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" |
900c9981 | 20 | #include "inode-map.h" |
6787bb9f NB |
21 | #include "block-group.h" |
22 | #include "space-info.h" | |
e02119d5 CM |
23 | |
24 | /* magic values for the inode_only field in btrfs_log_inode: | |
25 | * | |
26 | * LOG_INODE_ALL means to log everything | |
27 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
28 | * during log replay | |
29 | */ | |
e13976cf DS |
30 | enum { |
31 | LOG_INODE_ALL, | |
32 | LOG_INODE_EXISTS, | |
33 | LOG_OTHER_INODE, | |
34 | LOG_OTHER_INODE_ALL, | |
35 | }; | |
e02119d5 | 36 | |
12fcfd22 CM |
37 | /* |
38 | * directory trouble cases | |
39 | * | |
40 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
41 | * log, we must force a full commit before doing an fsync of the directory | |
42 | * where the unlink was done. | |
43 | * ---> record transid of last unlink/rename per directory | |
44 | * | |
45 | * mkdir foo/some_dir | |
46 | * normal commit | |
47 | * rename foo/some_dir foo2/some_dir | |
48 | * mkdir foo/some_dir | |
49 | * fsync foo/some_dir/some_file | |
50 | * | |
51 | * The fsync above will unlink the original some_dir without recording | |
52 | * it in its new location (foo2). After a crash, some_dir will be gone | |
53 | * unless the fsync of some_file forces a full commit | |
54 | * | |
55 | * 2) we must log any new names for any file or dir that is in the fsync | |
56 | * log. ---> check inode while renaming/linking. | |
57 | * | |
58 | * 2a) we must log any new names for any file or dir during rename | |
59 | * when the directory they are being removed from was logged. | |
60 | * ---> check inode and old parent dir during rename | |
61 | * | |
62 | * 2a is actually the more important variant. With the extra logging | |
63 | * a crash might unlink the old name without recreating the new one | |
64 | * | |
65 | * 3) after a crash, we must go through any directories with a link count | |
66 | * of zero and redo the rm -rf | |
67 | * | |
68 | * mkdir f1/foo | |
69 | * normal commit | |
70 | * rm -rf f1/foo | |
71 | * fsync(f1) | |
72 | * | |
73 | * The directory f1 was fully removed from the FS, but fsync was never | |
74 | * called on f1, only its parent dir. After a crash the rm -rf must | |
75 | * be replayed. This must be able to recurse down the entire | |
76 | * directory tree. The inode link count fixup code takes care of the | |
77 | * ugly details. | |
78 | */ | |
79 | ||
e02119d5 CM |
80 | /* |
81 | * stages for the tree walking. The first | |
82 | * stage (0) is to only pin down the blocks we find | |
83 | * the second stage (1) is to make sure that all the inodes | |
84 | * we find in the log are created in the subvolume. | |
85 | * | |
86 | * The last stage is to deal with directories and links and extents | |
87 | * and all the other fun semantics | |
88 | */ | |
e13976cf DS |
89 | enum { |
90 | LOG_WALK_PIN_ONLY, | |
91 | LOG_WALK_REPLAY_INODES, | |
92 | LOG_WALK_REPLAY_DIR_INDEX, | |
93 | LOG_WALK_REPLAY_ALL, | |
94 | }; | |
e02119d5 | 95 | |
12fcfd22 | 96 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 97 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc | 98 | int inode_only, |
7af59743 FM |
99 | const loff_t start, |
100 | const loff_t end, | |
8407f553 | 101 | struct btrfs_log_ctx *ctx); |
ec051c0f YZ |
102 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
103 | struct btrfs_root *root, | |
104 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
105 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
106 | struct btrfs_root *root, | |
107 | struct btrfs_root *log, | |
108 | struct btrfs_path *path, | |
109 | u64 dirid, int del_all); | |
e02119d5 CM |
110 | |
111 | /* | |
112 | * tree logging is a special write ahead log used to make sure that | |
113 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
114 | * | |
115 | * Full tree commits are expensive because they require commonly | |
116 | * modified blocks to be recowed, creating many dirty pages in the | |
117 | * extent tree an 4x-6x higher write load than ext3. | |
118 | * | |
119 | * Instead of doing a tree commit on every fsync, we use the | |
120 | * key ranges and transaction ids to find items for a given file or directory | |
121 | * that have changed in this transaction. Those items are copied into | |
122 | * a special tree (one per subvolume root), that tree is written to disk | |
123 | * and then the fsync is considered complete. | |
124 | * | |
125 | * After a crash, items are copied out of the log-tree back into the | |
126 | * subvolume tree. Any file data extents found are recorded in the extent | |
127 | * allocation tree, and the log-tree freed. | |
128 | * | |
129 | * The log tree is read three times, once to pin down all the extents it is | |
130 | * using in ram and once, once to create all the inodes logged in the tree | |
131 | * and once to do all the other items. | |
132 | */ | |
133 | ||
e02119d5 CM |
134 | /* |
135 | * start a sub transaction and setup the log tree | |
136 | * this increments the log tree writer count to make the people | |
137 | * syncing the tree wait for us to finish | |
138 | */ | |
139 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
140 | struct btrfs_root *root, |
141 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 142 | { |
0b246afa | 143 | struct btrfs_fs_info *fs_info = root->fs_info; |
34eb2a52 | 144 | int ret = 0; |
7237f183 YZ |
145 | |
146 | mutex_lock(&root->log_mutex); | |
34eb2a52 | 147 | |
7237f183 | 148 | if (root->log_root) { |
4884b8e8 | 149 | if (btrfs_need_log_full_commit(trans)) { |
50471a38 MX |
150 | ret = -EAGAIN; |
151 | goto out; | |
152 | } | |
34eb2a52 | 153 | |
ff782e0a | 154 | if (!root->log_start_pid) { |
27cdeb70 | 155 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 156 | root->log_start_pid = current->pid; |
ff782e0a | 157 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 158 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 159 | } |
34eb2a52 | 160 | } else { |
0b246afa JM |
161 | mutex_lock(&fs_info->tree_log_mutex); |
162 | if (!fs_info->log_root_tree) | |
163 | ret = btrfs_init_log_root_tree(trans, fs_info); | |
164 | mutex_unlock(&fs_info->tree_log_mutex); | |
34eb2a52 Z |
165 | if (ret) |
166 | goto out; | |
ff782e0a | 167 | |
e02119d5 | 168 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 169 | if (ret) |
e87ac136 | 170 | goto out; |
34eb2a52 Z |
171 | |
172 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); | |
173 | root->log_start_pid = current->pid; | |
e02119d5 | 174 | } |
34eb2a52 | 175 | |
2ecb7923 | 176 | atomic_inc(&root->log_batch); |
7237f183 | 177 | atomic_inc(&root->log_writers); |
8b050d35 | 178 | if (ctx) { |
34eb2a52 | 179 | int index = root->log_transid % 2; |
8b050d35 | 180 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 181 | ctx->log_transid = root->log_transid; |
8b050d35 | 182 | } |
34eb2a52 | 183 | |
e87ac136 | 184 | out: |
7237f183 | 185 | mutex_unlock(&root->log_mutex); |
e87ac136 | 186 | return ret; |
e02119d5 CM |
187 | } |
188 | ||
189 | /* | |
190 | * returns 0 if there was a log transaction running and we were able | |
191 | * to join, or returns -ENOENT if there were not transactions | |
192 | * in progress | |
193 | */ | |
194 | static int join_running_log_trans(struct btrfs_root *root) | |
195 | { | |
196 | int ret = -ENOENT; | |
197 | ||
7237f183 | 198 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
199 | if (root->log_root) { |
200 | ret = 0; | |
7237f183 | 201 | atomic_inc(&root->log_writers); |
e02119d5 | 202 | } |
7237f183 | 203 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
204 | return ret; |
205 | } | |
206 | ||
12fcfd22 CM |
207 | /* |
208 | * This either makes the current running log transaction wait | |
209 | * until you call btrfs_end_log_trans() or it makes any future | |
210 | * log transactions wait until you call btrfs_end_log_trans() | |
211 | */ | |
45128b08 | 212 | void btrfs_pin_log_trans(struct btrfs_root *root) |
12fcfd22 | 213 | { |
12fcfd22 CM |
214 | mutex_lock(&root->log_mutex); |
215 | atomic_inc(&root->log_writers); | |
216 | mutex_unlock(&root->log_mutex); | |
12fcfd22 CM |
217 | } |
218 | ||
e02119d5 CM |
219 | /* |
220 | * indicate we're done making changes to the log tree | |
221 | * and wake up anyone waiting to do a sync | |
222 | */ | |
143bede5 | 223 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 224 | { |
7237f183 | 225 | if (atomic_dec_and_test(&root->log_writers)) { |
093258e6 DS |
226 | /* atomic_dec_and_test implies a barrier */ |
227 | cond_wake_up_nomb(&root->log_writer_wait); | |
7237f183 | 228 | } |
e02119d5 CM |
229 | } |
230 | ||
247462a5 DS |
231 | static int btrfs_write_tree_block(struct extent_buffer *buf) |
232 | { | |
233 | return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, | |
234 | buf->start + buf->len - 1); | |
235 | } | |
236 | ||
237 | static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf) | |
238 | { | |
239 | filemap_fdatawait_range(buf->pages[0]->mapping, | |
240 | buf->start, buf->start + buf->len - 1); | |
241 | } | |
e02119d5 CM |
242 | |
243 | /* | |
244 | * the walk control struct is used to pass state down the chain when | |
245 | * processing the log tree. The stage field tells us which part | |
246 | * of the log tree processing we are currently doing. The others | |
247 | * are state fields used for that specific part | |
248 | */ | |
249 | struct walk_control { | |
250 | /* should we free the extent on disk when done? This is used | |
251 | * at transaction commit time while freeing a log tree | |
252 | */ | |
253 | int free; | |
254 | ||
255 | /* should we write out the extent buffer? This is used | |
256 | * while flushing the log tree to disk during a sync | |
257 | */ | |
258 | int write; | |
259 | ||
260 | /* should we wait for the extent buffer io to finish? Also used | |
261 | * while flushing the log tree to disk for a sync | |
262 | */ | |
263 | int wait; | |
264 | ||
265 | /* pin only walk, we record which extents on disk belong to the | |
266 | * log trees | |
267 | */ | |
268 | int pin; | |
269 | ||
270 | /* what stage of the replay code we're currently in */ | |
271 | int stage; | |
272 | ||
f2d72f42 FM |
273 | /* |
274 | * Ignore any items from the inode currently being processed. Needs | |
275 | * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in | |
276 | * the LOG_WALK_REPLAY_INODES stage. | |
277 | */ | |
278 | bool ignore_cur_inode; | |
279 | ||
e02119d5 CM |
280 | /* the root we are currently replaying */ |
281 | struct btrfs_root *replay_dest; | |
282 | ||
283 | /* the trans handle for the current replay */ | |
284 | struct btrfs_trans_handle *trans; | |
285 | ||
286 | /* the function that gets used to process blocks we find in the | |
287 | * tree. Note the extent_buffer might not be up to date when it is | |
288 | * passed in, and it must be checked or read if you need the data | |
289 | * inside it | |
290 | */ | |
291 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 292 | struct walk_control *wc, u64 gen, int level); |
e02119d5 CM |
293 | }; |
294 | ||
295 | /* | |
296 | * process_func used to pin down extents, write them or wait on them | |
297 | */ | |
298 | static int process_one_buffer(struct btrfs_root *log, | |
299 | struct extent_buffer *eb, | |
581c1760 | 300 | struct walk_control *wc, u64 gen, int level) |
e02119d5 | 301 | { |
0b246afa | 302 | struct btrfs_fs_info *fs_info = log->fs_info; |
b50c6e25 JB |
303 | int ret = 0; |
304 | ||
8c2a1a30 JB |
305 | /* |
306 | * If this fs is mixed then we need to be able to process the leaves to | |
307 | * pin down any logged extents, so we have to read the block. | |
308 | */ | |
0b246afa | 309 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
581c1760 | 310 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
8c2a1a30 JB |
311 | if (ret) |
312 | return ret; | |
313 | } | |
314 | ||
04018de5 | 315 | if (wc->pin) |
9fce5704 | 316 | ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start, |
2ff7e61e | 317 | eb->len); |
e02119d5 | 318 | |
b50c6e25 | 319 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 | 320 | if (wc->pin && btrfs_header_level(eb) == 0) |
bcdc428c | 321 | ret = btrfs_exclude_logged_extents(eb); |
e02119d5 CM |
322 | if (wc->write) |
323 | btrfs_write_tree_block(eb); | |
324 | if (wc->wait) | |
325 | btrfs_wait_tree_block_writeback(eb); | |
326 | } | |
b50c6e25 | 327 | return ret; |
e02119d5 CM |
328 | } |
329 | ||
330 | /* | |
331 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
332 | * to the src data we are copying out. | |
333 | * | |
334 | * root is the tree we are copying into, and path is a scratch | |
335 | * path for use in this function (it should be released on entry and | |
336 | * will be released on exit). | |
337 | * | |
338 | * If the key is already in the destination tree the existing item is | |
339 | * overwritten. If the existing item isn't big enough, it is extended. | |
340 | * If it is too large, it is truncated. | |
341 | * | |
342 | * If the key isn't in the destination yet, a new item is inserted. | |
343 | */ | |
344 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
345 | struct btrfs_root *root, | |
346 | struct btrfs_path *path, | |
347 | struct extent_buffer *eb, int slot, | |
348 | struct btrfs_key *key) | |
349 | { | |
350 | int ret; | |
351 | u32 item_size; | |
352 | u64 saved_i_size = 0; | |
353 | int save_old_i_size = 0; | |
354 | unsigned long src_ptr; | |
355 | unsigned long dst_ptr; | |
356 | int overwrite_root = 0; | |
4bc4bee4 | 357 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
358 | |
359 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
360 | overwrite_root = 1; | |
361 | ||
362 | item_size = btrfs_item_size_nr(eb, slot); | |
363 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
364 | ||
365 | /* look for the key in the destination tree */ | |
366 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
367 | if (ret < 0) |
368 | return ret; | |
369 | ||
e02119d5 CM |
370 | if (ret == 0) { |
371 | char *src_copy; | |
372 | char *dst_copy; | |
373 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
374 | path->slots[0]); | |
375 | if (dst_size != item_size) | |
376 | goto insert; | |
377 | ||
378 | if (item_size == 0) { | |
b3b4aa74 | 379 | btrfs_release_path(path); |
e02119d5 CM |
380 | return 0; |
381 | } | |
382 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
383 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 384 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 385 | btrfs_release_path(path); |
2a29edc6 | 386 | kfree(dst_copy); |
387 | kfree(src_copy); | |
388 | return -ENOMEM; | |
389 | } | |
e02119d5 CM |
390 | |
391 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
392 | ||
393 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
394 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
395 | item_size); | |
396 | ret = memcmp(dst_copy, src_copy, item_size); | |
397 | ||
398 | kfree(dst_copy); | |
399 | kfree(src_copy); | |
400 | /* | |
401 | * they have the same contents, just return, this saves | |
402 | * us from cowing blocks in the destination tree and doing | |
403 | * extra writes that may not have been done by a previous | |
404 | * sync | |
405 | */ | |
406 | if (ret == 0) { | |
b3b4aa74 | 407 | btrfs_release_path(path); |
e02119d5 CM |
408 | return 0; |
409 | } | |
410 | ||
4bc4bee4 JB |
411 | /* |
412 | * We need to load the old nbytes into the inode so when we | |
413 | * replay the extents we've logged we get the right nbytes. | |
414 | */ | |
415 | if (inode_item) { | |
416 | struct btrfs_inode_item *item; | |
417 | u64 nbytes; | |
d555438b | 418 | u32 mode; |
4bc4bee4 JB |
419 | |
420 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
421 | struct btrfs_inode_item); | |
422 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
423 | item = btrfs_item_ptr(eb, slot, | |
424 | struct btrfs_inode_item); | |
425 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
426 | |
427 | /* | |
428 | * If this is a directory we need to reset the i_size to | |
429 | * 0 so that we can set it up properly when replaying | |
430 | * the rest of the items in this log. | |
431 | */ | |
432 | mode = btrfs_inode_mode(eb, item); | |
433 | if (S_ISDIR(mode)) | |
434 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
435 | } |
436 | } else if (inode_item) { | |
437 | struct btrfs_inode_item *item; | |
d555438b | 438 | u32 mode; |
4bc4bee4 JB |
439 | |
440 | /* | |
441 | * New inode, set nbytes to 0 so that the nbytes comes out | |
442 | * properly when we replay the extents. | |
443 | */ | |
444 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
445 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
446 | |
447 | /* | |
448 | * If this is a directory we need to reset the i_size to 0 so | |
449 | * that we can set it up properly when replaying the rest of | |
450 | * the items in this log. | |
451 | */ | |
452 | mode = btrfs_inode_mode(eb, item); | |
453 | if (S_ISDIR(mode)) | |
454 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
455 | } |
456 | insert: | |
b3b4aa74 | 457 | btrfs_release_path(path); |
e02119d5 | 458 | /* try to insert the key into the destination tree */ |
df8d116f | 459 | path->skip_release_on_error = 1; |
e02119d5 CM |
460 | ret = btrfs_insert_empty_item(trans, root, path, |
461 | key, item_size); | |
df8d116f | 462 | path->skip_release_on_error = 0; |
e02119d5 CM |
463 | |
464 | /* make sure any existing item is the correct size */ | |
df8d116f | 465 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 CM |
466 | u32 found_size; |
467 | found_size = btrfs_item_size_nr(path->nodes[0], | |
468 | path->slots[0]); | |
143bede5 | 469 | if (found_size > item_size) |
78ac4f9e | 470 | btrfs_truncate_item(path, item_size, 1); |
143bede5 | 471 | else if (found_size < item_size) |
c71dd880 | 472 | btrfs_extend_item(path, item_size - found_size); |
e02119d5 | 473 | } else if (ret) { |
4a500fd1 | 474 | return ret; |
e02119d5 CM |
475 | } |
476 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
477 | path->slots[0]); | |
478 | ||
479 | /* don't overwrite an existing inode if the generation number | |
480 | * was logged as zero. This is done when the tree logging code | |
481 | * is just logging an inode to make sure it exists after recovery. | |
482 | * | |
483 | * Also, don't overwrite i_size on directories during replay. | |
484 | * log replay inserts and removes directory items based on the | |
485 | * state of the tree found in the subvolume, and i_size is modified | |
486 | * as it goes | |
487 | */ | |
488 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
489 | struct btrfs_inode_item *src_item; | |
490 | struct btrfs_inode_item *dst_item; | |
491 | ||
492 | src_item = (struct btrfs_inode_item *)src_ptr; | |
493 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
494 | ||
1a4bcf47 FM |
495 | if (btrfs_inode_generation(eb, src_item) == 0) { |
496 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 497 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 498 | |
2f2ff0ee FM |
499 | /* |
500 | * For regular files an ino_size == 0 is used only when | |
501 | * logging that an inode exists, as part of a directory | |
502 | * fsync, and the inode wasn't fsynced before. In this | |
503 | * case don't set the size of the inode in the fs/subvol | |
504 | * tree, otherwise we would be throwing valid data away. | |
505 | */ | |
1a4bcf47 | 506 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee | 507 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
60d48e2e DS |
508 | ino_size != 0) |
509 | btrfs_set_inode_size(dst_eb, dst_item, ino_size); | |
e02119d5 | 510 | goto no_copy; |
1a4bcf47 | 511 | } |
e02119d5 CM |
512 | |
513 | if (overwrite_root && | |
514 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
515 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
516 | save_old_i_size = 1; | |
517 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
518 | dst_item); | |
519 | } | |
520 | } | |
521 | ||
522 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
523 | src_ptr, item_size); | |
524 | ||
525 | if (save_old_i_size) { | |
526 | struct btrfs_inode_item *dst_item; | |
527 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
528 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
529 | } | |
530 | ||
531 | /* make sure the generation is filled in */ | |
532 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
533 | struct btrfs_inode_item *dst_item; | |
534 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
535 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
536 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
537 | trans->transid); | |
538 | } | |
539 | } | |
540 | no_copy: | |
541 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 542 | btrfs_release_path(path); |
e02119d5 CM |
543 | return 0; |
544 | } | |
545 | ||
546 | /* | |
547 | * simple helper to read an inode off the disk from a given root | |
548 | * This can only be called for subvolume roots and not for the log | |
549 | */ | |
550 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
551 | u64 objectid) | |
552 | { | |
553 | struct inode *inode; | |
e02119d5 | 554 | |
0202e83f | 555 | inode = btrfs_iget(root->fs_info->sb, objectid, root); |
2e19f1f9 | 556 | if (IS_ERR(inode)) |
5d4f98a2 | 557 | inode = NULL; |
e02119d5 CM |
558 | return inode; |
559 | } | |
560 | ||
561 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
562 | * subvolume 'root'. path is released on entry and should be released | |
563 | * on exit. | |
564 | * | |
565 | * extents in the log tree have not been allocated out of the extent | |
566 | * tree yet. So, this completes the allocation, taking a reference | |
567 | * as required if the extent already exists or creating a new extent | |
568 | * if it isn't in the extent allocation tree yet. | |
569 | * | |
570 | * The extent is inserted into the file, dropping any existing extents | |
571 | * from the file that overlap the new one. | |
572 | */ | |
573 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
574 | struct btrfs_root *root, | |
575 | struct btrfs_path *path, | |
576 | struct extent_buffer *eb, int slot, | |
577 | struct btrfs_key *key) | |
578 | { | |
0b246afa | 579 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 580 | int found_type; |
e02119d5 | 581 | u64 extent_end; |
e02119d5 | 582 | u64 start = key->offset; |
4bc4bee4 | 583 | u64 nbytes = 0; |
e02119d5 CM |
584 | struct btrfs_file_extent_item *item; |
585 | struct inode *inode = NULL; | |
586 | unsigned long size; | |
587 | int ret = 0; | |
588 | ||
589 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
590 | found_type = btrfs_file_extent_type(eb, item); | |
591 | ||
d899e052 | 592 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
593 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
594 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
595 | extent_end = start + nbytes; | |
596 | ||
597 | /* | |
598 | * We don't add to the inodes nbytes if we are prealloc or a | |
599 | * hole. | |
600 | */ | |
601 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
602 | nbytes = 0; | |
603 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 604 | size = btrfs_file_extent_ram_bytes(eb, item); |
4bc4bee4 | 605 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
da17066c | 606 | extent_end = ALIGN(start + size, |
0b246afa | 607 | fs_info->sectorsize); |
e02119d5 CM |
608 | } else { |
609 | ret = 0; | |
610 | goto out; | |
611 | } | |
612 | ||
613 | inode = read_one_inode(root, key->objectid); | |
614 | if (!inode) { | |
615 | ret = -EIO; | |
616 | goto out; | |
617 | } | |
618 | ||
619 | /* | |
620 | * first check to see if we already have this extent in the | |
621 | * file. This must be done before the btrfs_drop_extents run | |
622 | * so we don't try to drop this extent. | |
623 | */ | |
f85b7379 DS |
624 | ret = btrfs_lookup_file_extent(trans, root, path, |
625 | btrfs_ino(BTRFS_I(inode)), start, 0); | |
e02119d5 | 626 | |
d899e052 YZ |
627 | if (ret == 0 && |
628 | (found_type == BTRFS_FILE_EXTENT_REG || | |
629 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
630 | struct btrfs_file_extent_item cmp1; |
631 | struct btrfs_file_extent_item cmp2; | |
632 | struct btrfs_file_extent_item *existing; | |
633 | struct extent_buffer *leaf; | |
634 | ||
635 | leaf = path->nodes[0]; | |
636 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
637 | struct btrfs_file_extent_item); | |
638 | ||
639 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
640 | sizeof(cmp1)); | |
641 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
642 | sizeof(cmp2)); | |
643 | ||
644 | /* | |
645 | * we already have a pointer to this exact extent, | |
646 | * we don't have to do anything | |
647 | */ | |
648 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 649 | btrfs_release_path(path); |
e02119d5 CM |
650 | goto out; |
651 | } | |
652 | } | |
b3b4aa74 | 653 | btrfs_release_path(path); |
e02119d5 CM |
654 | |
655 | /* drop any overlapping extents */ | |
2671485d | 656 | ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1); |
3650860b JB |
657 | if (ret) |
658 | goto out; | |
e02119d5 | 659 | |
07d400a6 YZ |
660 | if (found_type == BTRFS_FILE_EXTENT_REG || |
661 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 662 | u64 offset; |
07d400a6 YZ |
663 | unsigned long dest_offset; |
664 | struct btrfs_key ins; | |
665 | ||
3168021c FM |
666 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0 && |
667 | btrfs_fs_incompat(fs_info, NO_HOLES)) | |
668 | goto update_inode; | |
669 | ||
07d400a6 YZ |
670 | ret = btrfs_insert_empty_item(trans, root, path, key, |
671 | sizeof(*item)); | |
3650860b JB |
672 | if (ret) |
673 | goto out; | |
07d400a6 YZ |
674 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
675 | path->slots[0]); | |
676 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
677 | (unsigned long)item, sizeof(*item)); | |
678 | ||
679 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
680 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
681 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 682 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 683 | |
df2c95f3 QW |
684 | /* |
685 | * Manually record dirty extent, as here we did a shallow | |
686 | * file extent item copy and skip normal backref update, | |
687 | * but modifying extent tree all by ourselves. | |
688 | * So need to manually record dirty extent for qgroup, | |
689 | * as the owner of the file extent changed from log tree | |
690 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
691 | */ | |
a95f3aaf | 692 | ret = btrfs_qgroup_trace_extent(trans, |
df2c95f3 QW |
693 | btrfs_file_extent_disk_bytenr(eb, item), |
694 | btrfs_file_extent_disk_num_bytes(eb, item), | |
695 | GFP_NOFS); | |
696 | if (ret < 0) | |
697 | goto out; | |
698 | ||
07d400a6 | 699 | if (ins.objectid > 0) { |
82fa113f | 700 | struct btrfs_ref ref = { 0 }; |
07d400a6 YZ |
701 | u64 csum_start; |
702 | u64 csum_end; | |
703 | LIST_HEAD(ordered_sums); | |
82fa113f | 704 | |
07d400a6 YZ |
705 | /* |
706 | * is this extent already allocated in the extent | |
707 | * allocation tree? If so, just add a reference | |
708 | */ | |
2ff7e61e | 709 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 YZ |
710 | ins.offset); |
711 | if (ret == 0) { | |
82fa113f QW |
712 | btrfs_init_generic_ref(&ref, |
713 | BTRFS_ADD_DELAYED_REF, | |
714 | ins.objectid, ins.offset, 0); | |
715 | btrfs_init_data_ref(&ref, | |
716 | root->root_key.objectid, | |
b06c4bf5 | 717 | key->objectid, offset); |
82fa113f | 718 | ret = btrfs_inc_extent_ref(trans, &ref); |
b50c6e25 JB |
719 | if (ret) |
720 | goto out; | |
07d400a6 YZ |
721 | } else { |
722 | /* | |
723 | * insert the extent pointer in the extent | |
724 | * allocation tree | |
725 | */ | |
5d4f98a2 | 726 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e | 727 | root->root_key.objectid, |
5d4f98a2 | 728 | key->objectid, offset, &ins); |
b50c6e25 JB |
729 | if (ret) |
730 | goto out; | |
07d400a6 | 731 | } |
b3b4aa74 | 732 | btrfs_release_path(path); |
07d400a6 YZ |
733 | |
734 | if (btrfs_file_extent_compression(eb, item)) { | |
735 | csum_start = ins.objectid; | |
736 | csum_end = csum_start + ins.offset; | |
737 | } else { | |
738 | csum_start = ins.objectid + | |
739 | btrfs_file_extent_offset(eb, item); | |
740 | csum_end = csum_start + | |
741 | btrfs_file_extent_num_bytes(eb, item); | |
742 | } | |
743 | ||
744 | ret = btrfs_lookup_csums_range(root->log_root, | |
745 | csum_start, csum_end - 1, | |
a2de733c | 746 | &ordered_sums, 0); |
3650860b JB |
747 | if (ret) |
748 | goto out; | |
b84b8390 FM |
749 | /* |
750 | * Now delete all existing cums in the csum root that | |
751 | * cover our range. We do this because we can have an | |
752 | * extent that is completely referenced by one file | |
753 | * extent item and partially referenced by another | |
754 | * file extent item (like after using the clone or | |
755 | * extent_same ioctls). In this case if we end up doing | |
756 | * the replay of the one that partially references the | |
757 | * extent first, and we do not do the csum deletion | |
758 | * below, we can get 2 csum items in the csum tree that | |
759 | * overlap each other. For example, imagine our log has | |
760 | * the two following file extent items: | |
761 | * | |
762 | * key (257 EXTENT_DATA 409600) | |
763 | * extent data disk byte 12845056 nr 102400 | |
764 | * extent data offset 20480 nr 20480 ram 102400 | |
765 | * | |
766 | * key (257 EXTENT_DATA 819200) | |
767 | * extent data disk byte 12845056 nr 102400 | |
