]>
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> |
9678c543 | 11 | #include "ctree.h" |
995946dd | 12 | #include "tree-log.h" |
e02119d5 CM |
13 | #include "disk-io.h" |
14 | #include "locking.h" | |
15 | #include "print-tree.h" | |
f186373f | 16 | #include "backref.h" |
ebb8765b | 17 | #include "compression.h" |
df2c95f3 | 18 | #include "qgroup.h" |
900c9981 | 19 | #include "inode-map.h" |
e02119d5 CM |
20 | |
21 | /* magic values for the inode_only field in btrfs_log_inode: | |
22 | * | |
23 | * LOG_INODE_ALL means to log everything | |
24 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
25 | * during log replay | |
26 | */ | |
27 | #define LOG_INODE_ALL 0 | |
28 | #define LOG_INODE_EXISTS 1 | |
781feef7 | 29 | #define LOG_OTHER_INODE 2 |
a3baaf0d | 30 | #define LOG_OTHER_INODE_ALL 3 |
e02119d5 | 31 | |
12fcfd22 CM |
32 | /* |
33 | * directory trouble cases | |
34 | * | |
35 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
36 | * log, we must force a full commit before doing an fsync of the directory | |
37 | * where the unlink was done. | |
38 | * ---> record transid of last unlink/rename per directory | |
39 | * | |
40 | * mkdir foo/some_dir | |
41 | * normal commit | |
42 | * rename foo/some_dir foo2/some_dir | |
43 | * mkdir foo/some_dir | |
44 | * fsync foo/some_dir/some_file | |
45 | * | |
46 | * The fsync above will unlink the original some_dir without recording | |
47 | * it in its new location (foo2). After a crash, some_dir will be gone | |
48 | * unless the fsync of some_file forces a full commit | |
49 | * | |
50 | * 2) we must log any new names for any file or dir that is in the fsync | |
51 | * log. ---> check inode while renaming/linking. | |
52 | * | |
53 | * 2a) we must log any new names for any file or dir during rename | |
54 | * when the directory they are being removed from was logged. | |
55 | * ---> check inode and old parent dir during rename | |
56 | * | |
57 | * 2a is actually the more important variant. With the extra logging | |
58 | * a crash might unlink the old name without recreating the new one | |
59 | * | |
60 | * 3) after a crash, we must go through any directories with a link count | |
61 | * of zero and redo the rm -rf | |
62 | * | |
63 | * mkdir f1/foo | |
64 | * normal commit | |
65 | * rm -rf f1/foo | |
66 | * fsync(f1) | |
67 | * | |
68 | * The directory f1 was fully removed from the FS, but fsync was never | |
69 | * called on f1, only its parent dir. After a crash the rm -rf must | |
70 | * be replayed. This must be able to recurse down the entire | |
71 | * directory tree. The inode link count fixup code takes care of the | |
72 | * ugly details. | |
73 | */ | |
74 | ||
e02119d5 CM |
75 | /* |
76 | * stages for the tree walking. The first | |
77 | * stage (0) is to only pin down the blocks we find | |
78 | * the second stage (1) is to make sure that all the inodes | |
79 | * we find in the log are created in the subvolume. | |
80 | * | |
81 | * The last stage is to deal with directories and links and extents | |
82 | * and all the other fun semantics | |
83 | */ | |
84 | #define LOG_WALK_PIN_ONLY 0 | |
85 | #define LOG_WALK_REPLAY_INODES 1 | |
dd8e7217 JB |
86 | #define LOG_WALK_REPLAY_DIR_INDEX 2 |
87 | #define LOG_WALK_REPLAY_ALL 3 | |
e02119d5 | 88 | |
12fcfd22 | 89 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 90 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc FM |
91 | int inode_only, |
92 | const loff_t start, | |
8407f553 FM |
93 | const loff_t end, |
94 | struct btrfs_log_ctx *ctx); | |
ec051c0f YZ |
95 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
96 | struct btrfs_root *root, | |
97 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
98 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
99 | struct btrfs_root *root, | |
100 | struct btrfs_root *log, | |
101 | struct btrfs_path *path, | |
102 | u64 dirid, int del_all); | |
e02119d5 CM |
103 | |
104 | /* | |
105 | * tree logging is a special write ahead log used to make sure that | |
106 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
107 | * | |
108 | * Full tree commits are expensive because they require commonly | |
109 | * modified blocks to be recowed, creating many dirty pages in the | |
110 | * extent tree an 4x-6x higher write load than ext3. | |
111 | * | |
112 | * Instead of doing a tree commit on every fsync, we use the | |
113 | * key ranges and transaction ids to find items for a given file or directory | |
114 | * that have changed in this transaction. Those items are copied into | |
115 | * a special tree (one per subvolume root), that tree is written to disk | |
116 | * and then the fsync is considered complete. | |
117 | * | |
118 | * After a crash, items are copied out of the log-tree back into the | |
119 | * subvolume tree. Any file data extents found are recorded in the extent | |
120 | * allocation tree, and the log-tree freed. | |
121 | * | |
122 | * The log tree is read three times, once to pin down all the extents it is | |
123 | * using in ram and once, once to create all the inodes logged in the tree | |
124 | * and once to do all the other items. | |
125 | */ | |
126 | ||
e02119d5 CM |
127 | /* |
128 | * start a sub transaction and setup the log tree | |
129 | * this increments the log tree writer count to make the people | |
130 | * syncing the tree wait for us to finish | |
131 | */ | |
132 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
8b050d35 MX |
133 | struct btrfs_root *root, |
134 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 135 | { |
0b246afa | 136 | struct btrfs_fs_info *fs_info = root->fs_info; |
34eb2a52 | 137 | int ret = 0; |
7237f183 YZ |
138 | |
139 | mutex_lock(&root->log_mutex); | |
34eb2a52 | 140 | |
7237f183 | 141 | if (root->log_root) { |
4884b8e8 | 142 | if (btrfs_need_log_full_commit(trans)) { |
50471a38 MX |
143 | ret = -EAGAIN; |
144 | goto out; | |
145 | } | |
34eb2a52 | 146 | |
ff782e0a | 147 | if (!root->log_start_pid) { |
27cdeb70 | 148 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
34eb2a52 | 149 | root->log_start_pid = current->pid; |
ff782e0a | 150 | } else if (root->log_start_pid != current->pid) { |
27cdeb70 | 151 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); |
ff782e0a | 152 | } |
34eb2a52 | 153 | } else { |
0b246afa JM |
154 | mutex_lock(&fs_info->tree_log_mutex); |
155 | if (!fs_info->log_root_tree) | |
156 | ret = btrfs_init_log_root_tree(trans, fs_info); | |
157 | mutex_unlock(&fs_info->tree_log_mutex); | |
34eb2a52 Z |
158 | if (ret) |
159 | goto out; | |
ff782e0a | 160 | |
e02119d5 | 161 | ret = btrfs_add_log_tree(trans, root); |
4a500fd1 | 162 | if (ret) |
e87ac136 | 163 | goto out; |
34eb2a52 Z |
164 | |
165 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); | |
166 | root->log_start_pid = current->pid; | |
e02119d5 | 167 | } |
34eb2a52 | 168 | |
2ecb7923 | 169 | atomic_inc(&root->log_batch); |
7237f183 | 170 | atomic_inc(&root->log_writers); |
8b050d35 | 171 | if (ctx) { |
34eb2a52 | 172 | int index = root->log_transid % 2; |
8b050d35 | 173 | list_add_tail(&ctx->list, &root->log_ctxs[index]); |
d1433deb | 174 | ctx->log_transid = root->log_transid; |
8b050d35 | 175 | } |
34eb2a52 | 176 | |
e87ac136 | 177 | out: |
7237f183 | 178 | mutex_unlock(&root->log_mutex); |
e87ac136 | 179 | return ret; |
e02119d5 CM |
180 | } |
181 | ||
182 | /* | |
183 | * returns 0 if there was a log transaction running and we were able | |
184 | * to join, or returns -ENOENT if there were not transactions | |
185 | * in progress | |
186 | */ | |
187 | static int join_running_log_trans(struct btrfs_root *root) | |
188 | { | |
189 | int ret = -ENOENT; | |
190 | ||
191 | smp_mb(); | |
192 | if (!root->log_root) | |
193 | return -ENOENT; | |
194 | ||
7237f183 | 195 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
196 | if (root->log_root) { |
197 | ret = 0; | |
7237f183 | 198 | atomic_inc(&root->log_writers); |
e02119d5 | 199 | } |
7237f183 | 200 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
201 | return ret; |
202 | } | |
203 | ||
12fcfd22 CM |
204 | /* |
205 | * This either makes the current running log transaction wait | |
206 | * until you call btrfs_end_log_trans() or it makes any future | |
207 | * log transactions wait until you call btrfs_end_log_trans() | |
208 | */ | |
45128b08 | 209 | void btrfs_pin_log_trans(struct btrfs_root *root) |
12fcfd22 | 210 | { |
12fcfd22 CM |
211 | mutex_lock(&root->log_mutex); |
212 | atomic_inc(&root->log_writers); | |
213 | mutex_unlock(&root->log_mutex); | |
12fcfd22 CM |
214 | } |
215 | ||
e02119d5 CM |
216 | /* |
217 | * indicate we're done making changes to the log tree | |
218 | * and wake up anyone waiting to do a sync | |
219 | */ | |
143bede5 | 220 | void btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 221 | { |
7237f183 | 222 | if (atomic_dec_and_test(&root->log_writers)) { |
093258e6 DS |
223 | /* atomic_dec_and_test implies a barrier */ |
224 | cond_wake_up_nomb(&root->log_writer_wait); | |
7237f183 | 225 | } |
e02119d5 CM |
226 | } |
227 | ||
247462a5 DS |
228 | static int btrfs_write_tree_block(struct extent_buffer *buf) |
229 | { | |
230 | return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, | |
231 | buf->start + buf->len - 1); | |
232 | } | |
233 | ||
234 | static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf) | |
235 | { | |
236 | filemap_fdatawait_range(buf->pages[0]->mapping, | |
237 | buf->start, buf->start + buf->len - 1); | |
238 | } | |
e02119d5 CM |
239 | |
240 | /* | |
241 | * the walk control struct is used to pass state down the chain when | |
242 | * processing the log tree. The stage field tells us which part | |
243 | * of the log tree processing we are currently doing. The others | |
244 | * are state fields used for that specific part | |
245 | */ | |
246 | struct walk_control { | |
247 | /* should we free the extent on disk when done? This is used | |
248 | * at transaction commit time while freeing a log tree | |
249 | */ | |
250 | int free; | |
251 | ||
252 | /* should we write out the extent buffer? This is used | |
253 | * while flushing the log tree to disk during a sync | |
254 | */ | |
255 | int write; | |
256 | ||
257 | /* should we wait for the extent buffer io to finish? Also used | |
258 | * while flushing the log tree to disk for a sync | |
259 | */ | |
260 | int wait; | |
261 | ||
262 | /* pin only walk, we record which extents on disk belong to the | |
263 | * log trees | |
264 | */ | |
265 | int pin; | |
266 | ||
267 | /* what stage of the replay code we're currently in */ | |
268 | int stage; | |
269 | ||
f2d72f42 FM |
270 | /* |
271 | * Ignore any items from the inode currently being processed. Needs | |
272 | * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in | |
273 | * the LOG_WALK_REPLAY_INODES stage. | |
274 | */ | |
275 | bool ignore_cur_inode; | |
276 | ||
e02119d5 CM |
277 | /* the root we are currently replaying */ |
278 | struct btrfs_root *replay_dest; | |
279 | ||
280 | /* the trans handle for the current replay */ | |
281 | struct btrfs_trans_handle *trans; | |
282 | ||
283 | /* the function that gets used to process blocks we find in the | |
284 | * tree. Note the extent_buffer might not be up to date when it is | |
285 | * passed in, and it must be checked or read if you need the data | |
286 | * inside it | |
287 | */ | |
288 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 289 | struct walk_control *wc, u64 gen, int level); |
e02119d5 CM |
290 | }; |
291 | ||
292 | /* | |
293 | * process_func used to pin down extents, write them or wait on them | |
294 | */ | |
295 | static int process_one_buffer(struct btrfs_root *log, | |
296 | struct extent_buffer *eb, | |
581c1760 | 297 | struct walk_control *wc, u64 gen, int level) |
e02119d5 | 298 | { |
0b246afa | 299 | struct btrfs_fs_info *fs_info = log->fs_info; |
b50c6e25 JB |
300 | int ret = 0; |
301 | ||
8c2a1a30 JB |
302 | /* |
303 | * If this fs is mixed then we need to be able to process the leaves to | |
304 | * pin down any logged extents, so we have to read the block. | |
305 | */ | |
0b246afa | 306 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
581c1760 | 307 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
8c2a1a30 JB |
308 | if (ret) |
309 | return ret; | |
310 | } | |
311 | ||
04018de5 | 312 | if (wc->pin) |
2ff7e61e JM |
313 | ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start, |
314 | eb->len); | |
e02119d5 | 315 | |
b50c6e25 | 316 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { |
8c2a1a30 | 317 | if (wc->pin && btrfs_header_level(eb) == 0) |
bcdc428c | 318 | ret = btrfs_exclude_logged_extents(eb); |
e02119d5 CM |
319 | if (wc->write) |
320 | btrfs_write_tree_block(eb); | |
321 | if (wc->wait) | |
322 | btrfs_wait_tree_block_writeback(eb); | |
323 | } | |
b50c6e25 | 324 | return ret; |
e02119d5 CM |
325 | } |
326 | ||
327 | /* | |
328 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
329 | * to the src data we are copying out. | |
330 | * | |
331 | * root is the tree we are copying into, and path is a scratch | |
332 | * path for use in this function (it should be released on entry and | |
333 | * will be released on exit). | |
334 | * | |
335 | * If the key is already in the destination tree the existing item is | |
336 | * overwritten. If the existing item isn't big enough, it is extended. | |
337 | * If it is too large, it is truncated. | |
338 | * | |
339 | * If the key isn't in the destination yet, a new item is inserted. | |
340 | */ | |
341 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
342 | struct btrfs_root *root, | |
343 | struct btrfs_path *path, | |
344 | struct extent_buffer *eb, int slot, | |
345 | struct btrfs_key *key) | |
346 | { | |
347 | int ret; | |
348 | u32 item_size; | |
349 | u64 saved_i_size = 0; | |
350 | int save_old_i_size = 0; | |
351 | unsigned long src_ptr; | |
352 | unsigned long dst_ptr; | |
353 | int overwrite_root = 0; | |
4bc4bee4 | 354 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; |
e02119d5 CM |
355 | |
356 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
357 | overwrite_root = 1; | |
358 | ||
359 | item_size = btrfs_item_size_nr(eb, slot); | |
360 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
361 | ||
362 | /* look for the key in the destination tree */ | |
363 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
4bc4bee4 JB |
364 | if (ret < 0) |
365 | return ret; | |
366 | ||
e02119d5 CM |
367 | if (ret == 0) { |
368 | char *src_copy; | |
369 | char *dst_copy; | |
370 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
371 | path->slots[0]); | |
372 | if (dst_size != item_size) | |
373 | goto insert; | |
374 | ||
375 | if (item_size == 0) { | |
b3b4aa74 | 376 | btrfs_release_path(path); |
e02119d5 CM |
377 | return 0; |
378 | } | |
379 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
380 | src_copy = kmalloc(item_size, GFP_NOFS); | |
2a29edc6 | 381 | if (!dst_copy || !src_copy) { |
b3b4aa74 | 382 | btrfs_release_path(path); |
2a29edc6 | 383 | kfree(dst_copy); |
384 | kfree(src_copy); | |
385 | return -ENOMEM; | |
386 | } | |
e02119d5 CM |
387 | |
388 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
389 | ||
390 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
391 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
392 | item_size); | |
393 | ret = memcmp(dst_copy, src_copy, item_size); | |
394 | ||
395 | kfree(dst_copy); | |
396 | kfree(src_copy); | |
397 | /* | |
398 | * they have the same contents, just return, this saves | |
399 | * us from cowing blocks in the destination tree and doing | |
400 | * extra writes that may not have been done by a previous | |
401 | * sync | |
402 | */ | |
403 | if (ret == 0) { | |
b3b4aa74 | 404 | btrfs_release_path(path); |
e02119d5 CM |
405 | return 0; |
406 | } | |
407 | ||
4bc4bee4 JB |
408 | /* |
409 | * We need to load the old nbytes into the inode so when we | |
410 | * replay the extents we've logged we get the right nbytes. | |
411 | */ | |
412 | if (inode_item) { | |
413 | struct btrfs_inode_item *item; | |
414 | u64 nbytes; | |
d555438b | 415 | u32 mode; |
4bc4bee4 JB |
416 | |
417 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
418 | struct btrfs_inode_item); | |
419 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
420 | item = btrfs_item_ptr(eb, slot, | |
421 | struct btrfs_inode_item); | |
422 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
d555438b JB |
423 | |
424 | /* | |
425 | * If this is a directory we need to reset the i_size to | |
426 | * 0 so that we can set it up properly when replaying | |
427 | * the rest of the items in this log. | |
428 | */ | |
429 | mode = btrfs_inode_mode(eb, item); | |
430 | if (S_ISDIR(mode)) | |
431 | btrfs_set_inode_size(eb, item, 0); | |
4bc4bee4 JB |
432 | } |
433 | } else if (inode_item) { | |
434 | struct btrfs_inode_item *item; | |
d555438b | 435 | u32 mode; |
4bc4bee4 JB |
436 | |
437 | /* | |
438 | * New inode, set nbytes to 0 so that the nbytes comes out | |
439 | * properly when we replay the extents. | |
440 | */ | |
441 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
442 | btrfs_set_inode_nbytes(eb, item, 0); | |
d555438b JB |
443 | |
444 | /* | |
445 | * If this is a directory we need to reset the i_size to 0 so | |
446 | * that we can set it up properly when replaying the rest of | |
447 | * the items in this log. | |
448 | */ | |
449 | mode = btrfs_inode_mode(eb, item); | |
450 | if (S_ISDIR(mode)) | |
451 | btrfs_set_inode_size(eb, item, 0); | |
e02119d5 CM |
452 | } |
453 | insert: | |
b3b4aa74 | 454 | btrfs_release_path(path); |
e02119d5 | 455 | /* try to insert the key into the destination tree */ |
df8d116f | 456 | path->skip_release_on_error = 1; |
e02119d5 CM |
457 | ret = btrfs_insert_empty_item(trans, root, path, |
458 | key, item_size); | |
df8d116f | 459 | path->skip_release_on_error = 0; |
e02119d5 CM |
460 | |
461 | /* make sure any existing item is the correct size */ | |
df8d116f | 462 | if (ret == -EEXIST || ret == -EOVERFLOW) { |
e02119d5 CM |
463 | u32 found_size; |
464 | found_size = btrfs_item_size_nr(path->nodes[0], | |
465 | path->slots[0]); | |
143bede5 | 466 | if (found_size > item_size) |
78ac4f9e | 467 | btrfs_truncate_item(path, item_size, 1); |
143bede5 | 468 | else if (found_size < item_size) |
c71dd880 | 469 | btrfs_extend_item(path, item_size - found_size); |
e02119d5 | 470 | } else if (ret) { |
4a500fd1 | 471 | return ret; |
e02119d5 CM |
472 | } |
473 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
474 | path->slots[0]); | |
475 | ||
476 | /* don't overwrite an existing inode if the generation number | |
477 | * was logged as zero. This is done when the tree logging code | |
478 | * is just logging an inode to make sure it exists after recovery. | |
479 | * | |
480 | * Also, don't overwrite i_size on directories during replay. | |
481 | * log replay inserts and removes directory items based on the | |
482 | * state of the tree found in the subvolume, and i_size is modified | |
483 | * as it goes | |
484 | */ | |
485 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
486 | struct btrfs_inode_item *src_item; | |
487 | struct btrfs_inode_item *dst_item; | |
488 | ||
489 | src_item = (struct btrfs_inode_item *)src_ptr; | |
490 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
491 | ||
1a4bcf47 FM |
492 | if (btrfs_inode_generation(eb, src_item) == 0) { |
493 | struct extent_buffer *dst_eb = path->nodes[0]; | |
2f2ff0ee | 494 | const u64 ino_size = btrfs_inode_size(eb, src_item); |
1a4bcf47 | 495 | |
2f2ff0ee FM |
496 | /* |
497 | * For regular files an ino_size == 0 is used only when | |
498 | * logging that an inode exists, as part of a directory | |
499 | * fsync, and the inode wasn't fsynced before. In this | |
500 | * case don't set the size of the inode in the fs/subvol | |
501 | * tree, otherwise we would be throwing valid data away. | |
502 | */ | |
1a4bcf47 | 503 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && |
2f2ff0ee FM |
504 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && |
505 | ino_size != 0) { | |
1a4bcf47 | 506 | struct btrfs_map_token token; |
