |
6579 | if (ret) |
6580 | goto end_trans; | |
12fcfd22 | 6581 | |
af4176b4 CM |
6582 | /* |
6583 | * for regular files, if its inode is already on disk, we don't | |
6584 | * have to worry about the parents at all. This is because | |
6585 | * we can use the last_unlink_trans field to record renames | |
6586 | * and other fun in this file. | |
6587 | */ | |
19df27a9 | 6588 | if (S_ISREG(inode->vfs_inode.i_mode) && |
47d3db41 FM |
6589 | inode->generation < trans->transid && |
6590 | inode->last_unlink_trans < trans->transid) { | |
4a500fd1 YZ |
6591 | ret = 0; |
6592 | goto end_trans; | |
6593 | } | |
af4176b4 | 6594 | |
289cffcb | 6595 | if (S_ISDIR(inode->vfs_inode.i_mode) && ctx->log_new_dentries) |
2f2ff0ee FM |
6596 | log_dentries = true; |
6597 | ||
18aa0922 | 6598 | /* |
01327610 | 6599 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
6600 | * inodes are fully logged. This is to prevent leaving dangling |
6601 | * directory index entries in directories that were our parents but are | |
6602 | * not anymore. Not doing this results in old parent directory being | |
6603 | * impossible to delete after log replay (rmdir will always fail with | |
6604 | * error -ENOTEMPTY). | |
6605 | * | |
6606 | * Example 1: | |
6607 | * | |
6608 | * mkdir testdir | |
6609 | * touch testdir/foo | |
6610 | * ln testdir/foo testdir/bar | |
6611 | * sync | |
6612 | * unlink testdir/bar | |
6613 | * xfs_io -c fsync testdir/foo | |
6614 | * <power failure> | |
6615 | * mount fs, triggers log replay | |
6616 | * | |
6617 | * If we don't log the parent directory (testdir), after log replay the | |
6618 | * directory still has an entry pointing to the file inode using the bar | |
6619 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
6620 | * the file inode has a link count of 1. | |
6621 | * | |
6622 | * Example 2: | |
6623 | * | |
6624 | * mkdir testdir | |
6625 | * touch foo | |
6626 | * ln foo testdir/foo2 | |
6627 | * ln foo testdir/foo3 | |
6628 | * sync | |
6629 | * unlink testdir/foo3 | |
6630 | * xfs_io -c fsync foo | |
6631 | * <power failure> | |
6632 | * mount fs, triggers log replay | |
6633 | * | |
6634 | * Similar as the first example, after log replay the parent directory | |
6635 | * testdir still has an entry pointing to the inode file with name foo3 | |
6636 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
6637 | * and has a link count of 2. | |
6638 | */ | |
47d3db41 | 6639 | if (inode->last_unlink_trans >= trans->transid) { |
b8aa330d | 6640 | ret = btrfs_log_all_parents(trans, inode, ctx); |
18aa0922 FM |
6641 | if (ret) |
6642 | goto end_trans; | |
6643 | } | |
6644 | ||
b8aa330d FM |
6645 | ret = log_all_new_ancestors(trans, inode, parent, ctx); |
6646 | if (ret) | |
41bd6067 | 6647 | goto end_trans; |
76dda93c | 6648 | |
2f2ff0ee | 6649 | if (log_dentries) |
b8aa330d | 6650 | ret = log_new_dir_dentries(trans, root, inode, ctx); |
2f2ff0ee FM |
6651 | else |
6652 | ret = 0; | |
4a500fd1 YZ |
6653 | end_trans: |
6654 | if (ret < 0) { | |
90787766 | 6655 | btrfs_set_log_full_commit(trans); |
4a500fd1 YZ |
6656 | ret = 1; |
6657 | } | |
8b050d35 MX |
6658 | |
6659 | if (ret) | |
6660 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
6661 | btrfs_end_log_trans(root); |
6662 | end_no_trans: | |
6663 | return ret; | |
e02119d5 CM |
6664 | } |
6665 | ||
6666 | /* | |
6667 | * it is not safe to log dentry if the chunk root has added new | |
6668 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
6669 | * If this returns 1, you must commit the transaction to safely get your | |
6670 | * data on disk. | |
6671 | */ | |
6672 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
e5b84f7a | 6673 | struct dentry *dentry, |
8b050d35 | 6674 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6675 | { |
6a912213 JB |
6676 | struct dentry *parent = dget_parent(dentry); |
6677 | int ret; | |
6678 | ||
f882274b | 6679 | ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent, |
48778179 | 6680 | LOG_INODE_ALL, ctx); |
6a912213 JB |
6681 | dput(parent); |
6682 | ||
6683 | return ret; | |
e02119d5 CM |
