]>
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" | |
e02119d5 CM |
22 | |
23 | /* magic values for the inode_only field in btrfs_log_inode: | |
24 | * | |
25 | * LOG_INODE_ALL means to log everything | |
26 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
27 | * during log replay | |
28 | */ | |
e13976cf DS |
29 | enum { |
30 | LOG_INODE_ALL, | |
31 | LOG_INODE_EXISTS, | |
32 | LOG_OTHER_INODE, | |
33 | LOG_OTHER_INODE_ALL, | |
34 | }; | |
e02119d5 | 35 | |
12fcfd22 CM |
36 | /* |
37 | * directory trouble cases | |
38 | * | |
39 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
40 | * log, we must force a full commit before doing an fsync of the directory | |
41 | * where the unlink was done. | |
42 | * ---> record transid of last unlink/rename per directory | |
43 | * | |
44 | * mkdir foo/some_dir | |
45 | * normal commit | |
46 | * rename foo/some_dir foo2/some_dir | |
47 | * mkdir foo/some_dir | |
48 | * fsync foo/some_dir/some_file | |
49 | * | |
50 | * The fsync above will unlink the original some_dir without recording | |
51 | * it in its new location (foo2). After a crash, some_dir will be gone | |
52 | * unless the fsync of some_file forces a full commit | |
53 | * | |
54 | * 2) we must log any new names for any file or dir that is in the fsync | |
55 | * log. ---> check inode while renaming/linking. | |
56 | * | |
57 | * 2a) we must log any new names for any file or dir during rename | |
58 | * when the directory they are being removed from was logged. | |
59 | * ---> check inode and old parent dir during rename | |
60 | * | |
61 | * 2a is actually the more important variant. With the extra logging | |
62 | * a crash might unlink the old name without recreating the new one | |
63 | * | |
64 | * 3) after a crash, we must go through any directories with a link count | |
65 | * of zero and redo the rm -rf | |
66 | * | |
67 | * mkdir f1/foo | |
68 | * normal commit | |
69 | * rm -rf f1/foo | |
70 | * fsync(f1) | |
71 | * | |
72 | * The directory f1 was fully removed from the FS, but fsync was never | |
73 | * called on f1, only its parent dir. After a crash the rm -rf must | |
74 | * be replayed. This must be able to recurse down the entire | |
75 | * directory tree. The inode link count fixup code takes care of the | |
76 | * ugly details. | |
77 | */ | |
78 | ||
e02119d5 CM |
79 | /* |
80 | * stages for the tree walking. The first | |
81 | * stage (0) is to only pin down the blocks we find | |
82 | * the second stage (1) is to make sure that all the inodes | |
83 | * we find in the log are created in the subvolume. | |
84 | * | |
85 | * The last stage is to deal with directories and links and extents | |
86 | * and all the other fun semantics | |
87 | */ | |
e13976cf DS |
88 | enum { |
89 | LOG_WALK_PIN_ONLY, | |
90 | LOG_WALK_REPLAY_INODES, | |
91 | LOG_WALK_REPLAY_DIR_INDEX, | |
92 | LOG_WALK_REPLAY_ALL, | |
93 | }; | |
e02119d5 | 94 | |
12fcfd22 | 95 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 96 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc | 97 | int inode_only, |
8407f553 | 98 | struct btrfs_log_ctx *ctx); |
ec051c0f YZ |
99 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
100 | struct btrfs_root *root, | |
101 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
102 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
103 | struct btrfs_root *root, | |
104 | struct btrfs_root *log, | |
105 | struct btrfs_path *path, | |
106 | u64 dirid, int del_all); | |
e02119d5 CM |
107 | |
108 | /* | |
109 | * tree logging is a special write ahead log used to make sure that | |
110 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
111 | * | |
112 | * Full tree commits are expensive because they require commonly | |
113 | * modified blocks to be recowed, creating many dirty pages in the | |
114 | * extent tree an 4x-6x higher write load than ext3. | |
115 | * | |
116 | * Instead of doing a tree commit on every fsync, we use the | |
117 | * key ranges and transaction ids to find items for a given file or directory | |
118 | * that have changed in this transaction. Those items are copied into | |
119 | * a special tree (one per subvolume root), that tree is written to disk | |
120 | * and then the fsync is considered complete. | |
121 | * | |
122 | * After a crash, items are copied out of the log-tree back into the | |
123 | * subvolume tree. Any file data extents found are recorded in the extent | |
124 | * allocation tree, and the log-tree freed. | |
125 | * | |
126 | * The log tree is read three times, once to pin down all the extents it is | |
127 | * using in ram and once, once to create all the inodes logged in the tree | |
128 | * and once to do all the other items. | |
129 | */ | |
130 | ||
e02119d5 CM |
131 | /* |
132 | * start a sub transaction and setup the log tree | |
133 | * this increments the log tree writer count to make the people | |
134 | * syncing the tree wait for us to finish | |
135 | */ | |
136 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
137 | struct btrfs_root *root, |
138 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 139 | { |
0b246afa | 140 | struct btrfs_fs_info *fs_info = root->fs_info; |
47876f7c | 141 | struct btrfs_root *tree_root = fs_info->tree_root; |
34eb2a52 | 142 | int ret = 0; |
7237f183 | 143 | |
47876f7c FM |
144 | /* |
145 | * First check if the log root tree was already created. If not, create | |
146 | * it before locking the root's log_mutex, just to keep lockdep happy. | |
147 | */ | |
148 | if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state)) { | |
149 | mutex_lock(&tree_root->log_mutex); | |
150 | if (!fs_info->log_root_tree) { | |
151 | ret = btrfs_init_log_root_tree(trans, fs_info); | |
152 | if (!ret) | |
153 | set_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state); | |
154 | } | |
155 | mutex_unlock(&tree_root->log_mutex); | |
156 | if (ret) | |
157 | return ret; | |
158 | } | |
159 | ||
7237f183 | 160 | mutex_lock(&root->log_mutex); |
34eb2a52 | 161 | |
7237f183 | 162 | if (root->log_root) { |
4884b8e8 | 163 | if (btrfs_need_log_full_commit(trans)) { |
50471a38 MX |
164 | ret = -EAGAIN; |
165 | goto out; | |
166 | } | |
34eb2a52 | 167 | |
ff782e0a | 168 | if (!root->log_start_pid) { |
27cdeb70 | 169 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 170 | root->log_start_pid = current->pid; |
ff782e0a | 171 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 172 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 173 | } |
34eb2a52 | 174 | } else { |
e02119d5 | 175 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 176 | if (ret) |
e87ac136 | 177 | goto out; |
34eb2a52 | 178 | |
e7a79811 | 179 | set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state); |
34eb2a52 Z |
180 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
181 | root->log_start_pid = current->pid; | |
e02119d5 | 182 | } |
34eb2a52 | 183 | |
7237f183 | 184 | atomic_inc(&root->log_writers); |
75b463d2 | 185 | if (ctx && !ctx->logging_new_name) { |
34eb2a52 | 186 | int index = root->log_transid % 2; |
8b050d35 | 187 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 188 | ctx->log_transid = root->log_transid; |
8b050d35 | 189 | } |
34eb2a52 | 190 | |
e87ac136 | 191 | out: |
7237f183 | 192 | mutex_unlock(&root->log_mutex); |
e87ac136 | 193 | return ret; |
e02119d5 CM |
194 | } |
195 | ||
196 | /* | |
197 | * returns 0 if there was a log transaction running and we were able | |
198 | * to join, or returns -ENOENT if there were not transactions | |
199 | * in progress | |
200 | */ | |
201 | static int join_running_log_trans(struct btrfs_root *root) | |
202 | { | |
203 | int ret = -ENOENT; | |
204 | ||
e7a79811 FM |
205 | if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state)) |
206 | return ret; | |
207 | ||
7237f183 | 208 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
209 | if (root->log_root) { |
210 | ret = 0; | |
7237f183 | 211 | atomic_inc(&root->log_writers); |
e02119d5 | 212 | } |
7237f183 | 213 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
214 | return ret; |
215 | } | |
216 | ||
12fcfd22 CM |
217 | /* |
218 | * This either makes the current running log transaction wait | |
219 | * until you call btrfs_end_log_trans() or it makes any future | |
220 | * log transactions wait until you call btrfs_end_log_trans() | |
221 | */ | |
45128b08 | 222 | void btrfs_pin_log_trans(struct btrfs_root *root) |
12fcfd22 | 223 | { |
12fcfd22 | 224 | atomic_inc(&root->log_writers); |
12fcfd22 CM |
225 | } |
226 | ||
e02119d5 CM |
227 | /* |
228 | * indicate we're done making changes to the log tree | |
229 | * and wake up anyone waiting to do a sync | |
230 | */ | |
143bede5 | 231 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 232 | { |
7237f183 | 233 | if (atomic_dec_and_test(&root->log_writers)) { |
093258e6 DS |
234 | /* atomic_dec_and_test implies a barrier */ |
235 | cond_wake_up_nomb(&root->log_writer_wait); | |
7237f183 | 236 | } |
e02119d5 CM |
237 | } |
238 | ||
247462a5 DS |
239 | static int btrfs_write_tree_block(struct extent_buffer *buf) |
240 | { | |
241 | return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, | |
242 | buf->start + buf->len - 1); | |
243 | } | |
244 | ||
245 | static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf) | |
246 | { | |
247 | filemap_fdatawait_range(buf->pages[0]->mapping, | |
248 | buf->start, buf->start + buf->len - 1); | |
249 | } | |
e02119d5 CM |
250 | |
251 | /* | |
252 | * the walk control struct is used to pass state down the chain when | |
253 | * processing the log tree. The stage field tells us which part | |
254 | * of the log tree processing we are currently doing. The others | |
255 | * are state fields used for that specific part | |
256 | */ | |
257 | struct walk_control { | |
258 | /* should we free the extent on disk when done? This is used | |
259 | * at transaction commit time while freeing a log tree | |
260 | */ | |
261 | int free; | |
262 | ||
263 | /* should we write out the extent buffer? This is used | |
264 | * while flushing the log tree to disk during a sync | |
265 | */ | |
266 | int write; | |
267 | ||
268 | /* should we wait for the extent buffer io to finish? Also used | |
269 | * while flushing the log tree to disk for a sync | |
270 | */ | |
271 | int wait; | |
272 | ||
273 | /* pin only walk, we record which extents on disk belong to the | |
274 | * log trees | |
275 | */ | |
276 | int pin; | |
277 | ||
278 | /* what stage of the replay code we're currently in */ | |
279 | int stage; | |
280 | ||
f2d72f42 FM |
281 | /* |
282 | * Ignore any items from the inode currently being processed. Needs | |
283 | * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in | |
284 | * the LOG_WALK_REPLAY_INODES stage. | |
285 | */ | |
286 | bool ignore_cur_inode; | |
287 | ||
e02119d5 CM |
288 | /* the root we are currently replaying */ |
289 | struct btrfs_root *replay_dest; | |
290 | ||
291 | /* the trans handle for the current replay */ | |
292 | struct btrfs_trans_handle *trans; | |
293 | ||
294 | /* the function that gets used to process blocks we find in the | |
295 | * tree. Note the extent_buffer might not be up to date when it is | |
296 | * passed in, and it must be checked or read if you need the data | |
297 | * inside it | |
298 | */ | |
299 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 300 | struct walk_control *wc, u64 gen, int level); |
e02119d5 CM |
301 | }; |
302 | ||
303 | /* | |
304 | * process_func used to pin down extents, write them or wait on them | |
305 | */ | |
306 | static int process_one_buffer(struct btrfs_root *log, | |
307 | struct extent_buffer *eb, | |
581c1760 | 308 | struct walk_control *wc, u64 gen, int level) |
e02119d5 | 309 | { |
0b246afa | 310 | struct btrfs_fs_info *fs_info = log->fs_info; |
b50c6e25 JB |
311 | int ret = 0; |
312 | ||
8c2a1a30 JB |
313 | /* |
314 | * If this fs is mixed then we need to be able to process the leaves to | |
315 | * pin down any logged extents, so we have to read the block. | |
316 | */ | |
0b246afa | 317 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
581c1760 | 318 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
8c2a1a30 JB |
319 | if (ret) |
320 | return ret; | |
321 | } | |
322 | ||
04018de5 | 323 | if (wc->pin) |
9fce5704 | 324 | ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start, |
2ff7e61e | 325 | eb->len); |
e02119d5 | 326 | |
b50c6e25 | 327 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 | 328 | if (wc->pin && btrfs_header_level(eb) == 0) |
bcdc428c | 329 | ret = btrfs_exclude_logged_extents(eb); |
e02119d5 CM |
330 | if (wc->write) |
331 | btrfs_write_tree_block(eb); | |
332 | if (wc->wait) | |
333 | btrfs_wait_tree_block_writeback(eb); | |
334 | } | |
b50c6e25 | 335 | return ret; |
e02119d5 CM |
336 | } |
337 | ||
338 | /* | |
339 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
340 | * to the src data we are copying out. | |
341 | * | |
342 | * root is the tree we are copying into, and path is a scratch | |
343 | * path for use in this function (it should be released on entry and | |
344 | * will be released on exit). | |
345 | * | |
346 | * If the key is already in the destination tree the existing item is | |
347 | * overwritten. If the existing item isn't big enough, it is extended. | |
348 | * If it is too large, it is truncated. | |
349 | * | |
350 | * If the key isn't in the destination yet, a new item is inserted. | |
351 | */ | |
352 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
353 | struct btrfs_root *root, | |
354 | struct btrfs_path *path, | |
355 | struct extent_buffer *eb, int slot, | |
356 | struct btrfs_key *key) | |
357 | { | |
358 | int ret; | |
359 | u32 item_size; | |
360 | u64 saved_i_size = 0; | |
361 | int save_old_i_size = 0; | |
362 | unsigned long src_ptr; | |
363 | unsigned long dst_ptr; | |
364 | int overwrite_root = 0; | |
4bc4bee4 | 365 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
366 | |
367 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
368 | overwrite_root = 1; | |
369 | ||
370 | item_size = btrfs_item_size_nr(eb, slot); | |
371 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
372 | ||
373 | /* look for the key in the destination tree */ | |
374 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
375 | if (ret < 0) |
376 | return ret; | |
377 | ||
e02119d5 CM |
378 | if (ret == 0) { |
379 | char *src_copy; | |
380 | char *dst_copy; | |
381 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
382 | path->slots[0]); | |
383 | if (dst_size != item_size) | |
384 | goto insert; | |
385 | ||
386 | if (item_size == 0) { | |
b3b4aa74 | 387 | btrfs_release_path(path); |
e02119d5 CM |
388 | return 0; |
389 | } | |
390 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
391 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 392 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 393 | btrfs_release_path(path); |
2a29edc6 | 394 | kfree(dst_copy); |
395 | kfree(src_copy); | |
396 | return -ENOMEM; | |
397 | } | |
e02119d5 CM |
398 | |
399 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
400 | ||
401 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
402 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
403 | item_size); | |
404 | ret = memcmp(dst_copy, src_copy, item_size); | |
405 | ||
406 | kfree(dst_copy); | |
407 | kfree(src_copy); | |
408 | /* | |
409 | * they have the same contents, just return, this saves | |
410 | * us from cowing blocks in the destination tree and doing | |
411 | * extra writes that may not have been done by a previous | |
412 | * sync | |
413 | */ | |
414 | if (ret == 0) { | |
b3b4aa74 | 415 | btrfs_release_path(path); |
e02119d5 CM |
416 | return 0; |
417 | } | |
418 | ||
4bc4bee4 JB |
419 | /* |
420 | * We need to load the old nbytes into the inode so when we | |
421 | * replay the extents we've logged we get the right nbytes. | |
422 | */ | |
423 | if (inode_item) { | |
424 | struct btrfs_inode_item *item; | |
425 | u64 nbytes; | |
d555438b | 426 | u32 mode; |
4bc4bee4 JB |
427 | |
428 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
429 | struct btrfs_inode_item); | |
430 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
431 | item = btrfs_item_ptr(eb, slot, | |
432 | struct btrfs_inode_item); | |
433 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
434 | |
435 | /* | |
436 | * If this is a directory we need to reset the i_size to | |
437 | * 0 so that we can set it up properly when replaying | |
438 | * the rest of the items in this log. | |
439 | */ | |
440 | mode = btrfs_inode_mode(eb, item); | |
441 | if (S_ISDIR(mode)) | |
442 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
443 | } |
444 | } else if (inode_item) { | |
445 | struct btrfs_inode_item *item; | |
d555438b | 446 | u32 mode; |
4bc4bee4 JB |
447 | |
448 | /* | |
449 | * New inode, set nbytes to 0 so that the nbytes comes out | |
450 | * properly when we replay the extents. | |
451 | */ | |
452 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
453 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
454 | |
455 | /* | |
456 | * If this is a directory we need to reset the i_size to 0 so | |
457 | * that we can set it up properly when replaying the rest of | |
458 | * the items in this log. | |
459 | */ | |
460 | mode = btrfs_inode_mode(eb, item); | |
461 | if (S_ISDIR(mode)) | |
462 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
463 | } |
464 | insert: | |
b3b4aa74 | 465 | btrfs_release_path(path); |
e02119d5 | 466 | /* try to insert the key into the destination tree */ |
df8d116f | 467 | path->skip_release_on_error = 1; |
e02119d5 CM |
468 | ret = btrfs_insert_empty_item(trans, root, path, |
469 | key, item_size); | |
df8d116f | 470 | path->skip_release_on_error = 0; |
e02119d5 CM |
471 | |
472 | /* make sure any existing item is the correct size */ | |
df8d116f | 473 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 CM |
474 | u32 found_size; |
475 | found_size = btrfs_item_size_nr(path->nodes[0], | |
476 | path->slots[0]); | |
143bede5 | 477 | if (found_size > item_size) |
78ac4f9e | 478 | btrfs_truncate_item(path, item_size, 1); |
143bede5 | 479 | else if (found_size < item_size) |
c71dd880 | 480 | btrfs_extend_item(path, item_size - found_size); |
e02119d5 | 481 | } else if (ret) { |
4a500fd1 | 482 | return ret; |
e02119d5 CM |
483 | } |
484 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
485 | path->slots[0]); | |
486 | ||
487 | /* don't overwrite an existing inode if the generation number | |
488 | * was logged as zero. This is done when the tree logging code | |
489 | * is just logging an inode to make sure it exists after recovery. | |
490 | * | |
491 | * Also, don't overwrite i_size on directories during replay. | |
492 | * log replay inserts and removes directory items based on the | |
493 | * state of the tree found in the subvolume, and i_size is modified | |
494 | * as it goes | |
495 | */ | |
496 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
497 | struct btrfs_inode_item *src_item; | |
498 | struct btrfs_inode_item *dst_item; | |
499 | ||
500 | src_item = (struct btrfs_inode_item *)src_ptr; | |
501 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
502 | ||
1a4bcf47 FM |
503 | if (btrfs_inode_generation(eb, src_item) == 0) { |
504 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 505 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 506 | |
2f2ff0ee FM |
507 | /* |
508 | * For regular files an ino_size == 0 is used only when | |
509 | * logging that an inode exists, as part of a directory | |
510 | * fsync, and the inode wasn't fsynced before. In this | |
511 | * case don't set the size of the inode in the fs/subvol | |
512 | * tree, otherwise we would be throwing valid data away. | |
513 | */ | |
1a4bcf47 | 514 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee | 515 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
60d48e2e DS |
516 | ino_size != 0) |
517 | btrfs_set_inode_size(dst_eb, dst_item, ino_size); | |
e02119d5 | 518 | goto no_copy; |
1a4bcf47 | 519 | } |
e02119d5 CM |
520 | |
521 | if (overwrite_root && | |
522 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
523 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
524 | save_old_i_size = 1; | |
525 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
526 | dst_item); | |
527 | } | |
528 | } | |
529 | ||
530 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
531 | src_ptr, item_size); | |
532 | ||
533 | if (save_old_i_size) { | |
534 | struct btrfs_inode_item *dst_item; | |
535 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
536 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
537 | } | |
538 | ||
539 | /* make sure the generation is filled in */ | |
540 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
541 | struct btrfs_inode_item *dst_item; | |
542 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
543 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
544 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
545 | trans->transid); | |
546 | } | |
547 | } | |
548 | no_copy: | |
549 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 550 | btrfs_release_path(path); |
e02119d5 CM |
551 | return 0; |
552 | } | |
553 | ||
554 | /* | |
555 | * simple helper to read an inode off the disk from a given root | |
556 | * This can only be called for subvolume roots and not for the log | |
557 | */ | |
558 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
559 | u64 objectid) | |
560 | { | |
561 | struct inode *inode; | |
e02119d5 | 562 | |
0202e83f | 563 | inode = btrfs_iget(root->fs_info->sb, objectid, root); |
2e19f1f9 | 564 | if (IS_ERR(inode)) |
5d4f98a2 | 565 | inode = NULL; |
e02119d5 CM |
566 | return inode; |
567 | } | |
568 | ||
569 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
570 | * subvolume 'root'. path is released on entry and should be released | |
571 | * on exit. | |
572 | * | |
573 | * extents in the log tree have not been allocated out of the extent | |
574 | * tree yet. So, this completes the allocation, taking a reference | |
575 | * as required if the extent already exists or creating a new extent | |
576 | * if it isn't in the extent allocation tree yet. | |
577 | * | |
578 | * The extent is inserted into the file, dropping any existing extents | |
579 | * from the file that overlap the new one. | |
580 | */ | |
581 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
582 | struct btrfs_root *root, | |
583 | struct btrfs_path *path, | |
584 | struct extent_buffer *eb, int slot, | |
585 | struct btrfs_key *key) | |
586 | { | |
5893dfb9 | 587 | struct btrfs_drop_extents_args drop_args = { 0 }; |
0b246afa | 588 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 589 | int found_type; |
e02119d5 | 590 | u64 extent_end; |
e02119d5 | 591 | u64 start = key->offset; |
4bc4bee4 | 592 | u64 nbytes = 0; |
e02119d5 CM |
593 | struct btrfs_file_extent_item *item; |
594 | struct inode *inode = NULL; | |
595 | unsigned long size; | |
596 | int ret = 0; | |
597 | ||
598 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
599 | found_type = btrfs_file_extent_type(eb, item); | |
600 | ||
d899e052 | 601 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
602 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
603 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
604 | extent_end = start + nbytes; | |
605 | ||
606 | /* | |
607 | * We don't add to the inodes nbytes if we are prealloc or a | |
608 | * hole. | |
609 | */ | |
610 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
611 | nbytes = 0; | |
612 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 613 | size = btrfs_file_extent_ram_bytes(eb, item); |
4bc4bee4 | 614 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
da17066c | 615 | extent_end = ALIGN(start + size, |
0b246afa | 616 | fs_info->sectorsize); |
e02119d5 CM |
617 | } else { |
618 | ret = 0; | |
619 | goto out; | |
620 | } | |
621 | ||
622 | inode = read_one_inode(root, key->objectid); | |
623 | if (!inode) { | |
624 | ret = -EIO; | |
625 | goto out; | |
626 | } | |
627 | ||
628 | /* | |
629 | * first check to see if we already have this extent in the | |
630 | * file. This must be done before the btrfs_drop_extents run | |
631 | * so we don't try to drop this extent. | |
632 | */ | |
f85b7379 DS |
633 | ret = btrfs_lookup_file_extent(trans, root, path, |
634 | btrfs_ino(BTRFS_I(inode)), start, 0); | |
e02119d5 | 635 | |
d899e052 YZ |
636 | if (ret == 0 && |
637 | (found_type == BTRFS_FILE_EXTENT_REG || | |
638 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
639 | struct btrfs_file_extent_item cmp1; |
640 | struct btrfs_file_extent_item cmp2; | |
641 | struct btrfs_file_extent_item *existing; | |
642 | struct extent_buffer *leaf; | |
643 | ||
644 | leaf = path->nodes[0]; | |
645 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
646 | struct btrfs_file_extent_item); | |
647 | ||
648 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
649 | sizeof(cmp1)); | |
650 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
651 | sizeof(cmp2)); | |
652 | ||
653 | /* | |
654 | * we already have a pointer to this exact extent, | |
655 | * we don't have to do anything | |
656 | */ | |
657 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 658 | btrfs_release_path(path); |
e02119d5 CM |
659 | goto out; |
660 | } | |
661 | } | |
b3b4aa74 | 662 | btrfs_release_path(path); |
e02119d5 CM |
663 | |
664 | /* drop any overlapping extents */ | |
5893dfb9 FM |
665 | drop_args.start = start; |
666 | drop_args.end = extent_end; | |
667 | drop_args.drop_cache = true; | |
668 | ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args); | |
3650860b JB |
669 | if (ret) |
670 | goto out; | |
e02119d5 | 671 | |
07d400a6 YZ |
672 | if (found_type == BTRFS_FILE_EXTENT_REG || |
673 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 674 | u64 offset; |
07d400a6 YZ |
675 | unsigned long dest_offset; |
676 | struct btrfs_key ins; | |
677 | ||
3168021c FM |
678 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0 && |
679 | btrfs_fs_incompat(fs_info, NO_HOLES)) | |
680 | goto update_inode; | |
681 | ||
07d400a6 YZ |
682 | ret = btrfs_insert_empty_item(trans, root, path, key, |
683 | sizeof(*item)); | |
3650860b JB |
684 | if (ret) |
685 | goto out; | |
07d400a6 YZ |
686 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
687 | path->slots[0]); | |
688 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
689 | (unsigned long)item, sizeof(*item)); | |
690 | ||
691 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
692 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
693 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 694 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 695 | |
df2c95f3 QW |
696 | /* |
697 | * Manually record dirty extent, as here we did a shallow | |
698 | * file extent item copy and skip normal backref update, | |
699 | * but modifying extent tree all by ourselves. | |
700 | * So need to manually record dirty extent for qgroup, | |
701 | * as the owner of the file extent changed from log tree | |
702 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
703 | */ | |
a95f3aaf | 704 | ret = btrfs_qgroup_trace_extent(trans, |
df2c95f3 QW |
705 | btrfs_file_extent_disk_bytenr(eb, item), |
706 | btrfs_file_extent_disk_num_bytes(eb, item), | |
707 | GFP_NOFS); | |
708 | if (ret < 0) | |
709 | goto out; | |
710 | ||
07d400a6 | 711 | if (ins.objectid > 0) { |
82fa113f | 712 | struct btrfs_ref ref = { 0 }; |
07d400a6 YZ |
713 | u64 csum_start; |
714 | u64 csum_end; | |
715 | LIST_HEAD(ordered_sums); | |
82fa113f | 716 | |
07d400a6 YZ |
717 | /* |
718 | * is this extent already allocated in the extent | |
719 | * allocation tree? If so, just add a reference | |
720 | */ | |
2ff7e61e | 721 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 YZ |
722 | ins.offset); |
723 | if (ret == 0) { | |
82fa113f QW |
724 | btrfs_init_generic_ref(&ref, |
725 | BTRFS_ADD_DELAYED_REF, | |
726 | ins.objectid, ins.offset, 0); | |
727 | btrfs_init_data_ref(&ref, | |
728 | root->root_key.objectid, | |
b06c4bf5 | 729 | key->objectid, offset); |
82fa113f | 730 | ret = btrfs_inc_extent_ref(trans, &ref); |
b50c6e25 JB |
731 | if (ret) |
732 | goto out; | |
07d400a6 YZ |
733 | } else { |
734 | /* | |
735 | * insert the extent pointer in the extent | |
736 | * allocation tree | |
737 | */ | |
5d4f98a2 | 738 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e | 739 | root->root_key.objectid, |
5d4f98a2 | 740 | key->objectid, offset, &ins); |
b50c6e25 JB |
741 | if (ret) |
742 | goto out; | |
07d400a6 | 743 | } |
b3b4aa74 | 744 | btrfs_release_path(path); |
07d400a6 YZ |
745 | |
746 | if (btrfs_file_extent_compression(eb, item)) { | |
747 | csum_start = ins.objectid; | |
748 | csum_end = csum_start + ins.offset; | |
749 | } else { | |
750 | csum_start = ins.objectid + | |
751 | btrfs_file_extent_offset(eb, item); | |
752 | csum_end = csum_start + | |
753 | btrfs_file_extent_num_bytes(eb, item); | |
754 | } | |
755 | ||
756 | ret = btrfs_lookup_csums_range(root->log_root, | |
757 | csum_start, csum_end - 1, | |
a2de733c | 758 | &ordered_sums, 0); |
3650860b JB |
759 | if (ret) |
760 | goto out; | |
b84b8390 FM |
761 | /* |
762 | * Now delete all existing cums in the csum root that | |
763 | * cover our range. We do this because we can have an | |
764 | * extent that is completely referenced by one file | |
765 | * extent item and partially referenced by another | |
766 | * file extent item (like after using the clone or | |
767 | * extent_same ioctls). In this case if we end up doing | |
768 | * the replay of the one that partially references the | |
769 | * extent first, and we do not do the csum deletion | |
770 | * below, we can get 2 csum items in the csum tree that | |
771 | * overlap each other. For example, imagine our log has | |
772 | * the two following file extent items: | |
773 | * | |
774 | * key (257 EXTENT_DATA 409600) | |
775 | * extent data disk byte 12845056 nr 102400 | |
776 | * extent data offset 20480 nr 20480 ram 102400 | |
777 | * | |
778 | * key (257 EXTENT_DATA 819200) | |
779 | * extent data disk byte 12845056 nr 102400 | |
780 | * extent data offset 0 nr 102400 ram 102400 | |
781 | * | |
782 | * Where the second one fully references the 100K extent | |
783 | * that starts at disk byte 12845056, and the log tree | |
784 | * has a single csum item that covers the entire range | |
785 | * of the extent: | |
786 | * | |
787 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