1a4bcf47 FM |
507 | |
508 | btrfs_init_map_token(&token); | |
509 | btrfs_set_token_inode_size(dst_eb, dst_item, | |
510 | ino_size, &token); | |
511 | } | |
e02119d5 | 512 | goto no_copy; |
1a4bcf47 | 513 | } |
e02119d5 CM |
514 | |
515 | if (overwrite_root && | |
516 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
517 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
518 | save_old_i_size = 1; | |
519 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
520 | dst_item); | |
521 | } | |
522 | } | |
523 | ||
524 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
525 | src_ptr, item_size); | |
526 | ||
527 | if (save_old_i_size) { | |
528 | struct btrfs_inode_item *dst_item; | |
529 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
530 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
531 | } | |
532 | ||
533 | /* make sure the generation is filled in */ | |
534 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
535 | struct btrfs_inode_item *dst_item; | |
536 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
537 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
538 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
539 | trans->transid); | |
540 | } | |
541 | } | |
542 | no_copy: | |
543 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 544 | btrfs_release_path(path); |
e02119d5 CM |
545 | return 0; |
546 | } | |
547 | ||
548 | /* | |
549 | * simple helper to read an inode off the disk from a given root | |
550 | * This can only be called for subvolume roots and not for the log | |
551 | */ | |
552 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
553 | u64 objectid) | |
554 | { | |
5d4f98a2 | 555 | struct btrfs_key key; |
e02119d5 | 556 | struct inode *inode; |
e02119d5 | 557 | |
5d4f98a2 YZ |
558 | key.objectid = objectid; |
559 | key.type = BTRFS_INODE_ITEM_KEY; | |
560 | key.offset = 0; | |
73f73415 | 561 | inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); |
2e19f1f9 | 562 | if (IS_ERR(inode)) |
5d4f98a2 | 563 | inode = NULL; |
e02119d5 CM |
564 | return inode; |
565 | } | |
566 | ||
567 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
568 | * subvolume 'root'. path is released on entry and should be released | |
569 | * on exit. | |
570 | * | |
571 | * extents in the log tree have not been allocated out of the extent | |
572 | * tree yet. So, this completes the allocation, taking a reference | |
573 | * as required if the extent already exists or creating a new extent | |
574 | * if it isn't in the extent allocation tree yet. | |
575 | * | |
576 | * The extent is inserted into the file, dropping any existing extents | |
577 | * from the file that overlap the new one. | |
578 | */ | |
579 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
580 | struct btrfs_root *root, | |
581 | struct btrfs_path *path, | |
582 | struct extent_buffer *eb, int slot, | |
583 | struct btrfs_key *key) | |
584 | { | |
0b246afa | 585 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 586 | int found_type; |
e02119d5 | 587 | u64 extent_end; |
e02119d5 | 588 | u64 start = key->offset; |
4bc4bee4 | 589 | u64 nbytes = 0; |
e02119d5 CM |
590 | struct btrfs_file_extent_item *item; |
591 | struct inode *inode = NULL; | |
592 | unsigned long size; | |
593 | int ret = 0; | |
594 | ||
595 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
596 | found_type = btrfs_file_extent_type(eb, item); | |
597 | ||
d899e052 | 598 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4bc4bee4 JB |
599 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
600 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
601 | extent_end = start + nbytes; | |
602 | ||
603 | /* | |
604 | * We don't add to the inodes nbytes if we are prealloc or a | |
605 | * hole. | |
606 | */ | |
607 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
608 | nbytes = 0; | |
609 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 610 | size = btrfs_file_extent_ram_bytes(eb, item); |
4bc4bee4 | 611 | nbytes = btrfs_file_extent_ram_bytes(eb, item); |
da17066c | 612 | extent_end = ALIGN(start + size, |
0b246afa | 613 | fs_info->sectorsize); |
e02119d5 CM |
614 | } else { |
615 | ret = 0; | |
616 | goto out; | |
617 | } | |
618 | ||
619 | inode = read_one_inode(root, key->objectid); | |
620 | if (!inode) { | |
621 | ret = -EIO; | |
622 | goto out; | |
623 | } | |
624 | ||
625 | /* | |
626 | * first check to see if we already have this extent in the | |
627 | * file. This must be done before the btrfs_drop_extents run | |
628 | * so we don't try to drop this extent. | |
629 | */ | |
f85b7379 DS |
630 | ret = btrfs_lookup_file_extent(trans, root, path, |
631 | btrfs_ino(BTRFS_I(inode)), start, 0); | |
e02119d5 | 632 | |
d899e052 YZ |
633 | if (ret == 0 && |
634 | (found_type == BTRFS_FILE_EXTENT_REG || | |
635 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
636 | struct btrfs_file_extent_item cmp1; |
637 | struct btrfs_file_extent_item cmp2; | |
638 | struct btrfs_file_extent_item *existing; | |
639 | struct extent_buffer *leaf; | |
640 | ||
641 | leaf = path->nodes[0]; | |
642 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
643 | struct btrfs_file_extent_item); | |
644 | ||
645 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
646 | sizeof(cmp1)); | |
647 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
648 | sizeof(cmp2)); | |
649 | ||
650 | /* | |
651 | * we already have a pointer to this exact extent, | |
652 | * we don't have to do anything | |
653 | */ | |
654 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
b3b4aa74 | 655 | btrfs_release_path(path); |
e02119d5 CM |
656 | goto out; |
657 | } | |
658 | } | |
b3b4aa74 | 659 | btrfs_release_path(path); |
e02119d5 CM |
660 | |
661 | /* drop any overlapping extents */ | |
2671485d | 662 | ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1); |
3650860b JB |
663 | if (ret) |
664 | goto out; | |
e02119d5 | 665 | |
07d400a6 YZ |
666 | if (found_type == BTRFS_FILE_EXTENT_REG || |
667 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 668 | u64 offset; |
07d400a6 YZ |
669 | unsigned long dest_offset; |
670 | struct btrfs_key ins; | |
671 | ||
3168021c FM |
672 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0 && |
673 | btrfs_fs_incompat(fs_info, NO_HOLES)) | |
674 | goto update_inode; | |
675 | ||
07d400a6 YZ |
676 | ret = btrfs_insert_empty_item(trans, root, path, key, |
677 | sizeof(*item)); | |
3650860b JB |
678 | if (ret) |
679 | goto out; | |
07d400a6 YZ |
680 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
681 | path->slots[0]); | |
682 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
683 | (unsigned long)item, sizeof(*item)); | |
684 | ||
685 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
686 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
687 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 688 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 | 689 | |
df2c95f3 QW |
690 | /* |
691 | * Manually record dirty extent, as here we did a shallow | |
692 | * file extent item copy and skip normal backref update, | |
693 | * but modifying extent tree all by ourselves. | |
694 | * So need to manually record dirty extent for qgroup, | |
695 | * as the owner of the file extent changed from log tree | |
696 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
697 | */ | |
a95f3aaf | 698 | ret = btrfs_qgroup_trace_extent(trans, |
df2c95f3 QW |
699 | btrfs_file_extent_disk_bytenr(eb, item), |
700 | btrfs_file_extent_disk_num_bytes(eb, item), | |
701 | GFP_NOFS); | |
702 | if (ret < 0) | |
703 | goto out; | |
704 | ||
07d400a6 | 705 | if (ins.objectid > 0) { |
82fa113f | 706 | struct btrfs_ref ref = { 0 }; |
07d400a6 YZ |
707 | u64 csum_start; |
708 | u64 csum_end; | |
709 | LIST_HEAD(ordered_sums); | |
82fa113f | 710 | |
07d400a6 YZ |
711 | /* |
712 | * is this extent already allocated in the extent | |
713 | * allocation tree? If so, just add a reference | |
714 | */ | |
2ff7e61e | 715 | ret = btrfs_lookup_data_extent(fs_info, ins.objectid, |
07d400a6 YZ |
716 | ins.offset); |
717 | if (ret == 0) { | |
82fa113f QW |
718 | btrfs_init_generic_ref(&ref, |
719 | BTRFS_ADD_DELAYED_REF, | |
720 | ins.objectid, ins.offset, 0); | |
721 | btrfs_init_data_ref(&ref, | |
722 | root->root_key.objectid, | |
b06c4bf5 | 723 | key->objectid, offset); |
82fa113f | 724 | ret = btrfs_inc_extent_ref(trans, &ref); |
b50c6e25 JB |
725 | if (ret) |
726 | goto out; | |
07d400a6 YZ |
727 | } else { |
728 | /* | |
729 | * insert the extent pointer in the extent | |
730 | * allocation tree | |
731 | */ | |
5d4f98a2 | 732 | ret = btrfs_alloc_logged_file_extent(trans, |
2ff7e61e | 733 | root->root_key.objectid, |
5d4f98a2 | 734 | key->objectid, offset, &ins); |
b50c6e25 JB |
735 | if (ret) |
736 | goto out; | |
07d400a6 | 737 | } |
b3b4aa74 | 738 | btrfs_release_path(path); |
07d400a6 YZ |
739 | |
740 | if (btrfs_file_extent_compression(eb, item)) { | |
741 | csum_start = ins.objectid; | |
742 | csum_end = csum_start + ins.offset; | |
743 | } else { | |
744 | csum_start = ins.objectid + | |
745 | btrfs_file_extent_offset(eb, item); | |
746 | csum_end = csum_start + | |
747 | btrfs_file_extent_num_bytes(eb, item); | |
748 | } | |
749 | ||
750 | ret = btrfs_lookup_csums_range(root->log_root, | |
751 | csum_start, csum_end - 1, | |
a2de733c | 752 | &ordered_sums, 0); |
3650860b JB |
753 | if (ret) |
754 | goto out; | |
b84b8390 FM |
755 | /* |
756 | * Now delete all existing cums in the csum root that | |
757 | * cover our range. We do this because we can have an | |
758 | * extent that is completely referenced by one file | |
759 | * extent item and partially referenced by another | |
760 | * file extent item (like after using the clone or | |
761 | * extent_same ioctls). In this case if we end up doing | |
762 | * the replay of the one that partially references the | |
763 | * extent first, and we do not do the csum deletion | |
764 | * below, we can get 2 csum items in the csum tree that | |
765 | * overlap each other. For example, imagine our log has | |
766 | * the two following file extent items: | |
767 | * | |
768 | * key (257 EXTENT_DATA 409600) | |
769 | * extent data disk byte 12845056 nr 102400 | |
770 | * extent data offset 20480 nr 20480 ram 102400 | |
771 | * | |
772 | * key (257 EXTENT_DATA 819200) | |
773 | * extent data disk byte 12845056 nr 102400 | |
774 | * extent data offset 0 nr 102400 ram 102400 | |
775 | * | |
776 | * Where the second one fully references the 100K extent | |
777 | * that starts at disk byte 12845056, and the log tree | |
778 | * has a single csum item that covers the entire range | |
779 | * of the extent: | |
780 | * | |
781 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
782 | * | |
783 | * After the first file extent item is replayed, the | |
784 | * csum tree gets the following csum item: | |
785 | * | |
786 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
787 | * | |
788 | * Which covers the 20K sub-range starting at offset 20K | |
789 | * of our extent. Now when we replay the second file | |
790 | * extent item, if we do not delete existing csum items | |
791 | * that cover any of its blocks, we end up getting two | |
792 | * csum items in our csum tree that overlap each other: | |
793 | * | |
794 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
795 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
796 | * | |
797 | * Which is a problem, because after this anyone trying | |
798 | * to lookup up for the checksum of any block of our | |
799 | * extent starting at an offset of 40K or higher, will | |
800 | * end up looking at the second csum item only, which | |
801 | * does not contain the checksum for any block starting | |
802 | * at offset 40K or higher of our extent. | |
803 | */ | |
07d400a6 YZ |
804 | while (!list_empty(&ordered_sums)) { |
805 | struct btrfs_ordered_sum *sums; | |
806 | sums = list_entry(ordered_sums.next, | |
807 | struct btrfs_ordered_sum, | |
808 | list); | |
b84b8390 | 809 | if (!ret) |
0b246afa | 810 | ret = btrfs_del_csums(trans, fs_info, |
5b4aacef JM |
811 | sums->bytenr, |
812 | sums->len); | |
3650860b JB |
813 | if (!ret) |
814 | ret = btrfs_csum_file_blocks(trans, | |
0b246afa | 815 | fs_info->csum_root, sums); |
07d400a6 YZ |
816 | list_del(&sums->list); |
817 | kfree(sums); | |
818 | } | |
3650860b JB |
819 | if (ret) |
820 | goto out; | |
07d400a6 | 821 | } else { |
b3b4aa74 | 822 | btrfs_release_path(path); |
07d400a6 YZ |
823 | } |
824 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
825 | /* inline extents are easy, we just overwrite them */ | |
826 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
3650860b JB |
827 | if (ret) |
828 | goto out; | |
07d400a6 | 829 | } |
e02119d5 | 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; | |
947 | struct btrfs_inode_ref *ref; | |
948 | unsigned long ptr; | |
949 | unsigned long ptr_end; | |
950 | unsigned long name_ptr; | |
951 | int found_name_len; | |
952 | int item_size; | |
953 | int ret; | |
954 | int match = 0; | |
955 | ||
956 | path = btrfs_alloc_path(); | |
2a29edc6 | 957 | if (!path) |
958 | return -ENOMEM; | |
959 | ||
e02119d5 CM |
960 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
961 | if (ret != 0) | |
962 | goto out; | |
963 | ||
e02119d5 | 964 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
f186373f MF |
965 | |
966 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1f250e92 FM |
967 | if (btrfs_find_name_in_ext_backref(path->nodes[0], |
968 | path->slots[0], | |
969 | ref_objectid, | |
f186373f MF |
970 | name, namelen, NULL)) |
971 | match = 1; | |
972 | ||
973 | goto out; | |
974 | } | |
975 | ||
976 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); | |
e02119d5 CM |
977 | ptr_end = ptr + item_size; |
978 | while (ptr < ptr_end) { | |
979 | ref = (struct btrfs_inode_ref *)ptr; | |
980 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); | |
981 | if (found_name_len == namelen) { | |
982 | name_ptr = (unsigned long)(ref + 1); | |
983 | ret = memcmp_extent_buffer(path->nodes[0], name, | |
984 | name_ptr, namelen); | |
985 | if (ret == 0) { | |
986 | match = 1; | |
987 | goto out; | |
988 | } | |
989 | } | |
990 | ptr = (unsigned long)(ref + 1) + found_name_len; | |
991 | } | |
992 | out: | |
993 | btrfs_free_path(path); | |
994 | return match; | |
995 | } | |
996 | ||
5a1d7843 | 997 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, |
e02119d5 | 998 | struct btrfs_root *root, |
e02119d5 | 999 | struct btrfs_path *path, |
5a1d7843 | 1000 | struct btrfs_root *log_root, |
94c91a1f NB |
1001 | struct btrfs_inode *dir, |
1002 | struct btrfs_inode *inode, | |
f186373f MF |
1003 | u64 inode_objectid, u64 parent_objectid, |
1004 | u64 ref_index, char *name, int namelen, | |
1005 | int *search_done) | |
e02119d5 | 1006 | { |
34f3e4f2 | 1007 | int ret; |
f186373f MF |
1008 | char *victim_name; |
1009 | int victim_name_len; | |
1010 | struct extent_buffer *leaf; | |
5a1d7843 | 1011 | struct btrfs_dir_item *di; |
f186373f MF |
1012 | struct btrfs_key search_key; |
1013 | struct btrfs_inode_extref *extref; | |
c622ae60 | 1014 | |
f186373f MF |
1015 | again: |
1016 | /* Search old style refs */ | |
1017 | search_key.objectid = inode_objectid; | |
1018 | search_key.type = BTRFS_INODE_REF_KEY; | |
1019 | search_key.offset = parent_objectid; | |
1020 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
e02119d5 | 1021 | if (ret == 0) { |
e02119d5 CM |
1022 | struct btrfs_inode_ref *victim_ref; |
1023 | unsigned long ptr; | |
1024 | unsigned long ptr_end; | |
f186373f MF |
1025 | |
1026 | leaf = path->nodes[0]; | |
e02119d5 CM |
1027 | |
1028 | /* are we trying to overwrite a back ref for the root directory | |
1029 | * if so, just jump out, we're done | |
1030 | */ | |
f186373f | 1031 | if (search_key.objectid == search_key.offset) |
5a1d7843 | 1032 | return 1; |
e02119d5 CM |
1033 | |
1034 | /* check all the names in this back reference to see | |
1035 | * if they are in the log. if so, we allow them to stay | |
1036 | * otherwise they must be unlinked as a conflict | |
1037 | */ | |
1038 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1039 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 1040 | while (ptr < ptr_end) { |
e02119d5 CM |
1041 | victim_ref = (struct btrfs_inode_ref *)ptr; |
1042 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1043 | victim_ref); | |
1044 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1045 | if (!victim_name) |
1046 | return -ENOMEM; | |
e02119d5 CM |
1047 | |
1048 | read_extent_buffer(leaf, victim_name, | |
1049 | (unsigned long)(victim_ref + 1), | |
1050 | victim_name_len); | |
1051 | ||
f186373f MF |
1052 | if (!backref_in_log(log_root, &search_key, |
1053 | parent_objectid, | |
1054 | victim_name, | |
e02119d5 | 1055 | victim_name_len)) { |
94c91a1f | 1056 | inc_nlink(&inode->vfs_inode); |
b3b4aa74 | 1057 | btrfs_release_path(path); |
12fcfd22 | 1058 | |
94c91a1f | 1059 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
4ec5934e | 1060 | victim_name, victim_name_len); |
f186373f | 1061 | kfree(victim_name); |
3650860b JB |
1062 | if (ret) |
1063 | return ret; | |
e5c304e6 | 1064 | ret = btrfs_run_delayed_items(trans); |
ada9af21 FDBM |
1065 | if (ret) |
1066 | return ret; | |
f186373f MF |
1067 | *search_done = 1; |
1068 | goto again; | |
e02119d5 CM |
1069 | } |
1070 | kfree(victim_name); | |
f186373f | 1071 | |
e02119d5 CM |
1072 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
1073 | } | |
e02119d5 | 1074 | |
c622ae60 | 1075 | /* |
1076 | * NOTE: we have searched root tree and checked the | |
bb7ab3b9 | 1077 | * corresponding ref, it does not need to check again. |
c622ae60 | 1078 | */ |
5a1d7843 | 1079 | *search_done = 1; |
e02119d5 | 1080 | } |
b3b4aa74 | 1081 | btrfs_release_path(path); |
e02119d5 | 1082 | |
f186373f MF |
1083 | /* Same search but for extended refs */ |
1084 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1085 | inode_objectid, parent_objectid, 0, | |
1086 | 0); | |
1087 | if (!IS_ERR_OR_NULL(extref)) { | |
1088 | u32 item_size; | |
1089 | u32 cur_offset = 0; | |
1090 | unsigned long base; | |
1091 | struct inode *victim_parent; | |
1092 | ||
1093 | leaf = path->nodes[0]; | |
1094 | ||
1095 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1096 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1097 | ||
1098 | while (cur_offset < item_size) { | |
dd9ef135 | 1099 | extref = (struct btrfs_inode_extref *)(base + cur_offset); |
f186373f MF |
1100 | |
1101 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1102 | ||
1103 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1104 | goto next; | |
1105 | ||
1106 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
3650860b JB |
1107 | if (!victim_name) |
1108 | return -ENOMEM; | |
f186373f MF |
1109 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, |
1110 | victim_name_len); | |
1111 | ||
1112 | search_key.objectid = inode_objectid; | |
1113 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1114 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1115 | victim_name, | |
1116 | victim_name_len); | |
1117 | ret = 0; | |
1118 | if (!backref_in_log(log_root, &search_key, | |
1119 | parent_objectid, victim_name, | |
1120 | victim_name_len)) { | |
1121 | ret = -ENOENT; | |
1122 | victim_parent = read_one_inode(root, | |
94c91a1f | 1123 | parent_objectid); |
f186373f | 1124 | if (victim_parent) { |
94c91a1f | 1125 | inc_nlink(&inode->vfs_inode); |
f186373f MF |
1126 | btrfs_release_path(path); |
1127 | ||
1128 | ret = btrfs_unlink_inode(trans, root, | |
4ec5934e | 1129 | BTRFS_I(victim_parent), |
94c91a1f | 1130 | inode, |
4ec5934e NB |
1131 | victim_name, |
1132 | victim_name_len); | |
ada9af21 FDBM |
1133 | if (!ret) |
1134 | ret = btrfs_run_delayed_items( | |
e5c304e6 | 1135 | trans); |
f186373f | 1136 | } |
f186373f MF |
1137 | iput(victim_parent); |
1138 | kfree(victim_name); | |
3650860b JB |
1139 | if (ret) |
1140 | return ret; | |
f186373f MF |
1141 | *search_done = 1; |
1142 | goto again; | |
1143 | } | |
1144 | kfree(victim_name); | |
f186373f MF |
1145 | next: |