768 | * extent data offset 0 nr 102400 ram 102400 | |
769 | * | |
770 | * Where the second one fully references the 100K extent | |
771 | * that starts at disk byte 12845056, and the log tree | |
772 | * has a single csum item that covers the entire range | |
773 | * of the extent: | |
774 | * | |
775 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
776 | * | |
777 | * After the first file extent item is replayed, the | |
778 | * csum tree gets the following csum item: | |
779 | * | |
780 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
781 | * | |
782 | * Which covers the 20K sub-range starting at offset 20K | |
783 | * of our extent. Now when we replay the second file | |
784 | * extent item, if we do not delete existing csum items | |
785 | * that cover any of its blocks, we end up getting two | |
786 | * csum items in our csum tree that overlap each other: | |
787 | * | |
788 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
789 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
790 | * | |
791 | * Which is a problem, because after this anyone trying | |
792 | * to lookup up for the checksum of any block of our | |
793 | * extent starting at an offset of 40K or higher, will | |
794 | * end up looking at the second csum item only, which | |
795 | * does not contain the checksum for any block starting | |
796 | * at offset 40K or higher of our extent. | |
797 | */ | |
07d400a6 YZ |
798 | while (!list_empty(&ordered_sums)) { |
799 | struct btrfs_ordered_sum *sums; | |
800 | sums = list_entry(ordered_sums.next, | |
801 | struct btrfs_ordered_sum, | |
802 | list); | |
b84b8390 | 803 | if (!ret) |
40e046ac FM |
804 | ret = btrfs_del_csums(trans, |
805 | fs_info->csum_root, | |
5b4aacef JM |
806 | sums->bytenr, |
807 | sums->len); | |
3650860b JB |
808 | if (!ret) |
809 | ret = btrfs_csum_file_blocks(trans, | |
0b246afa | 810 | fs_info->csum_root, sums); |
07d400a6 YZ |
811 | list_del(&sums->list); |
812 | kfree(sums); | |
813 | } | |
3650860b JB |
814 | if (ret) |
815 | goto out; | |
07d400a6 | 816 | } else { |
b3b4aa74 | 817 | btrfs_release_path(path); |
07d400a6 YZ |
818 | } |
819 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
820 | /* inline extents are easy, we just overwrite them */ | |
821 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
822 | if (ret) |
823 | goto out; | |
07d400a6 | 824 | } |
e02119d5 | 825 | |
9ddc959e JB |
826 | ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, |
827 | extent_end - start); | |
828 | if (ret) | |
829 | goto out; | |
830 | ||
4bc4bee4 | 831 | inode_add_bytes(inode, nbytes); |
3168021c | 832 | update_inode: |
b9959295 | 833 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
834 | out: |
835 | if (inode) | |
836 | iput(inode); | |
837 | return ret; | |
838 | } | |
839 | ||
840 | /* | |
841 | * when cleaning up conflicts between the directory names in the | |
842 | * subvolume, directory names in the log and directory names in the | |
843 | * inode back references, we may have to unlink inodes from directories. | |
844 | * | |
845 | * This is a helper function to do the unlink of a specific directory | |
846 | * item | |
847 | */ | |
848 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
849 | struct btrfs_root *root, | |
850 | struct btrfs_path *path, | |
207e7d92 | 851 | struct btrfs_inode *dir, |
e02119d5 CM |
852 | struct btrfs_dir_item *di) |
853 | { | |
854 | struct inode *inode; | |
855 | char *name; | |
856 | int name_len; | |
857 | struct extent_buffer *leaf; | |
858 | struct btrfs_key location; | |
859 | int ret; | |
860 | ||
861 | leaf = path->nodes[0]; | |
862 | ||
863 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
864 | name_len = btrfs_dir_name_len(leaf, di); | |
865 | name = kmalloc(name_len, GFP_NOFS); | |
2a29edc6 | 866 | if (!name) |
867 | return -ENOMEM; | |
868 | ||
e02119d5 | 869 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
b3b4aa74 | 870 | btrfs_release_path(path); |
e02119d5 CM |
871 | |
872 | inode = read_one_inode(root, location.objectid); | |
c00e9493 | 873 | if (!inode) { |
3650860b JB |
874 | ret = -EIO; |
875 | goto out; | |
c00e9493 | 876 | } |
e02119d5 | 877 | |
ec051c0f | 878 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
3650860b JB |
879 | if (ret) |
880 | goto out; | |
12fcfd22 | 881 | |
207e7d92 NB |
882 | ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name, |
883 | name_len); | |
3650860b JB |
884 | if (ret) |
885 | goto out; | |
ada9af21 | 886 | else |
e5c304e6 | 887 | ret = btrfs_run_delayed_items(trans); |
3650860b | 888 | out: |
e02119d5 | 889 | kfree(name); |
e02119d5 CM |
890 | iput(inode); |
891 | return ret; | |
892 | } | |
893 | ||
894 | /* | |
895 | * helper function to see if a given name and sequence number found | |
896 | * in an inode back reference are already in a directory and correctly | |
897 | * point to this inode | |
898 | */ | |
899 | static noinline int inode_in_dir(struct btrfs_root *root, | |
900 | struct btrfs_path *path, | |
901 | u64 dirid, u64 objectid, u64 index, | |
902 | const char *name, int name_len) | |
903 | { | |
904 | struct btrfs_dir_item *di; | |
905 | struct btrfs_key location; | |
906 | int match = 0; | |
907 | ||
908 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
909 | index, name, name_len, 0); | |
910 | if (di && !IS_ERR(di)) { | |
911 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
912 | if (location.objectid != objectid) | |
913 | goto out; | |
914 | } else | |
915 | goto out; | |
b3b4aa74 | 916 | btrfs_release_path(path); |
e02119d5 CM |
917 | |
918 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
919 | if (di && !IS_ERR(di)) { | |
920 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
921 | if (location.objectid != objectid) | |
922 | goto out; | |
923 | } else | |
924 | goto out; | |
925 | match = 1; | |
926 | out: | |
b3b4aa74 | 927 | btrfs_release_path(path); |
e02119d5 CM |
928 | return match; |
929 | } | |
930 | ||
931 | /* | |
932 | * helper function to check a log tree for a named back reference in | |
933 | * an inode. This is used to decide if a back reference that is | |
934 | * found in the subvolume conflicts with what we find in the log. | |
935 | * | |
936 | * inode backreferences may have multiple refs in a single item, | |
937 | * during replay we process one reference at a time, and we don't | |
938 | * want to delete valid links to a file from the subvolume if that | |
939 | * link is also in the log. | |
940 | */ | |
941 | static noinline int backref_in_log(struct btrfs_root *log, | |
942 | struct btrfs_key *key, | |
f186373f | 943 | u64 ref_objectid, |
df8d116f | 944 | const char *name, int namelen) |
e02119d5 CM |
945 | { |
946 | struct btrfs_path *path; | |
e02119d5 | 947 | int ret; |
e02119d5 CM |
948 | |
949 | path = btrfs_alloc_path(); | |
2a29edc6 | 950 | if (!path) |
951 | return -ENOMEM; | |
952 | ||
e02119d5 | 953 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
d3316c82 NB |
954 | if (ret < 0) { |
955 | goto out; | |
956 | } else if (ret == 1) { | |
89cbf5f6 | 957 | ret = 0; |
f186373f MF |
958 | goto out; |
959 | } | |
960 | ||
89cbf5f6 NB |
961 | if (key->type == BTRFS_INODE_EXTREF_KEY) |
962 | ret = !!btrfs_find_name_in_ext_backref(path->nodes[0], | |
963 | path->slots[0], | |
964 | ref_objectid, | |
965 | name, namelen); | |
966 | else | |
967 | ret = !!btrfs_find_name_in_backref(path->nodes[0], | |
968 | path->slots[0], | |
969 | name, namelen); | |
e02119d5 CM |
970 | out: |
971 | btrfs_free_path(path); | |
89cbf5f6 | 972 | return ret; |
e02119d5 CM |
973 | } |
974 | ||
5a1d7843 | 975 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 976 | struct btrfs_root *root, |
e02119d5 | 977 | struct btrfs_path *path, |
5a1d7843 | 978 | struct btrfs_root *log_root, |
94c91a1f NB |
979 | struct btrfs_inode *dir, |
980 | struct btrfs_inode *inode, | |
f186373f MF |
981 | u64 inode_objectid, u64 parent_objectid, |
982 | u64 ref_index, char *name, int namelen, | |
983 | int *search_done) | |
e02119d5 | 984 | { |
34f3e4f2 | 985 | int ret; |
f186373f MF |
986 | char *victim_name; |
987 | int victim_name_len; | |
988 | struct extent_buffer *leaf; | |
5a1d7843 | 989 | struct btrfs_dir_item *di; |
f186373f MF |
990 | struct btrfs_key search_key; |
991 | struct btrfs_inode_extref *extref; | |
c622ae60 | 992 | |
f186373f MF |
993 | again: |
994 | /* Search old style refs */ | |
995 | search_key.objectid = inode_objectid; | |
996 | search_key.type = BTRFS_INODE_REF_KEY; | |
997 | search_key.offset = parent_objectid; | |
998 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 999 | if (ret == 0) { |
e02119d5 CM |
1000 | struct btrfs_inode_ref *victim_ref; |
1001 | unsigned long ptr; | |
1002 | unsigned long ptr_end; | |
f186373f MF |
1003 | |
1004 | leaf = path->nodes[0]; | |
e02119d5 CM |
1005 | |
1006 | /* are we trying to overwrite a back ref for the root directory | |
1007 | * if so, just jump out, we're done | |
1008 | */ | |
f186373f | 1009 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1010 | return 1; |
e02119d5 CM |
1011 | |
1012 | /* check all the names in this back reference to see | |
1013 | * if they are in the log. if so, we allow them to stay | |
1014 | * otherwise they must be unlinked as a conflict | |
1015 | */ | |
1016 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1017 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 1018 | while (ptr < ptr_end) { |
e02119d5 CM |
1019 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1020 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1021 | victim_ref); | |
1022 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1023 | if (!victim_name) |
1024 | return -ENOMEM; | |
e02119d5 CM |
1025 | |
1026 | read_extent_buffer(leaf, victim_name, | |
1027 | (unsigned long)(victim_ref + 1), | |
1028 | victim_name_len); | |
1029 | ||
d3316c82 NB |
1030 | ret = backref_in_log(log_root, &search_key, |
1031 | parent_objectid, victim_name, | |
1032 | victim_name_len); | |
1033 | if (ret < 0) { | |
1034 | kfree(victim_name); | |
1035 | return ret; | |
1036 | } else if (!ret) { | |
94c91a1f | 1037 | inc_nlink(&inode->vfs_inode); |
b3b4aa74 | 1038 | btrfs_release_path(path); |
12fcfd22 | 1039 | |
94c91a1f | 1040 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
4ec5934e | 1041 | victim_name, victim_name_len); |
f186373f | 1042 | kfree(victim_name); |
3650860b JB |
1043 | if (ret) |
1044 | return ret; | |
e5c304e6 | 1045 | ret = btrfs_run_delayed_items(trans); |
ada9af21 FDBM |
1046 | if (ret) |
1047 | return ret; | |
f186373f MF |
1048 | *search_done = 1; |
1049 | goto again; | |
e02119d5 CM |
1050 | } |
1051 | kfree(victim_name); | |
f186373f | 1052 | |
e02119d5 CM |
1053 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1054 | } | |
e02119d5 | 1055 | |
c622ae60 | 1056 | /* |
1057 | * NOTE: we have searched root tree and checked the | |
bb7ab3b9 | 1058 | * corresponding ref, it does not need to check again. |
c622ae60 | 1059 | */ |
5a1d7843 | 1060 | *search_done = 1; |
e02119d5 | 1061 | } |
b3b4aa74 | 1062 | btrfs_release_path(path); |
e02119d5 | 1063 | |
f186373f MF |
1064 | /* Same search but for extended refs */ |
1065 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1066 | inode_objectid, parent_objectid, 0, | |
1067 | 0); | |
1068 | if (!IS_ERR_OR_NULL(extref)) { | |
1069 | u32 item_size; | |
1070 | u32 cur_offset = 0; | |
1071 | unsigned long base; | |
1072 | struct inode *victim_parent; | |
1073 | ||
1074 | leaf = path->nodes[0]; | |
1075 | ||
1076 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1077 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1078 | ||
1079 | while (cur_offset < item_size) { | |
dd9ef135 | 1080 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1081 | |
1082 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1083 | ||
1084 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1085 | goto next; | |
1086 | ||
1087 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1088 | if (!victim_name) |
1089 | return -ENOMEM; | |
f186373f MF |
1090 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1091 | victim_name_len); | |
1092 | ||
1093 | search_key.objectid = inode_objectid; | |
1094 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1095 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1096 | victim_name, | |
1097 | victim_name_len); | |
d3316c82 NB |
1098 | ret = backref_in_log(log_root, &search_key, |
1099 | parent_objectid, victim_name, | |
1100 | victim_name_len); | |
1101 | if (ret < 0) { | |
1102 | return ret; | |
1103 | } else if (!ret) { | |
f186373f MF |
1104 | ret = -ENOENT; |
1105 | victim_parent = read_one_inode(root, | |
94c91a1f | 1106 | parent_objectid); |
f186373f | 1107 | if (victim_parent) { |
94c91a1f | 1108 | inc_nlink(&inode->vfs_inode); |
f186373f MF |
1109 | btrfs_release_path(path); |
1110 | ||
1111 | ret = btrfs_unlink_inode(trans, root, | |
4ec5934e | 1112 | BTRFS_I(victim_parent), |
94c91a1f | 1113 | inode, |
4ec5934e NB |
1114 | victim_name, |
1115 | victim_name_len); | |
ada9af21 FDBM |
1116 | if (!ret) |
1117 | ret = btrfs_run_delayed_items( | |
e5c304e6 | 1118 | trans); |
f186373f | 1119 | } |
f186373f MF |
1120 | iput(victim_parent); |
1121 | kfree(victim_name); | |
3650860b JB |
1122 | if (ret) |
1123 | return ret; | |
f186373f MF |
1124 | *search_done = 1; |
1125 | goto again; | |
1126 | } | |
1127 | kfree(victim_name); | |
f186373f MF |
1128 | next: |
1129 | cur_offset += victim_name_len + sizeof(*extref); | |
1130 | } | |
1131 | *search_done = 1; | |
1132 | } | |
1133 | btrfs_release_path(path); | |
1134 | ||
34f3e4f2 | 1135 | /* look for a conflicting sequence number */ |
94c91a1f | 1136 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
f186373f | 1137 | ref_index, name, namelen, 0); |
34f3e4f2 | 1138 | if (di && !IS_ERR(di)) { |
94c91a1f | 1139 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1140 | if (ret) |
1141 | return ret; | |
34f3e4f2 | 1142 | } |
1143 | btrfs_release_path(path); | |
1144 | ||
52042d8e | 1145 | /* look for a conflicting name */ |
94c91a1f | 1146 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), |
34f3e4f2 | 1147 | name, namelen, 0); |
1148 | if (di && !IS_ERR(di)) { | |
94c91a1f | 1149 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1150 | if (ret) |
1151 | return ret; | |
34f3e4f2 | 1152 | } |
1153 | btrfs_release_path(path); | |
1154 | ||
5a1d7843 JS |
1155 | return 0; |
1156 | } | |
e02119d5 | 1157 | |
bae15d95 QW |
1158 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1159 | u32 *namelen, char **name, u64 *index, | |
1160 | u64 *parent_objectid) | |
f186373f MF |
1161 | { |
1162 | struct btrfs_inode_extref *extref; | |
1163 | ||
1164 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1165 | ||
1166 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1167 | *name = kmalloc(*namelen, GFP_NOFS); | |
1168 | if (*name == NULL) | |
1169 | return -ENOMEM; | |
1170 | ||
1171 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1172 | *namelen); | |
1173 | ||
1f250e92 FM |
1174 | if (index) |
1175 | *index = btrfs_inode_extref_index(eb, extref); | |
f186373f MF |
1176 | if (parent_objectid) |
1177 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1178 | ||
1179 | return 0; | |
1180 | } | |
1181 | ||
bae15d95 QW |
1182 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1183 | u32 *namelen, char **name, u64 *index) | |
f186373f MF |
1184 | { |
1185 | struct btrfs_inode_ref *ref; | |
1186 | ||
1187 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1188 | ||
1189 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1190 | *name = kmalloc(*namelen, GFP_NOFS); | |
1191 | if (*name == NULL) | |
1192 | return -ENOMEM; | |
1193 | ||
1194 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1195 | ||
1f250e92 FM |
1196 | if (index) |
1197 | *index = btrfs_inode_ref_index(eb, ref); | |
f186373f MF |
1198 | |
1199 | return 0; | |
1200 | } | |
1201 | ||
1f250e92 FM |
1202 | /* |
1203 | * Take an inode reference item from the log tree and iterate all names from the | |
1204 | * inode reference item in the subvolume tree with the same key (if it exists). | |
1205 | * For any name that is not in the inode reference item from the log tree, do a | |
1206 | * proper unlink of that name (that is, remove its entry from the inode | |
1207 | * reference item and both dir index keys). | |
1208 | */ | |
1209 | static int unlink_old_inode_refs(struct btrfs_trans_handle *trans, | |
1210 | struct btrfs_root *root, | |
1211 | struct btrfs_path *path, | |
1212 | struct btrfs_inode *inode, | |
1213 | struct extent_buffer *log_eb, | |
1214 | int log_slot, | |
1215 | struct btrfs_key *key) | |
1216 | { | |
1217 | int ret; | |
1218 | unsigned long ref_ptr; | |
1219 | unsigned long ref_end; | |
1220 | struct extent_buffer *eb; | |
1221 | ||
1222 | again: | |
1223 | btrfs_release_path(path); | |
1224 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1225 | if (ret > 0) { | |
1226 | ret = 0; | |
1227 | goto out; | |
1228 | } | |
1229 | if (ret < 0) | |
1230 | goto out; | |
1231 | ||
1232 | eb = path->nodes[0]; | |
1233 | ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]); | |
1234 | ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]); | |
1235 | while (ref_ptr < ref_end) { | |
1236 | char *name = NULL; | |
1237 | int namelen; | |
1238 | u64 parent_id; | |
1239 | ||
1240 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1241 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1242 | NULL, &parent_id); | |
1243 | } else { | |
1244 | parent_id = key->offset; | |
1245 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1246 | NULL); | |
1247 | } | |
1248 | if (ret) | |
1249 | goto out; | |
1250 | ||
1251 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
6ff49c6a NB |
1252 | ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot, |
1253 | parent_id, name, | |
1254 | namelen); | |
1f250e92 | 1255 | else |
9bb8407f NB |
1256 | ret = !!btrfs_find_name_in_backref(log_eb, log_slot, |
1257 | name, namelen); | |
1f250e92 FM |
1258 | |
1259 | if (!ret) { | |
1260 | struct inode *dir; | |
1261 | ||
1262 | btrfs_release_path(path); | |
1263 | dir = read_one_inode(root, parent_id); | |
1264 | if (!dir) { | |
1265 | ret = -ENOENT; | |
1266 | kfree(name); | |
1267 | goto out; | |
1268 | } | |
1269 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), | |
1270 | inode, name, namelen); | |
1271 | kfree(name); | |
1272 | iput(dir); | |
1273 | if (ret) | |
1274 | goto out; | |
1275 | goto again; | |
1276 | } | |
1277 | ||
1278 | kfree(name); | |
1279 | ref_ptr += namelen; | |
1280 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1281 | ref_ptr += sizeof(struct btrfs_inode_extref); | |
1282 | else | |
1283 | ref_ptr += sizeof(struct btrfs_inode_ref); | |
1284 | } | |
1285 | ret = 0; | |
1286 | out: | |
1287 | btrfs_release_path(path); | |
1288 | return ret; | |
1289 | } | |
1290 | ||
0d836392 FM |
1291 | static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir, |
1292 | const u8 ref_type, const char *name, | |
1293 | const int namelen) | |
1294 | { | |
1295 | struct btrfs_key key; | |
1296 | struct btrfs_path *path; | |
1297 | const u64 parent_id = btrfs_ino(BTRFS_I(dir)); | |
1298 | int ret; | |
1299 | ||
1300 | path = btrfs_alloc_path(); | |
1301 | if (!path) | |
1302 | return -ENOMEM; | |
1303 | ||
1304 | key.objectid = btrfs_ino(BTRFS_I(inode)); | |
1305 | key.type = ref_type; | |
1306 | if (key.type == BTRFS_INODE_REF_KEY) | |
1307 | key.offset = parent_id; | |
1308 | else | |
1309 | key.offset = btrfs_extref_hash(parent_id, name, namelen); | |
1310 | ||
1311 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0); | |
1312 | if (ret < 0) | |
1313 | goto out; | |
1314 | if (ret > 0) { | |
1315 | ret = 0; | |
1316 | goto out; | |
1317 | } | |
1318 | if (key.type == BTRFS_INODE_EXTREF_KEY) | |
6ff49c6a NB |
1319 | ret = !!btrfs_find_name_in_ext_backref(path->nodes[0], |
1320 | path->slots[0], parent_id, name, namelen); | |
0d836392 | 1321 | else |
9bb8407f NB |
1322 | ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0], |
1323 | name, namelen); | |
0d836392 FM |
1324 | |
1325 | out: | |
1326 | btrfs_free_path(path); | |
1327 | return ret; | |
1328 | } | |
1329 | ||
6b5fc433 FM |
1330 | static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1331 | struct inode *dir, struct inode *inode, const char *name, | |
1332 | int namelen, u64 ref_index) | |
1333 | { | |
1334 | struct btrfs_dir_item *dir_item; | |
1335 | struct btrfs_key key; | |
1336 | struct btrfs_path *path; | |
1337 | struct inode *other_inode = NULL; | |
1338 | int ret; | |
1339 | ||
1340 | path = btrfs_alloc_path(); | |
1341 | if (!path) | |
1342 | return -ENOMEM; | |
1343 | ||
1344 | dir_item = btrfs_lookup_dir_item(NULL, root, path, | |
1345 | btrfs_ino(BTRFS_I(dir)), | |
1346 | name, namelen, 0); | |
1347 | if (!dir_item) { | |
1348 | btrfs_release_path(path); | |
1349 | goto add_link; | |
1350 | } else if (IS_ERR(dir_item)) { | |
1351 | ret = PTR_ERR(dir_item); | |
1352 | goto out; | |
1353 | } | |
1354 | ||
1355 | /* | |
1356 | * Our inode's dentry collides with the dentry of another inode which is | |
1357 | * in the log but not yet processed since it has a higher inode number. | |
1358 | * So delete that other dentry. | |
1359 | */ | |
1360 | btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key); | |
1361 | btrfs_release_path(path); | |
1362 | other_inode = read_one_inode(root, key.objectid); | |
1363 | if (!other_inode) { | |
1364 | ret = -ENOENT; | |
1365 | goto out; | |
1366 | } | |
1367 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode), | |
1368 | name, namelen); | |
1369 | if (ret) | |
1370 | goto out; | |
1371 | /* | |
1372 | * If we dropped the link count to 0, bump it so that later the iput() | |
1373 | * on the inode will not free it. We will fixup the link count later. | |
1374 | */ | |
1375 | if (other_inode->i_nlink == 0) | |
1376 | inc_nlink(other_inode); | |
1377 | ||
1378 | ret = btrfs_run_delayed_items(trans); | |
1379 | if (ret) | |
1380 | goto out; | |
1381 | add_link: | |
1382 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), | |
1383 | name, namelen, 0, ref_index); | |
1384 | out: | |
1385 | iput(other_inode); | |
1386 | btrfs_free_path(path); | |
1387 | ||
1388 | return ret; | |
1389 | } | |
1390 | ||
5a1d7843 JS |
1391 | /* |
1392 | * replay one inode back reference item found in the log tree. | |
1393 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1394 | * root is the destination we are replaying into, and path is for temp | |
1395 | * use by this function. (it should be released on return). | |
1396 | */ | |
1397 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1398 | struct btrfs_root *root, | |
1399 | struct btrfs_root *log, | |
1400 | struct btrfs_path *path, | |
1401 | struct extent_buffer *eb, int slot, | |
1402 | struct btrfs_key *key) | |
1403 | { | |
03b2f08b GB |
1404 | struct inode *dir = NULL; |
1405 | struct inode *inode = NULL; | |
5a1d7843 JS |
1406 | unsigned long ref_ptr; |
1407 | unsigned long ref_end; | |
03b2f08b | 1408 | char *name = NULL; |
5a1d7843 JS |
1409 | int namelen; |
1410 | int ret; | |
1411 | int search_done = 0; | |
f186373f MF |
1412 | int log_ref_ver = 0; |
1413 | u64 parent_objectid; | |
1414 | u64 inode_objectid; | |
f46dbe3d | 1415 | u64 ref_index = 0; |
f186373f MF |
1416 | int ref_struct_size; |
1417 | ||
1418 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1419 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1420 | ||
1421 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1422 | struct btrfs_inode_extref *r; | |
1423 | ||
1424 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1425 | log_ref_ver = 1; | |
1426 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1427 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1428 | } else { | |
1429 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1430 | parent_objectid = key->offset; | |
1431 | } | |
1432 | inode_objectid = key->objectid; | |
e02119d5 | 1433 | |
5a1d7843 JS |
1434 | /* |
1435 | * it is possible that we didn't log all the parent directories | |
1436 | * for a given inode. If we don't find the dir, just don't | |
1437 | * copy the back ref in. The link count fixup code will take | |
1438 | * care of the rest | |
1439 | */ | |
f186373f | 1440 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1441 | if (!dir) { |
1442 | ret = -ENOENT; | |
1443 | goto out; | |
1444 | } | |
5a1d7843 | 1445 | |
f186373f | 1446 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1447 | if (!inode) { |
03b2f08b GB |
1448 | ret = -EIO; |
1449 | goto out; | |
5a1d7843 JS |
1450 | } |
1451 | ||
5a1d7843 | 1452 | while (ref_ptr < ref_end) { |
f186373f | 1453 | if (log_ref_ver) { |
bae15d95 QW |
1454 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, |
1455 | &ref_index, &parent_objectid); | |
f186373f MF |
1456 | /* |
1457 | * parent object can change from one array | |
1458 | * item to another. | |
1459 | */ | |
1460 | if (!dir) | |
1461 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1462 | if (!dir) { |
1463 | ret = -ENOENT; | |
1464 | goto out; | |
1465 | } | |
f186373f | 1466 | } else { |
bae15d95 QW |
1467 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, |
1468 | &ref_index); | |
f186373f MF |
1469 | } |
1470 | if (ret) | |
03b2f08b | 1471 | goto out; |
5a1d7843 JS |
1472 | |
1473 | /* if we already have a perfect match, we're done */ | |
f85b7379 DS |
1474 | if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), |
1475 | btrfs_ino(BTRFS_I(inode)), ref_index, | |
1476 | name, namelen)) { | |
5a1d7843 JS |
1477 | /* |
1478 | * look for a conflicting back reference in the | |
1479 | * metadata. if we find one we have to unlink that name | |
1480 | * of the file before we add our new link. Later on, we | |
1481 | * overwrite any existing back reference, and we don't | |
1482 | * want to create dangling pointers in the directory. | |
1483 | */ | |
1484 | ||
1485 | if (!search_done) { | |
1486 | ret = __add_inode_ref(trans, root, path, log, | |
94c91a1f | 1487 | BTRFS_I(dir), |
d75eefdf | 1488 | BTRFS_I(inode), |
f186373f MF |
1489 | inode_objectid, |
1490 | parent_objectid, | |
1491 | ref_index, name, namelen, | |
5a1d7843 | 1492 | &search_done); |
03b2f08b GB |
1493 | if (ret) { |
1494 | if (ret == 1) | |
1495 | ret = 0; | |
3650860b JB |
1496 | goto out; |
1497 | } | |
5a1d7843 JS |
1498 | } |
1499 | ||
0d836392 FM |
1500 | /* |
1501 | * If a reference item already exists for this inode | |
1502 | * with the same parent and name, but different index, | |
1503 | * drop it and the corresponding directory index entries | |
1504 | * from the parent before adding the new reference item | |
1505 | * and dir index entries, otherwise we would fail with | |
1506 | * -EEXIST returned from btrfs_add_link() below. | |
1507 | */ | |
1508 | ret = btrfs_inode_ref_exists(inode, dir, key->type, | |
1509 | name, namelen); | |
1510 | if (ret > 0) { | |
1511 | ret = btrfs_unlink_inode(trans, root, | |
1512 | BTRFS_I(dir), | |
1513 | BTRFS_I(inode), | |
1514 | name, namelen); | |
1515 | /* | |
1516 | * If we dropped the link count to 0, bump it so | |
1517 | * that later the iput() on the inode will not | |
1518 | * free it. We will fixup the link count later. | |
1519 | */ | |
1520 | if (!ret && inode->i_nlink == 0) | |
1521 | inc_nlink(inode); | |
1522 | } | |
1523 | if (ret < 0) | |
1524 | goto out; | |
1525 | ||
5a1d7843 | 1526 | /* insert our name */ |
6b5fc433 FM |
1527 | ret = add_link(trans, root, dir, inode, name, namelen, |
1528 | ref_index); | |
3650860b JB |
1529 | if (ret) |
1530 | goto out; | |
5a1d7843 JS |
1531 | |
1532 | btrfs_update_inode(trans, root, inode); | |
1533 | } | |
1534 | ||
f186373f | 1535 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1536 | kfree(name); |
03b2f08b | 1537 | name = NULL; |
f186373f MF |
1538 | if (log_ref_ver) { |
1539 | iput(dir); | |
1540 | dir = NULL; | |
1541 | } | |
5a1d7843 | 1542 | } |
e02119d5 | 1543 | |
1f250e92 FM |
1544 | /* |
1545 | * Before we overwrite the inode reference item in the subvolume tree | |
1546 | * with the item from the log tree, we must unlink all names from the | |
1547 | * parent directory that are in the subvolume's tree inode reference | |
1548 | * item, otherwise we end up with an inconsistent subvolume tree where | |
1549 | * dir index entries exist for a name but there is no inode reference | |
1550 | * item with the same name. | |
1551 | */ | |
1552 | ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot, | |
1553 | key); | |
1554 | if (ret) | |
1555 | goto out; | |
1556 | ||
e02119d5 CM |
1557 | /* finally write the back reference in the inode */ |
1558 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1559 | out: |
b3b4aa74 | 1560 | btrfs_release_path(path); |
03b2f08b | 1561 | kfree(name); |
e02119d5 CM |
1562 | iput(dir); |
1563 | iput(inode); | |
3650860b | 1564 | return ret; |
e02119d5 CM |
1565 | } |
1566 | ||
c71bf099 | 1567 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
9c4f61f0 | 1568 | struct btrfs_root *root, u64 ino) |
c71bf099 YZ |
1569 | { |
1570 | int ret; | |
381cf658 | 1571 | |
9c4f61f0 DS |
1572 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1573 | if (ret == -EEXIST) | |
1574 | ret = 0; | |
381cf658 | 1575 | |
c71bf099 YZ |
1576 | return ret; |
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); |
e02119d5 CM |
1719 | btrfs_update_inode(trans, root, inode); |
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 | } |
33345d01 | 1730 | ret = insert_orphan_item(trans, root, ino); |
12fcfd22 | 1731 | } |
12fcfd22 | 1732 | |
f186373f MF |
1733 | out: |
1734 | btrfs_free_path(path); | |