788 | * | |
789 | * After the first file extent item is replayed, the | |
790 | * csum tree gets the following csum item: | |
791 | * | |
792 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
793 | * | |
794 | * Which covers the 20K sub-range starting at offset 20K | |
795 | * of our extent. Now when we replay the second file | |
796 | * extent item, if we do not delete existing csum items | |
797 | * that cover any of its blocks, we end up getting two | |
798 | * csum items in our csum tree that overlap each other: | |
799 | * | |
800 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
801 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
802 | * | |
803 | * Which is a problem, because after this anyone trying | |
804 | * to lookup up for the checksum of any block of our | |
805 | * extent starting at an offset of 40K or higher, will | |
806 | * end up looking at the second csum item only, which | |
807 | * does not contain the checksum for any block starting | |
808 | * at offset 40K or higher of our extent. | |
809 | */ | |
07d400a6 YZ |
810 | while (!list_empty(&ordered_sums)) { |
811 | struct btrfs_ordered_sum *sums; | |
812 | sums = list_entry(ordered_sums.next, | |
813 | struct btrfs_ordered_sum, | |
814 | list); | |
b84b8390 | 815 | if (!ret) |
40e046ac FM |
816 | ret = btrfs_del_csums(trans, |
817 | fs_info->csum_root, | |
5b4aacef JM |
818 | sums->bytenr, |
819 | sums->len); | |
3650860b JB |
820 | if (!ret) |
821 | ret = btrfs_csum_file_blocks(trans, | |
0b246afa | 822 | fs_info->csum_root, sums); |
07d400a6 YZ |
823 | list_del(&sums->list); |
824 | kfree(sums); | |
825 | } | |
3650860b JB |
826 | if (ret) |
827 | goto out; | |
07d400a6 | 828 | } else { |
b3b4aa74 | 829 | btrfs_release_path(path); |
07d400a6 YZ |
830 | } |
831 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
832 | /* inline extents are easy, we just overwrite them */ | |
833 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
834 | if (ret) |
835 | goto out; | |
07d400a6 | 836 | } |
e02119d5 | 837 | |
9ddc959e JB |
838 | ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, |
839 | extent_end - start); | |
840 | if (ret) | |
841 | goto out; | |
842 | ||
3168021c | 843 | update_inode: |
2766ff61 | 844 | btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found); |
9a56fcd1 | 845 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e02119d5 CM |
846 | out: |
847 | if (inode) | |
848 | iput(inode); | |
849 | return ret; | |
850 | } | |
851 | ||
852 | /* | |
853 | * when cleaning up conflicts between the directory names in the | |
854 | * subvolume, directory names in the log and directory names in the | |
855 | * inode back references, we may have to unlink inodes from directories. | |
856 | * | |
857 | * This is a helper function to do the unlink of a specific directory | |
858 | * item | |
859 | */ | |
860 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
861 | struct btrfs_root *root, | |
862 | struct btrfs_path *path, | |
207e7d92 | 863 | struct btrfs_inode *dir, |
e02119d5 CM |
864 | struct btrfs_dir_item *di) |
865 | { | |
866 | struct inode *inode; | |
867 | char *name; | |
868 | int name_len; | |
869 | struct extent_buffer *leaf; | |
870 | struct btrfs_key location; | |
871 | int ret; | |
872 | ||
873 | leaf = path->nodes[0]; | |
874 | ||
875 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
876 | name_len = btrfs_dir_name_len(leaf, di); | |
877 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 878 | if (!name) |
879 | return -ENOMEM; | |
880 | ||
e02119d5 | 881 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 882 | btrfs_release_path(path); |
e02119d5 CM |
883 | |
884 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 885 | if (!inode) { |
3650860b JB |
886 | ret = -EIO; |
887 | goto out; | |
c00e9493 | 888 | } |
e02119d5 | 889 | |
ec051c0f | 890 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
891 | if (ret) |
892 | goto out; | |
12fcfd22 | 893 | |
207e7d92 NB |
894 | ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name, |
895 | name_len); | |
3650860b JB |
896 | if (ret) |
897 | goto out; | |
ada9af21 | 898 | else |
e5c304e6 | 899 | ret = btrfs_run_delayed_items(trans); |
3650860b | 900 | out: |
e02119d5 | 901 | kfree(name); |
e02119d5 CM |
902 | iput(inode); |
903 | return ret; | |
904 | } | |
905 | ||
906 | /* | |
907 | * helper function to see if a given name and sequence number found | |
908 | * in an inode back reference are already in a directory and correctly | |
909 | * point to this inode | |
910 | */ | |
911 | static noinline int inode_in_dir(struct btrfs_root *root, | |
912 | struct btrfs_path *path, | |
913 | u64 dirid, u64 objectid, u64 index, | |
914 | const char *name, int name_len) | |
915 | { | |
916 | struct btrfs_dir_item *di; | |
917 | struct btrfs_key location; | |
918 | int match = 0; | |
919 | ||
920 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
921 | index, name, name_len, 0); | |
922 | if (di && !IS_ERR(di)) { | |
923 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
924 | if (location.objectid != objectid) | |
925 | goto out; | |
926 | } else | |
927 | goto out; | |
b3b4aa74 | 928 | btrfs_release_path(path); |
e02119d5 CM |
929 | |
930 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
931 | if (di && !IS_ERR(di)) { | |
932 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
933 | if (location.objectid != objectid) | |
934 | goto out; | |
935 | } else | |
936 | goto out; | |
937 | match = 1; | |
938 | out: | |
b3b4aa74 | 939 | btrfs_release_path(path); |
e02119d5 CM |
940 | return match; |
941 | } | |
942 | ||
943 | /* | |
944 | * helper function to check a log tree for a named back reference in | |
945 | * an inode. This is used to decide if a back reference that is | |
946 | * found in the subvolume conflicts with what we find in the log. | |
947 | * | |
948 | * inode backreferences may have multiple refs in a single item, | |
949 | * during replay we process one reference at a time, and we don't | |
950 | * want to delete valid links to a file from the subvolume if that | |
951 | * link is also in the log. | |
952 | */ | |
953 | static noinline int backref_in_log(struct btrfs_root *log, | |
954 | struct btrfs_key *key, | |
f186373f | 955 | u64 ref_objectid, |
df8d116f | 956 | const char *name, int namelen) |
e02119d5 CM |
957 | { |
958 | struct btrfs_path *path; | |
e02119d5 | 959 | int ret; |
e02119d5 CM |
960 | |
961 | path = btrfs_alloc_path(); | |
2a29edc6 | 962 | if (!path) |
963 | return -ENOMEM; | |
964 | ||
e02119d5 | 965 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
d3316c82 NB |
966 | if (ret < 0) { |
967 | goto out; | |
968 | } else if (ret == 1) { | |
89cbf5f6 | 969 | ret = 0; |
f186373f MF |
970 | goto out; |
971 | } | |
972 | ||
89cbf5f6 NB |
973 | if (key->type == BTRFS_INODE_EXTREF_KEY) |
974 | ret = !!btrfs_find_name_in_ext_backref(path->nodes[0], | |
975 | path->slots[0], | |
976 | ref_objectid, | |
977 | name, namelen); | |
978 | else | |
979 | ret = !!btrfs_find_name_in_backref(path->nodes[0], | |
980 | path->slots[0], | |
981 | name, namelen); | |
e02119d5 CM |
982 | out: |
983 | btrfs_free_path(path); | |
89cbf5f6 | 984 | return ret; |
e02119d5 CM |
985 | } |
986 | ||
5a1d7843 | 987 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 988 | struct btrfs_root *root, |
e02119d5 | 989 | struct btrfs_path *path, |
5a1d7843 | 990 | struct btrfs_root *log_root, |
94c91a1f NB |
991 | struct btrfs_inode *dir, |
992 | struct btrfs_inode *inode, | |
f186373f MF |
993 | u64 inode_objectid, u64 parent_objectid, |
994 | u64 ref_index, char *name, int namelen, | |
995 | int *search_done) | |
e02119d5 | 996 | { |
34f3e4f2 | 997 | int ret; |
f186373f MF |
998 | char *victim_name; |
999 | int victim_name_len; | |
1000 | struct extent_buffer *leaf; | |
5a1d7843 | 1001 | struct btrfs_dir_item *di; |
f186373f MF |
1002 | struct btrfs_key search_key; |
1003 | struct btrfs_inode_extref *extref; | |
c622ae60 | 1004 | |
f186373f MF |
1005 | again: |
1006 | /* Search old style refs */ | |
1007 | search_key.objectid = inode_objectid; | |
1008 | search_key.type = BTRFS_INODE_REF_KEY; | |
1009 | search_key.offset = parent_objectid; | |
1010 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 1011 | if (ret == 0) { |
e02119d5 CM |
1012 | struct btrfs_inode_ref *victim_ref; |
1013 | unsigned long ptr; | |
1014 | unsigned long ptr_end; | |
f186373f MF |
1015 | |
1016 | leaf = path->nodes[0]; | |
e02119d5 CM |
1017 | |
1018 | /* are we trying to overwrite a back ref for the root directory | |
1019 | * if so, just jump out, we're done | |
1020 | */ | |
f186373f | 1021 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1022 | return 1; |
e02119d5 CM |
1023 | |
1024 | /* check all the names in this back reference to see | |
1025 | * if they are in the log. if so, we allow them to stay | |
1026 | * otherwise they must be unlinked as a conflict | |
1027 | */ | |
1028 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1029 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 1030 | while (ptr < ptr_end) { |
e02119d5 CM |
1031 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1032 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1033 | victim_ref); | |
1034 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1035 | if (!victim_name) |
1036 | return -ENOMEM; | |
e02119d5 CM |
1037 | |
1038 | read_extent_buffer(leaf, victim_name, | |
1039 | (unsigned long)(victim_ref + 1), | |
1040 | victim_name_len); | |
1041 | ||
d3316c82 NB |
1042 | ret = backref_in_log(log_root, &search_key, |
1043 | parent_objectid, victim_name, | |
1044 | victim_name_len); | |
1045 | if (ret < 0) { | |
1046 | kfree(victim_name); | |
1047 | return ret; | |
1048 | } else if (!ret) { | |
94c91a1f | 1049 | inc_nlink(&inode->vfs_inode); |
b3b4aa74 | 1050 | btrfs_release_path(path); |
12fcfd22 | 1051 | |
94c91a1f | 1052 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
4ec5934e | 1053 | victim_name, victim_name_len); |
f186373f | 1054 | kfree(victim_name); |
3650860b JB |
1055 | if (ret) |
1056 | return ret; | |
e5c304e6 | 1057 | ret = btrfs_run_delayed_items(trans); |
ada9af21 FDBM |
1058 | if (ret) |
1059 | return ret; | |
f186373f MF |
1060 | *search_done = 1; |
1061 | goto again; | |
e02119d5 CM |
1062 | } |
1063 | kfree(victim_name); | |
f186373f | 1064 | |
e02119d5 CM |
1065 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1066 | } | |
e02119d5 | 1067 | |
c622ae60 | 1068 | /* |
1069 | * NOTE: we have searched root tree and checked the | |
bb7ab3b9 | 1070 | * corresponding ref, it does not need to check again. |
c622ae60 | 1071 | */ |
5a1d7843 | 1072 | *search_done = 1; |
e02119d5 | 1073 | } |
b3b4aa74 | 1074 | btrfs_release_path(path); |
e02119d5 | 1075 | |
f186373f MF |
1076 | /* Same search but for extended refs */ |
1077 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1078 | inode_objectid, parent_objectid, 0, | |
1079 | 0); | |
1080 | if (!IS_ERR_OR_NULL(extref)) { | |
1081 | u32 item_size; | |
1082 | u32 cur_offset = 0; | |
1083 | unsigned long base; | |
1084 | struct inode *victim_parent; | |
1085 | ||
1086 | leaf = path->nodes[0]; | |
1087 | ||
1088 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1089 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1090 | ||
1091 | while (cur_offset < item_size) { | |
dd9ef135 | 1092 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1093 | |
1094 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1095 | ||
1096 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1097 | goto next; | |
1098 | ||
1099 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1100 | if (!victim_name) |
1101 | return -ENOMEM; | |
f186373f MF |
1102 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1103 | victim_name_len); | |
1104 | ||
1105 | search_key.objectid = inode_objectid; | |
1106 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1107 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1108 | victim_name, | |
1109 | victim_name_len); | |
d3316c82 NB |
1110 | ret = backref_in_log(log_root, &search_key, |
1111 | parent_objectid, victim_name, | |
1112 | victim_name_len); | |
1113 | if (ret < 0) { | |
1114 | return ret; | |
1115 | } else if (!ret) { | |
f186373f MF |
1116 | ret = -ENOENT; |
1117 | victim_parent = read_one_inode(root, | |
94c91a1f | 1118 | parent_objectid); |
f186373f | 1119 | if (victim_parent) { |
94c91a1f | 1120 | inc_nlink(&inode->vfs_inode); |
f186373f MF |
1121 | btrfs_release_path(path); |
1122 | ||
1123 | ret = btrfs_unlink_inode(trans, root, | |
4ec5934e | 1124 | BTRFS_I(victim_parent), |
94c91a1f | 1125 | inode, |
4ec5934e NB |
1126 | victim_name, |
1127 | victim_name_len); | |
ada9af21 FDBM |
1128 | if (!ret) |
1129 | ret = btrfs_run_delayed_items( | |
e5c304e6 | 1130 | trans); |
f186373f | 1131 | } |
f186373f MF |
1132 | iput(victim_parent); |
1133 | kfree(victim_name); | |
3650860b JB |
1134 | if (ret) |
1135 | return ret; | |
f186373f MF |
1136 | *search_done = 1; |
1137 | goto again; | |
1138 | } | |
1139 | kfree(victim_name); | |
f186373f MF |
1140 | next: |
1141 | cur_offset += victim_name_len + sizeof(*extref); | |
1142 | } | |
1143 | *search_done = 1; | |
1144 | } | |
1145 | btrfs_release_path(path); | |
1146 | ||
34f3e4f2 | 1147 | /* look for a conflicting sequence number */ |
94c91a1f | 1148 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
f186373f | 1149 | ref_index, name, namelen, 0); |
34f3e4f2 | 1150 | if (di && !IS_ERR(di)) { |
94c91a1f | 1151 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1152 | if (ret) |
1153 | return ret; | |
34f3e4f2 | 1154 | } |
1155 | btrfs_release_path(path); | |
1156 | ||
52042d8e | 1157 | /* look for a conflicting name */ |
94c91a1f | 1158 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), |
34f3e4f2 | 1159 | name, namelen, 0); |
1160 | if (di && !IS_ERR(di)) { | |
94c91a1f | 1161 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1162 | if (ret) |
1163 | return ret; | |
34f3e4f2 | 1164 | } |
1165 | btrfs_release_path(path); | |
1166 | ||
5a1d7843 JS |
1167 | return 0; |
1168 | } | |
e02119d5 | 1169 | |
bae15d95 QW |
1170 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1171 | u32 *namelen, char **name, u64 *index, | |
1172 | u64 *parent_objectid) | |
f186373f MF |
1173 | { |
1174 | struct btrfs_inode_extref *extref; | |
1175 | ||
1176 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1177 | ||
1178 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1179 | *name = kmalloc(*namelen, GFP_NOFS); | |
1180 | if (*name == NULL) | |
1181 | return -ENOMEM; | |
1182 | ||
1183 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1184 | *namelen); | |
1185 | ||
1f250e92 FM |
1186 | if (index) |
1187 | *index = btrfs_inode_extref_index(eb, extref); | |
f186373f MF |
1188 | if (parent_objectid) |
1189 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1190 | ||
1191 | return 0; | |
1192 | } | |
1193 | ||
bae15d95 QW |
1194 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1195 | u32 *namelen, char **name, u64 *index) | |
f186373f MF |
1196 | { |
1197 | struct btrfs_inode_ref *ref; | |
1198 | ||
1199 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1200 | ||
1201 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1202 | *name = kmalloc(*namelen, GFP_NOFS); | |
1203 | if (*name == NULL) | |
1204 | return -ENOMEM; | |
1205 | ||
1206 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1207 | ||
1f250e92 FM |
1208 | if (index) |
1209 | *index = btrfs_inode_ref_index(eb, ref); | |
f186373f MF |
1210 | |
1211 | return 0; | |
1212 | } | |
1213 | ||
1f250e92 FM |
1214 | /* |
1215 | * Take an inode reference item from the log tree and iterate all names from the | |
1216 | * inode reference item in the subvolume tree with the same key (if it exists). | |
1217 | * For any name that is not in the inode reference item from the log tree, do a | |
1218 | * proper unlink of that name (that is, remove its entry from the inode | |
1219 | * reference item and both dir index keys). | |
1220 | */ | |
1221 | static int unlink_old_inode_refs(struct btrfs_trans_handle *trans, | |
1222 | struct btrfs_root *root, | |
1223 | struct btrfs_path *path, | |
1224 | struct btrfs_inode *inode, | |
1225 | struct extent_buffer *log_eb, | |
1226 | int log_slot, | |
1227 | struct btrfs_key *key) | |
1228 | { | |
1229 | int ret; | |
1230 | unsigned long ref_ptr; | |
1231 | unsigned long ref_end; | |
1232 | struct extent_buffer *eb; | |
1233 | ||
1234 | again: | |
1235 | btrfs_release_path(path); | |
1236 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1237 | if (ret > 0) { | |
1238 | ret = 0; | |
1239 | goto out; | |
1240 | } | |
1241 | if (ret < 0) | |
1242 | goto out; | |
1243 | ||
1244 | eb = path->nodes[0]; | |
1245 | ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]); | |
1246 | ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]); | |
1247 | while (ref_ptr < ref_end) { | |
1248 | char *name = NULL; | |
1249 | int namelen; | |
1250 | u64 parent_id; | |
1251 | ||
1252 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1253 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1254 | NULL, &parent_id); | |
1255 | } else { | |
1256 | parent_id = key->offset; | |
1257 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1258 | NULL); | |
1259 | } | |
1260 | if (ret) | |
1261 | goto out; | |
1262 | ||
1263 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
6ff49c6a NB |
1264 | ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot, |
1265 | parent_id, name, | |
1266 | namelen); | |
1f250e92 | 1267 | else |
9bb8407f NB |
1268 | ret = !!btrfs_find_name_in_backref(log_eb, log_slot, |
1269 | name, namelen); | |
1f250e92 FM |
1270 | |
1271 | if (!ret) { | |
1272 | struct inode *dir; | |
1273 | ||
1274 | btrfs_release_path(path); | |
1275 | dir = read_one_inode(root, parent_id); | |
1276 | if (!dir) { | |
1277 | ret = -ENOENT; | |
1278 | kfree(name); | |
1279 | goto out; | |
1280 | } | |
1281 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), | |
1282 | inode, name, namelen); | |
1283 | kfree(name); | |
1284 | iput(dir); | |
1285 | if (ret) | |
1286 | goto out; | |
1287 | goto again; | |
1288 | } | |
1289 | ||
1290 | kfree(name); | |
1291 | ref_ptr += namelen; | |
1292 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1293 | ref_ptr += sizeof(struct btrfs_inode_extref); | |
1294 | else | |
1295 | ref_ptr += sizeof(struct btrfs_inode_ref); | |
1296 | } | |
1297 | ret = 0; | |
1298 | out: | |
1299 | btrfs_release_path(path); | |
1300 | return ret; | |
1301 | } | |
1302 | ||
0d836392 FM |
1303 | static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir, |
1304 | const u8 ref_type, const char *name, | |
1305 | const int namelen) | |
1306 | { | |
1307 | struct btrfs_key key; | |
1308 | struct btrfs_path *path; | |
1309 | const u64 parent_id = btrfs_ino(BTRFS_I(dir)); | |
1310 | int ret; | |
1311 | ||
1312 | path = btrfs_alloc_path(); | |
1313 | if (!path) | |
1314 | return -ENOMEM; | |
1315 | ||
1316 | key.objectid = btrfs_ino(BTRFS_I(inode)); | |
1317 | key.type = ref_type; | |
1318 | if (key.type == BTRFS_INODE_REF_KEY) | |
1319 | key.offset = parent_id; | |
1320 | else | |
1321 | key.offset = btrfs_extref_hash(parent_id, name, namelen); | |
1322 | ||
1323 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0); | |
1324 | if (ret < 0) | |
1325 | goto out; | |
1326 | if (ret > 0) { | |
1327 | ret = 0; | |
1328 | goto out; | |
1329 | } | |
1330 | if (key.type == BTRFS_INODE_EXTREF_KEY) | |
6ff49c6a NB |
1331 | ret = !!btrfs_find_name_in_ext_backref(path->nodes[0], |
1332 | path->slots[0], parent_id, name, namelen); | |
0d836392 | 1333 | else |
9bb8407f NB |
1334 | ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0], |
1335 | name, namelen); | |
0d836392 FM |
1336 | |
1337 | out: | |
1338 | btrfs_free_path(path); | |
1339 | return ret; | |
1340 | } | |
1341 | ||
6b5fc433 FM |
1342 | static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1343 | struct inode *dir, struct inode *inode, const char *name, | |
1344 | int namelen, u64 ref_index) | |
1345 | { | |
1346 | struct btrfs_dir_item *dir_item; | |
1347 | struct btrfs_key key; | |
1348 | struct btrfs_path *path; | |
1349 | struct inode *other_inode = NULL; | |
1350 | int ret; | |
1351 | ||
1352 | path = btrfs_alloc_path(); | |
1353 | if (!path) | |
1354 | return -ENOMEM; | |
1355 | ||
1356 | dir_item = btrfs_lookup_dir_item(NULL, root, path, | |
1357 | btrfs_ino(BTRFS_I(dir)), | |
1358 | name, namelen, 0); | |
1359 | if (!dir_item) { | |
1360 | btrfs_release_path(path); | |
1361 | goto add_link; | |
1362 | } else if (IS_ERR(dir_item)) { | |
1363 | ret = PTR_ERR(dir_item); | |
1364 | goto out; | |
1365 | } | |
1366 | ||
1367 | /* | |
1368 | * Our inode's dentry collides with the dentry of another inode which is | |
1369 | * in the log but not yet processed since it has a higher inode number. | |
1370 | * So delete that other dentry. | |
1371 | */ | |
1372 | btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key); | |
1373 | btrfs_release_path(path); | |
1374 | other_inode = read_one_inode(root, key.objectid); | |
1375 | if (!other_inode) { | |
1376 | ret = -ENOENT; | |
1377 | goto out; | |
1378 | } | |
1379 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode), | |
1380 | name, namelen); | |
1381 | if (ret) | |
1382 | goto out; | |
1383 | /* | |
1384 | * If we dropped the link count to 0, bump it so that later the iput() | |
1385 | * on the inode will not free it. We will fixup the link count later. | |
1386 | */ | |
1387 | if (other_inode->i_nlink == 0) | |
1388 | inc_nlink(other_inode); | |
1389 | ||
1390 | ret = btrfs_run_delayed_items(trans); | |
1391 | if (ret) | |
1392 | goto out; | |
1393 | add_link: | |
1394 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), | |
1395 | name, namelen, 0, ref_index); | |
1396 | out: | |
1397 | iput(other_inode); | |
1398 | btrfs_free_path(path); | |
1399 | ||
1400 | return ret; | |
1401 | } | |
1402 | ||
5a1d7843 JS |
1403 | /* |
1404 | * replay one inode back reference item found in the log tree. | |
1405 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1406 | * root is the destination we are replaying into, and path is for temp | |
1407 | * use by this function. (it should be released on return). | |
1408 | */ | |
1409 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1410 | struct btrfs_root *root, | |
1411 | struct btrfs_root *log, | |
1412 | struct btrfs_path *path, | |
1413 | struct extent_buffer *eb, int slot, | |
1414 | struct btrfs_key *key) | |
1415 | { | |
03b2f08b GB |
1416 | struct inode *dir = NULL; |
1417 | struct inode *inode = NULL; | |
5a1d7843 JS |
1418 | unsigned long ref_ptr; |
1419 | unsigned long ref_end; | |
03b2f08b | 1420 | char *name = NULL; |
5a1d7843 JS |
1421 | int namelen; |
1422 | int ret; | |
1423 | int search_done = 0; | |
f186373f MF |
1424 | int log_ref_ver = 0; |
1425 | u64 parent_objectid; | |
1426 | u64 inode_objectid; | |
f46dbe3d | 1427 | u64 ref_index = 0; |
f186373f MF |
1428 | int ref_struct_size; |
1429 | ||
1430 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1431 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1432 | ||
1433 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1434 | struct btrfs_inode_extref *r; | |
1435 | ||
1436 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1437 | log_ref_ver = 1; | |
1438 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1439 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1440 | } else { | |
1441 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1442 | parent_objectid = key->offset; | |
1443 | } | |
1444 | inode_objectid = key->objectid; | |
e02119d5 | 1445 | |
5a1d7843 JS |
1446 | /* |
1447 | * it is possible that we didn't log all the parent directories | |
1448 | * for a given inode. If we don't find the dir, just don't | |
1449 | * copy the back ref in. The link count fixup code will take | |
1450 | * care of the rest | |
1451 | */ | |
f186373f | 1452 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1453 | if (!dir) { |
1454 | ret = -ENOENT; | |
1455 | goto out; | |
1456 | } | |
5a1d7843 | 1457 | |
f186373f | 1458 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1459 | if (!inode) { |
03b2f08b GB |
1460 | ret = -EIO; |
1461 | goto out; | |
5a1d7843 JS |
1462 | } |
1463 | ||
5a1d7843 | 1464 | while (ref_ptr < ref_end) { |
f186373f | 1465 | if (log_ref_ver) { |
bae15d95 QW |
1466 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, |
1467 | &ref_index, &parent_objectid); | |
f186373f MF |
1468 | /* |
1469 | * parent object can change from one array | |
1470 | * item to another. | |
1471 | */ | |
1472 | if (!dir) | |
1473 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1474 | if (!dir) { |
1475 | ret = -ENOENT; | |
1476 | goto out; | |
1477 | } | |
f186373f | 1478 | } else { |
bae15d95 QW |
1479 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, |
1480 | &ref_index); | |
f186373f MF |
1481 | } |
1482 | if (ret) | |
03b2f08b | 1483 | goto out; |
5a1d7843 JS |
1484 | |
1485 | /* if we already have a perfect match, we're done */ | |
f85b7379 DS |
1486 | if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), |
1487 | btrfs_ino(BTRFS_I(inode)), ref_index, | |
1488 | name, namelen)) { | |
5a1d7843 JS |
1489 | /* |
1490 | * look for a conflicting back reference in the | |
1491 | * metadata. if we find one we have to unlink that name | |
1492 | * of the file before we add our new link. Later on, we | |
1493 | * overwrite any existing back reference, and we don't | |
1494 | * want to create dangling pointers in the directory. | |
1495 | */ | |
1496 | ||
1497 | if (!search_done) { | |
1498 | ret = __add_inode_ref(trans, root, path, log, | |
94c91a1f | 1499 | BTRFS_I(dir), |
d75eefdf | 1500 | BTRFS_I(inode), |
f186373f MF |
1501 | inode_objectid, |
1502 | parent_objectid, | |
1503 | ref_index, name, namelen, | |
5a1d7843 | 1504 | &search_done); |
03b2f08b GB |
1505 | if (ret) { |
1506 | if (ret == 1) | |
1507 | ret = 0; | |
3650860b JB |
1508 | goto out; |
1509 | } | |
5a1d7843 JS |
1510 | } |
1511 | ||
0d836392 FM |
1512 | /* |
1513 | * If a reference item already exists for this inode | |
1514 | * with the same parent and name, but different index, | |
1515 | * drop it and the corresponding directory index entries | |
1516 | * from the parent before adding the new reference item | |
1517 | * and dir index entries, otherwise we would fail with | |
1518 | * -EEXIST returned from btrfs_add_link() below. | |
1519 | */ | |
1520 | ret = btrfs_inode_ref_exists(inode, dir, key->type, | |
1521 | name, namelen); | |
1522 | if (ret > 0) { | |
1523 | ret = btrfs_unlink_inode(trans, root, | |
1524 | BTRFS_I(dir), | |
1525 | BTRFS_I(inode), | |
1526 | name, namelen); | |
1527 | /* | |
1528 | * If we dropped the link count to 0, bump it so | |
1529 | * that later the iput() on the inode will not | |
1530 | * free it. We will fixup the link count later. | |
1531 | */ | |
1532 | if (!ret && inode->i_nlink == 0) | |
1533 | inc_nlink(inode); | |
1534 | } | |
1535 | if (ret < 0) | |
1536 | goto out; | |
1537 | ||
5a1d7843 | 1538 | /* insert our name */ |
6b5fc433 FM |
1539 | ret = add_link(trans, root, dir, inode, name, namelen, |
1540 | ref_index); | |
3650860b JB |
1541 | if (ret) |
1542 | goto out; | |
5a1d7843 | 1543 | |
9a56fcd1 | 1544 | btrfs_update_inode(trans, root, BTRFS_I(inode)); |
5a1d7843 JS |
1545 | } |
1546 | ||
f186373f | 1547 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1548 | kfree(name); |
03b2f08b | 1549 | name = NULL; |
f186373f MF |
1550 | if (log_ref_ver) { |
1551 | iput(dir); | |
1552 | dir = NULL; | |
1553 | } | |
5a1d7843 | 1554 | } |
e02119d5 | 1555 | |
1f250e92 FM |
1556 | /* |
1557 | * Before we overwrite the inode reference item in the subvolume tree | |
1558 | * with the item from the log tree, we must unlink all names from the | |
1559 | * parent directory that are in the subvolume's tree inode reference | |
1560 | * item, otherwise we end up with an inconsistent subvolume tree where | |
1561 | * dir index entries exist for a name but there is no inode reference | |
1562 | * item with the same name. | |
1563 | */ | |
1564 | ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot, | |
1565 | key); | |
1566 | if (ret) | |
1567 | goto out; | |
1568 | ||
e02119d5 CM |
1569 | /* finally write the back reference in the inode */ |
1570 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1571 | out: |
b3b4aa74 | 1572 | btrfs_release_path(path); |
03b2f08b | 1573 | kfree(name); |
e02119d5 CM |
1574 | iput(dir); |
1575 | iput(inode); | |
3650860b | 1576 | return ret; |
e02119d5 CM |
1577 | } |
1578 | ||
f186373f | 1579 | static int count_inode_extrefs(struct btrfs_root *root, |
36283658 | 1580 | struct btrfs_inode *inode, struct btrfs_path *path) |
f186373f MF |
1581 | { |
1582 | int ret = 0; | |
1583 | int name_len; | |
1584 | unsigned int nlink = 0; | |
1585 | u32 item_size; | |
1586 | u32 cur_offset = 0; | |
36283658 | 1587 | u64 inode_objectid = btrfs_ino(inode); |
f186373f MF |
1588 | u64 offset = 0; |
1589 | unsigned long ptr; | |
1590 | struct btrfs_inode_extref *extref; | |
1591 | struct extent_buffer *leaf; | |
1592 | ||
1593 | while (1) { | |
1594 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1595 | &extref, &offset); | |
1596 | if (ret) | |
1597 | break; | |
c71bf099 | 1598 | |
f186373f MF |
1599 | leaf = path->nodes[0]; |
1600 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1601 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2c2c452b | 1602 | cur_offset = 0; |
f186373f MF |
1603 | |
1604 | while (cur_offset < item_size) { | |
1605 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1606 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1607 | ||
1608 | nlink++; | |
1609 | ||
1610 | cur_offset += name_len + sizeof(*extref); | |
1611 | } | |
1612 | ||
1613 | offset++; | |
1614 | btrfs_release_path(path); | |
1615 | } | |
1616 | btrfs_release_path(path); | |
1617 | ||
2c2c452b | 1618 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1619 | return ret; |
1620 | return nlink; | |
1621 | } | |
1622 | ||
1623 | static int count_inode_refs(struct btrfs_root *root, | |
f329e319 | 1624 | struct btrfs_inode *inode, struct btrfs_path *path) |
e02119d5 | 1625 | { |
e02119d5 CM |
1626 | int ret; |
1627 | struct btrfs_key key; | |
f186373f | 1628 | unsigned int nlink = 0; |
e02119d5 CM |
1629 | unsigned long ptr; |
1630 | unsigned long ptr_end; | |
1631 | int name_len; | |
f329e319 | 1632 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1633 | |