1146 | cur_offset += victim_name_len + sizeof(*extref); | |
1147 | } | |
1148 | *search_done = 1; | |
1149 | } | |
1150 | btrfs_release_path(path); | |
1151 | ||
34f3e4f2 | 1152 | /* look for a conflicting sequence number */ |
94c91a1f | 1153 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
f186373f | 1154 | ref_index, name, namelen, 0); |
34f3e4f2 | 1155 | if (di && !IS_ERR(di)) { |
94c91a1f | 1156 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1157 | if (ret) |
1158 | return ret; | |
34f3e4f2 | 1159 | } |
1160 | btrfs_release_path(path); | |
1161 | ||
52042d8e | 1162 | /* look for a conflicting name */ |
94c91a1f | 1163 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), |
34f3e4f2 | 1164 | name, namelen, 0); |
1165 | if (di && !IS_ERR(di)) { | |
94c91a1f | 1166 | ret = drop_one_dir_item(trans, root, path, dir, di); |
3650860b JB |
1167 | if (ret) |
1168 | return ret; | |
34f3e4f2 | 1169 | } |
1170 | btrfs_release_path(path); | |
1171 | ||
5a1d7843 JS |
1172 | return 0; |
1173 | } | |
e02119d5 | 1174 | |
bae15d95 QW |
1175 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1176 | u32 *namelen, char **name, u64 *index, | |
1177 | u64 *parent_objectid) | |
f186373f MF |
1178 | { |
1179 | struct btrfs_inode_extref *extref; | |
1180 | ||
1181 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1182 | ||
1183 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1184 | *name = kmalloc(*namelen, GFP_NOFS); | |
1185 | if (*name == NULL) | |
1186 | return -ENOMEM; | |
1187 | ||
1188 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1189 | *namelen); | |
1190 | ||
1f250e92 FM |
1191 | if (index) |
1192 | *index = btrfs_inode_extref_index(eb, extref); | |
f186373f MF |
1193 | if (parent_objectid) |
1194 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1195 | ||
1196 | return 0; | |
1197 | } | |
1198 | ||
bae15d95 QW |
1199 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, |
1200 | u32 *namelen, char **name, u64 *index) | |
f186373f MF |
1201 | { |
1202 | struct btrfs_inode_ref *ref; | |
1203 | ||
1204 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1205 | ||
1206 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1207 | *name = kmalloc(*namelen, GFP_NOFS); | |
1208 | if (*name == NULL) | |
1209 | return -ENOMEM; | |
1210 | ||
1211 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1212 | ||
1f250e92 FM |
1213 | if (index) |
1214 | *index = btrfs_inode_ref_index(eb, ref); | |
f186373f MF |
1215 | |
1216 | return 0; | |
1217 | } | |
1218 | ||
1f250e92 FM |
1219 | /* |
1220 | * Take an inode reference item from the log tree and iterate all names from the | |
1221 | * inode reference item in the subvolume tree with the same key (if it exists). | |
1222 | * For any name that is not in the inode reference item from the log tree, do a | |
1223 | * proper unlink of that name (that is, remove its entry from the inode | |
1224 | * reference item and both dir index keys). | |
1225 | */ | |
1226 | static int unlink_old_inode_refs(struct btrfs_trans_handle *trans, | |
1227 | struct btrfs_root *root, | |
1228 | struct btrfs_path *path, | |
1229 | struct btrfs_inode *inode, | |
1230 | struct extent_buffer *log_eb, | |
1231 | int log_slot, | |
1232 | struct btrfs_key *key) | |
1233 | { | |
1234 | int ret; | |
1235 | unsigned long ref_ptr; | |
1236 | unsigned long ref_end; | |
1237 | struct extent_buffer *eb; | |
1238 | ||
1239 | again: | |
1240 | btrfs_release_path(path); | |
1241 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
1242 | if (ret > 0) { | |
1243 | ret = 0; | |
1244 | goto out; | |
1245 | } | |
1246 | if (ret < 0) | |
1247 | goto out; | |
1248 | ||
1249 | eb = path->nodes[0]; | |
1250 | ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]); | |
1251 | ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]); | |
1252 | while (ref_ptr < ref_end) { | |
1253 | char *name = NULL; | |
1254 | int namelen; | |
1255 | u64 parent_id; | |
1256 | ||
1257 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1258 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1259 | NULL, &parent_id); | |
1260 | } else { | |
1261 | parent_id = key->offset; | |
1262 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1263 | NULL); | |
1264 | } | |
1265 | if (ret) | |
1266 | goto out; | |
1267 | ||
1268 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1269 | ret = btrfs_find_name_in_ext_backref(log_eb, log_slot, | |
1270 | parent_id, name, | |
1271 | namelen, NULL); | |
1272 | else | |
1273 | ret = btrfs_find_name_in_backref(log_eb, log_slot, name, | |
1274 | namelen, NULL); | |
1275 | ||
1276 | if (!ret) { | |
1277 | struct inode *dir; | |
1278 | ||
1279 | btrfs_release_path(path); | |
1280 | dir = read_one_inode(root, parent_id); | |
1281 | if (!dir) { | |
1282 | ret = -ENOENT; | |
1283 | kfree(name); | |
1284 | goto out; | |
1285 | } | |
1286 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), | |
1287 | inode, name, namelen); | |
1288 | kfree(name); | |
1289 | iput(dir); | |
1290 | if (ret) | |
1291 | goto out; | |
1292 | goto again; | |
1293 | } | |
1294 | ||
1295 | kfree(name); | |
1296 | ref_ptr += namelen; | |
1297 | if (key->type == BTRFS_INODE_EXTREF_KEY) | |
1298 | ref_ptr += sizeof(struct btrfs_inode_extref); | |
1299 | else | |
1300 | ref_ptr += sizeof(struct btrfs_inode_ref); | |
1301 | } | |
1302 | ret = 0; | |
1303 | out: | |
1304 | btrfs_release_path(path); | |
1305 | return ret; | |
1306 | } | |
1307 | ||
0d836392 FM |
1308 | static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir, |
1309 | const u8 ref_type, const char *name, | |
1310 | const int namelen) | |
1311 | { | |
1312 | struct btrfs_key key; | |
1313 | struct btrfs_path *path; | |
1314 | const u64 parent_id = btrfs_ino(BTRFS_I(dir)); | |
1315 | int ret; | |
1316 | ||
1317 | path = btrfs_alloc_path(); | |
1318 | if (!path) | |
1319 | return -ENOMEM; | |
1320 | ||
1321 | key.objectid = btrfs_ino(BTRFS_I(inode)); | |
1322 | key.type = ref_type; | |
1323 | if (key.type == BTRFS_INODE_REF_KEY) | |
1324 | key.offset = parent_id; | |
1325 | else | |
1326 | key.offset = btrfs_extref_hash(parent_id, name, namelen); | |
1327 | ||
1328 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0); | |
1329 | if (ret < 0) | |
1330 | goto out; | |
1331 | if (ret > 0) { | |
1332 | ret = 0; | |
1333 | goto out; | |
1334 | } | |
1335 | if (key.type == BTRFS_INODE_EXTREF_KEY) | |
1336 | ret = btrfs_find_name_in_ext_backref(path->nodes[0], | |
1337 | path->slots[0], parent_id, | |
1338 | name, namelen, NULL); | |
1339 | else | |
1340 | ret = btrfs_find_name_in_backref(path->nodes[0], path->slots[0], | |
1341 | name, namelen, NULL); | |
1342 | ||
1343 | out: | |
1344 | btrfs_free_path(path); | |
1345 | return ret; | |
1346 | } | |
1347 | ||
6b5fc433 FM |
1348 | static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1349 | struct inode *dir, struct inode *inode, const char *name, | |
1350 | int namelen, u64 ref_index) | |
1351 | { | |
1352 | struct btrfs_dir_item *dir_item; | |
1353 | struct btrfs_key key; | |
1354 | struct btrfs_path *path; | |
1355 | struct inode *other_inode = NULL; | |
1356 | int ret; | |
1357 | ||
1358 | path = btrfs_alloc_path(); | |
1359 | if (!path) | |
1360 | return -ENOMEM; | |
1361 | ||
1362 | dir_item = btrfs_lookup_dir_item(NULL, root, path, | |
1363 | btrfs_ino(BTRFS_I(dir)), | |
1364 | name, namelen, 0); | |
1365 | if (!dir_item) { | |
1366 | btrfs_release_path(path); | |
1367 | goto add_link; | |
1368 | } else if (IS_ERR(dir_item)) { | |
1369 | ret = PTR_ERR(dir_item); | |
1370 | goto out; | |
1371 | } | |
1372 | ||
1373 | /* | |
1374 | * Our inode's dentry collides with the dentry of another inode which is | |
1375 | * in the log but not yet processed since it has a higher inode number. | |
1376 | * So delete that other dentry. | |
1377 | */ | |
1378 | btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key); | |
1379 | btrfs_release_path(path); | |
1380 | other_inode = read_one_inode(root, key.objectid); | |
1381 | if (!other_inode) { | |
1382 | ret = -ENOENT; | |
1383 | goto out; | |
1384 | } | |
1385 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode), | |
1386 | name, namelen); | |
1387 | if (ret) | |
1388 | goto out; | |
1389 | /* | |
1390 | * If we dropped the link count to 0, bump it so that later the iput() | |
1391 | * on the inode will not free it. We will fixup the link count later. | |
1392 | */ | |
1393 | if (other_inode->i_nlink == 0) | |
1394 | inc_nlink(other_inode); | |
1395 | ||
1396 | ret = btrfs_run_delayed_items(trans); | |
1397 | if (ret) | |
1398 | goto out; | |
1399 | add_link: | |
1400 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), | |
1401 | name, namelen, 0, ref_index); | |
1402 | out: | |
1403 | iput(other_inode); | |
1404 | btrfs_free_path(path); | |
1405 | ||
1406 | return ret; | |
1407 | } | |
1408 | ||
5a1d7843 JS |
1409 | /* |
1410 | * replay one inode back reference item found in the log tree. | |
1411 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1412 | * root is the destination we are replaying into, and path is for temp | |
1413 | * use by this function. (it should be released on return). | |
1414 | */ | |
1415 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1416 | struct btrfs_root *root, | |
1417 | struct btrfs_root *log, | |
1418 | struct btrfs_path *path, | |
1419 | struct extent_buffer *eb, int slot, | |
1420 | struct btrfs_key *key) | |
1421 | { | |
03b2f08b GB |
1422 | struct inode *dir = NULL; |
1423 | struct inode *inode = NULL; | |
5a1d7843 JS |
1424 | unsigned long ref_ptr; |
1425 | unsigned long ref_end; | |
03b2f08b | 1426 | char *name = NULL; |
5a1d7843 JS |
1427 | int namelen; |
1428 | int ret; | |
1429 | int search_done = 0; | |
f186373f MF |
1430 | int log_ref_ver = 0; |
1431 | u64 parent_objectid; | |
1432 | u64 inode_objectid; | |
f46dbe3d | 1433 | u64 ref_index = 0; |
f186373f MF |
1434 | int ref_struct_size; |
1435 | ||
1436 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1437 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1438 | ||
1439 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1440 | struct btrfs_inode_extref *r; | |
1441 | ||
1442 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1443 | log_ref_ver = 1; | |
1444 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1445 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1446 | } else { | |
1447 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1448 | parent_objectid = key->offset; | |
1449 | } | |
1450 | inode_objectid = key->objectid; | |
e02119d5 | 1451 | |
5a1d7843 JS |
1452 | /* |
1453 | * it is possible that we didn't log all the parent directories | |
1454 | * for a given inode. If we don't find the dir, just don't | |
1455 | * copy the back ref in. The link count fixup code will take | |
1456 | * care of the rest | |
1457 | */ | |
f186373f | 1458 | dir = read_one_inode(root, parent_objectid); |
03b2f08b GB |
1459 | if (!dir) { |
1460 | ret = -ENOENT; | |
1461 | goto out; | |
1462 | } | |
5a1d7843 | 1463 | |
f186373f | 1464 | inode = read_one_inode(root, inode_objectid); |
5a1d7843 | 1465 | if (!inode) { |
03b2f08b GB |
1466 | ret = -EIO; |
1467 | goto out; | |
5a1d7843 JS |
1468 | } |
1469 | ||
5a1d7843 | 1470 | while (ref_ptr < ref_end) { |
f186373f | 1471 | if (log_ref_ver) { |
bae15d95 QW |
1472 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, |
1473 | &ref_index, &parent_objectid); | |
f186373f MF |
1474 | /* |
1475 | * parent object can change from one array | |
1476 | * item to another. | |
1477 | */ | |
1478 | if (!dir) | |
1479 | dir = read_one_inode(root, parent_objectid); | |
03b2f08b GB |
1480 | if (!dir) { |
1481 | ret = -ENOENT; | |
1482 | goto out; | |
1483 | } | |
f186373f | 1484 | } else { |
bae15d95 QW |
1485 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, |
1486 | &ref_index); | |
f186373f MF |
1487 | } |
1488 | if (ret) | |
03b2f08b | 1489 | goto out; |
5a1d7843 JS |
1490 | |
1491 | /* if we already have a perfect match, we're done */ | |
f85b7379 DS |
1492 | if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), |
1493 | btrfs_ino(BTRFS_I(inode)), ref_index, | |
1494 | name, namelen)) { | |
5a1d7843 JS |
1495 | /* |
1496 | * look for a conflicting back reference in the | |
1497 | * metadata. if we find one we have to unlink that name | |
1498 | * of the file before we add our new link. Later on, we | |
1499 | * overwrite any existing back reference, and we don't | |
1500 | * want to create dangling pointers in the directory. | |
1501 | */ | |
1502 | ||
1503 | if (!search_done) { | |
1504 | ret = __add_inode_ref(trans, root, path, log, | |
94c91a1f | 1505 | BTRFS_I(dir), |
d75eefdf | 1506 | BTRFS_I(inode), |
f186373f MF |
1507 | inode_objectid, |
1508 | parent_objectid, | |
1509 | ref_index, name, namelen, | |
5a1d7843 | 1510 | &search_done); |
03b2f08b GB |
1511 | if (ret) { |
1512 | if (ret == 1) | |
1513 | ret = 0; | |
3650860b JB |
1514 | goto out; |
1515 | } | |
5a1d7843 JS |
1516 | } |
1517 | ||
0d836392 FM |
1518 | /* |
1519 | * If a reference item already exists for this inode | |
1520 | * with the same parent and name, but different index, | |
1521 | * drop it and the corresponding directory index entries | |
1522 | * from the parent before adding the new reference item | |
1523 | * and dir index entries, otherwise we would fail with | |
1524 | * -EEXIST returned from btrfs_add_link() below. | |
1525 | */ | |
1526 | ret = btrfs_inode_ref_exists(inode, dir, key->type, | |
1527 | name, namelen); | |
1528 | if (ret > 0) { | |
1529 | ret = btrfs_unlink_inode(trans, root, | |
1530 | BTRFS_I(dir), | |
1531 | BTRFS_I(inode), | |
1532 | name, namelen); | |
1533 | /* | |
1534 | * If we dropped the link count to 0, bump it so | |
1535 | * that later the iput() on the inode will not | |
1536 | * free it. We will fixup the link count later. | |
1537 | */ | |
1538 | if (!ret && inode->i_nlink == 0) | |
1539 | inc_nlink(inode); | |
1540 | } | |
1541 | if (ret < 0) | |
1542 | goto out; | |
1543 | ||
5a1d7843 | 1544 | /* insert our name */ |
6b5fc433 FM |
1545 | ret = add_link(trans, root, dir, inode, name, namelen, |
1546 | ref_index); | |
3650860b JB |
1547 | if (ret) |
1548 | goto out; | |
5a1d7843 JS |
1549 | |
1550 | btrfs_update_inode(trans, root, inode); | |
1551 | } | |
1552 | ||
f186373f | 1553 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; |
5a1d7843 | 1554 | kfree(name); |
03b2f08b | 1555 | name = NULL; |
f186373f MF |
1556 | if (log_ref_ver) { |
1557 | iput(dir); | |
1558 | dir = NULL; | |
1559 | } | |
5a1d7843 | 1560 | } |
e02119d5 | 1561 | |
1f250e92 FM |
1562 | /* |
1563 | * Before we overwrite the inode reference item in the subvolume tree | |
1564 | * with the item from the log tree, we must unlink all names from the | |
1565 | * parent directory that are in the subvolume's tree inode reference | |
1566 | * item, otherwise we end up with an inconsistent subvolume tree where | |
1567 | * dir index entries exist for a name but there is no inode reference | |
1568 | * item with the same name. | |
1569 | */ | |
1570 | ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot, | |
1571 | key); | |
1572 | if (ret) | |
1573 | goto out; | |
1574 | ||
e02119d5 CM |
1575 | /* finally write the back reference in the inode */ |
1576 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
5a1d7843 | 1577 | out: |
b3b4aa74 | 1578 | btrfs_release_path(path); |
03b2f08b | 1579 | kfree(name); |
e02119d5 CM |
1580 | iput(dir); |
1581 | iput(inode); | |
3650860b | 1582 | return ret; |
e02119d5 CM |
1583 | } |
1584 | ||
c71bf099 | 1585 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
9c4f61f0 | 1586 | struct btrfs_root *root, u64 ino) |
c71bf099 YZ |
1587 | { |
1588 | int ret; | |
381cf658 | 1589 | |
9c4f61f0 DS |
1590 | ret = btrfs_insert_orphan_item(trans, root, ino); |
1591 | if (ret == -EEXIST) | |
1592 | ret = 0; | |
381cf658 | 1593 | |
c71bf099 YZ |
1594 | return ret; |
1595 | } | |
1596 | ||
f186373f | 1597 | static int count_inode_extrefs(struct btrfs_root *root, |
36283658 | 1598 | struct btrfs_inode *inode, struct btrfs_path *path) |
f186373f MF |
1599 | { |
1600 | int ret = 0; | |
1601 | int name_len; | |
1602 | unsigned int nlink = 0; | |
1603 | u32 item_size; | |
1604 | u32 cur_offset = 0; | |
36283658 | 1605 | u64 inode_objectid = btrfs_ino(inode); |
f186373f MF |
1606 | u64 offset = 0; |
1607 | unsigned long ptr; | |
1608 | struct btrfs_inode_extref *extref; | |
1609 | struct extent_buffer *leaf; | |
1610 | ||
1611 | while (1) { | |
1612 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1613 | &extref, &offset); | |
1614 | if (ret) | |
1615 | break; | |
c71bf099 | 1616 | |
f186373f MF |
1617 | leaf = path->nodes[0]; |
1618 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1619 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2c2c452b | 1620 | cur_offset = 0; |
f186373f MF |
1621 | |
1622 | while (cur_offset < item_size) { | |
1623 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1624 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1625 | ||
1626 | nlink++; | |
1627 | ||
1628 | cur_offset += name_len + sizeof(*extref); | |
1629 | } | |
1630 | ||
1631 | offset++; | |
1632 | btrfs_release_path(path); | |
1633 | } | |
1634 | btrfs_release_path(path); | |
1635 | ||
2c2c452b | 1636 | if (ret < 0 && ret != -ENOENT) |
f186373f MF |
1637 | return ret; |
1638 | return nlink; | |
1639 | } | |
1640 | ||
1641 | static int count_inode_refs(struct btrfs_root *root, | |
f329e319 | 1642 | struct btrfs_inode *inode, struct btrfs_path *path) |
e02119d5 | 1643 | { |
e02119d5 CM |
1644 | int ret; |
1645 | struct btrfs_key key; | |
f186373f | 1646 | unsigned int nlink = 0; |
e02119d5 CM |
1647 | unsigned long ptr; |
1648 | unsigned long ptr_end; | |
1649 | int name_len; | |
f329e319 | 1650 | u64 ino = btrfs_ino(inode); |
e02119d5 | 1651 | |
33345d01 | 1652 | key.objectid = ino; |
e02119d5 CM |
1653 | key.type = BTRFS_INODE_REF_KEY; |
1654 | key.offset = (u64)-1; | |
1655 | ||
d397712b | 1656 | while (1) { |
e02119d5 CM |
1657 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1658 | if (ret < 0) | |
1659 | break; | |
1660 | if (ret > 0) { | |
1661 | if (path->slots[0] == 0) | |
1662 | break; | |
1663 | path->slots[0]--; | |
1664 | } | |
e93ae26f | 1665 | process_slot: |
e02119d5 CM |
1666 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
1667 | path->slots[0]); | |
33345d01 | 1668 | if (key.objectid != ino || |
e02119d5 CM |
1669 | key.type != BTRFS_INODE_REF_KEY) |
1670 | break; | |
1671 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1672 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1673 | path->slots[0]); | |
d397712b | 1674 | while (ptr < ptr_end) { |
e02119d5 CM |
1675 | struct btrfs_inode_ref *ref; |
1676 | ||
1677 | ref = (struct btrfs_inode_ref *)ptr; | |
1678 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1679 | ref); | |
1680 | ptr = (unsigned long)(ref + 1) + name_len; | |
1681 | nlink++; | |
1682 | } | |
1683 | ||
1684 | if (key.offset == 0) | |
1685 | break; | |
e93ae26f FDBM |
1686 | if (path->slots[0] > 0) { |
1687 | path->slots[0]--; | |
1688 | goto process_slot; | |
1689 | } | |
e02119d5 | 1690 | key.offset--; |
b3b4aa74 | 1691 | btrfs_release_path(path); |
e02119d5 | 1692 | } |
b3b4aa74 | 1693 | btrfs_release_path(path); |
f186373f MF |
1694 | |
1695 | return nlink; | |
1696 | } | |
1697 | ||
1698 | /* | |
1699 | * There are a few corners where the link count of the file can't | |
1700 | * be properly maintained during replay. So, instead of adding | |
1701 | * lots of complexity to the log code, we just scan the backrefs | |
1702 | * for any file that has been through replay. | |
1703 | * | |
1704 | * The scan will update the link count on the inode to reflect the | |
1705 | * number of back refs found. If it goes down to zero, the iput | |
1706 | * will free the inode. | |
1707 | */ | |
1708 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1709 | struct btrfs_root *root, | |
1710 | struct inode *inode) | |
1711 | { | |
1712 | struct btrfs_path *path; | |
1713 | int ret; | |
1714 | u64 nlink = 0; | |
4a0cc7ca | 1715 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