1735 | return ret; | |
e02119d5 CM |
1736 | } |
1737 | ||
1738 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1739 | struct btrfs_root *root, | |
1740 | struct btrfs_path *path) | |
1741 | { | |
1742 | int ret; | |
1743 | struct btrfs_key key; | |
1744 | struct inode *inode; | |
1745 | ||
1746 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1747 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1748 | key.offset = (u64)-1; | |
d397712b | 1749 | while (1) { |
e02119d5 CM |
1750 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1751 | if (ret < 0) | |
1752 | break; | |
1753 | ||
1754 | if (ret == 1) { | |
1755 | if (path->slots[0] == 0) | |
1756 | break; | |
1757 | path->slots[0]--; | |
1758 | } | |
1759 | ||
1760 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1761 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1762 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1763 | break; | |
1764 | ||
1765 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1766 | if (ret) |
1767 | goto out; | |
e02119d5 | 1768 | |
b3b4aa74 | 1769 | btrfs_release_path(path); |
e02119d5 | 1770 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1771 | if (!inode) |
1772 | return -EIO; | |
e02119d5 CM |
1773 | |
1774 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1775 | iput(inode); |
3650860b JB |
1776 | if (ret) |
1777 | goto out; | |
e02119d5 | 1778 | |
12fcfd22 CM |
1779 | /* |
1780 | * fixup on a directory may create new entries, | |
1781 | * make sure we always look for the highset possible | |
1782 | * offset | |
1783 | */ | |
1784 | key.offset = (u64)-1; | |
e02119d5 | 1785 | } |
65a246c5 TI |
1786 | ret = 0; |
1787 | out: | |
b3b4aa74 | 1788 | btrfs_release_path(path); |
65a246c5 | 1789 | return ret; |
e02119d5 CM |
1790 | } |
1791 | ||
1792 | ||
1793 | /* | |
1794 | * record a given inode in the fixup dir so we can check its link | |
1795 | * count when replay is done. The link count is incremented here | |
1796 | * so the inode won't go away until we check it | |
1797 | */ | |
1798 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1799 | struct btrfs_root *root, | |
1800 | struct btrfs_path *path, | |
1801 | u64 objectid) | |
1802 | { | |
1803 | struct btrfs_key key; | |
1804 | int ret = 0; | |
1805 | struct inode *inode; | |
1806 | ||
1807 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1808 | if (!inode) |
1809 | return -EIO; | |
e02119d5 CM |
1810 | |
1811 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1812 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1813 | key.offset = objectid; |
1814 | ||
1815 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1816 | ||
b3b4aa74 | 1817 | btrfs_release_path(path); |
e02119d5 | 1818 | if (ret == 0) { |
9bf7a489 JB |
1819 | if (!inode->i_nlink) |
1820 | set_nlink(inode, 1); | |
1821 | else | |
8b558c5f | 1822 | inc_nlink(inode); |
b9959295 | 1823 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
1824 | } else if (ret == -EEXIST) { |
1825 | ret = 0; | |
1826 | } else { | |
3650860b | 1827 | BUG(); /* Logic Error */ |
e02119d5 CM |
1828 | } |
1829 | iput(inode); | |
1830 | ||
1831 | return ret; | |
1832 | } | |
1833 | ||
1834 | /* | |
1835 | * when replaying the log for a directory, we only insert names | |
1836 | * for inodes that actually exist. This means an fsync on a directory | |
1837 | * does not implicitly fsync all the new files in it | |
1838 | */ | |
1839 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1840 | struct btrfs_root *root, | |
e02119d5 | 1841 | u64 dirid, u64 index, |
60d53eb3 | 1842 | char *name, int name_len, |
e02119d5 CM |
1843 | struct btrfs_key *location) |
1844 | { | |
1845 | struct inode *inode; | |
1846 | struct inode *dir; | |
1847 | int ret; | |
1848 | ||
1849 | inode = read_one_inode(root, location->objectid); | |
1850 | if (!inode) | |
1851 | return -ENOENT; | |
1852 | ||
1853 | dir = read_one_inode(root, dirid); | |
1854 | if (!dir) { | |
1855 | iput(inode); | |
1856 | return -EIO; | |
1857 | } | |
d555438b | 1858 | |
db0a669f NB |
1859 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, |
1860 | name_len, 1, index); | |
e02119d5 CM |
1861 | |
1862 | /* FIXME, put inode into FIXUP list */ | |
1863 | ||
1864 | iput(inode); | |
1865 | iput(dir); | |
1866 | return ret; | |
1867 | } | |
1868 | ||
1869 | /* | |
1870 | * take a single entry in a log directory item and replay it into | |
1871 | * the subvolume. | |
1872 | * | |
1873 | * if a conflicting item exists in the subdirectory already, | |
1874 | * the inode it points to is unlinked and put into the link count | |
1875 | * fix up tree. | |
1876 | * | |
1877 | * If a name from the log points to a file or directory that does | |
1878 | * not exist in the FS, it is skipped. fsyncs on directories | |
1879 | * do not force down inodes inside that directory, just changes to the | |
1880 | * names or unlinks in a directory. | |
bb53eda9 FM |
1881 | * |
1882 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1883 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1884 | */ |
1885 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1886 | struct btrfs_root *root, | |
1887 | struct btrfs_path *path, | |
1888 | struct extent_buffer *eb, | |
1889 | struct btrfs_dir_item *di, | |
1890 | struct btrfs_key *key) | |
1891 | { | |
1892 | char *name; | |
1893 | int name_len; | |
1894 | struct btrfs_dir_item *dst_di; | |
1895 | struct btrfs_key found_key; | |
1896 | struct btrfs_key log_key; | |
1897 | struct inode *dir; | |
e02119d5 | 1898 | u8 log_type; |
4bef0848 | 1899 | int exists; |
3650860b | 1900 | int ret = 0; |
d555438b | 1901 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1902 | bool name_added = false; |
e02119d5 CM |
1903 | |
1904 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1905 | if (!dir) |
1906 | return -EIO; | |
e02119d5 CM |
1907 | |
1908 | name_len = btrfs_dir_name_len(eb, di); | |
1909 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1910 | if (!name) { |
1911 | ret = -ENOMEM; | |
1912 | goto out; | |
1913 | } | |
2a29edc6 | 1914 | |
e02119d5 CM |
1915 | log_type = btrfs_dir_type(eb, di); |
1916 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1917 | name_len); | |
1918 | ||
1919 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1920 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1921 | if (exists == 0) | |
1922 | exists = 1; | |
1923 | else | |
1924 | exists = 0; | |
b3b4aa74 | 1925 | btrfs_release_path(path); |
4bef0848 | 1926 | |
e02119d5 CM |
1927 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1928 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1929 | name, name_len, 1); | |
d397712b | 1930 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1931 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1932 | key->objectid, | |
1933 | key->offset, name, | |
1934 | name_len, 1); | |
1935 | } else { | |
3650860b JB |
1936 | /* Corruption */ |
1937 | ret = -EINVAL; | |
1938 | goto out; | |
e02119d5 | 1939 | } |
c704005d | 1940 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1941 | /* we need a sequence number to insert, so we only |
1942 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1943 | */ | |
1944 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1945 | goto out; | |
1946 | goto insert; | |
1947 | } | |
1948 | ||
1949 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1950 | /* the existing item matches the logged item */ | |
1951 | if (found_key.objectid == log_key.objectid && | |
1952 | found_key.type == log_key.type && | |
1953 | found_key.offset == log_key.offset && | |
1954 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1955 | update_size = false; |
e02119d5 CM |
1956 | goto out; |
1957 | } | |
1958 | ||
1959 | /* | |
1960 | * don't drop the conflicting directory entry if the inode | |
1961 | * for the new entry doesn't exist | |
1962 | */ | |
4bef0848 | 1963 | if (!exists) |
e02119d5 CM |
1964 | goto out; |
1965 | ||
207e7d92 | 1966 | ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di); |
3650860b JB |
1967 | if (ret) |
1968 | goto out; | |
e02119d5 CM |
1969 | |
1970 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1971 | goto insert; | |
1972 | out: | |
b3b4aa74 | 1973 | btrfs_release_path(path); |
d555438b | 1974 | if (!ret && update_size) { |
6ef06d27 | 1975 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2); |
d555438b JB |
1976 | ret = btrfs_update_inode(trans, root, dir); |
1977 | } | |
e02119d5 CM |
1978 | kfree(name); |
1979 | iput(dir); | |
bb53eda9 FM |
1980 | if (!ret && name_added) |
1981 | ret = 1; | |
3650860b | 1982 | return ret; |
e02119d5 CM |
1983 | |
1984 | insert: | |
725af92a NB |
1985 | /* |
1986 | * Check if the inode reference exists in the log for the given name, | |
1987 | * inode and parent inode | |
1988 | */ | |
1989 | found_key.objectid = log_key.objectid; | |
1990 | found_key.type = BTRFS_INODE_REF_KEY; | |
1991 | found_key.offset = key->objectid; | |
1992 | ret = backref_in_log(root->log_root, &found_key, 0, name, name_len); | |
1993 | if (ret < 0) { | |
1994 | goto out; | |
1995 | } else if (ret) { | |
1996 | /* The dentry will be added later. */ | |
1997 | ret = 0; | |
1998 | update_size = false; | |
1999 | goto out; | |
2000 | } | |
2001 | ||
2002 | found_key.objectid = log_key.objectid; | |
2003 | found_key.type = BTRFS_INODE_EXTREF_KEY; | |
2004 | found_key.offset = key->objectid; | |
2005 | ret = backref_in_log(root->log_root, &found_key, key->objectid, name, | |
2006 | name_len); | |
2007 | if (ret < 0) { | |
2008 | goto out; | |
2009 | } else if (ret) { | |
df8d116f FM |
2010 | /* The dentry will be added later. */ |
2011 | ret = 0; | |
2012 | update_size = false; | |
2013 | goto out; | |
2014 | } | |
b3b4aa74 | 2015 | btrfs_release_path(path); |
60d53eb3 Z |
2016 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
2017 | name, name_len, &log_key); | |
df8d116f | 2018 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 2019 | goto out; |
bb53eda9 FM |
2020 | if (!ret) |
2021 | name_added = true; | |
d555438b | 2022 | update_size = false; |
3650860b | 2023 | ret = 0; |
e02119d5 CM |
2024 | goto out; |
2025 | } | |
2026 | ||
2027 | /* | |
2028 | * find all the names in a directory item and reconcile them into | |
2029 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
2030 | * one name in a directory item, but the same code gets used for | |
2031 | * both directory index types | |
2032 | */ | |
2033 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
2034 | struct btrfs_root *root, | |
2035 | struct btrfs_path *path, | |
2036 | struct extent_buffer *eb, int slot, | |
2037 | struct btrfs_key *key) | |
2038 | { | |
bb53eda9 | 2039 | int ret = 0; |
e02119d5 CM |
2040 | u32 item_size = btrfs_item_size_nr(eb, slot); |
2041 | struct btrfs_dir_item *di; | |
2042 | int name_len; | |
2043 | unsigned long ptr; | |
2044 | unsigned long ptr_end; | |
bb53eda9 | 2045 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
2046 | |
2047 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2048 | ptr_end = ptr + item_size; | |
d397712b | 2049 | while (ptr < ptr_end) { |
e02119d5 CM |
2050 | di = (struct btrfs_dir_item *)ptr; |
2051 | name_len = btrfs_dir_name_len(eb, di); | |
2052 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
2053 | if (ret < 0) |
2054 | break; | |
e02119d5 CM |
2055 | ptr = (unsigned long)(di + 1); |
2056 | ptr += name_len; | |
bb53eda9 FM |
2057 | |
2058 | /* | |
2059 | * If this entry refers to a non-directory (directories can not | |
2060 | * have a link count > 1) and it was added in the transaction | |
2061 | * that was not committed, make sure we fixup the link count of | |
2062 | * the inode it the entry points to. Otherwise something like | |
2063 | * the following would result in a directory pointing to an | |
2064 | * inode with a wrong link that does not account for this dir | |
2065 | * entry: | |
2066 | * | |
2067 | * mkdir testdir | |
2068 | * touch testdir/foo | |
2069 | * touch testdir/bar | |
2070 | * sync | |
2071 | * | |
2072 | * ln testdir/bar testdir/bar_link | |
2073 | * ln testdir/foo testdir/foo_link | |
2074 | * xfs_io -c "fsync" testdir/bar | |
2075 | * | |
2076 | * <power failure> | |
2077 | * | |
2078 | * mount fs, log replay happens | |
2079 | * | |
2080 | * File foo would remain with a link count of 1 when it has two | |
2081 | * entries pointing to it in the directory testdir. This would | |
2082 | * make it impossible to ever delete the parent directory has | |
2083 | * it would result in stale dentries that can never be deleted. | |
2084 | */ | |
2085 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
2086 | struct btrfs_key di_key; | |
2087 | ||
2088 | if (!fixup_path) { | |
2089 | fixup_path = btrfs_alloc_path(); | |
2090 | if (!fixup_path) { | |
2091 | ret = -ENOMEM; | |
2092 | break; | |
2093 | } | |
2094 | } | |
2095 | ||
2096 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2097 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
2098 | di_key.objectid); | |
2099 | if (ret) | |
2100 | break; | |
2101 | } | |
2102 | ret = 0; | |
e02119d5 | 2103 | } |
bb53eda9 FM |
2104 | btrfs_free_path(fixup_path); |
2105 | return ret; | |
e02119d5 CM |
2106 | } |
2107 | ||
2108 | /* | |
2109 | * directory replay has two parts. There are the standard directory | |
2110 | * items in the log copied from the subvolume, and range items | |
2111 | * created in the log while the subvolume was logged. | |
2112 | * | |
2113 | * The range items tell us which parts of the key space the log | |
2114 | * is authoritative for. During replay, if a key in the subvolume | |
2115 | * directory is in a logged range item, but not actually in the log | |
2116 | * that means it was deleted from the directory before the fsync | |
2117 | * and should be removed. | |
2118 | */ | |
2119 | static noinline int find_dir_range(struct btrfs_root *root, | |
2120 | struct btrfs_path *path, | |
2121 | u64 dirid, int key_type, | |
2122 | u64 *start_ret, u64 *end_ret) | |
2123 | { | |
2124 | struct btrfs_key key; | |
2125 | u64 found_end; | |
2126 | struct btrfs_dir_log_item *item; | |
2127 | int ret; | |
2128 | int nritems; | |
2129 | ||
2130 | if (*start_ret == (u64)-1) | |
2131 | return 1; | |
2132 | ||
2133 | key.objectid = dirid; | |
2134 | key.type = key_type; | |
2135 | key.offset = *start_ret; | |
2136 | ||
2137 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2138 | if (ret < 0) | |
2139 | goto out; | |
2140 | if (ret > 0) { | |
2141 | if (path->slots[0] == 0) | |
2142 | goto out; | |
2143 | path->slots[0]--; | |
2144 | } | |
2145 | if (ret != 0) | |
2146 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2147 | ||
2148 | if (key.type != key_type || key.objectid != dirid) { | |
2149 | ret = 1; | |
2150 | goto next; | |
2151 | } | |
2152 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2153 | struct btrfs_dir_log_item); | |
2154 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2155 | ||
2156 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
2157 | ret = 0; | |
2158 | *start_ret = key.offset; | |
2159 | *end_ret = found_end; | |
2160 | goto out; | |
2161 | } | |
2162 | ret = 1; | |
2163 | next: | |
2164 | /* check the next slot in the tree to see if it is a valid item */ | |
2165 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 2166 | path->slots[0]++; |
e02119d5 CM |
2167 | if (path->slots[0] >= nritems) { |
2168 | ret = btrfs_next_leaf(root, path); | |
2169 | if (ret) | |
2170 | goto out; | |
e02119d5 CM |
2171 | } |
2172 | ||
2173 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2174 | ||
2175 | if (key.type != key_type || key.objectid != dirid) { | |
2176 | ret = 1; | |
2177 | goto out; | |
2178 | } | |
2179 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2180 | struct btrfs_dir_log_item); | |
2181 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2182 | *start_ret = key.offset; | |
2183 | *end_ret = found_end; | |
2184 | ret = 0; | |
2185 | out: | |
b3b4aa74 | 2186 | btrfs_release_path(path); |
e02119d5 CM |
2187 | return ret; |
2188 | } | |
2189 | ||
2190 | /* | |
2191 | * this looks for a given directory item in the log. If the directory | |
2192 | * item is not in the log, the item is removed and the inode it points | |
2193 | * to is unlinked | |
2194 | */ | |
2195 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
2196 | struct btrfs_root *root, | |
2197 | struct btrfs_root *log, | |
2198 | struct btrfs_path *path, | |
2199 | struct btrfs_path *log_path, | |
2200 | struct inode *dir, | |
2201 | struct btrfs_key *dir_key) | |
2202 | { | |
2203 | int ret; | |
2204 | struct extent_buffer *eb; | |
2205 | int slot; | |
2206 | u32 item_size; | |
2207 | struct btrfs_dir_item *di; | |
2208 | struct btrfs_dir_item *log_di; | |
2209 | int name_len; | |
2210 | unsigned long ptr; | |
2211 | unsigned long ptr_end; | |
2212 | char *name; | |
2213 | struct inode *inode; | |
2214 | struct btrfs_key location; | |
2215 | ||
2216 | again: | |
2217 | eb = path->nodes[0]; | |
2218 | slot = path->slots[0]; | |
2219 | item_size = btrfs_item_size_nr(eb, slot); | |
2220 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2221 | ptr_end = ptr + item_size; | |
d397712b | 2222 | while (ptr < ptr_end) { |
e02119d5 CM |
2223 | di = (struct btrfs_dir_item *)ptr; |
2224 | name_len = btrfs_dir_name_len(eb, di); | |
2225 | name = kmalloc(name_len, GFP_NOFS); | |
2226 | if (!name) { | |
2227 | ret = -ENOMEM; | |
2228 | goto out; | |
2229 | } | |
2230 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2231 | name_len); | |
2232 | log_di = NULL; | |
12fcfd22 | 2233 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2234 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
2235 | dir_key->objectid, | |
2236 | name, name_len, 0); | |
12fcfd22 | 2237 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
2238 | log_di = btrfs_lookup_dir_index_item(trans, log, |
2239 | log_path, | |
2240 | dir_key->objectid, | |
2241 | dir_key->offset, | |
2242 | name, name_len, 0); | |
2243 | } | |
8d9e220c | 2244 | if (!log_di || log_di == ERR_PTR(-ENOENT)) { |
e02119d5 | 2245 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
2246 | btrfs_release_path(path); |
2247 | btrfs_release_path(log_path); | |
e02119d5 | 2248 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
2249 | if (!inode) { |
2250 | kfree(name); | |
2251 | return -EIO; | |
2252 | } | |
e02119d5 CM |
2253 | |
2254 | ret = link_to_fixup_dir(trans, root, | |
2255 | path, location.objectid); | |
3650860b JB |
2256 | if (ret) { |
2257 | kfree(name); | |
2258 | iput(inode); | |
2259 | goto out; | |
2260 | } | |
2261 | ||
8b558c5f | 2262 | inc_nlink(inode); |
4ec5934e NB |
2263 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
2264 | BTRFS_I(inode), name, name_len); | |
3650860b | 2265 | if (!ret) |
e5c304e6 | 2266 | ret = btrfs_run_delayed_items(trans); |
e02119d5 CM |
2267 | kfree(name); |
2268 | iput(inode); | |
3650860b JB |
2269 | if (ret) |
2270 | goto out; | |
e02119d5 CM |
2271 | |
2272 | /* there might still be more names under this key | |
2273 | * check and repeat if required | |
2274 | */ | |
2275 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2276 | 0, 0); | |
2277 | if (ret == 0) | |
2278 | goto again; | |
2279 | ret = 0; | |
2280 | goto out; | |
269d040f FDBM |
2281 | } else if (IS_ERR(log_di)) { |
2282 | kfree(name); | |
2283 | return PTR_ERR(log_di); | |
e02119d5 | 2284 | } |
b3b4aa74 | 2285 | btrfs_release_path(log_path); |
e02119d5 CM |
2286 | kfree(name); |
2287 | ||
2288 | ptr = (unsigned long)(di + 1); | |
2289 | ptr += name_len; | |
2290 | } | |
2291 | ret = 0; | |
2292 | out: | |
b3b4aa74 DS |
2293 | btrfs_release_path(path); |
2294 | btrfs_release_path(log_path); | |
e02119d5 CM |
2295 | return ret; |
2296 | } | |
2297 | ||
4f764e51 FM |
2298 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2299 | struct btrfs_root *root, | |
2300 | struct btrfs_root *log, | |
2301 | struct btrfs_path *path, | |
2302 | const u64 ino) | |
2303 | { | |
2304 | struct btrfs_key search_key; | |
2305 | struct btrfs_path *log_path; | |
2306 | int i; | |
2307 | int nritems; | |
2308 | int ret; | |
2309 | ||
2310 | log_path = btrfs_alloc_path(); | |
2311 | if (!log_path) | |
2312 | return -ENOMEM; | |
2313 | ||
2314 | search_key.objectid = ino; | |
2315 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2316 | search_key.offset = 0; | |
2317 | again: | |
2318 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2319 | if (ret < 0) | |
2320 | goto out; | |
2321 | process_leaf: | |
2322 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2323 | for (i = path->slots[0]; i < nritems; i++) { | |
2324 | struct btrfs_key key; | |
2325 | struct btrfs_dir_item *di; | |
2326 | struct btrfs_dir_item *log_di; | |
2327 | u32 total_size; | |
2328 | u32 cur; | |
2329 | ||
2330 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2331 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2332 | ret = 0; | |
2333 | goto out; | |
2334 | } | |
2335 | ||
2336 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2337 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2338 | cur = 0; | |
2339 | while (cur < total_size) { | |
2340 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2341 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2342 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2343 | char *name; | |
2344 | ||
2345 | name = kmalloc(name_len, GFP_NOFS); | |
2346 | if (!name) { | |
2347 | ret = -ENOMEM; | |
2348 | goto out; | |
2349 | } | |
2350 | read_extent_buffer(path->nodes[0], name, | |
2351 | (unsigned long)(di + 1), name_len); | |
2352 | ||
2353 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2354 | name, name_len, 0); | |
2355 | btrfs_release_path(log_path); | |
2356 | if (!log_di) { | |
2357 | /* Doesn't exist in log tree, so delete it. */ | |
2358 | btrfs_release_path(path); | |
2359 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2360 | name, name_len, -1); | |
2361 | kfree(name); | |
2362 | if (IS_ERR(di)) { | |
2363 | ret = PTR_ERR(di); | |
2364 | goto out; | |
2365 | } | |
2366 | ASSERT(di); | |
2367 | ret = btrfs_delete_one_dir_name(trans, root, | |
2368 | path, di); | |
2369 | if (ret) | |
2370 | goto out; | |
2371 | btrfs_release_path(path); | |
2372 | search_key = key; | |
2373 | goto again; | |
2374 | } | |
2375 | kfree(name); | |
2376 | if (IS_ERR(log_di)) { | |
2377 | ret = PTR_ERR(log_di); | |
2378 | goto out; | |
2379 | } | |
2380 | cur += this_len; | |
2381 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2382 | } | |
2383 | } | |
2384 | ret = btrfs_next_leaf(root, path); | |
2385 | if (ret > 0) | |
2386 | ret = 0; | |
2387 | else if (ret == 0) | |
2388 | goto process_leaf; | |
2389 | out: | |
2390 | btrfs_free_path(log_path); | |
2391 | btrfs_release_path(path); | |
2392 | return ret; | |
2393 | } | |
2394 | ||
2395 | ||
e02119d5 CM |
2396 | /* |
2397 | * deletion replay happens before we copy any new directory items | |
2398 | * out of the log or out of backreferences from inodes. It | |
2399 | * scans the log to find ranges of keys that log is authoritative for, | |
2400 | * and then scans the directory to find items in those ranges that are | |
2401 | * not present in the log. | |
2402 | * | |
2403 | * Anything we don't find in the log is unlinked and removed from the | |
2404 | * directory. | |
2405 | */ | |
2406 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2407 | struct btrfs_root *root, | |
2408 | struct btrfs_root *log, | |
2409 | struct btrfs_path *path, | |
12fcfd22 | 2410 | u64 dirid, int del_all) |
e02119d5 CM |
2411 | { |
2412 | u64 range_start; | |
2413 | u64 range_end; | |
2414 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2415 | int ret = 0; | |
2416 | struct btrfs_key dir_key; | |
2417 | struct btrfs_key found_key; | |
2418 | struct btrfs_path *log_path; | |
2419 | struct inode *dir; | |
2420 | ||
2421 | dir_key.objectid = dirid; | |
2422 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2423 | log_path = btrfs_alloc_path(); | |
2424 | if (!log_path) | |
2425 | return -ENOMEM; | |
2426 | ||
2427 | dir = read_one_inode(root, dirid); | |
2428 | /* it isn't an error if the inode isn't there, that can happen | |
2429 | * because we replay the deletes before we copy in the inode item | |
2430 | * from the log | |
2431 | */ | |
2432 | if (!dir) { | |
2433 | btrfs_free_path(log_path); | |
2434 | return 0; | |
2435 | } | |
2436 | again: | |
2437 | range_start = 0; | |
2438 | range_end = 0; | |
d397712b | 2439 | while (1) { |
12fcfd22 CM |
2440 | if (del_all) |
2441 | range_end = (u64)-1; | |
2442 | else { | |
2443 | ret = find_dir_range(log, path, dirid, key_type, | |
2444 | &range_start, &range_end); | |
2445 | if (ret != 0) | |
2446 | break; | |
2447 | } | |
e02119d5 CM |
2448 | |
2449 | dir_key.offset = range_start; | |
d397712b | 2450 | while (1) { |
e02119d5 CM |
2451 | int nritems; |
2452 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2453 | 0, 0); | |
2454 | if (ret < 0) | |
2455 | goto out; | |
2456 | ||
2457 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2458 | if (path->slots[0] >= nritems) { | |
2459 | ret = btrfs_next_leaf(root, path); | |
b98def7c | 2460 | if (ret == 1) |
e02119d5 | 2461 | break; |
b98def7c LB |
2462 | else if (ret < 0) |
2463 | goto out; | |
e02119d5 CM |
2464 | } |
2465 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2466 | path->slots[0]); | |
2467 | if (found_key.objectid != dirid || | |
2468 | found_key.type != dir_key.type) | |
2469 | goto next_type; | |
2470 | ||
2471 | if (found_key.offset > range_end) | |
2472 | break; | |
2473 | ||
2474 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2475 | log_path, dir, |
2476 | &found_key); | |
3650860b JB |
2477 | if (ret) |
2478 | goto out; | |
e02119d5 CM |
2479 | if (found_key.offset == (u64)-1) |
2480 | break; | |
2481 | dir_key.offset = found_key.offset + 1; | |
2482 | } | |
b3b4aa74 | 2483 | btrfs_release_path(path); |
e02119d5 CM |
2484 | if (range_end == (u64)-1) |
2485 | break; | |
2486 | range_start = range_end + 1; | |
2487 | } | |
2488 | ||
2489 | next_type: | |
2490 | ret = 0; | |
2491 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2492 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2493 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2494 | btrfs_release_path(path); |
e02119d5 CM |
2495 | goto again; |
2496 | } | |
2497 | out: | |
b3b4aa74 | 2498 | btrfs_release_path(path); |
e02119d5 CM |
2499 | btrfs_free_path(log_path); |
2500 | iput(dir); | |
2501 | return ret; | |
2502 | } | |
2503 | ||
2504 | /* | |
2505 | * the process_func used to replay items from the log tree. This | |
2506 | * gets called in two different stages. The first stage just looks | |
2507 | * for inodes and makes sure they are all copied into the subvolume. | |
2508 | * | |
2509 | * The second stage copies all the other item types from the log into | |
2510 | * the subvolume. The two stage approach is slower, but gets rid of | |
2511 | * lots of complexity around inodes referencing other inodes that exist | |
2512 | * only in the log (references come from either directory items or inode | |
2513 | * back refs). | |
2514 | */ | |
2515 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 2516 | struct walk_control *wc, u64 gen, int level) |
e02119d5 CM |
2517 | { |
2518 | int nritems; | |
2519 | struct btrfs_path *path; | |
2520 | struct btrfs_root *root = wc->replay_dest; | |
2521 | struct btrfs_key key; | |
e02119d5 CM |
2522 | int i; |
2523 | int ret; | |
2524 | ||
581c1760 | 2525 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
018642a1 TI |
2526 | if (ret) |
2527 | return ret; | |
e02119d5 CM |
2528 | |
2529 | level = btrfs_header_level(eb); | |
2530 | ||
2531 | if (level != 0) | |
2532 | return 0; | |
2533 | ||
2534 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2535 | if (!path) |
2536 | return -ENOMEM; | |
e02119d5 CM |
2537 | |
2538 | nritems = btrfs_header_nritems(eb); | |
2539 | for (i = 0; i < nritems; i++) { | |
2540 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2541 | |
2542 | /* inode keys are done during the first stage */ | |
2543 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2544 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2545 | struct btrfs_inode_item *inode_item; |
2546 | u32 mode; | |
2547 | ||
2548 | inode_item = btrfs_item_ptr(eb, i, | |
2549 | struct btrfs_inode_item); | |
f2d72f42 FM |
2550 | /* |
2551 | * If we have a tmpfile (O_TMPFILE) that got fsync'ed | |
2552 | * and never got linked before the fsync, skip it, as | |
2553 | * replaying it is pointless since it would be deleted | |
2554 | * later. We skip logging tmpfiles, but it's always | |
2555 | * possible we are replaying a log created with a kernel | |
2556 | * that used to log tmpfiles. | |
2557 | */ | |
2558 | if (btrfs_inode_nlink(eb, inode_item) == 0) { | |
2559 | wc->ignore_cur_inode = true; | |
2560 | continue; | |
2561 | } else { | |
2562 | wc->ignore_cur_inode = false; | |
2563 | } | |
4f764e51 FM |
2564 | ret = replay_xattr_deletes(wc->trans, root, log, |
2565 | path, key.objectid); | |
2566 | if (ret) | |
2567 | break; | |
e02119d5 CM |
2568 | mode = btrfs_inode_mode(eb, inode_item); |
2569 | if (S_ISDIR(mode)) { | |
2570 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2571 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2572 | if (ret) |
2573 | break; | |
e02119d5 CM |
2574 | } |
2575 | ret = overwrite_item(wc->trans, root, path, | |
2576 | eb, i, &key); | |
b50c6e25 JB |
2577 | if (ret) |
2578 | break; | |
e02119d5 | 2579 | |
471d557a FM |
2580 | /* |
2581 | * Before replaying extents, truncate the inode to its | |