33345d01 | 1634 | key.objectid = ino; |
e02119d5 CM |
1635 | key.type = BTRFS_INODE_REF_KEY; |
1636 | key.offset = (u64)-1; | |
1637 | ||
d397712b | 1638 | while (1) { |
e02119d5 CM |
1639 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1640 | if (ret < 0) | |
1641 | break; | |
1642 | if (ret > 0) { | |
1643 | if (path->slots[0] == 0) | |
1644 | break; | |
1645 | path->slots[0]--; | |
1646 | } | |
e93ae26f | 1647 | process_slot: |
e02119d5 CM |
1648 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1649 | path->slots[0]); | |
33345d01 | 1650 | if (key.objectid != ino || |
e02119d5 CM |
1651 | key.type != BTRFS_INODE_REF_KEY) |
1652 | break; | |
1653 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1654 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1655 | path->slots[0]); | |
d397712b | 1656 | while (ptr < ptr_end) { |
e02119d5 CM |
1657 | struct btrfs_inode_ref *ref; |
1658 | ||
1659 | ref = (struct btrfs_inode_ref *)ptr; | |
1660 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1661 | ref); | |
1662 | ptr = (unsigned long)(ref + 1) + name_len; | |
1663 | nlink++; | |
1664 | } | |
1665 | ||
1666 | if (key.offset == 0) | |
1667 | break; | |
e93ae26f FDBM |
1668 | if (path->slots[0] > 0) { |
1669 | path->slots[0]--; | |
1670 | goto process_slot; | |
1671 | } | |
e02119d5 | 1672 | key.offset--; |
b3b4aa74 | 1673 | btrfs_release_path(path); |
e02119d5 | 1674 | } |
b3b4aa74 | 1675 | btrfs_release_path(path); |
f186373f MF |
1676 | |
1677 | return nlink; | |
1678 | } | |
1679 | ||
1680 | /* | |
1681 | * There are a few corners where the link count of the file can't | |
1682 | * be properly maintained during replay. So, instead of adding | |
1683 | * lots of complexity to the log code, we just scan the backrefs | |
1684 | * for any file that has been through replay. | |
1685 | * | |
1686 | * The scan will update the link count on the inode to reflect the | |
1687 | * number of back refs found. If it goes down to zero, the iput | |
1688 | * will free the inode. | |
1689 | */ | |
1690 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1691 | struct btrfs_root *root, | |
1692 | struct inode *inode) | |
1693 | { | |
1694 | struct btrfs_path *path; | |
1695 | int ret; | |
1696 | u64 nlink = 0; | |
4a0cc7ca | 1697 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1698 | |
1699 | path = btrfs_alloc_path(); | |
1700 | if (!path) | |
1701 | return -ENOMEM; | |
1702 | ||
f329e319 | 1703 | ret = count_inode_refs(root, BTRFS_I(inode), path); |
f186373f MF |
1704 | if (ret < 0) |
1705 | goto out; | |
1706 | ||
1707 | nlink = ret; | |
1708 | ||
36283658 | 1709 | ret = count_inode_extrefs(root, BTRFS_I(inode), path); |
f186373f MF |
1710 | if (ret < 0) |
1711 | goto out; | |
1712 | ||
1713 | nlink += ret; | |
1714 | ||
1715 | ret = 0; | |
1716 | ||
e02119d5 | 1717 | if (nlink != inode->i_nlink) { |
bfe86848 | 1718 | set_nlink(inode, nlink); |
9a56fcd1 | 1719 | btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e02119d5 | 1720 | } |
8d5bf1cb | 1721 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1722 | |
c71bf099 YZ |
1723 | if (inode->i_nlink == 0) { |
1724 | if (S_ISDIR(inode->i_mode)) { | |
1725 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1726 | ino, 1); |
3650860b JB |
1727 | if (ret) |
1728 | goto out; | |
c71bf099 | 1729 | } |
ecdcf3c2 NB |
1730 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1731 | if (ret == -EEXIST) | |
1732 | ret = 0; | |
12fcfd22 | 1733 | } |
12fcfd22 | 1734 | |
f186373f MF |
1735 | out: |
1736 | btrfs_free_path(path); | |
1737 | return ret; | |
e02119d5 CM |
1738 | } |
1739 | ||
1740 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1741 | struct btrfs_root *root, | |
1742 | struct btrfs_path *path) | |
1743 | { | |
1744 | int ret; | |
1745 | struct btrfs_key key; | |
1746 | struct inode *inode; | |
1747 | ||
1748 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1749 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1750 | key.offset = (u64)-1; | |
d397712b | 1751 | while (1) { |
e02119d5 CM |
1752 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1753 | if (ret < 0) | |
1754 | break; | |
1755 | ||
1756 | if (ret == 1) { | |
1757 | if (path->slots[0] == 0) | |
1758 | break; | |
1759 | path->slots[0]--; | |
1760 | } | |
1761 | ||
1762 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1763 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1764 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1765 | break; | |
1766 | ||
1767 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1768 | if (ret) |
1769 | goto out; | |
e02119d5 | 1770 | |
b3b4aa74 | 1771 | btrfs_release_path(path); |
e02119d5 | 1772 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1773 | if (!inode) |
1774 | return -EIO; | |
e02119d5 CM |
1775 | |
1776 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1777 | iput(inode); |
3650860b JB |
1778 | if (ret) |
1779 | goto out; | |
e02119d5 | 1780 | |
12fcfd22 CM |
1781 | /* |
1782 | * fixup on a directory may create new entries, | |
1783 | * make sure we always look for the highset possible | |
1784 | * offset | |
1785 | */ | |
1786 | key.offset = (u64)-1; | |
e02119d5 | 1787 | } |
65a246c5 TI |
1788 | ret = 0; |
1789 | out: | |
b3b4aa74 | 1790 | btrfs_release_path(path); |
65a246c5 | 1791 | return ret; |
e02119d5 CM |
1792 | } |
1793 | ||
1794 | ||
1795 | /* | |
1796 | * record a given inode in the fixup dir so we can check its link | |
1797 | * count when replay is done. The link count is incremented here | |
1798 | * so the inode won't go away until we check it | |
1799 | */ | |
1800 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1801 | struct btrfs_root *root, | |
1802 | struct btrfs_path *path, | |
1803 | u64 objectid) | |
1804 | { | |
1805 | struct btrfs_key key; | |
1806 | int ret = 0; | |
1807 | struct inode *inode; | |
1808 | ||
1809 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1810 | if (!inode) |
1811 | return -EIO; | |
e02119d5 CM |
1812 | |
1813 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1814 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1815 | key.offset = objectid; |
1816 | ||
1817 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1818 | ||
b3b4aa74 | 1819 | btrfs_release_path(path); |
e02119d5 | 1820 | if (ret == 0) { |
9bf7a489 JB |
1821 | if (!inode->i_nlink) |
1822 | set_nlink(inode, 1); | |
1823 | else | |
8b558c5f | 1824 | inc_nlink(inode); |
9a56fcd1 | 1825 | ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); |
e02119d5 CM |
1826 | } else if (ret == -EEXIST) { |
1827 | ret = 0; | |
1828 | } else { | |
3650860b | 1829 | BUG(); /* Logic Error */ |
e02119d5 CM |
1830 | } |
1831 | iput(inode); | |
1832 | ||
1833 | return ret; | |
1834 | } | |
1835 | ||
1836 | /* | |
1837 | * when replaying the log for a directory, we only insert names | |
1838 | * for inodes that actually exist. This means an fsync on a directory | |
1839 | * does not implicitly fsync all the new files in it | |
1840 | */ | |
1841 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1842 | struct btrfs_root *root, | |
e02119d5 | 1843 | u64 dirid, u64 index, |
60d53eb3 | 1844 | char *name, int name_len, |
e02119d5 CM |
1845 | struct btrfs_key *location) |
1846 | { | |
1847 | struct inode *inode; | |
1848 | struct inode *dir; | |
1849 | int ret; | |
1850 | ||
1851 | inode = read_one_inode(root, location->objectid); | |
1852 | if (!inode) | |
1853 | return -ENOENT; | |
1854 | ||
1855 | dir = read_one_inode(root, dirid); | |
1856 | if (!dir) { | |
1857 | iput(inode); | |
1858 | return -EIO; | |
1859 | } | |
d555438b | 1860 | |
db0a669f NB |
1861 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, |
1862 | name_len, 1, index); | |
e02119d5 CM |
1863 | |
1864 | /* FIXME, put inode into FIXUP list */ | |
1865 | ||
1866 | iput(inode); | |
1867 | iput(dir); | |
1868 | return ret; | |
1869 | } | |
1870 | ||
1871 | /* | |
1872 | * take a single entry in a log directory item and replay it into | |
1873 | * the subvolume. | |
1874 | * | |
1875 | * if a conflicting item exists in the subdirectory already, | |
1876 | * the inode it points to is unlinked and put into the link count | |
1877 | * fix up tree. | |
1878 | * | |
1879 | * If a name from the log points to a file or directory that does | |
1880 | * not exist in the FS, it is skipped. fsyncs on directories | |
1881 | * do not force down inodes inside that directory, just changes to the | |
1882 | * names or unlinks in a directory. | |
bb53eda9 FM |
1883 | * |
1884 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1885 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1886 | */ |
1887 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1888 | struct btrfs_root *root, | |
1889 | struct btrfs_path *path, | |
1890 | struct extent_buffer *eb, | |
1891 | struct btrfs_dir_item *di, | |
1892 | struct btrfs_key *key) | |
1893 | { | |
1894 | char *name; | |
1895 | int name_len; | |
1896 | struct btrfs_dir_item *dst_di; | |
1897 | struct btrfs_key found_key; | |
1898 | struct btrfs_key log_key; | |
1899 | struct inode *dir; | |
e02119d5 | 1900 | u8 log_type; |
4bef0848 | 1901 | int exists; |
3650860b | 1902 | int ret = 0; |
d555438b | 1903 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1904 | bool name_added = false; |
e02119d5 CM |
1905 | |
1906 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1907 | if (!dir) |
1908 | return -EIO; | |
e02119d5 CM |
1909 | |
1910 | name_len = btrfs_dir_name_len(eb, di); | |
1911 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1912 | if (!name) { |
1913 | ret = -ENOMEM; | |
1914 | goto out; | |
1915 | } | |
2a29edc6 | 1916 | |
e02119d5 CM |
1917 | log_type = btrfs_dir_type(eb, di); |
1918 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1919 | name_len); | |
1920 | ||
1921 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1922 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1923 | if (exists == 0) | |
1924 | exists = 1; | |
1925 | else | |
1926 | exists = 0; | |
b3b4aa74 | 1927 | btrfs_release_path(path); |
4bef0848 | 1928 | |
e02119d5 CM |
1929 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1930 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1931 | name, name_len, 1); | |
d397712b | 1932 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1933 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1934 | key->objectid, | |
1935 | key->offset, name, | |
1936 | name_len, 1); | |
1937 | } else { | |
3650860b JB |
1938 | /* Corruption */ |
1939 | ret = -EINVAL; | |
1940 | goto out; | |
e02119d5 | 1941 | } |
c704005d | 1942 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1943 | /* we need a sequence number to insert, so we only |
1944 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1945 | */ | |
1946 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1947 | goto out; | |
1948 | goto insert; | |
1949 | } | |
1950 | ||
1951 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1952 | /* the existing item matches the logged item */ | |
1953 | if (found_key.objectid == log_key.objectid && | |
1954 | found_key.type == log_key.type && | |
1955 | found_key.offset == log_key.offset && | |
1956 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1957 | update_size = false; |
e02119d5 CM |
1958 | goto out; |
1959 | } | |
1960 | ||
1961 | /* | |
1962 | * don't drop the conflicting directory entry if the inode | |
1963 | * for the new entry doesn't exist | |
1964 | */ | |
4bef0848 | 1965 | if (!exists) |
e02119d5 CM |
1966 | goto out; |
1967 | ||
207e7d92 | 1968 | ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di); |
3650860b JB |
1969 | if (ret) |
1970 | goto out; | |
e02119d5 CM |
1971 | |
1972 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1973 | goto insert; | |
1974 | out: | |
b3b4aa74 | 1975 | btrfs_release_path(path); |
d555438b | 1976 | if (!ret && update_size) { |
6ef06d27 | 1977 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2); |
9a56fcd1 | 1978 | ret = btrfs_update_inode(trans, root, BTRFS_I(dir)); |
d555438b | 1979 | } |
e02119d5 CM |
1980 | kfree(name); |
1981 | iput(dir); | |
bb53eda9 FM |
1982 | if (!ret && name_added) |
1983 | ret = 1; | |
3650860b | 1984 | return ret; |
e02119d5 CM |
1985 | |
1986 | insert: | |
725af92a NB |
1987 | /* |
1988 | * Check if the inode reference exists in the log for the given name, | |
1989 | * inode and parent inode | |
1990 | */ | |
1991 | found_key.objectid = log_key.objectid; | |
1992 | found_key.type = BTRFS_INODE_REF_KEY; | |
1993 | found_key.offset = key->objectid; | |
1994 | ret = backref_in_log(root->log_root, &found_key, 0, name, name_len); | |
1995 | if (ret < 0) { | |
1996 | goto out; | |
1997 | } else if (ret) { | |
1998 | /* The dentry will be added later. */ | |
1999 | ret = 0; | |
2000 | update_size = false; | |
2001 | goto out; | |
2002 | } | |
2003 | ||
2004 | found_key.objectid = log_key.objectid; | |
2005 | found_key.type = BTRFS_INODE_EXTREF_KEY; | |
2006 | found_key.offset = key->objectid; | |
2007 | ret = backref_in_log(root->log_root, &found_key, key->objectid, name, | |
2008 | name_len); | |
2009 | if (ret < 0) { | |
2010 | goto out; | |
2011 | } else if (ret) { | |
df8d116f FM |
2012 | /* The dentry will be added later. */ |
2013 | ret = 0; | |
2014 | update_size = false; | |
2015 | goto out; | |
2016 | } | |
b3b4aa74 | 2017 | btrfs_release_path(path); |
60d53eb3 Z |
2018 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
2019 | name, name_len, &log_key); | |
df8d116f | 2020 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 2021 | goto out; |
bb53eda9 FM |
2022 | if (!ret) |
2023 | name_added = true; | |
d555438b | 2024 | update_size = false; |
3650860b | 2025 | ret = 0; |
e02119d5 CM |
2026 | goto out; |
2027 | } | |
2028 | ||
2029 | /* | |
2030 | * find all the names in a directory item and reconcile them into | |
2031 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
2032 | * one name in a directory item, but the same code gets used for | |
2033 | * both directory index types | |
2034 | */ | |
2035 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
2036 | struct btrfs_root *root, | |
2037 | struct btrfs_path *path, | |
2038 | struct extent_buffer *eb, int slot, | |
2039 | struct btrfs_key *key) | |
2040 | { | |
bb53eda9 | 2041 | int ret = 0; |
e02119d5 CM |
2042 | u32 item_size = btrfs_item_size_nr(eb, slot); |
2043 | struct btrfs_dir_item *di; | |
2044 | int name_len; | |
2045 | unsigned long ptr; | |
2046 | unsigned long ptr_end; | |
bb53eda9 | 2047 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
2048 | |
2049 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2050 | ptr_end = ptr + item_size; | |
d397712b | 2051 | while (ptr < ptr_end) { |
e02119d5 CM |
2052 | di = (struct btrfs_dir_item *)ptr; |
2053 | name_len = btrfs_dir_name_len(eb, di); | |
2054 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
2055 | if (ret < 0) |
2056 | break; | |
e02119d5 CM |
2057 | ptr = (unsigned long)(di + 1); |
2058 | ptr += name_len; | |
bb53eda9 FM |
2059 | |
2060 | /* | |
2061 | * If this entry refers to a non-directory (directories can not | |
2062 | * have a link count > 1) and it was added in the transaction | |
2063 | * that was not committed, make sure we fixup the link count of | |
2064 | * the inode it the entry points to. Otherwise something like | |
2065 | * the following would result in a directory pointing to an | |
2066 | * inode with a wrong link that does not account for this dir | |
2067 | * entry: | |
2068 | * | |
2069 | * mkdir testdir | |
2070 | * touch testdir/foo | |
2071 | * touch testdir/bar | |
2072 | * sync | |
2073 | * | |
2074 | * ln testdir/bar testdir/bar_link | |
2075 | * ln testdir/foo testdir/foo_link | |
2076 | * xfs_io -c "fsync" testdir/bar | |
2077 | * | |
2078 | * <power failure> | |
2079 | * | |
2080 | * mount fs, log replay happens | |
2081 | * | |
2082 | * File foo would remain with a link count of 1 when it has two | |
2083 | * entries pointing to it in the directory testdir. This would | |
2084 | * make it impossible to ever delete the parent directory has | |
2085 | * it would result in stale dentries that can never be deleted. | |
2086 | */ | |
2087 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
2088 | struct btrfs_key di_key; | |
2089 | ||
2090 | if (!fixup_path) { | |
2091 | fixup_path = btrfs_alloc_path(); | |
2092 | if (!fixup_path) { | |
2093 | ret = -ENOMEM; | |
2094 | break; | |
2095 | } | |
2096 | } | |
2097 | ||
2098 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2099 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
2100 | di_key.objectid); | |
2101 | if (ret) | |
2102 | break; | |
2103 | } | |
2104 | ret = 0; | |
e02119d5 | 2105 | } |
bb53eda9 FM |
2106 | btrfs_free_path(fixup_path); |
2107 | return ret; | |
e02119d5 CM |
2108 | } |
2109 | ||
2110 | /* | |
2111 | * directory replay has two parts. There are the standard directory | |
2112 | * items in the log copied from the subvolume, and range items | |
2113 | * created in the log while the subvolume was logged. | |
2114 | * | |
2115 | * The range items tell us which parts of the key space the log | |
2116 | * is authoritative for. During replay, if a key in the subvolume | |
2117 | * directory is in a logged range item, but not actually in the log | |
2118 | * that means it was deleted from the directory before the fsync | |
2119 | * and should be removed. | |
2120 | */ | |
2121 | static noinline int find_dir_range(struct btrfs_root *root, | |
2122 | struct btrfs_path *path, | |
2123 | u64 dirid, int key_type, | |
2124 | u64 *start_ret, u64 *end_ret) | |
2125 | { | |
2126 | struct btrfs_key key; | |
2127 | u64 found_end; | |
2128 | struct btrfs_dir_log_item *item; | |
2129 | int ret; | |
2130 | int nritems; | |
2131 | ||
2132 | if (*start_ret == (u64)-1) | |
2133 | return 1; | |
2134 | ||
2135 | key.objectid = dirid; | |
2136 | key.type = key_type; | |
2137 | key.offset = *start_ret; | |
2138 | ||
2139 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2140 | if (ret < 0) | |
2141 | goto out; | |
2142 | if (ret > 0) { | |
2143 | if (path->slots[0] == 0) | |
2144 | goto out; | |
2145 | path->slots[0]--; | |
2146 | } | |
2147 | if (ret != 0) | |
2148 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2149 | ||
2150 | if (key.type != key_type || key.objectid != dirid) { | |
2151 | ret = 1; | |
2152 | goto next; | |
2153 | } | |
2154 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2155 | struct btrfs_dir_log_item); | |
2156 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2157 | ||
2158 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
2159 | ret = 0; | |
2160 | *start_ret = key.offset; | |
2161 | *end_ret = found_end; | |
2162 | goto out; | |
2163 | } | |
2164 | ret = 1; | |
2165 | next: | |
2166 | /* check the next slot in the tree to see if it is a valid item */ | |
2167 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 2168 | path->slots[0]++; |
e02119d5 CM |
2169 | if (path->slots[0] >= nritems) { |
2170 | ret = btrfs_next_leaf(root, path); | |
2171 | if (ret) | |
2172 | goto out; | |
e02119d5 CM |
2173 | } |
2174 | ||
2175 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2176 | ||
2177 | if (key.type != key_type || key.objectid != dirid) { | |
2178 | ret = 1; | |
2179 | goto out; | |
2180 | } | |
2181 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2182 | struct btrfs_dir_log_item); | |
2183 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2184 | *start_ret = key.offset; | |
2185 | *end_ret = found_end; | |
2186 | ret = 0; | |
2187 | out: | |
b3b4aa74 | 2188 | btrfs_release_path(path); |
e02119d5 CM |
2189 | return ret; |
2190 | } | |
2191 | ||
2192 | /* | |
2193 | * this looks for a given directory item in the log. If the directory | |
2194 | * item is not in the log, the item is removed and the inode it points | |
2195 | * to is unlinked | |
2196 | */ | |
2197 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
2198 | struct btrfs_root *root, | |
2199 | struct btrfs_root *log, | |
2200 | struct btrfs_path *path, | |
2201 | struct btrfs_path *log_path, | |
2202 | struct inode *dir, | |
2203 | struct btrfs_key *dir_key) | |
2204 | { | |
2205 | int ret; | |
2206 | struct extent_buffer *eb; | |
2207 | int slot; | |
2208 | u32 item_size; | |
2209 | struct btrfs_dir_item *di; | |
2210 | struct btrfs_dir_item *log_di; | |
2211 | int name_len; | |
2212 | unsigned long ptr; | |
2213 | unsigned long ptr_end; | |
2214 | char *name; | |
2215 | struct inode *inode; | |
2216 | struct btrfs_key location; | |
2217 | ||
2218 | again: | |
2219 | eb = path->nodes[0]; | |
2220 | slot = path->slots[0]; | |
2221 | item_size = btrfs_item_size_nr(eb, slot); | |
2222 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2223 | ptr_end = ptr + item_size; | |
d397712b | 2224 | while (ptr < ptr_end) { |
e02119d5 CM |
2225 | di = (struct btrfs_dir_item *)ptr; |
2226 | name_len = btrfs_dir_name_len(eb, di); | |
2227 | name = kmalloc(name_len, GFP_NOFS); | |
2228 | if (!name) { | |
2229 | ret = -ENOMEM; | |
2230 | goto out; | |
2231 | } | |
2232 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2233 | name_len); | |
2234 | log_di = NULL; | |
12fcfd22 | 2235 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2236 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
2237 | dir_key->objectid, | |
2238 | name, name_len, 0); | |
12fcfd22 | 2239 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
2240 | log_di = btrfs_lookup_dir_index_item(trans, log, |
2241 | log_path, | |
2242 | dir_key->objectid, | |
2243 | dir_key->offset, | |
2244 | name, name_len, 0); | |
2245 | } | |
8d9e220c | 2246 | if (!log_di || log_di == ERR_PTR(-ENOENT)) { |
e02119d5 | 2247 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
2248 | btrfs_release_path(path); |
2249 | btrfs_release_path(log_path); | |
e02119d5 | 2250 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
2251 | if (!inode) { |
2252 | kfree(name); | |
2253 | return -EIO; | |
2254 | } | |
e02119d5 CM |
2255 | |
2256 | ret = link_to_fixup_dir(trans, root, | |
2257 | path, location.objectid); | |
3650860b JB |
2258 | if (ret) { |
2259 | kfree(name); | |
2260 | iput(inode); | |
2261 | goto out; | |
2262 | } | |
2263 | ||
8b558c5f | 2264 | inc_nlink(inode); |
4ec5934e NB |
2265 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
2266 | BTRFS_I(inode), name, name_len); | |
3650860b | 2267 | if (!ret) |
e5c304e6 | 2268 | ret = btrfs_run_delayed_items(trans); |
e02119d5 CM |
2269 | kfree(name); |
2270 | iput(inode); | |
3650860b JB |
2271 | if (ret) |
2272 | goto out; | |
e02119d5 CM |
2273 | |
2274 | /* there might still be more names under this key | |
2275 | * check and repeat if required | |
2276 | */ | |
2277 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2278 | 0, 0); | |
2279 | if (ret == 0) | |
2280 | goto again; | |
2281 | ret = 0; | |
2282 | goto out; | |
269d040f FDBM |
2283 | } else if (IS_ERR(log_di)) { |
2284 | kfree(name); | |
2285 | return PTR_ERR(log_di); | |
e02119d5 | 2286 | } |
b3b4aa74 | 2287 | btrfs_release_path(log_path); |
e02119d5 CM |
2288 | kfree(name); |
2289 | ||
2290 | ptr = (unsigned long)(di + 1); | |
2291 | ptr += name_len; | |
2292 | } | |
2293 | ret = 0; | |
2294 | out: | |
b3b4aa74 DS |
2295 | btrfs_release_path(path); |
2296 | btrfs_release_path(log_path); | |
e02119d5 CM |
2297 | return ret; |
2298 | } | |
2299 | ||
4f764e51 FM |
2300 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2301 | struct btrfs_root *root, | |
2302 | struct btrfs_root *log, | |
2303 | struct btrfs_path *path, | |
2304 | const u64 ino) | |
2305 | { | |
2306 | struct btrfs_key search_key; | |
2307 | struct btrfs_path *log_path; | |
2308 | int i; | |
2309 | int nritems; | |
2310 | int ret; | |
2311 | ||
2312 | log_path = btrfs_alloc_path(); | |
2313 | if (!log_path) | |
2314 | return -ENOMEM; | |
2315 | ||
2316 | search_key.objectid = ino; | |
2317 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2318 | search_key.offset = 0; | |
2319 | again: | |
2320 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2321 | if (ret < 0) | |
2322 | goto out; | |
2323 | process_leaf: | |
2324 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2325 | for (i = path->slots[0]; i < nritems; i++) { | |
2326 | struct btrfs_key key; | |
2327 | struct btrfs_dir_item *di; | |
2328 | struct btrfs_dir_item *log_di; | |
2329 | u32 total_size; | |
2330 | u32 cur; | |
2331 | ||
2332 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2333 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2334 | ret = 0; | |
2335 | goto out; | |
2336 | } | |
2337 | ||
2338 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2339 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2340 | cur = 0; | |
2341 | while (cur < total_size) { | |
2342 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2343 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2344 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2345 | char *name; | |
2346 | ||
2347 | name = kmalloc(name_len, GFP_NOFS); | |
2348 | if (!name) { | |
2349 | ret = -ENOMEM; | |
2350 | goto out; | |
2351 | } | |
2352 | read_extent_buffer(path->nodes[0], name, | |
2353 | (unsigned long)(di + 1), name_len); | |
2354 | ||
2355 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2356 | name, name_len, 0); | |
2357 | btrfs_release_path(log_path); | |
2358 | if (!log_di) { | |
2359 | /* Doesn't exist in log tree, so delete it. */ | |
2360 | btrfs_release_path(path); | |
2361 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2362 | name, name_len, -1); | |
2363 | kfree(name); | |
2364 | if (IS_ERR(di)) { | |
2365 | ret = PTR_ERR(di); | |
2366 | goto out; | |
2367 | } | |
2368 | ASSERT(di); | |
2369 | ret = btrfs_delete_one_dir_name(trans, root, | |
2370 | path, di); | |
2371 | if (ret) | |
2372 | goto out; | |
2373 | btrfs_release_path(path); | |
2374 | search_key = key; | |
2375 | goto again; | |
2376 | } | |
2377 | kfree(name); | |
2378 | if (IS_ERR(log_di)) { | |
2379 | ret = PTR_ERR(log_di); | |
2380 | goto out; | |
2381 | } | |
2382 | cur += this_len; | |
2383 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2384 | } | |
2385 | } | |
2386 | ret = btrfs_next_leaf(root, path); | |
2387 | if (ret > 0) | |
2388 | ret = 0; | |
2389 | else if (ret == 0) | |
2390 | goto process_leaf; | |
2391 | out: | |
2392 | btrfs_free_path(log_path); | |
2393 | btrfs_release_path(path); | |
2394 | return ret; | |
2395 | } | |
2396 | ||
2397 | ||
e02119d5 CM |
2398 | /* |
2399 | * deletion replay happens before we copy any new directory items | |
2400 | * out of the log or out of backreferences from inodes. It | |
2401 | * scans the log to find ranges of keys that log is authoritative for, | |
2402 | * and then scans the directory to find items in those ranges that are | |
2403 | * not present in the log. | |
2404 | * | |
2405 | * Anything we don't find in the log is unlinked and removed from the | |
2406 | * directory. | |
2407 | */ | |
2408 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2409 | struct btrfs_root *root, | |
2410 | struct btrfs_root *log, | |
2411 | struct btrfs_path *path, | |
12fcfd22 | 2412 | u64 dirid, int del_all) |
e02119d5 CM |
2413 | { |
2414 | u64 range_start; | |
2415 | u64 range_end; | |
2416 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2417 | int ret = 0; | |
2418 | struct btrfs_key dir_key; | |
2419 | struct btrfs_key found_key; | |
2420 | struct btrfs_path *log_path; | |
2421 | struct inode *dir; | |
2422 | ||
2423 | dir_key.objectid = dirid; | |
2424 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2425 | log_path = btrfs_alloc_path(); | |
2426 | if (!log_path) | |
2427 | return -ENOMEM; | |
2428 | ||
2429 | dir = read_one_inode(root, dirid); | |
2430 | /* it isn't an error if the inode isn't there, that can happen | |
2431 | * because we replay the deletes before we copy in the inode item | |
2432 | * from the log | |
2433 | */ | |
2434 | if (!dir) { | |
2435 | btrfs_free_path(log_path); | |
2436 | return 0; | |
2437 | } | |
2438 | again: | |
2439 | range_start = 0; | |
2440 | range_end = 0; | |
d397712b | 2441 | while (1) { |
12fcfd22 CM |
2442 | if (del_all) |
2443 | range_end = (u64)-1; | |
2444 | else { | |
2445 | ret = find_dir_range(log, path, dirid, key_type, | |
2446 | &range_start, &range_end); | |
2447 | if (ret != 0) | |
2448 | break; | |
2449 | } | |
e02119d5 CM |
2450 | |
2451 | dir_key.offset = range_start; | |
d397712b | 2452 | while (1) { |
e02119d5 CM |
2453 | int nritems; |
2454 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2455 | 0, 0); | |
2456 | if (ret < 0) | |
2457 | goto out; | |
2458 | ||
2459 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2460 | if (path->slots[0] >= nritems) { | |
2461 | ret = btrfs_next_leaf(root, path); | |
b98def7c | 2462 | if (ret == 1) |
e02119d5 | 2463 | break; |
b98def7c LB |
2464 | else if (ret < 0) |
2465 | goto out; | |
e02119d5 CM |
2466 | } |
2467 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2468 | path->slots[0]); | |
2469 | if (found_key.objectid != dirid || | |
2470 | found_key.type != dir_key.type) | |
2471 | goto next_type; | |
2472 | ||
2473 | if (found_key.offset > range_end) | |
2474 | break; | |
2475 | ||
2476 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2477 | log_path, dir, |
2478 | &found_key); | |
3650860b JB |
2479 | if (ret) |
2480 | goto out; | |
e02119d5 CM |
2481 | if (found_key.offset == (u64)-1) |
2482 | break; | |
2483 | dir_key.offset = found_key.offset + 1; | |
2484 | } | |
b3b4aa74 | 2485 | btrfs_release_path(path); |
e02119d5 CM |
2486 | if (range_end == (u64)-1) |
2487 | break; | |
2488 | range_start = range_end + 1; | |
2489 | } | |
2490 | ||
2491 | next_type: | |
2492 | ret = 0; | |
2493 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2494 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2495 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2496 | btrfs_release_path(path); |
e02119d5 CM |
2497 | goto again; |
2498 | } | |
2499 | out: | |
b3b4aa74 | 2500 | btrfs_release_path(path); |
e02119d5 CM |
2501 | btrfs_free_path(log_path); |
2502 | iput(dir); | |
2503 | return ret; | |
2504 | } | |
2505 | ||
2506 | /* | |
2507 | * the process_func used to replay items from the log tree. This | |
2508 | * gets called in two different stages. The first stage just looks | |
2509 | * for inodes and makes sure they are all copied into the subvolume. | |
2510 | * | |
2511 | * The second stage copies all the other item types from the log into | |
2512 | * the subvolume. The two stage approach is slower, but gets rid of | |