f186373f MF |
1716 | |
1717 | path = btrfs_alloc_path(); | |
1718 | if (!path) | |
1719 | return -ENOMEM; | |
1720 | ||
f329e319 | 1721 | ret = count_inode_refs(root, BTRFS_I(inode), path); |
f186373f MF |
1722 | if (ret < 0) |
1723 | goto out; | |
1724 | ||
1725 | nlink = ret; | |
1726 | ||
36283658 | 1727 | ret = count_inode_extrefs(root, BTRFS_I(inode), path); |
f186373f MF |
1728 | if (ret < 0) |
1729 | goto out; | |
1730 | ||
1731 | nlink += ret; | |
1732 | ||
1733 | ret = 0; | |
1734 | ||
e02119d5 | 1735 | if (nlink != inode->i_nlink) { |
bfe86848 | 1736 | set_nlink(inode, nlink); |
e02119d5 CM |
1737 | btrfs_update_inode(trans, root, inode); |
1738 | } | |
8d5bf1cb | 1739 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 1740 | |
c71bf099 YZ |
1741 | if (inode->i_nlink == 0) { |
1742 | if (S_ISDIR(inode->i_mode)) { | |
1743 | ret = replay_dir_deletes(trans, root, NULL, path, | |
33345d01 | 1744 | ino, 1); |
3650860b JB |
1745 | if (ret) |
1746 | goto out; | |
c71bf099 | 1747 | } |
33345d01 | 1748 | ret = insert_orphan_item(trans, root, ino); |
12fcfd22 | 1749 | } |
12fcfd22 | 1750 | |
f186373f MF |
1751 | out: |
1752 | btrfs_free_path(path); | |
1753 | return ret; | |
e02119d5 CM |
1754 | } |
1755 | ||
1756 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1757 | struct btrfs_root *root, | |
1758 | struct btrfs_path *path) | |
1759 | { | |
1760 | int ret; | |
1761 | struct btrfs_key key; | |
1762 | struct inode *inode; | |
1763 | ||
1764 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1765 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1766 | key.offset = (u64)-1; | |
d397712b | 1767 | while (1) { |
e02119d5 CM |
1768 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1769 | if (ret < 0) | |
1770 | break; | |
1771 | ||
1772 | if (ret == 1) { | |
1773 | if (path->slots[0] == 0) | |
1774 | break; | |
1775 | path->slots[0]--; | |
1776 | } | |
1777 | ||
1778 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1779 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1780 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1781 | break; | |
1782 | ||
1783 | ret = btrfs_del_item(trans, root, path); | |
65a246c5 TI |
1784 | if (ret) |
1785 | goto out; | |
e02119d5 | 1786 | |
b3b4aa74 | 1787 | btrfs_release_path(path); |
e02119d5 | 1788 | inode = read_one_inode(root, key.offset); |
c00e9493 TI |
1789 | if (!inode) |
1790 | return -EIO; | |
e02119d5 CM |
1791 | |
1792 | ret = fixup_inode_link_count(trans, root, inode); | |
e02119d5 | 1793 | iput(inode); |
3650860b JB |
1794 | if (ret) |
1795 | goto out; | |
e02119d5 | 1796 | |
12fcfd22 CM |
1797 | /* |
1798 | * fixup on a directory may create new entries, | |
1799 | * make sure we always look for the highset possible | |
1800 | * offset | |
1801 | */ | |
1802 | key.offset = (u64)-1; | |
e02119d5 | 1803 | } |
65a246c5 TI |
1804 | ret = 0; |
1805 | out: | |
b3b4aa74 | 1806 | btrfs_release_path(path); |
65a246c5 | 1807 | return ret; |
e02119d5 CM |
1808 | } |
1809 | ||
1810 | ||
1811 | /* | |
1812 | * record a given inode in the fixup dir so we can check its link | |
1813 | * count when replay is done. The link count is incremented here | |
1814 | * so the inode won't go away until we check it | |
1815 | */ | |
1816 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1817 | struct btrfs_root *root, | |
1818 | struct btrfs_path *path, | |
1819 | u64 objectid) | |
1820 | { | |
1821 | struct btrfs_key key; | |
1822 | int ret = 0; | |
1823 | struct inode *inode; | |
1824 | ||
1825 | inode = read_one_inode(root, objectid); | |
c00e9493 TI |
1826 | if (!inode) |
1827 | return -EIO; | |
e02119d5 CM |
1828 | |
1829 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
962a298f | 1830 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
e02119d5 CM |
1831 | key.offset = objectid; |
1832 | ||
1833 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1834 | ||
b3b4aa74 | 1835 | btrfs_release_path(path); |
e02119d5 | 1836 | if (ret == 0) { |
9bf7a489 JB |
1837 | if (!inode->i_nlink) |
1838 | set_nlink(inode, 1); | |
1839 | else | |
8b558c5f | 1840 | inc_nlink(inode); |
b9959295 | 1841 | ret = btrfs_update_inode(trans, root, inode); |
e02119d5 CM |
1842 | } else if (ret == -EEXIST) { |
1843 | ret = 0; | |
1844 | } else { | |
3650860b | 1845 | BUG(); /* Logic Error */ |
e02119d5 CM |
1846 | } |
1847 | iput(inode); | |
1848 | ||
1849 | return ret; | |
1850 | } | |
1851 | ||
1852 | /* | |
1853 | * when replaying the log for a directory, we only insert names | |
1854 | * for inodes that actually exist. This means an fsync on a directory | |
1855 | * does not implicitly fsync all the new files in it | |
1856 | */ | |
1857 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1858 | struct btrfs_root *root, | |
e02119d5 | 1859 | u64 dirid, u64 index, |
60d53eb3 | 1860 | char *name, int name_len, |
e02119d5 CM |
1861 | struct btrfs_key *location) |
1862 | { | |
1863 | struct inode *inode; | |
1864 | struct inode *dir; | |
1865 | int ret; | |
1866 | ||
1867 | inode = read_one_inode(root, location->objectid); | |
1868 | if (!inode) | |
1869 | return -ENOENT; | |
1870 | ||
1871 | dir = read_one_inode(root, dirid); | |
1872 | if (!dir) { | |
1873 | iput(inode); | |
1874 | return -EIO; | |
1875 | } | |
d555438b | 1876 | |
db0a669f NB |
1877 | ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name, |
1878 | name_len, 1, index); | |
e02119d5 CM |
1879 | |
1880 | /* FIXME, put inode into FIXUP list */ | |
1881 | ||
1882 | iput(inode); | |
1883 | iput(dir); | |
1884 | return ret; | |
1885 | } | |
1886 | ||
df8d116f FM |
1887 | /* |
1888 | * Return true if an inode reference exists in the log for the given name, | |
1889 | * inode and parent inode. | |
1890 | */ | |
1891 | static bool name_in_log_ref(struct btrfs_root *log_root, | |
1892 | const char *name, const int name_len, | |
1893 | const u64 dirid, const u64 ino) | |
1894 | { | |
1895 | struct btrfs_key search_key; | |
1896 | ||
1897 | search_key.objectid = ino; | |
1898 | search_key.type = BTRFS_INODE_REF_KEY; | |
1899 | search_key.offset = dirid; | |
1900 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1901 | return true; | |
1902 | ||
1903 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1904 | search_key.offset = btrfs_extref_hash(dirid, name, name_len); | |
1905 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1906 | return true; | |
1907 | ||
1908 | return false; | |
1909 | } | |
1910 | ||
e02119d5 CM |
1911 | /* |
1912 | * take a single entry in a log directory item and replay it into | |
1913 | * the subvolume. | |
1914 | * | |
1915 | * if a conflicting item exists in the subdirectory already, | |
1916 | * the inode it points to is unlinked and put into the link count | |
1917 | * fix up tree. | |
1918 | * | |
1919 | * If a name from the log points to a file or directory that does | |
1920 | * not exist in the FS, it is skipped. fsyncs on directories | |
1921 | * do not force down inodes inside that directory, just changes to the | |
1922 | * names or unlinks in a directory. | |
bb53eda9 FM |
1923 | * |
1924 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1925 | * non-existing inode) and 1 if the name was replayed. | |
e02119d5 CM |
1926 | */ |
1927 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1928 | struct btrfs_root *root, | |
1929 | struct btrfs_path *path, | |
1930 | struct extent_buffer *eb, | |
1931 | struct btrfs_dir_item *di, | |
1932 | struct btrfs_key *key) | |
1933 | { | |
1934 | char *name; | |
1935 | int name_len; | |
1936 | struct btrfs_dir_item *dst_di; | |
1937 | struct btrfs_key found_key; | |
1938 | struct btrfs_key log_key; | |
1939 | struct inode *dir; | |
e02119d5 | 1940 | u8 log_type; |
4bef0848 | 1941 | int exists; |
3650860b | 1942 | int ret = 0; |
d555438b | 1943 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); |
bb53eda9 | 1944 | bool name_added = false; |
e02119d5 CM |
1945 | |
1946 | dir = read_one_inode(root, key->objectid); | |
c00e9493 TI |
1947 | if (!dir) |
1948 | return -EIO; | |
e02119d5 CM |
1949 | |
1950 | name_len = btrfs_dir_name_len(eb, di); | |
1951 | name = kmalloc(name_len, GFP_NOFS); | |
2bac325e FDBM |
1952 | if (!name) { |
1953 | ret = -ENOMEM; | |
1954 | goto out; | |
1955 | } | |
2a29edc6 | 1956 | |
e02119d5 CM |
1957 | log_type = btrfs_dir_type(eb, di); |
1958 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1959 | name_len); | |
1960 | ||
1961 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1962 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1963 | if (exists == 0) | |
1964 | exists = 1; | |
1965 | else | |
1966 | exists = 0; | |
b3b4aa74 | 1967 | btrfs_release_path(path); |
4bef0848 | 1968 | |
e02119d5 CM |
1969 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1970 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1971 | name, name_len, 1); | |
d397712b | 1972 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1973 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1974 | key->objectid, | |
1975 | key->offset, name, | |
1976 | name_len, 1); | |
1977 | } else { | |
3650860b JB |
1978 | /* Corruption */ |
1979 | ret = -EINVAL; | |
1980 | goto out; | |
e02119d5 | 1981 | } |
c704005d | 1982 | if (IS_ERR_OR_NULL(dst_di)) { |
e02119d5 CM |
1983 | /* we need a sequence number to insert, so we only |
1984 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1985 | */ | |
1986 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1987 | goto out; | |
1988 | goto insert; | |
1989 | } | |
1990 | ||
1991 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1992 | /* the existing item matches the logged item */ | |
1993 | if (found_key.objectid == log_key.objectid && | |
1994 | found_key.type == log_key.type && | |
1995 | found_key.offset == log_key.offset && | |
1996 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
a2cc11db | 1997 | update_size = false; |
e02119d5 CM |
1998 | goto out; |
1999 | } | |
2000 | ||
2001 | /* | |
2002 | * don't drop the conflicting directory entry if the inode | |
2003 | * for the new entry doesn't exist | |
2004 | */ | |
4bef0848 | 2005 | if (!exists) |
e02119d5 CM |
2006 | goto out; |
2007 | ||
207e7d92 | 2008 | ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di); |
3650860b JB |
2009 | if (ret) |
2010 | goto out; | |
e02119d5 CM |
2011 | |
2012 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
2013 | goto insert; | |
2014 | out: | |
b3b4aa74 | 2015 | btrfs_release_path(path); |
d555438b | 2016 | if (!ret && update_size) { |
6ef06d27 | 2017 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2); |
d555438b JB |
2018 | ret = btrfs_update_inode(trans, root, dir); |
2019 | } | |
e02119d5 CM |
2020 | kfree(name); |
2021 | iput(dir); | |
bb53eda9 FM |
2022 | if (!ret && name_added) |
2023 | ret = 1; | |
3650860b | 2024 | return ret; |
e02119d5 CM |
2025 | |
2026 | insert: | |
df8d116f FM |
2027 | if (name_in_log_ref(root->log_root, name, name_len, |
2028 | key->objectid, log_key.objectid)) { | |
2029 | /* The dentry will be added later. */ | |
2030 | ret = 0; | |
2031 | update_size = false; | |
2032 | goto out; | |
2033 | } | |
b3b4aa74 | 2034 | btrfs_release_path(path); |
60d53eb3 Z |
2035 | ret = insert_one_name(trans, root, key->objectid, key->offset, |
2036 | name, name_len, &log_key); | |
df8d116f | 2037 | if (ret && ret != -ENOENT && ret != -EEXIST) |
3650860b | 2038 | goto out; |
bb53eda9 FM |
2039 | if (!ret) |
2040 | name_added = true; | |
d555438b | 2041 | update_size = false; |
3650860b | 2042 | ret = 0; |
e02119d5 CM |
2043 | goto out; |
2044 | } | |
2045 | ||
2046 | /* | |
2047 | * find all the names in a directory item and reconcile them into | |
2048 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
2049 | * one name in a directory item, but the same code gets used for | |
2050 | * both directory index types | |
2051 | */ | |
2052 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
2053 | struct btrfs_root *root, | |
2054 | struct btrfs_path *path, | |
2055 | struct extent_buffer *eb, int slot, | |
2056 | struct btrfs_key *key) | |
2057 | { | |
bb53eda9 | 2058 | int ret = 0; |
e02119d5 CM |
2059 | u32 item_size = btrfs_item_size_nr(eb, slot); |
2060 | struct btrfs_dir_item *di; | |
2061 | int name_len; | |
2062 | unsigned long ptr; | |
2063 | unsigned long ptr_end; | |
bb53eda9 | 2064 | struct btrfs_path *fixup_path = NULL; |
e02119d5 CM |
2065 | |
2066 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2067 | ptr_end = ptr + item_size; | |
d397712b | 2068 | while (ptr < ptr_end) { |
e02119d5 CM |
2069 | di = (struct btrfs_dir_item *)ptr; |
2070 | name_len = btrfs_dir_name_len(eb, di); | |
2071 | ret = replay_one_name(trans, root, path, eb, di, key); | |
bb53eda9 FM |
2072 | if (ret < 0) |
2073 | break; | |
e02119d5 CM |
2074 | ptr = (unsigned long)(di + 1); |
2075 | ptr += name_len; | |
bb53eda9 FM |
2076 | |
2077 | /* | |
2078 | * If this entry refers to a non-directory (directories can not | |
2079 | * have a link count > 1) and it was added in the transaction | |
2080 | * that was not committed, make sure we fixup the link count of | |
2081 | * the inode it the entry points to. Otherwise something like | |
2082 | * the following would result in a directory pointing to an | |
2083 | * inode with a wrong link that does not account for this dir | |
2084 | * entry: | |
2085 | * | |
2086 | * mkdir testdir | |
2087 | * touch testdir/foo | |
2088 | * touch testdir/bar | |
2089 | * sync | |
2090 | * | |
2091 | * ln testdir/bar testdir/bar_link | |
2092 | * ln testdir/foo testdir/foo_link | |
2093 | * xfs_io -c "fsync" testdir/bar | |
2094 | * | |
2095 | * <power failure> | |
2096 | * | |
2097 | * mount fs, log replay happens | |
2098 | * | |
2099 | * File foo would remain with a link count of 1 when it has two | |
2100 | * entries pointing to it in the directory testdir. This would | |
2101 | * make it impossible to ever delete the parent directory has | |
2102 | * it would result in stale dentries that can never be deleted. | |
2103 | */ | |
2104 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
2105 | struct btrfs_key di_key; | |
2106 | ||
2107 | if (!fixup_path) { | |
2108 | fixup_path = btrfs_alloc_path(); | |
2109 | if (!fixup_path) { | |
2110 | ret = -ENOMEM; | |
2111 | break; | |
2112 | } | |
2113 | } | |
2114 | ||
2115 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2116 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
2117 | di_key.objectid); | |
2118 | if (ret) | |
2119 | break; | |
2120 | } | |
2121 | ret = 0; | |
e02119d5 | 2122 | } |
bb53eda9 FM |
2123 | btrfs_free_path(fixup_path); |
2124 | return ret; | |
e02119d5 CM |
2125 | } |
2126 | ||
2127 | /* | |
2128 | * directory replay has two parts. There are the standard directory | |
2129 | * items in the log copied from the subvolume, and range items | |
2130 | * created in the log while the subvolume was logged. | |
2131 | * | |
2132 | * The range items tell us which parts of the key space the log | |
2133 | * is authoritative for. During replay, if a key in the subvolume | |
2134 | * directory is in a logged range item, but not actually in the log | |
2135 | * that means it was deleted from the directory before the fsync | |
2136 | * and should be removed. | |
2137 | */ | |
2138 | static noinline int find_dir_range(struct btrfs_root *root, | |
2139 | struct btrfs_path *path, | |
2140 | u64 dirid, int key_type, | |
2141 | u64 *start_ret, u64 *end_ret) | |
2142 | { | |
2143 | struct btrfs_key key; | |
2144 | u64 found_end; | |
2145 | struct btrfs_dir_log_item *item; | |
2146 | int ret; | |
2147 | int nritems; | |
2148 | ||
2149 | if (*start_ret == (u64)-1) | |
2150 | return 1; | |
2151 | ||
2152 | key.objectid = dirid; | |
2153 | key.type = key_type; | |
2154 | key.offset = *start_ret; | |
2155 | ||
2156 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2157 | if (ret < 0) | |
2158 | goto out; | |
2159 | if (ret > 0) { | |
2160 | if (path->slots[0] == 0) | |
2161 | goto out; | |
2162 | path->slots[0]--; | |
2163 | } | |
2164 | if (ret != 0) | |
2165 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2166 | ||
2167 | if (key.type != key_type || key.objectid != dirid) { | |
2168 | ret = 1; | |
2169 | goto next; | |
2170 | } | |
2171 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2172 | struct btrfs_dir_log_item); | |
2173 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2174 | ||
2175 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
2176 | ret = 0; | |
2177 | *start_ret = key.offset; | |
2178 | *end_ret = found_end; | |
2179 | goto out; | |
2180 | } | |
2181 | ret = 1; | |
2182 | next: | |
2183 | /* check the next slot in the tree to see if it is a valid item */ | |
2184 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2a7bf53f | 2185 | path->slots[0]++; |
e02119d5 CM |
2186 | if (path->slots[0] >= nritems) { |
2187 | ret = btrfs_next_leaf(root, path); | |
2188 | if (ret) | |
2189 | goto out; | |
e02119d5 CM |
2190 | } |
2191 | ||
2192 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
2193 | ||
2194 | if (key.type != key_type || key.objectid != dirid) { | |
2195 | ret = 1; | |
2196 | goto out; | |
2197 | } | |
2198 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2199 | struct btrfs_dir_log_item); | |
2200 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
2201 | *start_ret = key.offset; | |
2202 | *end_ret = found_end; | |
2203 | ret = 0; | |
2204 | out: | |
b3b4aa74 | 2205 | btrfs_release_path(path); |
e02119d5 CM |
2206 | return ret; |
2207 | } | |
2208 | ||
2209 | /* | |
2210 | * this looks for a given directory item in the log. If the directory | |
2211 | * item is not in the log, the item is removed and the inode it points | |
2212 | * to is unlinked | |
2213 | */ | |
2214 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
2215 | struct btrfs_root *root, | |
2216 | struct btrfs_root *log, | |
2217 | struct btrfs_path *path, | |
2218 | struct btrfs_path *log_path, | |
2219 | struct inode *dir, | |
2220 | struct btrfs_key *dir_key) | |
2221 | { | |
2222 | int ret; | |
2223 | struct extent_buffer *eb; | |
2224 | int slot; | |
2225 | u32 item_size; | |
2226 | struct btrfs_dir_item *di; | |
2227 | struct btrfs_dir_item *log_di; | |
2228 | int name_len; | |
2229 | unsigned long ptr; | |
2230 | unsigned long ptr_end; | |
2231 | char *name; | |
2232 | struct inode *inode; | |
2233 | struct btrfs_key location; | |
2234 | ||
2235 | again: | |
2236 | eb = path->nodes[0]; | |
2237 | slot = path->slots[0]; | |
2238 | item_size = btrfs_item_size_nr(eb, slot); | |
2239 | ptr = btrfs_item_ptr_offset(eb, slot); | |
2240 | ptr_end = ptr + item_size; | |
d397712b | 2241 | while (ptr < ptr_end) { |
e02119d5 CM |
2242 | di = (struct btrfs_dir_item *)ptr; |
2243 | name_len = btrfs_dir_name_len(eb, di); | |
2244 | name = kmalloc(name_len, GFP_NOFS); | |
2245 | if (!name) { | |
2246 | ret = -ENOMEM; | |
2247 | goto out; | |
2248 | } | |
2249 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2250 | name_len); | |
2251 | log_di = NULL; | |
12fcfd22 | 2252 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2253 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
2254 | dir_key->objectid, | |
2255 | name, name_len, 0); | |
12fcfd22 | 2256 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
2257 | log_di = btrfs_lookup_dir_index_item(trans, log, |
2258 | log_path, | |
2259 | dir_key->objectid, | |
2260 | dir_key->offset, | |
2261 | name, name_len, 0); | |
2262 | } | |
8d9e220c | 2263 | if (!log_di || log_di == ERR_PTR(-ENOENT)) { |
e02119d5 | 2264 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
b3b4aa74 DS |
2265 | btrfs_release_path(path); |
2266 | btrfs_release_path(log_path); | |
e02119d5 | 2267 | inode = read_one_inode(root, location.objectid); |
c00e9493 TI |
2268 | if (!inode) { |
2269 | kfree(name); | |
2270 | return -EIO; | |
2271 | } | |
e02119d5 CM |
2272 | |
2273 | ret = link_to_fixup_dir(trans, root, | |
2274 | path, location.objectid); | |
3650860b JB |
2275 | if (ret) { |
2276 | kfree(name); | |
2277 | iput(inode); | |
2278 | goto out; | |
2279 | } | |
2280 | ||
8b558c5f | 2281 | inc_nlink(inode); |