6684 | } |
6685 | ||
6686 | /* | |
6687 | * should be called during mount to recover any replay any log trees | |
6688 | * from the FS | |
6689 | */ | |
6690 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
6691 | { | |
6692 | int ret; | |
6693 | struct btrfs_path *path; | |
6694 | struct btrfs_trans_handle *trans; | |
6695 | struct btrfs_key key; | |
6696 | struct btrfs_key found_key; | |
e02119d5 CM |
6697 | struct btrfs_root *log; |
6698 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
6699 | struct walk_control wc = { | |
6700 | .process_func = process_one_buffer, | |
430a6626 | 6701 | .stage = LOG_WALK_PIN_ONLY, |
e02119d5 CM |
6702 | }; |
6703 | ||
e02119d5 | 6704 | path = btrfs_alloc_path(); |
db5b493a TI |
6705 | if (!path) |
6706 | return -ENOMEM; | |
6707 | ||
afcdd129 | 6708 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 6709 | |
4a500fd1 | 6710 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
6711 | if (IS_ERR(trans)) { |
6712 | ret = PTR_ERR(trans); | |
6713 | goto error; | |
6714 | } | |
e02119d5 CM |
6715 | |
6716 | wc.trans = trans; | |
6717 | wc.pin = 1; | |
6718 | ||
db5b493a | 6719 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 6720 | if (ret) { |
ba51e2a1 | 6721 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
6722 | goto error; |
6723 | } | |
e02119d5 CM |
6724 | |
6725 | again: | |
6726 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
6727 | key.offset = (u64)-1; | |
962a298f | 6728 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 6729 | |
d397712b | 6730 | while (1) { |
e02119d5 | 6731 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
6732 | |
6733 | if (ret < 0) { | |
ba51e2a1 | 6734 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
6735 | goto error; |
6736 | } | |
e02119d5 CM |
6737 | if (ret > 0) { |
6738 | if (path->slots[0] == 0) | |
6739 | break; | |
6740 | path->slots[0]--; | |
6741 | } | |
6742 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
6743 | path->slots[0]); | |
b3b4aa74 | 6744 | btrfs_release_path(path); |
e02119d5 CM |
6745 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
6746 | break; | |
6747 | ||
62a2c73e | 6748 | log = btrfs_read_tree_root(log_root_tree, &found_key); |
79787eaa JM |
6749 | if (IS_ERR(log)) { |
6750 | ret = PTR_ERR(log); | |
ba51e2a1 | 6751 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
6752 | goto error; |
6753 | } | |
e02119d5 | 6754 | |
56e9357a DS |
6755 | wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset, |
6756 | true); | |
79787eaa JM |
6757 | if (IS_ERR(wc.replay_dest)) { |
6758 | ret = PTR_ERR(wc.replay_dest); | |
9bc574de JB |
6759 | |
6760 | /* | |
6761 | * We didn't find the subvol, likely because it was | |
6762 | * deleted. This is ok, simply skip this log and go to | |
6763 | * the next one. | |
6764 | * | |
6765 | * We need to exclude the root because we can't have | |
6766 | * other log replays overwriting this log as we'll read | |
6767 | * it back in a few more times. This will keep our | |
6768 | * block from being modified, and we'll just bail for | |
6769 | * each subsequent pass. | |
6770 | */ | |
6771 | if (ret == -ENOENT) | |
9fce5704 | 6772 | ret = btrfs_pin_extent_for_log_replay(trans, |
9bc574de JB |
6773 | log->node->start, |
6774 | log->node->len); | |
00246528 | 6775 | btrfs_put_root(log); |
9bc574de JB |
6776 | |
6777 | if (!ret) | |
6778 | goto next; | |
ba51e2a1 | 6779 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
6780 | goto error; |
6781 | } | |
e02119d5 | 6782 | |
07d400a6 | 6783 | wc.replay_dest->log_root = log; |
2002ae11 JB |
6784 | ret = btrfs_record_root_in_trans(trans, wc.replay_dest); |
6785 | if (ret) | |
6786 | /* The loop needs to continue due to the root refs */ | |
ba51e2a1 | 6787 | btrfs_abort_transaction(trans, ret); |
2002ae11 JB |
6788 | else |
6789 | ret = walk_log_tree(trans, log, &wc); | |
e02119d5 | 6790 | |
b50c6e25 | 6791 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
6792 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
6793 | path); | |
ba51e2a1 JB |
6794 | if (ret) |
6795 | btrfs_abort_transaction(trans, ret); | |