2513 | * lots of complexity around inodes referencing other inodes that exist | |
2514 | * only in the log (references come from either directory items or inode | |
2515 | * back refs). | |
2516 | */ | |
2517 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 2518 | struct walk_control *wc, u64 gen, int level) |
e02119d5 CM |
2519 | { |
2520 | int nritems; | |
2521 | struct btrfs_path *path; | |
2522 | struct btrfs_root *root = wc->replay_dest; | |
2523 | struct btrfs_key key; | |
e02119d5 CM |
2524 | int i; |
2525 | int ret; | |
2526 | ||
581c1760 | 2527 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
018642a1 TI |
2528 | if (ret) |
2529 | return ret; | |
e02119d5 CM |
2530 | |
2531 | level = btrfs_header_level(eb); | |
2532 | ||
2533 | if (level != 0) | |
2534 | return 0; | |
2535 | ||
2536 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2537 | if (!path) |
2538 | return -ENOMEM; | |
e02119d5 CM |
2539 | |
2540 | nritems = btrfs_header_nritems(eb); | |
2541 | for (i = 0; i < nritems; i++) { | |
2542 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2543 | |
2544 | /* inode keys are done during the first stage */ | |
2545 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2546 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2547 | struct btrfs_inode_item *inode_item; |
2548 | u32 mode; | |
2549 | ||
2550 | inode_item = btrfs_item_ptr(eb, i, | |
2551 | struct btrfs_inode_item); | |
f2d72f42 FM |
2552 | /* |
2553 | * If we have a tmpfile (O_TMPFILE) that got fsync'ed | |
2554 | * and never got linked before the fsync, skip it, as | |
2555 | * replaying it is pointless since it would be deleted | |
2556 | * later. We skip logging tmpfiles, but it's always | |
2557 | * possible we are replaying a log created with a kernel | |
2558 | * that used to log tmpfiles. | |
2559 | */ | |
2560 | if (btrfs_inode_nlink(eb, inode_item) == 0) { | |
2561 | wc->ignore_cur_inode = true; | |
2562 | continue; | |
2563 | } else { | |
2564 | wc->ignore_cur_inode = false; | |
2565 | } | |
4f764e51 FM |
2566 | ret = replay_xattr_deletes(wc->trans, root, log, |
2567 | path, key.objectid); | |
2568 | if (ret) | |
2569 | break; | |
e02119d5 CM |
2570 | mode = btrfs_inode_mode(eb, inode_item); |
2571 | if (S_ISDIR(mode)) { | |
2572 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2573 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2574 | if (ret) |
2575 | break; | |
e02119d5 CM |
2576 | } |
2577 | ret = overwrite_item(wc->trans, root, path, | |
2578 | eb, i, &key); | |
b50c6e25 JB |
2579 | if (ret) |
2580 | break; | |
e02119d5 | 2581 | |
471d557a FM |
2582 | /* |
2583 | * Before replaying extents, truncate the inode to its | |
2584 | * size. We need to do it now and not after log replay | |
2585 | * because before an fsync we can have prealloc extents | |
2586 | * added beyond the inode's i_size. If we did it after, | |
2587 | * through orphan cleanup for example, we would drop | |
2588 | * those prealloc extents just after replaying them. | |
e02119d5 CM |
2589 | */ |
2590 | if (S_ISREG(mode)) { | |
5893dfb9 | 2591 | struct btrfs_drop_extents_args drop_args = { 0 }; |
471d557a FM |
2592 | struct inode *inode; |
2593 | u64 from; | |
2594 | ||
2595 | inode = read_one_inode(root, key.objectid); | |
2596 | if (!inode) { | |
2597 | ret = -EIO; | |
2598 | break; | |
2599 | } | |
2600 | from = ALIGN(i_size_read(inode), | |
2601 | root->fs_info->sectorsize); | |
5893dfb9 FM |
2602 | drop_args.start = from; |
2603 | drop_args.end = (u64)-1; | |
2604 | drop_args.drop_cache = true; | |
2605 | ret = btrfs_drop_extents(wc->trans, root, | |
2606 | BTRFS_I(inode), | |
2607 | &drop_args); | |
471d557a | 2608 | if (!ret) { |
2766ff61 FM |
2609 | inode_sub_bytes(inode, |
2610 | drop_args.bytes_found); | |
f2d72f42 | 2611 | /* Update the inode's nbytes. */ |
471d557a | 2612 | ret = btrfs_update_inode(wc->trans, |
9a56fcd1 | 2613 | root, BTRFS_I(inode)); |
471d557a FM |
2614 | } |
2615 | iput(inode); | |
b50c6e25 JB |
2616 | if (ret) |
2617 | break; | |
e02119d5 | 2618 | } |
c71bf099 | 2619 | |
e02119d5 CM |
2620 | ret = link_to_fixup_dir(wc->trans, root, |
2621 | path, key.objectid); | |
b50c6e25 JB |
2622 | if (ret) |
2623 | break; | |
e02119d5 | 2624 | } |
dd8e7217 | 2625 | |
f2d72f42 FM |
2626 | if (wc->ignore_cur_inode) |
2627 | continue; | |
2628 | ||
dd8e7217 JB |
2629 | if (key.type == BTRFS_DIR_INDEX_KEY && |
2630 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2631 | ret = replay_one_dir_item(wc->trans, root, path, | |
2632 | eb, i, &key); | |
2633 | if (ret) | |
2634 | break; | |
2635 | } | |
2636 | ||
e02119d5 CM |
2637 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2638 | continue; | |
2639 | ||
2640 | /* these keys are simply copied */ | |
2641 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2642 | ret = overwrite_item(wc->trans, root, path, | |
2643 | eb, i, &key); | |
b50c6e25 JB |
2644 | if (ret) |
2645 | break; | |
2da1c669 LB |
2646 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2647 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2648 | ret = add_inode_ref(wc->trans, root, log, path, |
2649 | eb, i, &key); | |
b50c6e25 JB |
2650 | if (ret && ret != -ENOENT) |
2651 | break; | |
2652 | ret = 0; | |
e02119d5 CM |
2653 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2654 | ret = replay_one_extent(wc->trans, root, path, | |
2655 | eb, i, &key); | |
b50c6e25 JB |
2656 | if (ret) |
2657 | break; | |
dd8e7217 | 2658 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2659 | ret = replay_one_dir_item(wc->trans, root, path, |
2660 | eb, i, &key); | |
b50c6e25 JB |
2661 | if (ret) |
2662 | break; | |
e02119d5 CM |
2663 | } |
2664 | } | |
2665 | btrfs_free_path(path); | |
b50c6e25 | 2666 | return ret; |
e02119d5 CM |
2667 | } |
2668 | ||
6787bb9f NB |
2669 | /* |
2670 | * Correctly adjust the reserved bytes occupied by a log tree extent buffer | |
2671 | */ | |
2672 | static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start) | |
2673 | { | |
2674 | struct btrfs_block_group *cache; | |
2675 | ||
2676 | cache = btrfs_lookup_block_group(fs_info, start); | |
2677 | if (!cache) { | |
2678 | btrfs_err(fs_info, "unable to find block group for %llu", start); | |
2679 | return; | |
2680 | } | |
2681 | ||
2682 | spin_lock(&cache->space_info->lock); | |
2683 | spin_lock(&cache->lock); | |
2684 | cache->reserved -= fs_info->nodesize; | |
2685 | cache->space_info->bytes_reserved -= fs_info->nodesize; | |
2686 | spin_unlock(&cache->lock); | |
2687 | spin_unlock(&cache->space_info->lock); | |
2688 | ||
2689 | btrfs_put_block_group(cache); | |
2690 | } | |
2691 | ||
d397712b | 2692 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2693 | struct btrfs_root *root, |
2694 | struct btrfs_path *path, int *level, | |
2695 | struct walk_control *wc) | |
2696 | { | |
0b246afa | 2697 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2698 | u64 bytenr; |
2699 | u64 ptr_gen; | |
2700 | struct extent_buffer *next; | |
2701 | struct extent_buffer *cur; | |
e02119d5 CM |
2702 | u32 blocksize; |
2703 | int ret = 0; | |
2704 | ||
d397712b | 2705 | while (*level > 0) { |
581c1760 QW |
2706 | struct btrfs_key first_key; |
2707 | ||
e02119d5 CM |
2708 | cur = path->nodes[*level]; |
2709 | ||
fae7f21c | 2710 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2711 | |
2712 | if (path->slots[*level] >= | |
2713 | btrfs_header_nritems(cur)) | |
2714 | break; | |
2715 | ||
2716 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2717 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
581c1760 | 2718 | btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]); |
0b246afa | 2719 | blocksize = fs_info->nodesize; |
e02119d5 | 2720 | |
3fbaf258 JB |
2721 | next = btrfs_find_create_tree_block(fs_info, bytenr, |
2722 | btrfs_header_owner(cur), | |
2723 | *level - 1); | |
c871b0f2 LB |
2724 | if (IS_ERR(next)) |
2725 | return PTR_ERR(next); | |
e02119d5 | 2726 | |
e02119d5 | 2727 | if (*level == 1) { |
581c1760 QW |
2728 | ret = wc->process_func(root, next, wc, ptr_gen, |
2729 | *level - 1); | |
b50c6e25 JB |
2730 | if (ret) { |
2731 | free_extent_buffer(next); | |
1e5063d0 | 2732 | return ret; |
b50c6e25 | 2733 | } |
4a500fd1 | 2734 | |
e02119d5 CM |
2735 | path->slots[*level]++; |
2736 | if (wc->free) { | |
581c1760 QW |
2737 | ret = btrfs_read_buffer(next, ptr_gen, |
2738 | *level - 1, &first_key); | |
018642a1 TI |
2739 | if (ret) { |
2740 | free_extent_buffer(next); | |
2741 | return ret; | |
2742 | } | |
e02119d5 | 2743 | |
681ae509 JB |
2744 | if (trans) { |
2745 | btrfs_tree_lock(next); | |
6a884d7d | 2746 | btrfs_clean_tree_block(next); |
681ae509 JB |
2747 | btrfs_wait_tree_block_writeback(next); |
2748 | btrfs_tree_unlock(next); | |
7bfc1007 | 2749 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2750 | bytenr, blocksize); |
2751 | if (ret) { | |
2752 | free_extent_buffer(next); | |
2753 | return ret; | |
2754 | } | |
1846430c LB |
2755 | } else { |
2756 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2757 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2758 | unaccount_log_buffer(fs_info, bytenr); |
3650860b | 2759 | } |
e02119d5 CM |
2760 | } |
2761 | free_extent_buffer(next); | |
2762 | continue; | |
2763 | } | |
581c1760 | 2764 | ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key); |
018642a1 TI |
2765 | if (ret) { |
2766 | free_extent_buffer(next); | |
2767 | return ret; | |
2768 | } | |
e02119d5 | 2769 | |
e02119d5 CM |
2770 | if (path->nodes[*level-1]) |
2771 | free_extent_buffer(path->nodes[*level-1]); | |
2772 | path->nodes[*level-1] = next; | |
2773 | *level = btrfs_header_level(next); | |
2774 | path->slots[*level] = 0; | |
2775 | cond_resched(); | |
2776 | } | |
4a500fd1 | 2777 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2778 | |
2779 | cond_resched(); | |
2780 | return 0; | |
2781 | } | |
2782 | ||
d397712b | 2783 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2784 | struct btrfs_root *root, |
2785 | struct btrfs_path *path, int *level, | |
2786 | struct walk_control *wc) | |
2787 | { | |
0b246afa | 2788 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2789 | int i; |
2790 | int slot; | |
2791 | int ret; | |
2792 | ||
d397712b | 2793 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2794 | slot = path->slots[i]; |
4a500fd1 | 2795 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2796 | path->slots[i]++; |
2797 | *level = i; | |
2798 | WARN_ON(*level == 0); | |
2799 | return 0; | |
2800 | } else { | |
1e5063d0 | 2801 | ret = wc->process_func(root, path->nodes[*level], wc, |
581c1760 QW |
2802 | btrfs_header_generation(path->nodes[*level]), |
2803 | *level); | |
1e5063d0 MF |
2804 | if (ret) |
2805 | return ret; | |
2806 | ||
e02119d5 CM |
2807 | if (wc->free) { |
2808 | struct extent_buffer *next; | |
2809 | ||
2810 | next = path->nodes[*level]; | |
2811 | ||
681ae509 JB |
2812 | if (trans) { |
2813 | btrfs_tree_lock(next); | |
6a884d7d | 2814 | btrfs_clean_tree_block(next); |
681ae509 JB |
2815 | btrfs_wait_tree_block_writeback(next); |
2816 | btrfs_tree_unlock(next); | |
7bfc1007 | 2817 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2818 | path->nodes[*level]->start, |
2819 | path->nodes[*level]->len); | |
2820 | if (ret) | |
2821 | return ret; | |
1846430c LB |
2822 | } else { |
2823 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2824 | clear_extent_buffer_dirty(next); | |
e02119d5 | 2825 | |
10e958d5 NB |
2826 | unaccount_log_buffer(fs_info, |
2827 | path->nodes[*level]->start); | |
2828 | } | |
e02119d5 CM |
2829 | } |
2830 | free_extent_buffer(path->nodes[*level]); | |
2831 | path->nodes[*level] = NULL; | |
2832 | *level = i + 1; | |
2833 | } | |
2834 | } | |
2835 | return 1; | |
2836 | } | |
2837 | ||
2838 | /* | |
2839 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2840 | * the tree freeing any blocks that have a ref count of zero after being | |
2841 | * decremented. | |
2842 | */ | |
2843 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2844 | struct btrfs_root *log, struct walk_control *wc) | |
2845 | { | |
2ff7e61e | 2846 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2847 | int ret = 0; |
2848 | int wret; | |
2849 | int level; | |
2850 | struct btrfs_path *path; | |
e02119d5 CM |
2851 | int orig_level; |
2852 | ||
2853 | path = btrfs_alloc_path(); | |
db5b493a TI |
2854 | if (!path) |
2855 | return -ENOMEM; | |
e02119d5 CM |
2856 | |
2857 | level = btrfs_header_level(log->node); | |
2858 | orig_level = level; | |
2859 | path->nodes[level] = log->node; | |
67439dad | 2860 | atomic_inc(&log->node->refs); |
e02119d5 CM |
2861 | path->slots[level] = 0; |
2862 | ||
d397712b | 2863 | while (1) { |
e02119d5 CM |
2864 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2865 | if (wret > 0) | |
2866 | break; | |
79787eaa | 2867 | if (wret < 0) { |
e02119d5 | 2868 | ret = wret; |
79787eaa JM |
2869 | goto out; |
2870 | } | |
e02119d5 CM |
2871 | |
2872 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2873 | if (wret > 0) | |
2874 | break; | |
79787eaa | 2875 | if (wret < 0) { |
e02119d5 | 2876 | ret = wret; |
79787eaa JM |
2877 | goto out; |
2878 | } | |
e02119d5 CM |
2879 | } |
2880 | ||
2881 | /* was the root node processed? if not, catch it here */ | |
2882 | if (path->nodes[orig_level]) { | |
79787eaa | 2883 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
581c1760 QW |
2884 | btrfs_header_generation(path->nodes[orig_level]), |
2885 | orig_level); | |
79787eaa JM |
2886 | if (ret) |
2887 | goto out; | |
e02119d5 CM |
2888 | if (wc->free) { |
2889 | struct extent_buffer *next; | |
2890 | ||
2891 | next = path->nodes[orig_level]; | |
2892 | ||
681ae509 JB |
2893 | if (trans) { |
2894 | btrfs_tree_lock(next); | |
6a884d7d | 2895 | btrfs_clean_tree_block(next); |
681ae509 JB |
2896 | btrfs_wait_tree_block_writeback(next); |
2897 | btrfs_tree_unlock(next); | |
7bfc1007 | 2898 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2899 | next->start, next->len); |
2900 | if (ret) | |
2901 | goto out; | |
1846430c LB |
2902 | } else { |
2903 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2904 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2905 | unaccount_log_buffer(fs_info, next->start); |
681ae509 | 2906 | } |
e02119d5 CM |
2907 | } |
2908 | } | |
2909 | ||
79787eaa | 2910 | out: |
e02119d5 | 2911 | btrfs_free_path(path); |
e02119d5 CM |
2912 | return ret; |
2913 | } | |
2914 | ||
7237f183 YZ |
2915 | /* |
2916 | * helper function to update the item for a given subvolumes log root | |
2917 | * in the tree of log roots | |
2918 | */ | |
2919 | static int update_log_root(struct btrfs_trans_handle *trans, | |
4203e968 JB |
2920 | struct btrfs_root *log, |
2921 | struct btrfs_root_item *root_item) | |
7237f183 | 2922 | { |
0b246afa | 2923 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
2924 | int ret; |
2925 | ||
2926 | if (log->log_transid == 1) { | |
2927 | /* insert root item on the first sync */ | |
0b246afa | 2928 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
4203e968 | 2929 | &log->root_key, root_item); |
7237f183 | 2930 | } else { |
0b246afa | 2931 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
4203e968 | 2932 | &log->root_key, root_item); |
7237f183 YZ |
2933 | } |
2934 | return ret; | |
2935 | } | |
2936 | ||
60d53eb3 | 2937 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2938 | { |
2939 | DEFINE_WAIT(wait); | |
7237f183 | 2940 | int index = transid % 2; |
e02119d5 | 2941 | |
7237f183 YZ |
2942 | /* |
2943 | * we only allow two pending log transactions at a time, | |
2944 | * so we know that if ours is more than 2 older than the | |
2945 | * current transaction, we're done | |
2946 | */ | |
49e83f57 | 2947 | for (;;) { |
7237f183 YZ |
2948 | prepare_to_wait(&root->log_commit_wait[index], |
2949 | &wait, TASK_UNINTERRUPTIBLE); | |
12fcfd22 | 2950 | |
49e83f57 LB |
2951 | if (!(root->log_transid_committed < transid && |
2952 | atomic_read(&root->log_commit[index]))) | |
2953 | break; | |
12fcfd22 | 2954 | |
49e83f57 LB |
2955 | mutex_unlock(&root->log_mutex); |
2956 | schedule(); | |
7237f183 | 2957 | mutex_lock(&root->log_mutex); |
49e83f57 LB |
2958 | } |
2959 | finish_wait(&root->log_commit_wait[index], &wait); | |
7237f183 YZ |
2960 | } |
2961 | ||
60d53eb3 | 2962 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
2963 | { |
2964 | DEFINE_WAIT(wait); | |
8b050d35 | 2965 | |
49e83f57 LB |
2966 | for (;;) { |
2967 | prepare_to_wait(&root->log_writer_wait, &wait, | |
2968 | TASK_UNINTERRUPTIBLE); | |
2969 | if (!atomic_read(&root->log_writers)) | |
2970 | break; | |
2971 | ||
7237f183 | 2972 | mutex_unlock(&root->log_mutex); |
49e83f57 | 2973 | schedule(); |
575849ec | 2974 | mutex_lock(&root->log_mutex); |
7237f183 | 2975 | } |
49e83f57 | 2976 | finish_wait(&root->log_writer_wait, &wait); |
e02119d5 CM |
2977 | } |
2978 | ||
8b050d35 MX |
2979 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2980 | struct btrfs_log_ctx *ctx) | |
2981 | { | |
2982 | if (!ctx) | |
2983 | return; | |
2984 | ||
2985 | mutex_lock(&root->log_mutex); | |
2986 | list_del_init(&ctx->list); | |
2987 | mutex_unlock(&root->log_mutex); | |
2988 | } | |
2989 | ||
2990 | /* | |
2991 | * Invoked in log mutex context, or be sure there is no other task which | |
2992 | * can access the list. | |
2993 | */ | |
2994 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2995 | int index, int error) | |
2996 | { | |
2997 | struct btrfs_log_ctx *ctx; | |
570dd450 | 2998 | struct btrfs_log_ctx *safe; |
8b050d35 | 2999 | |
570dd450 CM |
3000 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
3001 | list_del_init(&ctx->list); | |
8b050d35 | 3002 | ctx->log_ret = error; |
570dd450 | 3003 | } |
8b050d35 MX |
3004 | |
3005 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
3006 | } | |
3007 | ||
e02119d5 CM |
3008 | /* |
3009 | * btrfs_sync_log does sends a given tree log down to the disk and | |
3010 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
3011 | * you know that any inodes previously logged are safely on disk only |
3012 | * if it returns 0. | |
3013 | * | |
3014 | * Any other return value means you need to call btrfs_commit_transaction. | |
3015 | * Some of the edge cases for fsyncing directories that have had unlinks | |
3016 | * or renames done in the past mean that sometimes the only safe | |
3017 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
3018 | * that has happened. | |
e02119d5 CM |
3019 | */ |
3020 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 3021 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 3022 | { |
7237f183 YZ |
3023 | int index1; |
3024 | int index2; | |
8cef4e16 | 3025 | int mark; |
e02119d5 | 3026 | int ret; |
0b246afa | 3027 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 3028 | struct btrfs_root *log = root->log_root; |
0b246afa | 3029 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
4203e968 | 3030 | struct btrfs_root_item new_root_item; |
bb14a59b | 3031 | int log_transid = 0; |
8b050d35 | 3032 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 3033 | struct blk_plug plug; |
47876f7c FM |
3034 | u64 log_root_start; |
3035 | u64 log_root_level; | |
e02119d5 | 3036 | |
7237f183 | 3037 | mutex_lock(&root->log_mutex); |
d1433deb MX |
3038 | log_transid = ctx->log_transid; |
3039 | if (root->log_transid_committed >= log_transid) { | |
3040 | mutex_unlock(&root->log_mutex); | |
3041 | return ctx->log_ret; | |
3042 | } | |
3043 | ||
3044 | index1 = log_transid % 2; | |
7237f183 | 3045 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 3046 | wait_log_commit(root, log_transid); |
7237f183 | 3047 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3048 | return ctx->log_ret; |
e02119d5 | 3049 | } |
d1433deb | 3050 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
3051 | atomic_set(&root->log_commit[index1], 1); |
3052 | ||
3053 | /* wait for previous tree log sync to complete */ | |
3054 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 3055 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 3056 | |
86df7eb9 | 3057 | while (1) { |
2ecb7923 | 3058 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 3059 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 3060 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 3061 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
3062 | mutex_unlock(&root->log_mutex); |
3063 | schedule_timeout_uninterruptible(1); | |
3064 | mutex_lock(&root->log_mutex); | |
3065 | } | |
60d53eb3 | 3066 | wait_for_writer(root); |
2ecb7923 | 3067 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
3068 | break; |
3069 | } | |
e02119d5 | 3070 | |
12fcfd22 | 3071 | /* bail out if we need to do a full commit */ |
4884b8e8 | 3072 | if (btrfs_need_log_full_commit(trans)) { |
12fcfd22 CM |
3073 | ret = -EAGAIN; |
3074 | mutex_unlock(&root->log_mutex); | |
3075 | goto out; | |
3076 | } | |
3077 | ||
8cef4e16 YZ |
3078 | if (log_transid % 2 == 0) |
3079 | mark = EXTENT_DIRTY; | |
3080 | else | |
3081 | mark = EXTENT_NEW; | |
3082 | ||
690587d1 CM |
3083 | /* we start IO on all the marked extents here, but we don't actually |
3084 | * wait for them until later. | |
3085 | */ | |
c6adc9cc | 3086 | blk_start_plug(&plug); |
2ff7e61e | 3087 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
79787eaa | 3088 | if (ret) { |
c6adc9cc | 3089 | blk_finish_plug(&plug); |
66642832 | 3090 | btrfs_abort_transaction(trans, ret); |
90787766 | 3091 | btrfs_set_log_full_commit(trans); |
79787eaa JM |
3092 | mutex_unlock(&root->log_mutex); |
3093 | goto out; | |
3094 | } | |
7237f183 | 3095 | |
4203e968 JB |
3096 | /* |
3097 | * We _must_ update under the root->log_mutex in order to make sure we | |
3098 | * have a consistent view of the log root we are trying to commit at | |
3099 | * this moment. | |
3100 | * | |
3101 | * We _must_ copy this into a local copy, because we are not holding the | |
3102 | * log_root_tree->log_mutex yet. This is important because when we | |
3103 | * commit the log_root_tree we must have a consistent view of the | |
3104 | * log_root_tree when we update the super block to point at the | |
3105 | * log_root_tree bytenr. If we update the log_root_tree here we'll race | |
3106 | * with the commit and possibly point at the new block which we may not | |
3107 | * have written out. | |
3108 | */ | |
5d4f98a2 | 3109 | btrfs_set_root_node(&log->root_item, log->node); |
4203e968 | 3110 | memcpy(&new_root_item, &log->root_item, sizeof(new_root_item)); |
7237f183 | 3111 | |
7237f183 YZ |
3112 | root->log_transid++; |
3113 | log->log_transid = root->log_transid; | |
ff782e0a | 3114 | root->log_start_pid = 0; |
7237f183 | 3115 | /* |
8cef4e16 YZ |
3116 | * IO has been started, blocks of the log tree have WRITTEN flag set |
3117 | * in their headers. new modifications of the log will be written to | |
3118 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
3119 | */ |
3120 | mutex_unlock(&root->log_mutex); | |
3121 | ||
28a23593 | 3122 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
d1433deb | 3123 | |
7237f183 | 3124 | mutex_lock(&log_root_tree->log_mutex); |
d1433deb MX |
3125 | |
3126 | index2 = log_root_tree->log_transid % 2; | |
3127 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
3128 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
3129 | ||
4203e968 JB |
3130 | /* |
3131 | * Now we are safe to update the log_root_tree because we're under the | |
3132 | * log_mutex, and we're a current writer so we're holding the commit | |
3133 | * open until we drop the log_mutex. | |
3134 | */ | |
3135 | ret = update_log_root(trans, log, &new_root_item); | |
4a500fd1 | 3136 | if (ret) { |
d1433deb MX |
3137 | if (!list_empty(&root_log_ctx.list)) |
3138 | list_del_init(&root_log_ctx.list); | |
3139 | ||
c6adc9cc | 3140 | blk_finish_plug(&plug); |
90787766 | 3141 | btrfs_set_log_full_commit(trans); |
995946dd | 3142 | |
79787eaa | 3143 | if (ret != -ENOSPC) { |
66642832 | 3144 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3145 | mutex_unlock(&log_root_tree->log_mutex); |
3146 | goto out; | |
3147 | } | |
bf89d38f | 3148 | btrfs_wait_tree_log_extents(log, mark); |
4a500fd1 YZ |
3149 | mutex_unlock(&log_root_tree->log_mutex); |
3150 | ret = -EAGAIN; | |
3151 | goto out; | |
3152 | } | |
3153 | ||
d1433deb | 3154 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 3155 | blk_finish_plug(&plug); |
cbd60aa7 | 3156 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
3157 | mutex_unlock(&log_root_tree->log_mutex); |
3158 | ret = root_log_ctx.log_ret; | |
3159 | goto out; | |
3160 | } | |
8b050d35 | 3161 | |
d1433deb | 3162 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 3163 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 3164 | blk_finish_plug(&plug); |
bf89d38f | 3165 | ret = btrfs_wait_tree_log_extents(log, mark); |
60d53eb3 | 3166 | wait_log_commit(log_root_tree, |
d1433deb | 3167 | root_log_ctx.log_transid); |
7237f183 | 3168 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
3169 | if (!ret) |
3170 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
3171 | goto out; |
3172 | } | |
d1433deb | 3173 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
3174 | atomic_set(&log_root_tree->log_commit[index2], 1); |
3175 | ||
12fcfd22 | 3176 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 3177 | wait_log_commit(log_root_tree, |
d1433deb | 3178 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
3179 | } |
3180 | ||
12fcfd22 CM |
3181 | /* |
3182 | * now that we've moved on to the tree of log tree roots, | |
3183 | * check the full commit flag again | |
3184 | */ | |
4884b8e8 | 3185 | if (btrfs_need_log_full_commit(trans)) { |
c6adc9cc | 3186 | blk_finish_plug(&plug); |
bf89d38f | 3187 | btrfs_wait_tree_log_extents(log, mark); |
12fcfd22 CM |
3188 | mutex_unlock(&log_root_tree->log_mutex); |
3189 | ret = -EAGAIN; | |
3190 | goto out_wake_log_root; | |
3191 | } | |
7237f183 | 3192 | |
2ff7e61e | 3193 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
3194 | &log_root_tree->dirty_log_pages, |
3195 | EXTENT_DIRTY | EXTENT_NEW); | |
3196 | blk_finish_plug(&plug); | |
79787eaa | 3197 | if (ret) { |
90787766 | 3198 | btrfs_set_log_full_commit(trans); |
66642832 | 3199 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3200 | mutex_unlock(&log_root_tree->log_mutex); |
3201 | goto out_wake_log_root; | |
3202 | } | |
bf89d38f | 3203 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 3204 | if (!ret) |
bf89d38f JM |
3205 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
3206 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 3207 | if (ret) { |
90787766 | 3208 | btrfs_set_log_full_commit(trans); |
5ab5e44a FM |
3209 | mutex_unlock(&log_root_tree->log_mutex); |
3210 | goto out_wake_log_root; | |
3211 | } | |
e02119d5 | 3212 | |
47876f7c FM |
3213 | log_root_start = log_root_tree->node->start; |
3214 | log_root_level = btrfs_header_level(log_root_tree->node); | |
7237f183 | 3215 | log_root_tree->log_transid++; |
7237f183 YZ |
3216 | mutex_unlock(&log_root_tree->log_mutex); |
3217 | ||
3218 | /* | |
47876f7c FM |
3219 | * Here we are guaranteed that nobody is going to write the superblock |
3220 | * for the current transaction before us and that neither we do write | |
3221 | * our superblock before the previous transaction finishes its commit | |
3222 | * and writes its superblock, because: | |
3223 | * | |
3224 | * 1) We are holding a handle on the current transaction, so no body | |
3225 | * can commit it until we release the handle; | |
3226 | * | |
3227 | * 2) Before writing our superblock we acquire the tree_log_mutex, so | |
3228 | * if the previous transaction is still committing, and hasn't yet | |
3229 | * written its superblock, we wait for it to do it, because a | |
3230 | * transaction commit acquires the tree_log_mutex when the commit | |
3231 | * begins and releases it only after writing its superblock. | |
7237f183 | 3232 | */ |
47876f7c FM |
3233 | mutex_lock(&fs_info->tree_log_mutex); |
3234 | btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start); | |
3235 | btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level); | |
eece6a9c | 3236 | ret = write_all_supers(fs_info, 1); |
47876f7c | 3237 | mutex_unlock(&fs_info->tree_log_mutex); |
5af3e8cc | 3238 | if (ret) { |
90787766 | 3239 | btrfs_set_log_full_commit(trans); |
66642832 | 3240 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
3241 | goto out_wake_log_root; |
3242 | } | |
7237f183 | 3243 | |
257c62e1 CM |
3244 | mutex_lock(&root->log_mutex); |
3245 | if (root->last_log_commit < log_transid) | |
3246 | root->last_log_commit = log_transid; | |
3247 | mutex_unlock(&root->log_mutex); | |
3248 | ||
12fcfd22 | 3249 | out_wake_log_root: |
570dd450 | 3250 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
3251 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
3252 | ||
d1433deb | 3253 | log_root_tree->log_transid_committed++; |
7237f183 | 3254 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
3255 | mutex_unlock(&log_root_tree->log_mutex); |
3256 | ||
33a9eca7 | 3257 | /* |
093258e6 DS |
3258 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3259 | * all the updates above are seen by the woken threads. It might not be | |
3260 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3261 | */ |
093258e6 | 3262 | cond_wake_up(&log_root_tree->log_commit_wait[index2]); |
e02119d5 | 3263 | out: |
d1433deb | 3264 | mutex_lock(&root->log_mutex); |
570dd450 | 3265 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 3266 | root->log_transid_committed++; |
7237f183 | 3267 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 3268 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3269 | |
33a9eca7 | 3270 | /* |
093258e6 DS |
3271 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3272 | * all the updates above are seen by the woken threads. It might not be | |
3273 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3274 | */ |
093258e6 | 3275 | cond_wake_up(&root->log_commit_wait[index1]); |
b31eabd8 | 3276 | return ret; |
e02119d5 CM |
3277 | } |
3278 | ||
4a500fd1 YZ |
3279 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3280 | struct btrfs_root *log) | |
e02119d5 CM |
3281 | { |
3282 | int ret; | |
e02119d5 CM |
3283 | struct walk_control wc = { |
3284 | .free = 1, | |
3285 | .process_func = process_one_buffer | |
3286 | }; | |
3287 | ||
681ae509 | 3288 | ret = walk_log_tree(trans, log, &wc); |
374b0e2d JM |
3289 | if (ret) { |
3290 | if (trans) | |
3291 | btrfs_abort_transaction(trans, ret); | |
3292 | else | |
3293 | btrfs_handle_fs_error(log->fs_info, ret, NULL); | |