4ec5934e NB |
2282 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
2283 | BTRFS_I(inode), name, name_len); | |
3650860b | 2284 | if (!ret) |
e5c304e6 | 2285 | ret = btrfs_run_delayed_items(trans); |
e02119d5 CM |
2286 | kfree(name); |
2287 | iput(inode); | |
3650860b JB |
2288 | if (ret) |
2289 | goto out; | |
e02119d5 CM |
2290 | |
2291 | /* there might still be more names under this key | |
2292 | * check and repeat if required | |
2293 | */ | |
2294 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2295 | 0, 0); | |
2296 | if (ret == 0) | |
2297 | goto again; | |
2298 | ret = 0; | |
2299 | goto out; | |
269d040f FDBM |
2300 | } else if (IS_ERR(log_di)) { |
2301 | kfree(name); | |
2302 | return PTR_ERR(log_di); | |
e02119d5 | 2303 | } |
b3b4aa74 | 2304 | btrfs_release_path(log_path); |
e02119d5 CM |
2305 | kfree(name); |
2306 | ||
2307 | ptr = (unsigned long)(di + 1); | |
2308 | ptr += name_len; | |
2309 | } | |
2310 | ret = 0; | |
2311 | out: | |
b3b4aa74 DS |
2312 | btrfs_release_path(path); |
2313 | btrfs_release_path(log_path); | |
e02119d5 CM |
2314 | return ret; |
2315 | } | |
2316 | ||
4f764e51 FM |
2317 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, |
2318 | struct btrfs_root *root, | |
2319 | struct btrfs_root *log, | |
2320 | struct btrfs_path *path, | |
2321 | const u64 ino) | |
2322 | { | |
2323 | struct btrfs_key search_key; | |
2324 | struct btrfs_path *log_path; | |
2325 | int i; | |
2326 | int nritems; | |
2327 | int ret; | |
2328 | ||
2329 | log_path = btrfs_alloc_path(); | |
2330 | if (!log_path) | |
2331 | return -ENOMEM; | |
2332 | ||
2333 | search_key.objectid = ino; | |
2334 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2335 | search_key.offset = 0; | |
2336 | again: | |
2337 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2338 | if (ret < 0) | |
2339 | goto out; | |
2340 | process_leaf: | |
2341 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2342 | for (i = path->slots[0]; i < nritems; i++) { | |
2343 | struct btrfs_key key; | |
2344 | struct btrfs_dir_item *di; | |
2345 | struct btrfs_dir_item *log_di; | |
2346 | u32 total_size; | |
2347 | u32 cur; | |
2348 | ||
2349 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2350 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2351 | ret = 0; | |
2352 | goto out; | |
2353 | } | |
2354 | ||
2355 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2356 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2357 | cur = 0; | |
2358 | while (cur < total_size) { | |
2359 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2360 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2361 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2362 | char *name; | |
2363 | ||
2364 | name = kmalloc(name_len, GFP_NOFS); | |
2365 | if (!name) { | |
2366 | ret = -ENOMEM; | |
2367 | goto out; | |
2368 | } | |
2369 | read_extent_buffer(path->nodes[0], name, | |
2370 | (unsigned long)(di + 1), name_len); | |
2371 | ||
2372 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2373 | name, name_len, 0); | |
2374 | btrfs_release_path(log_path); | |
2375 | if (!log_di) { | |
2376 | /* Doesn't exist in log tree, so delete it. */ | |
2377 | btrfs_release_path(path); | |
2378 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2379 | name, name_len, -1); | |
2380 | kfree(name); | |
2381 | if (IS_ERR(di)) { | |
2382 | ret = PTR_ERR(di); | |
2383 | goto out; | |
2384 | } | |
2385 | ASSERT(di); | |
2386 | ret = btrfs_delete_one_dir_name(trans, root, | |
2387 | path, di); | |
2388 | if (ret) | |
2389 | goto out; | |
2390 | btrfs_release_path(path); | |
2391 | search_key = key; | |
2392 | goto again; | |
2393 | } | |
2394 | kfree(name); | |
2395 | if (IS_ERR(log_di)) { | |
2396 | ret = PTR_ERR(log_di); | |
2397 | goto out; | |
2398 | } | |
2399 | cur += this_len; | |
2400 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2401 | } | |
2402 | } | |
2403 | ret = btrfs_next_leaf(root, path); | |
2404 | if (ret > 0) | |
2405 | ret = 0; | |
2406 | else if (ret == 0) | |
2407 | goto process_leaf; | |
2408 | out: | |
2409 | btrfs_free_path(log_path); | |
2410 | btrfs_release_path(path); | |
2411 | return ret; | |
2412 | } | |
2413 | ||
2414 | ||
e02119d5 CM |
2415 | /* |
2416 | * deletion replay happens before we copy any new directory items | |
2417 | * out of the log or out of backreferences from inodes. It | |
2418 | * scans the log to find ranges of keys that log is authoritative for, | |
2419 | * and then scans the directory to find items in those ranges that are | |
2420 | * not present in the log. | |
2421 | * | |
2422 | * Anything we don't find in the log is unlinked and removed from the | |
2423 | * directory. | |
2424 | */ | |
2425 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2426 | struct btrfs_root *root, | |
2427 | struct btrfs_root *log, | |
2428 | struct btrfs_path *path, | |
12fcfd22 | 2429 | u64 dirid, int del_all) |
e02119d5 CM |
2430 | { |
2431 | u64 range_start; | |
2432 | u64 range_end; | |
2433 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2434 | int ret = 0; | |
2435 | struct btrfs_key dir_key; | |
2436 | struct btrfs_key found_key; | |
2437 | struct btrfs_path *log_path; | |
2438 | struct inode *dir; | |
2439 | ||
2440 | dir_key.objectid = dirid; | |
2441 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2442 | log_path = btrfs_alloc_path(); | |
2443 | if (!log_path) | |
2444 | return -ENOMEM; | |
2445 | ||
2446 | dir = read_one_inode(root, dirid); | |
2447 | /* it isn't an error if the inode isn't there, that can happen | |
2448 | * because we replay the deletes before we copy in the inode item | |
2449 | * from the log | |
2450 | */ | |
2451 | if (!dir) { | |
2452 | btrfs_free_path(log_path); | |
2453 | return 0; | |
2454 | } | |
2455 | again: | |
2456 | range_start = 0; | |
2457 | range_end = 0; | |
d397712b | 2458 | while (1) { |
12fcfd22 CM |
2459 | if (del_all) |
2460 | range_end = (u64)-1; | |
2461 | else { | |
2462 | ret = find_dir_range(log, path, dirid, key_type, | |
2463 | &range_start, &range_end); | |
2464 | if (ret != 0) | |
2465 | break; | |
2466 | } | |
e02119d5 CM |
2467 | |
2468 | dir_key.offset = range_start; | |
d397712b | 2469 | while (1) { |
e02119d5 CM |
2470 | int nritems; |
2471 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2472 | 0, 0); | |
2473 | if (ret < 0) | |
2474 | goto out; | |
2475 | ||
2476 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2477 | if (path->slots[0] >= nritems) { | |
2478 | ret = btrfs_next_leaf(root, path); | |
b98def7c | 2479 | if (ret == 1) |
e02119d5 | 2480 | break; |
b98def7c LB |
2481 | else if (ret < 0) |
2482 | goto out; | |
e02119d5 CM |
2483 | } |
2484 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2485 | path->slots[0]); | |
2486 | if (found_key.objectid != dirid || | |
2487 | found_key.type != dir_key.type) | |
2488 | goto next_type; | |
2489 | ||
2490 | if (found_key.offset > range_end) | |
2491 | break; | |
2492 | ||
2493 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
2494 | log_path, dir, |
2495 | &found_key); | |
3650860b JB |
2496 | if (ret) |
2497 | goto out; | |
e02119d5 CM |
2498 | if (found_key.offset == (u64)-1) |
2499 | break; | |
2500 | dir_key.offset = found_key.offset + 1; | |
2501 | } | |
b3b4aa74 | 2502 | btrfs_release_path(path); |
e02119d5 CM |
2503 | if (range_end == (u64)-1) |
2504 | break; | |
2505 | range_start = range_end + 1; | |
2506 | } | |
2507 | ||
2508 | next_type: | |
2509 | ret = 0; | |
2510 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2511 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2512 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
b3b4aa74 | 2513 | btrfs_release_path(path); |
e02119d5 CM |
2514 | goto again; |
2515 | } | |
2516 | out: | |
b3b4aa74 | 2517 | btrfs_release_path(path); |
e02119d5 CM |
2518 | btrfs_free_path(log_path); |
2519 | iput(dir); | |
2520 | return ret; | |
2521 | } | |
2522 | ||
2523 | /* | |
2524 | * the process_func used to replay items from the log tree. This | |
2525 | * gets called in two different stages. The first stage just looks | |
2526 | * for inodes and makes sure they are all copied into the subvolume. | |
2527 | * | |
2528 | * The second stage copies all the other item types from the log into | |
2529 | * the subvolume. The two stage approach is slower, but gets rid of | |
2530 | * lots of complexity around inodes referencing other inodes that exist | |
2531 | * only in the log (references come from either directory items or inode | |
2532 | * back refs). | |
2533 | */ | |
2534 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
581c1760 | 2535 | struct walk_control *wc, u64 gen, int level) |
e02119d5 CM |
2536 | { |
2537 | int nritems; | |
2538 | struct btrfs_path *path; | |
2539 | struct btrfs_root *root = wc->replay_dest; | |
2540 | struct btrfs_key key; | |
e02119d5 CM |
2541 | int i; |
2542 | int ret; | |
2543 | ||
581c1760 | 2544 | ret = btrfs_read_buffer(eb, gen, level, NULL); |
018642a1 TI |
2545 | if (ret) |
2546 | return ret; | |
e02119d5 CM |
2547 | |
2548 | level = btrfs_header_level(eb); | |
2549 | ||
2550 | if (level != 0) | |
2551 | return 0; | |
2552 | ||
2553 | path = btrfs_alloc_path(); | |
1e5063d0 MF |
2554 | if (!path) |
2555 | return -ENOMEM; | |
e02119d5 CM |
2556 | |
2557 | nritems = btrfs_header_nritems(eb); | |
2558 | for (i = 0; i < nritems; i++) { | |
2559 | btrfs_item_key_to_cpu(eb, &key, i); | |
e02119d5 CM |
2560 | |
2561 | /* inode keys are done during the first stage */ | |
2562 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2563 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
e02119d5 CM |
2564 | struct btrfs_inode_item *inode_item; |
2565 | u32 mode; | |
2566 | ||
2567 | inode_item = btrfs_item_ptr(eb, i, | |
2568 | struct btrfs_inode_item); | |
f2d72f42 FM |
2569 | /* |
2570 | * If we have a tmpfile (O_TMPFILE) that got fsync'ed | |
2571 | * and never got linked before the fsync, skip it, as | |
2572 | * replaying it is pointless since it would be deleted | |
2573 | * later. We skip logging tmpfiles, but it's always | |
2574 | * possible we are replaying a log created with a kernel | |
2575 | * that used to log tmpfiles. | |
2576 | */ | |
2577 | if (btrfs_inode_nlink(eb, inode_item) == 0) { | |
2578 | wc->ignore_cur_inode = true; | |
2579 | continue; | |
2580 | } else { | |
2581 | wc->ignore_cur_inode = false; | |
2582 | } | |
4f764e51 FM |
2583 | ret = replay_xattr_deletes(wc->trans, root, log, |
2584 | path, key.objectid); | |
2585 | if (ret) | |
2586 | break; | |
e02119d5 CM |
2587 | mode = btrfs_inode_mode(eb, inode_item); |
2588 | if (S_ISDIR(mode)) { | |
2589 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 2590 | root, log, path, key.objectid, 0); |
b50c6e25 JB |
2591 | if (ret) |
2592 | break; | |
e02119d5 CM |
2593 | } |
2594 | ret = overwrite_item(wc->trans, root, path, | |
2595 | eb, i, &key); | |
b50c6e25 JB |
2596 | if (ret) |
2597 | break; | |
e02119d5 | 2598 | |
471d557a FM |
2599 | /* |
2600 | * Before replaying extents, truncate the inode to its | |
2601 | * size. We need to do it now and not after log replay | |
2602 | * because before an fsync we can have prealloc extents | |
2603 | * added beyond the inode's i_size. If we did it after, | |
2604 | * through orphan cleanup for example, we would drop | |
2605 | * those prealloc extents just after replaying them. | |
e02119d5 CM |
2606 | */ |
2607 | if (S_ISREG(mode)) { | |
471d557a FM |
2608 | struct inode *inode; |
2609 | u64 from; | |
2610 | ||
2611 | inode = read_one_inode(root, key.objectid); | |
2612 | if (!inode) { | |
2613 | ret = -EIO; | |
2614 | break; | |
2615 | } | |
2616 | from = ALIGN(i_size_read(inode), | |
2617 | root->fs_info->sectorsize); | |
2618 | ret = btrfs_drop_extents(wc->trans, root, inode, | |
2619 | from, (u64)-1, 1); | |
471d557a | 2620 | if (!ret) { |
f2d72f42 | 2621 | /* Update the inode's nbytes. */ |
471d557a FM |
2622 | ret = btrfs_update_inode(wc->trans, |
2623 | root, inode); | |
2624 | } | |
2625 | iput(inode); | |
b50c6e25 JB |
2626 | if (ret) |
2627 | break; | |
e02119d5 | 2628 | } |
c71bf099 | 2629 | |
e02119d5 CM |
2630 | ret = link_to_fixup_dir(wc->trans, root, |
2631 | path, key.objectid); | |
b50c6e25 JB |
2632 | if (ret) |
2633 | break; | |
e02119d5 | 2634 | } |
dd8e7217 | 2635 | |
f2d72f42 FM |
2636 | if (wc->ignore_cur_inode) |
2637 | continue; | |
2638 | ||
dd8e7217 JB |
2639 | if (key.type == BTRFS_DIR_INDEX_KEY && |
2640 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2641 | ret = replay_one_dir_item(wc->trans, root, path, | |
2642 | eb, i, &key); | |
2643 | if (ret) | |
2644 | break; | |
2645 | } | |
2646 | ||
e02119d5 CM |
2647 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
2648 | continue; | |
2649 | ||
2650 | /* these keys are simply copied */ | |
2651 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2652 | ret = overwrite_item(wc->trans, root, path, | |
2653 | eb, i, &key); | |
b50c6e25 JB |
2654 | if (ret) |
2655 | break; | |
2da1c669 LB |
2656 | } else if (key.type == BTRFS_INODE_REF_KEY || |
2657 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
f186373f MF |
2658 | ret = add_inode_ref(wc->trans, root, log, path, |
2659 | eb, i, &key); | |
b50c6e25 JB |
2660 | if (ret && ret != -ENOENT) |
2661 | break; | |
2662 | ret = 0; | |
e02119d5 CM |
2663 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
2664 | ret = replay_one_extent(wc->trans, root, path, | |
2665 | eb, i, &key); | |
b50c6e25 JB |
2666 | if (ret) |
2667 | break; | |
dd8e7217 | 2668 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
2669 | ret = replay_one_dir_item(wc->trans, root, path, |
2670 | eb, i, &key); | |
b50c6e25 JB |
2671 | if (ret) |
2672 | break; | |
e02119d5 CM |
2673 | } |
2674 | } | |
2675 | btrfs_free_path(path); | |
b50c6e25 | 2676 | return ret; |
e02119d5 CM |
2677 | } |
2678 | ||
d397712b | 2679 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2680 | struct btrfs_root *root, |
2681 | struct btrfs_path *path, int *level, | |
2682 | struct walk_control *wc) | |
2683 | { | |
0b246afa | 2684 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2685 | u64 root_owner; |
e02119d5 CM |
2686 | u64 bytenr; |
2687 | u64 ptr_gen; | |
2688 | struct extent_buffer *next; | |
2689 | struct extent_buffer *cur; | |
2690 | struct extent_buffer *parent; | |
2691 | u32 blocksize; | |
2692 | int ret = 0; | |
2693 | ||
2694 | WARN_ON(*level < 0); | |
2695 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2696 | ||
d397712b | 2697 | while (*level > 0) { |
581c1760 QW |
2698 | struct btrfs_key first_key; |
2699 | ||
e02119d5 CM |
2700 | WARN_ON(*level < 0); |
2701 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2702 | cur = path->nodes[*level]; | |
2703 | ||
fae7f21c | 2704 | WARN_ON(btrfs_header_level(cur) != *level); |
e02119d5 CM |
2705 | |
2706 | if (path->slots[*level] >= | |
2707 | btrfs_header_nritems(cur)) | |
2708 | break; | |
2709 | ||
2710 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2711 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
581c1760 | 2712 | btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]); |
0b246afa | 2713 | blocksize = fs_info->nodesize; |
e02119d5 CM |
2714 | |
2715 | parent = path->nodes[*level]; | |
2716 | root_owner = btrfs_header_owner(parent); | |
e02119d5 | 2717 | |
2ff7e61e | 2718 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
2719 | if (IS_ERR(next)) |
2720 | return PTR_ERR(next); | |
e02119d5 | 2721 | |
e02119d5 | 2722 | if (*level == 1) { |
581c1760 QW |
2723 | ret = wc->process_func(root, next, wc, ptr_gen, |
2724 | *level - 1); | |
b50c6e25 JB |
2725 | if (ret) { |
2726 | free_extent_buffer(next); | |
1e5063d0 | 2727 | return ret; |
b50c6e25 | 2728 | } |
4a500fd1 | 2729 | |
e02119d5 CM |
2730 | path->slots[*level]++; |
2731 | if (wc->free) { | |
581c1760 QW |
2732 | ret = btrfs_read_buffer(next, ptr_gen, |
2733 | *level - 1, &first_key); | |
018642a1 TI |
2734 | if (ret) { |
2735 | free_extent_buffer(next); | |
2736 | return ret; | |
2737 | } | |
e02119d5 | 2738 | |
681ae509 JB |
2739 | if (trans) { |
2740 | btrfs_tree_lock(next); | |
8bead258 | 2741 | btrfs_set_lock_blocking_write(next); |
6a884d7d | 2742 | btrfs_clean_tree_block(next); |
681ae509 JB |
2743 | btrfs_wait_tree_block_writeback(next); |
2744 | btrfs_tree_unlock(next); | |
1846430c LB |
2745 | } else { |
2746 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2747 | clear_extent_buffer_dirty(next); | |
681ae509 | 2748 | } |
e02119d5 | 2749 | |
e02119d5 CM |
2750 | WARN_ON(root_owner != |
2751 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2752 | ret = btrfs_free_and_pin_reserved_extent( |
2753 | fs_info, bytenr, | |
2754 | blocksize); | |
3650860b JB |
2755 | if (ret) { |
2756 | free_extent_buffer(next); | |
2757 | return ret; | |
2758 | } | |
e02119d5 CM |
2759 | } |
2760 | free_extent_buffer(next); | |
2761 | continue; | |
2762 | } | |
581c1760 | 2763 | ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key); |
018642a1 TI |
2764 | if (ret) { |
2765 | free_extent_buffer(next); | |
2766 | return ret; | |
2767 | } | |
e02119d5 CM |
2768 | |
2769 | WARN_ON(*level <= 0); | |
2770 | if (path->nodes[*level-1]) | |
2771 | free_extent_buffer(path->nodes[*level-1]); | |
2772 | path->nodes[*level-1] = next; | |
2773 | *level = btrfs_header_level(next); | |
2774 | path->slots[*level] = 0; | |
2775 | cond_resched(); | |
2776 | } | |
2777 | WARN_ON(*level < 0); | |
2778 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2779 | ||
4a500fd1 | 2780 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
e02119d5 CM |
2781 | |
2782 | cond_resched(); | |
2783 | return 0; | |
2784 | } | |
2785 | ||
d397712b | 2786 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2787 | struct btrfs_root *root, |
2788 | struct btrfs_path *path, int *level, | |
2789 | struct walk_control *wc) | |
2790 | { | |
0b246afa | 2791 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 2792 | u64 root_owner; |
e02119d5 CM |
2793 | int i; |
2794 | int slot; | |
2795 | int ret; | |
2796 | ||
d397712b | 2797 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 | 2798 | slot = path->slots[i]; |
4a500fd1 | 2799 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
e02119d5 CM |
2800 | path->slots[i]++; |
2801 | *level = i; | |
2802 | WARN_ON(*level == 0); | |
2803 | return 0; | |
2804 | } else { | |
31840ae1 ZY |
2805 | struct extent_buffer *parent; |
2806 | if (path->nodes[*level] == root->node) | |
2807 | parent = path->nodes[*level]; | |
2808 | else | |
2809 | parent = path->nodes[*level + 1]; | |
2810 | ||
2811 | root_owner = btrfs_header_owner(parent); | |
1e5063d0 | 2812 | ret = wc->process_func(root, path->nodes[*level], wc, |
581c1760 QW |
2813 | btrfs_header_generation(path->nodes[*level]), |
2814 | *level); | |
1e5063d0 MF |
2815 | if (ret) |
2816 | return ret; | |
2817 | ||
e02119d5 CM |
2818 | if (wc->free) { |
2819 | struct extent_buffer *next; | |
2820 | ||
2821 | next = path->nodes[*level]; | |
2822 | ||
681ae509 JB |
2823 | if (trans) { |
2824 | btrfs_tree_lock(next); | |
8bead258 | 2825 | btrfs_set_lock_blocking_write(next); |
6a884d7d | 2826 | btrfs_clean_tree_block(next); |
681ae509 JB |
2827 | btrfs_wait_tree_block_writeback(next); |
2828 | btrfs_tree_unlock(next); | |
1846430c LB |
2829 | } else { |
2830 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2831 | clear_extent_buffer_dirty(next); | |
681ae509 | 2832 | } |
e02119d5 | 2833 | |
e02119d5 | 2834 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
2ff7e61e JM |
2835 | ret = btrfs_free_and_pin_reserved_extent( |
2836 | fs_info, | |
e02119d5 | 2837 | path->nodes[*level]->start, |
d00aff00 | 2838 | path->nodes[*level]->len); |
3650860b JB |
2839 | if (ret) |
2840 | return ret; | |
e02119d5 CM |
2841 | } |
2842 | free_extent_buffer(path->nodes[*level]); | |
2843 | path->nodes[*level] = NULL; | |
2844 | *level = i + 1; | |
2845 | } | |
2846 | } | |
2847 | return 1; | |
2848 | } | |
2849 | ||
2850 | /* | |
2851 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2852 | * the tree freeing any blocks that have a ref count of zero after being | |
2853 | * decremented. | |