2582 | * size. We need to do it now and not after log replay | |
2583 | * because before an fsync we can have prealloc extents | |
2584 | * added beyond the inode's i_size. If we did it after, | |
2585 | * through orphan cleanup for example, we would drop | |
2586 | * those prealloc extents just after replaying them. | |
e02119d5 CM |
2587 | */ |
2588 | if (S_ISREG(mode)) { | |
471d557a FM |
2589 | struct inode *inode; |
2590 | u64 from; | |
2591 | ||
2592 | inode = read_one_inode(root, key.objectid); | |
2593 | if (!inode) { | |
2594 | ret = -EIO; | |
2595 | break; | |
2596 | } | |
2597 | from = ALIGN(i_size_read(inode), | |
2598 | root->fs_info->sectorsize); | |
2599 | ret = btrfs_drop_extents(wc->trans, root, inode, | |
2600 | from, (u64)-1, 1); | |
471d557a | 2601 | if (!ret) { |
f2d72f42 | 2602 | /* Update the inode's nbytes. */ |
471d557a FM |
2603 | ret = btrfs_update_inode(wc->trans, |
2604 | root, inode); | |
2605 | } | |
2606 | iput(inode); | |
b50c6e25 JB |
2607 | if (ret) |
2608 | break; | |
e02119d5 | 2609 | } |
c71bf099 | 2610 | |
e02119d5 CM |
2611 | ret = link_to_fixup_dir(wc->trans, root, |
2612 | path, key.objectid); | |
b50c6e25 JB |
2613 | if (ret) |
2614 | break; | |
e02119d5 | 2615 | } |
dd8e7217 | 2616 | |
f2d72f42 FM |
2617 | if (wc->ignore_cur_inode) |
2618 | continue; | |
2619 | ||
dd8e7217 JB |
2620 | if (key.type == BTRFS_DIR_INDEX_KEY && |
2621 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2622 | ret = replay_one_dir_item(wc->trans, root, path, | |
2623 | eb, i, &key); | |
2624 | if (ret) | |
2625 | break; | |
2626 | } | |
2627 | ||
e02119d5 CM |
2628 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2629 | continue; | |
2630 | ||
2631 | /* these keys are simply copied */ | |
2632 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2633 | ret = overwrite_item(wc->trans, root, path, | |
2634 | eb, i, &key); | |
b50c6e25 JB |
2635 | if (ret) |
2636 | break; | |
2da1c669 LB |
2637 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2638 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2639 | ret = add_inode_ref(wc->trans, root, log, path, |
2640 | eb, i, &key); | |
b50c6e25 JB |
2641 | if (ret && ret != -ENOENT) |
2642 | break; | |
2643 | ret = 0; | |
e02119d5 CM |
2644 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2645 | ret = replay_one_extent(wc->trans, root, path, | |
2646 | eb, i, &key); | |
b50c6e25 JB |
2647 | if (ret) |
2648 | break; | |
dd8e7217 | 2649 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2650 | ret = replay_one_dir_item(wc->trans, root, path, |
2651 | eb, i, &key); | |
b50c6e25 JB |
2652 | if (ret) |
2653 | break; | |
e02119d5 CM |
2654 | } |
2655 | } | |
2656 | btrfs_free_path(path); | |
b50c6e25 | 2657 | return ret; |
e02119d5 CM |
2658 | } |
2659 | ||
6787bb9f NB |
2660 | /* |
2661 | * Correctly adjust the reserved bytes occupied by a log tree extent buffer | |
2662 | */ | |
2663 | static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start) | |
2664 | { | |
2665 | struct btrfs_block_group *cache; | |
2666 | ||
2667 | cache = btrfs_lookup_block_group(fs_info, start); | |
2668 | if (!cache) { | |
2669 | btrfs_err(fs_info, "unable to find block group for %llu", start); | |
2670 | return; | |
2671 | } | |
2672 | ||
2673 | spin_lock(&cache->space_info->lock); | |
2674 | spin_lock(&cache->lock); | |
2675 | cache->reserved -= fs_info->nodesize; | |
2676 | cache->space_info->bytes_reserved -= fs_info->nodesize; | |
2677 | spin_unlock(&cache->lock); | |
2678 | spin_unlock(&cache->space_info->lock); | |
2679 | ||
2680 | btrfs_put_block_group(cache); | |
2681 | } | |
2682 | ||
d397712b | 2683 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2684 | struct btrfs_root *root, |
2685 | struct btrfs_path *path, int *level, | |
2686 | struct walk_control *wc) | |
2687 | { | |
0b246afa | 2688 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2689 | u64 bytenr; |
2690 | u64 ptr_gen; | |
2691 | struct extent_buffer *next; | |
2692 | struct extent_buffer *cur; | |
e02119d5 CM |
2693 | u32 blocksize; |
2694 | int ret = 0; | |
2695 | ||
d397712b | 2696 | while (*level > 0) { |
581c1760 QW |
2697 | struct btrfs_key first_key; |
2698 | ||
e02119d5 CM |
2699 | cur = path->nodes[*level]; |
2700 | ||
fae7f21c | 2701 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2702 | |
2703 | if (path->slots[*level] >= | |
2704 | btrfs_header_nritems(cur)) | |
2705 | break; | |
2706 | ||
2707 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2708 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
581c1760 | 2709 | btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]); |
0b246afa | 2710 | blocksize = fs_info->nodesize; |
e02119d5 | 2711 | |
2ff7e61e | 2712 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
2713 | if (IS_ERR(next)) |
2714 | return PTR_ERR(next); | |
e02119d5 | 2715 | |
e02119d5 | 2716 | if (*level == 1) { |
581c1760 QW |
2717 | ret = wc->process_func(root, next, wc, ptr_gen, |
2718 | *level - 1); | |
b50c6e25 JB |
2719 | if (ret) { |
2720 | free_extent_buffer(next); | |
1e5063d0 | 2721 | return ret; |
b50c6e25 | 2722 | } |
4a500fd1 | 2723 | |
e02119d5 CM |
2724 | path->slots[*level]++; |
2725 | if (wc->free) { | |
581c1760 QW |
2726 | ret = btrfs_read_buffer(next, ptr_gen, |
2727 | *level - 1, &first_key); | |
018642a1 TI |
2728 | if (ret) { |
2729 | free_extent_buffer(next); | |
2730 | return ret; | |
2731 | } | |
e02119d5 | 2732 | |
681ae509 JB |
2733 | if (trans) { |
2734 | btrfs_tree_lock(next); | |
8bead258 | 2735 | btrfs_set_lock_blocking_write(next); |
6a884d7d | 2736 | btrfs_clean_tree_block(next); |
681ae509 JB |
2737 | btrfs_wait_tree_block_writeback(next); |
2738 | btrfs_tree_unlock(next); | |
7bfc1007 | 2739 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2740 | bytenr, blocksize); |
2741 | if (ret) { | |
2742 | free_extent_buffer(next); | |
2743 | return ret; | |
2744 | } | |
1846430c LB |
2745 | } else { |
2746 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2747 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2748 | unaccount_log_buffer(fs_info, bytenr); |
3650860b | 2749 | } |
e02119d5 CM |
2750 | } |
2751 | free_extent_buffer(next); | |
2752 | continue; | |
2753 | } | |
581c1760 | 2754 | ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key); |
018642a1 TI |
2755 | if (ret) { |
2756 | free_extent_buffer(next); | |
2757 | return ret; | |
2758 | } | |
e02119d5 | 2759 | |
e02119d5 CM |
2760 | if (path->nodes[*level-1]) |
2761 | free_extent_buffer(path->nodes[*level-1]); | |
2762 | path->nodes[*level-1] = next; | |
2763 | *level = btrfs_header_level(next); | |
2764 | path->slots[*level] = 0; | |
2765 | cond_resched(); | |
2766 | } | |
4a500fd1 | 2767 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2768 | |
2769 | cond_resched(); | |
2770 | return 0; | |
2771 | } | |
2772 | ||
d397712b | 2773 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2774 | struct btrfs_root *root, |
2775 | struct btrfs_path *path, int *level, | |
2776 | struct walk_control *wc) | |
2777 | { | |
0b246afa | 2778 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
2779 | int i; |
2780 | int slot; | |
2781 | int ret; | |
2782 | ||
d397712b | 2783 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2784 | slot = path->slots[i]; |
4a500fd1 | 2785 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2786 | path->slots[i]++; |
2787 | *level = i; | |
2788 | WARN_ON(*level == 0); | |
2789 | return 0; | |
2790 | } else { | |
1e5063d0 | 2791 | ret = wc->process_func(root, path->nodes[*level], wc, |
581c1760 QW |
2792 | btrfs_header_generation(path->nodes[*level]), |
2793 | *level); | |
1e5063d0 MF |
2794 | if (ret) |
2795 | return ret; | |
2796 | ||
e02119d5 CM |
2797 | if (wc->free) { |
2798 | struct extent_buffer *next; | |
2799 | ||
2800 | next = path->nodes[*level]; | |
2801 | ||
681ae509 JB |
2802 | if (trans) { |
2803 | btrfs_tree_lock(next); | |
8bead258 | 2804 | btrfs_set_lock_blocking_write(next); |
6a884d7d | 2805 | btrfs_clean_tree_block(next); |
681ae509 JB |
2806 | btrfs_wait_tree_block_writeback(next); |
2807 | btrfs_tree_unlock(next); | |
7bfc1007 | 2808 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2809 | path->nodes[*level]->start, |
2810 | path->nodes[*level]->len); | |
2811 | if (ret) | |
2812 | return ret; | |
1846430c LB |
2813 | } else { |
2814 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2815 | clear_extent_buffer_dirty(next); | |
e02119d5 | 2816 | |
10e958d5 NB |
2817 | unaccount_log_buffer(fs_info, |
2818 | path->nodes[*level]->start); | |
2819 | } | |
e02119d5 CM |
2820 | } |
2821 | free_extent_buffer(path->nodes[*level]); | |
2822 | path->nodes[*level] = NULL; | |
2823 | *level = i + 1; | |
2824 | } | |
2825 | } | |
2826 | return 1; | |
2827 | } | |
2828 | ||
2829 | /* | |
2830 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2831 | * the tree freeing any blocks that have a ref count of zero after being | |
2832 | * decremented. | |
2833 | */ | |
2834 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2835 | struct btrfs_root *log, struct walk_control *wc) | |
2836 | { | |
2ff7e61e | 2837 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2838 | int ret = 0; |
2839 | int wret; | |
2840 | int level; | |
2841 | struct btrfs_path *path; | |
e02119d5 CM |
2842 | int orig_level; |
2843 | ||
2844 | path = btrfs_alloc_path(); | |
db5b493a TI |
2845 | if (!path) |
2846 | return -ENOMEM; | |
e02119d5 CM |
2847 | |
2848 | level = btrfs_header_level(log->node); | |
2849 | orig_level = level; | |
2850 | path->nodes[level] = log->node; | |
67439dad | 2851 | atomic_inc(&log->node->refs); |
e02119d5 CM |
2852 | path->slots[level] = 0; |
2853 | ||
d397712b | 2854 | while (1) { |
e02119d5 CM |
2855 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2856 | if (wret > 0) | |
2857 | break; | |
79787eaa | 2858 | if (wret < 0) { |
e02119d5 | 2859 | ret = wret; |
79787eaa JM |
2860 | goto out; |
2861 | } | |
e02119d5 CM |
2862 | |
2863 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2864 | if (wret > 0) | |
2865 | break; | |
79787eaa | 2866 | if (wret < 0) { |
e02119d5 | 2867 | ret = wret; |
79787eaa JM |
2868 | goto out; |
2869 | } | |
e02119d5 CM |
2870 | } |
2871 | ||
2872 | /* was the root node processed? if not, catch it here */ | |
2873 | if (path->nodes[orig_level]) { | |
79787eaa | 2874 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
581c1760 QW |
2875 | btrfs_header_generation(path->nodes[orig_level]), |
2876 | orig_level); | |
79787eaa JM |
2877 | if (ret) |
2878 | goto out; | |
e02119d5 CM |
2879 | if (wc->free) { |
2880 | struct extent_buffer *next; | |
2881 | ||
2882 | next = path->nodes[orig_level]; | |
2883 | ||
681ae509 JB |
2884 | if (trans) { |
2885 | btrfs_tree_lock(next); | |
8bead258 | 2886 | btrfs_set_lock_blocking_write(next); |
6a884d7d | 2887 | btrfs_clean_tree_block(next); |
681ae509 JB |
2888 | btrfs_wait_tree_block_writeback(next); |
2889 | btrfs_tree_unlock(next); | |
7bfc1007 | 2890 | ret = btrfs_pin_reserved_extent(trans, |
10e958d5 NB |
2891 | next->start, next->len); |
2892 | if (ret) | |
2893 | goto out; | |
1846430c LB |
2894 | } else { |
2895 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2896 | clear_extent_buffer_dirty(next); | |
10e958d5 | 2897 | unaccount_log_buffer(fs_info, next->start); |
681ae509 | 2898 | } |
e02119d5 CM |
2899 | } |
2900 | } | |
2901 | ||
79787eaa | 2902 | out: |
e02119d5 | 2903 | btrfs_free_path(path); |
e02119d5 CM |
2904 | return ret; |
2905 | } | |
2906 | ||
7237f183 YZ |
2907 | /* |
2908 | * helper function to update the item for a given subvolumes log root | |
2909 | * in the tree of log roots | |
2910 | */ | |
2911 | static int update_log_root(struct btrfs_trans_handle *trans, | |
4203e968 JB |
2912 | struct btrfs_root *log, |
2913 | struct btrfs_root_item *root_item) | |
7237f183 | 2914 | { |
0b246afa | 2915 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
2916 | int ret; |
2917 | ||
2918 | if (log->log_transid == 1) { | |
2919 | /* insert root item on the first sync */ | |
0b246afa | 2920 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
4203e968 | 2921 | &log->root_key, root_item); |
7237f183 | 2922 | } else { |
0b246afa | 2923 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
4203e968 | 2924 | &log->root_key, root_item); |
7237f183 YZ |
2925 | } |
2926 | return ret; | |
2927 | } | |
2928 | ||
60d53eb3 | 2929 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2930 | { |
2931 | DEFINE_WAIT(wait); | |
7237f183 | 2932 | int index = transid % 2; |
e02119d5 | 2933 | |
7237f183 YZ |
2934 | /* |
2935 | * we only allow two pending log transactions at a time, | |
2936 | * so we know that if ours is more than 2 older than the | |
2937 | * current transaction, we're done | |
2938 | */ | |
49e83f57 | 2939 | for (;;) { |
7237f183 YZ |
2940 | prepare_to_wait(&root->log_commit_wait[index], |
2941 | &wait, TASK_UNINTERRUPTIBLE); | |
12fcfd22 | 2942 | |
49e83f57 LB |
2943 | if (!(root->log_transid_committed < transid && |
2944 | atomic_read(&root->log_commit[index]))) | |
2945 | break; | |
12fcfd22 | 2946 | |
49e83f57 LB |
2947 | mutex_unlock(&root->log_mutex); |
2948 | schedule(); | |
7237f183 | 2949 | mutex_lock(&root->log_mutex); |
49e83f57 LB |
2950 | } |
2951 | finish_wait(&root->log_commit_wait[index], &wait); | |
7237f183 YZ |
2952 | } |
2953 | ||
60d53eb3 | 2954 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
2955 | { |
2956 | DEFINE_WAIT(wait); | |
8b050d35 | 2957 | |
49e83f57 LB |
2958 | for (;;) { |
2959 | prepare_to_wait(&root->log_writer_wait, &wait, | |
2960 | TASK_UNINTERRUPTIBLE); | |
2961 | if (!atomic_read(&root->log_writers)) | |
2962 | break; | |
2963 | ||
7237f183 | 2964 | mutex_unlock(&root->log_mutex); |
49e83f57 | 2965 | schedule(); |
575849ec | 2966 | mutex_lock(&root->log_mutex); |
7237f183 | 2967 | } |
49e83f57 | 2968 | finish_wait(&root->log_writer_wait, &wait); |
e02119d5 CM |
2969 | } |
2970 | ||
8b050d35 MX |
2971 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2972 | struct btrfs_log_ctx *ctx) | |
2973 | { | |
2974 | if (!ctx) | |
2975 | return; | |
2976 | ||
2977 | mutex_lock(&root->log_mutex); | |
2978 | list_del_init(&ctx->list); | |
2979 | mutex_unlock(&root->log_mutex); | |
2980 | } | |
2981 | ||
2982 | /* | |
2983 | * Invoked in log mutex context, or be sure there is no other task which | |
2984 | * can access the list. | |
2985 | */ | |
2986 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2987 | int index, int error) | |
2988 | { | |
2989 | struct btrfs_log_ctx *ctx; | |
570dd450 | 2990 | struct btrfs_log_ctx *safe; |
8b050d35 | 2991 | |
570dd450 CM |
2992 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
2993 | list_del_init(&ctx->list); | |
8b050d35 | 2994 | ctx->log_ret = error; |
570dd450 | 2995 | } |
8b050d35 MX |
2996 | |
2997 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2998 | } | |
2999 | ||
e02119d5 CM |
3000 | /* |
3001 | * btrfs_sync_log does sends a given tree log down to the disk and | |
3002 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
3003 | * you know that any inodes previously logged are safely on disk only |
3004 | * if it returns 0. | |
3005 | * | |
3006 | * Any other return value means you need to call btrfs_commit_transaction. | |
3007 | * Some of the edge cases for fsyncing directories that have had unlinks | |
3008 | * or renames done in the past mean that sometimes the only safe | |
3009 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
3010 | * that has happened. | |
e02119d5 CM |
3011 | */ |
3012 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 3013 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 3014 | { |
7237f183 YZ |
3015 | int index1; |
3016 | int index2; | |
8cef4e16 | 3017 | int mark; |
e02119d5 | 3018 | int ret; |
0b246afa | 3019 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 3020 | struct btrfs_root *log = root->log_root; |
0b246afa | 3021 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
4203e968 | 3022 | struct btrfs_root_item new_root_item; |
bb14a59b | 3023 | int log_transid = 0; |
8b050d35 | 3024 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 3025 | struct blk_plug plug; |
e02119d5 | 3026 | |
7237f183 | 3027 | mutex_lock(&root->log_mutex); |
d1433deb MX |
3028 | log_transid = ctx->log_transid; |
3029 | if (root->log_transid_committed >= log_transid) { | |
3030 | mutex_unlock(&root->log_mutex); | |
3031 | return ctx->log_ret; | |
3032 | } | |
3033 | ||
3034 | index1 = log_transid % 2; | |
7237f183 | 3035 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 3036 | wait_log_commit(root, log_transid); |
7237f183 | 3037 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3038 | return ctx->log_ret; |
e02119d5 | 3039 | } |
d1433deb | 3040 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
3041 | atomic_set(&root->log_commit[index1], 1); |
3042 | ||
3043 | /* wait for previous tree log sync to complete */ | |
3044 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 3045 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 3046 | |
86df7eb9 | 3047 | while (1) { |
2ecb7923 | 3048 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 3049 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 3050 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 3051 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
3052 | mutex_unlock(&root->log_mutex); |
3053 | schedule_timeout_uninterruptible(1); | |
3054 | mutex_lock(&root->log_mutex); | |
3055 | } | |
60d53eb3 | 3056 | wait_for_writer(root); |
2ecb7923 | 3057 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
3058 | break; |
3059 | } | |
e02119d5 | 3060 | |
12fcfd22 | 3061 | /* bail out if we need to do a full commit */ |
4884b8e8 | 3062 | if (btrfs_need_log_full_commit(trans)) { |
12fcfd22 CM |
3063 | ret = -EAGAIN; |
3064 | mutex_unlock(&root->log_mutex); | |
3065 | goto out; | |
3066 | } | |
3067 | ||
8cef4e16 YZ |
3068 | if (log_transid % 2 == 0) |
3069 | mark = EXTENT_DIRTY; | |
3070 | else | |
3071 | mark = EXTENT_NEW; | |
3072 | ||
690587d1 CM |
3073 | /* we start IO on all the marked extents here, but we don't actually |
3074 | * wait for them until later. | |
3075 | */ | |
c6adc9cc | 3076 | blk_start_plug(&plug); |
2ff7e61e | 3077 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
79787eaa | 3078 | if (ret) { |
c6adc9cc | 3079 | blk_finish_plug(&plug); |
66642832 | 3080 | btrfs_abort_transaction(trans, ret); |
90787766 | 3081 | btrfs_set_log_full_commit(trans); |
79787eaa JM |
3082 | mutex_unlock(&root->log_mutex); |
3083 | goto out; | |
3084 | } | |
7237f183 | 3085 | |
4203e968 JB |
3086 | /* |
3087 | * We _must_ update under the root->log_mutex in order to make sure we | |
3088 | * have a consistent view of the log root we are trying to commit at | |
3089 | * this moment. | |
3090 | * | |
3091 | * We _must_ copy this into a local copy, because we are not holding the | |
3092 | * log_root_tree->log_mutex yet. This is important because when we | |
3093 | * commit the log_root_tree we must have a consistent view of the | |
3094 | * log_root_tree when we update the super block to point at the | |
3095 | * log_root_tree bytenr. If we update the log_root_tree here we'll race | |
3096 | * with the commit and possibly point at the new block which we may not | |
3097 | * have written out. | |
3098 | */ | |
5d4f98a2 | 3099 | btrfs_set_root_node(&log->root_item, log->node); |
4203e968 | 3100 | memcpy(&new_root_item, &log->root_item, sizeof(new_root_item)); |
7237f183 | 3101 | |
7237f183 YZ |
3102 | root->log_transid++; |
3103 | log->log_transid = root->log_transid; | |
ff782e0a | 3104 | root->log_start_pid = 0; |
7237f183 | 3105 | /* |
8cef4e16 YZ |
3106 | * IO has been started, blocks of the log tree have WRITTEN flag set |
3107 | * in their headers. new modifications of the log will be written to | |
3108 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
3109 | */ |
3110 | mutex_unlock(&root->log_mutex); | |
3111 | ||
28a23593 | 3112 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
d1433deb | 3113 | |
7237f183 | 3114 | mutex_lock(&log_root_tree->log_mutex); |
2ecb7923 | 3115 | atomic_inc(&log_root_tree->log_batch); |
7237f183 | 3116 | atomic_inc(&log_root_tree->log_writers); |
d1433deb MX |
3117 | |
3118 | index2 = log_root_tree->log_transid % 2; | |
3119 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
3120 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
3121 | ||
7237f183 YZ |
3122 | mutex_unlock(&log_root_tree->log_mutex); |
3123 | ||
7237f183 | 3124 | mutex_lock(&log_root_tree->log_mutex); |
4203e968 JB |
3125 | |
3126 | /* | |
3127 | * Now we are safe to update the log_root_tree because we're under the | |
3128 | * log_mutex, and we're a current writer so we're holding the commit | |
3129 | * open until we drop the log_mutex. | |
3130 | */ | |
3131 | ret = update_log_root(trans, log, &new_root_item); | |
3132 | ||
7237f183 | 3133 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { |
093258e6 DS |
3134 | /* atomic_dec_and_test implies a barrier */ |
3135 | cond_wake_up_nomb(&log_root_tree->log_writer_wait); | |
7237f183 YZ |
3136 | } |
3137 | ||
4a500fd1 | 3138 | if (ret) { |
d1433deb MX |
3139 | if (!list_empty(&root_log_ctx.list)) |
3140 | list_del_init(&root_log_ctx.list); | |
3141 | ||
c6adc9cc | 3142 | blk_finish_plug(&plug); |
90787766 | 3143 | btrfs_set_log_full_commit(trans); |
995946dd | 3144 | |
79787eaa | 3145 | if (ret != -ENOSPC) { |
66642832 | 3146 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3147 | mutex_unlock(&log_root_tree->log_mutex); |
3148 | goto out; | |
3149 | } | |
bf89d38f | 3150 | btrfs_wait_tree_log_extents(log, mark); |
4a500fd1 YZ |
3151 | mutex_unlock(&log_root_tree->log_mutex); |
3152 | ret = -EAGAIN; | |
3153 | goto out; | |
3154 | } | |
3155 | ||
d1433deb | 3156 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 3157 | blk_finish_plug(&plug); |
cbd60aa7 | 3158 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
3159 | mutex_unlock(&log_root_tree->log_mutex); |
3160 | ret = root_log_ctx.log_ret; | |
3161 | goto out; | |
3162 | } | |
8b050d35 | 3163 | |
d1433deb | 3164 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 3165 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 3166 | blk_finish_plug(&plug); |
bf89d38f | 3167 | ret = btrfs_wait_tree_log_extents(log, mark); |
60d53eb3 | 3168 | wait_log_commit(log_root_tree, |
d1433deb | 3169 | root_log_ctx.log_transid); |
7237f183 | 3170 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
3171 | if (!ret) |
3172 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
3173 | goto out; |
3174 | } | |
d1433deb | 3175 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
3176 | atomic_set(&log_root_tree->log_commit[index2], 1); |
3177 | ||
12fcfd22 | 3178 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 3179 | wait_log_commit(log_root_tree, |
d1433deb | 3180 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
3181 | } |
3182 | ||
60d53eb3 | 3183 | wait_for_writer(log_root_tree); |
7237f183 | 3184 | |
12fcfd22 CM |
3185 | /* |
3186 | * now that we've moved on to the tree of log tree roots, | |
3187 | * check the full commit flag again | |
3188 | */ | |
4884b8e8 | 3189 | if (btrfs_need_log_full_commit(trans)) { |
c6adc9cc | 3190 | blk_finish_plug(&plug); |
bf89d38f | 3191 | btrfs_wait_tree_log_extents(log, mark); |
12fcfd22 CM |
3192 | mutex_unlock(&log_root_tree->log_mutex); |
3193 | ret = -EAGAIN; | |
3194 | goto out_wake_log_root; | |
3195 | } | |
7237f183 | 3196 | |
2ff7e61e | 3197 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
3198 | &log_root_tree->dirty_log_pages, |
3199 | EXTENT_DIRTY | EXTENT_NEW); | |
3200 | blk_finish_plug(&plug); | |
79787eaa | 3201 | if (ret) { |
90787766 | 3202 | btrfs_set_log_full_commit(trans); |
66642832 | 3203 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3204 | mutex_unlock(&log_root_tree->log_mutex); |
3205 | goto out_wake_log_root; | |
3206 | } | |
bf89d38f | 3207 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 3208 | if (!ret) |
bf89d38f JM |
3209 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
3210 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 3211 | if (ret) { |
90787766 | 3212 | btrfs_set_log_full_commit(trans); |
5ab5e44a FM |
3213 | mutex_unlock(&log_root_tree->log_mutex); |
3214 | goto out_wake_log_root; | |
3215 | } | |
e02119d5 | 3216 | |
0b246afa JM |
3217 | btrfs_set_super_log_root(fs_info->super_for_commit, |
3218 | log_root_tree->node->start); | |
3219 | btrfs_set_super_log_root_level(fs_info->super_for_commit, | |
3220 | btrfs_header_level(log_root_tree->node)); | |
e02119d5 | 3221 | |
7237f183 | 3222 | log_root_tree->log_transid++; |
7237f183 YZ |
3223 | mutex_unlock(&log_root_tree->log_mutex); |
3224 | ||
3225 | /* | |
52042d8e | 3226 | * Nobody else is going to jump in and write the ctree |
7237f183 YZ |
3227 | * super here because the log_commit atomic below is protecting |
3228 | * us. We must be called with a transaction handle pinning | |
3229 | * the running transaction open, so a full commit can't hop | |
3230 | * in and cause problems either. | |
3231 | */ | |
eece6a9c | 3232 | ret = write_all_supers(fs_info, 1); |
5af3e8cc | 3233 | if (ret) { |
90787766 | 3234 | btrfs_set_log_full_commit(trans); |
66642832 | 3235 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
3236 | goto out_wake_log_root; |
3237 | } | |
7237f183 | 3238 | |
257c62e1 CM |
3239 | mutex_lock(&root->log_mutex); |
3240 | if (root->last_log_commit < log_transid) | |
3241 | root->last_log_commit = log_transid; | |
3242 | mutex_unlock(&root->log_mutex); | |
3243 | ||
12fcfd22 | 3244 | out_wake_log_root: |
570dd450 | 3245 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
3246 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
3247 | ||
d1433deb | 3248 | log_root_tree->log_transid_committed++; |
7237f183 | 3249 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
3250 | mutex_unlock(&log_root_tree->log_mutex); |
3251 | ||
33a9eca7 | 3252 | /* |
093258e6 DS |
3253 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3254 | * all the updates above are seen by the woken threads. It might not be | |
3255 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3256 | */ |
093258e6 | 3257 | cond_wake_up(&log_root_tree->log_commit_wait[index2]); |
e02119d5 | 3258 | out: |
d1433deb | 3259 | mutex_lock(&root->log_mutex); |
570dd450 | 3260 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 3261 | root->log_transid_committed++; |
7237f183 | 3262 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 3263 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3264 | |
33a9eca7 | 3265 | /* |
093258e6 DS |
3266 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3267 | * all the updates above are seen by the woken threads. It might not be | |
3268 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3269 | */ |
093258e6 | 3270 | cond_wake_up(&root->log_commit_wait[index1]); |
b31eabd8 | 3271 | return ret; |
e02119d5 CM |
3272 | } |
3273 | ||
4a500fd1 YZ |
3274 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3275 | struct btrfs_root *log) | |
e02119d5 CM |
3276 | { |
3277 | int ret; | |
e02119d5 CM |
3278 | struct walk_control wc = { |
3279 | .free = 1, | |
3280 | .process_func = process_one_buffer | |
3281 | }; | |
3282 | ||
681ae509 | 3283 | ret = walk_log_tree(trans, log, &wc); |
374b0e2d JM |
3284 | if (ret) { |
3285 | if (trans) | |
3286 | btrfs_abort_transaction(trans, ret); | |
3287 | else | |
3288 | btrfs_handle_fs_error(log->fs_info, ret, NULL); | |
3289 | } | |
e02119d5 | 3290 | |
59b0713a FM |
3291 | clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1, |
3292 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT); | |
00246528 | 3293 | btrfs_put_root(log); |
4a500fd1 YZ |
3294 | } |
3295 | ||
3296 | /* | |
3297 | * free all the extents used by the tree log. This should be called | |
3298 | * at commit time of the full transaction | |
3299 | */ | |
3300 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3301 | { | |
3302 | if (root->log_root) { | |
3303 | free_log_tree(trans, root->log_root); | |
3304 | root->log_root = NULL; | |
3305 | } | |
3306 | return 0; | |
3307 | } | |
3308 | ||
3309 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3310 | struct btrfs_fs_info *fs_info) | |
3311 | { | |
3312 | if (fs_info->log_root_tree) { | |
3313 | free_log_tree(trans, fs_info->log_root_tree); | |
3314 | fs_info->log_root_tree = NULL; | |
3315 | } | |
e02119d5 CM |
3316 | return 0; |
3317 | } | |
3318 | ||
803f0f64 FM |
3319 | /* |
3320 | * Check if an inode was logged in the current transaction. We can't always rely | |
3321 | * on an inode's logged_trans value, because it's an in-memory only field and | |
3322 | * therefore not persisted. This means that its value is lost if the inode gets | |
3323 | * evicted and loaded again from disk (in which case it has a value of 0, and | |
3324 | * certainly it is smaller then any possible transaction ID), when that happens | |
3325 | * the full_sync flag is set in the inode's runtime flags, so on that case we | |
3326 | * assume eviction happened and ignore the logged_trans value, assuming the | |
3327 | * worst case, that the inode was logged before in the current transaction. | |
3328 | */ | |
3329 | static bool inode_logged(struct btrfs_trans_handle *trans, | |
3330 | struct btrfs_inode *inode) | |
3331 | { | |
3332 | if (inode->logged_trans == trans->transid) | |
3333 | return true; | |
3334 | ||
3335 | if (inode->last_trans == trans->transid && | |
3336 | test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) && | |
3337 | !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags)) | |
3338 | return true; | |