3294 | } | |
e02119d5 | 3295 | |
59b0713a FM |
3296 | clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1, |
3297 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT); | |
e289f03e | 3298 | extent_io_tree_release(&log->log_csum_range); |
00246528 | 3299 | btrfs_put_root(log); |
4a500fd1 YZ |
3300 | } |
3301 | ||
3302 | /* | |
3303 | * free all the extents used by the tree log. This should be called | |
3304 | * at commit time of the full transaction | |
3305 | */ | |
3306 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3307 | { | |
3308 | if (root->log_root) { | |
3309 | free_log_tree(trans, root->log_root); | |
3310 | root->log_root = NULL; | |
e7a79811 | 3311 | clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state); |
4a500fd1 YZ |
3312 | } |
3313 | return 0; | |
3314 | } | |
3315 | ||
3316 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3317 | struct btrfs_fs_info *fs_info) | |
3318 | { | |
3319 | if (fs_info->log_root_tree) { | |
3320 | free_log_tree(trans, fs_info->log_root_tree); | |
3321 | fs_info->log_root_tree = NULL; | |
47876f7c | 3322 | clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state); |
4a500fd1 | 3323 | } |
e02119d5 CM |
3324 | return 0; |
3325 | } | |
3326 | ||
803f0f64 FM |
3327 | /* |
3328 | * Check if an inode was logged in the current transaction. We can't always rely | |
3329 | * on an inode's logged_trans value, because it's an in-memory only field and | |
3330 | * therefore not persisted. This means that its value is lost if the inode gets | |
3331 | * evicted and loaded again from disk (in which case it has a value of 0, and | |
3332 | * certainly it is smaller then any possible transaction ID), when that happens | |
3333 | * the full_sync flag is set in the inode's runtime flags, so on that case we | |
3334 | * assume eviction happened and ignore the logged_trans value, assuming the | |
3335 | * worst case, that the inode was logged before in the current transaction. | |
3336 | */ | |
3337 | static bool inode_logged(struct btrfs_trans_handle *trans, | |
3338 | struct btrfs_inode *inode) | |
3339 | { | |
3340 | if (inode->logged_trans == trans->transid) | |
3341 | return true; | |
3342 | ||
3343 | if (inode->last_trans == trans->transid && | |
3344 | test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) && | |
3345 | !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags)) | |
3346 | return true; | |
3347 | ||
3348 | return false; | |
3349 | } | |
3350 | ||
e02119d5 CM |
3351 | /* |
3352 | * If both a file and directory are logged, and unlinks or renames are | |
3353 | * mixed in, we have a few interesting corners: | |
3354 | * | |
3355 | * create file X in dir Y | |
3356 | * link file X to X.link in dir Y | |
3357 | * fsync file X | |
3358 | * unlink file X but leave X.link | |
3359 | * fsync dir Y | |
3360 | * | |
3361 | * After a crash we would expect only X.link to exist. But file X | |
3362 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3363 | * | |
3364 | * We solve this by removing directory entries and inode backrefs from the | |
3365 | * log when a file that was logged in the current transaction is | |
3366 | * unlinked. Any later fsync will include the updated log entries, and | |
3367 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3368 | * | |
3369 | * This optimizations allows us to avoid relogging the entire inode | |
3370 | * or the entire directory. | |
3371 | */ | |
3372 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3373 | struct btrfs_root *root, | |
3374 | const char *name, int name_len, | |
49f34d1f | 3375 | struct btrfs_inode *dir, u64 index) |
e02119d5 CM |
3376 | { |
3377 | struct btrfs_root *log; | |
3378 | struct btrfs_dir_item *di; | |
3379 | struct btrfs_path *path; | |
3380 | int ret; | |
4a500fd1 | 3381 | int err = 0; |
49f34d1f | 3382 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3383 | |
803f0f64 | 3384 | if (!inode_logged(trans, dir)) |
3a5f1d45 CM |
3385 | return 0; |
3386 | ||
e02119d5 CM |
3387 | ret = join_running_log_trans(root); |
3388 | if (ret) | |
3389 | return 0; | |
3390 | ||
49f34d1f | 3391 | mutex_lock(&dir->log_mutex); |
e02119d5 CM |
3392 | |
3393 | log = root->log_root; | |
3394 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3395 | if (!path) { |
3396 | err = -ENOMEM; | |
3397 | goto out_unlock; | |
3398 | } | |
2a29edc6 | 3399 | |
33345d01 | 3400 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3401 | name, name_len, -1); |
4a500fd1 YZ |
3402 | if (IS_ERR(di)) { |
3403 | err = PTR_ERR(di); | |
3404 | goto fail; | |
3405 | } | |
3406 | if (di) { | |
e02119d5 | 3407 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3650860b JB |
3408 | if (ret) { |
3409 | err = ret; | |
3410 | goto fail; | |
3411 | } | |
e02119d5 | 3412 | } |
b3b4aa74 | 3413 | btrfs_release_path(path); |
33345d01 | 3414 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3415 | index, name, name_len, -1); |
4a500fd1 YZ |
3416 | if (IS_ERR(di)) { |
3417 | err = PTR_ERR(di); | |
3418 | goto fail; | |
3419 | } | |
3420 | if (di) { | |
e02119d5 | 3421 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3650860b JB |
3422 | if (ret) { |
3423 | err = ret; | |
3424 | goto fail; | |
3425 | } | |
e02119d5 CM |
3426 | } |
3427 | ||
ddffcf6f FM |
3428 | /* |
3429 | * We do not need to update the size field of the directory's inode item | |
3430 | * because on log replay we update the field to reflect all existing | |
3431 | * entries in the directory (see overwrite_item()). | |
e02119d5 | 3432 | */ |
4a500fd1 | 3433 | fail: |
e02119d5 | 3434 | btrfs_free_path(path); |
a62f44a5 | 3435 | out_unlock: |
49f34d1f | 3436 | mutex_unlock(&dir->log_mutex); |
fb2fecba | 3437 | if (err == -ENOSPC) { |
90787766 | 3438 | btrfs_set_log_full_commit(trans); |
fb2fecba JB |
3439 | err = 0; |
3440 | } else if (err < 0 && err != -ENOENT) { | |
3441 | /* ENOENT can be returned if the entry hasn't been fsynced yet */ | |
3442 | btrfs_abort_transaction(trans, err); | |
3443 | } | |
79787eaa | 3444 | |
12fcfd22 | 3445 | btrfs_end_log_trans(root); |
e02119d5 | 3446 | |
411fc6bc | 3447 | return err; |
e02119d5 CM |
3448 | } |
3449 | ||
3450 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3451 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3452 | struct btrfs_root *root, | |
3453 | const char *name, int name_len, | |
a491abb2 | 3454 | struct btrfs_inode *inode, u64 dirid) |
e02119d5 CM |
3455 | { |
3456 | struct btrfs_root *log; | |
3457 | u64 index; | |
3458 | int ret; | |
3459 | ||
803f0f64 | 3460 | if (!inode_logged(trans, inode)) |
3a5f1d45 CM |
3461 | return 0; |
3462 | ||
e02119d5 CM |
3463 | ret = join_running_log_trans(root); |
3464 | if (ret) | |
3465 | return 0; | |
3466 | log = root->log_root; | |
a491abb2 | 3467 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3468 | |
a491abb2 | 3469 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 | 3470 | dirid, &index); |
a491abb2 | 3471 | mutex_unlock(&inode->log_mutex); |
4a500fd1 | 3472 | if (ret == -ENOSPC) { |
90787766 | 3473 | btrfs_set_log_full_commit(trans); |
4a500fd1 | 3474 | ret = 0; |
79787eaa | 3475 | } else if (ret < 0 && ret != -ENOENT) |
66642832 | 3476 | btrfs_abort_transaction(trans, ret); |
12fcfd22 | 3477 | btrfs_end_log_trans(root); |
e02119d5 | 3478 | |
e02119d5 CM |
3479 | return ret; |
3480 | } | |
3481 | ||
3482 | /* | |
3483 | * creates a range item in the log for 'dirid'. first_offset and | |
3484 | * last_offset tell us which parts of the key space the log should | |
3485 | * be considered authoritative for. | |
3486 | */ | |
3487 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3488 | struct btrfs_root *log, | |
3489 | struct btrfs_path *path, | |
3490 | int key_type, u64 dirid, | |
3491 | u64 first_offset, u64 last_offset) | |
3492 | { | |
3493 | int ret; | |
3494 | struct btrfs_key key; | |
3495 | struct btrfs_dir_log_item *item; | |
3496 | ||
3497 | key.objectid = dirid; | |
3498 | key.offset = first_offset; | |
3499 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3500 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3501 | else | |
3502 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3503 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3504 | if (ret) |
3505 | return ret; | |
e02119d5 CM |
3506 | |
3507 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3508 | struct btrfs_dir_log_item); | |
3509 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3510 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3511 | btrfs_release_path(path); |
e02119d5 CM |
3512 | return 0; |
3513 | } | |
3514 | ||
3515 | /* | |
3516 | * log all the items included in the current transaction for a given | |
3517 | * directory. This also creates the range items in the log tree required | |
3518 | * to replay anything deleted before the fsync | |
3519 | */ | |
3520 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
684a5773 | 3521 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 CM |
3522 | struct btrfs_path *path, |
3523 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3524 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3525 | u64 min_offset, u64 *last_offset_ret) |
3526 | { | |
3527 | struct btrfs_key min_key; | |
e02119d5 CM |
3528 | struct btrfs_root *log = root->log_root; |
3529 | struct extent_buffer *src; | |
4a500fd1 | 3530 | int err = 0; |
e02119d5 CM |
3531 | int ret; |
3532 | int i; | |
3533 | int nritems; | |
3534 | u64 first_offset = min_offset; | |
3535 | u64 last_offset = (u64)-1; | |
684a5773 | 3536 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
3537 | |
3538 | log = root->log_root; | |
e02119d5 | 3539 | |
33345d01 | 3540 | min_key.objectid = ino; |
e02119d5 CM |
3541 | min_key.type = key_type; |
3542 | min_key.offset = min_offset; | |
3543 | ||
6174d3cb | 3544 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3545 | |
3546 | /* | |
3547 | * we didn't find anything from this transaction, see if there | |
3548 | * is anything at all | |
3549 | */ | |
33345d01 LZ |
3550 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3551 | min_key.objectid = ino; | |
e02119d5 CM |
3552 | min_key.type = key_type; |
3553 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3554 | btrfs_release_path(path); |
e02119d5 CM |
3555 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3556 | if (ret < 0) { | |
b3b4aa74 | 3557 | btrfs_release_path(path); |
e02119d5 CM |
3558 | return ret; |
3559 | } | |
33345d01 | 3560 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3561 | |
3562 | /* if ret == 0 there are items for this type, | |
3563 | * create a range to tell us the last key of this type. | |
3564 | * otherwise, there are no items in this directory after | |
3565 | * *min_offset, and we create a range to indicate that. | |
3566 | */ | |
3567 | if (ret == 0) { | |
3568 | struct btrfs_key tmp; | |
3569 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3570 | path->slots[0]); | |
d397712b | 3571 | if (key_type == tmp.type) |
e02119d5 | 3572 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3573 | } |
3574 | goto done; | |
3575 | } | |
3576 | ||
3577 | /* go backward to find any previous key */ | |
33345d01 | 3578 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3579 | if (ret == 0) { |
3580 | struct btrfs_key tmp; | |
3581 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3582 | if (key_type == tmp.type) { | |
3583 | first_offset = tmp.offset; | |
3584 | ret = overwrite_item(trans, log, dst_path, | |
3585 | path->nodes[0], path->slots[0], | |
3586 | &tmp); | |
4a500fd1 YZ |
3587 | if (ret) { |
3588 | err = ret; | |
3589 | goto done; | |
3590 | } | |
e02119d5 CM |
3591 | } |
3592 | } | |
b3b4aa74 | 3593 | btrfs_release_path(path); |
e02119d5 | 3594 | |
2cc83342 JB |
3595 | /* |
3596 | * Find the first key from this transaction again. See the note for | |
3597 | * log_new_dir_dentries, if we're logging a directory recursively we | |
3598 | * won't be holding its i_mutex, which means we can modify the directory | |
3599 | * while we're logging it. If we remove an entry between our first | |
3600 | * search and this search we'll not find the key again and can just | |
3601 | * bail. | |
3602 | */ | |
bb56f02f | 3603 | search: |
e02119d5 | 3604 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
2cc83342 | 3605 | if (ret != 0) |
e02119d5 | 3606 | goto done; |
e02119d5 CM |
3607 | |
3608 | /* | |
3609 | * we have a block from this transaction, log every item in it | |
3610 | * from our directory | |
3611 | */ | |
d397712b | 3612 | while (1) { |
e02119d5 CM |
3613 | struct btrfs_key tmp; |
3614 | src = path->nodes[0]; | |
3615 | nritems = btrfs_header_nritems(src); | |
3616 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3617 | struct btrfs_dir_item *di; |
3618 | ||
e02119d5 CM |
3619 | btrfs_item_key_to_cpu(src, &min_key, i); |
3620 | ||
33345d01 | 3621 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 | 3622 | goto done; |
bb56f02f FM |
3623 | |
3624 | if (need_resched()) { | |
3625 | btrfs_release_path(path); | |
3626 | cond_resched(); | |
3627 | goto search; | |
3628 | } | |
3629 | ||
e02119d5 CM |
3630 | ret = overwrite_item(trans, log, dst_path, src, i, |
3631 | &min_key); | |
4a500fd1 YZ |
3632 | if (ret) { |
3633 | err = ret; | |
3634 | goto done; | |
3635 | } | |
2f2ff0ee FM |
3636 | |
3637 | /* | |
3638 | * We must make sure that when we log a directory entry, | |
3639 | * the corresponding inode, after log replay, has a | |
3640 | * matching link count. For example: | |
3641 | * | |
3642 | * touch foo | |
3643 | * mkdir mydir | |
3644 | * sync | |
3645 | * ln foo mydir/bar | |
3646 | * xfs_io -c "fsync" mydir | |
3647 | * <crash> | |
3648 | * <mount fs and log replay> | |
3649 | * | |
3650 | * Would result in a fsync log that when replayed, our | |
3651 | * file inode would have a link count of 1, but we get | |
3652 | * two directory entries pointing to the same inode. | |
3653 | * After removing one of the names, it would not be | |
3654 | * possible to remove the other name, which resulted | |
3655 | * always in stale file handle errors, and would not | |
3656 | * be possible to rmdir the parent directory, since | |
3657 | * its i_size could never decrement to the value | |
3658 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3659 | */ | |
3660 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3661 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3662 | if (ctx && | |
3663 | (btrfs_dir_transid(src, di) == trans->transid || | |
3664 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3665 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3666 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3667 | } |
3668 | path->slots[0] = nritems; | |
3669 | ||
3670 | /* | |
3671 | * look ahead to the next item and see if it is also | |
3672 | * from this directory and from this transaction | |
3673 | */ | |
3674 | ret = btrfs_next_leaf(root, path); | |
80c0b421 LB |
3675 | if (ret) { |
3676 | if (ret == 1) | |
3677 | last_offset = (u64)-1; | |
3678 | else | |
3679 | err = ret; | |
e02119d5 CM |
3680 | goto done; |
3681 | } | |
3682 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3683 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3684 | last_offset = (u64)-1; |
3685 | goto done; | |
3686 | } | |
3687 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3688 | ret = overwrite_item(trans, log, dst_path, | |
3689 | path->nodes[0], path->slots[0], | |
3690 | &tmp); | |
4a500fd1 YZ |
3691 | if (ret) |
3692 | err = ret; | |
3693 | else | |
3694 | last_offset = tmp.offset; | |
e02119d5 CM |
3695 | goto done; |
3696 | } | |
3697 | } | |
3698 | done: | |
b3b4aa74 DS |
3699 | btrfs_release_path(path); |
3700 | btrfs_release_path(dst_path); | |
e02119d5 | 3701 | |
4a500fd1 YZ |
3702 | if (err == 0) { |
3703 | *last_offset_ret = last_offset; | |
3704 | /* | |
3705 | * insert the log range keys to indicate where the log | |
3706 | * is valid | |
3707 | */ | |
3708 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3709 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3710 | if (ret) |
3711 | err = ret; | |
3712 | } | |
3713 | return err; | |
e02119d5 CM |
3714 | } |
3715 | ||
3716 | /* | |
3717 | * logging directories is very similar to logging inodes, We find all the items | |
3718 | * from the current transaction and write them to the log. | |
3719 | * | |
3720 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3721 | * key in the range logged that is not present in the log tree, then it means | |
3722 | * that dir entry was unlinked during the transaction. | |
3723 | * | |
3724 | * In order for that scan to work, we must include one key smaller than | |
3725 | * the smallest logged by this transaction and one key larger than the largest | |
3726 | * key logged by this transaction. | |
3727 | */ | |
3728 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
dbf39ea4 | 3729 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 | 3730 | struct btrfs_path *path, |
2f2ff0ee FM |
3731 | struct btrfs_path *dst_path, |
3732 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3733 | { |
3734 | u64 min_key; | |
3735 | u64 max_key; | |
3736 | int ret; | |
3737 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3738 | ||
3739 | again: | |
3740 | min_key = 0; | |
3741 | max_key = 0; | |
d397712b | 3742 | while (1) { |
dbf39ea4 NB |
3743 | ret = log_dir_items(trans, root, inode, path, dst_path, key_type, |
3744 | ctx, min_key, &max_key); | |
4a500fd1 YZ |
3745 | if (ret) |
3746 | return ret; | |
e02119d5 CM |
3747 | if (max_key == (u64)-1) |
3748 | break; | |
3749 | min_key = max_key + 1; | |
3750 | } | |
3751 | ||
3752 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3753 | key_type = BTRFS_DIR_INDEX_KEY; | |
3754 | goto again; | |
3755 | } | |
3756 | return 0; | |
3757 | } | |
3758 | ||
3759 | /* | |
3760 | * a helper function to drop items from the log before we relog an | |
3761 | * inode. max_key_type indicates the highest item type to remove. | |
3762 | * This cannot be run for file data extents because it does not | |
3763 | * free the extents they point to. | |
3764 | */ | |
3765 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3766 | struct btrfs_root *log, | |
3767 | struct btrfs_path *path, | |
3768 | u64 objectid, int max_key_type) | |
3769 | { | |
3770 | int ret; | |
3771 | struct btrfs_key key; | |
3772 | struct btrfs_key found_key; | |
18ec90d6 | 3773 | int start_slot; |
e02119d5 CM |
3774 | |
3775 | key.objectid = objectid; | |
3776 | key.type = max_key_type; | |
3777 | key.offset = (u64)-1; | |
3778 | ||
d397712b | 3779 | while (1) { |
e02119d5 | 3780 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3781 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3782 | if (ret < 0) |
e02119d5 CM |
3783 | break; |
3784 | ||
3785 | if (path->slots[0] == 0) | |
3786 | break; | |
3787 | ||
3788 | path->slots[0]--; | |
3789 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3790 | path->slots[0]); | |
3791 | ||
3792 | if (found_key.objectid != objectid) | |
3793 | break; | |
3794 | ||
18ec90d6 JB |
3795 | found_key.offset = 0; |
3796 | found_key.type = 0; | |
e3b83361 | 3797 | ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot); |
cbca7d59 FM |
3798 | if (ret < 0) |
3799 | break; | |
18ec90d6 JB |
3800 | |
3801 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3802 | path->slots[0] - start_slot + 1); | |
3803 | /* | |
3804 | * If start slot isn't 0 then we don't need to re-search, we've | |
3805 | * found the last guy with the objectid in this tree. | |
3806 | */ | |
3807 | if (ret || start_slot != 0) | |
65a246c5 | 3808 | break; |
b3b4aa74 | 3809 | btrfs_release_path(path); |
e02119d5 | 3810 | } |
b3b4aa74 | 3811 | btrfs_release_path(path); |
5bdbeb21 JB |
3812 | if (ret > 0) |
3813 | ret = 0; | |
4a500fd1 | 3814 | return ret; |
e02119d5 CM |
3815 | } |
3816 | ||
94edf4ae JB |
3817 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3818 | struct extent_buffer *leaf, | |
3819 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3820 | struct inode *inode, int log_inode_only, |
3821 | u64 logged_isize) | |
94edf4ae | 3822 | { |
0b1c6cca JB |
3823 | struct btrfs_map_token token; |
3824 | ||
c82f823c | 3825 | btrfs_init_map_token(&token, leaf); |
94edf4ae JB |
3826 | |
3827 | if (log_inode_only) { | |
3828 | /* set the generation to zero so the recover code | |
3829 | * can tell the difference between an logging | |
3830 | * just to say 'this inode exists' and a logging | |
3831 | * to say 'update this inode with these values' | |
3832 | */ | |
cc4c13d5 DS |
3833 | btrfs_set_token_inode_generation(&token, item, 0); |
3834 | btrfs_set_token_inode_size(&token, item, logged_isize); | |
94edf4ae | 3835 | } else { |
cc4c13d5 DS |
3836 | btrfs_set_token_inode_generation(&token, item, |
3837 | BTRFS_I(inode)->generation); | |
3838 | btrfs_set_token_inode_size(&token, item, inode->i_size); | |
0b1c6cca JB |
3839 | } |
3840 | ||
cc4c13d5 DS |
3841 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
3842 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
3843 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
3844 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
3845 | ||
3846 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
3847 | inode->i_atime.tv_sec); | |
3848 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
3849 | inode->i_atime.tv_nsec); | |
3850 | ||
3851 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
3852 | inode->i_mtime.tv_sec); | |
3853 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
3854 | inode->i_mtime.tv_nsec); | |
3855 | ||
3856 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
3857 | inode->i_ctime.tv_sec); | |
3858 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
3859 | inode->i_ctime.tv_nsec); | |
3860 | ||
3861 | btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode)); | |
3862 | ||
3863 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
3864 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
3865 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
3866 | btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags); | |
3867 | btrfs_set_token_inode_block_group(&token, item, 0); | |
94edf4ae JB |
3868 | } |
3869 | ||
a95249b3 JB |
3870 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3871 | struct btrfs_root *log, struct btrfs_path *path, | |
6d889a3b | 3872 | struct btrfs_inode *inode) |
a95249b3 JB |
3873 | { |
3874 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3875 | int ret; |
3876 | ||
efd0c405 | 3877 | ret = btrfs_insert_empty_item(trans, log, path, |
6d889a3b | 3878 | &inode->location, sizeof(*inode_item)); |
a95249b3 JB |
3879 | if (ret && ret != -EEXIST) |
3880 | return ret; | |
3881 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3882 | struct btrfs_inode_item); | |
6d889a3b NB |
3883 | fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode, |
3884 | 0, 0); | |
a95249b3 JB |
3885 | btrfs_release_path(path); |
3886 | return 0; | |
3887 | } | |
3888 | ||
40e046ac | 3889 | static int log_csums(struct btrfs_trans_handle *trans, |
3ebac17c | 3890 | struct btrfs_inode *inode, |
40e046ac FM |
3891 | struct btrfs_root *log_root, |
3892 | struct btrfs_ordered_sum *sums) | |
3893 | { | |
e289f03e FM |
3894 | const u64 lock_end = sums->bytenr + sums->len - 1; |
3895 | struct extent_state *cached_state = NULL; | |
40e046ac FM |
3896 | int ret; |
3897 | ||
3ebac17c FM |
3898 | /* |
3899 | * If this inode was not used for reflink operations in the current | |
3900 | * transaction with new extents, then do the fast path, no need to | |
3901 | * worry about logging checksum items with overlapping ranges. | |
3902 | */ | |
3903 | if (inode->last_reflink_trans < trans->transid) | |
3904 | return btrfs_csum_file_blocks(trans, log_root, sums); | |
3905 | ||
e289f03e FM |
3906 | /* |
3907 | * Serialize logging for checksums. This is to avoid racing with the | |
3908 | * same checksum being logged by another task that is logging another | |
3909 | * file which happens to refer to the same extent as well. Such races | |
3910 | * can leave checksum items in the log with overlapping ranges. | |
3911 | */ | |
3912 | ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr, | |
3913 | lock_end, &cached_state); | |
3914 | if (ret) | |
3915 | return ret; | |
40e046ac FM |
3916 | /* |
3917 | * Due to extent cloning, we might have logged a csum item that covers a | |
3918 | * subrange of a cloned extent, and later we can end up logging a csum | |
3919 | * item for a larger subrange of the same extent or the entire range. | |
3920 | * This would leave csum items in the log tree that cover the same range | |
3921 | * and break the searches for checksums in the log tree, resulting in | |
3922 | * some checksums missing in the fs/subvolume tree. So just delete (or | |
3923 | * trim and adjust) any existing csum items in the log for this range. | |
3924 | */ | |
3925 | ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len); | |
e289f03e FM |
3926 | if (!ret) |
3927 | ret = btrfs_csum_file_blocks(trans, log_root, sums); | |
40e046ac | 3928 | |
e289f03e FM |
3929 | unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end, |
3930 | &cached_state); | |
3931 | ||
3932 | return ret; | |
40e046ac FM |
3933 | } |
3934 | ||
31ff1cd2 | 3935 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 3936 | struct btrfs_inode *inode, |
31ff1cd2 | 3937 | struct btrfs_path *dst_path, |
0e56315c | 3938 | struct btrfs_path *src_path, |
1a4bcf47 FM |
3939 | int start_slot, int nr, int inode_only, |
3940 | u64 logged_isize) | |
31ff1cd2 | 3941 | { |
3ffbd68c | 3942 | struct btrfs_fs_info *fs_info = trans->fs_info; |
31ff1cd2 CM |
3943 | unsigned long src_offset; |
3944 | unsigned long dst_offset; | |
44d70e19 | 3945 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 CM |
3946 | struct btrfs_file_extent_item *extent; |
3947 | struct btrfs_inode_item *inode_item; | |
16e7549f | 3948 | struct extent_buffer *src = src_path->nodes[0]; |
31ff1cd2 CM |
3949 | int ret; |
3950 | struct btrfs_key *ins_keys; | |
3951 | u32 *ins_sizes; | |
3952 | char *ins_data; | |
3953 | int i; | |
d20f7043 | 3954 | struct list_head ordered_sums; |
44d70e19 | 3955 | int skip_csum = inode->flags & BTRFS_INODE_NODATASUM; |
d20f7043 CM |
3956 | |
3957 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3958 | |
3959 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3960 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3961 | if (!ins_data) |
3962 | return -ENOMEM; | |
3963 | ||
31ff1cd2 CM |
3964 | ins_sizes = (u32 *)ins_data; |
3965 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3966 | ||
3967 | for (i = 0; i < nr; i++) { | |
3968 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3969 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3970 | } | |
3971 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3972 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3973 | if (ret) { |
3974 | kfree(ins_data); | |
3975 | return ret; | |
3976 | } | |
31ff1cd2 | 3977 | |
5d4f98a2 | 3978 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3979 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3980 | dst_path->slots[0]); | |
3981 | ||
3982 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3983 | ||
94edf4ae | 3984 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3985 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3986 | dst_path->slots[0], | |
3987 | struct btrfs_inode_item); | |
94edf4ae | 3988 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
f85b7379 DS |
3989 | &inode->vfs_inode, |
3990 | inode_only == LOG_INODE_EXISTS, | |
1a4bcf47 | 3991 | logged_isize); |
94edf4ae JB |
3992 | } else { |
3993 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3994 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 3995 | } |
94edf4ae | 3996 | |
31ff1cd2 CM |
3997 | /* take a reference on file data extents so that truncates |
3998 | * or deletes of this inode don't have to relog the inode | |
3999 | * again | |
4000 | */ | |
962a298f | 4001 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 4002 | !skip_csum) { |
31ff1cd2 CM |
4003 | int found_type; |
4004 | extent = btrfs_item_ptr(src, start_slot + i, | |
4005 | struct btrfs_file_extent_item); | |
4006 | ||
8e531cdf | 4007 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
4008 | continue; | |
4009 | ||
31ff1cd2 | 4010 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 4011 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
4012 | u64 ds, dl, cs, cl; |
4013 | ds = btrfs_file_extent_disk_bytenr(src, | |
4014 | extent); | |
4015 | /* ds == 0 is a hole */ | |
4016 | if (ds == 0) | |
4017 | continue; | |
4018 | ||
4019 | dl = btrfs_file_extent_disk_num_bytes(src, | |
4020 | extent); | |
4021 | cs = btrfs_file_extent_offset(src, extent); | |
4022 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 4023 | extent); |
580afd76 CM |
4024 | if (btrfs_file_extent_compression(src, |
4025 | extent)) { | |
4026 | cs = 0; | |
4027 | cl = dl; | |
4028 | } | |
5d4f98a2 YZ |
4029 | |
4030 | ret = btrfs_lookup_csums_range( | |
0b246afa | 4031 | fs_info->csum_root, |
5d4f98a2 | 4032 | ds + cs, ds + cs + cl - 1, |
a2de733c | 4033 | &ordered_sums, 0); |
4f26433e FM |
4034 | if (ret) |
4035 | break; | |
31ff1cd2 CM |
4036 | } |
4037 | } | |
31ff1cd2 CM |
4038 | } |
4039 | ||
4040 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 4041 | btrfs_release_path(dst_path); |
31ff1cd2 | 4042 | kfree(ins_data); |
d20f7043 CM |
4043 | |
4044 | /* | |
4045 | * we have to do this after the loop above to avoid changing the | |
4046 | * log tree while trying to change the log tree. | |
4047 | */ | |
d397712b | 4048 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
4049 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
4050 | struct btrfs_ordered_sum, | |
4051 | list); | |
4a500fd1 | 4052 | if (!ret) |
3ebac17c | 4053 | ret = log_csums(trans, inode, log, sums); |
d20f7043 CM |
4054 | list_del(&sums->list); |
4055 | kfree(sums); | |
4056 | } | |
16e7549f | 4057 | |
4a500fd1 | 4058 | return ret; |
31ff1cd2 CM |
4059 | } |
4060 | ||
5dc562c5 JB |
4061 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
4062 | { | |
4063 | struct extent_map *em1, *em2; | |
4064 | ||
4065 | em1 = list_entry(a, struct extent_map, list); | |
4066 | em2 = list_entry(b, struct extent_map, list); | |
4067 | ||
4068 | if (em1->start < em2->start) | |
4069 | return -1; | |
4070 | else if (em1->start > em2->start) | |
4071 | return 1; | |
4072 | return 0; | |