2854 | */ | |
2855 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2856 | struct btrfs_root *log, struct walk_control *wc) | |
2857 | { | |
2ff7e61e | 2858 | struct btrfs_fs_info *fs_info = log->fs_info; |
e02119d5 CM |
2859 | int ret = 0; |
2860 | int wret; | |
2861 | int level; | |
2862 | struct btrfs_path *path; | |
e02119d5 CM |
2863 | int orig_level; |
2864 | ||
2865 | path = btrfs_alloc_path(); | |
db5b493a TI |
2866 | if (!path) |
2867 | return -ENOMEM; | |
e02119d5 CM |
2868 | |
2869 | level = btrfs_header_level(log->node); | |
2870 | orig_level = level; | |
2871 | path->nodes[level] = log->node; | |
2872 | extent_buffer_get(log->node); | |
2873 | path->slots[level] = 0; | |
2874 | ||
d397712b | 2875 | while (1) { |
e02119d5 CM |
2876 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
2877 | if (wret > 0) | |
2878 | break; | |
79787eaa | 2879 | if (wret < 0) { |
e02119d5 | 2880 | ret = wret; |
79787eaa JM |
2881 | goto out; |
2882 | } | |
e02119d5 CM |
2883 | |
2884 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2885 | if (wret > 0) | |
2886 | break; | |
79787eaa | 2887 | if (wret < 0) { |
e02119d5 | 2888 | ret = wret; |
79787eaa JM |
2889 | goto out; |
2890 | } | |
e02119d5 CM |
2891 | } |
2892 | ||
2893 | /* was the root node processed? if not, catch it here */ | |
2894 | if (path->nodes[orig_level]) { | |
79787eaa | 2895 | ret = wc->process_func(log, path->nodes[orig_level], wc, |
581c1760 QW |
2896 | btrfs_header_generation(path->nodes[orig_level]), |
2897 | orig_level); | |
79787eaa JM |
2898 | if (ret) |
2899 | goto out; | |
e02119d5 CM |
2900 | if (wc->free) { |
2901 | struct extent_buffer *next; | |
2902 | ||
2903 | next = path->nodes[orig_level]; | |
2904 | ||
681ae509 JB |
2905 | if (trans) { |
2906 | btrfs_tree_lock(next); | |
8bead258 | 2907 | btrfs_set_lock_blocking_write(next); |
6a884d7d | 2908 | btrfs_clean_tree_block(next); |
681ae509 JB |
2909 | btrfs_wait_tree_block_writeback(next); |
2910 | btrfs_tree_unlock(next); | |
1846430c LB |
2911 | } else { |
2912 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags)) | |
2913 | clear_extent_buffer_dirty(next); | |
681ae509 | 2914 | } |
e02119d5 | 2915 | |
e02119d5 CM |
2916 | WARN_ON(log->root_key.objectid != |
2917 | BTRFS_TREE_LOG_OBJECTID); | |
2ff7e61e JM |
2918 | ret = btrfs_free_and_pin_reserved_extent(fs_info, |
2919 | next->start, next->len); | |
3650860b JB |
2920 | if (ret) |
2921 | goto out; | |
e02119d5 CM |
2922 | } |
2923 | } | |
2924 | ||
79787eaa | 2925 | out: |
e02119d5 | 2926 | btrfs_free_path(path); |
e02119d5 CM |
2927 | return ret; |
2928 | } | |
2929 | ||
7237f183 YZ |
2930 | /* |
2931 | * helper function to update the item for a given subvolumes log root | |
2932 | * in the tree of log roots | |
2933 | */ | |
2934 | static int update_log_root(struct btrfs_trans_handle *trans, | |
2935 | struct btrfs_root *log) | |
2936 | { | |
0b246afa | 2937 | struct btrfs_fs_info *fs_info = log->fs_info; |
7237f183 YZ |
2938 | int ret; |
2939 | ||
2940 | if (log->log_transid == 1) { | |
2941 | /* insert root item on the first sync */ | |
0b246afa | 2942 | ret = btrfs_insert_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2943 | &log->root_key, &log->root_item); |
2944 | } else { | |
0b246afa | 2945 | ret = btrfs_update_root(trans, fs_info->log_root_tree, |
7237f183 YZ |
2946 | &log->root_key, &log->root_item); |
2947 | } | |
2948 | return ret; | |
2949 | } | |
2950 | ||
60d53eb3 | 2951 | static void wait_log_commit(struct btrfs_root *root, int transid) |
e02119d5 CM |
2952 | { |
2953 | DEFINE_WAIT(wait); | |
7237f183 | 2954 | int index = transid % 2; |
e02119d5 | 2955 | |
7237f183 YZ |
2956 | /* |
2957 | * we only allow two pending log transactions at a time, | |
2958 | * so we know that if ours is more than 2 older than the | |
2959 | * current transaction, we're done | |
2960 | */ | |
49e83f57 | 2961 | for (;;) { |
7237f183 YZ |
2962 | prepare_to_wait(&root->log_commit_wait[index], |
2963 | &wait, TASK_UNINTERRUPTIBLE); | |
12fcfd22 | 2964 | |
49e83f57 LB |
2965 | if (!(root->log_transid_committed < transid && |
2966 | atomic_read(&root->log_commit[index]))) | |
2967 | break; | |
12fcfd22 | 2968 | |
49e83f57 LB |
2969 | mutex_unlock(&root->log_mutex); |
2970 | schedule(); | |
7237f183 | 2971 | mutex_lock(&root->log_mutex); |
49e83f57 LB |
2972 | } |
2973 | finish_wait(&root->log_commit_wait[index], &wait); | |
7237f183 YZ |
2974 | } |
2975 | ||
60d53eb3 | 2976 | static void wait_for_writer(struct btrfs_root *root) |
7237f183 YZ |
2977 | { |
2978 | DEFINE_WAIT(wait); | |
8b050d35 | 2979 | |
49e83f57 LB |
2980 | for (;;) { |
2981 | prepare_to_wait(&root->log_writer_wait, &wait, | |
2982 | TASK_UNINTERRUPTIBLE); | |
2983 | if (!atomic_read(&root->log_writers)) | |
2984 | break; | |
2985 | ||
7237f183 | 2986 | mutex_unlock(&root->log_mutex); |
49e83f57 | 2987 | schedule(); |
575849ec | 2988 | mutex_lock(&root->log_mutex); |
7237f183 | 2989 | } |
49e83f57 | 2990 | finish_wait(&root->log_writer_wait, &wait); |
e02119d5 CM |
2991 | } |
2992 | ||
8b050d35 MX |
2993 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, |
2994 | struct btrfs_log_ctx *ctx) | |
2995 | { | |
2996 | if (!ctx) | |
2997 | return; | |
2998 | ||
2999 | mutex_lock(&root->log_mutex); | |
3000 | list_del_init(&ctx->list); | |
3001 | mutex_unlock(&root->log_mutex); | |
3002 | } | |
3003 | ||
3004 | /* | |
3005 | * Invoked in log mutex context, or be sure there is no other task which | |
3006 | * can access the list. | |
3007 | */ | |
3008 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
3009 | int index, int error) | |
3010 | { | |
3011 | struct btrfs_log_ctx *ctx; | |
570dd450 | 3012 | struct btrfs_log_ctx *safe; |
8b050d35 | 3013 | |
570dd450 CM |
3014 | list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) { |
3015 | list_del_init(&ctx->list); | |
8b050d35 | 3016 | ctx->log_ret = error; |
570dd450 | 3017 | } |
8b050d35 MX |
3018 | |
3019 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
3020 | } | |
3021 | ||
e02119d5 CM |
3022 | /* |
3023 | * btrfs_sync_log does sends a given tree log down to the disk and | |
3024 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
3025 | * you know that any inodes previously logged are safely on disk only |
3026 | * if it returns 0. | |
3027 | * | |
3028 | * Any other return value means you need to call btrfs_commit_transaction. | |
3029 | * Some of the edge cases for fsyncing directories that have had unlinks | |
3030 | * or renames done in the past mean that sometimes the only safe | |
3031 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
3032 | * that has happened. | |
e02119d5 CM |
3033 | */ |
3034 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
8b050d35 | 3035 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) |
e02119d5 | 3036 | { |
7237f183 YZ |
3037 | int index1; |
3038 | int index2; | |
8cef4e16 | 3039 | int mark; |
e02119d5 | 3040 | int ret; |
0b246afa | 3041 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 3042 | struct btrfs_root *log = root->log_root; |
0b246afa | 3043 | struct btrfs_root *log_root_tree = fs_info->log_root_tree; |
bb14a59b | 3044 | int log_transid = 0; |
8b050d35 | 3045 | struct btrfs_log_ctx root_log_ctx; |
c6adc9cc | 3046 | struct blk_plug plug; |
e02119d5 | 3047 | |
7237f183 | 3048 | mutex_lock(&root->log_mutex); |
d1433deb MX |
3049 | log_transid = ctx->log_transid; |
3050 | if (root->log_transid_committed >= log_transid) { | |
3051 | mutex_unlock(&root->log_mutex); | |
3052 | return ctx->log_ret; | |
3053 | } | |
3054 | ||
3055 | index1 = log_transid % 2; | |
7237f183 | 3056 | if (atomic_read(&root->log_commit[index1])) { |
60d53eb3 | 3057 | wait_log_commit(root, log_transid); |
7237f183 | 3058 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3059 | return ctx->log_ret; |
e02119d5 | 3060 | } |
d1433deb | 3061 | ASSERT(log_transid == root->log_transid); |
7237f183 YZ |
3062 | atomic_set(&root->log_commit[index1], 1); |
3063 | ||
3064 | /* wait for previous tree log sync to complete */ | |
3065 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
60d53eb3 | 3066 | wait_log_commit(root, log_transid - 1); |
48cab2e0 | 3067 | |
86df7eb9 | 3068 | while (1) { |
2ecb7923 | 3069 | int batch = atomic_read(&root->log_batch); |
cd354ad6 | 3070 | /* when we're on an ssd, just kick the log commit out */ |
0b246afa | 3071 | if (!btrfs_test_opt(fs_info, SSD) && |
27cdeb70 | 3072 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { |
86df7eb9 YZ |
3073 | mutex_unlock(&root->log_mutex); |
3074 | schedule_timeout_uninterruptible(1); | |
3075 | mutex_lock(&root->log_mutex); | |
3076 | } | |
60d53eb3 | 3077 | wait_for_writer(root); |
2ecb7923 | 3078 | if (batch == atomic_read(&root->log_batch)) |
e02119d5 CM |
3079 | break; |
3080 | } | |
e02119d5 | 3081 | |
12fcfd22 | 3082 | /* bail out if we need to do a full commit */ |
4884b8e8 | 3083 | if (btrfs_need_log_full_commit(trans)) { |
12fcfd22 CM |
3084 | ret = -EAGAIN; |
3085 | mutex_unlock(&root->log_mutex); | |
3086 | goto out; | |
3087 | } | |
3088 | ||
8cef4e16 YZ |
3089 | if (log_transid % 2 == 0) |
3090 | mark = EXTENT_DIRTY; | |
3091 | else | |
3092 | mark = EXTENT_NEW; | |
3093 | ||
690587d1 CM |
3094 | /* we start IO on all the marked extents here, but we don't actually |
3095 | * wait for them until later. | |
3096 | */ | |
c6adc9cc | 3097 | blk_start_plug(&plug); |
2ff7e61e | 3098 | ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark); |
79787eaa | 3099 | if (ret) { |
c6adc9cc | 3100 | blk_finish_plug(&plug); |
66642832 | 3101 | btrfs_abort_transaction(trans, ret); |
90787766 | 3102 | btrfs_set_log_full_commit(trans); |
79787eaa JM |
3103 | mutex_unlock(&root->log_mutex); |
3104 | goto out; | |
3105 | } | |
7237f183 | 3106 | |
5d4f98a2 | 3107 | btrfs_set_root_node(&log->root_item, log->node); |
7237f183 | 3108 | |
7237f183 YZ |
3109 | root->log_transid++; |
3110 | log->log_transid = root->log_transid; | |
ff782e0a | 3111 | root->log_start_pid = 0; |
06989c79 FM |
3112 | /* |
3113 | * Update or create log root item under the root's log_mutex to prevent | |
3114 | * races with concurrent log syncs that can lead to failure to update | |
3115 | * log root item because it was not created yet. | |
3116 | */ | |
3117 | ret = update_log_root(trans, log); | |
7237f183 | 3118 | /* |
8cef4e16 YZ |
3119 | * IO has been started, blocks of the log tree have WRITTEN flag set |
3120 | * in their headers. new modifications of the log will be written to | |
3121 | * new positions. so it's safe to allow log writers to go in. | |
7237f183 YZ |
3122 | */ |
3123 | mutex_unlock(&root->log_mutex); | |
3124 | ||
28a23593 | 3125 | btrfs_init_log_ctx(&root_log_ctx, NULL); |
d1433deb | 3126 | |
7237f183 | 3127 | mutex_lock(&log_root_tree->log_mutex); |
2ecb7923 | 3128 | atomic_inc(&log_root_tree->log_batch); |
7237f183 | 3129 | atomic_inc(&log_root_tree->log_writers); |
d1433deb MX |
3130 | |
3131 | index2 = log_root_tree->log_transid % 2; | |
3132 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
3133 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
3134 | ||
7237f183 YZ |
3135 | mutex_unlock(&log_root_tree->log_mutex); |
3136 | ||
7237f183 YZ |
3137 | mutex_lock(&log_root_tree->log_mutex); |
3138 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | |
093258e6 DS |
3139 | /* atomic_dec_and_test implies a barrier */ |
3140 | cond_wake_up_nomb(&log_root_tree->log_writer_wait); | |
7237f183 YZ |
3141 | } |
3142 | ||
4a500fd1 | 3143 | if (ret) { |
d1433deb MX |
3144 | if (!list_empty(&root_log_ctx.list)) |
3145 | list_del_init(&root_log_ctx.list); | |
3146 | ||
c6adc9cc | 3147 | blk_finish_plug(&plug); |
90787766 | 3148 | btrfs_set_log_full_commit(trans); |
995946dd | 3149 | |
79787eaa | 3150 | if (ret != -ENOSPC) { |
66642832 | 3151 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3152 | mutex_unlock(&log_root_tree->log_mutex); |
3153 | goto out; | |
3154 | } | |
bf89d38f | 3155 | btrfs_wait_tree_log_extents(log, mark); |
4a500fd1 YZ |
3156 | mutex_unlock(&log_root_tree->log_mutex); |
3157 | ret = -EAGAIN; | |
3158 | goto out; | |
3159 | } | |
3160 | ||
d1433deb | 3161 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { |
3da5ab56 | 3162 | blk_finish_plug(&plug); |
cbd60aa7 | 3163 | list_del_init(&root_log_ctx.list); |
d1433deb MX |
3164 | mutex_unlock(&log_root_tree->log_mutex); |
3165 | ret = root_log_ctx.log_ret; | |
3166 | goto out; | |
3167 | } | |
8b050d35 | 3168 | |
d1433deb | 3169 | index2 = root_log_ctx.log_transid % 2; |
7237f183 | 3170 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
c6adc9cc | 3171 | blk_finish_plug(&plug); |
bf89d38f | 3172 | ret = btrfs_wait_tree_log_extents(log, mark); |
60d53eb3 | 3173 | wait_log_commit(log_root_tree, |
d1433deb | 3174 | root_log_ctx.log_transid); |
7237f183 | 3175 | mutex_unlock(&log_root_tree->log_mutex); |
5ab5e44a FM |
3176 | if (!ret) |
3177 | ret = root_log_ctx.log_ret; | |
7237f183 YZ |
3178 | goto out; |
3179 | } | |
d1433deb | 3180 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); |
7237f183 YZ |
3181 | atomic_set(&log_root_tree->log_commit[index2], 1); |
3182 | ||
12fcfd22 | 3183 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
60d53eb3 | 3184 | wait_log_commit(log_root_tree, |
d1433deb | 3185 | root_log_ctx.log_transid - 1); |
12fcfd22 CM |
3186 | } |
3187 | ||
60d53eb3 | 3188 | wait_for_writer(log_root_tree); |
7237f183 | 3189 | |
12fcfd22 CM |
3190 | /* |
3191 | * now that we've moved on to the tree of log tree roots, | |
3192 | * check the full commit flag again | |
3193 | */ | |
4884b8e8 | 3194 | if (btrfs_need_log_full_commit(trans)) { |
c6adc9cc | 3195 | blk_finish_plug(&plug); |
bf89d38f | 3196 | btrfs_wait_tree_log_extents(log, mark); |
12fcfd22 CM |
3197 | mutex_unlock(&log_root_tree->log_mutex); |
3198 | ret = -EAGAIN; | |
3199 | goto out_wake_log_root; | |
3200 | } | |
7237f183 | 3201 | |
2ff7e61e | 3202 | ret = btrfs_write_marked_extents(fs_info, |
c6adc9cc MX |
3203 | &log_root_tree->dirty_log_pages, |
3204 | EXTENT_DIRTY | EXTENT_NEW); | |
3205 | blk_finish_plug(&plug); | |
79787eaa | 3206 | if (ret) { |
90787766 | 3207 | btrfs_set_log_full_commit(trans); |
66642832 | 3208 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3209 | mutex_unlock(&log_root_tree->log_mutex); |
3210 | goto out_wake_log_root; | |
3211 | } | |
bf89d38f | 3212 | ret = btrfs_wait_tree_log_extents(log, mark); |
5ab5e44a | 3213 | if (!ret) |
bf89d38f JM |
3214 | ret = btrfs_wait_tree_log_extents(log_root_tree, |
3215 | EXTENT_NEW | EXTENT_DIRTY); | |
5ab5e44a | 3216 | if (ret) { |
90787766 | 3217 | btrfs_set_log_full_commit(trans); |
5ab5e44a FM |
3218 | mutex_unlock(&log_root_tree->log_mutex); |
3219 | goto out_wake_log_root; | |
3220 | } | |
e02119d5 | 3221 | |
0b246afa JM |
3222 | btrfs_set_super_log_root(fs_info->super_for_commit, |
3223 | log_root_tree->node->start); | |
3224 | btrfs_set_super_log_root_level(fs_info->super_for_commit, | |
3225 | btrfs_header_level(log_root_tree->node)); | |
e02119d5 | 3226 | |
7237f183 | 3227 | log_root_tree->log_transid++; |
7237f183 YZ |
3228 | mutex_unlock(&log_root_tree->log_mutex); |
3229 | ||
3230 | /* | |
52042d8e | 3231 | * Nobody else is going to jump in and write the ctree |
7237f183 YZ |
3232 | * super here because the log_commit atomic below is protecting |
3233 | * us. We must be called with a transaction handle pinning | |
3234 | * the running transaction open, so a full commit can't hop | |
3235 | * in and cause problems either. | |
3236 | */ | |
eece6a9c | 3237 | ret = write_all_supers(fs_info, 1); |
5af3e8cc | 3238 | if (ret) { |
90787766 | 3239 | btrfs_set_log_full_commit(trans); |
66642832 | 3240 | btrfs_abort_transaction(trans, ret); |
5af3e8cc SB |
3241 | goto out_wake_log_root; |
3242 | } | |
7237f183 | 3243 | |
257c62e1 CM |
3244 | mutex_lock(&root->log_mutex); |
3245 | if (root->last_log_commit < log_transid) | |
3246 | root->last_log_commit = log_transid; | |
3247 | mutex_unlock(&root->log_mutex); | |
3248 | ||
12fcfd22 | 3249 | out_wake_log_root: |
570dd450 | 3250 | mutex_lock(&log_root_tree->log_mutex); |
8b050d35 MX |
3251 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); |
3252 | ||
d1433deb | 3253 | log_root_tree->log_transid_committed++; |
7237f183 | 3254 | atomic_set(&log_root_tree->log_commit[index2], 0); |
d1433deb MX |
3255 | mutex_unlock(&log_root_tree->log_mutex); |
3256 | ||
33a9eca7 | 3257 | /* |
093258e6 DS |
3258 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3259 | * all the updates above are seen by the woken threads. It might not be | |
3260 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3261 | */ |
093258e6 | 3262 | cond_wake_up(&log_root_tree->log_commit_wait[index2]); |
e02119d5 | 3263 | out: |
d1433deb | 3264 | mutex_lock(&root->log_mutex); |
570dd450 | 3265 | btrfs_remove_all_log_ctxs(root, index1, ret); |
d1433deb | 3266 | root->log_transid_committed++; |
7237f183 | 3267 | atomic_set(&root->log_commit[index1], 0); |
d1433deb | 3268 | mutex_unlock(&root->log_mutex); |
8b050d35 | 3269 | |
33a9eca7 | 3270 | /* |
093258e6 DS |
3271 | * The barrier before waitqueue_active (in cond_wake_up) is needed so |
3272 | * all the updates above are seen by the woken threads. It might not be | |
3273 | * necessary, but proving that seems to be hard. | |
33a9eca7 | 3274 | */ |
093258e6 | 3275 | cond_wake_up(&root->log_commit_wait[index1]); |
b31eabd8 | 3276 | return ret; |
e02119d5 CM |
3277 | } |
3278 | ||
4a500fd1 YZ |
3279 | static void free_log_tree(struct btrfs_trans_handle *trans, |
3280 | struct btrfs_root *log) | |
e02119d5 CM |
3281 | { |
3282 | int ret; | |
e02119d5 CM |
3283 | struct walk_control wc = { |
3284 | .free = 1, | |
3285 | .process_func = process_one_buffer | |
3286 | }; | |
3287 | ||
681ae509 | 3288 | ret = walk_log_tree(trans, log, &wc); |
374b0e2d JM |
3289 | if (ret) { |
3290 | if (trans) | |
3291 | btrfs_abort_transaction(trans, ret); | |
3292 | else | |
3293 | btrfs_handle_fs_error(log->fs_info, ret, NULL); | |
3294 | } | |
e02119d5 | 3295 | |
59b0713a FM |
3296 | clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1, |
3297 | EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT); | |
7237f183 YZ |
3298 | free_extent_buffer(log->node); |
3299 | kfree(log); | |
4a500fd1 YZ |
3300 | } |
3301 | ||
3302 | /* | |
3303 | * free all the extents used by the tree log. This should be called | |
3304 | * at commit time of the full transaction | |
3305 | */ | |
3306 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3307 | { | |
3308 | if (root->log_root) { | |
3309 | free_log_tree(trans, root->log_root); | |
3310 | root->log_root = NULL; | |
3311 | } | |
3312 | return 0; | |
3313 | } | |
3314 | ||
3315 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3316 | struct btrfs_fs_info *fs_info) | |
3317 | { | |
3318 | if (fs_info->log_root_tree) { | |
3319 | free_log_tree(trans, fs_info->log_root_tree); | |
3320 | fs_info->log_root_tree = NULL; | |
3321 | } | |
e02119d5 CM |
3322 | return 0; |
3323 | } | |
3324 | ||
e02119d5 CM |
3325 | /* |
3326 | * If both a file and directory are logged, and unlinks or renames are | |
3327 | * mixed in, we have a few interesting corners: | |
3328 | * | |
3329 | * create file X in dir Y | |
3330 | * link file X to X.link in dir Y | |
3331 | * fsync file X | |
3332 | * unlink file X but leave X.link | |
3333 | * fsync dir Y | |
3334 | * | |
3335 | * After a crash we would expect only X.link to exist. But file X | |
3336 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3337 | * | |
3338 | * We solve this by removing directory entries and inode backrefs from the | |
3339 | * log when a file that was logged in the current transaction is | |
3340 | * unlinked. Any later fsync will include the updated log entries, and | |
3341 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3342 | * | |
3343 | * This optimizations allows us to avoid relogging the entire inode | |
3344 | * or the entire directory. | |
3345 | */ | |
3346 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3347 | struct btrfs_root *root, | |
3348 | const char *name, int name_len, | |