3339 | ||
3340 | return false; | |
3341 | } | |
3342 | ||
e02119d5 CM |
3343 | /* |
3344 | * If both a file and directory are logged, and unlinks or renames are | |
3345 | * mixed in, we have a few interesting corners: | |
3346 | * | |
3347 | * create file X in dir Y | |
3348 | * link file X to X.link in dir Y | |
3349 | * fsync file X | |
3350 | * unlink file X but leave X.link | |
3351 | * fsync dir Y | |
3352 | * | |
3353 | * After a crash we would expect only X.link to exist. But file X | |
3354 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3355 | * | |
3356 | * We solve this by removing directory entries and inode backrefs from the | |
3357 | * log when a file that was logged in the current transaction is | |
3358 | * unlinked. Any later fsync will include the updated log entries, and | |
3359 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3360 | * | |
3361 | * This optimizations allows us to avoid relogging the entire inode | |
3362 | * or the entire directory. | |
3363 | */ | |
3364 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3365 | struct btrfs_root *root, | |
3366 | const char *name, int name_len, | |
49f34d1f | 3367 | struct btrfs_inode *dir, u64 index) |
e02119d5 CM |
3368 | { |
3369 | struct btrfs_root *log; | |
3370 | struct btrfs_dir_item *di; | |
3371 | struct btrfs_path *path; | |
3372 | int ret; | |
4a500fd1 | 3373 | int err = 0; |
e02119d5 | 3374 | int bytes_del = 0; |
49f34d1f | 3375 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3376 | |
803f0f64 | 3377 | if (!inode_logged(trans, dir)) |
3a5f1d45 CM |
3378 | return 0; |
3379 | ||
e02119d5 CM |
3380 | ret = join_running_log_trans(root); |
3381 | if (ret) | |
3382 | return 0; | |
3383 | ||
49f34d1f | 3384 | mutex_lock(&dir->log_mutex); |
e02119d5 CM |
3385 | |
3386 | log = root->log_root; | |
3387 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3388 | if (!path) { |
3389 | err = -ENOMEM; | |
3390 | goto out_unlock; | |
3391 | } | |
2a29edc6 | 3392 | |
33345d01 | 3393 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3394 | name, name_len, -1); |
4a500fd1 YZ |
3395 | if (IS_ERR(di)) { |
3396 | err = PTR_ERR(di); | |
3397 | goto fail; | |
3398 | } | |
3399 | if (di) { | |
e02119d5 CM |
3400 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3401 | bytes_del += name_len; | |
3650860b JB |
3402 | if (ret) { |
3403 | err = ret; | |
3404 | goto fail; | |
3405 | } | |
e02119d5 | 3406 | } |
b3b4aa74 | 3407 | btrfs_release_path(path); |
33345d01 | 3408 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3409 | index, name, name_len, -1); |
4a500fd1 YZ |
3410 | if (IS_ERR(di)) { |
3411 | err = PTR_ERR(di); | |
3412 | goto fail; | |
3413 | } | |
3414 | if (di) { | |
e02119d5 CM |
3415 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3416 | bytes_del += name_len; | |
3650860b JB |
3417 | if (ret) { |
3418 | err = ret; | |
3419 | goto fail; | |
3420 | } | |
e02119d5 CM |
3421 | } |
3422 | ||
3423 | /* update the directory size in the log to reflect the names | |
3424 | * we have removed | |
3425 | */ | |
3426 | if (bytes_del) { | |
3427 | struct btrfs_key key; | |
3428 | ||
33345d01 | 3429 | key.objectid = dir_ino; |
e02119d5 CM |
3430 | key.offset = 0; |
3431 | key.type = BTRFS_INODE_ITEM_KEY; | |
b3b4aa74 | 3432 | btrfs_release_path(path); |
e02119d5 CM |
3433 | |
3434 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
4a500fd1 YZ |
3435 | if (ret < 0) { |
3436 | err = ret; | |
3437 | goto fail; | |
3438 | } | |
e02119d5 CM |
3439 | if (ret == 0) { |
3440 | struct btrfs_inode_item *item; | |
3441 | u64 i_size; | |
3442 | ||
3443 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3444 | struct btrfs_inode_item); | |
3445 | i_size = btrfs_inode_size(path->nodes[0], item); | |
3446 | if (i_size > bytes_del) | |
3447 | i_size -= bytes_del; | |
3448 | else | |
3449 | i_size = 0; | |
3450 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
3451 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3452 | } else | |
3453 | ret = 0; | |
b3b4aa74 | 3454 | btrfs_release_path(path); |
e02119d5 | 3455 | } |
4a500fd1 | 3456 | fail: |
e02119d5 | 3457 | btrfs_free_path(path); |
a62f44a5 | 3458 | out_unlock: |
49f34d1f | 3459 | mutex_unlock(&dir->log_mutex); |
4a500fd1 | 3460 | if (ret == -ENOSPC) { |
90787766 | 3461 | btrfs_set_log_full_commit(trans); |
4a500fd1 | 3462 | ret = 0; |
79787eaa | 3463 | } else if (ret < 0) |
66642832 | 3464 | btrfs_abort_transaction(trans, ret); |
79787eaa | 3465 | |
12fcfd22 | 3466 | btrfs_end_log_trans(root); |
e02119d5 | 3467 | |
411fc6bc | 3468 | return err; |
e02119d5 CM |
3469 | } |
3470 | ||
3471 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3472 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3473 | struct btrfs_root *root, | |
3474 | const char *name, int name_len, | |
a491abb2 | 3475 | struct btrfs_inode *inode, u64 dirid) |
e02119d5 CM |
3476 | { |
3477 | struct btrfs_root *log; | |
3478 | u64 index; | |
3479 | int ret; | |
3480 | ||
803f0f64 | 3481 | if (!inode_logged(trans, inode)) |
3a5f1d45 CM |
3482 | return 0; |
3483 | ||
e02119d5 CM |
3484 | ret = join_running_log_trans(root); |
3485 | if (ret) | |
3486 | return 0; | |
3487 | log = root->log_root; | |
a491abb2 | 3488 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3489 | |
a491abb2 | 3490 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 | 3491 | dirid, &index); |
a491abb2 | 3492 | mutex_unlock(&inode->log_mutex); |
4a500fd1 | 3493 | if (ret == -ENOSPC) { |
90787766 | 3494 | btrfs_set_log_full_commit(trans); |
4a500fd1 | 3495 | ret = 0; |
79787eaa | 3496 | } else if (ret < 0 && ret != -ENOENT) |
66642832 | 3497 | btrfs_abort_transaction(trans, ret); |
12fcfd22 | 3498 | btrfs_end_log_trans(root); |
e02119d5 | 3499 | |
e02119d5 CM |
3500 | return ret; |
3501 | } | |
3502 | ||
3503 | /* | |
3504 | * creates a range item in the log for 'dirid'. first_offset and | |
3505 | * last_offset tell us which parts of the key space the log should | |
3506 | * be considered authoritative for. | |
3507 | */ | |
3508 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3509 | struct btrfs_root *log, | |
3510 | struct btrfs_path *path, | |
3511 | int key_type, u64 dirid, | |
3512 | u64 first_offset, u64 last_offset) | |
3513 | { | |
3514 | int ret; | |
3515 | struct btrfs_key key; | |
3516 | struct btrfs_dir_log_item *item; | |
3517 | ||
3518 | key.objectid = dirid; | |
3519 | key.offset = first_offset; | |
3520 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3521 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3522 | else | |
3523 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3524 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3525 | if (ret) |
3526 | return ret; | |
e02119d5 CM |
3527 | |
3528 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3529 | struct btrfs_dir_log_item); | |
3530 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3531 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3532 | btrfs_release_path(path); |
e02119d5 CM |
3533 | return 0; |
3534 | } | |
3535 | ||
3536 | /* | |
3537 | * log all the items included in the current transaction for a given | |
3538 | * directory. This also creates the range items in the log tree required | |
3539 | * to replay anything deleted before the fsync | |
3540 | */ | |
3541 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
684a5773 | 3542 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 CM |
3543 | struct btrfs_path *path, |
3544 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3545 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3546 | u64 min_offset, u64 *last_offset_ret) |
3547 | { | |
3548 | struct btrfs_key min_key; | |
e02119d5 CM |
3549 | struct btrfs_root *log = root->log_root; |
3550 | struct extent_buffer *src; | |
4a500fd1 | 3551 | int err = 0; |
e02119d5 CM |
3552 | int ret; |
3553 | int i; | |
3554 | int nritems; | |
3555 | u64 first_offset = min_offset; | |
3556 | u64 last_offset = (u64)-1; | |
684a5773 | 3557 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
3558 | |
3559 | log = root->log_root; | |
e02119d5 | 3560 | |
33345d01 | 3561 | min_key.objectid = ino; |
e02119d5 CM |
3562 | min_key.type = key_type; |
3563 | min_key.offset = min_offset; | |
3564 | ||
6174d3cb | 3565 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3566 | |
3567 | /* | |
3568 | * we didn't find anything from this transaction, see if there | |
3569 | * is anything at all | |
3570 | */ | |
33345d01 LZ |
3571 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3572 | min_key.objectid = ino; | |
e02119d5 CM |
3573 | min_key.type = key_type; |
3574 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3575 | btrfs_release_path(path); |
e02119d5 CM |
3576 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3577 | if (ret < 0) { | |
b3b4aa74 | 3578 | btrfs_release_path(path); |
e02119d5 CM |
3579 | return ret; |
3580 | } | |
33345d01 | 3581 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3582 | |
3583 | /* if ret == 0 there are items for this type, | |
3584 | * create a range to tell us the last key of this type. | |
3585 | * otherwise, there are no items in this directory after | |
3586 | * *min_offset, and we create a range to indicate that. | |
3587 | */ | |
3588 | if (ret == 0) { | |
3589 | struct btrfs_key tmp; | |
3590 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3591 | path->slots[0]); | |
d397712b | 3592 | if (key_type == tmp.type) |
e02119d5 | 3593 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3594 | } |
3595 | goto done; | |
3596 | } | |
3597 | ||
3598 | /* go backward to find any previous key */ | |
33345d01 | 3599 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3600 | if (ret == 0) { |
3601 | struct btrfs_key tmp; | |
3602 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3603 | if (key_type == tmp.type) { | |
3604 | first_offset = tmp.offset; | |
3605 | ret = overwrite_item(trans, log, dst_path, | |
3606 | path->nodes[0], path->slots[0], | |
3607 | &tmp); | |
4a500fd1 YZ |
3608 | if (ret) { |
3609 | err = ret; | |
3610 | goto done; | |
3611 | } | |
e02119d5 CM |
3612 | } |
3613 | } | |
b3b4aa74 | 3614 | btrfs_release_path(path); |
e02119d5 | 3615 | |
2cc83342 JB |
3616 | /* |
3617 | * Find the first key from this transaction again. See the note for | |
3618 | * log_new_dir_dentries, if we're logging a directory recursively we | |
3619 | * won't be holding its i_mutex, which means we can modify the directory | |
3620 | * while we're logging it. If we remove an entry between our first | |
3621 | * search and this search we'll not find the key again and can just | |
3622 | * bail. | |
3623 | */ | |
e02119d5 | 3624 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
2cc83342 | 3625 | if (ret != 0) |
e02119d5 | 3626 | goto done; |
e02119d5 CM |
3627 | |
3628 | /* | |
3629 | * we have a block from this transaction, log every item in it | |
3630 | * from our directory | |
3631 | */ | |
d397712b | 3632 | while (1) { |
e02119d5 CM |
3633 | struct btrfs_key tmp; |
3634 | src = path->nodes[0]; | |
3635 | nritems = btrfs_header_nritems(src); | |
3636 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3637 | struct btrfs_dir_item *di; |
3638 | ||
e02119d5 CM |
3639 | btrfs_item_key_to_cpu(src, &min_key, i); |
3640 | ||
33345d01 | 3641 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 CM |
3642 | goto done; |
3643 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3644 | &min_key); | |
4a500fd1 YZ |
3645 | if (ret) { |
3646 | err = ret; | |
3647 | goto done; | |
3648 | } | |
2f2ff0ee FM |
3649 | |
3650 | /* | |
3651 | * We must make sure that when we log a directory entry, | |
3652 | * the corresponding inode, after log replay, has a | |
3653 | * matching link count. For example: | |
3654 | * | |
3655 | * touch foo | |
3656 | * mkdir mydir | |
3657 | * sync | |
3658 | * ln foo mydir/bar | |
3659 | * xfs_io -c "fsync" mydir | |
3660 | * <crash> | |
3661 | * <mount fs and log replay> | |
3662 | * | |
3663 | * Would result in a fsync log that when replayed, our | |
3664 | * file inode would have a link count of 1, but we get | |
3665 | * two directory entries pointing to the same inode. | |
3666 | * After removing one of the names, it would not be | |
3667 | * possible to remove the other name, which resulted | |
3668 | * always in stale file handle errors, and would not | |
3669 | * be possible to rmdir the parent directory, since | |
3670 | * its i_size could never decrement to the value | |
3671 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3672 | */ | |
3673 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3674 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3675 | if (ctx && | |
3676 | (btrfs_dir_transid(src, di) == trans->transid || | |
3677 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3678 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3679 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3680 | } |
3681 | path->slots[0] = nritems; | |
3682 | ||
3683 | /* | |
3684 | * look ahead to the next item and see if it is also | |
3685 | * from this directory and from this transaction | |
3686 | */ | |
3687 | ret = btrfs_next_leaf(root, path); | |
80c0b421 LB |
3688 | if (ret) { |
3689 | if (ret == 1) | |
3690 | last_offset = (u64)-1; | |
3691 | else | |
3692 | err = ret; | |
e02119d5 CM |
3693 | goto done; |
3694 | } | |
3695 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3696 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3697 | last_offset = (u64)-1; |
3698 | goto done; | |
3699 | } | |
3700 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3701 | ret = overwrite_item(trans, log, dst_path, | |
3702 | path->nodes[0], path->slots[0], | |
3703 | &tmp); | |
4a500fd1 YZ |
3704 | if (ret) |
3705 | err = ret; | |
3706 | else | |
3707 | last_offset = tmp.offset; | |
e02119d5 CM |
3708 | goto done; |
3709 | } | |
3710 | } | |
3711 | done: | |
b3b4aa74 DS |
3712 | btrfs_release_path(path); |
3713 | btrfs_release_path(dst_path); | |
e02119d5 | 3714 | |
4a500fd1 YZ |
3715 | if (err == 0) { |
3716 | *last_offset_ret = last_offset; | |
3717 | /* | |
3718 | * insert the log range keys to indicate where the log | |
3719 | * is valid | |
3720 | */ | |
3721 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3722 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3723 | if (ret) |
3724 | err = ret; | |
3725 | } | |
3726 | return err; | |
e02119d5 CM |
3727 | } |
3728 | ||
3729 | /* | |
3730 | * logging directories is very similar to logging inodes, We find all the items | |
3731 | * from the current transaction and write them to the log. | |
3732 | * | |
3733 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3734 | * key in the range logged that is not present in the log tree, then it means | |
3735 | * that dir entry was unlinked during the transaction. | |
3736 | * | |
3737 | * In order for that scan to work, we must include one key smaller than | |
3738 | * the smallest logged by this transaction and one key larger than the largest | |
3739 | * key logged by this transaction. | |
3740 | */ | |
3741 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
dbf39ea4 | 3742 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 | 3743 | struct btrfs_path *path, |
2f2ff0ee FM |
3744 | struct btrfs_path *dst_path, |
3745 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3746 | { |
3747 | u64 min_key; | |
3748 | u64 max_key; | |
3749 | int ret; | |
3750 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3751 | ||
3752 | again: | |
3753 | min_key = 0; | |
3754 | max_key = 0; | |
d397712b | 3755 | while (1) { |
dbf39ea4 NB |
3756 | ret = log_dir_items(trans, root, inode, path, dst_path, key_type, |
3757 | ctx, min_key, &max_key); | |
4a500fd1 YZ |
3758 | if (ret) |
3759 | return ret; | |
e02119d5 CM |
3760 | if (max_key == (u64)-1) |
3761 | break; | |
3762 | min_key = max_key + 1; | |
3763 | } | |
3764 | ||
3765 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3766 | key_type = BTRFS_DIR_INDEX_KEY; | |
3767 | goto again; | |
3768 | } | |
3769 | return 0; | |
3770 | } | |
3771 | ||
3772 | /* | |
3773 | * a helper function to drop items from the log before we relog an | |
3774 | * inode. max_key_type indicates the highest item type to remove. | |
3775 | * This cannot be run for file data extents because it does not | |
3776 | * free the extents they point to. | |
3777 | */ | |
3778 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3779 | struct btrfs_root *log, | |
3780 | struct btrfs_path *path, | |
3781 | u64 objectid, int max_key_type) | |
3782 | { | |
3783 | int ret; | |
3784 | struct btrfs_key key; | |
3785 | struct btrfs_key found_key; | |
18ec90d6 | 3786 | int start_slot; |
e02119d5 CM |
3787 | |
3788 | key.objectid = objectid; | |
3789 | key.type = max_key_type; | |
3790 | key.offset = (u64)-1; | |
3791 | ||
d397712b | 3792 | while (1) { |
e02119d5 | 3793 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3794 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3795 | if (ret < 0) |
e02119d5 CM |
3796 | break; |
3797 | ||
3798 | if (path->slots[0] == 0) | |
3799 | break; | |
3800 | ||
3801 | path->slots[0]--; | |
3802 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3803 | path->slots[0]); | |
3804 | ||
3805 | if (found_key.objectid != objectid) | |
3806 | break; | |
3807 | ||
18ec90d6 JB |
3808 | found_key.offset = 0; |
3809 | found_key.type = 0; | |
e3b83361 | 3810 | ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot); |
cbca7d59 FM |
3811 | if (ret < 0) |
3812 | break; | |
18ec90d6 JB |
3813 | |
3814 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3815 | path->slots[0] - start_slot + 1); | |
3816 | /* | |
3817 | * If start slot isn't 0 then we don't need to re-search, we've | |
3818 | * found the last guy with the objectid in this tree. | |
3819 | */ | |
3820 | if (ret || start_slot != 0) | |
65a246c5 | 3821 | break; |
b3b4aa74 | 3822 | btrfs_release_path(path); |
e02119d5 | 3823 | } |
b3b4aa74 | 3824 | btrfs_release_path(path); |
5bdbeb21 JB |
3825 | if (ret > 0) |
3826 | ret = 0; | |
4a500fd1 | 3827 | return ret; |
e02119d5 CM |
3828 | } |
3829 | ||
94edf4ae JB |
3830 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3831 | struct extent_buffer *leaf, | |
3832 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3833 | struct inode *inode, int log_inode_only, |
3834 | u64 logged_isize) | |
94edf4ae | 3835 | { |
0b1c6cca JB |
3836 | struct btrfs_map_token token; |
3837 | ||
c82f823c | 3838 | btrfs_init_map_token(&token, leaf); |
94edf4ae JB |
3839 | |
3840 | if (log_inode_only) { | |
3841 | /* set the generation to zero so the recover code | |
3842 | * can tell the difference between an logging | |
3843 | * just to say 'this inode exists' and a logging | |
3844 | * to say 'update this inode with these values' | |
3845 | */ | |
cc4c13d5 DS |
3846 | btrfs_set_token_inode_generation(&token, item, 0); |
3847 | btrfs_set_token_inode_size(&token, item, logged_isize); | |
94edf4ae | 3848 | } else { |
cc4c13d5 DS |
3849 | btrfs_set_token_inode_generation(&token, item, |
3850 | BTRFS_I(inode)->generation); | |
3851 | btrfs_set_token_inode_size(&token, item, inode->i_size); | |
0b1c6cca JB |
3852 | } |
3853 | ||
cc4c13d5 DS |
3854 | btrfs_set_token_inode_uid(&token, item, i_uid_read(inode)); |
3855 | btrfs_set_token_inode_gid(&token, item, i_gid_read(inode)); | |
3856 | btrfs_set_token_inode_mode(&token, item, inode->i_mode); | |
3857 | btrfs_set_token_inode_nlink(&token, item, inode->i_nlink); | |
3858 | ||
3859 | btrfs_set_token_timespec_sec(&token, &item->atime, | |
3860 | inode->i_atime.tv_sec); | |
3861 | btrfs_set_token_timespec_nsec(&token, &item->atime, | |
3862 | inode->i_atime.tv_nsec); | |
3863 | ||
3864 | btrfs_set_token_timespec_sec(&token, &item->mtime, | |
3865 | inode->i_mtime.tv_sec); | |
3866 | btrfs_set_token_timespec_nsec(&token, &item->mtime, | |
3867 | inode->i_mtime.tv_nsec); | |
3868 | ||
3869 | btrfs_set_token_timespec_sec(&token, &item->ctime, | |
3870 | inode->i_ctime.tv_sec); | |
3871 | btrfs_set_token_timespec_nsec(&token, &item->ctime, | |
3872 | inode->i_ctime.tv_nsec); | |
3873 | ||
3874 | btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode)); | |
3875 | ||
3876 | btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode)); | |
3877 | btrfs_set_token_inode_transid(&token, item, trans->transid); | |
3878 | btrfs_set_token_inode_rdev(&token, item, inode->i_rdev); | |
3879 | btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags); | |
3880 | btrfs_set_token_inode_block_group(&token, item, 0); | |
94edf4ae JB |
3881 | } |
3882 | ||
a95249b3 JB |
3883 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3884 | struct btrfs_root *log, struct btrfs_path *path, | |
6d889a3b | 3885 | struct btrfs_inode *inode) |
a95249b3 JB |
3886 | { |
3887 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3888 | int ret; |
3889 | ||
efd0c405 | 3890 | ret = btrfs_insert_empty_item(trans, log, path, |
6d889a3b | 3891 | &inode->location, sizeof(*inode_item)); |
a95249b3 JB |
3892 | if (ret && ret != -EEXIST) |
3893 | return ret; | |
3894 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3895 | struct btrfs_inode_item); | |
6d889a3b NB |
3896 | fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode, |
3897 | 0, 0); | |
a95249b3 JB |
3898 | btrfs_release_path(path); |
3899 | return 0; | |
3900 | } | |
3901 | ||
40e046ac FM |
3902 | static int log_csums(struct btrfs_trans_handle *trans, |
3903 | struct btrfs_root *log_root, | |
3904 | struct btrfs_ordered_sum *sums) | |
3905 | { | |
3906 | int ret; | |
3907 | ||
3908 | /* | |
3909 | * Due to extent cloning, we might have logged a csum item that covers a | |
3910 | * subrange of a cloned extent, and later we can end up logging a csum | |
3911 | * item for a larger subrange of the same extent or the entire range. | |
3912 | * This would leave csum items in the log tree that cover the same range | |
3913 | * and break the searches for checksums in the log tree, resulting in | |
3914 | * some checksums missing in the fs/subvolume tree. So just delete (or | |
3915 | * trim and adjust) any existing csum items in the log for this range. | |
3916 | */ | |
3917 | ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len); | |
3918 | if (ret) | |
3919 | return ret; | |
3920 | ||
3921 | return btrfs_csum_file_blocks(trans, log_root, sums); | |
3922 | } | |
3923 | ||
31ff1cd2 | 3924 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 3925 | struct btrfs_inode *inode, |
31ff1cd2 | 3926 | struct btrfs_path *dst_path, |
0e56315c | 3927 | struct btrfs_path *src_path, |
1a4bcf47 FM |
3928 | int start_slot, int nr, int inode_only, |
3929 | u64 logged_isize) | |
31ff1cd2 | 3930 | { |
3ffbd68c | 3931 | struct btrfs_fs_info *fs_info = trans->fs_info; |
31ff1cd2 CM |
3932 | unsigned long src_offset; |
3933 | unsigned long dst_offset; | |
44d70e19 | 3934 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 CM |
3935 | struct btrfs_file_extent_item *extent; |
3936 | struct btrfs_inode_item *inode_item; | |
16e7549f | 3937 | struct extent_buffer *src = src_path->nodes[0]; |
31ff1cd2 CM |
3938 | int ret; |
3939 | struct btrfs_key *ins_keys; | |
3940 | u32 *ins_sizes; | |
3941 | char *ins_data; | |
3942 | int i; | |
d20f7043 | 3943 | struct list_head ordered_sums; |
44d70e19 | 3944 | int skip_csum = inode->flags & BTRFS_INODE_NODATASUM; |
d20f7043 CM |
3945 | |
3946 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3947 | |
3948 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3949 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3950 | if (!ins_data) |
3951 | return -ENOMEM; | |
3952 | ||
31ff1cd2 CM |
3953 | ins_sizes = (u32 *)ins_data; |
3954 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3955 | ||
3956 | for (i = 0; i < nr; i++) { | |
3957 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3958 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3959 | } | |
3960 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3961 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3962 | if (ret) { |
3963 | kfree(ins_data); | |
3964 | return ret; | |
3965 | } | |
31ff1cd2 | 3966 | |
5d4f98a2 | 3967 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3968 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3969 | dst_path->slots[0]); | |
3970 | ||
3971 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3972 | ||
94edf4ae | 3973 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3974 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3975 | dst_path->slots[0], | |
3976 | struct btrfs_inode_item); | |
94edf4ae | 3977 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
f85b7379 DS |
3978 | &inode->vfs_inode, |
3979 | inode_only == LOG_INODE_EXISTS, | |
1a4bcf47 | 3980 | logged_isize); |
94edf4ae JB |
3981 | } else { |
3982 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3983 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 3984 | } |
94edf4ae | 3985 | |
31ff1cd2 CM |
3986 | /* take a reference on file data extents so that truncates |
3987 | * or deletes of this inode don't have to relog the inode | |
3988 | * again | |
3989 | */ | |
962a298f | 3990 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 3991 | !skip_csum) { |
31ff1cd2 CM |
3992 | int found_type; |
3993 | extent = btrfs_item_ptr(src, start_slot + i, | |
3994 | struct btrfs_file_extent_item); | |
3995 | ||
8e531cdf | 3996 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
3997 | continue; | |
3998 | ||
31ff1cd2 | 3999 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 4000 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
4001 | u64 ds, dl, cs, cl; |
4002 | ds = btrfs_file_extent_disk_bytenr(src, | |
4003 | extent); | |
4004 | /* ds == 0 is a hole */ | |
4005 | if (ds == 0) | |
4006 | continue; | |
4007 | ||
4008 | dl = btrfs_file_extent_disk_num_bytes(src, | |
4009 | extent); | |
4010 | cs = btrfs_file_extent_offset(src, extent); | |
4011 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 4012 | extent); |
580afd76 CM |
4013 | if (btrfs_file_extent_compression(src, |
4014 | extent)) { | |
4015 | cs = 0; | |
4016 | cl = dl; | |
4017 | } | |
5d4f98a2 YZ |
4018 | |
4019 | ret = btrfs_lookup_csums_range( | |
0b246afa | 4020 | fs_info->csum_root, |
5d4f98a2 | 4021 | ds + cs, ds + cs + cl - 1, |
a2de733c | 4022 | &ordered_sums, 0); |
3650860b JB |
4023 | if (ret) { |
4024 | btrfs_release_path(dst_path); | |
4025 | kfree(ins_data); | |
4026 | return ret; | |
4027 | } | |
31ff1cd2 CM |
4028 | } |
4029 | } | |
31ff1cd2 CM |
4030 | } |
4031 | ||
4032 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 4033 | btrfs_release_path(dst_path); |
31ff1cd2 | 4034 | kfree(ins_data); |
d20f7043 CM |
4035 | |
4036 | /* | |
4037 | * we have to do this after the loop above to avoid changing the | |
4038 | * log tree while trying to change the log tree. | |
4039 | */ | |
4a500fd1 | 4040 | ret = 0; |
d397712b | 4041 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
4042 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
4043 | struct btrfs_ordered_sum, | |
4044 | list); | |
4a500fd1 | 4045 | if (!ret) |
40e046ac | 4046 | ret = log_csums(trans, log, sums); |
d20f7043 CM |
4047 | list_del(&sums->list); |
4048 | kfree(sums); | |
4049 | } | |
16e7549f | 4050 | |
4a500fd1 | 4051 | return ret; |
31ff1cd2 CM |
4052 | } |
4053 | ||
5dc562c5 JB |
4054 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
4055 | { | |
4056 | struct extent_map *em1, *em2; | |
4057 | ||
4058 | em1 = list_entry(a, struct extent_map, list); | |
4059 | em2 = list_entry(b, struct extent_map, list); | |
4060 | ||
4061 | if (em1->start < em2->start) | |
4062 | return -1; | |
4063 | else if (em1->start > em2->start) | |
4064 | return 1; | |
4065 | return 0; | |
4066 | } | |
4067 | ||
e7175a69 JB |
4068 | static int log_extent_csums(struct btrfs_trans_handle *trans, |
4069 | struct btrfs_inode *inode, | |
a9ecb653 | 4070 | struct btrfs_root *log_root, |
e7175a69 | 4071 | const struct extent_map *em) |
5dc562c5 | 4072 | { |
2ab28f32 JB |
4073 | u64 csum_offset; |
4074 | u64 csum_len; | |
8407f553 FM |
4075 | LIST_HEAD(ordered_sums); |
4076 | int ret = 0; | |
0aa4a17d | 4077 | |
e7175a69 JB |
4078 | if (inode->flags & BTRFS_INODE_NODATASUM || |
4079 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
8407f553 | 4080 | em->block_start == EXTENT_MAP_HOLE) |
70c8a91c | 4081 | return 0; |
5dc562c5 | 4082 | |
e7175a69 | 4083 | /* If we're compressed we have to save the entire range of csums. */ |
488111aa FDBM |
4084 | if (em->compress_type) { |
4085 | csum_offset = 0; | |
8407f553 | 4086 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa | 4087 | } else { |
e7175a69 JB |
4088 | csum_offset = em->mod_start - em->start; |
4089 | csum_len = em->mod_len; | |
488111aa | 4090 | } |
2ab28f32 | 4091 | |
70c8a91c | 4092 | /* block start is already adjusted for the file extent offset. */ |
a9ecb653 | 4093 | ret = btrfs_lookup_csums_range(trans->fs_info->csum_root, |
70c8a91c JB |
4094 | em->block_start + csum_offset, |
4095 | em->block_start + csum_offset + | |
4096 | csum_len - 1, &ordered_sums, 0); | |
4097 | if (ret) | |
4098 | return ret; | |
5dc562c5 | 4099 | |
70c8a91c JB |
4100 | while (!list_empty(&ordered_sums)) { |
4101 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4102 | struct btrfs_ordered_sum, | |
4103 | list); | |
4104 | if (!ret) | |
40e046ac | 4105 | ret = log_csums(trans, log_root, sums); |
70c8a91c JB |
4106 | list_del(&sums->list); |
4107 | kfree(sums); | |
5dc562c5 JB |
4108 | } |
4109 | ||
70c8a91c | 4110 | return ret; |
5dc562c5 JB |
4111 | } |
4112 | ||
8407f553 | 4113 | static int log_one_extent(struct btrfs_trans_handle *trans, |