4073 | } | |
4074 | ||
e7175a69 JB |
4075 | static int log_extent_csums(struct btrfs_trans_handle *trans, |
4076 | struct btrfs_inode *inode, | |
a9ecb653 | 4077 | struct btrfs_root *log_root, |
48778179 FM |
4078 | const struct extent_map *em, |
4079 | struct btrfs_log_ctx *ctx) | |
5dc562c5 | 4080 | { |
48778179 | 4081 | struct btrfs_ordered_extent *ordered; |
2ab28f32 JB |
4082 | u64 csum_offset; |
4083 | u64 csum_len; | |
48778179 FM |
4084 | u64 mod_start = em->mod_start; |
4085 | u64 mod_len = em->mod_len; | |
8407f553 FM |
4086 | LIST_HEAD(ordered_sums); |
4087 | int ret = 0; | |
0aa4a17d | 4088 | |
e7175a69 JB |
4089 | if (inode->flags & BTRFS_INODE_NODATASUM || |
4090 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
8407f553 | 4091 | em->block_start == EXTENT_MAP_HOLE) |
70c8a91c | 4092 | return 0; |
5dc562c5 | 4093 | |
48778179 FM |
4094 | list_for_each_entry(ordered, &ctx->ordered_extents, log_list) { |
4095 | const u64 ordered_end = ordered->file_offset + ordered->num_bytes; | |
4096 | const u64 mod_end = mod_start + mod_len; | |
4097 | struct btrfs_ordered_sum *sums; | |
4098 | ||
4099 | if (mod_len == 0) | |
4100 | break; | |
4101 | ||
4102 | if (ordered_end <= mod_start) | |
4103 | continue; | |
4104 | if (mod_end <= ordered->file_offset) | |
4105 | break; | |
4106 | ||
4107 | /* | |
4108 | * We are going to copy all the csums on this ordered extent, so | |
4109 | * go ahead and adjust mod_start and mod_len in case this ordered | |
4110 | * extent has already been logged. | |
4111 | */ | |
4112 | if (ordered->file_offset > mod_start) { | |
4113 | if (ordered_end >= mod_end) | |
4114 | mod_len = ordered->file_offset - mod_start; | |
4115 | /* | |
4116 | * If we have this case | |
4117 | * | |
4118 | * |--------- logged extent ---------| | |
4119 | * |----- ordered extent ----| | |
4120 | * | |
4121 | * Just don't mess with mod_start and mod_len, we'll | |
4122 | * just end up logging more csums than we need and it | |
4123 | * will be ok. | |
4124 | */ | |
4125 | } else { | |
4126 | if (ordered_end < mod_end) { | |
4127 | mod_len = mod_end - ordered_end; | |
4128 | mod_start = ordered_end; | |
4129 | } else { | |
4130 | mod_len = 0; | |
4131 | } | |
4132 | } | |
4133 | ||
4134 | /* | |
4135 | * To keep us from looping for the above case of an ordered | |
4136 | * extent that falls inside of the logged extent. | |
4137 | */ | |
4138 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags)) | |
4139 | continue; | |
4140 | ||
4141 | list_for_each_entry(sums, &ordered->list, list) { | |
4142 | ret = log_csums(trans, inode, log_root, sums); | |
4143 | if (ret) | |
4144 | return ret; | |
4145 | } | |
4146 | } | |
4147 | ||
4148 | /* We're done, found all csums in the ordered extents. */ | |
4149 | if (mod_len == 0) | |
4150 | return 0; | |
4151 | ||
e7175a69 | 4152 | /* If we're compressed we have to save the entire range of csums. */ |
488111aa FDBM |
4153 | if (em->compress_type) { |
4154 | csum_offset = 0; | |
8407f553 | 4155 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa | 4156 | } else { |
48778179 FM |
4157 | csum_offset = mod_start - em->start; |
4158 | csum_len = mod_len; | |
488111aa | 4159 | } |
2ab28f32 | 4160 | |
70c8a91c | 4161 | /* block start is already adjusted for the file extent offset. */ |
a9ecb653 | 4162 | ret = btrfs_lookup_csums_range(trans->fs_info->csum_root, |
70c8a91c JB |
4163 | em->block_start + csum_offset, |
4164 | em->block_start + csum_offset + | |
4165 | csum_len - 1, &ordered_sums, 0); | |
4166 | if (ret) | |
4167 | return ret; | |
5dc562c5 | 4168 | |
70c8a91c JB |
4169 | while (!list_empty(&ordered_sums)) { |
4170 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4171 | struct btrfs_ordered_sum, | |
4172 | list); | |
4173 | if (!ret) | |
3ebac17c | 4174 | ret = log_csums(trans, inode, log_root, sums); |
70c8a91c JB |
4175 | list_del(&sums->list); |
4176 | kfree(sums); | |
5dc562c5 JB |
4177 | } |
4178 | ||
70c8a91c | 4179 | return ret; |
5dc562c5 JB |
4180 | } |
4181 | ||
8407f553 | 4182 | static int log_one_extent(struct btrfs_trans_handle *trans, |
9d122629 | 4183 | struct btrfs_inode *inode, struct btrfs_root *root, |
8407f553 FM |
4184 | const struct extent_map *em, |
4185 | struct btrfs_path *path, | |
8407f553 FM |
4186 | struct btrfs_log_ctx *ctx) |
4187 | { | |
5893dfb9 | 4188 | struct btrfs_drop_extents_args drop_args = { 0 }; |
8407f553 FM |
4189 | struct btrfs_root *log = root->log_root; |
4190 | struct btrfs_file_extent_item *fi; | |
4191 | struct extent_buffer *leaf; | |
4192 | struct btrfs_map_token token; | |
4193 | struct btrfs_key key; | |
4194 | u64 extent_offset = em->start - em->orig_start; | |
4195 | u64 block_len; | |
4196 | int ret; | |
8407f553 | 4197 | |
48778179 | 4198 | ret = log_extent_csums(trans, inode, log, em, ctx); |
8407f553 FM |
4199 | if (ret) |
4200 | return ret; | |
4201 | ||
5893dfb9 FM |
4202 | drop_args.path = path; |
4203 | drop_args.start = em->start; | |
4204 | drop_args.end = em->start + em->len; | |
4205 | drop_args.replace_extent = true; | |
4206 | drop_args.extent_item_size = sizeof(*fi); | |
4207 | ret = btrfs_drop_extents(trans, log, inode, &drop_args); | |
8407f553 FM |
4208 | if (ret) |
4209 | return ret; | |
4210 | ||
5893dfb9 | 4211 | if (!drop_args.extent_inserted) { |
9d122629 | 4212 | key.objectid = btrfs_ino(inode); |
8407f553 FM |
4213 | key.type = BTRFS_EXTENT_DATA_KEY; |
4214 | key.offset = em->start; | |
4215 | ||
4216 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4217 | sizeof(*fi)); | |
4218 | if (ret) | |
4219 | return ret; | |
4220 | } | |
4221 | leaf = path->nodes[0]; | |
c82f823c | 4222 | btrfs_init_map_token(&token, leaf); |
8407f553 FM |
4223 | fi = btrfs_item_ptr(leaf, path->slots[0], |
4224 | struct btrfs_file_extent_item); | |
4225 | ||
cc4c13d5 | 4226 | btrfs_set_token_file_extent_generation(&token, fi, trans->transid); |
8407f553 | 4227 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
cc4c13d5 DS |
4228 | btrfs_set_token_file_extent_type(&token, fi, |
4229 | BTRFS_FILE_EXTENT_PREALLOC); | |
8407f553 | 4230 | else |
cc4c13d5 DS |
4231 | btrfs_set_token_file_extent_type(&token, fi, |
4232 | BTRFS_FILE_EXTENT_REG); | |
8407f553 FM |
4233 | |
4234 | block_len = max(em->block_len, em->orig_block_len); | |
4235 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
cc4c13d5 DS |
4236 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, |
4237 | em->block_start); | |
4238 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len); | |
8407f553 | 4239 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { |
cc4c13d5 | 4240 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, |
8407f553 | 4241 | em->block_start - |
cc4c13d5 DS |
4242 | extent_offset); |
4243 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len); | |
8407f553 | 4244 | } else { |
cc4c13d5 DS |
4245 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0); |
4246 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0); | |
8407f553 FM |
4247 | } |
4248 | ||
cc4c13d5 DS |
4249 | btrfs_set_token_file_extent_offset(&token, fi, extent_offset); |
4250 | btrfs_set_token_file_extent_num_bytes(&token, fi, em->len); | |
4251 | btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes); | |
4252 | btrfs_set_token_file_extent_compression(&token, fi, em->compress_type); | |
4253 | btrfs_set_token_file_extent_encryption(&token, fi, 0); | |
4254 | btrfs_set_token_file_extent_other_encoding(&token, fi, 0); | |
8407f553 FM |
4255 | btrfs_mark_buffer_dirty(leaf); |
4256 | ||
4257 | btrfs_release_path(path); | |
4258 | ||
4259 | return ret; | |
4260 | } | |
4261 | ||
31d11b83 FM |
4262 | /* |
4263 | * Log all prealloc extents beyond the inode's i_size to make sure we do not | |
4264 | * lose them after doing a fast fsync and replaying the log. We scan the | |
4265 | * subvolume's root instead of iterating the inode's extent map tree because | |
4266 | * otherwise we can log incorrect extent items based on extent map conversion. | |
4267 | * That can happen due to the fact that extent maps are merged when they | |
4268 | * are not in the extent map tree's list of modified extents. | |
4269 | */ | |
4270 | static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans, | |
4271 | struct btrfs_inode *inode, | |
4272 | struct btrfs_path *path) | |
4273 | { | |
4274 | struct btrfs_root *root = inode->root; | |
4275 | struct btrfs_key key; | |
4276 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
4277 | const u64 ino = btrfs_ino(inode); | |
4278 | struct btrfs_path *dst_path = NULL; | |
0e56315c | 4279 | bool dropped_extents = false; |
f135cea3 FM |
4280 | u64 truncate_offset = i_size; |
4281 | struct extent_buffer *leaf; | |
4282 | int slot; | |
31d11b83 FM |
4283 | int ins_nr = 0; |
4284 | int start_slot; | |
4285 | int ret; | |
4286 | ||
4287 | if (!(inode->flags & BTRFS_INODE_PREALLOC)) | |
4288 | return 0; | |
4289 | ||
4290 | key.objectid = ino; | |
4291 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4292 | key.offset = i_size; | |
4293 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4294 | if (ret < 0) | |
4295 | goto out; | |
4296 | ||
f135cea3 FM |
4297 | /* |
4298 | * We must check if there is a prealloc extent that starts before the | |
4299 | * i_size and crosses the i_size boundary. This is to ensure later we | |
4300 | * truncate down to the end of that extent and not to the i_size, as | |
4301 | * otherwise we end up losing part of the prealloc extent after a log | |
4302 | * replay and with an implicit hole if there is another prealloc extent | |
4303 | * that starts at an offset beyond i_size. | |
4304 | */ | |
4305 | ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY); | |
4306 | if (ret < 0) | |
4307 | goto out; | |
4308 | ||
4309 | if (ret == 0) { | |
4310 | struct btrfs_file_extent_item *ei; | |
4311 | ||
4312 | leaf = path->nodes[0]; | |
4313 | slot = path->slots[0]; | |
4314 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4315 | ||
4316 | if (btrfs_file_extent_type(leaf, ei) == | |
4317 | BTRFS_FILE_EXTENT_PREALLOC) { | |
4318 | u64 extent_end; | |
4319 | ||
4320 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4321 | extent_end = key.offset + | |
4322 | btrfs_file_extent_num_bytes(leaf, ei); | |
4323 | ||
4324 | if (extent_end > i_size) | |
4325 | truncate_offset = extent_end; | |
4326 | } | |
4327 | } else { | |
4328 | ret = 0; | |
4329 | } | |
4330 | ||
31d11b83 | 4331 | while (true) { |
f135cea3 FM |
4332 | leaf = path->nodes[0]; |
4333 | slot = path->slots[0]; | |
31d11b83 FM |
4334 | |
4335 | if (slot >= btrfs_header_nritems(leaf)) { | |
4336 | if (ins_nr > 0) { | |
4337 | ret = copy_items(trans, inode, dst_path, path, | |
0e56315c | 4338 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4339 | if (ret < 0) |
4340 | goto out; | |
4341 | ins_nr = 0; | |
4342 | } | |
4343 | ret = btrfs_next_leaf(root, path); | |
4344 | if (ret < 0) | |
4345 | goto out; | |
4346 | if (ret > 0) { | |
4347 | ret = 0; | |
4348 | break; | |
4349 | } | |
4350 | continue; | |
4351 | } | |
4352 | ||
4353 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4354 | if (key.objectid > ino) | |
4355 | break; | |
4356 | if (WARN_ON_ONCE(key.objectid < ino) || | |
4357 | key.type < BTRFS_EXTENT_DATA_KEY || | |
4358 | key.offset < i_size) { | |
4359 | path->slots[0]++; | |
4360 | continue; | |
4361 | } | |
0e56315c | 4362 | if (!dropped_extents) { |
31d11b83 FM |
4363 | /* |
4364 | * Avoid logging extent items logged in past fsync calls | |
4365 | * and leading to duplicate keys in the log tree. | |
4366 | */ | |
4367 | do { | |
4368 | ret = btrfs_truncate_inode_items(trans, | |
4369 | root->log_root, | |
50743398 | 4370 | inode, truncate_offset, |
31d11b83 FM |
4371 | BTRFS_EXTENT_DATA_KEY); |
4372 | } while (ret == -EAGAIN); | |
4373 | if (ret) | |
4374 | goto out; | |
0e56315c | 4375 | dropped_extents = true; |
31d11b83 FM |
4376 | } |
4377 | if (ins_nr == 0) | |
4378 | start_slot = slot; | |
4379 | ins_nr++; | |
4380 | path->slots[0]++; | |
4381 | if (!dst_path) { | |
4382 | dst_path = btrfs_alloc_path(); | |
4383 | if (!dst_path) { | |
4384 | ret = -ENOMEM; | |
4385 | goto out; | |
4386 | } | |
4387 | } | |
4388 | } | |
0bc2d3c0 | 4389 | if (ins_nr > 0) |
0e56315c | 4390 | ret = copy_items(trans, inode, dst_path, path, |
31d11b83 | 4391 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4392 | out: |
4393 | btrfs_release_path(path); | |
4394 | btrfs_free_path(dst_path); | |
4395 | return ret; | |
4396 | } | |
4397 | ||
5dc562c5 JB |
4398 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4399 | struct btrfs_root *root, | |
9d122629 | 4400 | struct btrfs_inode *inode, |
827463c4 | 4401 | struct btrfs_path *path, |
48778179 | 4402 | struct btrfs_log_ctx *ctx) |
5dc562c5 | 4403 | { |
48778179 FM |
4404 | struct btrfs_ordered_extent *ordered; |
4405 | struct btrfs_ordered_extent *tmp; | |
5dc562c5 JB |
4406 | struct extent_map *em, *n; |
4407 | struct list_head extents; | |
9d122629 | 4408 | struct extent_map_tree *tree = &inode->extent_tree; |
5dc562c5 | 4409 | int ret = 0; |
2ab28f32 | 4410 | int num = 0; |
5dc562c5 JB |
4411 | |
4412 | INIT_LIST_HEAD(&extents); | |
4413 | ||
5dc562c5 | 4414 | write_lock(&tree->lock); |
5dc562c5 JB |
4415 | |
4416 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
4417 | list_del_init(&em->list); | |
2ab28f32 JB |
4418 | /* |
4419 | * Just an arbitrary number, this can be really CPU intensive | |
4420 | * once we start getting a lot of extents, and really once we | |
4421 | * have a bunch of extents we just want to commit since it will | |
4422 | * be faster. | |
4423 | */ | |
4424 | if (++num > 32768) { | |
4425 | list_del_init(&tree->modified_extents); | |
4426 | ret = -EFBIG; | |
4427 | goto process; | |
4428 | } | |
4429 | ||
5f96bfb7 | 4430 | if (em->generation < trans->transid) |
5dc562c5 | 4431 | continue; |
8c6c5928 | 4432 | |
31d11b83 FM |
4433 | /* We log prealloc extents beyond eof later. */ |
4434 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && | |
4435 | em->start >= i_size_read(&inode->vfs_inode)) | |
4436 | continue; | |
4437 | ||
ff44c6e3 | 4438 | /* Need a ref to keep it from getting evicted from cache */ |
490b54d6 | 4439 | refcount_inc(&em->refs); |
ff44c6e3 | 4440 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); |
5dc562c5 | 4441 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4442 | num++; |
5dc562c5 JB |
4443 | } |
4444 | ||
4445 | list_sort(NULL, &extents, extent_cmp); | |
2ab28f32 | 4446 | process: |
5dc562c5 JB |
4447 | while (!list_empty(&extents)) { |
4448 | em = list_entry(extents.next, struct extent_map, list); | |
4449 | ||
4450 | list_del_init(&em->list); | |
4451 | ||
4452 | /* | |
4453 | * If we had an error we just need to delete everybody from our | |
4454 | * private list. | |
4455 | */ | |
ff44c6e3 | 4456 | if (ret) { |
201a9038 | 4457 | clear_em_logging(tree, em); |
ff44c6e3 | 4458 | free_extent_map(em); |
5dc562c5 | 4459 | continue; |
ff44c6e3 JB |
4460 | } |
4461 | ||
4462 | write_unlock(&tree->lock); | |
5dc562c5 | 4463 | |
a2120a47 | 4464 | ret = log_one_extent(trans, inode, root, em, path, ctx); |
ff44c6e3 | 4465 | write_lock(&tree->lock); |
201a9038 JB |
4466 | clear_em_logging(tree, em); |
4467 | free_extent_map(em); | |
5dc562c5 | 4468 | } |
ff44c6e3 JB |
4469 | WARN_ON(!list_empty(&extents)); |
4470 | write_unlock(&tree->lock); | |
5dc562c5 | 4471 | |
5dc562c5 | 4472 | btrfs_release_path(path); |
31d11b83 FM |
4473 | if (!ret) |
4474 | ret = btrfs_log_prealloc_extents(trans, inode, path); | |
48778179 FM |
4475 | if (ret) |
4476 | return ret; | |
31d11b83 | 4477 | |
48778179 FM |
4478 | /* |
4479 | * We have logged all extents successfully, now make sure the commit of | |
4480 | * the current transaction waits for the ordered extents to complete | |
4481 | * before it commits and wipes out the log trees, otherwise we would | |
4482 | * lose data if an ordered extents completes after the transaction | |
4483 | * commits and a power failure happens after the transaction commit. | |
4484 | */ | |
4485 | list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) { | |
4486 | list_del_init(&ordered->log_list); | |
4487 | set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags); | |
4488 | ||
4489 | if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) { | |
4490 | spin_lock_irq(&inode->ordered_tree.lock); | |
4491 | if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) { | |
4492 | set_bit(BTRFS_ORDERED_PENDING, &ordered->flags); | |
4493 | atomic_inc(&trans->transaction->pending_ordered); | |
4494 | } | |
4495 | spin_unlock_irq(&inode->ordered_tree.lock); | |
4496 | } | |
4497 | btrfs_put_ordered_extent(ordered); | |
4498 | } | |
4499 | ||
4500 | return 0; | |
5dc562c5 JB |
4501 | } |
4502 | ||
481b01c0 | 4503 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
4504 | struct btrfs_path *path, u64 *size_ret) |
4505 | { | |
4506 | struct btrfs_key key; | |
4507 | int ret; | |
4508 | ||
481b01c0 | 4509 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
4510 | key.type = BTRFS_INODE_ITEM_KEY; |
4511 | key.offset = 0; | |
4512 | ||
4513 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4514 | if (ret < 0) { | |
4515 | return ret; | |
4516 | } else if (ret > 0) { | |
2f2ff0ee | 4517 | *size_ret = 0; |
1a4bcf47 FM |
4518 | } else { |
4519 | struct btrfs_inode_item *item; | |
4520 | ||
4521 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4522 | struct btrfs_inode_item); | |
4523 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
bf504110 FM |
4524 | /* |
4525 | * If the in-memory inode's i_size is smaller then the inode | |
4526 | * size stored in the btree, return the inode's i_size, so | |
4527 | * that we get a correct inode size after replaying the log | |
4528 | * when before a power failure we had a shrinking truncate | |
4529 | * followed by addition of a new name (rename / new hard link). | |
4530 | * Otherwise return the inode size from the btree, to avoid | |
4531 | * data loss when replaying a log due to previously doing a | |
4532 | * write that expands the inode's size and logging a new name | |
4533 | * immediately after. | |
4534 | */ | |
4535 | if (*size_ret > inode->vfs_inode.i_size) | |
4536 | *size_ret = inode->vfs_inode.i_size; | |
1a4bcf47 FM |
4537 | } |
4538 | ||
4539 | btrfs_release_path(path); | |
4540 | return 0; | |
4541 | } | |
4542 | ||
36283bf7 FM |
4543 | /* |
4544 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4545 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4546 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4547 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4548 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4549 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4550 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4551 | */ | |
4552 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4553 | struct btrfs_root *root, | |
1a93c36a | 4554 | struct btrfs_inode *inode, |
36283bf7 FM |
4555 | struct btrfs_path *path, |
4556 | struct btrfs_path *dst_path) | |
4557 | { | |
4558 | int ret; | |
4559 | struct btrfs_key key; | |
1a93c36a | 4560 | const u64 ino = btrfs_ino(inode); |
36283bf7 FM |
4561 | int ins_nr = 0; |
4562 | int start_slot = 0; | |
f2f121ab FM |
4563 | bool found_xattrs = false; |
4564 | ||
4565 | if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags)) | |
4566 | return 0; | |
36283bf7 FM |
4567 | |
4568 | key.objectid = ino; | |
4569 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4570 | key.offset = 0; | |
4571 | ||
4572 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4573 | if (ret < 0) | |
4574 | return ret; | |
4575 | ||
4576 | while (true) { | |
4577 | int slot = path->slots[0]; | |
4578 | struct extent_buffer *leaf = path->nodes[0]; | |
4579 | int nritems = btrfs_header_nritems(leaf); | |
4580 | ||
4581 | if (slot >= nritems) { | |
4582 | if (ins_nr > 0) { | |
1a93c36a | 4583 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 4584 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
4585 | if (ret < 0) |
4586 | return ret; | |
4587 | ins_nr = 0; | |
4588 | } | |
4589 | ret = btrfs_next_leaf(root, path); | |
4590 | if (ret < 0) | |
4591 | return ret; | |
4592 | else if (ret > 0) | |
4593 | break; | |
4594 | continue; | |
4595 | } | |
4596 | ||
4597 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4598 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4599 | break; | |
4600 | ||
4601 | if (ins_nr == 0) | |
4602 | start_slot = slot; | |
4603 | ins_nr++; | |
4604 | path->slots[0]++; | |
f2f121ab | 4605 | found_xattrs = true; |
36283bf7 FM |
4606 | cond_resched(); |
4607 | } | |
4608 | if (ins_nr > 0) { | |
1a93c36a | 4609 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 4610 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
4611 | if (ret < 0) |
4612 | return ret; | |
4613 | } | |
4614 | ||
f2f121ab FM |
4615 | if (!found_xattrs) |
4616 | set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags); | |
4617 | ||
36283bf7 FM |
4618 | return 0; |
4619 | } | |
4620 | ||
a89ca6f2 | 4621 | /* |
0e56315c FM |
4622 | * When using the NO_HOLES feature if we punched a hole that causes the |
4623 | * deletion of entire leafs or all the extent items of the first leaf (the one | |
4624 | * that contains the inode item and references) we may end up not processing | |
4625 | * any extents, because there are no leafs with a generation matching the | |
4626 | * current transaction that have extent items for our inode. So we need to find | |
4627 | * if any holes exist and then log them. We also need to log holes after any | |
4628 | * truncate operation that changes the inode's size. | |
a89ca6f2 | 4629 | */ |
0e56315c FM |
4630 | static int btrfs_log_holes(struct btrfs_trans_handle *trans, |
4631 | struct btrfs_root *root, | |
4632 | struct btrfs_inode *inode, | |
7af59743 | 4633 | struct btrfs_path *path) |
a89ca6f2 | 4634 | { |
0b246afa | 4635 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 | 4636 | struct btrfs_key key; |
a0308dd7 NB |
4637 | const u64 ino = btrfs_ino(inode); |
4638 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
7af59743 | 4639 | u64 prev_extent_end = 0; |
0e56315c | 4640 | int ret; |
a89ca6f2 | 4641 | |
0e56315c | 4642 | if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0) |
a89ca6f2 FM |
4643 | return 0; |
4644 | ||
4645 | key.objectid = ino; | |
4646 | key.type = BTRFS_EXTENT_DATA_KEY; | |
7af59743 | 4647 | key.offset = 0; |
a89ca6f2 FM |
4648 | |
4649 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
a89ca6f2 FM |
4650 | if (ret < 0) |
4651 | return ret; | |
4652 | ||
0e56315c | 4653 | while (true) { |
0e56315c | 4654 | struct extent_buffer *leaf = path->nodes[0]; |
a89ca6f2 | 4655 | |
0e56315c FM |
4656 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
4657 | ret = btrfs_next_leaf(root, path); | |
4658 | if (ret < 0) | |
4659 | return ret; | |
4660 | if (ret > 0) { | |
4661 | ret = 0; | |
4662 | break; | |
4663 | } | |
4664 | leaf = path->nodes[0]; | |
4665 | } | |
4666 | ||
4667 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4668 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
4669 | break; | |
4670 | ||
4671 | /* We have a hole, log it. */ | |
4672 | if (prev_extent_end < key.offset) { | |
7af59743 | 4673 | const u64 hole_len = key.offset - prev_extent_end; |
0e56315c FM |
4674 | |
4675 | /* | |
4676 | * Release the path to avoid deadlocks with other code | |
4677 | * paths that search the root while holding locks on | |
4678 | * leafs from the log root. | |
4679 | */ | |
4680 | btrfs_release_path(path); | |
4681 | ret = btrfs_insert_file_extent(trans, root->log_root, | |
4682 | ino, prev_extent_end, 0, | |
4683 | 0, hole_len, 0, hole_len, | |
4684 | 0, 0, 0); | |
4685 | if (ret < 0) | |
4686 | return ret; | |
4687 | ||
4688 | /* | |
4689 | * Search for the same key again in the root. Since it's | |
4690 | * an extent item and we are holding the inode lock, the | |
4691 | * key must still exist. If it doesn't just emit warning | |
4692 | * and return an error to fall back to a transaction | |
4693 | * commit. | |
4694 | */ | |
4695 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4696 | if (ret < 0) | |
4697 | return ret; | |
4698 | if (WARN_ON(ret > 0)) | |
4699 | return -ENOENT; | |
4700 | leaf = path->nodes[0]; | |
4701 | } | |
a89ca6f2 | 4702 | |
7af59743 | 4703 | prev_extent_end = btrfs_file_extent_end(path); |
0e56315c FM |
4704 | path->slots[0]++; |
4705 | cond_resched(); | |
a89ca6f2 | 4706 | } |
a89ca6f2 | 4707 | |
7af59743 | 4708 | if (prev_extent_end < i_size) { |
0e56315c | 4709 | u64 hole_len; |
a89ca6f2 | 4710 | |
0e56315c | 4711 | btrfs_release_path(path); |
7af59743 | 4712 | hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize); |
0e56315c FM |
4713 | ret = btrfs_insert_file_extent(trans, root->log_root, |
4714 | ino, prev_extent_end, 0, 0, | |
4715 | hole_len, 0, hole_len, | |
4716 | 0, 0, 0); | |
4717 | if (ret < 0) | |
4718 | return ret; | |
4719 | } | |
4720 | ||
4721 | return 0; | |
a89ca6f2 FM |
4722 | } |
4723 | ||
56f23fdb FM |
4724 | /* |
4725 | * When we are logging a new inode X, check if it doesn't have a reference that | |
4726 | * matches the reference from some other inode Y created in a past transaction | |
4727 | * and that was renamed in the current transaction. If we don't do this, then at | |
4728 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
4729 | * | |
4730 | * mkdir /mnt/x | |
4731 | * echo "hello world" > /mnt/x/foobar | |
4732 | * sync | |
4733 | * mv /mnt/x /mnt/y | |
4734 | * mkdir /mnt/x # or touch /mnt/x | |
4735 | * xfs_io -c fsync /mnt/x | |
4736 | * <power fail> | |
4737 | * mount fs, trigger log replay | |
4738 | * | |
4739 | * After the log replay procedure, we would lose the first directory and all its | |
4740 | * files (file foobar). | |
4741 | * For the case where inode Y is not a directory we simply end up losing it: | |
4742 | * | |
4743 | * echo "123" > /mnt/foo | |
4744 | * sync | |
4745 | * mv /mnt/foo /mnt/bar | |
4746 | * echo "abc" > /mnt/foo | |
4747 | * xfs_io -c fsync /mnt/foo | |
4748 | * <power fail> | |
4749 | * | |
4750 | * We also need this for cases where a snapshot entry is replaced by some other | |
4751 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
4752 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
4753 | * if it were a regular entry: | |
4754 | * | |
4755 | * mkdir /mnt/x | |
4756 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
4757 | * btrfs subvolume delete /mnt/x/snap | |
4758 | * rmdir /mnt/x | |
4759 | * mkdir /mnt/x | |
4760 | * fsync /mnt/x or fsync some new file inside it | |
4761 | * <power fail> | |
4762 | * | |
4763 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
4764 | * the same transaction. | |
4765 | */ | |
4766 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
4767 | const int slot, | |
4768 | const struct btrfs_key *key, | |
4791c8f1 | 4769 | struct btrfs_inode *inode, |
a3baaf0d | 4770 | u64 *other_ino, u64 *other_parent) |
56f23fdb FM |
4771 | { |
4772 | int ret; | |
4773 | struct btrfs_path *search_path; | |
4774 | char *name = NULL; | |
4775 | u32 name_len = 0; | |
4776 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
4777 | u32 cur_offset = 0; | |
4778 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
4779 | ||
4780 | search_path = btrfs_alloc_path(); | |
4781 | if (!search_path) | |
4782 | return -ENOMEM; | |
4783 | search_path->search_commit_root = 1; | |
4784 | search_path->skip_locking = 1; | |
4785 | ||
4786 | while (cur_offset < item_size) { | |
4787 | u64 parent; | |
4788 | u32 this_name_len; | |
4789 | u32 this_len; | |
4790 | unsigned long name_ptr; | |
4791 | struct btrfs_dir_item *di; | |
4792 | ||
4793 | if (key->type == BTRFS_INODE_REF_KEY) { | |
4794 | struct btrfs_inode_ref *iref; | |
4795 | ||
4796 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
4797 | parent = key->offset; | |
4798 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
4799 | name_ptr = (unsigned long)(iref + 1); | |
4800 | this_len = sizeof(*iref) + this_name_len; | |
4801 | } else { | |
4802 | struct btrfs_inode_extref *extref; | |
4803 | ||
4804 | extref = (struct btrfs_inode_extref *)(ptr + | |
4805 | cur_offset); | |
4806 | parent = btrfs_inode_extref_parent(eb, extref); | |
4807 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
4808 | name_ptr = (unsigned long)&extref->name; | |
4809 | this_len = sizeof(*extref) + this_name_len; | |
4810 | } | |
4811 | ||
4812 | if (this_name_len > name_len) { | |
4813 | char *new_name; | |
4814 | ||
4815 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
4816 | if (!new_name) { | |
4817 | ret = -ENOMEM; | |
4818 | goto out; | |
4819 | } | |
4820 | name_len = this_name_len; | |
4821 | name = new_name; | |
4822 | } | |
4823 | ||
4824 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4791c8f1 NB |
4825 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
4826 | parent, name, this_name_len, 0); | |
56f23fdb | 4827 | if (di && !IS_ERR(di)) { |
44f714da FM |
4828 | struct btrfs_key di_key; |
4829 | ||
4830 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
4831 | di, &di_key); | |
4832 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
6b5fc433 FM |
4833 | if (di_key.objectid != key->objectid) { |
4834 | ret = 1; | |
4835 | *other_ino = di_key.objectid; | |
a3baaf0d | 4836 | *other_parent = parent; |
6b5fc433 FM |
4837 | } else { |
4838 | ret = 0; | |
4839 | } | |
44f714da FM |
4840 | } else { |
4841 | ret = -EAGAIN; | |
4842 | } | |
56f23fdb FM |
4843 | goto out; |
4844 | } else if (IS_ERR(di)) { | |
4845 | ret = PTR_ERR(di); | |
4846 | goto out; | |
4847 | } | |
4848 | btrfs_release_path(search_path); | |
4849 | ||
4850 | cur_offset += this_len; | |
4851 | } | |
4852 | ret = 0; | |
4853 | out: | |
4854 | btrfs_free_path(search_path); | |
4855 | kfree(name); | |
4856 | return ret; | |
4857 | } | |
4858 | ||
6b5fc433 FM |
4859 | struct btrfs_ino_list { |
4860 | u64 ino; | |
a3baaf0d | 4861 | u64 parent; |
6b5fc433 FM |
4862 | struct list_head list; |
4863 | }; | |
4864 | ||
4865 | static int log_conflicting_inodes(struct btrfs_trans_handle *trans, | |
4866 | struct btrfs_root *root, | |
4867 | struct btrfs_path *path, | |
4868 | struct btrfs_log_ctx *ctx, | |
a3baaf0d | 4869 | u64 ino, u64 parent) |
6b5fc433 FM |
4870 | { |
4871 | struct btrfs_ino_list *ino_elem; | |
4872 | LIST_HEAD(inode_list); | |
4873 | int ret = 0; | |
4874 | ||
4875 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
4876 | if (!ino_elem) | |