49f34d1f | 3349 | struct btrfs_inode *dir, u64 index) |
e02119d5 CM |
3350 | { |
3351 | struct btrfs_root *log; | |
3352 | struct btrfs_dir_item *di; | |
3353 | struct btrfs_path *path; | |
3354 | int ret; | |
4a500fd1 | 3355 | int err = 0; |
e02119d5 | 3356 | int bytes_del = 0; |
49f34d1f | 3357 | u64 dir_ino = btrfs_ino(dir); |
e02119d5 | 3358 | |
49f34d1f | 3359 | if (dir->logged_trans < trans->transid) |
3a5f1d45 CM |
3360 | return 0; |
3361 | ||
e02119d5 CM |
3362 | ret = join_running_log_trans(root); |
3363 | if (ret) | |
3364 | return 0; | |
3365 | ||
49f34d1f | 3366 | mutex_lock(&dir->log_mutex); |
e02119d5 CM |
3367 | |
3368 | log = root->log_root; | |
3369 | path = btrfs_alloc_path(); | |
a62f44a5 TI |
3370 | if (!path) { |
3371 | err = -ENOMEM; | |
3372 | goto out_unlock; | |
3373 | } | |
2a29edc6 | 3374 | |
33345d01 | 3375 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
e02119d5 | 3376 | name, name_len, -1); |
4a500fd1 YZ |
3377 | if (IS_ERR(di)) { |
3378 | err = PTR_ERR(di); | |
3379 | goto fail; | |
3380 | } | |
3381 | if (di) { | |
e02119d5 CM |
3382 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3383 | bytes_del += name_len; | |
3650860b JB |
3384 | if (ret) { |
3385 | err = ret; | |
3386 | goto fail; | |
3387 | } | |
e02119d5 | 3388 | } |
b3b4aa74 | 3389 | btrfs_release_path(path); |
33345d01 | 3390 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
e02119d5 | 3391 | index, name, name_len, -1); |
4a500fd1 YZ |
3392 | if (IS_ERR(di)) { |
3393 | err = PTR_ERR(di); | |
3394 | goto fail; | |
3395 | } | |
3396 | if (di) { | |
e02119d5 CM |
3397 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
3398 | bytes_del += name_len; | |
3650860b JB |
3399 | if (ret) { |
3400 | err = ret; | |
3401 | goto fail; | |
3402 | } | |
e02119d5 CM |
3403 | } |
3404 | ||
3405 | /* update the directory size in the log to reflect the names | |
3406 | * we have removed | |
3407 | */ | |
3408 | if (bytes_del) { | |
3409 | struct btrfs_key key; | |
3410 | ||
33345d01 | 3411 | key.objectid = dir_ino; |
e02119d5 CM |
3412 | key.offset = 0; |
3413 | key.type = BTRFS_INODE_ITEM_KEY; | |
b3b4aa74 | 3414 | btrfs_release_path(path); |
e02119d5 CM |
3415 | |
3416 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
4a500fd1 YZ |
3417 | if (ret < 0) { |
3418 | err = ret; | |
3419 | goto fail; | |
3420 | } | |
e02119d5 CM |
3421 | if (ret == 0) { |
3422 | struct btrfs_inode_item *item; | |
3423 | u64 i_size; | |
3424 | ||
3425 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3426 | struct btrfs_inode_item); | |
3427 | i_size = btrfs_inode_size(path->nodes[0], item); | |
3428 | if (i_size > bytes_del) | |
3429 | i_size -= bytes_del; | |
3430 | else | |
3431 | i_size = 0; | |
3432 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
3433 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3434 | } else | |
3435 | ret = 0; | |
b3b4aa74 | 3436 | btrfs_release_path(path); |
e02119d5 | 3437 | } |
4a500fd1 | 3438 | fail: |
e02119d5 | 3439 | btrfs_free_path(path); |
a62f44a5 | 3440 | out_unlock: |
49f34d1f | 3441 | mutex_unlock(&dir->log_mutex); |
4a500fd1 | 3442 | if (ret == -ENOSPC) { |
90787766 | 3443 | btrfs_set_log_full_commit(trans); |
4a500fd1 | 3444 | ret = 0; |
79787eaa | 3445 | } else if (ret < 0) |
66642832 | 3446 | btrfs_abort_transaction(trans, ret); |
79787eaa | 3447 | |
12fcfd22 | 3448 | btrfs_end_log_trans(root); |
e02119d5 | 3449 | |
411fc6bc | 3450 | return err; |
e02119d5 CM |
3451 | } |
3452 | ||
3453 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3454 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3455 | struct btrfs_root *root, | |
3456 | const char *name, int name_len, | |
a491abb2 | 3457 | struct btrfs_inode *inode, u64 dirid) |
e02119d5 CM |
3458 | { |
3459 | struct btrfs_root *log; | |
3460 | u64 index; | |
3461 | int ret; | |
3462 | ||
a491abb2 | 3463 | if (inode->logged_trans < trans->transid) |
3a5f1d45 CM |
3464 | return 0; |
3465 | ||
e02119d5 CM |
3466 | ret = join_running_log_trans(root); |
3467 | if (ret) | |
3468 | return 0; | |
3469 | log = root->log_root; | |
a491abb2 | 3470 | mutex_lock(&inode->log_mutex); |
e02119d5 | 3471 | |
a491abb2 | 3472 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
e02119d5 | 3473 | dirid, &index); |
a491abb2 | 3474 | mutex_unlock(&inode->log_mutex); |
4a500fd1 | 3475 | if (ret == -ENOSPC) { |
90787766 | 3476 | btrfs_set_log_full_commit(trans); |
4a500fd1 | 3477 | ret = 0; |
79787eaa | 3478 | } else if (ret < 0 && ret != -ENOENT) |
66642832 | 3479 | btrfs_abort_transaction(trans, ret); |
12fcfd22 | 3480 | btrfs_end_log_trans(root); |
e02119d5 | 3481 | |
e02119d5 CM |
3482 | return ret; |
3483 | } | |
3484 | ||
3485 | /* | |
3486 | * creates a range item in the log for 'dirid'. first_offset and | |
3487 | * last_offset tell us which parts of the key space the log should | |
3488 | * be considered authoritative for. | |
3489 | */ | |
3490 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3491 | struct btrfs_root *log, | |
3492 | struct btrfs_path *path, | |
3493 | int key_type, u64 dirid, | |
3494 | u64 first_offset, u64 last_offset) | |
3495 | { | |
3496 | int ret; | |
3497 | struct btrfs_key key; | |
3498 | struct btrfs_dir_log_item *item; | |
3499 | ||
3500 | key.objectid = dirid; | |
3501 | key.offset = first_offset; | |
3502 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3503 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3504 | else | |
3505 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3506 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
4a500fd1 YZ |
3507 | if (ret) |
3508 | return ret; | |
e02119d5 CM |
3509 | |
3510 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3511 | struct btrfs_dir_log_item); | |
3512 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3513 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
b3b4aa74 | 3514 | btrfs_release_path(path); |
e02119d5 CM |
3515 | return 0; |
3516 | } | |
3517 | ||
3518 | /* | |
3519 | * log all the items included in the current transaction for a given | |
3520 | * directory. This also creates the range items in the log tree required | |
3521 | * to replay anything deleted before the fsync | |
3522 | */ | |
3523 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
684a5773 | 3524 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 CM |
3525 | struct btrfs_path *path, |
3526 | struct btrfs_path *dst_path, int key_type, | |
2f2ff0ee | 3527 | struct btrfs_log_ctx *ctx, |
e02119d5 CM |
3528 | u64 min_offset, u64 *last_offset_ret) |
3529 | { | |
3530 | struct btrfs_key min_key; | |
e02119d5 CM |
3531 | struct btrfs_root *log = root->log_root; |
3532 | struct extent_buffer *src; | |
4a500fd1 | 3533 | int err = 0; |
e02119d5 CM |
3534 | int ret; |
3535 | int i; | |
3536 | int nritems; | |
3537 | u64 first_offset = min_offset; | |
3538 | u64 last_offset = (u64)-1; | |
684a5773 | 3539 | u64 ino = btrfs_ino(inode); |
e02119d5 CM |
3540 | |
3541 | log = root->log_root; | |
e02119d5 | 3542 | |
33345d01 | 3543 | min_key.objectid = ino; |
e02119d5 CM |
3544 | min_key.type = key_type; |
3545 | min_key.offset = min_offset; | |
3546 | ||
6174d3cb | 3547 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); |
e02119d5 CM |
3548 | |
3549 | /* | |
3550 | * we didn't find anything from this transaction, see if there | |
3551 | * is anything at all | |
3552 | */ | |
33345d01 LZ |
3553 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
3554 | min_key.objectid = ino; | |
e02119d5 CM |
3555 | min_key.type = key_type; |
3556 | min_key.offset = (u64)-1; | |
b3b4aa74 | 3557 | btrfs_release_path(path); |
e02119d5 CM |
3558 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
3559 | if (ret < 0) { | |
b3b4aa74 | 3560 | btrfs_release_path(path); |
e02119d5 CM |
3561 | return ret; |
3562 | } | |
33345d01 | 3563 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3564 | |
3565 | /* if ret == 0 there are items for this type, | |
3566 | * create a range to tell us the last key of this type. | |
3567 | * otherwise, there are no items in this directory after | |
3568 | * *min_offset, and we create a range to indicate that. | |
3569 | */ | |
3570 | if (ret == 0) { | |
3571 | struct btrfs_key tmp; | |
3572 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3573 | path->slots[0]); | |
d397712b | 3574 | if (key_type == tmp.type) |
e02119d5 | 3575 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
3576 | } |
3577 | goto done; | |
3578 | } | |
3579 | ||
3580 | /* go backward to find any previous key */ | |
33345d01 | 3581 | ret = btrfs_previous_item(root, path, ino, key_type); |
e02119d5 CM |
3582 | if (ret == 0) { |
3583 | struct btrfs_key tmp; | |
3584 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3585 | if (key_type == tmp.type) { | |
3586 | first_offset = tmp.offset; | |
3587 | ret = overwrite_item(trans, log, dst_path, | |
3588 | path->nodes[0], path->slots[0], | |
3589 | &tmp); | |
4a500fd1 YZ |
3590 | if (ret) { |
3591 | err = ret; | |
3592 | goto done; | |
3593 | } | |
e02119d5 CM |
3594 | } |
3595 | } | |
b3b4aa74 | 3596 | btrfs_release_path(path); |
e02119d5 | 3597 | |
2cc83342 JB |
3598 | /* |
3599 | * Find the first key from this transaction again. See the note for | |
3600 | * log_new_dir_dentries, if we're logging a directory recursively we | |
3601 | * won't be holding its i_mutex, which means we can modify the directory | |
3602 | * while we're logging it. If we remove an entry between our first | |
3603 | * search and this search we'll not find the key again and can just | |
3604 | * bail. | |
3605 | */ | |
e02119d5 | 3606 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
2cc83342 | 3607 | if (ret != 0) |
e02119d5 | 3608 | goto done; |
e02119d5 CM |
3609 | |
3610 | /* | |
3611 | * we have a block from this transaction, log every item in it | |
3612 | * from our directory | |
3613 | */ | |
d397712b | 3614 | while (1) { |
e02119d5 CM |
3615 | struct btrfs_key tmp; |
3616 | src = path->nodes[0]; | |
3617 | nritems = btrfs_header_nritems(src); | |
3618 | for (i = path->slots[0]; i < nritems; i++) { | |
2f2ff0ee FM |
3619 | struct btrfs_dir_item *di; |
3620 | ||
e02119d5 CM |
3621 | btrfs_item_key_to_cpu(src, &min_key, i); |
3622 | ||
33345d01 | 3623 | if (min_key.objectid != ino || min_key.type != key_type) |
e02119d5 CM |
3624 | goto done; |
3625 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3626 | &min_key); | |
4a500fd1 YZ |
3627 | if (ret) { |
3628 | err = ret; | |
3629 | goto done; | |
3630 | } | |
2f2ff0ee FM |
3631 | |
3632 | /* | |
3633 | * We must make sure that when we log a directory entry, | |
3634 | * the corresponding inode, after log replay, has a | |
3635 | * matching link count. For example: | |
3636 | * | |
3637 | * touch foo | |
3638 | * mkdir mydir | |
3639 | * sync | |
3640 | * ln foo mydir/bar | |
3641 | * xfs_io -c "fsync" mydir | |
3642 | * <crash> | |
3643 | * <mount fs and log replay> | |
3644 | * | |
3645 | * Would result in a fsync log that when replayed, our | |
3646 | * file inode would have a link count of 1, but we get | |
3647 | * two directory entries pointing to the same inode. | |
3648 | * After removing one of the names, it would not be | |
3649 | * possible to remove the other name, which resulted | |
3650 | * always in stale file handle errors, and would not | |
3651 | * be possible to rmdir the parent directory, since | |
3652 | * its i_size could never decrement to the value | |
3653 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3654 | */ | |
3655 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3656 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3657 | if (ctx && | |
3658 | (btrfs_dir_transid(src, di) == trans->transid || | |
3659 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3660 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3661 | ctx->log_new_dentries = true; | |
e02119d5 CM |
3662 | } |
3663 | path->slots[0] = nritems; | |
3664 | ||
3665 | /* | |
3666 | * look ahead to the next item and see if it is also | |
3667 | * from this directory and from this transaction | |
3668 | */ | |
3669 | ret = btrfs_next_leaf(root, path); | |
80c0b421 LB |
3670 | if (ret) { |
3671 | if (ret == 1) | |
3672 | last_offset = (u64)-1; | |
3673 | else | |
3674 | err = ret; | |
e02119d5 CM |
3675 | goto done; |
3676 | } | |
3677 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
33345d01 | 3678 | if (tmp.objectid != ino || tmp.type != key_type) { |
e02119d5 CM |
3679 | last_offset = (u64)-1; |
3680 | goto done; | |
3681 | } | |
3682 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3683 | ret = overwrite_item(trans, log, dst_path, | |
3684 | path->nodes[0], path->slots[0], | |
3685 | &tmp); | |
4a500fd1 YZ |
3686 | if (ret) |
3687 | err = ret; | |
3688 | else | |
3689 | last_offset = tmp.offset; | |
e02119d5 CM |
3690 | goto done; |
3691 | } | |
3692 | } | |
3693 | done: | |
b3b4aa74 DS |
3694 | btrfs_release_path(path); |
3695 | btrfs_release_path(dst_path); | |
e02119d5 | 3696 | |
4a500fd1 YZ |
3697 | if (err == 0) { |
3698 | *last_offset_ret = last_offset; | |
3699 | /* | |
3700 | * insert the log range keys to indicate where the log | |
3701 | * is valid | |
3702 | */ | |
3703 | ret = insert_dir_log_key(trans, log, path, key_type, | |
33345d01 | 3704 | ino, first_offset, last_offset); |
4a500fd1 YZ |
3705 | if (ret) |
3706 | err = ret; | |
3707 | } | |
3708 | return err; | |
e02119d5 CM |
3709 | } |
3710 | ||
3711 | /* | |
3712 | * logging directories is very similar to logging inodes, We find all the items | |
3713 | * from the current transaction and write them to the log. | |
3714 | * | |
3715 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3716 | * key in the range logged that is not present in the log tree, then it means | |
3717 | * that dir entry was unlinked during the transaction. | |
3718 | * | |
3719 | * In order for that scan to work, we must include one key smaller than | |
3720 | * the smallest logged by this transaction and one key larger than the largest | |
3721 | * key logged by this transaction. | |
3722 | */ | |
3723 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
dbf39ea4 | 3724 | struct btrfs_root *root, struct btrfs_inode *inode, |
e02119d5 | 3725 | struct btrfs_path *path, |
2f2ff0ee FM |
3726 | struct btrfs_path *dst_path, |
3727 | struct btrfs_log_ctx *ctx) | |
e02119d5 CM |
3728 | { |
3729 | u64 min_key; | |
3730 | u64 max_key; | |
3731 | int ret; | |
3732 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3733 | ||
3734 | again: | |
3735 | min_key = 0; | |
3736 | max_key = 0; | |
d397712b | 3737 | while (1) { |
dbf39ea4 NB |
3738 | ret = log_dir_items(trans, root, inode, path, dst_path, key_type, |
3739 | ctx, min_key, &max_key); | |
4a500fd1 YZ |
3740 | if (ret) |
3741 | return ret; | |
e02119d5 CM |
3742 | if (max_key == (u64)-1) |
3743 | break; | |
3744 | min_key = max_key + 1; | |
3745 | } | |
3746 | ||
3747 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3748 | key_type = BTRFS_DIR_INDEX_KEY; | |
3749 | goto again; | |
3750 | } | |
3751 | return 0; | |
3752 | } | |
3753 | ||
3754 | /* | |
3755 | * a helper function to drop items from the log before we relog an | |
3756 | * inode. max_key_type indicates the highest item type to remove. | |
3757 | * This cannot be run for file data extents because it does not | |
3758 | * free the extents they point to. | |
3759 | */ | |
3760 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3761 | struct btrfs_root *log, | |
3762 | struct btrfs_path *path, | |
3763 | u64 objectid, int max_key_type) | |
3764 | { | |
3765 | int ret; | |
3766 | struct btrfs_key key; | |
3767 | struct btrfs_key found_key; | |
18ec90d6 | 3768 | int start_slot; |
e02119d5 CM |
3769 | |
3770 | key.objectid = objectid; | |
3771 | key.type = max_key_type; | |
3772 | key.offset = (u64)-1; | |
3773 | ||
d397712b | 3774 | while (1) { |
e02119d5 | 3775 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
3650860b | 3776 | BUG_ON(ret == 0); /* Logic error */ |
4a500fd1 | 3777 | if (ret < 0) |
e02119d5 CM |
3778 | break; |
3779 | ||
3780 | if (path->slots[0] == 0) | |
3781 | break; | |
3782 | ||
3783 | path->slots[0]--; | |
3784 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3785 | path->slots[0]); | |
3786 | ||
3787 | if (found_key.objectid != objectid) | |
3788 | break; | |
3789 | ||
18ec90d6 JB |
3790 | found_key.offset = 0; |
3791 | found_key.type = 0; | |
3792 | ret = btrfs_bin_search(path->nodes[0], &found_key, 0, | |
3793 | &start_slot); | |
cbca7d59 FM |
3794 | if (ret < 0) |
3795 | break; | |
18ec90d6 JB |
3796 | |
3797 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3798 | path->slots[0] - start_slot + 1); | |
3799 | /* | |
3800 | * If start slot isn't 0 then we don't need to re-search, we've | |
3801 | * found the last guy with the objectid in this tree. | |
3802 | */ | |
3803 | if (ret || start_slot != 0) | |
65a246c5 | 3804 | break; |
b3b4aa74 | 3805 | btrfs_release_path(path); |
e02119d5 | 3806 | } |
b3b4aa74 | 3807 | btrfs_release_path(path); |
5bdbeb21 JB |
3808 | if (ret > 0) |
3809 | ret = 0; | |
4a500fd1 | 3810 | return ret; |
e02119d5 CM |
3811 | } |
3812 | ||
94edf4ae JB |
3813 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3814 | struct extent_buffer *leaf, | |
3815 | struct btrfs_inode_item *item, | |
1a4bcf47 FM |
3816 | struct inode *inode, int log_inode_only, |
3817 | u64 logged_isize) | |
94edf4ae | 3818 | { |
0b1c6cca JB |
3819 | struct btrfs_map_token token; |
3820 | ||
3821 | btrfs_init_map_token(&token); | |
94edf4ae JB |
3822 | |
3823 | if (log_inode_only) { | |
3824 | /* set the generation to zero so the recover code | |
3825 | * can tell the difference between an logging | |
3826 | * just to say 'this inode exists' and a logging | |
3827 | * to say 'update this inode with these values' | |
3828 | */ | |
0b1c6cca | 3829 | btrfs_set_token_inode_generation(leaf, item, 0, &token); |
1a4bcf47 | 3830 | btrfs_set_token_inode_size(leaf, item, logged_isize, &token); |
94edf4ae | 3831 | } else { |
0b1c6cca JB |
3832 | btrfs_set_token_inode_generation(leaf, item, |
3833 | BTRFS_I(inode)->generation, | |
3834 | &token); | |
3835 | btrfs_set_token_inode_size(leaf, item, inode->i_size, &token); | |
3836 | } | |
3837 | ||
3838 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
3839 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3840 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3841 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
3842 | ||
a937b979 | 3843 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
0b1c6cca | 3844 | inode->i_atime.tv_sec, &token); |
a937b979 | 3845 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
0b1c6cca JB |
3846 | inode->i_atime.tv_nsec, &token); |
3847 | ||
a937b979 | 3848 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
0b1c6cca | 3849 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3850 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
0b1c6cca JB |
3851 | inode->i_mtime.tv_nsec, &token); |
3852 | ||
a937b979 | 3853 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
0b1c6cca | 3854 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3855 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
0b1c6cca JB |
3856 | inode->i_ctime.tv_nsec, &token); |
3857 | ||
3858 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
3859 | &token); | |
3860 | ||
c7f88c4e JL |
3861 | btrfs_set_token_inode_sequence(leaf, item, |
3862 | inode_peek_iversion(inode), &token); | |
0b1c6cca JB |
3863 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); |
3864 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3865 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3866 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
94edf4ae JB |
3867 | } |
3868 | ||
a95249b3 JB |
3869 | static int log_inode_item(struct btrfs_trans_handle *trans, |
3870 | struct btrfs_root *log, struct btrfs_path *path, | |
6d889a3b | 3871 | struct btrfs_inode *inode) |
a95249b3 JB |
3872 | { |
3873 | struct btrfs_inode_item *inode_item; | |
a95249b3 JB |
3874 | int ret; |
3875 | ||
efd0c405 | 3876 | ret = btrfs_insert_empty_item(trans, log, path, |