9d122629 | 4114 | struct btrfs_inode *inode, struct btrfs_root *root, |
8407f553 FM |
4115 | const struct extent_map *em, |
4116 | struct btrfs_path *path, | |
8407f553 FM |
4117 | struct btrfs_log_ctx *ctx) |
4118 | { | |
4119 | struct btrfs_root *log = root->log_root; | |
4120 | struct btrfs_file_extent_item *fi; | |
4121 | struct extent_buffer *leaf; | |
4122 | struct btrfs_map_token token; | |
4123 | struct btrfs_key key; | |
4124 | u64 extent_offset = em->start - em->orig_start; | |
4125 | u64 block_len; | |
4126 | int ret; | |
4127 | int extent_inserted = 0; | |
8407f553 | 4128 | |
a9ecb653 | 4129 | ret = log_extent_csums(trans, inode, log, em); |
8407f553 FM |
4130 | if (ret) |
4131 | return ret; | |
4132 | ||
9d122629 | 4133 | ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start, |
8407f553 FM |
4134 | em->start + em->len, NULL, 0, 1, |
4135 | sizeof(*fi), &extent_inserted); | |
4136 | if (ret) | |
4137 | return ret; | |
4138 | ||
4139 | if (!extent_inserted) { | |
9d122629 | 4140 | key.objectid = btrfs_ino(inode); |
8407f553 FM |
4141 | key.type = BTRFS_EXTENT_DATA_KEY; |
4142 | key.offset = em->start; | |
4143 | ||
4144 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4145 | sizeof(*fi)); | |
4146 | if (ret) | |
4147 | return ret; | |
4148 | } | |
4149 | leaf = path->nodes[0]; | |
c82f823c | 4150 | btrfs_init_map_token(&token, leaf); |
8407f553 FM |
4151 | fi = btrfs_item_ptr(leaf, path->slots[0], |
4152 | struct btrfs_file_extent_item); | |
4153 | ||
cc4c13d5 | 4154 | btrfs_set_token_file_extent_generation(&token, fi, trans->transid); |
8407f553 | 4155 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
cc4c13d5 DS |
4156 | btrfs_set_token_file_extent_type(&token, fi, |
4157 | BTRFS_FILE_EXTENT_PREALLOC); | |
8407f553 | 4158 | else |
cc4c13d5 DS |
4159 | btrfs_set_token_file_extent_type(&token, fi, |
4160 | BTRFS_FILE_EXTENT_REG); | |
8407f553 FM |
4161 | |
4162 | block_len = max(em->block_len, em->orig_block_len); | |
4163 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
cc4c13d5 DS |
4164 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, |
4165 | em->block_start); | |
4166 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len); | |
8407f553 | 4167 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { |
cc4c13d5 | 4168 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, |
8407f553 | 4169 | em->block_start - |
cc4c13d5 DS |
4170 | extent_offset); |
4171 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len); | |
8407f553 | 4172 | } else { |
cc4c13d5 DS |
4173 | btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0); |
4174 | btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0); | |
8407f553 FM |
4175 | } |
4176 | ||
cc4c13d5 DS |
4177 | btrfs_set_token_file_extent_offset(&token, fi, extent_offset); |
4178 | btrfs_set_token_file_extent_num_bytes(&token, fi, em->len); | |
4179 | btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes); | |
4180 | btrfs_set_token_file_extent_compression(&token, fi, em->compress_type); | |
4181 | btrfs_set_token_file_extent_encryption(&token, fi, 0); | |
4182 | btrfs_set_token_file_extent_other_encoding(&token, fi, 0); | |
8407f553 FM |
4183 | btrfs_mark_buffer_dirty(leaf); |
4184 | ||
4185 | btrfs_release_path(path); | |
4186 | ||
4187 | return ret; | |
4188 | } | |
4189 | ||
31d11b83 FM |
4190 | /* |
4191 | * Log all prealloc extents beyond the inode's i_size to make sure we do not | |
4192 | * lose them after doing a fast fsync and replaying the log. We scan the | |
4193 | * subvolume's root instead of iterating the inode's extent map tree because | |
4194 | * otherwise we can log incorrect extent items based on extent map conversion. | |
4195 | * That can happen due to the fact that extent maps are merged when they | |
4196 | * are not in the extent map tree's list of modified extents. | |
4197 | */ | |
4198 | static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans, | |
4199 | struct btrfs_inode *inode, | |
4200 | struct btrfs_path *path) | |
4201 | { | |
4202 | struct btrfs_root *root = inode->root; | |
4203 | struct btrfs_key key; | |
4204 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
4205 | const u64 ino = btrfs_ino(inode); | |
4206 | struct btrfs_path *dst_path = NULL; | |
0e56315c | 4207 | bool dropped_extents = false; |
f135cea3 FM |
4208 | u64 truncate_offset = i_size; |
4209 | struct extent_buffer *leaf; | |
4210 | int slot; | |
31d11b83 FM |
4211 | int ins_nr = 0; |
4212 | int start_slot; | |
4213 | int ret; | |
4214 | ||
4215 | if (!(inode->flags & BTRFS_INODE_PREALLOC)) | |
4216 | return 0; | |
4217 | ||
4218 | key.objectid = ino; | |
4219 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4220 | key.offset = i_size; | |
4221 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4222 | if (ret < 0) | |
4223 | goto out; | |
4224 | ||
f135cea3 FM |
4225 | /* |
4226 | * We must check if there is a prealloc extent that starts before the | |
4227 | * i_size and crosses the i_size boundary. This is to ensure later we | |
4228 | * truncate down to the end of that extent and not to the i_size, as | |
4229 | * otherwise we end up losing part of the prealloc extent after a log | |
4230 | * replay and with an implicit hole if there is another prealloc extent | |
4231 | * that starts at an offset beyond i_size. | |
4232 | */ | |
4233 | ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY); | |
4234 | if (ret < 0) | |
4235 | goto out; | |
4236 | ||
4237 | if (ret == 0) { | |
4238 | struct btrfs_file_extent_item *ei; | |
4239 | ||
4240 | leaf = path->nodes[0]; | |
4241 | slot = path->slots[0]; | |
4242 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4243 | ||
4244 | if (btrfs_file_extent_type(leaf, ei) == | |
4245 | BTRFS_FILE_EXTENT_PREALLOC) { | |
4246 | u64 extent_end; | |
4247 | ||
4248 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4249 | extent_end = key.offset + | |
4250 | btrfs_file_extent_num_bytes(leaf, ei); | |
4251 | ||
4252 | if (extent_end > i_size) | |
4253 | truncate_offset = extent_end; | |
4254 | } | |
4255 | } else { | |
4256 | ret = 0; | |
4257 | } | |
4258 | ||
31d11b83 | 4259 | while (true) { |
f135cea3 FM |
4260 | leaf = path->nodes[0]; |
4261 | slot = path->slots[0]; | |
31d11b83 FM |
4262 | |
4263 | if (slot >= btrfs_header_nritems(leaf)) { | |
4264 | if (ins_nr > 0) { | |
4265 | ret = copy_items(trans, inode, dst_path, path, | |
0e56315c | 4266 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4267 | if (ret < 0) |
4268 | goto out; | |
4269 | ins_nr = 0; | |
4270 | } | |
4271 | ret = btrfs_next_leaf(root, path); | |
4272 | if (ret < 0) | |
4273 | goto out; | |
4274 | if (ret > 0) { | |
4275 | ret = 0; | |
4276 | break; | |
4277 | } | |
4278 | continue; | |
4279 | } | |
4280 | ||
4281 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4282 | if (key.objectid > ino) | |
4283 | break; | |
4284 | if (WARN_ON_ONCE(key.objectid < ino) || | |
4285 | key.type < BTRFS_EXTENT_DATA_KEY || | |
4286 | key.offset < i_size) { | |
4287 | path->slots[0]++; | |
4288 | continue; | |
4289 | } | |
0e56315c | 4290 | if (!dropped_extents) { |
31d11b83 FM |
4291 | /* |
4292 | * Avoid logging extent items logged in past fsync calls | |
4293 | * and leading to duplicate keys in the log tree. | |
4294 | */ | |
4295 | do { | |
4296 | ret = btrfs_truncate_inode_items(trans, | |
4297 | root->log_root, | |
4298 | &inode->vfs_inode, | |
f135cea3 | 4299 | truncate_offset, |
31d11b83 FM |
4300 | BTRFS_EXTENT_DATA_KEY); |
4301 | } while (ret == -EAGAIN); | |
4302 | if (ret) | |
4303 | goto out; | |
0e56315c | 4304 | dropped_extents = true; |
31d11b83 FM |
4305 | } |
4306 | if (ins_nr == 0) | |
4307 | start_slot = slot; | |
4308 | ins_nr++; | |
4309 | path->slots[0]++; | |
4310 | if (!dst_path) { | |
4311 | dst_path = btrfs_alloc_path(); | |
4312 | if (!dst_path) { | |
4313 | ret = -ENOMEM; | |
4314 | goto out; | |
4315 | } | |
4316 | } | |
4317 | } | |
0bc2d3c0 | 4318 | if (ins_nr > 0) |
0e56315c | 4319 | ret = copy_items(trans, inode, dst_path, path, |
31d11b83 | 4320 | start_slot, ins_nr, 1, 0); |
31d11b83 FM |
4321 | out: |
4322 | btrfs_release_path(path); | |
4323 | btrfs_free_path(dst_path); | |
4324 | return ret; | |
4325 | } | |
4326 | ||
5dc562c5 JB |
4327 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4328 | struct btrfs_root *root, | |
9d122629 | 4329 | struct btrfs_inode *inode, |
827463c4 | 4330 | struct btrfs_path *path, |
de0ee0ed FM |
4331 | struct btrfs_log_ctx *ctx, |
4332 | const u64 start, | |
4333 | const u64 end) | |
5dc562c5 | 4334 | { |
5dc562c5 JB |
4335 | struct extent_map *em, *n; |
4336 | struct list_head extents; | |
9d122629 | 4337 | struct extent_map_tree *tree = &inode->extent_tree; |
5dc562c5 JB |
4338 | u64 test_gen; |
4339 | int ret = 0; | |
2ab28f32 | 4340 | int num = 0; |
5dc562c5 JB |
4341 | |
4342 | INIT_LIST_HEAD(&extents); | |
4343 | ||
5dc562c5 JB |
4344 | write_lock(&tree->lock); |
4345 | test_gen = root->fs_info->last_trans_committed; | |
4346 | ||
4347 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
008c6753 FM |
4348 | /* |
4349 | * Skip extents outside our logging range. It's important to do | |
4350 | * it for correctness because if we don't ignore them, we may | |
4351 | * log them before their ordered extent completes, and therefore | |
4352 | * we could log them without logging their respective checksums | |
4353 | * (the checksum items are added to the csum tree at the very | |
4354 | * end of btrfs_finish_ordered_io()). Also leave such extents | |
4355 | * outside of our range in the list, since we may have another | |
4356 | * ranged fsync in the near future that needs them. If an extent | |
4357 | * outside our range corresponds to a hole, log it to avoid | |
4358 | * leaving gaps between extents (fsck will complain when we are | |
4359 | * not using the NO_HOLES feature). | |
4360 | */ | |
4361 | if ((em->start > end || em->start + em->len <= start) && | |
4362 | em->block_start != EXTENT_MAP_HOLE) | |
4363 | continue; | |
4364 | ||
5dc562c5 | 4365 | list_del_init(&em->list); |
2ab28f32 JB |
4366 | /* |
4367 | * Just an arbitrary number, this can be really CPU intensive | |
4368 | * once we start getting a lot of extents, and really once we | |
4369 | * have a bunch of extents we just want to commit since it will | |
4370 | * be faster. | |
4371 | */ | |
4372 | if (++num > 32768) { | |
4373 | list_del_init(&tree->modified_extents); | |
4374 | ret = -EFBIG; | |
4375 | goto process; | |
4376 | } | |
4377 | ||
5dc562c5 JB |
4378 | if (em->generation <= test_gen) |
4379 | continue; | |
8c6c5928 | 4380 | |
31d11b83 FM |
4381 | /* We log prealloc extents beyond eof later. */ |
4382 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && | |
4383 | em->start >= i_size_read(&inode->vfs_inode)) | |
4384 | continue; | |
4385 | ||
ff44c6e3 | 4386 | /* Need a ref to keep it from getting evicted from cache */ |
490b54d6 | 4387 | refcount_inc(&em->refs); |
ff44c6e3 | 4388 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); |
5dc562c5 | 4389 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4390 | num++; |
5dc562c5 JB |
4391 | } |
4392 | ||
4393 | list_sort(NULL, &extents, extent_cmp); | |
2ab28f32 | 4394 | process: |
5dc562c5 JB |
4395 | while (!list_empty(&extents)) { |
4396 | em = list_entry(extents.next, struct extent_map, list); | |
4397 | ||
4398 | list_del_init(&em->list); | |
4399 | ||
4400 | /* | |
4401 | * If we had an error we just need to delete everybody from our | |
4402 | * private list. | |
4403 | */ | |
ff44c6e3 | 4404 | if (ret) { |
201a9038 | 4405 | clear_em_logging(tree, em); |
ff44c6e3 | 4406 | free_extent_map(em); |
5dc562c5 | 4407 | continue; |
ff44c6e3 JB |
4408 | } |
4409 | ||
4410 | write_unlock(&tree->lock); | |
5dc562c5 | 4411 | |
a2120a47 | 4412 | ret = log_one_extent(trans, inode, root, em, path, ctx); |
ff44c6e3 | 4413 | write_lock(&tree->lock); |
201a9038 JB |
4414 | clear_em_logging(tree, em); |
4415 | free_extent_map(em); | |
5dc562c5 | 4416 | } |
ff44c6e3 JB |
4417 | WARN_ON(!list_empty(&extents)); |
4418 | write_unlock(&tree->lock); | |
5dc562c5 | 4419 | |
5dc562c5 | 4420 | btrfs_release_path(path); |
31d11b83 FM |
4421 | if (!ret) |
4422 | ret = btrfs_log_prealloc_extents(trans, inode, path); | |
4423 | ||
5dc562c5 JB |
4424 | return ret; |
4425 | } | |
4426 | ||
481b01c0 | 4427 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
4428 | struct btrfs_path *path, u64 *size_ret) |
4429 | { | |
4430 | struct btrfs_key key; | |
4431 | int ret; | |
4432 | ||
481b01c0 | 4433 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
4434 | key.type = BTRFS_INODE_ITEM_KEY; |
4435 | key.offset = 0; | |
4436 | ||
4437 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4438 | if (ret < 0) { | |
4439 | return ret; | |
4440 | } else if (ret > 0) { | |
2f2ff0ee | 4441 | *size_ret = 0; |
1a4bcf47 FM |
4442 | } else { |
4443 | struct btrfs_inode_item *item; | |
4444 | ||
4445 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4446 | struct btrfs_inode_item); | |
4447 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
bf504110 FM |
4448 | /* |
4449 | * If the in-memory inode's i_size is smaller then the inode | |
4450 | * size stored in the btree, return the inode's i_size, so | |
4451 | * that we get a correct inode size after replaying the log | |
4452 | * when before a power failure we had a shrinking truncate | |
4453 | * followed by addition of a new name (rename / new hard link). | |
4454 | * Otherwise return the inode size from the btree, to avoid | |
4455 | * data loss when replaying a log due to previously doing a | |
4456 | * write that expands the inode's size and logging a new name | |
4457 | * immediately after. | |
4458 | */ | |
4459 | if (*size_ret > inode->vfs_inode.i_size) | |
4460 | *size_ret = inode->vfs_inode.i_size; | |
1a4bcf47 FM |
4461 | } |
4462 | ||
4463 | btrfs_release_path(path); | |
4464 | return 0; | |
4465 | } | |
4466 | ||
36283bf7 FM |
4467 | /* |
4468 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4469 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4470 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4471 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4472 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4473 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4474 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4475 | */ | |
4476 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4477 | struct btrfs_root *root, | |
1a93c36a | 4478 | struct btrfs_inode *inode, |
36283bf7 FM |
4479 | struct btrfs_path *path, |
4480 | struct btrfs_path *dst_path) | |
4481 | { | |
4482 | int ret; | |
4483 | struct btrfs_key key; | |
1a93c36a | 4484 | const u64 ino = btrfs_ino(inode); |
36283bf7 FM |
4485 | int ins_nr = 0; |
4486 | int start_slot = 0; | |
4487 | ||
4488 | key.objectid = ino; | |
4489 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4490 | key.offset = 0; | |
4491 | ||
4492 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4493 | if (ret < 0) | |
4494 | return ret; | |
4495 | ||
4496 | while (true) { | |
4497 | int slot = path->slots[0]; | |
4498 | struct extent_buffer *leaf = path->nodes[0]; | |
4499 | int nritems = btrfs_header_nritems(leaf); | |
4500 | ||
4501 | if (slot >= nritems) { | |
4502 | if (ins_nr > 0) { | |
1a93c36a | 4503 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 4504 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
4505 | if (ret < 0) |
4506 | return ret; | |
4507 | ins_nr = 0; | |
4508 | } | |
4509 | ret = btrfs_next_leaf(root, path); | |
4510 | if (ret < 0) | |
4511 | return ret; | |
4512 | else if (ret > 0) | |
4513 | break; | |
4514 | continue; | |
4515 | } | |
4516 | ||
4517 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4518 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4519 | break; | |
4520 | ||
4521 | if (ins_nr == 0) | |
4522 | start_slot = slot; | |
4523 | ins_nr++; | |
4524 | path->slots[0]++; | |
4525 | cond_resched(); | |
4526 | } | |
4527 | if (ins_nr > 0) { | |
1a93c36a | 4528 | ret = copy_items(trans, inode, dst_path, path, |
0e56315c | 4529 | start_slot, ins_nr, 1, 0); |
36283bf7 FM |
4530 | if (ret < 0) |
4531 | return ret; | |
4532 | } | |
4533 | ||
4534 | return 0; | |
4535 | } | |
4536 | ||
a89ca6f2 | 4537 | /* |
0e56315c FM |
4538 | * When using the NO_HOLES feature if we punched a hole that causes the |
4539 | * deletion of entire leafs or all the extent items of the first leaf (the one | |
4540 | * that contains the inode item and references) we may end up not processing | |
4541 | * any extents, because there are no leafs with a generation matching the | |
4542 | * current transaction that have extent items for our inode. So we need to find | |
4543 | * if any holes exist and then log them. We also need to log holes after any | |
4544 | * truncate operation that changes the inode's size. | |
a89ca6f2 | 4545 | */ |
0e56315c FM |
4546 | static int btrfs_log_holes(struct btrfs_trans_handle *trans, |
4547 | struct btrfs_root *root, | |
4548 | struct btrfs_inode *inode, | |
7af59743 | 4549 | struct btrfs_path *path) |
a89ca6f2 | 4550 | { |
0b246afa | 4551 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 | 4552 | struct btrfs_key key; |
a0308dd7 NB |
4553 | const u64 ino = btrfs_ino(inode); |
4554 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
7af59743 | 4555 | u64 prev_extent_end = 0; |
0e56315c | 4556 | int ret; |
a89ca6f2 | 4557 | |
0e56315c | 4558 | if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0) |
a89ca6f2 FM |
4559 | return 0; |
4560 | ||
4561 | key.objectid = ino; | |
4562 | key.type = BTRFS_EXTENT_DATA_KEY; | |
7af59743 | 4563 | key.offset = 0; |
a89ca6f2 FM |
4564 | |
4565 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
a89ca6f2 FM |
4566 | if (ret < 0) |
4567 | return ret; | |
4568 | ||
0e56315c | 4569 | while (true) { |
0e56315c | 4570 | struct extent_buffer *leaf = path->nodes[0]; |
a89ca6f2 | 4571 | |
0e56315c FM |
4572 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
4573 | ret = btrfs_next_leaf(root, path); | |
4574 | if (ret < 0) | |
4575 | return ret; | |
4576 | if (ret > 0) { | |
4577 | ret = 0; | |
4578 | break; | |
4579 | } | |
4580 | leaf = path->nodes[0]; | |
4581 | } | |
4582 | ||
4583 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4584 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
4585 | break; | |
4586 | ||
4587 | /* We have a hole, log it. */ | |
4588 | if (prev_extent_end < key.offset) { | |
7af59743 | 4589 | const u64 hole_len = key.offset - prev_extent_end; |
0e56315c FM |
4590 | |
4591 | /* | |
4592 | * Release the path to avoid deadlocks with other code | |
4593 | * paths that search the root while holding locks on | |
4594 | * leafs from the log root. | |
4595 | */ | |
4596 | btrfs_release_path(path); | |
4597 | ret = btrfs_insert_file_extent(trans, root->log_root, | |
4598 | ino, prev_extent_end, 0, | |
4599 | 0, hole_len, 0, hole_len, | |
4600 | 0, 0, 0); | |
4601 | if (ret < 0) | |
4602 | return ret; | |
4603 | ||
4604 | /* | |
4605 | * Search for the same key again in the root. Since it's | |
4606 | * an extent item and we are holding the inode lock, the | |
4607 | * key must still exist. If it doesn't just emit warning | |
4608 | * and return an error to fall back to a transaction | |
4609 | * commit. | |
4610 | */ | |
4611 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4612 | if (ret < 0) | |
4613 | return ret; | |
4614 | if (WARN_ON(ret > 0)) | |
4615 | return -ENOENT; | |
4616 | leaf = path->nodes[0]; | |
4617 | } | |
a89ca6f2 | 4618 | |
7af59743 | 4619 | prev_extent_end = btrfs_file_extent_end(path); |
0e56315c FM |
4620 | path->slots[0]++; |
4621 | cond_resched(); | |
a89ca6f2 | 4622 | } |
a89ca6f2 | 4623 | |
7af59743 | 4624 | if (prev_extent_end < i_size) { |
0e56315c | 4625 | u64 hole_len; |
a89ca6f2 | 4626 | |
0e56315c | 4627 | btrfs_release_path(path); |
7af59743 | 4628 | hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize); |
0e56315c FM |
4629 | ret = btrfs_insert_file_extent(trans, root->log_root, |
4630 | ino, prev_extent_end, 0, 0, | |
4631 | hole_len, 0, hole_len, | |
4632 | 0, 0, 0); | |
4633 | if (ret < 0) | |
4634 | return ret; | |
4635 | } | |
4636 | ||
4637 | return 0; | |
a89ca6f2 FM |
4638 | } |
4639 | ||
56f23fdb FM |
4640 | /* |
4641 | * When we are logging a new inode X, check if it doesn't have a reference that | |
4642 | * matches the reference from some other inode Y created in a past transaction | |
4643 | * and that was renamed in the current transaction. If we don't do this, then at | |
4644 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
4645 | * | |
4646 | * mkdir /mnt/x | |
4647 | * echo "hello world" > /mnt/x/foobar | |
4648 | * sync | |
4649 | * mv /mnt/x /mnt/y | |
4650 | * mkdir /mnt/x # or touch /mnt/x | |
4651 | * xfs_io -c fsync /mnt/x | |
4652 | * <power fail> | |
4653 | * mount fs, trigger log replay | |
4654 | * | |
4655 | * After the log replay procedure, we would lose the first directory and all its | |
4656 | * files (file foobar). | |
4657 | * For the case where inode Y is not a directory we simply end up losing it: | |
4658 | * | |
4659 | * echo "123" > /mnt/foo | |
4660 | * sync | |
4661 | * mv /mnt/foo /mnt/bar | |
4662 | * echo "abc" > /mnt/foo | |
4663 | * xfs_io -c fsync /mnt/foo | |
4664 | * <power fail> | |
4665 | * | |
4666 | * We also need this for cases where a snapshot entry is replaced by some other | |
4667 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
4668 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
4669 | * if it were a regular entry: | |
4670 | * | |
4671 | * mkdir /mnt/x | |
4672 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
4673 | * btrfs subvolume delete /mnt/x/snap | |
4674 | * rmdir /mnt/x | |
4675 | * mkdir /mnt/x | |
4676 | * fsync /mnt/x or fsync some new file inside it | |
4677 | * <power fail> | |
4678 | * | |
4679 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
4680 | * the same transaction. | |
4681 | */ | |
4682 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
4683 | const int slot, | |
4684 | const struct btrfs_key *key, | |
4791c8f1 | 4685 | struct btrfs_inode *inode, |
a3baaf0d | 4686 | u64 *other_ino, u64 *other_parent) |
56f23fdb FM |
4687 | { |
4688 | int ret; | |
4689 | struct btrfs_path *search_path; | |
4690 | char *name = NULL; | |
4691 | u32 name_len = 0; | |
4692 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
4693 | u32 cur_offset = 0; | |
4694 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
4695 | ||
4696 | search_path = btrfs_alloc_path(); | |
4697 | if (!search_path) | |
4698 | return -ENOMEM; | |
4699 | search_path->search_commit_root = 1; | |
4700 | search_path->skip_locking = 1; | |
4701 | ||
4702 | while (cur_offset < item_size) { | |
4703 | u64 parent; | |
4704 | u32 this_name_len; | |
4705 | u32 this_len; | |
4706 | unsigned long name_ptr; | |
4707 | struct btrfs_dir_item *di; | |
4708 | ||
4709 | if (key->type == BTRFS_INODE_REF_KEY) { | |
4710 | struct btrfs_inode_ref *iref; | |
4711 | ||
4712 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
4713 | parent = key->offset; | |
4714 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
4715 | name_ptr = (unsigned long)(iref + 1); | |
4716 | this_len = sizeof(*iref) + this_name_len; | |
4717 | } else { | |
4718 | struct btrfs_inode_extref *extref; | |
4719 | ||
4720 | extref = (struct btrfs_inode_extref *)(ptr + | |
4721 | cur_offset); | |
4722 | parent = btrfs_inode_extref_parent(eb, extref); | |
4723 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
4724 | name_ptr = (unsigned long)&extref->name; | |
4725 | this_len = sizeof(*extref) + this_name_len; | |
4726 | } | |
4727 | ||
4728 | if (this_name_len > name_len) { | |
4729 | char *new_name; | |
4730 | ||
4731 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
4732 | if (!new_name) { | |
4733 | ret = -ENOMEM; | |
4734 | goto out; | |
4735 | } | |
4736 | name_len = this_name_len; | |
4737 | name = new_name; | |
4738 | } | |
4739 | ||
4740 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4791c8f1 NB |
4741 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
4742 | parent, name, this_name_len, 0); | |
56f23fdb | 4743 | if (di && !IS_ERR(di)) { |
44f714da FM |
4744 | struct btrfs_key di_key; |
4745 | ||
4746 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
4747 | di, &di_key); | |
4748 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
6b5fc433 FM |
4749 | if (di_key.objectid != key->objectid) { |
4750 | ret = 1; | |
4751 | *other_ino = di_key.objectid; | |
a3baaf0d | 4752 | *other_parent = parent; |
6b5fc433 FM |
4753 | } else { |
4754 | ret = 0; | |
4755 | } | |
44f714da FM |
4756 | } else { |
4757 | ret = -EAGAIN; | |
4758 | } | |
56f23fdb FM |
4759 | goto out; |
4760 | } else if (IS_ERR(di)) { | |
4761 | ret = PTR_ERR(di); | |
4762 | goto out; | |
4763 | } | |
4764 | btrfs_release_path(search_path); | |
4765 | ||
4766 | cur_offset += this_len; | |
4767 | } | |
4768 | ret = 0; | |
4769 | out: | |
4770 | btrfs_free_path(search_path); | |
4771 | kfree(name); | |
4772 | return ret; | |
4773 | } | |
4774 | ||
6b5fc433 FM |
4775 | struct btrfs_ino_list { |
4776 | u64 ino; | |
a3baaf0d | 4777 | u64 parent; |
6b5fc433 FM |
4778 | struct list_head list; |
4779 | }; | |
4780 | ||
4781 | static int log_conflicting_inodes(struct btrfs_trans_handle *trans, | |
4782 | struct btrfs_root *root, | |
4783 | struct btrfs_path *path, | |
4784 | struct btrfs_log_ctx *ctx, | |
a3baaf0d | 4785 | u64 ino, u64 parent) |
6b5fc433 FM |
4786 | { |
4787 | struct btrfs_ino_list *ino_elem; | |
4788 | LIST_HEAD(inode_list); | |
4789 | int ret = 0; | |
4790 | ||
4791 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
4792 | if (!ino_elem) | |
4793 | return -ENOMEM; | |
4794 | ino_elem->ino = ino; | |
a3baaf0d | 4795 | ino_elem->parent = parent; |
6b5fc433 FM |
4796 | list_add_tail(&ino_elem->list, &inode_list); |
4797 | ||
4798 | while (!list_empty(&inode_list)) { | |
4799 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4800 | struct btrfs_key key; | |
4801 | struct inode *inode; | |
4802 | ||
4803 | ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list, | |
4804 | list); | |
4805 | ino = ino_elem->ino; | |
a3baaf0d | 4806 | parent = ino_elem->parent; |
6b5fc433 FM |
4807 | list_del(&ino_elem->list); |
4808 | kfree(ino_elem); | |
4809 | if (ret) | |
4810 | continue; | |
4811 | ||
4812 | btrfs_release_path(path); | |
4813 | ||
0202e83f | 4814 | inode = btrfs_iget(fs_info->sb, ino, root); |
6b5fc433 FM |
4815 | /* |
4816 | * If the other inode that had a conflicting dir entry was | |
a3baaf0d FM |
4817 | * deleted in the current transaction, we need to log its parent |
4818 | * directory. | |
6b5fc433 FM |
4819 | */ |
4820 | if (IS_ERR(inode)) { | |
4821 | ret = PTR_ERR(inode); | |
a3baaf0d | 4822 | if (ret == -ENOENT) { |
0202e83f | 4823 | inode = btrfs_iget(fs_info->sb, parent, root); |
a3baaf0d FM |
4824 | if (IS_ERR(inode)) { |
4825 | ret = PTR_ERR(inode); | |
4826 | } else { | |
4827 | ret = btrfs_log_inode(trans, root, | |
4828 | BTRFS_I(inode), | |
4829 | LOG_OTHER_INODE_ALL, | |
4830 | 0, LLONG_MAX, ctx); | |
410f954c | 4831 | btrfs_add_delayed_iput(inode); |
a3baaf0d FM |
4832 | } |
4833 | } | |
6b5fc433 FM |
4834 | continue; |
4835 | } | |
b5e4ff9d FM |
4836 | /* |
4837 | * If the inode was already logged skip it - otherwise we can | |
4838 | * hit an infinite loop. Example: | |
4839 | * | |
4840 | * From the commit root (previous transaction) we have the | |
4841 | * following inodes: | |
4842 | * | |
4843 | * inode 257 a directory | |
4844 | * inode 258 with references "zz" and "zz_link" on inode 257 | |
4845 | * inode 259 with reference "a" on inode 257 | |
4846 | * | |
4847 | * And in the current (uncommitted) transaction we have: | |
4848 | * | |
4849 | * inode 257 a directory, unchanged | |
4850 | * inode 258 with references "a" and "a2" on inode 257 | |
4851 | * inode 259 with reference "zz_link" on inode 257 | |
4852 | * inode 261 with reference "zz" on inode 257 | |
4853 | * | |
4854 | * When logging inode 261 the following infinite loop could | |
4855 | * happen if we don't skip already logged inodes: | |
4856 | * | |
4857 | * - we detect inode 258 as a conflicting inode, with inode 261 | |
4858 | * on reference "zz", and log it; | |
4859 | * | |
4860 | * - we detect inode 259 as a conflicting inode, with inode 258 | |
4861 | * on reference "a", and log it; | |
4862 | * | |
4863 | * - we detect inode 258 as a conflicting inode, with inode 259 | |
4864 | * on reference "zz_link", and log it - again! After this we | |
4865 | * repeat the above steps forever. | |
4866 | */ | |
4867 | spin_lock(&BTRFS_I(inode)->lock); | |
4868 | /* | |
4869 | * Check the inode's logged_trans only instead of | |
4870 | * btrfs_inode_in_log(). This is because the last_log_commit of | |
4871 | * the inode is not updated when we only log that it exists and | |
4872 | * and it has the full sync bit set (see btrfs_log_inode()). | |
4873 | */ | |
4874 | if (BTRFS_I(inode)->logged_trans == trans->transid) { | |
4875 | spin_unlock(&BTRFS_I(inode)->lock); | |
4876 | btrfs_add_delayed_iput(inode); | |
4877 | continue; | |
4878 | } | |
4879 | spin_unlock(&BTRFS_I(inode)->lock); | |
6b5fc433 FM |
4880 | /* |
4881 | * We are safe logging the other inode without acquiring its | |
4882 | * lock as long as we log with the LOG_INODE_EXISTS mode. We | |
4883 | * are safe against concurrent renames of the other inode as | |
4884 | * well because during a rename we pin the log and update the | |
4885 | * log with the new name before we unpin it. | |
4886 | */ | |