4877 | return -ENOMEM; | |
4878 | ino_elem->ino = ino; | |
a3baaf0d | 4879 | ino_elem->parent = parent; |
6b5fc433 FM |
4880 | list_add_tail(&ino_elem->list, &inode_list); |
4881 | ||
4882 | while (!list_empty(&inode_list)) { | |
4883 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4884 | struct btrfs_key key; | |
4885 | struct inode *inode; | |
4886 | ||
4887 | ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list, | |
4888 | list); | |
4889 | ino = ino_elem->ino; | |
a3baaf0d | 4890 | parent = ino_elem->parent; |
6b5fc433 FM |
4891 | list_del(&ino_elem->list); |
4892 | kfree(ino_elem); | |
4893 | if (ret) | |
4894 | continue; | |
4895 | ||
4896 | btrfs_release_path(path); | |
4897 | ||
0202e83f | 4898 | inode = btrfs_iget(fs_info->sb, ino, root); |
6b5fc433 FM |
4899 | /* |
4900 | * If the other inode that had a conflicting dir entry was | |
a3baaf0d FM |
4901 | * deleted in the current transaction, we need to log its parent |
4902 | * directory. | |
6b5fc433 FM |
4903 | */ |
4904 | if (IS_ERR(inode)) { | |
4905 | ret = PTR_ERR(inode); | |
a3baaf0d | 4906 | if (ret == -ENOENT) { |
0202e83f | 4907 | inode = btrfs_iget(fs_info->sb, parent, root); |
a3baaf0d FM |
4908 | if (IS_ERR(inode)) { |
4909 | ret = PTR_ERR(inode); | |
4910 | } else { | |
4911 | ret = btrfs_log_inode(trans, root, | |
4912 | BTRFS_I(inode), | |
4913 | LOG_OTHER_INODE_ALL, | |
48778179 | 4914 | ctx); |
410f954c | 4915 | btrfs_add_delayed_iput(inode); |
a3baaf0d FM |
4916 | } |
4917 | } | |
6b5fc433 FM |
4918 | continue; |
4919 | } | |
b5e4ff9d FM |
4920 | /* |
4921 | * If the inode was already logged skip it - otherwise we can | |
4922 | * hit an infinite loop. Example: | |
4923 | * | |
4924 | * From the commit root (previous transaction) we have the | |
4925 | * following inodes: | |
4926 | * | |
4927 | * inode 257 a directory | |
4928 | * inode 258 with references "zz" and "zz_link" on inode 257 | |
4929 | * inode 259 with reference "a" on inode 257 | |
4930 | * | |
4931 | * And in the current (uncommitted) transaction we have: | |
4932 | * | |
4933 | * inode 257 a directory, unchanged | |
4934 | * inode 258 with references "a" and "a2" on inode 257 | |
4935 | * inode 259 with reference "zz_link" on inode 257 | |
4936 | * inode 261 with reference "zz" on inode 257 | |
4937 | * | |
4938 | * When logging inode 261 the following infinite loop could | |
4939 | * happen if we don't skip already logged inodes: | |
4940 | * | |
4941 | * - we detect inode 258 as a conflicting inode, with inode 261 | |
4942 | * on reference "zz", and log it; | |
4943 | * | |
4944 | * - we detect inode 259 as a conflicting inode, with inode 258 | |
4945 | * on reference "a", and log it; | |
4946 | * | |
4947 | * - we detect inode 258 as a conflicting inode, with inode 259 | |
4948 | * on reference "zz_link", and log it - again! After this we | |
4949 | * repeat the above steps forever. | |
4950 | */ | |
4951 | spin_lock(&BTRFS_I(inode)->lock); | |
4952 | /* | |
4953 | * Check the inode's logged_trans only instead of | |
4954 | * btrfs_inode_in_log(). This is because the last_log_commit of | |
4955 | * the inode is not updated when we only log that it exists and | |
260db43c | 4956 | * it has the full sync bit set (see btrfs_log_inode()). |
b5e4ff9d FM |
4957 | */ |
4958 | if (BTRFS_I(inode)->logged_trans == trans->transid) { | |
4959 | spin_unlock(&BTRFS_I(inode)->lock); | |
4960 | btrfs_add_delayed_iput(inode); | |
4961 | continue; | |
4962 | } | |
4963 | spin_unlock(&BTRFS_I(inode)->lock); | |
6b5fc433 FM |
4964 | /* |
4965 | * We are safe logging the other inode without acquiring its | |
4966 | * lock as long as we log with the LOG_INODE_EXISTS mode. We | |
4967 | * are safe against concurrent renames of the other inode as | |
4968 | * well because during a rename we pin the log and update the | |
4969 | * log with the new name before we unpin it. | |
4970 | */ | |
4971 | ret = btrfs_log_inode(trans, root, BTRFS_I(inode), | |
48778179 | 4972 | LOG_OTHER_INODE, ctx); |
6b5fc433 | 4973 | if (ret) { |
410f954c | 4974 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
4975 | continue; |
4976 | } | |
4977 | ||
4978 | key.objectid = ino; | |
4979 | key.type = BTRFS_INODE_REF_KEY; | |
4980 | key.offset = 0; | |
4981 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4982 | if (ret < 0) { | |
410f954c | 4983 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
4984 | continue; |
4985 | } | |
4986 | ||
4987 | while (true) { | |
4988 | struct extent_buffer *leaf = path->nodes[0]; | |
4989 | int slot = path->slots[0]; | |
4990 | u64 other_ino = 0; | |
a3baaf0d | 4991 | u64 other_parent = 0; |
6b5fc433 FM |
4992 | |
4993 | if (slot >= btrfs_header_nritems(leaf)) { | |
4994 | ret = btrfs_next_leaf(root, path); | |
4995 | if (ret < 0) { | |
4996 | break; | |
4997 | } else if (ret > 0) { | |
4998 | ret = 0; | |
4999 | break; | |
5000 | } | |
5001 | continue; | |
5002 | } | |
5003 | ||
5004 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5005 | if (key.objectid != ino || | |
5006 | (key.type != BTRFS_INODE_REF_KEY && | |
5007 | key.type != BTRFS_INODE_EXTREF_KEY)) { | |
5008 | ret = 0; | |
5009 | break; | |
5010 | } | |
5011 | ||
5012 | ret = btrfs_check_ref_name_override(leaf, slot, &key, | |
a3baaf0d FM |
5013 | BTRFS_I(inode), &other_ino, |
5014 | &other_parent); | |
6b5fc433 FM |
5015 | if (ret < 0) |
5016 | break; | |
5017 | if (ret > 0) { | |
5018 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
5019 | if (!ino_elem) { | |
5020 | ret = -ENOMEM; | |
5021 | break; | |
5022 | } | |
5023 | ino_elem->ino = other_ino; | |
a3baaf0d | 5024 | ino_elem->parent = other_parent; |
6b5fc433 FM |
5025 | list_add_tail(&ino_elem->list, &inode_list); |
5026 | ret = 0; | |
5027 | } | |
5028 | path->slots[0]++; | |
5029 | } | |
410f954c | 5030 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
5031 | } |
5032 | ||
5033 | return ret; | |
5034 | } | |
5035 | ||
da447009 FM |
5036 | static int copy_inode_items_to_log(struct btrfs_trans_handle *trans, |
5037 | struct btrfs_inode *inode, | |
5038 | struct btrfs_key *min_key, | |
5039 | const struct btrfs_key *max_key, | |
5040 | struct btrfs_path *path, | |
5041 | struct btrfs_path *dst_path, | |
5042 | const u64 logged_isize, | |
5043 | const bool recursive_logging, | |
5044 | const int inode_only, | |
5045 | struct btrfs_log_ctx *ctx, | |
5046 | bool *need_log_inode_item) | |
5047 | { | |
5048 | struct btrfs_root *root = inode->root; | |
5049 | int ins_start_slot = 0; | |
5050 | int ins_nr = 0; | |
5051 | int ret; | |
5052 | ||
5053 | while (1) { | |
5054 | ret = btrfs_search_forward(root, min_key, path, trans->transid); | |
5055 | if (ret < 0) | |
5056 | return ret; | |
5057 | if (ret > 0) { | |
5058 | ret = 0; | |
5059 | break; | |
5060 | } | |
5061 | again: | |
5062 | /* Note, ins_nr might be > 0 here, cleanup outside the loop */ | |
5063 | if (min_key->objectid != max_key->objectid) | |
5064 | break; | |
5065 | if (min_key->type > max_key->type) | |
5066 | break; | |
5067 | ||
5068 | if (min_key->type == BTRFS_INODE_ITEM_KEY) | |
5069 | *need_log_inode_item = false; | |
5070 | ||
5071 | if ((min_key->type == BTRFS_INODE_REF_KEY || | |
5072 | min_key->type == BTRFS_INODE_EXTREF_KEY) && | |
5073 | inode->generation == trans->transid && | |
5074 | !recursive_logging) { | |
5075 | u64 other_ino = 0; | |
5076 | u64 other_parent = 0; | |
5077 | ||
5078 | ret = btrfs_check_ref_name_override(path->nodes[0], | |
5079 | path->slots[0], min_key, inode, | |
5080 | &other_ino, &other_parent); | |
5081 | if (ret < 0) { | |
5082 | return ret; | |
5083 | } else if (ret > 0 && ctx && | |
5084 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { | |
5085 | if (ins_nr > 0) { | |
5086 | ins_nr++; | |
5087 | } else { | |
5088 | ins_nr = 1; | |
5089 | ins_start_slot = path->slots[0]; | |
5090 | } | |
5091 | ret = copy_items(trans, inode, dst_path, path, | |
5092 | ins_start_slot, ins_nr, | |
5093 | inode_only, logged_isize); | |
5094 | if (ret < 0) | |
5095 | return ret; | |
5096 | ins_nr = 0; | |
5097 | ||
5098 | ret = log_conflicting_inodes(trans, root, path, | |
5099 | ctx, other_ino, other_parent); | |
5100 | if (ret) | |
5101 | return ret; | |
5102 | btrfs_release_path(path); | |
5103 | goto next_key; | |
5104 | } | |
5105 | } | |
5106 | ||
5107 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ | |
5108 | if (min_key->type == BTRFS_XATTR_ITEM_KEY) { | |
5109 | if (ins_nr == 0) | |
5110 | goto next_slot; | |
5111 | ret = copy_items(trans, inode, dst_path, path, | |
5112 | ins_start_slot, | |
5113 | ins_nr, inode_only, logged_isize); | |
5114 | if (ret < 0) | |
5115 | return ret; | |
5116 | ins_nr = 0; | |
5117 | goto next_slot; | |
5118 | } | |
5119 | ||
5120 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { | |
5121 | ins_nr++; | |
5122 | goto next_slot; | |
5123 | } else if (!ins_nr) { | |
5124 | ins_start_slot = path->slots[0]; | |
5125 | ins_nr = 1; | |
5126 | goto next_slot; | |
5127 | } | |
5128 | ||
5129 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, | |
5130 | ins_nr, inode_only, logged_isize); | |
5131 | if (ret < 0) | |
5132 | return ret; | |
5133 | ins_nr = 1; | |
5134 | ins_start_slot = path->slots[0]; | |
5135 | next_slot: | |
5136 | path->slots[0]++; | |
5137 | if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) { | |
5138 | btrfs_item_key_to_cpu(path->nodes[0], min_key, | |
5139 | path->slots[0]); | |
5140 | goto again; | |
5141 | } | |
5142 | if (ins_nr) { | |
5143 | ret = copy_items(trans, inode, dst_path, path, | |
5144 | ins_start_slot, ins_nr, inode_only, | |
5145 | logged_isize); | |
5146 | if (ret < 0) | |
5147 | return ret; | |
5148 | ins_nr = 0; | |
5149 | } | |
5150 | btrfs_release_path(path); | |
5151 | next_key: | |
5152 | if (min_key->offset < (u64)-1) { | |
5153 | min_key->offset++; | |
5154 | } else if (min_key->type < max_key->type) { | |
5155 | min_key->type++; | |
5156 | min_key->offset = 0; | |
5157 | } else { | |
5158 | break; | |
5159 | } | |
5160 | } | |
5161 | if (ins_nr) | |
5162 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, | |
5163 | ins_nr, inode_only, logged_isize); | |
5164 | ||
5165 | return ret; | |
5166 | } | |
5167 | ||
e02119d5 CM |
5168 | /* log a single inode in the tree log. |
5169 | * At least one parent directory for this inode must exist in the tree | |
5170 | * or be logged already. | |
5171 | * | |
5172 | * Any items from this inode changed by the current transaction are copied | |
5173 | * to the log tree. An extra reference is taken on any extents in this | |
5174 | * file, allowing us to avoid a whole pile of corner cases around logging | |
5175 | * blocks that have been removed from the tree. | |
5176 | * | |
5177 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
5178 | * does. | |
5179 | * | |
5180 | * This handles both files and directories. | |
5181 | */ | |
12fcfd22 | 5182 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 5183 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc | 5184 | int inode_only, |
8407f553 | 5185 | struct btrfs_log_ctx *ctx) |
e02119d5 CM |
5186 | { |
5187 | struct btrfs_path *path; | |
5188 | struct btrfs_path *dst_path; | |
5189 | struct btrfs_key min_key; | |
5190 | struct btrfs_key max_key; | |
5191 | struct btrfs_root *log = root->log_root; | |
4a500fd1 | 5192 | int err = 0; |
8c8648dd | 5193 | int ret = 0; |
5dc562c5 | 5194 | bool fast_search = false; |
a59108a7 NB |
5195 | u64 ino = btrfs_ino(inode); |
5196 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
1a4bcf47 | 5197 | u64 logged_isize = 0; |
e4545de5 | 5198 | bool need_log_inode_item = true; |
9a8fca62 | 5199 | bool xattrs_logged = false; |
a3baaf0d | 5200 | bool recursive_logging = false; |
e02119d5 | 5201 | |
e02119d5 | 5202 | path = btrfs_alloc_path(); |
5df67083 TI |
5203 | if (!path) |
5204 | return -ENOMEM; | |
e02119d5 | 5205 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
5206 | if (!dst_path) { |
5207 | btrfs_free_path(path); | |
5208 | return -ENOMEM; | |
5209 | } | |
e02119d5 | 5210 | |
33345d01 | 5211 | min_key.objectid = ino; |
e02119d5 CM |
5212 | min_key.type = BTRFS_INODE_ITEM_KEY; |
5213 | min_key.offset = 0; | |
5214 | ||
33345d01 | 5215 | max_key.objectid = ino; |
12fcfd22 | 5216 | |
12fcfd22 | 5217 | |
5dc562c5 | 5218 | /* today the code can only do partial logging of directories */ |
a59108a7 | 5219 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5269b67e | 5220 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5221 | &inode->runtime_flags) && |
781feef7 | 5222 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
5223 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5224 | else | |
5225 | max_key.type = (u8)-1; | |
5226 | max_key.offset = (u64)-1; | |
5227 | ||
2c2c452b | 5228 | /* |
5aa7d1a7 FM |
5229 | * Only run delayed items if we are a directory. We want to make sure |
5230 | * all directory indexes hit the fs/subvolume tree so we can find them | |
5231 | * and figure out which index ranges have to be logged. | |
5232 | * | |
8c8648dd FM |
5233 | * Otherwise commit the delayed inode only if the full sync flag is set, |
5234 | * as we want to make sure an up to date version is in the subvolume | |
5235 | * tree so copy_inode_items_to_log() / copy_items() can find it and copy | |
5236 | * it to the log tree. For a non full sync, we always log the inode item | |
5237 | * based on the in-memory struct btrfs_inode which is always up to date. | |
2c2c452b | 5238 | */ |
5aa7d1a7 | 5239 | if (S_ISDIR(inode->vfs_inode.i_mode)) |
a59108a7 | 5240 | ret = btrfs_commit_inode_delayed_items(trans, inode); |
8c8648dd | 5241 | else if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags)) |
a59108a7 | 5242 | ret = btrfs_commit_inode_delayed_inode(inode); |
2c2c452b FM |
5243 | |
5244 | if (ret) { | |
5245 | btrfs_free_path(path); | |
5246 | btrfs_free_path(dst_path); | |
5247 | return ret; | |
16cdcec7 MX |
5248 | } |
5249 | ||
a3baaf0d FM |
5250 | if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) { |
5251 | recursive_logging = true; | |
5252 | if (inode_only == LOG_OTHER_INODE) | |
5253 | inode_only = LOG_INODE_EXISTS; | |
5254 | else | |
5255 | inode_only = LOG_INODE_ALL; | |
a59108a7 | 5256 | mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING); |
781feef7 | 5257 | } else { |
a59108a7 | 5258 | mutex_lock(&inode->log_mutex); |
781feef7 | 5259 | } |
e02119d5 CM |
5260 | |
5261 | /* | |
5262 | * a brute force approach to making sure we get the most uptodate | |
5263 | * copies of everything. | |
5264 | */ | |
a59108a7 | 5265 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
e02119d5 CM |
5266 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; |
5267 | ||
4f764e51 FM |
5268 | if (inode_only == LOG_INODE_EXISTS) |
5269 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 5270 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 5271 | } else { |
1a4bcf47 FM |
5272 | if (inode_only == LOG_INODE_EXISTS) { |
5273 | /* | |
5274 | * Make sure the new inode item we write to the log has | |
5275 | * the same isize as the current one (if it exists). | |
5276 | * This is necessary to prevent data loss after log | |
5277 | * replay, and also to prevent doing a wrong expanding | |
5278 | * truncate - for e.g. create file, write 4K into offset | |
5279 | * 0, fsync, write 4K into offset 4096, add hard link, | |
5280 | * fsync some other file (to sync log), power fail - if | |
5281 | * we use the inode's current i_size, after log replay | |
5282 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
5283 | * (zeroes), as if an expanding truncate happened, | |
5284 | * instead of getting a file of 4Kb only. | |
5285 | */ | |
a59108a7 | 5286 | err = logged_inode_size(log, inode, path, &logged_isize); |
1a4bcf47 FM |
5287 | if (err) |
5288 | goto out_unlock; | |
5289 | } | |
a742994a | 5290 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5291 | &inode->runtime_flags)) { |
a742994a | 5292 | if (inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5293 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
5294 | ret = drop_objectid_items(trans, log, path, ino, |
5295 | max_key.type); | |
5296 | } else { | |
5297 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a59108a7 | 5298 | &inode->runtime_flags); |
a742994a | 5299 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5300 | &inode->runtime_flags); |
28ed1345 CM |
5301 | while(1) { |
5302 | ret = btrfs_truncate_inode_items(trans, | |
50743398 | 5303 | log, inode, 0, 0); |
28ed1345 CM |
5304 | if (ret != -EAGAIN) |
5305 | break; | |
5306 | } | |
a742994a | 5307 | } |
4f764e51 | 5308 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5309 | &inode->runtime_flags) || |
6cfab851 | 5310 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5311 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 5312 | fast_search = true; |
4f764e51 | 5313 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 5314 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 5315 | max_key.type); |
a95249b3 JB |
5316 | } else { |
5317 | if (inode_only == LOG_INODE_ALL) | |
5318 | fast_search = true; | |
a95249b3 | 5319 | goto log_extents; |
5dc562c5 | 5320 | } |
a95249b3 | 5321 | |
e02119d5 | 5322 | } |
4a500fd1 YZ |
5323 | if (ret) { |
5324 | err = ret; | |
5325 | goto out_unlock; | |
5326 | } | |
e02119d5 | 5327 | |
da447009 FM |
5328 | err = copy_inode_items_to_log(trans, inode, &min_key, &max_key, |
5329 | path, dst_path, logged_isize, | |
7af59743 FM |
5330 | recursive_logging, inode_only, ctx, |
5331 | &need_log_inode_item); | |
da447009 FM |
5332 | if (err) |
5333 | goto out_unlock; | |
5dc562c5 | 5334 | |
36283bf7 FM |
5335 | btrfs_release_path(path); |
5336 | btrfs_release_path(dst_path); | |
a59108a7 | 5337 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); |
36283bf7 FM |
5338 | if (err) |
5339 | goto out_unlock; | |
9a8fca62 | 5340 | xattrs_logged = true; |
a89ca6f2 FM |
5341 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
5342 | btrfs_release_path(path); | |
5343 | btrfs_release_path(dst_path); | |
7af59743 | 5344 | err = btrfs_log_holes(trans, root, inode, path); |
a89ca6f2 FM |
5345 | if (err) |
5346 | goto out_unlock; | |
5347 | } | |
a95249b3 | 5348 | log_extents: |
f3b15ccd JB |
5349 | btrfs_release_path(path); |
5350 | btrfs_release_path(dst_path); | |
e4545de5 | 5351 | if (need_log_inode_item) { |
a59108a7 | 5352 | err = log_inode_item(trans, log, dst_path, inode); |
9a8fca62 FM |
5353 | if (!err && !xattrs_logged) { |
5354 | err = btrfs_log_all_xattrs(trans, root, inode, path, | |
5355 | dst_path); | |
5356 | btrfs_release_path(path); | |
5357 | } | |
e4545de5 FM |
5358 | if (err) |
5359 | goto out_unlock; | |
5360 | } | |
5dc562c5 | 5361 | if (fast_search) { |
a59108a7 | 5362 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
48778179 | 5363 | ctx); |
5dc562c5 JB |
5364 | if (ret) { |
5365 | err = ret; | |
5366 | goto out_unlock; | |
5367 | } | |
d006a048 | 5368 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
5369 | struct extent_map *em, *n; |
5370 | ||
49dae1bc | 5371 | write_lock(&em_tree->lock); |
48778179 FM |
5372 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, list) |
5373 | list_del_init(&em->list); | |
49dae1bc | 5374 | write_unlock(&em_tree->lock); |
5dc562c5 JB |
5375 | } |
5376 | ||
a59108a7 NB |
5377 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) { |
5378 | ret = log_directory_changes(trans, root, inode, path, dst_path, | |
5379 | ctx); | |
4a500fd1 YZ |
5380 | if (ret) { |
5381 | err = ret; | |
5382 | goto out_unlock; | |
5383 | } | |
e02119d5 | 5384 | } |
49dae1bc | 5385 | |
d1d832a0 | 5386 | /* |
75b463d2 FM |
5387 | * If we are logging that an ancestor inode exists as part of logging a |
5388 | * new name from a link or rename operation, don't mark the inode as | |
5389 | * logged - otherwise if an explicit fsync is made against an ancestor, | |
5390 | * the fsync considers the inode in the log and doesn't sync the log, | |
5391 | * resulting in the ancestor missing after a power failure unless the | |
5392 | * log was synced as part of an fsync against any other unrelated inode. | |
5393 | * So keep it simple for this case and just don't flag the ancestors as | |
5394 | * logged. | |
d1d832a0 | 5395 | */ |
75b463d2 FM |
5396 | if (!ctx || |
5397 | !(S_ISDIR(inode->vfs_inode.i_mode) && ctx->logging_new_name && | |
5398 | &inode->vfs_inode != ctx->inode)) { | |
5399 | spin_lock(&inode->lock); | |
5400 | inode->logged_trans = trans->transid; | |
5401 | /* | |
5402 | * Don't update last_log_commit if we logged that an inode exists | |
5403 | * after it was loaded to memory (full_sync bit set). | |
5404 | * This is to prevent data loss when we do a write to the inode, | |
5405 | * then the inode gets evicted after all delalloc was flushed, | |
5406 | * then we log it exists (due to a rename for example) and then | |
5407 | * fsync it. This last fsync would do nothing (not logging the | |
5408 | * extents previously written). | |
5409 | */ | |
5410 | if (inode_only != LOG_INODE_EXISTS || | |
5411 | !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags)) | |
5412 | inode->last_log_commit = inode->last_sub_trans; | |
5413 | spin_unlock(&inode->lock); | |
5414 | } | |
4a500fd1 | 5415 | out_unlock: |
a59108a7 | 5416 | mutex_unlock(&inode->log_mutex); |
e02119d5 CM |
5417 | |
5418 | btrfs_free_path(path); | |
5419 | btrfs_free_path(dst_path); | |
4a500fd1 | 5420 | return err; |
e02119d5 CM |
5421 | } |
5422 | ||
2be63d5c FM |
5423 | /* |
5424 | * Check if we must fallback to a transaction commit when logging an inode. | |
5425 | * This must be called after logging the inode and is used only in the context | |
5426 | * when fsyncing an inode requires the need to log some other inode - in which | |
5427 | * case we can't lock the i_mutex of each other inode we need to log as that | |
5428 | * can lead to deadlocks with concurrent fsync against other inodes (as we can | |
5429 | * log inodes up or down in the hierarchy) or rename operations for example. So | |
5430 | * we take the log_mutex of the inode after we have logged it and then check for | |
5431 | * its last_unlink_trans value - this is safe because any task setting | |
5432 | * last_unlink_trans must take the log_mutex and it must do this before it does | |
5433 | * the actual unlink operation, so if we do this check before a concurrent task | |
5434 | * sets last_unlink_trans it means we've logged a consistent version/state of | |
5435 | * all the inode items, otherwise we are not sure and must do a transaction | |
01327610 | 5436 | * commit (the concurrent task might have only updated last_unlink_trans before |
2be63d5c FM |
5437 | * we logged the inode or it might have also done the unlink). |
5438 | */ | |
5439 | static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans, | |
ab1717b2 | 5440 | struct btrfs_inode *inode) |
2be63d5c | 5441 | { |
2be63d5c FM |
5442 | bool ret = false; |
5443 | ||
ab1717b2 | 5444 | mutex_lock(&inode->log_mutex); |
47d3db41 | 5445 | if (inode->last_unlink_trans >= trans->transid) { |
2be63d5c FM |
5446 | /* |
5447 | * Make sure any commits to the log are forced to be full | |
5448 | * commits. | |
5449 | */ | |
90787766 | 5450 | btrfs_set_log_full_commit(trans); |
2be63d5c FM |
5451 | ret = true; |
5452 | } | |
ab1717b2 | 5453 | mutex_unlock(&inode->log_mutex); |
2be63d5c FM |
5454 | |
5455 | return ret; | |
5456 | } | |
5457 | ||
12fcfd22 CM |
5458 | /* |
5459 | * follow the dentry parent pointers up the chain and see if any | |
5460 | * of the directories in it require a full commit before they can | |
5461 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
5462 | * a full commit is required. | |
5463 | */ | |
5464 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
aefa6115 | 5465 | struct btrfs_inode *inode, |
12fcfd22 | 5466 | struct dentry *parent, |
47d3db41 | 5467 | struct super_block *sb) |
e02119d5 | 5468 | { |
12fcfd22 | 5469 | int ret = 0; |
6a912213 | 5470 | struct dentry *old_parent = NULL; |
e02119d5 | 5471 | |
af4176b4 CM |
5472 | /* |
5473 | * for regular files, if its inode is already on disk, we don't | |
5474 | * have to worry about the parents at all. This is because | |
5475 | * we can use the last_unlink_trans field to record renames | |
5476 | * and other fun in this file. | |
5477 | */ | |
aefa6115 | 5478 | if (S_ISREG(inode->vfs_inode.i_mode) && |
47d3db41 FM |
5479 | inode->generation < trans->transid && |
5480 | inode->last_unlink_trans < trans->transid) | |
aefa6115 | 5481 | goto out; |
af4176b4 | 5482 | |
aefa6115 | 5483 | if (!S_ISDIR(inode->vfs_inode.i_mode)) { |
fc64005c | 5484 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 | 5485 | goto out; |
aefa6115 | 5486 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5487 | } |
5488 | ||
5489 | while (1) { | |
aefa6115 | 5490 | if (btrfs_must_commit_transaction(trans, inode)) { |
12fcfd22 CM |
5491 | ret = 1; |
5492 | break; | |
5493 | } | |
5494 | ||
fc64005c | 5495 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 CM |
5496 | break; |
5497 | ||
44f714da | 5498 | if (IS_ROOT(parent)) { |
aefa6115 NB |
5499 | inode = BTRFS_I(d_inode(parent)); |
5500 | if (btrfs_must_commit_transaction(trans, inode)) | |
44f714da | 5501 | ret = 1; |
12fcfd22 | 5502 | break; |
44f714da | 5503 | } |
12fcfd22 | 5504 | |
6a912213 JB |
5505 | parent = dget_parent(parent); |
5506 | dput(old_parent); | |
5507 | old_parent = parent; | |
aefa6115 | 5508 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5509 | |
5510 | } | |
6a912213 | 5511 | dput(old_parent); |
12fcfd22 | 5512 | out: |
e02119d5 CM |
5513 | return ret; |
5514 | } | |
5515 | ||
2f2ff0ee FM |
5516 | struct btrfs_dir_list { |
5517 | u64 ino; | |
5518 | struct list_head list; | |
5519 | }; | |
5520 | ||
5521 | /* | |
5522 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
5523 | * details about the why it is needed. | |
5524 | * This is a recursive operation - if an existing dentry corresponds to a | |
5525 | * directory, that directory's new entries are logged too (same behaviour as | |
5526 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5527 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
5528 | * complains about the following circular lock dependency / possible deadlock: | |
5529 | * | |
5530 | * CPU0 CPU1 | |
5531 | * ---- ---- | |
5532 | * lock(&type->i_mutex_dir_key#3/2); | |
5533 | * lock(sb_internal#2); | |
5534 | * lock(&type->i_mutex_dir_key#3/2); | |
5535 | * lock(&sb->s_type->i_mutex_key#14); | |
5536 | * | |
5537 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5538 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5539 | * Not locking i_mutex of the inodes is still safe because: | |
5540 | * | |
5541 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5542 | * that while logging the inode new references (names) are added or removed | |
5543 | * from the inode, leaving the logged inode item with a link count that does | |
5544 | * not match the number of logged inode reference items. This is fine because | |
5545 | * at log replay time we compute the real number of links and correct the | |
5546 | * link count in the inode item (see replay_one_buffer() and | |
5547 | * link_to_fixup_dir()); | |
5548 | * | |
5549 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5550 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
5551 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
5552 | * has a size that doesn't match the sum of the lengths of all the logged | |
5553 | * names. This does not result in a problem because if a dir_item key is | |
5554 | * logged but its matching dir_index key is not logged, at log replay time we | |
5555 | * don't use it to replay the respective name (see replay_one_name()). On the | |
5556 | * other hand if only the dir_index key ends up being logged, the respective | |
5557 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
5558 | * keys created (see replay_one_name()). | |
5559 | * The directory's inode item with a wrong i_size is not a problem as well, | |
5560 | * since we don't use it at log replay time to set the i_size in the inode | |
5561 | * item of the fs/subvol tree (see overwrite_item()). | |
5562 | */ | |
5563 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5564 | struct btrfs_root *root, | |
51cc0d32 | 5565 | struct btrfs_inode *start_inode, |
2f2ff0ee FM |
5566 | struct btrfs_log_ctx *ctx) |
5567 | { | |
0b246afa | 5568 | struct btrfs_fs_info *fs_info = root->fs_info; |
2f2ff0ee FM |
5569 | struct btrfs_root *log = root->log_root; |
5570 | struct btrfs_path *path; | |
5571 | LIST_HEAD(dir_list); | |
5572 | struct btrfs_dir_list *dir_elem; | |
5573 | int ret = 0; | |
5574 | ||
5575 | path = btrfs_alloc_path(); | |
5576 | if (!path) | |
5577 | return -ENOMEM; | |
5578 | ||
5579 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5580 | if (!dir_elem) { | |
5581 | btrfs_free_path(path); | |
5582 | return -ENOMEM; | |
5583 | } | |
51cc0d32 | 5584 | dir_elem->ino = btrfs_ino(start_inode); |
2f2ff0ee FM |
5585 | list_add_tail(&dir_elem->list, &dir_list); |
5586 | ||
5587 | while (!list_empty(&dir_list)) { | |
5588 | struct extent_buffer *leaf; | |
5589 | struct btrfs_key min_key; | |
5590 | int nritems; | |
5591 | int i; | |
5592 | ||
5593 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
5594 | list); | |
5595 | if (ret) | |
5596 | goto next_dir_inode; | |
5597 | ||
5598 | min_key.objectid = dir_elem->ino; | |
5599 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
5600 | min_key.offset = 0; | |
5601 | again: | |
5602 | btrfs_release_path(path); | |
5603 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
5604 | if (ret < 0) { | |
5605 | goto next_dir_inode; | |
5606 | } else if (ret > 0) { | |
5607 | ret = 0; | |
5608 | goto next_dir_inode; | |
5609 | } | |
5610 | ||
5611 | process_leaf: | |
5612 | leaf = path->nodes[0]; | |
5613 | nritems = btrfs_header_nritems(leaf); | |
5614 | for (i = path->slots[0]; i < nritems; i++) { | |
5615 | struct btrfs_dir_item *di; | |
5616 | struct btrfs_key di_key; | |
5617 | struct inode *di_inode; | |
5618 | struct btrfs_dir_list *new_dir_elem; | |
5619 | int log_mode = LOG_INODE_EXISTS; | |
5620 | int type; | |
5621 | ||
5622 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5623 | if (min_key.objectid != dir_elem->ino || | |
5624 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5625 | goto next_dir_inode; | |
5626 | ||
5627 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5628 | type = btrfs_dir_type(leaf, di); | |