e02119d5 CM |
6796 | } |
6797 | ||
900c9981 LB |
6798 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
6799 | struct btrfs_root *root = wc.replay_dest; | |
6800 | ||
6801 | btrfs_release_path(path); | |
6802 | ||
6803 | /* | |
6804 | * We have just replayed everything, and the highest | |
6805 | * objectid of fs roots probably has changed in case | |
6806 | * some inode_item's got replayed. | |
6807 | * | |
6808 | * root->objectid_mutex is not acquired as log replay | |
6809 | * could only happen during mount. | |
6810 | */ | |
453e4873 | 6811 | ret = btrfs_init_root_free_objectid(root); |
ba51e2a1 JB |
6812 | if (ret) |
6813 | btrfs_abort_transaction(trans, ret); | |
900c9981 LB |
6814 | } |
6815 | ||
07d400a6 | 6816 | wc.replay_dest->log_root = NULL; |
00246528 | 6817 | btrfs_put_root(wc.replay_dest); |
00246528 | 6818 | btrfs_put_root(log); |
e02119d5 | 6819 | |
b50c6e25 JB |
6820 | if (ret) |
6821 | goto error; | |
9bc574de | 6822 | next: |
e02119d5 CM |
6823 | if (found_key.offset == 0) |
6824 | break; | |
9bc574de | 6825 | key.offset = found_key.offset - 1; |
e02119d5 | 6826 | } |
b3b4aa74 | 6827 | btrfs_release_path(path); |
e02119d5 CM |
6828 | |
6829 | /* step one is to pin it all, step two is to replay just inodes */ | |
6830 | if (wc.pin) { | |
6831 | wc.pin = 0; | |
6832 | wc.process_func = replay_one_buffer; | |
6833 | wc.stage = LOG_WALK_REPLAY_INODES; | |
6834 | goto again; | |
6835 | } | |
6836 | /* step three is to replay everything */ | |
6837 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
6838 | wc.stage++; | |
6839 | goto again; | |
6840 | } | |
6841 | ||
6842 | btrfs_free_path(path); | |
6843 | ||
abefa55a | 6844 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 6845 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
6846 | if (ret) |
6847 | return ret; | |
6848 | ||
e02119d5 | 6849 | log_root_tree->log_root = NULL; |
afcdd129 | 6850 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
00246528 | 6851 | btrfs_put_root(log_root_tree); |
79787eaa | 6852 | |
abefa55a | 6853 | return 0; |
79787eaa | 6854 | error: |
b50c6e25 | 6855 | if (wc.trans) |
3a45bb20 | 6856 | btrfs_end_transaction(wc.trans); |
1aeb6b56 | 6857 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
79787eaa JM |
6858 | btrfs_free_path(path); |
6859 | return ret; | |
e02119d5 | 6860 | } |
12fcfd22 CM |
6861 | |
6862 | /* | |
6863 | * there are some corner cases where we want to force a full | |
6864 | * commit instead of allowing a directory to be logged. | |
6865 | * | |
6866 | * They revolve around files there were unlinked from the directory, and | |
6867 | * this function updates the parent directory so that a full commit is | |
6868 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
6869 | * |
6870 | * Must be called before the unlink operations (updates to the subvolume tree, | |
6871 | * inodes, etc) are done. | |
12fcfd22 CM |
6872 | */ |
6873 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 6874 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
6875 | int for_rename) |
6876 | { | |
af4176b4 CM |
6877 | /* |
6878 | * when we're logging a file, if it hasn't been renamed | |
6879 | * or unlinked, and its inode is fully committed on disk, | |
6880 | * we don't have to worry about walking up the directory chain | |
6881 | * to log its parents. | |
6882 | * | |
6883 | * So, we use the last_unlink_trans field to put this transid | |
6884 | * into the file. When the file is logged we check it and | |
6885 | * don't log the parents if the file is fully on disk. | |
6886 | */ | |
4176bdbf NB |
6887 | mutex_lock(&inode->log_mutex); |
6888 | inode->last_unlink_trans = trans->transid; | |
6889 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 6890 | |
12fcfd22 CM |
6891 | /* |
6892 | * if this directory was already logged any new | |
6893 | * names for this file/dir will get recorded | |
6894 | */ | |
4176bdbf | 6895 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
6896 | return; |
6897 | ||
6898 | /* | |
6899 | * if the inode we're about to unlink was logged, | |