4887 | ret = btrfs_log_inode(trans, root, BTRFS_I(inode), | |
4888 | LOG_OTHER_INODE, 0, LLONG_MAX, ctx); | |
4889 | if (ret) { | |
410f954c | 4890 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
4891 | continue; |
4892 | } | |
4893 | ||
4894 | key.objectid = ino; | |
4895 | key.type = BTRFS_INODE_REF_KEY; | |
4896 | key.offset = 0; | |
4897 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4898 | if (ret < 0) { | |
410f954c | 4899 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
4900 | continue; |
4901 | } | |
4902 | ||
4903 | while (true) { | |
4904 | struct extent_buffer *leaf = path->nodes[0]; | |
4905 | int slot = path->slots[0]; | |
4906 | u64 other_ino = 0; | |
a3baaf0d | 4907 | u64 other_parent = 0; |
6b5fc433 FM |
4908 | |
4909 | if (slot >= btrfs_header_nritems(leaf)) { | |
4910 | ret = btrfs_next_leaf(root, path); | |
4911 | if (ret < 0) { | |
4912 | break; | |
4913 | } else if (ret > 0) { | |
4914 | ret = 0; | |
4915 | break; | |
4916 | } | |
4917 | continue; | |
4918 | } | |
4919 | ||
4920 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4921 | if (key.objectid != ino || | |
4922 | (key.type != BTRFS_INODE_REF_KEY && | |
4923 | key.type != BTRFS_INODE_EXTREF_KEY)) { | |
4924 | ret = 0; | |
4925 | break; | |
4926 | } | |
4927 | ||
4928 | ret = btrfs_check_ref_name_override(leaf, slot, &key, | |
a3baaf0d FM |
4929 | BTRFS_I(inode), &other_ino, |
4930 | &other_parent); | |
6b5fc433 FM |
4931 | if (ret < 0) |
4932 | break; | |
4933 | if (ret > 0) { | |
4934 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
4935 | if (!ino_elem) { | |
4936 | ret = -ENOMEM; | |
4937 | break; | |
4938 | } | |
4939 | ino_elem->ino = other_ino; | |
a3baaf0d | 4940 | ino_elem->parent = other_parent; |
6b5fc433 FM |
4941 | list_add_tail(&ino_elem->list, &inode_list); |
4942 | ret = 0; | |
4943 | } | |
4944 | path->slots[0]++; | |
4945 | } | |
410f954c | 4946 | btrfs_add_delayed_iput(inode); |
6b5fc433 FM |
4947 | } |
4948 | ||
4949 | return ret; | |
4950 | } | |
4951 | ||
da447009 FM |
4952 | static int copy_inode_items_to_log(struct btrfs_trans_handle *trans, |
4953 | struct btrfs_inode *inode, | |
4954 | struct btrfs_key *min_key, | |
4955 | const struct btrfs_key *max_key, | |
4956 | struct btrfs_path *path, | |
4957 | struct btrfs_path *dst_path, | |
4958 | const u64 logged_isize, | |
4959 | const bool recursive_logging, | |
4960 | const int inode_only, | |
4961 | struct btrfs_log_ctx *ctx, | |
4962 | bool *need_log_inode_item) | |
4963 | { | |
4964 | struct btrfs_root *root = inode->root; | |
4965 | int ins_start_slot = 0; | |
4966 | int ins_nr = 0; | |
4967 | int ret; | |
4968 | ||
4969 | while (1) { | |
4970 | ret = btrfs_search_forward(root, min_key, path, trans->transid); | |
4971 | if (ret < 0) | |
4972 | return ret; | |
4973 | if (ret > 0) { | |
4974 | ret = 0; | |
4975 | break; | |
4976 | } | |
4977 | again: | |
4978 | /* Note, ins_nr might be > 0 here, cleanup outside the loop */ | |
4979 | if (min_key->objectid != max_key->objectid) | |
4980 | break; | |
4981 | if (min_key->type > max_key->type) | |
4982 | break; | |
4983 | ||
4984 | if (min_key->type == BTRFS_INODE_ITEM_KEY) | |
4985 | *need_log_inode_item = false; | |
4986 | ||
4987 | if ((min_key->type == BTRFS_INODE_REF_KEY || | |
4988 | min_key->type == BTRFS_INODE_EXTREF_KEY) && | |
4989 | inode->generation == trans->transid && | |
4990 | !recursive_logging) { | |
4991 | u64 other_ino = 0; | |
4992 | u64 other_parent = 0; | |
4993 | ||
4994 | ret = btrfs_check_ref_name_override(path->nodes[0], | |
4995 | path->slots[0], min_key, inode, | |
4996 | &other_ino, &other_parent); | |
4997 | if (ret < 0) { | |
4998 | return ret; | |
4999 | } else if (ret > 0 && ctx && | |
5000 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { | |
5001 | if (ins_nr > 0) { | |
5002 | ins_nr++; | |
5003 | } else { | |
5004 | ins_nr = 1; | |
5005 | ins_start_slot = path->slots[0]; | |
5006 | } | |
5007 | ret = copy_items(trans, inode, dst_path, path, | |
5008 | ins_start_slot, ins_nr, | |
5009 | inode_only, logged_isize); | |
5010 | if (ret < 0) | |
5011 | return ret; | |
5012 | ins_nr = 0; | |
5013 | ||
5014 | ret = log_conflicting_inodes(trans, root, path, | |
5015 | ctx, other_ino, other_parent); | |
5016 | if (ret) | |
5017 | return ret; | |
5018 | btrfs_release_path(path); | |
5019 | goto next_key; | |
5020 | } | |
5021 | } | |
5022 | ||
5023 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ | |
5024 | if (min_key->type == BTRFS_XATTR_ITEM_KEY) { | |
5025 | if (ins_nr == 0) | |
5026 | goto next_slot; | |
5027 | ret = copy_items(trans, inode, dst_path, path, | |
5028 | ins_start_slot, | |
5029 | ins_nr, inode_only, logged_isize); | |
5030 | if (ret < 0) | |
5031 | return ret; | |
5032 | ins_nr = 0; | |
5033 | goto next_slot; | |
5034 | } | |
5035 | ||
5036 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { | |
5037 | ins_nr++; | |
5038 | goto next_slot; | |
5039 | } else if (!ins_nr) { | |
5040 | ins_start_slot = path->slots[0]; | |
5041 | ins_nr = 1; | |
5042 | goto next_slot; | |
5043 | } | |
5044 | ||
5045 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, | |
5046 | ins_nr, inode_only, logged_isize); | |
5047 | if (ret < 0) | |
5048 | return ret; | |
5049 | ins_nr = 1; | |
5050 | ins_start_slot = path->slots[0]; | |
5051 | next_slot: | |
5052 | path->slots[0]++; | |
5053 | if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) { | |
5054 | btrfs_item_key_to_cpu(path->nodes[0], min_key, | |
5055 | path->slots[0]); | |
5056 | goto again; | |
5057 | } | |
5058 | if (ins_nr) { | |
5059 | ret = copy_items(trans, inode, dst_path, path, | |
5060 | ins_start_slot, ins_nr, inode_only, | |
5061 | logged_isize); | |
5062 | if (ret < 0) | |
5063 | return ret; | |
5064 | ins_nr = 0; | |
5065 | } | |
5066 | btrfs_release_path(path); | |
5067 | next_key: | |
5068 | if (min_key->offset < (u64)-1) { | |
5069 | min_key->offset++; | |
5070 | } else if (min_key->type < max_key->type) { | |
5071 | min_key->type++; | |
5072 | min_key->offset = 0; | |
5073 | } else { | |
5074 | break; | |
5075 | } | |
5076 | } | |
5077 | if (ins_nr) | |
5078 | ret = copy_items(trans, inode, dst_path, path, ins_start_slot, | |
5079 | ins_nr, inode_only, logged_isize); | |
5080 | ||
5081 | return ret; | |
5082 | } | |
5083 | ||
e02119d5 CM |
5084 | /* log a single inode in the tree log. |
5085 | * At least one parent directory for this inode must exist in the tree | |
5086 | * or be logged already. | |
5087 | * | |
5088 | * Any items from this inode changed by the current transaction are copied | |
5089 | * to the log tree. An extra reference is taken on any extents in this | |
5090 | * file, allowing us to avoid a whole pile of corner cases around logging | |
5091 | * blocks that have been removed from the tree. | |
5092 | * | |
5093 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
5094 | * does. | |
5095 | * | |
5096 | * This handles both files and directories. | |
5097 | */ | |
12fcfd22 | 5098 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 5099 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc | 5100 | int inode_only, |
7af59743 FM |
5101 | const loff_t start, |
5102 | const loff_t end, | |
8407f553 | 5103 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5104 | { |
0b246afa | 5105 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
5106 | struct btrfs_path *path; |
5107 | struct btrfs_path *dst_path; | |
5108 | struct btrfs_key min_key; | |
5109 | struct btrfs_key max_key; | |
5110 | struct btrfs_root *log = root->log_root; | |
4a500fd1 | 5111 | int err = 0; |
e02119d5 | 5112 | int ret; |
5dc562c5 | 5113 | bool fast_search = false; |
a59108a7 NB |
5114 | u64 ino = btrfs_ino(inode); |
5115 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
1a4bcf47 | 5116 | u64 logged_isize = 0; |
e4545de5 | 5117 | bool need_log_inode_item = true; |
9a8fca62 | 5118 | bool xattrs_logged = false; |
a3baaf0d | 5119 | bool recursive_logging = false; |
e02119d5 | 5120 | |
e02119d5 | 5121 | path = btrfs_alloc_path(); |
5df67083 TI |
5122 | if (!path) |
5123 | return -ENOMEM; | |
e02119d5 | 5124 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
5125 | if (!dst_path) { |
5126 | btrfs_free_path(path); | |
5127 | return -ENOMEM; | |
5128 | } | |
e02119d5 | 5129 | |
33345d01 | 5130 | min_key.objectid = ino; |
e02119d5 CM |
5131 | min_key.type = BTRFS_INODE_ITEM_KEY; |
5132 | min_key.offset = 0; | |
5133 | ||
33345d01 | 5134 | max_key.objectid = ino; |
12fcfd22 | 5135 | |
12fcfd22 | 5136 | |
5dc562c5 | 5137 | /* today the code can only do partial logging of directories */ |
a59108a7 | 5138 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5269b67e | 5139 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5140 | &inode->runtime_flags) && |
781feef7 | 5141 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
5142 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5143 | else | |
5144 | max_key.type = (u8)-1; | |
5145 | max_key.offset = (u64)-1; | |
5146 | ||
2c2c452b FM |
5147 | /* |
5148 | * Only run delayed items if we are a dir or a new file. | |
5149 | * Otherwise commit the delayed inode only, which is needed in | |
5150 | * order for the log replay code to mark inodes for link count | |
5151 | * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items). | |
5152 | */ | |
a59108a7 NB |
5153 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5154 | inode->generation > fs_info->last_trans_committed) | |
5155 | ret = btrfs_commit_inode_delayed_items(trans, inode); | |
2c2c452b | 5156 | else |
a59108a7 | 5157 | ret = btrfs_commit_inode_delayed_inode(inode); |
2c2c452b FM |
5158 | |
5159 | if (ret) { | |
5160 | btrfs_free_path(path); | |
5161 | btrfs_free_path(dst_path); | |
5162 | return ret; | |
16cdcec7 MX |
5163 | } |
5164 | ||
a3baaf0d FM |
5165 | if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) { |
5166 | recursive_logging = true; | |
5167 | if (inode_only == LOG_OTHER_INODE) | |
5168 | inode_only = LOG_INODE_EXISTS; | |
5169 | else | |
5170 | inode_only = LOG_INODE_ALL; | |
a59108a7 | 5171 | mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING); |
781feef7 | 5172 | } else { |
a59108a7 | 5173 | mutex_lock(&inode->log_mutex); |
781feef7 | 5174 | } |
e02119d5 CM |
5175 | |
5176 | /* | |
5177 | * a brute force approach to making sure we get the most uptodate | |
5178 | * copies of everything. | |
5179 | */ | |
a59108a7 | 5180 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
e02119d5 CM |
5181 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; |
5182 | ||
4f764e51 FM |
5183 | if (inode_only == LOG_INODE_EXISTS) |
5184 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 5185 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 5186 | } else { |
1a4bcf47 FM |
5187 | if (inode_only == LOG_INODE_EXISTS) { |
5188 | /* | |
5189 | * Make sure the new inode item we write to the log has | |
5190 | * the same isize as the current one (if it exists). | |
5191 | * This is necessary to prevent data loss after log | |
5192 | * replay, and also to prevent doing a wrong expanding | |
5193 | * truncate - for e.g. create file, write 4K into offset | |
5194 | * 0, fsync, write 4K into offset 4096, add hard link, | |
5195 | * fsync some other file (to sync log), power fail - if | |
5196 | * we use the inode's current i_size, after log replay | |
5197 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
5198 | * (zeroes), as if an expanding truncate happened, | |
5199 | * instead of getting a file of 4Kb only. | |
5200 | */ | |
a59108a7 | 5201 | err = logged_inode_size(log, inode, path, &logged_isize); |
1a4bcf47 FM |
5202 | if (err) |
5203 | goto out_unlock; | |
5204 | } | |
a742994a | 5205 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5206 | &inode->runtime_flags)) { |
a742994a | 5207 | if (inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5208 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
5209 | ret = drop_objectid_items(trans, log, path, ino, |
5210 | max_key.type); | |
5211 | } else { | |
5212 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a59108a7 | 5213 | &inode->runtime_flags); |
a742994a | 5214 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5215 | &inode->runtime_flags); |
28ed1345 CM |
5216 | while(1) { |
5217 | ret = btrfs_truncate_inode_items(trans, | |
a59108a7 | 5218 | log, &inode->vfs_inode, 0, 0); |
28ed1345 CM |
5219 | if (ret != -EAGAIN) |
5220 | break; | |
5221 | } | |
a742994a | 5222 | } |
4f764e51 | 5223 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5224 | &inode->runtime_flags) || |
6cfab851 | 5225 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5226 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 5227 | fast_search = true; |
4f764e51 | 5228 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 5229 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 5230 | max_key.type); |
a95249b3 JB |
5231 | } else { |
5232 | if (inode_only == LOG_INODE_ALL) | |
5233 | fast_search = true; | |
a95249b3 | 5234 | goto log_extents; |
5dc562c5 | 5235 | } |
a95249b3 | 5236 | |
e02119d5 | 5237 | } |
4a500fd1 YZ |
5238 | if (ret) { |
5239 | err = ret; | |
5240 | goto out_unlock; | |
5241 | } | |
e02119d5 | 5242 | |
da447009 FM |
5243 | err = copy_inode_items_to_log(trans, inode, &min_key, &max_key, |
5244 | path, dst_path, logged_isize, | |
7af59743 FM |
5245 | recursive_logging, inode_only, ctx, |
5246 | &need_log_inode_item); | |
da447009 FM |
5247 | if (err) |
5248 | goto out_unlock; | |
5dc562c5 | 5249 | |
36283bf7 FM |
5250 | btrfs_release_path(path); |
5251 | btrfs_release_path(dst_path); | |
a59108a7 | 5252 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); |
36283bf7 FM |
5253 | if (err) |
5254 | goto out_unlock; | |
9a8fca62 | 5255 | xattrs_logged = true; |
a89ca6f2 FM |
5256 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
5257 | btrfs_release_path(path); | |
5258 | btrfs_release_path(dst_path); | |
7af59743 | 5259 | err = btrfs_log_holes(trans, root, inode, path); |
a89ca6f2 FM |
5260 | if (err) |
5261 | goto out_unlock; | |
5262 | } | |
a95249b3 | 5263 | log_extents: |
f3b15ccd JB |
5264 | btrfs_release_path(path); |
5265 | btrfs_release_path(dst_path); | |
e4545de5 | 5266 | if (need_log_inode_item) { |
a59108a7 | 5267 | err = log_inode_item(trans, log, dst_path, inode); |
9a8fca62 FM |
5268 | if (!err && !xattrs_logged) { |
5269 | err = btrfs_log_all_xattrs(trans, root, inode, path, | |
5270 | dst_path); | |
5271 | btrfs_release_path(path); | |
5272 | } | |
e4545de5 FM |
5273 | if (err) |
5274 | goto out_unlock; | |
5275 | } | |
5dc562c5 | 5276 | if (fast_search) { |
a59108a7 | 5277 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
a2120a47 | 5278 | ctx, start, end); |
5dc562c5 JB |
5279 | if (ret) { |
5280 | err = ret; | |
5281 | goto out_unlock; | |
5282 | } | |
d006a048 | 5283 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
5284 | struct extent_map *em, *n; |
5285 | ||
49dae1bc FM |
5286 | write_lock(&em_tree->lock); |
5287 | /* | |
5288 | * We can't just remove every em if we're called for a ranged | |
5289 | * fsync - that is, one that doesn't cover the whole possible | |
5290 | * file range (0 to LLONG_MAX). This is because we can have | |
5291 | * em's that fall outside the range we're logging and therefore | |
5292 | * their ordered operations haven't completed yet | |
5293 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
5294 | * didn't get their respective file extent item in the fs/subvol | |
5295 | * tree yet, and need to let the next fast fsync (one which | |
5296 | * consults the list of modified extent maps) find the em so | |
5297 | * that it logs a matching file extent item and waits for the | |
5298 | * respective ordered operation to complete (if it's still | |
5299 | * running). | |
5300 | * | |
5301 | * Removing every em outside the range we're logging would make | |
5302 | * the next fast fsync not log their matching file extent items, | |
5303 | * therefore making us lose data after a log replay. | |
5304 | */ | |
5305 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
5306 | list) { | |
5307 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
5308 | ||
5309 | if (em->mod_start >= start && mod_end <= end) | |
5310 | list_del_init(&em->list); | |
5311 | } | |
5312 | write_unlock(&em_tree->lock); | |
5dc562c5 JB |
5313 | } |
5314 | ||
a59108a7 NB |
5315 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) { |
5316 | ret = log_directory_changes(trans, root, inode, path, dst_path, | |
5317 | ctx); | |
4a500fd1 YZ |
5318 | if (ret) { |
5319 | err = ret; | |
5320 | goto out_unlock; | |
5321 | } | |
e02119d5 | 5322 | } |
49dae1bc | 5323 | |
d1d832a0 FM |
5324 | /* |
5325 | * Don't update last_log_commit if we logged that an inode exists after | |
5326 | * it was loaded to memory (full_sync bit set). | |
5327 | * This is to prevent data loss when we do a write to the inode, then | |
5328 | * the inode gets evicted after all delalloc was flushed, then we log | |
5329 | * it exists (due to a rename for example) and then fsync it. This last | |
5330 | * fsync would do nothing (not logging the extents previously written). | |
5331 | */ | |
a59108a7 NB |
5332 | spin_lock(&inode->lock); |
5333 | inode->logged_trans = trans->transid; | |
d1d832a0 FM |
5334 | if (inode_only != LOG_INODE_EXISTS || |
5335 | !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags)) | |
5336 | inode->last_log_commit = inode->last_sub_trans; | |
a59108a7 | 5337 | spin_unlock(&inode->lock); |
4a500fd1 | 5338 | out_unlock: |
a59108a7 | 5339 | mutex_unlock(&inode->log_mutex); |
e02119d5 CM |
5340 | |
5341 | btrfs_free_path(path); | |
5342 | btrfs_free_path(dst_path); | |
4a500fd1 | 5343 | return err; |
e02119d5 CM |
5344 | } |
5345 | ||
2be63d5c FM |
5346 | /* |
5347 | * Check if we must fallback to a transaction commit when logging an inode. | |
5348 | * This must be called after logging the inode and is used only in the context | |
5349 | * when fsyncing an inode requires the need to log some other inode - in which | |
5350 | * case we can't lock the i_mutex of each other inode we need to log as that | |
5351 | * can lead to deadlocks with concurrent fsync against other inodes (as we can | |
5352 | * log inodes up or down in the hierarchy) or rename operations for example. So | |
5353 | * we take the log_mutex of the inode after we have logged it and then check for | |
5354 | * its last_unlink_trans value - this is safe because any task setting | |
5355 | * last_unlink_trans must take the log_mutex and it must do this before it does | |
5356 | * the actual unlink operation, so if we do this check before a concurrent task | |
5357 | * sets last_unlink_trans it means we've logged a consistent version/state of | |
5358 | * all the inode items, otherwise we are not sure and must do a transaction | |
01327610 | 5359 | * commit (the concurrent task might have only updated last_unlink_trans before |
2be63d5c FM |
5360 | * we logged the inode or it might have also done the unlink). |
5361 | */ | |
5362 | static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans, | |
ab1717b2 | 5363 | struct btrfs_inode *inode) |
2be63d5c | 5364 | { |
ab1717b2 | 5365 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2be63d5c FM |
5366 | bool ret = false; |
5367 | ||
ab1717b2 NB |
5368 | mutex_lock(&inode->log_mutex); |
5369 | if (inode->last_unlink_trans > fs_info->last_trans_committed) { | |
2be63d5c FM |
5370 | /* |
5371 | * Make sure any commits to the log are forced to be full | |
5372 | * commits. | |
5373 | */ | |
90787766 | 5374 | btrfs_set_log_full_commit(trans); |
2be63d5c FM |
5375 | ret = true; |
5376 | } | |
ab1717b2 | 5377 | mutex_unlock(&inode->log_mutex); |
2be63d5c FM |
5378 | |
5379 | return ret; | |
5380 | } | |
5381 | ||
12fcfd22 CM |
5382 | /* |
5383 | * follow the dentry parent pointers up the chain and see if any | |
5384 | * of the directories in it require a full commit before they can | |
5385 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
5386 | * a full commit is required. | |
5387 | */ | |
5388 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
aefa6115 | 5389 | struct btrfs_inode *inode, |
12fcfd22 CM |
5390 | struct dentry *parent, |
5391 | struct super_block *sb, | |
5392 | u64 last_committed) | |
e02119d5 | 5393 | { |
12fcfd22 | 5394 | int ret = 0; |
6a912213 | 5395 | struct dentry *old_parent = NULL; |
e02119d5 | 5396 | |
af4176b4 CM |
5397 | /* |
5398 | * for regular files, if its inode is already on disk, we don't | |
5399 | * have to worry about the parents at all. This is because | |
5400 | * we can use the last_unlink_trans field to record renames | |
5401 | * and other fun in this file. | |
5402 | */ | |
aefa6115 NB |
5403 | if (S_ISREG(inode->vfs_inode.i_mode) && |
5404 | inode->generation <= last_committed && | |
5405 | inode->last_unlink_trans <= last_committed) | |
5406 | goto out; | |
af4176b4 | 5407 | |
aefa6115 | 5408 | if (!S_ISDIR(inode->vfs_inode.i_mode)) { |
fc64005c | 5409 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 | 5410 | goto out; |
aefa6115 | 5411 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5412 | } |
5413 | ||
5414 | while (1) { | |
aefa6115 | 5415 | if (btrfs_must_commit_transaction(trans, inode)) { |
12fcfd22 CM |
5416 | ret = 1; |
5417 | break; | |
5418 | } | |
5419 | ||
fc64005c | 5420 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 CM |
5421 | break; |
5422 | ||
44f714da | 5423 | if (IS_ROOT(parent)) { |
aefa6115 NB |
5424 | inode = BTRFS_I(d_inode(parent)); |
5425 | if (btrfs_must_commit_transaction(trans, inode)) | |
44f714da | 5426 | ret = 1; |
12fcfd22 | 5427 | break; |
44f714da | 5428 | } |
12fcfd22 | 5429 | |
6a912213 JB |
5430 | parent = dget_parent(parent); |
5431 | dput(old_parent); | |
5432 | old_parent = parent; | |
aefa6115 | 5433 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5434 | |
5435 | } | |
6a912213 | 5436 | dput(old_parent); |
12fcfd22 | 5437 | out: |
e02119d5 CM |
5438 | return ret; |
5439 | } | |
5440 | ||
2f2ff0ee FM |
5441 | struct btrfs_dir_list { |
5442 | u64 ino; | |
5443 | struct list_head list; | |
5444 | }; | |
5445 | ||
5446 | /* | |
5447 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
5448 | * details about the why it is needed. | |
5449 | * This is a recursive operation - if an existing dentry corresponds to a | |
5450 | * directory, that directory's new entries are logged too (same behaviour as | |
5451 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5452 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
5453 | * complains about the following circular lock dependency / possible deadlock: | |
5454 | * | |
5455 | * CPU0 CPU1 | |
5456 | * ---- ---- | |
5457 | * lock(&type->i_mutex_dir_key#3/2); | |
5458 | * lock(sb_internal#2); | |
5459 | * lock(&type->i_mutex_dir_key#3/2); | |
5460 | * lock(&sb->s_type->i_mutex_key#14); | |
5461 | * | |
5462 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5463 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5464 | * Not locking i_mutex of the inodes is still safe because: | |
5465 | * | |
5466 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5467 | * that while logging the inode new references (names) are added or removed | |
5468 | * from the inode, leaving the logged inode item with a link count that does | |
5469 | * not match the number of logged inode reference items. This is fine because | |
5470 | * at log replay time we compute the real number of links and correct the | |
5471 | * link count in the inode item (see replay_one_buffer() and | |
5472 | * link_to_fixup_dir()); | |
5473 | * | |
5474 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5475 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
5476 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
5477 | * has a size that doesn't match the sum of the lengths of all the logged | |
5478 | * names. This does not result in a problem because if a dir_item key is | |
5479 | * logged but its matching dir_index key is not logged, at log replay time we | |
5480 | * don't use it to replay the respective name (see replay_one_name()). On the | |
5481 | * other hand if only the dir_index key ends up being logged, the respective | |
5482 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
5483 | * keys created (see replay_one_name()). | |
5484 | * The directory's inode item with a wrong i_size is not a problem as well, | |
5485 | * since we don't use it at log replay time to set the i_size in the inode | |
5486 | * item of the fs/subvol tree (see overwrite_item()). | |
5487 | */ | |
5488 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5489 | struct btrfs_root *root, | |
51cc0d32 | 5490 | struct btrfs_inode *start_inode, |
2f2ff0ee FM |
5491 | struct btrfs_log_ctx *ctx) |
5492 | { | |
0b246afa | 5493 | struct btrfs_fs_info *fs_info = root->fs_info; |
2f2ff0ee FM |
5494 | struct btrfs_root *log = root->log_root; |
5495 | struct btrfs_path *path; | |
5496 | LIST_HEAD(dir_list); | |
5497 | struct btrfs_dir_list *dir_elem; | |
5498 | int ret = 0; | |
5499 | ||
5500 | path = btrfs_alloc_path(); | |
5501 | if (!path) | |
5502 | return -ENOMEM; | |
5503 | ||
5504 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5505 | if (!dir_elem) { | |
5506 | btrfs_free_path(path); | |
5507 | return -ENOMEM; | |
5508 | } | |
51cc0d32 | 5509 | dir_elem->ino = btrfs_ino(start_inode); |
2f2ff0ee FM |
5510 | list_add_tail(&dir_elem->list, &dir_list); |
5511 | ||
5512 | while (!list_empty(&dir_list)) { | |
5513 | struct extent_buffer *leaf; | |
5514 | struct btrfs_key min_key; | |
5515 | int nritems; | |
5516 | int i; | |
5517 | ||
5518 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
5519 | list); | |
5520 | if (ret) | |
5521 | goto next_dir_inode; | |
5522 | ||
5523 | min_key.objectid = dir_elem->ino; | |
5524 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
5525 | min_key.offset = 0; | |
5526 | again: | |
5527 | btrfs_release_path(path); | |
5528 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
5529 | if (ret < 0) { | |
5530 | goto next_dir_inode; | |
5531 | } else if (ret > 0) { | |
5532 | ret = 0; | |
5533 | goto next_dir_inode; | |
5534 | } | |
5535 | ||
5536 | process_leaf: | |
5537 | leaf = path->nodes[0]; | |
5538 | nritems = btrfs_header_nritems(leaf); | |
5539 | for (i = path->slots[0]; i < nritems; i++) { | |
5540 | struct btrfs_dir_item *di; | |
5541 | struct btrfs_key di_key; | |
5542 | struct inode *di_inode; | |
5543 | struct btrfs_dir_list *new_dir_elem; | |
5544 | int log_mode = LOG_INODE_EXISTS; | |
5545 | int type; | |
5546 | ||
5547 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5548 | if (min_key.objectid != dir_elem->ino || | |
5549 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5550 | goto next_dir_inode; | |
5551 | ||
5552 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5553 | type = btrfs_dir_type(leaf, di); | |
5554 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5555 | type != BTRFS_FT_DIR) | |
5556 | continue; | |
5557 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5558 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5559 | continue; | |
5560 | ||
ec125cfb | 5561 | btrfs_release_path(path); |
0202e83f | 5562 | di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root); |
2f2ff0ee FM |
5563 | if (IS_ERR(di_inode)) { |
5564 | ret = PTR_ERR(di_inode); | |
5565 | goto next_dir_inode; | |
5566 | } | |
5567 | ||
0f8939b8 | 5568 | if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) { |
410f954c | 5569 | btrfs_add_delayed_iput(di_inode); |
ec125cfb | 5570 | break; |
2f2ff0ee FM |
5571 | } |
5572 | ||
5573 | ctx->log_new_dentries = false; | |
3f9749f6 | 5574 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) |
2f2ff0ee | 5575 | log_mode = LOG_INODE_ALL; |
a59108a7 | 5576 | ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode), |
2f2ff0ee | 5577 | log_mode, 0, LLONG_MAX, ctx); |
2be63d5c | 5578 | if (!ret && |
ab1717b2 | 5579 | btrfs_must_commit_transaction(trans, BTRFS_I(di_inode))) |
2be63d5c | 5580 | ret = 1; |
410f954c | 5581 | btrfs_add_delayed_iput(di_inode); |
2f2ff0ee FM |
5582 | if (ret) |
5583 | goto next_dir_inode; | |
5584 | if (ctx->log_new_dentries) { | |
5585 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5586 | GFP_NOFS); | |
5587 | if (!new_dir_elem) { | |
5588 | ret = -ENOMEM; | |
5589 | goto next_dir_inode; | |
5590 | } | |
5591 | new_dir_elem->ino = di_key.objectid; | |
5592 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5593 | } | |
5594 | break; | |
5595 | } | |
5596 | if (i == nritems) { | |
5597 | ret = btrfs_next_leaf(log, path); | |
5598 | if (ret < 0) { | |
5599 | goto next_dir_inode; | |
5600 | } else if (ret > 0) { | |
5601 | ret = 0; | |
5602 | goto next_dir_inode; | |
5603 | } | |
5604 | goto process_leaf; | |
5605 | } | |
5606 | if (min_key.offset < (u64)-1) { | |
5607 | min_key.offset++; | |
5608 | goto again; | |
5609 | } | |
5610 | next_dir_inode: | |
5611 | list_del(&dir_elem->list); | |
5612 | kfree(dir_elem); | |
5613 | } | |
5614 | ||
5615 | btrfs_free_path(path); | |
5616 | return ret; | |
5617 | } | |
5618 | ||
18aa0922 | 5619 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
d0a0b78d | 5620 | struct btrfs_inode *inode, |
18aa0922 FM |
5621 | struct btrfs_log_ctx *ctx) |
5622 | { | |
3ffbd68c | 5623 | struct btrfs_fs_info *fs_info = trans->fs_info; |
18aa0922 FM |
5624 | int ret; |
5625 | struct btrfs_path *path; | |
5626 | struct btrfs_key key; | |
d0a0b78d NB |
5627 | struct btrfs_root *root = inode->root; |
5628 | const u64 ino = btrfs_ino(inode); | |
18aa0922 FM |
5629 | |
5630 | path = btrfs_alloc_path(); | |
5631 | if (!path) | |
5632 | return -ENOMEM; | |