5629 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5630 | type != BTRFS_FT_DIR) | |
5631 | continue; | |
5632 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5633 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5634 | continue; | |
5635 | ||
ec125cfb | 5636 | btrfs_release_path(path); |
0202e83f | 5637 | di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root); |
2f2ff0ee FM |
5638 | if (IS_ERR(di_inode)) { |
5639 | ret = PTR_ERR(di_inode); | |
5640 | goto next_dir_inode; | |
5641 | } | |
5642 | ||
0f8939b8 | 5643 | if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) { |
410f954c | 5644 | btrfs_add_delayed_iput(di_inode); |
ec125cfb | 5645 | break; |
2f2ff0ee FM |
5646 | } |
5647 | ||
5648 | ctx->log_new_dentries = false; | |
3f9749f6 | 5649 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) |
2f2ff0ee | 5650 | log_mode = LOG_INODE_ALL; |
a59108a7 | 5651 | ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode), |
48778179 | 5652 | log_mode, ctx); |
2be63d5c | 5653 | if (!ret && |
ab1717b2 | 5654 | btrfs_must_commit_transaction(trans, BTRFS_I(di_inode))) |
2be63d5c | 5655 | ret = 1; |
410f954c | 5656 | btrfs_add_delayed_iput(di_inode); |
2f2ff0ee FM |
5657 | if (ret) |
5658 | goto next_dir_inode; | |
5659 | if (ctx->log_new_dentries) { | |
5660 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5661 | GFP_NOFS); | |
5662 | if (!new_dir_elem) { | |
5663 | ret = -ENOMEM; | |
5664 | goto next_dir_inode; | |
5665 | } | |
5666 | new_dir_elem->ino = di_key.objectid; | |
5667 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5668 | } | |
5669 | break; | |
5670 | } | |
5671 | if (i == nritems) { | |
5672 | ret = btrfs_next_leaf(log, path); | |
5673 | if (ret < 0) { | |
5674 | goto next_dir_inode; | |
5675 | } else if (ret > 0) { | |
5676 | ret = 0; | |
5677 | goto next_dir_inode; | |
5678 | } | |
5679 | goto process_leaf; | |
5680 | } | |
5681 | if (min_key.offset < (u64)-1) { | |
5682 | min_key.offset++; | |
5683 | goto again; | |
5684 | } | |
5685 | next_dir_inode: | |
5686 | list_del(&dir_elem->list); | |
5687 | kfree(dir_elem); | |
5688 | } | |
5689 | ||
5690 | btrfs_free_path(path); | |
5691 | return ret; | |
5692 | } | |
5693 | ||
18aa0922 | 5694 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
d0a0b78d | 5695 | struct btrfs_inode *inode, |
18aa0922 FM |
5696 | struct btrfs_log_ctx *ctx) |
5697 | { | |
3ffbd68c | 5698 | struct btrfs_fs_info *fs_info = trans->fs_info; |
18aa0922 FM |
5699 | int ret; |
5700 | struct btrfs_path *path; | |
5701 | struct btrfs_key key; | |
d0a0b78d NB |
5702 | struct btrfs_root *root = inode->root; |
5703 | const u64 ino = btrfs_ino(inode); | |
18aa0922 FM |
5704 | |
5705 | path = btrfs_alloc_path(); | |
5706 | if (!path) | |
5707 | return -ENOMEM; | |
5708 | path->skip_locking = 1; | |
5709 | path->search_commit_root = 1; | |
5710 | ||
5711 | key.objectid = ino; | |
5712 | key.type = BTRFS_INODE_REF_KEY; | |
5713 | key.offset = 0; | |
5714 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5715 | if (ret < 0) | |
5716 | goto out; | |
5717 | ||
5718 | while (true) { | |
5719 | struct extent_buffer *leaf = path->nodes[0]; | |
5720 | int slot = path->slots[0]; | |
5721 | u32 cur_offset = 0; | |
5722 | u32 item_size; | |
5723 | unsigned long ptr; | |
5724 | ||
5725 | if (slot >= btrfs_header_nritems(leaf)) { | |
5726 | ret = btrfs_next_leaf(root, path); | |
5727 | if (ret < 0) | |
5728 | goto out; | |
5729 | else if (ret > 0) | |
5730 | break; | |
5731 | continue; | |
5732 | } | |
5733 | ||
5734 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5735 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5736 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5737 | break; | |
5738 | ||
5739 | item_size = btrfs_item_size_nr(leaf, slot); | |
5740 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5741 | while (cur_offset < item_size) { | |
5742 | struct btrfs_key inode_key; | |
5743 | struct inode *dir_inode; | |
5744 | ||
5745 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5746 | inode_key.offset = 0; | |
5747 | ||
5748 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5749 | struct btrfs_inode_extref *extref; | |
5750 | ||
5751 | extref = (struct btrfs_inode_extref *) | |
5752 | (ptr + cur_offset); | |
5753 | inode_key.objectid = btrfs_inode_extref_parent( | |
5754 | leaf, extref); | |
5755 | cur_offset += sizeof(*extref); | |
5756 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5757 | extref); | |
5758 | } else { | |
5759 | inode_key.objectid = key.offset; | |
5760 | cur_offset = item_size; | |
5761 | } | |
5762 | ||
0202e83f DS |
5763 | dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid, |
5764 | root); | |
0f375eed FM |
5765 | /* |
5766 | * If the parent inode was deleted, return an error to | |
5767 | * fallback to a transaction commit. This is to prevent | |
5768 | * getting an inode that was moved from one parent A to | |
5769 | * a parent B, got its former parent A deleted and then | |
5770 | * it got fsync'ed, from existing at both parents after | |
5771 | * a log replay (and the old parent still existing). | |
5772 | * Example: | |
5773 | * | |
5774 | * mkdir /mnt/A | |
5775 | * mkdir /mnt/B | |
5776 | * touch /mnt/B/bar | |
5777 | * sync | |
5778 | * mv /mnt/B/bar /mnt/A/bar | |
5779 | * mv -T /mnt/A /mnt/B | |
5780 | * fsync /mnt/B/bar | |
5781 | * <power fail> | |
5782 | * | |
5783 | * If we ignore the old parent B which got deleted, | |
5784 | * after a log replay we would have file bar linked | |
5785 | * at both parents and the old parent B would still | |
5786 | * exist. | |
5787 | */ | |
5788 | if (IS_ERR(dir_inode)) { | |
5789 | ret = PTR_ERR(dir_inode); | |
5790 | goto out; | |
5791 | } | |
18aa0922 | 5792 | |
657ed1aa FM |
5793 | if (ctx) |
5794 | ctx->log_new_dentries = false; | |
a59108a7 | 5795 | ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode), |
48778179 | 5796 | LOG_INODE_ALL, ctx); |
2be63d5c | 5797 | if (!ret && |
ab1717b2 | 5798 | btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode))) |
2be63d5c | 5799 | ret = 1; |
657ed1aa FM |
5800 | if (!ret && ctx && ctx->log_new_dentries) |
5801 | ret = log_new_dir_dentries(trans, root, | |
f85b7379 | 5802 | BTRFS_I(dir_inode), ctx); |
410f954c | 5803 | btrfs_add_delayed_iput(dir_inode); |
18aa0922 FM |
5804 | if (ret) |
5805 | goto out; | |
5806 | } | |
5807 | path->slots[0]++; | |
5808 | } | |
5809 | ret = 0; | |
5810 | out: | |
5811 | btrfs_free_path(path); | |
5812 | return ret; | |
5813 | } | |
5814 | ||
b8aa330d FM |
5815 | static int log_new_ancestors(struct btrfs_trans_handle *trans, |
5816 | struct btrfs_root *root, | |
5817 | struct btrfs_path *path, | |
5818 | struct btrfs_log_ctx *ctx) | |
5819 | { | |
5820 | struct btrfs_key found_key; | |
5821 | ||
5822 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
5823 | ||
5824 | while (true) { | |
5825 | struct btrfs_fs_info *fs_info = root->fs_info; | |
b8aa330d FM |
5826 | struct extent_buffer *leaf = path->nodes[0]; |
5827 | int slot = path->slots[0]; | |
5828 | struct btrfs_key search_key; | |
5829 | struct inode *inode; | |
0202e83f | 5830 | u64 ino; |
b8aa330d FM |
5831 | int ret = 0; |
5832 | ||
5833 | btrfs_release_path(path); | |
5834 | ||
0202e83f DS |
5835 | ino = found_key.offset; |
5836 | ||
b8aa330d FM |
5837 | search_key.objectid = found_key.offset; |
5838 | search_key.type = BTRFS_INODE_ITEM_KEY; | |
5839 | search_key.offset = 0; | |
0202e83f | 5840 | inode = btrfs_iget(fs_info->sb, ino, root); |
b8aa330d FM |
5841 | if (IS_ERR(inode)) |
5842 | return PTR_ERR(inode); | |
5843 | ||
4d6221d7 | 5844 | if (BTRFS_I(inode)->generation >= trans->transid) |
b8aa330d | 5845 | ret = btrfs_log_inode(trans, root, BTRFS_I(inode), |
48778179 | 5846 | LOG_INODE_EXISTS, ctx); |
410f954c | 5847 | btrfs_add_delayed_iput(inode); |
b8aa330d FM |
5848 | if (ret) |
5849 | return ret; | |
5850 | ||
5851 | if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID) | |
5852 | break; | |
5853 | ||
5854 | search_key.type = BTRFS_INODE_REF_KEY; | |
5855 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
5856 | if (ret < 0) | |
5857 | return ret; | |
5858 | ||
5859 | leaf = path->nodes[0]; | |
5860 | slot = path->slots[0]; | |
5861 | if (slot >= btrfs_header_nritems(leaf)) { | |
5862 | ret = btrfs_next_leaf(root, path); | |
5863 | if (ret < 0) | |
5864 | return ret; | |
5865 | else if (ret > 0) | |
5866 | return -ENOENT; | |
5867 | leaf = path->nodes[0]; | |
5868 | slot = path->slots[0]; | |
5869 | } | |
5870 | ||
5871 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5872 | if (found_key.objectid != search_key.objectid || | |
5873 | found_key.type != BTRFS_INODE_REF_KEY) | |
5874 | return -ENOENT; | |
5875 | } | |
5876 | return 0; | |
5877 | } | |
5878 | ||
5879 | static int log_new_ancestors_fast(struct btrfs_trans_handle *trans, | |
5880 | struct btrfs_inode *inode, | |
5881 | struct dentry *parent, | |
5882 | struct btrfs_log_ctx *ctx) | |
5883 | { | |
5884 | struct btrfs_root *root = inode->root; | |
b8aa330d FM |
5885 | struct dentry *old_parent = NULL; |
5886 | struct super_block *sb = inode->vfs_inode.i_sb; | |
5887 | int ret = 0; | |
5888 | ||
5889 | while (true) { | |
5890 | if (!parent || d_really_is_negative(parent) || | |
5891 | sb != parent->d_sb) | |
5892 | break; | |
5893 | ||
5894 | inode = BTRFS_I(d_inode(parent)); | |
5895 | if (root != inode->root) | |
5896 | break; | |
5897 | ||
4d6221d7 | 5898 | if (inode->generation >= trans->transid) { |
b8aa330d | 5899 | ret = btrfs_log_inode(trans, root, inode, |
48778179 | 5900 | LOG_INODE_EXISTS, ctx); |
b8aa330d FM |
5901 | if (ret) |
5902 | break; | |
5903 | } | |
5904 | if (IS_ROOT(parent)) | |
5905 | break; | |
5906 | ||
5907 | parent = dget_parent(parent); | |
5908 | dput(old_parent); | |
5909 | old_parent = parent; | |
5910 | } | |
5911 | dput(old_parent); | |
5912 | ||
5913 | return ret; | |
5914 | } | |
5915 | ||
5916 | static int log_all_new_ancestors(struct btrfs_trans_handle *trans, | |
5917 | struct btrfs_inode *inode, | |
5918 | struct dentry *parent, | |
5919 | struct btrfs_log_ctx *ctx) | |
5920 | { | |
5921 | struct btrfs_root *root = inode->root; | |
5922 | const u64 ino = btrfs_ino(inode); | |
5923 | struct btrfs_path *path; | |
5924 | struct btrfs_key search_key; | |
5925 | int ret; | |
5926 | ||
5927 | /* | |
5928 | * For a single hard link case, go through a fast path that does not | |
5929 | * need to iterate the fs/subvolume tree. | |
5930 | */ | |
5931 | if (inode->vfs_inode.i_nlink < 2) | |
5932 | return log_new_ancestors_fast(trans, inode, parent, ctx); | |
5933 | ||
5934 | path = btrfs_alloc_path(); | |
5935 | if (!path) | |
5936 | return -ENOMEM; | |
5937 | ||
5938 | search_key.objectid = ino; | |
5939 | search_key.type = BTRFS_INODE_REF_KEY; | |
5940 | search_key.offset = 0; | |
5941 | again: | |
5942 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
5943 | if (ret < 0) | |
5944 | goto out; | |
5945 | if (ret == 0) | |
5946 | path->slots[0]++; | |
5947 | ||
5948 | while (true) { | |
5949 | struct extent_buffer *leaf = path->nodes[0]; | |
5950 | int slot = path->slots[0]; | |
5951 | struct btrfs_key found_key; | |
5952 | ||
5953 | if (slot >= btrfs_header_nritems(leaf)) { | |
5954 | ret = btrfs_next_leaf(root, path); | |
5955 | if (ret < 0) | |
5956 | goto out; | |
5957 | else if (ret > 0) | |
5958 | break; | |
5959 | continue; | |
5960 | } | |
5961 | ||
5962 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5963 | if (found_key.objectid != ino || | |
5964 | found_key.type > BTRFS_INODE_EXTREF_KEY) | |
5965 | break; | |
5966 | ||
5967 | /* | |
5968 | * Don't deal with extended references because they are rare | |
5969 | * cases and too complex to deal with (we would need to keep | |
5970 | * track of which subitem we are processing for each item in | |
5971 | * this loop, etc). So just return some error to fallback to | |
5972 | * a transaction commit. | |
5973 | */ | |
5974 | if (found_key.type == BTRFS_INODE_EXTREF_KEY) { | |
5975 | ret = -EMLINK; | |
5976 | goto out; | |
5977 | } | |
5978 | ||
5979 | /* | |
5980 | * Logging ancestors needs to do more searches on the fs/subvol | |
5981 | * tree, so it releases the path as needed to avoid deadlocks. | |
5982 | * Keep track of the last inode ref key and resume from that key | |
5983 | * after logging all new ancestors for the current hard link. | |
5984 | */ | |
5985 | memcpy(&search_key, &found_key, sizeof(search_key)); | |
5986 | ||
5987 | ret = log_new_ancestors(trans, root, path, ctx); | |
5988 | if (ret) | |
5989 | goto out; | |
5990 | btrfs_release_path(path); | |
5991 | goto again; | |
5992 | } | |
5993 | ret = 0; | |
5994 | out: | |
5995 | btrfs_free_path(path); | |
5996 | return ret; | |
5997 | } | |
5998 | ||
e02119d5 CM |
5999 | /* |
6000 | * helper function around btrfs_log_inode to make sure newly created | |
6001 | * parent directories also end up in the log. A minimal inode and backref | |
6002 | * only logging is done of any parent directories that are older than | |
6003 | * the last committed transaction | |
6004 | */ | |
48a3b636 | 6005 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
19df27a9 | 6006 | struct btrfs_inode *inode, |
49dae1bc | 6007 | struct dentry *parent, |
41a1eada | 6008 | int inode_only, |
8b050d35 | 6009 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6010 | { |
f882274b | 6011 | struct btrfs_root *root = inode->root; |
0b246afa | 6012 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 6013 | struct super_block *sb; |
12fcfd22 | 6014 | int ret = 0; |
2f2ff0ee | 6015 | bool log_dentries = false; |
12fcfd22 | 6016 | |
19df27a9 | 6017 | sb = inode->vfs_inode.i_sb; |
12fcfd22 | 6018 | |
0b246afa | 6019 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
3a5e1404 SW |
6020 | ret = 1; |
6021 | goto end_no_trans; | |
6022 | } | |
6023 | ||
f882274b | 6024 | if (btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
6025 | ret = 1; |
6026 | goto end_no_trans; | |
6027 | } | |
6028 | ||
47d3db41 | 6029 | ret = check_parent_dirs_for_sync(trans, inode, parent, sb); |
12fcfd22 CM |
6030 | if (ret) |
6031 | goto end_no_trans; | |
e02119d5 | 6032 | |
f2d72f42 FM |
6033 | /* |
6034 | * Skip already logged inodes or inodes corresponding to tmpfiles | |
6035 | * (since logging them is pointless, a link count of 0 means they | |
6036 | * will never be accessible). | |
6037 | */ | |
6038 | if (btrfs_inode_in_log(inode, trans->transid) || | |
6039 | inode->vfs_inode.i_nlink == 0) { | |
257c62e1 CM |
6040 | ret = BTRFS_NO_LOG_SYNC; |
6041 | goto end_no_trans; | |
6042 | } | |
6043 | ||
8b050d35 | 6044 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 6045 | if (ret) |
e87ac136 | 6046 | goto end_no_trans; |
e02119d5 | 6047 | |
48778179 | 6048 | ret = btrfs_log_inode(trans, root, inode, inode_only, ctx); |
4a500fd1 YZ |
6049 | if (ret) |
6050 | goto end_trans; | |
12fcfd22 | 6051 | |
af4176b4 CM |
6052 | /* |
6053 | * for regular files, if its inode is already on disk, we don't | |
6054 | * have to worry about the parents at all. This is because | |
6055 | * we can use the last_unlink_trans field to record renames | |
6056 | * and other fun in this file. | |
6057 | */ | |
19df27a9 | 6058 | if (S_ISREG(inode->vfs_inode.i_mode) && |
47d3db41 FM |
6059 | inode->generation < trans->transid && |
6060 | inode->last_unlink_trans < trans->transid) { | |
4a500fd1 YZ |
6061 | ret = 0; |
6062 | goto end_trans; | |
6063 | } | |
af4176b4 | 6064 | |
19df27a9 | 6065 | if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries) |
2f2ff0ee FM |
6066 | log_dentries = true; |
6067 | ||
18aa0922 | 6068 | /* |
01327610 | 6069 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
6070 | * inodes are fully logged. This is to prevent leaving dangling |
6071 | * directory index entries in directories that were our parents but are | |
6072 | * not anymore. Not doing this results in old parent directory being | |
6073 | * impossible to delete after log replay (rmdir will always fail with | |
6074 | * error -ENOTEMPTY). | |
6075 | * | |
6076 | * Example 1: | |
6077 | * | |
6078 | * mkdir testdir | |
6079 | * touch testdir/foo | |
6080 | * ln testdir/foo testdir/bar | |
6081 | * sync | |
6082 | * unlink testdir/bar | |
6083 | * xfs_io -c fsync testdir/foo | |
6084 | * <power failure> | |
6085 | * mount fs, triggers log replay | |
6086 | * | |
6087 | * If we don't log the parent directory (testdir), after log replay the | |
6088 | * directory still has an entry pointing to the file inode using the bar | |
6089 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
6090 | * the file inode has a link count of 1. | |
6091 | * | |
6092 | * Example 2: | |
6093 | * | |
6094 | * mkdir testdir | |
6095 | * touch foo | |
6096 | * ln foo testdir/foo2 | |
6097 | * ln foo testdir/foo3 | |
6098 | * sync | |
6099 | * unlink testdir/foo3 | |
6100 | * xfs_io -c fsync foo | |
6101 | * <power failure> | |
6102 | * mount fs, triggers log replay | |
6103 | * | |
6104 | * Similar as the first example, after log replay the parent directory | |
6105 | * testdir still has an entry pointing to the inode file with name foo3 | |
6106 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
6107 | * and has a link count of 2. | |
6108 | */ | |
47d3db41 | 6109 | if (inode->last_unlink_trans >= trans->transid) { |
b8aa330d | 6110 | ret = btrfs_log_all_parents(trans, inode, ctx); |
18aa0922 FM |
6111 | if (ret) |
6112 | goto end_trans; | |
6113 | } | |
6114 | ||
b8aa330d FM |
6115 | ret = log_all_new_ancestors(trans, inode, parent, ctx); |
6116 | if (ret) | |
41bd6067 | 6117 | goto end_trans; |
76dda93c | 6118 | |
2f2ff0ee | 6119 | if (log_dentries) |
b8aa330d | 6120 | ret = log_new_dir_dentries(trans, root, inode, ctx); |
2f2ff0ee FM |
6121 | else |
6122 | ret = 0; | |
4a500fd1 YZ |
6123 | end_trans: |
6124 | if (ret < 0) { | |
90787766 | 6125 | btrfs_set_log_full_commit(trans); |
4a500fd1 YZ |
6126 | ret = 1; |
6127 | } | |
8b050d35 MX |
6128 | |
6129 | if (ret) | |
6130 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
6131 | btrfs_end_log_trans(root); |
6132 | end_no_trans: | |
6133 | return ret; | |
e02119d5 CM |
6134 | } |
6135 | ||
6136 | /* | |
6137 | * it is not safe to log dentry if the chunk root has added new | |
6138 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
6139 | * If this returns 1, you must commit the transaction to safely get your | |
6140 | * data on disk. | |
6141 | */ | |
6142 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
e5b84f7a | 6143 | struct dentry *dentry, |
8b050d35 | 6144 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6145 | { |
6a912213 JB |
6146 | struct dentry *parent = dget_parent(dentry); |
6147 | int ret; | |
6148 | ||
f882274b | 6149 | ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent, |
48778179 | 6150 | LOG_INODE_ALL, ctx); |
6a912213 JB |
6151 | dput(parent); |
6152 | ||
6153 | return ret; | |
e02119d5 CM |
6154 | } |
6155 | ||
6156 | /* | |
6157 | * should be called during mount to recover any replay any log trees | |
6158 | * from the FS | |
6159 | */ | |
6160 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
6161 | { | |
6162 | int ret; | |
6163 | struct btrfs_path *path; | |
6164 | struct btrfs_trans_handle *trans; | |
6165 | struct btrfs_key key; | |
6166 | struct btrfs_key found_key; | |
e02119d5 CM |
6167 | struct btrfs_root *log; |
6168 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
6169 | struct walk_control wc = { | |
6170 | .process_func = process_one_buffer, | |
430a6626 | 6171 | .stage = LOG_WALK_PIN_ONLY, |
e02119d5 CM |
6172 | }; |
6173 | ||
e02119d5 | 6174 | path = btrfs_alloc_path(); |
db5b493a TI |
6175 | if (!path) |
6176 | return -ENOMEM; | |
6177 | ||
afcdd129 | 6178 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 6179 | |
4a500fd1 | 6180 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
6181 | if (IS_ERR(trans)) { |
6182 | ret = PTR_ERR(trans); | |
6183 | goto error; | |
6184 | } | |
e02119d5 CM |
6185 | |
6186 | wc.trans = trans; | |
6187 | wc.pin = 1; | |
6188 | ||
db5b493a | 6189 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 6190 | if (ret) { |
5d163e0e JM |
6191 | btrfs_handle_fs_error(fs_info, ret, |
6192 | "Failed to pin buffers while recovering log root tree."); | |
79787eaa JM |
6193 | goto error; |
6194 | } | |
e02119d5 CM |
6195 | |
6196 | again: | |
6197 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
6198 | key.offset = (u64)-1; | |
962a298f | 6199 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 6200 | |
d397712b | 6201 | while (1) { |
e02119d5 | 6202 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
6203 | |
6204 | if (ret < 0) { | |
34d97007 | 6205 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6206 | "Couldn't find tree log root."); |
6207 | goto error; | |
6208 | } | |
e02119d5 CM |
6209 | if (ret > 0) { |
6210 | if (path->slots[0] == 0) | |
6211 | break; | |
6212 | path->slots[0]--; | |
6213 | } | |
6214 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
6215 | path->slots[0]); | |
b3b4aa74 | 6216 | btrfs_release_path(path); |
e02119d5 CM |
6217 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
6218 | break; | |
6219 | ||
62a2c73e | 6220 | log = btrfs_read_tree_root(log_root_tree, &found_key); |
79787eaa JM |
6221 | if (IS_ERR(log)) { |
6222 | ret = PTR_ERR(log); | |
34d97007 | 6223 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6224 | "Couldn't read tree log root."); |
6225 | goto error; | |
6226 | } | |
e02119d5 | 6227 | |
56e9357a DS |
6228 | wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset, |
6229 | true); | |
79787eaa JM |
6230 | if (IS_ERR(wc.replay_dest)) { |
6231 | ret = PTR_ERR(wc.replay_dest); | |
9bc574de JB |
6232 | |
6233 | /* | |
6234 | * We didn't find the subvol, likely because it was | |
6235 | * deleted. This is ok, simply skip this log and go to | |
6236 | * the next one. | |
6237 | * | |
6238 | * We need to exclude the root because we can't have | |
6239 | * other log replays overwriting this log as we'll read | |
6240 | * it back in a few more times. This will keep our | |
6241 | * block from being modified, and we'll just bail for | |
6242 | * each subsequent pass. | |
6243 | */ | |
6244 | if (ret == -ENOENT) | |
9fce5704 | 6245 | ret = btrfs_pin_extent_for_log_replay(trans, |
9bc574de JB |
6246 | log->node->start, |
6247 | log->node->len); | |
00246528 | 6248 | btrfs_put_root(log); |
9bc574de JB |
6249 | |
6250 | if (!ret) | |
6251 | goto next; | |
5d163e0e JM |
6252 | btrfs_handle_fs_error(fs_info, ret, |
6253 | "Couldn't read target root for tree log recovery."); | |
79787eaa JM |
6254 | goto error; |
6255 | } | |
e02119d5 | 6256 | |
07d400a6 | 6257 | wc.replay_dest->log_root = log; |
5d4f98a2 | 6258 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 6259 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 6260 | |
b50c6e25 | 6261 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
6262 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
6263 | path); | |
e02119d5 CM |
6264 | } |
6265 | ||
900c9981 LB |
6266 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
6267 | struct btrfs_root *root = wc.replay_dest; | |
6268 | ||
6269 | btrfs_release_path(path); | |
6270 | ||
6271 | /* | |
6272 | * We have just replayed everything, and the highest | |
6273 | * objectid of fs roots probably has changed in case | |
6274 | * some inode_item's got replayed. | |
6275 | * | |
6276 | * root->objectid_mutex is not acquired as log replay | |
6277 | * could only happen during mount. | |
6278 | */ | |
453e4873 | 6279 | ret = btrfs_init_root_free_objectid(root); |
900c9981 LB |
6280 | } |
6281 | ||
07d400a6 | 6282 | wc.replay_dest->log_root = NULL; |
00246528 | 6283 | btrfs_put_root(wc.replay_dest); |
00246528 | 6284 | btrfs_put_root(log); |
e02119d5 | 6285 | |
b50c6e25 JB |
6286 | if (ret) |
6287 | goto error; | |
9bc574de | 6288 | next: |
e02119d5 CM |
6289 | if (found_key.offset == 0) |
6290 | break; | |
9bc574de | 6291 | key.offset = found_key.offset - 1; |
e02119d5 | 6292 | } |
b3b4aa74 | 6293 | btrfs_release_path(path); |
e02119d5 CM |
6294 | |
6295 | /* step one is to pin it all, step two is to replay just inodes */ | |
6296 | if (wc.pin) { | |
6297 | wc.pin = 0; | |
6298 | wc.process_func = replay_one_buffer; | |
6299 | wc.stage = LOG_WALK_REPLAY_INODES; | |
6300 | goto again; | |
6301 | } | |
6302 | /* step three is to replay everything */ | |
6303 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
6304 | wc.stage++; | |
6305 | goto again; | |
6306 | } | |
6307 | ||
6308 | btrfs_free_path(path); | |
6309 | ||
abefa55a | 6310 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 6311 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
6312 | if (ret) |
6313 | return ret; | |
6314 | ||
e02119d5 | 6315 | log_root_tree->log_root = NULL; |
afcdd129 | 6316 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
00246528 | 6317 | btrfs_put_root(log_root_tree); |
79787eaa | 6318 | |
abefa55a | 6319 | return 0; |
79787eaa | 6320 | error: |
b50c6e25 | 6321 | if (wc.trans) |
3a45bb20 | 6322 | btrfs_end_transaction(wc.trans); |
79787eaa JM |
6323 | btrfs_free_path(path); |
6324 | return ret; | |
e02119d5 | 6325 | } |
12fcfd22 CM |
6326 | |
6327 | /* | |
6328 | * there are some corner cases where we want to force a full | |
6329 | * commit instead of allowing a directory to be logged. | |
6330 | * | |
6331 | * They revolve around files there were unlinked from the directory, and | |
6332 | * this function updates the parent directory so that a full commit is | |
6333 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
6334 | * |
6335 | * Must be called before the unlink operations (updates to the subvolume tree, | |
6336 | * inodes, etc) are done. | |
12fcfd22 CM |
6337 | */ |
6338 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 6339 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
6340 | int for_rename) |
6341 | { | |
af4176b4 CM |
6342 | /* |
6343 | * when we're logging a file, if it hasn't been renamed | |
6344 | * or unlinked, and its inode is fully committed on disk, | |
6345 | * we don't have to worry about walking up the directory chain | |
6346 | * to log its parents. | |
6347 | * | |
6348 | * So, we use the last_unlink_trans field to put this transid | |
6349 | * into the file. When the file is logged we check it and | |
6350 | * don't log the parents if the file is fully on disk. | |
6351 | */ | |
4176bdbf NB |
6352 | mutex_lock(&inode->log_mutex); |
6353 | inode->last_unlink_trans = trans->transid; | |
6354 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 6355 | |
12fcfd22 CM |
6356 | /* |
6357 | * if this directory was already logged any new | |
6358 | * names for this file/dir will get recorded | |
6359 | */ | |
4176bdbf | 6360 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
6361 | return; |
6362 | ||
6363 | /* | |
6364 | * if the inode we're about to unlink was logged, | |
6365 | * the log will be properly updated for any new names | |
6366 | */ | |
4176bdbf | 6367 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
6368 | return; |
6369 | ||
6370 | /* | |
6371 | * when renaming files across directories, if the directory | |
6372 | * there we're unlinking from gets fsync'd later on, there's | |
6373 | * no way to find the destination directory later and fsync it | |
6374 | * properly. So, we have to be conservative and force commits | |
6375 | * so the new name gets discovered. | |
6376 | */ | |
6377 | if (for_rename) | |
6378 | goto record; | |
6379 | ||
6380 | /* we can safely do the unlink without any special recording */ | |
6381 | return; | |
6382 | ||
6383 | record: | |
4176bdbf NB |
6384 | mutex_lock(&dir->log_mutex); |
6385 | dir->last_unlink_trans = trans->transid; | |
6386 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
6387 | } |
6388 | ||
6389 | /* | |
6390 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
6391 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
6392 | * that after replaying the log tree of the parent directory's root we will not | |
6393 | * see the snapshot anymore and at log replay time we will not see any log tree | |
6394 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
6395 | * it after replaying the log tree of the parent directory (which would replay | |
6396 | * the snapshot delete operation). | |
2be63d5c FM |
6397 | * |
6398 | * Must be called before the actual snapshot destroy operation (updates to the | |
6399 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
6400 | */ |
6401 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 6402 | struct btrfs_inode *dir) |
1ec9a1ae | 6403 | { |
43663557 NB |
6404 | mutex_lock(&dir->log_mutex); |
6405 | dir->last_unlink_trans = trans->transid; | |
6406 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
6407 | } |
6408 | ||
6409 | /* | |
6410 | * Call this after adding a new name for a file and it will properly | |
6411 | * update the log to reflect the new name. | |
12fcfd22 | 6412 | */ |
75b463d2 | 6413 | void btrfs_log_new_name(struct btrfs_trans_handle *trans, |
9ca5fbfb | 6414 | struct btrfs_inode *inode, struct btrfs_inode *old_dir, |
75b463d2 | 6415 | struct dentry *parent) |
12fcfd22 | 6416 | { |
75b463d2 | 6417 | struct btrfs_log_ctx ctx; |
12fcfd22 | 6418 | |
af4176b4 CM |
6419 | /* |
6420 | * this will force the logging code to walk the dentry chain | |
6421 | * up for the file | |
6422 | */ | |
9a6509c4 | 6423 | if (!S_ISDIR(inode->vfs_inode.i_mode)) |
9ca5fbfb | 6424 | inode->last_unlink_trans = trans->transid; |
af4176b4 | 6425 | |
12fcfd22 CM |
6426 | /* |
6427 | * if this inode hasn't been logged and directory we're renaming it | |
6428 | * from hasn't been logged, we don't need to log it | |
6429 | */ | |
de53d892 FM |
6430 | if (inode->logged_trans < trans->transid && |
6431 | (!old_dir || old_dir->logged_trans < trans->transid)) | |
75b463d2 | 6432 | return; |
12fcfd22 | 6433 | |
75b463d2 FM |
6434 | btrfs_init_log_ctx(&ctx, &inode->vfs_inode); |
6435 | ctx.logging_new_name = true; | |
6436 | /* | |
6437 | * We don't care about the return value. If we fail to log the new name | |
6438 | * then we know the next attempt to sync the log will fallback to a full | |
6439 | * transaction commit (due to a call to btrfs_set_log_full_commit()), so | |
6440 | * we don't need to worry about getting a log committed that has an | |
6441 | * inconsistent state after a rename operation. | |
6442 | */ | |
48778179 | 6443 | btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx); |
12fcfd22 CM |
6444 | } |
6445 |