6900 | * the log will be properly updated for any new names | |
6901 | */ | |
4176bdbf | 6902 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
6903 | return; |
6904 | ||
6905 | /* | |
6906 | * when renaming files across directories, if the directory | |
6907 | * there we're unlinking from gets fsync'd later on, there's | |
6908 | * no way to find the destination directory later and fsync it | |
6909 | * properly. So, we have to be conservative and force commits | |
6910 | * so the new name gets discovered. | |
6911 | */ | |
6912 | if (for_rename) | |
6913 | goto record; | |
6914 | ||
6915 | /* we can safely do the unlink without any special recording */ | |
6916 | return; | |
6917 | ||
6918 | record: | |
4176bdbf NB |
6919 | mutex_lock(&dir->log_mutex); |
6920 | dir->last_unlink_trans = trans->transid; | |
6921 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
6922 | } |
6923 | ||
6924 | /* | |
6925 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
6926 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
6927 | * that after replaying the log tree of the parent directory's root we will not | |
6928 | * see the snapshot anymore and at log replay time we will not see any log tree | |
6929 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
6930 | * it after replaying the log tree of the parent directory (which would replay | |
6931 | * the snapshot delete operation). | |
2be63d5c FM |
6932 | * |
6933 | * Must be called before the actual snapshot destroy operation (updates to the | |
6934 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
6935 | */ |
6936 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 6937 | struct btrfs_inode *dir) |
1ec9a1ae | 6938 | { |
43663557 NB |
6939 | mutex_lock(&dir->log_mutex); |
6940 | dir->last_unlink_trans = trans->transid; | |
6941 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
6942 | } |
6943 | ||
d5f5bd54 FM |
6944 | /** |
6945 | * Update the log after adding a new name for an inode. | |
6946 | * | |
6947 | * @trans: Transaction handle. | |
6948 | * @old_dentry: The dentry associated with the old name and the old | |
6949 | * parent directory. | |
6950 | * @old_dir: The inode of the previous parent directory for the case | |
6951 | * of a rename. For a link operation, it must be NULL. | |
88d2beec FM |
6952 | * @old_dir_index: The index number associated with the old name, meaningful |
6953 | * only for rename operations (when @old_dir is not NULL). | |
6954 | * Ignored for link operations. | |
d5f5bd54 FM |
6955 | * @parent: The dentry associated with the directory under which the |
6956 | * new name is located. | |
6957 | * | |
6958 | * Call this after adding a new name for an inode, as a result of a link or | |
6959 | * rename operation, and it will properly update the log to reflect the new name. | |
12fcfd22 | 6960 | */ |
75b463d2 | 6961 | void btrfs_log_new_name(struct btrfs_trans_handle *trans, |
d5f5bd54 | 6962 | struct dentry *old_dentry, struct btrfs_inode *old_dir, |
88d2beec | 6963 | u64 old_dir_index, struct dentry *parent) |
12fcfd22 | 6964 | { |
d5f5bd54 | 6965 | struct btrfs_inode *inode = BTRFS_I(d_inode(old_dentry)); |
259c4b96 | 6966 | struct btrfs_root *root = inode->root; |
75b463d2 | 6967 | struct btrfs_log_ctx ctx; |
259c4b96 | 6968 | bool log_pinned = false; |
0f8ce498 | 6969 | int ret; |
12fcfd22 | 6970 | |
af4176b4 CM |
6971 | /* |
6972 | * this will force the logging code to walk the dentry chain | |
6973 | * up for the file | |
6974 | */ | |
9a6509c4 | 6975 | if (!S_ISDIR(inode->vfs_inode.i_mode)) |
9ca5fbfb | 6976 | inode->last_unlink_trans = trans->transid; |
af4176b4 | 6977 | |
12fcfd22 CM |
6978 | /* |
6979 | * if this inode hasn't been logged and directory we're renaming it | |
6980 | * from hasn't been logged, we don't need to log it | |
6981 | */ | |
0f8ce498 FM |
6982 | ret = inode_logged(trans, inode, NULL); |
6983 | if (ret < 0) { | |
6984 | goto out; | |
6985 | } else if (ret == 0) { | |
6986 | if (!old_dir) | |
6987 | return; | |
6988 | /* | |
6989 | * If the inode was not logged and we are doing a rename (old_dir is not | |