5633 | path->skip_locking = 1; | |
5634 | path->search_commit_root = 1; | |
5635 | ||
5636 | key.objectid = ino; | |
5637 | key.type = BTRFS_INODE_REF_KEY; | |
5638 | key.offset = 0; | |
5639 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5640 | if (ret < 0) | |
5641 | goto out; | |
5642 | ||
5643 | while (true) { | |
5644 | struct extent_buffer *leaf = path->nodes[0]; | |
5645 | int slot = path->slots[0]; | |
5646 | u32 cur_offset = 0; | |
5647 | u32 item_size; | |
5648 | unsigned long ptr; | |
5649 | ||
5650 | if (slot >= btrfs_header_nritems(leaf)) { | |
5651 | ret = btrfs_next_leaf(root, path); | |
5652 | if (ret < 0) | |
5653 | goto out; | |
5654 | else if (ret > 0) | |
5655 | break; | |
5656 | continue; | |
5657 | } | |
5658 | ||
5659 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5660 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5661 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5662 | break; | |
5663 | ||
5664 | item_size = btrfs_item_size_nr(leaf, slot); | |
5665 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5666 | while (cur_offset < item_size) { | |
5667 | struct btrfs_key inode_key; | |
5668 | struct inode *dir_inode; | |
5669 | ||
5670 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5671 | inode_key.offset = 0; | |
5672 | ||
5673 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5674 | struct btrfs_inode_extref *extref; | |
5675 | ||
5676 | extref = (struct btrfs_inode_extref *) | |
5677 | (ptr + cur_offset); | |
5678 | inode_key.objectid = btrfs_inode_extref_parent( | |
5679 | leaf, extref); | |
5680 | cur_offset += sizeof(*extref); | |
5681 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5682 | extref); | |
5683 | } else { | |
5684 | inode_key.objectid = key.offset; | |
5685 | cur_offset = item_size; | |
5686 | } | |
5687 | ||
0202e83f DS |
5688 | dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid, |
5689 | root); | |
0f375eed FM |
5690 | /* |
5691 | * If the parent inode was deleted, return an error to | |
5692 | * fallback to a transaction commit. This is to prevent | |
5693 | * getting an inode that was moved from one parent A to | |
5694 | * a parent B, got its former parent A deleted and then | |
5695 | * it got fsync'ed, from existing at both parents after | |
5696 | * a log replay (and the old parent still existing). | |
5697 | * Example: | |
5698 | * | |
5699 | * mkdir /mnt/A | |
5700 | * mkdir /mnt/B | |
5701 | * touch /mnt/B/bar | |
5702 | * sync | |
5703 | * mv /mnt/B/bar /mnt/A/bar | |
5704 | * mv -T /mnt/A /mnt/B | |
5705 | * fsync /mnt/B/bar | |
5706 | * <power fail> | |
5707 | * | |
5708 | * If we ignore the old parent B which got deleted, | |
5709 | * after a log replay we would have file bar linked | |
5710 | * at both parents and the old parent B would still | |
5711 | * exist. | |
5712 | */ | |
5713 | if (IS_ERR(dir_inode)) { | |
5714 | ret = PTR_ERR(dir_inode); | |
5715 | goto out; | |
5716 | } | |
18aa0922 | 5717 | |
657ed1aa FM |
5718 | if (ctx) |
5719 | ctx->log_new_dentries = false; | |
a59108a7 | 5720 | ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode), |
18aa0922 | 5721 | LOG_INODE_ALL, 0, LLONG_MAX, ctx); |
2be63d5c | 5722 | if (!ret && |
ab1717b2 | 5723 | btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode))) |
2be63d5c | 5724 | ret = 1; |
657ed1aa FM |
5725 | if (!ret && ctx && ctx->log_new_dentries) |
5726 | ret = log_new_dir_dentries(trans, root, | |
f85b7379 | 5727 | BTRFS_I(dir_inode), ctx); |
410f954c | 5728 | btrfs_add_delayed_iput(dir_inode); |
18aa0922 FM |
5729 | if (ret) |
5730 | goto out; | |
5731 | } | |
5732 | path->slots[0]++; | |
5733 | } | |
5734 | ret = 0; | |
5735 | out: | |
5736 | btrfs_free_path(path); | |
5737 | return ret; | |
5738 | } | |
5739 | ||
b8aa330d FM |
5740 | static int log_new_ancestors(struct btrfs_trans_handle *trans, |
5741 | struct btrfs_root *root, | |
5742 | struct btrfs_path *path, | |
5743 | struct btrfs_log_ctx *ctx) | |
5744 | { | |
5745 | struct btrfs_key found_key; | |
5746 | ||
5747 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
5748 | ||
5749 | while (true) { | |
5750 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5751 | const u64 last_committed = fs_info->last_trans_committed; | |
5752 | struct extent_buffer *leaf = path->nodes[0]; | |
5753 | int slot = path->slots[0]; | |
5754 | struct btrfs_key search_key; | |
5755 | struct inode *inode; | |
0202e83f | 5756 | u64 ino; |
b8aa330d FM |
5757 | int ret = 0; |
5758 | ||
5759 | btrfs_release_path(path); | |
5760 | ||
0202e83f DS |
5761 | ino = found_key.offset; |
5762 | ||
b8aa330d FM |
5763 | search_key.objectid = found_key.offset; |
5764 | search_key.type = BTRFS_INODE_ITEM_KEY; | |
5765 | search_key.offset = 0; | |
0202e83f | 5766 | inode = btrfs_iget(fs_info->sb, ino, root); |
b8aa330d FM |
5767 | if (IS_ERR(inode)) |
5768 | return PTR_ERR(inode); | |
5769 | ||
5770 | if (BTRFS_I(inode)->generation > last_committed) | |
5771 | ret = btrfs_log_inode(trans, root, BTRFS_I(inode), | |
5772 | LOG_INODE_EXISTS, | |
5773 | 0, LLONG_MAX, ctx); | |
410f954c | 5774 | btrfs_add_delayed_iput(inode); |
b8aa330d FM |
5775 | if (ret) |
5776 | return ret; | |
5777 | ||
5778 | if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID) | |
5779 | break; | |
5780 | ||
5781 | search_key.type = BTRFS_INODE_REF_KEY; | |
5782 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
5783 | if (ret < 0) | |
5784 | return ret; | |
5785 | ||
5786 | leaf = path->nodes[0]; | |
5787 | slot = path->slots[0]; | |
5788 | if (slot >= btrfs_header_nritems(leaf)) { | |
5789 | ret = btrfs_next_leaf(root, path); | |
5790 | if (ret < 0) | |
5791 | return ret; | |
5792 | else if (ret > 0) | |
5793 | return -ENOENT; | |
5794 | leaf = path->nodes[0]; | |
5795 | slot = path->slots[0]; | |
5796 | } | |
5797 | ||
5798 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5799 | if (found_key.objectid != search_key.objectid || | |
5800 | found_key.type != BTRFS_INODE_REF_KEY) | |
5801 | return -ENOENT; | |
5802 | } | |
5803 | return 0; | |
5804 | } | |
5805 | ||
5806 | static int log_new_ancestors_fast(struct btrfs_trans_handle *trans, | |
5807 | struct btrfs_inode *inode, | |
5808 | struct dentry *parent, | |
5809 | struct btrfs_log_ctx *ctx) | |
5810 | { | |
5811 | struct btrfs_root *root = inode->root; | |
5812 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5813 | struct dentry *old_parent = NULL; | |
5814 | struct super_block *sb = inode->vfs_inode.i_sb; | |
5815 | int ret = 0; | |
5816 | ||
5817 | while (true) { | |
5818 | if (!parent || d_really_is_negative(parent) || | |
5819 | sb != parent->d_sb) | |
5820 | break; | |
5821 | ||
5822 | inode = BTRFS_I(d_inode(parent)); | |
5823 | if (root != inode->root) | |
5824 | break; | |
5825 | ||
5826 | if (inode->generation > fs_info->last_trans_committed) { | |
5827 | ret = btrfs_log_inode(trans, root, inode, | |
5828 | LOG_INODE_EXISTS, 0, LLONG_MAX, ctx); | |
5829 | if (ret) | |
5830 | break; | |
5831 | } | |
5832 | if (IS_ROOT(parent)) | |
5833 | break; | |
5834 | ||
5835 | parent = dget_parent(parent); | |
5836 | dput(old_parent); | |
5837 | old_parent = parent; | |
5838 | } | |
5839 | dput(old_parent); | |
5840 | ||
5841 | return ret; | |
5842 | } | |
5843 | ||
5844 | static int log_all_new_ancestors(struct btrfs_trans_handle *trans, | |
5845 | struct btrfs_inode *inode, | |
5846 | struct dentry *parent, | |
5847 | struct btrfs_log_ctx *ctx) | |
5848 | { | |
5849 | struct btrfs_root *root = inode->root; | |
5850 | const u64 ino = btrfs_ino(inode); | |
5851 | struct btrfs_path *path; | |
5852 | struct btrfs_key search_key; | |
5853 | int ret; | |
5854 | ||
5855 | /* | |
5856 | * For a single hard link case, go through a fast path that does not | |
5857 | * need to iterate the fs/subvolume tree. | |
5858 | */ | |
5859 | if (inode->vfs_inode.i_nlink < 2) | |
5860 | return log_new_ancestors_fast(trans, inode, parent, ctx); | |
5861 | ||
5862 | path = btrfs_alloc_path(); | |
5863 | if (!path) | |
5864 | return -ENOMEM; | |
5865 | ||
5866 | search_key.objectid = ino; | |
5867 | search_key.type = BTRFS_INODE_REF_KEY; | |
5868 | search_key.offset = 0; | |
5869 | again: | |
5870 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
5871 | if (ret < 0) | |
5872 | goto out; | |
5873 | if (ret == 0) | |
5874 | path->slots[0]++; | |
5875 | ||
5876 | while (true) { | |
5877 | struct extent_buffer *leaf = path->nodes[0]; | |
5878 | int slot = path->slots[0]; | |
5879 | struct btrfs_key found_key; | |
5880 | ||
5881 | if (slot >= btrfs_header_nritems(leaf)) { | |
5882 | ret = btrfs_next_leaf(root, path); | |
5883 | if (ret < 0) | |
5884 | goto out; | |
5885 | else if (ret > 0) | |
5886 | break; | |
5887 | continue; | |
5888 | } | |
5889 | ||
5890 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5891 | if (found_key.objectid != ino || | |
5892 | found_key.type > BTRFS_INODE_EXTREF_KEY) | |
5893 | break; | |
5894 | ||
5895 | /* | |
5896 | * Don't deal with extended references because they are rare | |
5897 | * cases and too complex to deal with (we would need to keep | |
5898 | * track of which subitem we are processing for each item in | |
5899 | * this loop, etc). So just return some error to fallback to | |
5900 | * a transaction commit. | |
5901 | */ | |
5902 | if (found_key.type == BTRFS_INODE_EXTREF_KEY) { | |
5903 | ret = -EMLINK; | |
5904 | goto out; | |
5905 | } | |
5906 | ||
5907 | /* | |
5908 | * Logging ancestors needs to do more searches on the fs/subvol | |
5909 | * tree, so it releases the path as needed to avoid deadlocks. | |
5910 | * Keep track of the last inode ref key and resume from that key | |
5911 | * after logging all new ancestors for the current hard link. | |
5912 | */ | |
5913 | memcpy(&search_key, &found_key, sizeof(search_key)); | |
5914 | ||
5915 | ret = log_new_ancestors(trans, root, path, ctx); | |
5916 | if (ret) | |
5917 | goto out; | |
5918 | btrfs_release_path(path); | |
5919 | goto again; | |
5920 | } | |
5921 | ret = 0; | |
5922 | out: | |
5923 | btrfs_free_path(path); | |
5924 | return ret; | |
5925 | } | |
5926 | ||
e02119d5 CM |
5927 | /* |
5928 | * helper function around btrfs_log_inode to make sure newly created | |
5929 | * parent directories also end up in the log. A minimal inode and backref | |
5930 | * only logging is done of any parent directories that are older than | |
5931 | * the last committed transaction | |
5932 | */ | |
48a3b636 | 5933 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
19df27a9 | 5934 | struct btrfs_inode *inode, |
49dae1bc FM |
5935 | struct dentry *parent, |
5936 | const loff_t start, | |
5937 | const loff_t end, | |
41a1eada | 5938 | int inode_only, |
8b050d35 | 5939 | struct btrfs_log_ctx *ctx) |
e02119d5 | 5940 | { |
f882274b | 5941 | struct btrfs_root *root = inode->root; |
0b246afa | 5942 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 5943 | struct super_block *sb; |
12fcfd22 | 5944 | int ret = 0; |
0b246afa | 5945 | u64 last_committed = fs_info->last_trans_committed; |
2f2ff0ee | 5946 | bool log_dentries = false; |
12fcfd22 | 5947 | |
19df27a9 | 5948 | sb = inode->vfs_inode.i_sb; |
12fcfd22 | 5949 | |
0b246afa | 5950 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
3a5e1404 SW |
5951 | ret = 1; |
5952 | goto end_no_trans; | |
5953 | } | |
5954 | ||
995946dd MX |
5955 | /* |
5956 | * The prev transaction commit doesn't complete, we need do | |
5957 | * full commit by ourselves. | |
5958 | */ | |
0b246afa JM |
5959 | if (fs_info->last_trans_log_full_commit > |
5960 | fs_info->last_trans_committed) { | |
12fcfd22 CM |
5961 | ret = 1; |
5962 | goto end_no_trans; | |
5963 | } | |
5964 | ||
f882274b | 5965 | if (btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5966 | ret = 1; |
5967 | goto end_no_trans; | |
5968 | } | |
5969 | ||
19df27a9 NB |
5970 | ret = check_parent_dirs_for_sync(trans, inode, parent, sb, |
5971 | last_committed); | |
12fcfd22 CM |
5972 | if (ret) |
5973 | goto end_no_trans; | |
e02119d5 | 5974 | |
f2d72f42 FM |
5975 | /* |
5976 | * Skip already logged inodes or inodes corresponding to tmpfiles | |
5977 | * (since logging them is pointless, a link count of 0 means they | |
5978 | * will never be accessible). | |
5979 | */ | |
5980 | if (btrfs_inode_in_log(inode, trans->transid) || | |
5981 | inode->vfs_inode.i_nlink == 0) { | |
257c62e1 CM |
5982 | ret = BTRFS_NO_LOG_SYNC; |
5983 | goto end_no_trans; | |
5984 | } | |
5985 | ||
8b050d35 | 5986 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 5987 | if (ret) |
e87ac136 | 5988 | goto end_no_trans; |
e02119d5 | 5989 | |
19df27a9 | 5990 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); |
4a500fd1 YZ |
5991 | if (ret) |
5992 | goto end_trans; | |
12fcfd22 | 5993 | |
af4176b4 CM |
5994 | /* |
5995 | * for regular files, if its inode is already on disk, we don't | |
5996 | * have to worry about the parents at all. This is because | |
5997 | * we can use the last_unlink_trans field to record renames | |
5998 | * and other fun in this file. | |
5999 | */ | |
19df27a9 NB |
6000 | if (S_ISREG(inode->vfs_inode.i_mode) && |
6001 | inode->generation <= last_committed && | |
6002 | inode->last_unlink_trans <= last_committed) { | |
4a500fd1 YZ |
6003 | ret = 0; |
6004 | goto end_trans; | |
6005 | } | |
af4176b4 | 6006 | |
19df27a9 | 6007 | if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries) |
2f2ff0ee FM |
6008 | log_dentries = true; |
6009 | ||
18aa0922 | 6010 | /* |
01327610 | 6011 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
6012 | * inodes are fully logged. This is to prevent leaving dangling |
6013 | * directory index entries in directories that were our parents but are | |
6014 | * not anymore. Not doing this results in old parent directory being | |
6015 | * impossible to delete after log replay (rmdir will always fail with | |
6016 | * error -ENOTEMPTY). | |
6017 | * | |
6018 | * Example 1: | |
6019 | * | |
6020 | * mkdir testdir | |
6021 | * touch testdir/foo | |
6022 | * ln testdir/foo testdir/bar | |
6023 | * sync | |
6024 | * unlink testdir/bar | |
6025 | * xfs_io -c fsync testdir/foo | |
6026 | * <power failure> | |
6027 | * mount fs, triggers log replay | |
6028 | * | |
6029 | * If we don't log the parent directory (testdir), after log replay the | |
6030 | * directory still has an entry pointing to the file inode using the bar | |
6031 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
6032 | * the file inode has a link count of 1. | |
6033 | * | |
6034 | * Example 2: | |
6035 | * | |
6036 | * mkdir testdir | |
6037 | * touch foo | |
6038 | * ln foo testdir/foo2 | |
6039 | * ln foo testdir/foo3 | |
6040 | * sync | |
6041 | * unlink testdir/foo3 | |
6042 | * xfs_io -c fsync foo | |
6043 | * <power failure> | |
6044 | * mount fs, triggers log replay | |
6045 | * | |
6046 | * Similar as the first example, after log replay the parent directory | |
6047 | * testdir still has an entry pointing to the inode file with name foo3 | |
6048 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
6049 | * and has a link count of 2. | |
6050 | */ | |
19df27a9 | 6051 | if (inode->last_unlink_trans > last_committed) { |
b8aa330d | 6052 | ret = btrfs_log_all_parents(trans, inode, ctx); |
18aa0922 FM |
6053 | if (ret) |
6054 | goto end_trans; | |
6055 | } | |
6056 | ||
b8aa330d FM |
6057 | ret = log_all_new_ancestors(trans, inode, parent, ctx); |
6058 | if (ret) | |
41bd6067 | 6059 | goto end_trans; |
76dda93c | 6060 | |
2f2ff0ee | 6061 | if (log_dentries) |
b8aa330d | 6062 | ret = log_new_dir_dentries(trans, root, inode, ctx); |
2f2ff0ee FM |
6063 | else |
6064 | ret = 0; | |
4a500fd1 YZ |
6065 | end_trans: |
6066 | if (ret < 0) { | |
90787766 | 6067 | btrfs_set_log_full_commit(trans); |
4a500fd1 YZ |
6068 | ret = 1; |
6069 | } | |
8b050d35 MX |
6070 | |
6071 | if (ret) | |
6072 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
6073 | btrfs_end_log_trans(root); |
6074 | end_no_trans: | |
6075 | return ret; | |
e02119d5 CM |
6076 | } |
6077 | ||
6078 | /* | |
6079 | * it is not safe to log dentry if the chunk root has added new | |
6080 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
6081 | * If this returns 1, you must commit the transaction to safely get your | |
6082 | * data on disk. | |
6083 | */ | |
6084 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
e5b84f7a | 6085 | struct dentry *dentry, |
49dae1bc FM |
6086 | const loff_t start, |
6087 | const loff_t end, | |
8b050d35 | 6088 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6089 | { |
6a912213 JB |
6090 | struct dentry *parent = dget_parent(dentry); |
6091 | int ret; | |
6092 | ||
f882274b NB |
6093 | ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent, |
6094 | start, end, LOG_INODE_ALL, ctx); | |
6a912213 JB |
6095 | dput(parent); |
6096 | ||
6097 | return ret; | |
e02119d5 CM |
6098 | } |
6099 | ||
6100 | /* | |
6101 | * should be called during mount to recover any replay any log trees | |
6102 | * from the FS | |
6103 | */ | |
6104 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
6105 | { | |
6106 | int ret; | |
6107 | struct btrfs_path *path; | |
6108 | struct btrfs_trans_handle *trans; | |
6109 | struct btrfs_key key; | |
6110 | struct btrfs_key found_key; | |
e02119d5 CM |
6111 | struct btrfs_root *log; |
6112 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
6113 | struct walk_control wc = { | |
6114 | .process_func = process_one_buffer, | |
430a6626 | 6115 | .stage = LOG_WALK_PIN_ONLY, |
e02119d5 CM |
6116 | }; |
6117 | ||
e02119d5 | 6118 | path = btrfs_alloc_path(); |
db5b493a TI |
6119 | if (!path) |
6120 | return -ENOMEM; | |
6121 | ||
afcdd129 | 6122 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 6123 | |
4a500fd1 | 6124 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
6125 | if (IS_ERR(trans)) { |
6126 | ret = PTR_ERR(trans); | |
6127 | goto error; | |
6128 | } | |
e02119d5 CM |
6129 | |
6130 | wc.trans = trans; | |
6131 | wc.pin = 1; | |
6132 | ||
db5b493a | 6133 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 6134 | if (ret) { |
5d163e0e JM |
6135 | btrfs_handle_fs_error(fs_info, ret, |
6136 | "Failed to pin buffers while recovering log root tree."); | |
79787eaa JM |
6137 | goto error; |
6138 | } | |
e02119d5 CM |
6139 | |
6140 | again: | |
6141 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
6142 | key.offset = (u64)-1; | |
962a298f | 6143 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 6144 | |
d397712b | 6145 | while (1) { |
e02119d5 | 6146 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
6147 | |
6148 | if (ret < 0) { | |
34d97007 | 6149 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6150 | "Couldn't find tree log root."); |
6151 | goto error; | |
6152 | } | |
e02119d5 CM |
6153 | if (ret > 0) { |
6154 | if (path->slots[0] == 0) | |
6155 | break; | |
6156 | path->slots[0]--; | |
6157 | } | |
6158 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
6159 | path->slots[0]); | |
b3b4aa74 | 6160 | btrfs_release_path(path); |
e02119d5 CM |
6161 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
6162 | break; | |
6163 | ||
62a2c73e | 6164 | log = btrfs_read_tree_root(log_root_tree, &found_key); |
79787eaa JM |
6165 | if (IS_ERR(log)) { |
6166 | ret = PTR_ERR(log); | |
34d97007 | 6167 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6168 | "Couldn't read tree log root."); |
6169 | goto error; | |
6170 | } | |
e02119d5 | 6171 | |
56e9357a DS |
6172 | wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset, |
6173 | true); | |
79787eaa JM |
6174 | if (IS_ERR(wc.replay_dest)) { |
6175 | ret = PTR_ERR(wc.replay_dest); | |
9bc574de JB |
6176 | |
6177 | /* | |
6178 | * We didn't find the subvol, likely because it was | |
6179 | * deleted. This is ok, simply skip this log and go to | |
6180 | * the next one. | |
6181 | * | |
6182 | * We need to exclude the root because we can't have | |
6183 | * other log replays overwriting this log as we'll read | |
6184 | * it back in a few more times. This will keep our | |
6185 | * block from being modified, and we'll just bail for | |
6186 | * each subsequent pass. | |
6187 | */ | |
6188 | if (ret == -ENOENT) | |
9fce5704 | 6189 | ret = btrfs_pin_extent_for_log_replay(trans, |
9bc574de JB |
6190 | log->node->start, |
6191 | log->node->len); | |
00246528 | 6192 | btrfs_put_root(log); |
9bc574de JB |
6193 | |
6194 | if (!ret) | |
6195 | goto next; | |
5d163e0e JM |
6196 | btrfs_handle_fs_error(fs_info, ret, |
6197 | "Couldn't read target root for tree log recovery."); | |
79787eaa JM |
6198 | goto error; |
6199 | } | |
e02119d5 | 6200 | |
07d400a6 | 6201 | wc.replay_dest->log_root = log; |
5d4f98a2 | 6202 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 6203 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 6204 | |
b50c6e25 | 6205 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
6206 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
6207 | path); | |
e02119d5 CM |
6208 | } |
6209 | ||
900c9981 LB |
6210 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
6211 | struct btrfs_root *root = wc.replay_dest; | |
6212 | ||
6213 | btrfs_release_path(path); | |
6214 | ||
6215 | /* | |
6216 | * We have just replayed everything, and the highest | |
6217 | * objectid of fs roots probably has changed in case | |
6218 | * some inode_item's got replayed. | |
6219 | * | |
6220 | * root->objectid_mutex is not acquired as log replay | |
6221 | * could only happen during mount. | |
6222 | */ | |
6223 | ret = btrfs_find_highest_objectid(root, | |
6224 | &root->highest_objectid); | |
6225 | } | |
6226 | ||
07d400a6 | 6227 | wc.replay_dest->log_root = NULL; |
00246528 | 6228 | btrfs_put_root(wc.replay_dest); |
00246528 | 6229 | btrfs_put_root(log); |
e02119d5 | 6230 | |
b50c6e25 JB |
6231 | if (ret) |
6232 | goto error; | |
9bc574de | 6233 | next: |
e02119d5 CM |
6234 | if (found_key.offset == 0) |
6235 | break; | |
9bc574de | 6236 | key.offset = found_key.offset - 1; |
e02119d5 | 6237 | } |
b3b4aa74 | 6238 | btrfs_release_path(path); |
e02119d5 CM |
6239 | |
6240 | /* step one is to pin it all, step two is to replay just inodes */ | |
6241 | if (wc.pin) { | |
6242 | wc.pin = 0; | |
6243 | wc.process_func = replay_one_buffer; | |
6244 | wc.stage = LOG_WALK_REPLAY_INODES; | |
6245 | goto again; | |
6246 | } | |
6247 | /* step three is to replay everything */ | |
6248 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
6249 | wc.stage++; | |
6250 | goto again; | |
6251 | } | |
6252 | ||
6253 | btrfs_free_path(path); | |
6254 | ||
abefa55a | 6255 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 6256 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
6257 | if (ret) |
6258 | return ret; | |
6259 | ||
e02119d5 | 6260 | log_root_tree->log_root = NULL; |
afcdd129 | 6261 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
00246528 | 6262 | btrfs_put_root(log_root_tree); |
79787eaa | 6263 | |
abefa55a | 6264 | return 0; |
79787eaa | 6265 | error: |
b50c6e25 | 6266 | if (wc.trans) |
3a45bb20 | 6267 | btrfs_end_transaction(wc.trans); |
79787eaa JM |
6268 | btrfs_free_path(path); |
6269 | return ret; | |
e02119d5 | 6270 | } |
12fcfd22 CM |
6271 | |
6272 | /* | |
6273 | * there are some corner cases where we want to force a full | |
6274 | * commit instead of allowing a directory to be logged. | |
6275 | * | |
6276 | * They revolve around files there were unlinked from the directory, and | |
6277 | * this function updates the parent directory so that a full commit is | |
6278 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
6279 | * |
6280 | * Must be called before the unlink operations (updates to the subvolume tree, | |
6281 | * inodes, etc) are done. | |
12fcfd22 CM |
6282 | */ |
6283 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 6284 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
6285 | int for_rename) |
6286 | { | |
af4176b4 CM |
6287 | /* |
6288 | * when we're logging a file, if it hasn't been renamed | |
6289 | * or unlinked, and its inode is fully committed on disk, | |
6290 | * we don't have to worry about walking up the directory chain | |
6291 | * to log its parents. | |
6292 | * | |
6293 | * So, we use the last_unlink_trans field to put this transid | |
6294 | * into the file. When the file is logged we check it and | |
6295 | * don't log the parents if the file is fully on disk. | |
6296 | */ | |
4176bdbf NB |
6297 | mutex_lock(&inode->log_mutex); |
6298 | inode->last_unlink_trans = trans->transid; | |
6299 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 6300 | |
12fcfd22 CM |
6301 | /* |
6302 | * if this directory was already logged any new | |
6303 | * names for this file/dir will get recorded | |
6304 | */ | |
4176bdbf | 6305 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
6306 | return; |
6307 | ||
6308 | /* | |
6309 | * if the inode we're about to unlink was logged, | |
6310 | * the log will be properly updated for any new names | |
6311 | */ | |
4176bdbf | 6312 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
6313 | return; |
6314 | ||
6315 | /* | |
6316 | * when renaming files across directories, if the directory | |
6317 | * there we're unlinking from gets fsync'd later on, there's | |
6318 | * no way to find the destination directory later and fsync it | |
6319 | * properly. So, we have to be conservative and force commits | |
6320 | * so the new name gets discovered. | |
6321 | */ | |
6322 | if (for_rename) | |
6323 | goto record; | |
6324 | ||
6325 | /* we can safely do the unlink without any special recording */ | |
6326 | return; | |
6327 | ||
6328 | record: | |
4176bdbf NB |
6329 | mutex_lock(&dir->log_mutex); |
6330 | dir->last_unlink_trans = trans->transid; | |
6331 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
6332 | } |
6333 | ||
6334 | /* | |
6335 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
6336 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
6337 | * that after replaying the log tree of the parent directory's root we will not | |
6338 | * see the snapshot anymore and at log replay time we will not see any log tree | |
6339 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
6340 | * it after replaying the log tree of the parent directory (which would replay | |
6341 | * the snapshot delete operation). | |
2be63d5c FM |
6342 | * |
6343 | * Must be called before the actual snapshot destroy operation (updates to the | |
6344 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
6345 | */ |
6346 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 6347 | struct btrfs_inode *dir) |
1ec9a1ae | 6348 | { |
43663557 NB |
6349 | mutex_lock(&dir->log_mutex); |
6350 | dir->last_unlink_trans = trans->transid; | |
6351 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
6352 | } |
6353 | ||
6354 | /* | |
6355 | * Call this after adding a new name for a file and it will properly | |
6356 | * update the log to reflect the new name. | |
6357 | * | |
d4682ba0 FM |
6358 | * @ctx can not be NULL when @sync_log is false, and should be NULL when it's |
6359 | * true (because it's not used). | |
6360 | * | |
6361 | * Return value depends on whether @sync_log is true or false. | |
6362 | * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be | |
6363 | * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT | |
6364 | * otherwise. | |
6365 | * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to | |
6366 | * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log, | |
6367 | * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be | |
6368 | * committed (without attempting to sync the log). | |
12fcfd22 CM |
6369 | */ |
6370 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
9ca5fbfb | 6371 | struct btrfs_inode *inode, struct btrfs_inode *old_dir, |
d4682ba0 FM |
6372 | struct dentry *parent, |
6373 | bool sync_log, struct btrfs_log_ctx *ctx) | |
12fcfd22 | 6374 | { |
3ffbd68c | 6375 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d4682ba0 | 6376 | int ret; |
12fcfd22 | 6377 | |
af4176b4 CM |
6378 | /* |
6379 | * this will force the logging code to walk the dentry chain | |
6380 | * up for the file | |
6381 | */ | |
9a6509c4 | 6382 | if (!S_ISDIR(inode->vfs_inode.i_mode)) |
9ca5fbfb | 6383 | inode->last_unlink_trans = trans->transid; |
af4176b4 | 6384 | |
12fcfd22 CM |
6385 | /* |
6386 | * if this inode hasn't been logged and directory we're renaming it | |
6387 | * from hasn't been logged, we don't need to log it | |
6388 | */ | |
9ca5fbfb NB |
6389 | if (inode->logged_trans <= fs_info->last_trans_committed && |
6390 | (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed)) | |
d4682ba0 FM |
6391 | return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT : |
6392 | BTRFS_DONT_NEED_LOG_SYNC; | |
6393 | ||
6394 | if (sync_log) { | |
6395 | struct btrfs_log_ctx ctx2; | |
6396 | ||
6397 | btrfs_init_log_ctx(&ctx2, &inode->vfs_inode); | |
6398 | ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX, | |
6399 | LOG_INODE_EXISTS, &ctx2); | |
6400 | if (ret == BTRFS_NO_LOG_SYNC) | |
6401 | return BTRFS_DONT_NEED_TRANS_COMMIT; | |
6402 | else if (ret) | |
6403 | return BTRFS_NEED_TRANS_COMMIT; | |
6404 | ||
6405 | ret = btrfs_sync_log(trans, inode->root, &ctx2); | |
6406 | if (ret) | |
6407 | return BTRFS_NEED_TRANS_COMMIT; | |
6408 | return BTRFS_DONT_NEED_TRANS_COMMIT; | |
6409 | } | |
6410 | ||
6411 | ASSERT(ctx); | |
6412 | ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX, | |
6413 | LOG_INODE_EXISTS, ctx); | |
6414 | if (ret == BTRFS_NO_LOG_SYNC) | |
6415 | return BTRFS_DONT_NEED_LOG_SYNC; | |
6416 | else if (ret) | |
6417 | return BTRFS_NEED_TRANS_COMMIT; | |
12fcfd22 | 6418 | |
d4682ba0 | 6419 | return BTRFS_NEED_LOG_SYNC; |
12fcfd22 CM |
6420 | } |
6421 |