6990 | * NULL), check if old_dir was logged - if it was not we can return and | |
6991 | * do nothing. | |
6992 | */ | |
6993 | ret = inode_logged(trans, old_dir, NULL); | |
6994 | if (ret < 0) | |
6995 | goto out; | |
6996 | else if (ret == 0) | |
6997 | return; | |
6998 | } | |
6999 | ret = 0; | |
12fcfd22 | 7000 | |
54a40fc3 FM |
7001 | /* |
7002 | * If we are doing a rename (old_dir is not NULL) from a directory that | |
88d2beec FM |
7003 | * was previously logged, make sure that on log replay we get the old |
7004 | * dir entry deleted. This is needed because we will also log the new | |
7005 | * name of the renamed inode, so we need to make sure that after log | |
7006 | * replay we don't end up with both the new and old dir entries existing. | |
54a40fc3 | 7007 | */ |
88d2beec FM |
7008 | if (old_dir && old_dir->logged_trans == trans->transid) { |
7009 | struct btrfs_root *log = old_dir->root->log_root; | |
7010 | struct btrfs_path *path; | |
88d2beec FM |
7011 | |
7012 | ASSERT(old_dir_index >= BTRFS_DIR_START_INDEX); | |
7013 | ||
259c4b96 FM |
7014 | /* |
7015 | * We have two inodes to update in the log, the old directory and | |
7016 | * the inode that got renamed, so we must pin the log to prevent | |
7017 | * anyone from syncing the log until we have updated both inodes | |
7018 | * in the log. | |
7019 | */ | |
7020 | log_pinned = true; | |
7021 | btrfs_pin_log_trans(root); | |
7022 | ||
88d2beec FM |
7023 | path = btrfs_alloc_path(); |
7024 | if (!path) { | |
259c4b96 FM |
7025 | ret = -ENOMEM; |
7026 | goto out; | |
88d2beec FM |
7027 | } |
7028 | ||
7029 | /* | |
7030 | * Other concurrent task might be logging the old directory, | |
7031 | * as it can be triggered when logging other inode that had or | |
7032 | * still has a dentry in the old directory. So take the old | |
7033 | * directory's log_mutex to prevent getting an -EEXIST when | |
7034 | * logging a key to record the deletion, or having that other | |
7035 | * task logging the old directory get an -EEXIST if it attempts | |
7036 | * to log the same key after we just did it. In both cases that | |
7037 | * would result in falling back to a transaction commit. | |
7038 | */ | |
7039 | mutex_lock(&old_dir->log_mutex); | |
7040 | ret = del_logged_dentry(trans, log, path, btrfs_ino(old_dir), | |
7041 | old_dentry->d_name.name, | |
7042 | old_dentry->d_name.len, old_dir_index); | |
7043 | if (ret > 0) { | |
7044 | /* | |
7045 | * The dentry does not exist in the log, so record its | |
7046 | * deletion. | |
7047 | */ | |
7048 | btrfs_release_path(path); | |
7049 | ret = insert_dir_log_key(trans, log, path, | |
7050 | btrfs_ino(old_dir), | |
7051 | old_dir_index, old_dir_index); | |
7052 | } | |
7053 | mutex_unlock(&old_dir->log_mutex); | |
7054 | ||
7055 | btrfs_free_path(path); | |
259c4b96 FM |
7056 | if (ret < 0) |
7057 | goto out; | |
88d2beec | 7058 | } |
54a40fc3 | 7059 | |
75b463d2 FM |
7060 | btrfs_init_log_ctx(&ctx, &inode->vfs_inode); |
7061 | ctx.logging_new_name = true; | |
7062 | /* | |
7063 | * We don't care about the return value. If we fail to log the new name | |
7064 | * then we know the next attempt to sync the log will fallback to a full | |
7065 | * transaction commit (due to a call to btrfs_set_log_full_commit()), so | |
7066 | * we don't need to worry about getting a log committed that has an | |
7067 | * inconsistent state after a rename operation. | |
7068 | */ | |
48778179 | 7069 | btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx); |
259c4b96 | 7070 | out: |
0f8ce498 FM |
7071 | /* |
7072 | * If an error happened mark the log for a full commit because it's not | |
7073 | * consistent and up to date or we couldn't find out if one of the | |
7074 | * inodes was logged before in this transaction. Do it before unpinning | |
7075 | * the log, to avoid any races with someone else trying to commit it. | |
7076 | */ | |
7077 | if (ret < 0) | |
7078 | btrfs_set_log_full_commit(trans); | |
7079 | if (log_pinned) | |
259c4b96 | 7080 | btrfs_end_log_trans(root); |
12fcfd22 CM |
7081 | } |
7082 |