6d889a3b | 3877 | &inode->location, sizeof(*inode_item)); |
a95249b3 JB |
3878 | if (ret && ret != -EEXIST) |
3879 | return ret; | |
3880 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3881 | struct btrfs_inode_item); | |
6d889a3b NB |
3882 | fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode, |
3883 | 0, 0); | |
a95249b3 JB |
3884 | btrfs_release_path(path); |
3885 | return 0; | |
3886 | } | |
3887 | ||
31ff1cd2 | 3888 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
44d70e19 | 3889 | struct btrfs_inode *inode, |
31ff1cd2 | 3890 | struct btrfs_path *dst_path, |
16e7549f | 3891 | struct btrfs_path *src_path, u64 *last_extent, |
1a4bcf47 FM |
3892 | int start_slot, int nr, int inode_only, |
3893 | u64 logged_isize) | |
31ff1cd2 | 3894 | { |
3ffbd68c | 3895 | struct btrfs_fs_info *fs_info = trans->fs_info; |
31ff1cd2 CM |
3896 | unsigned long src_offset; |
3897 | unsigned long dst_offset; | |
44d70e19 | 3898 | struct btrfs_root *log = inode->root->log_root; |
31ff1cd2 CM |
3899 | struct btrfs_file_extent_item *extent; |
3900 | struct btrfs_inode_item *inode_item; | |
16e7549f JB |
3901 | struct extent_buffer *src = src_path->nodes[0]; |
3902 | struct btrfs_key first_key, last_key, key; | |
31ff1cd2 CM |
3903 | int ret; |
3904 | struct btrfs_key *ins_keys; | |
3905 | u32 *ins_sizes; | |
3906 | char *ins_data; | |
3907 | int i; | |
d20f7043 | 3908 | struct list_head ordered_sums; |
44d70e19 | 3909 | int skip_csum = inode->flags & BTRFS_INODE_NODATASUM; |
16e7549f | 3910 | bool has_extents = false; |
74121f7c | 3911 | bool need_find_last_extent = true; |
16e7549f | 3912 | bool done = false; |
d20f7043 CM |
3913 | |
3914 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
3915 | |
3916 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3917 | nr * sizeof(u32), GFP_NOFS); | |
2a29edc6 | 3918 | if (!ins_data) |
3919 | return -ENOMEM; | |
3920 | ||
16e7549f JB |
3921 | first_key.objectid = (u64)-1; |
3922 | ||
31ff1cd2 CM |
3923 | ins_sizes = (u32 *)ins_data; |
3924 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3925 | ||
3926 | for (i = 0; i < nr; i++) { | |
3927 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3928 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3929 | } | |
3930 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3931 | ins_keys, ins_sizes, nr); | |
4a500fd1 YZ |
3932 | if (ret) { |
3933 | kfree(ins_data); | |
3934 | return ret; | |
3935 | } | |
31ff1cd2 | 3936 | |
5d4f98a2 | 3937 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
3938 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
3939 | dst_path->slots[0]); | |
3940 | ||
3941 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3942 | ||
0dde10be | 3943 | if (i == nr - 1) |
16e7549f JB |
3944 | last_key = ins_keys[i]; |
3945 | ||
94edf4ae | 3946 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
31ff1cd2 CM |
3947 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
3948 | dst_path->slots[0], | |
3949 | struct btrfs_inode_item); | |
94edf4ae | 3950 | fill_inode_item(trans, dst_path->nodes[0], inode_item, |
f85b7379 DS |
3951 | &inode->vfs_inode, |
3952 | inode_only == LOG_INODE_EXISTS, | |
1a4bcf47 | 3953 | logged_isize); |
94edf4ae JB |
3954 | } else { |
3955 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3956 | src_offset, ins_sizes[i]); | |
31ff1cd2 | 3957 | } |
94edf4ae | 3958 | |
16e7549f JB |
3959 | /* |
3960 | * We set need_find_last_extent here in case we know we were | |
3961 | * processing other items and then walk into the first extent in | |
3962 | * the inode. If we don't hit an extent then nothing changes, | |
3963 | * we'll do the last search the next time around. | |
3964 | */ | |
3965 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) { | |
3966 | has_extents = true; | |
74121f7c | 3967 | if (first_key.objectid == (u64)-1) |
16e7549f JB |
3968 | first_key = ins_keys[i]; |
3969 | } else { | |
3970 | need_find_last_extent = false; | |
3971 | } | |
3972 | ||
31ff1cd2 CM |
3973 | /* take a reference on file data extents so that truncates |
3974 | * or deletes of this inode don't have to relog the inode | |
3975 | * again | |
3976 | */ | |
962a298f | 3977 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && |
d2794405 | 3978 | !skip_csum) { |
31ff1cd2 CM |
3979 | int found_type; |
3980 | extent = btrfs_item_ptr(src, start_slot + i, | |
3981 | struct btrfs_file_extent_item); | |
3982 | ||
8e531cdf | 3983 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
3984 | continue; | |
3985 | ||
31ff1cd2 | 3986 | found_type = btrfs_file_extent_type(src, extent); |
6f1fed77 | 3987 | if (found_type == BTRFS_FILE_EXTENT_REG) { |
5d4f98a2 YZ |
3988 | u64 ds, dl, cs, cl; |
3989 | ds = btrfs_file_extent_disk_bytenr(src, | |
3990 | extent); | |
3991 | /* ds == 0 is a hole */ | |
3992 | if (ds == 0) | |
3993 | continue; | |
3994 | ||
3995 | dl = btrfs_file_extent_disk_num_bytes(src, | |
3996 | extent); | |
3997 | cs = btrfs_file_extent_offset(src, extent); | |
3998 | cl = btrfs_file_extent_num_bytes(src, | |
a419aef8 | 3999 | extent); |
580afd76 CM |
4000 | if (btrfs_file_extent_compression(src, |
4001 | extent)) { | |
4002 | cs = 0; | |
4003 | cl = dl; | |
4004 | } | |
5d4f98a2 YZ |
4005 | |
4006 | ret = btrfs_lookup_csums_range( | |
0b246afa | 4007 | fs_info->csum_root, |
5d4f98a2 | 4008 | ds + cs, ds + cs + cl - 1, |
a2de733c | 4009 | &ordered_sums, 0); |
3650860b JB |
4010 | if (ret) { |
4011 | btrfs_release_path(dst_path); | |
4012 | kfree(ins_data); | |
4013 | return ret; | |
4014 | } | |
31ff1cd2 CM |
4015 | } |
4016 | } | |
31ff1cd2 CM |
4017 | } |
4018 | ||
4019 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
b3b4aa74 | 4020 | btrfs_release_path(dst_path); |
31ff1cd2 | 4021 | kfree(ins_data); |
d20f7043 CM |
4022 | |
4023 | /* | |
4024 | * we have to do this after the loop above to avoid changing the | |
4025 | * log tree while trying to change the log tree. | |
4026 | */ | |
4a500fd1 | 4027 | ret = 0; |
d397712b | 4028 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
4029 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
4030 | struct btrfs_ordered_sum, | |
4031 | list); | |
4a500fd1 YZ |
4032 | if (!ret) |
4033 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
d20f7043 CM |
4034 | list_del(&sums->list); |
4035 | kfree(sums); | |
4036 | } | |
16e7549f JB |
4037 | |
4038 | if (!has_extents) | |
4039 | return ret; | |
4040 | ||
74121f7c FM |
4041 | if (need_find_last_extent && *last_extent == first_key.offset) { |
4042 | /* | |
4043 | * We don't have any leafs between our current one and the one | |
4044 | * we processed before that can have file extent items for our | |
4045 | * inode (and have a generation number smaller than our current | |
4046 | * transaction id). | |
4047 | */ | |
4048 | need_find_last_extent = false; | |
4049 | } | |
4050 | ||
16e7549f JB |
4051 | /* |
4052 | * Because we use btrfs_search_forward we could skip leaves that were | |
4053 | * not modified and then assume *last_extent is valid when it really | |
4054 | * isn't. So back up to the previous leaf and read the end of the last | |
4055 | * extent before we go and fill in holes. | |
4056 | */ | |
4057 | if (need_find_last_extent) { | |
4058 | u64 len; | |
4059 | ||
44d70e19 | 4060 | ret = btrfs_prev_leaf(inode->root, src_path); |
16e7549f JB |
4061 | if (ret < 0) |
4062 | return ret; | |
4063 | if (ret) | |
4064 | goto fill_holes; | |
4065 | if (src_path->slots[0]) | |
4066 | src_path->slots[0]--; | |
4067 | src = src_path->nodes[0]; | |
4068 | btrfs_item_key_to_cpu(src, &key, src_path->slots[0]); | |
44d70e19 | 4069 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
4070 | key.type != BTRFS_EXTENT_DATA_KEY) |
4071 | goto fill_holes; | |
4072 | extent = btrfs_item_ptr(src, src_path->slots[0], | |
4073 | struct btrfs_file_extent_item); | |
4074 | if (btrfs_file_extent_type(src, extent) == | |
4075 | BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 4076 | len = btrfs_file_extent_ram_bytes(src, extent); |
16e7549f | 4077 | *last_extent = ALIGN(key.offset + len, |
0b246afa | 4078 | fs_info->sectorsize); |
16e7549f JB |
4079 | } else { |
4080 | len = btrfs_file_extent_num_bytes(src, extent); | |
4081 | *last_extent = key.offset + len; | |
4082 | } | |
4083 | } | |
4084 | fill_holes: | |
4085 | /* So we did prev_leaf, now we need to move to the next leaf, but a few | |
4086 | * things could have happened | |
4087 | * | |
4088 | * 1) A merge could have happened, so we could currently be on a leaf | |
4089 | * that holds what we were copying in the first place. | |
4090 | * 2) A split could have happened, and now not all of the items we want | |
4091 | * are on the same leaf. | |
4092 | * | |
4093 | * So we need to adjust how we search for holes, we need to drop the | |
4094 | * path and re-search for the first extent key we found, and then walk | |
4095 | * forward until we hit the last one we copied. | |
4096 | */ | |
4097 | if (need_find_last_extent) { | |
4098 | /* btrfs_prev_leaf could return 1 without releasing the path */ | |
4099 | btrfs_release_path(src_path); | |
f85b7379 DS |
4100 | ret = btrfs_search_slot(NULL, inode->root, &first_key, |
4101 | src_path, 0, 0); | |
16e7549f JB |
4102 | if (ret < 0) |
4103 | return ret; | |
4104 | ASSERT(ret == 0); | |
4105 | src = src_path->nodes[0]; | |
4106 | i = src_path->slots[0]; | |
4107 | } else { | |
4108 | i = start_slot; | |
4109 | } | |
4110 | ||
4111 | /* | |
4112 | * Ok so here we need to go through and fill in any holes we may have | |
4113 | * to make sure that holes are punched for those areas in case they had | |
4114 | * extents previously. | |
4115 | */ | |
4116 | while (!done) { | |
4117 | u64 offset, len; | |
4118 | u64 extent_end; | |
4119 | ||
4120 | if (i >= btrfs_header_nritems(src_path->nodes[0])) { | |
44d70e19 | 4121 | ret = btrfs_next_leaf(inode->root, src_path); |
16e7549f JB |
4122 | if (ret < 0) |
4123 | return ret; | |
4124 | ASSERT(ret == 0); | |
4125 | src = src_path->nodes[0]; | |
4126 | i = 0; | |
8434ec46 | 4127 | need_find_last_extent = true; |
16e7549f JB |
4128 | } |
4129 | ||
4130 | btrfs_item_key_to_cpu(src, &key, i); | |
4131 | if (!btrfs_comp_cpu_keys(&key, &last_key)) | |
4132 | done = true; | |
44d70e19 | 4133 | if (key.objectid != btrfs_ino(inode) || |
16e7549f JB |
4134 | key.type != BTRFS_EXTENT_DATA_KEY) { |
4135 | i++; | |
4136 | continue; | |
4137 | } | |
4138 | extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item); | |
4139 | if (btrfs_file_extent_type(src, extent) == | |
4140 | BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 4141 | len = btrfs_file_extent_ram_bytes(src, extent); |
da17066c | 4142 | extent_end = ALIGN(key.offset + len, |
0b246afa | 4143 | fs_info->sectorsize); |
16e7549f JB |
4144 | } else { |
4145 | len = btrfs_file_extent_num_bytes(src, extent); | |
4146 | extent_end = key.offset + len; | |
4147 | } | |
4148 | i++; | |
4149 | ||
4150 | if (*last_extent == key.offset) { | |
4151 | *last_extent = extent_end; | |
4152 | continue; | |
4153 | } | |
4154 | offset = *last_extent; | |
4155 | len = key.offset - *last_extent; | |
44d70e19 | 4156 | ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode), |
f85b7379 | 4157 | offset, 0, 0, len, 0, len, 0, 0, 0); |
16e7549f JB |
4158 | if (ret) |
4159 | break; | |
74121f7c | 4160 | *last_extent = extent_end; |
16e7549f | 4161 | } |
4ee3fad3 FM |
4162 | |
4163 | /* | |
4164 | * Check if there is a hole between the last extent found in our leaf | |
4165 | * and the first extent in the next leaf. If there is one, we need to | |
4166 | * log an explicit hole so that at replay time we can punch the hole. | |
4167 | */ | |
4168 | if (ret == 0 && | |
4169 | key.objectid == btrfs_ino(inode) && | |
4170 | key.type == BTRFS_EXTENT_DATA_KEY && | |
4171 | i == btrfs_header_nritems(src_path->nodes[0])) { | |
4172 | ret = btrfs_next_leaf(inode->root, src_path); | |
4173 | need_find_last_extent = true; | |
4174 | if (ret > 0) { | |
4175 | ret = 0; | |
4176 | } else if (ret == 0) { | |
4177 | btrfs_item_key_to_cpu(src_path->nodes[0], &key, | |
4178 | src_path->slots[0]); | |
4179 | if (key.objectid == btrfs_ino(inode) && | |
4180 | key.type == BTRFS_EXTENT_DATA_KEY && | |
4181 | *last_extent < key.offset) { | |
4182 | const u64 len = key.offset - *last_extent; | |
4183 | ||
4184 | ret = btrfs_insert_file_extent(trans, log, | |
4185 | btrfs_ino(inode), | |
4186 | *last_extent, 0, | |
4187 | 0, len, 0, len, | |
4188 | 0, 0, 0); | |
ebb92906 | 4189 | *last_extent += len; |
4ee3fad3 FM |
4190 | } |
4191 | } | |
4192 | } | |
16e7549f JB |
4193 | /* |
4194 | * Need to let the callers know we dropped the path so they should | |
4195 | * re-search. | |
4196 | */ | |
4197 | if (!ret && need_find_last_extent) | |
4198 | ret = 1; | |
4a500fd1 | 4199 | return ret; |
31ff1cd2 CM |
4200 | } |
4201 | ||
5dc562c5 JB |
4202 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) |
4203 | { | |
4204 | struct extent_map *em1, *em2; | |
4205 | ||
4206 | em1 = list_entry(a, struct extent_map, list); | |
4207 | em2 = list_entry(b, struct extent_map, list); | |
4208 | ||
4209 | if (em1->start < em2->start) | |
4210 | return -1; | |
4211 | else if (em1->start > em2->start) | |
4212 | return 1; | |
4213 | return 0; | |
4214 | } | |
4215 | ||
e7175a69 JB |
4216 | static int log_extent_csums(struct btrfs_trans_handle *trans, |
4217 | struct btrfs_inode *inode, | |
a9ecb653 | 4218 | struct btrfs_root *log_root, |
e7175a69 | 4219 | const struct extent_map *em) |
5dc562c5 | 4220 | { |
2ab28f32 JB |
4221 | u64 csum_offset; |
4222 | u64 csum_len; | |
8407f553 FM |
4223 | LIST_HEAD(ordered_sums); |
4224 | int ret = 0; | |
0aa4a17d | 4225 | |
e7175a69 JB |
4226 | if (inode->flags & BTRFS_INODE_NODATASUM || |
4227 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
8407f553 | 4228 | em->block_start == EXTENT_MAP_HOLE) |
70c8a91c | 4229 | return 0; |
5dc562c5 | 4230 | |
e7175a69 | 4231 | /* If we're compressed we have to save the entire range of csums. */ |
488111aa FDBM |
4232 | if (em->compress_type) { |
4233 | csum_offset = 0; | |
8407f553 | 4234 | csum_len = max(em->block_len, em->orig_block_len); |
488111aa | 4235 | } else { |
e7175a69 JB |
4236 | csum_offset = em->mod_start - em->start; |
4237 | csum_len = em->mod_len; | |
488111aa | 4238 | } |
2ab28f32 | 4239 | |
70c8a91c | 4240 | /* block start is already adjusted for the file extent offset. */ |
a9ecb653 | 4241 | ret = btrfs_lookup_csums_range(trans->fs_info->csum_root, |
70c8a91c JB |
4242 | em->block_start + csum_offset, |
4243 | em->block_start + csum_offset + | |
4244 | csum_len - 1, &ordered_sums, 0); | |
4245 | if (ret) | |
4246 | return ret; | |
5dc562c5 | 4247 | |
70c8a91c JB |
4248 | while (!list_empty(&ordered_sums)) { |
4249 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4250 | struct btrfs_ordered_sum, | |
4251 | list); | |
4252 | if (!ret) | |
a9ecb653 | 4253 | ret = btrfs_csum_file_blocks(trans, log_root, sums); |
70c8a91c JB |
4254 | list_del(&sums->list); |
4255 | kfree(sums); | |
5dc562c5 JB |
4256 | } |
4257 | ||
70c8a91c | 4258 | return ret; |
5dc562c5 JB |
4259 | } |
4260 | ||
8407f553 | 4261 | static int log_one_extent(struct btrfs_trans_handle *trans, |
9d122629 | 4262 | struct btrfs_inode *inode, struct btrfs_root *root, |
8407f553 FM |
4263 | const struct extent_map *em, |
4264 | struct btrfs_path *path, | |
8407f553 FM |
4265 | struct btrfs_log_ctx *ctx) |
4266 | { | |
4267 | struct btrfs_root *log = root->log_root; | |
4268 | struct btrfs_file_extent_item *fi; | |
4269 | struct extent_buffer *leaf; | |
4270 | struct btrfs_map_token token; | |
4271 | struct btrfs_key key; | |
4272 | u64 extent_offset = em->start - em->orig_start; | |
4273 | u64 block_len; | |
4274 | int ret; | |
4275 | int extent_inserted = 0; | |
8407f553 | 4276 | |
a9ecb653 | 4277 | ret = log_extent_csums(trans, inode, log, em); |
8407f553 FM |
4278 | if (ret) |
4279 | return ret; | |
4280 | ||
8407f553 FM |
4281 | btrfs_init_map_token(&token); |
4282 | ||
9d122629 | 4283 | ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start, |
8407f553 FM |
4284 | em->start + em->len, NULL, 0, 1, |
4285 | sizeof(*fi), &extent_inserted); | |
4286 | if (ret) | |
4287 | return ret; | |
4288 | ||
4289 | if (!extent_inserted) { | |
9d122629 | 4290 | key.objectid = btrfs_ino(inode); |
8407f553 FM |
4291 | key.type = BTRFS_EXTENT_DATA_KEY; |
4292 | key.offset = em->start; | |
4293 | ||
4294 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4295 | sizeof(*fi)); | |
4296 | if (ret) | |
4297 | return ret; | |
4298 | } | |
4299 | leaf = path->nodes[0]; | |
4300 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4301 | struct btrfs_file_extent_item); | |
4302 | ||
50d9aa99 | 4303 | btrfs_set_token_file_extent_generation(leaf, fi, trans->transid, |
8407f553 FM |
4304 | &token); |
4305 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4306 | btrfs_set_token_file_extent_type(leaf, fi, | |
4307 | BTRFS_FILE_EXTENT_PREALLOC, | |
4308 | &token); | |
4309 | else | |
4310 | btrfs_set_token_file_extent_type(leaf, fi, | |
4311 | BTRFS_FILE_EXTENT_REG, | |
4312 | &token); | |
4313 | ||
4314 | block_len = max(em->block_len, em->orig_block_len); | |
4315 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
4316 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4317 | em->block_start, | |
4318 | &token); | |
4319 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4320 | &token); | |
4321 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
4322 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4323 | em->block_start - | |
4324 | extent_offset, &token); | |
4325 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4326 | &token); | |
4327 | } else { | |
4328 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token); | |
4329 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0, | |
4330 | &token); | |
4331 | } | |
4332 | ||
4333 | btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token); | |
4334 | btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token); | |
4335 | btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token); | |
4336 | btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type, | |
4337 | &token); | |
4338 | btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token); | |
4339 | btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token); | |
4340 | btrfs_mark_buffer_dirty(leaf); | |
4341 | ||
4342 | btrfs_release_path(path); | |
4343 | ||
4344 | return ret; | |
4345 | } | |
4346 | ||
31d11b83 FM |
4347 | /* |
4348 | * Log all prealloc extents beyond the inode's i_size to make sure we do not | |
4349 | * lose them after doing a fast fsync and replaying the log. We scan the | |
4350 | * subvolume's root instead of iterating the inode's extent map tree because | |
4351 | * otherwise we can log incorrect extent items based on extent map conversion. | |
4352 | * That can happen due to the fact that extent maps are merged when they | |
4353 | * are not in the extent map tree's list of modified extents. | |
4354 | */ | |
4355 | static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans, | |
4356 | struct btrfs_inode *inode, | |
4357 | struct btrfs_path *path) | |
4358 | { | |
4359 | struct btrfs_root *root = inode->root; | |
4360 | struct btrfs_key key; | |
4361 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
4362 | const u64 ino = btrfs_ino(inode); | |
4363 | struct btrfs_path *dst_path = NULL; | |
4364 | u64 last_extent = (u64)-1; | |
4365 | int ins_nr = 0; | |
4366 | int start_slot; | |
4367 | int ret; | |
4368 | ||
4369 | if (!(inode->flags & BTRFS_INODE_PREALLOC)) | |
4370 | return 0; | |
4371 | ||
4372 | key.objectid = ino; | |
4373 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4374 | key.offset = i_size; | |
4375 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4376 | if (ret < 0) | |
4377 | goto out; | |
4378 | ||
4379 | while (true) { | |
4380 | struct extent_buffer *leaf = path->nodes[0]; | |
4381 | int slot = path->slots[0]; | |
4382 | ||
4383 | if (slot >= btrfs_header_nritems(leaf)) { | |
4384 | if (ins_nr > 0) { | |
4385 | ret = copy_items(trans, inode, dst_path, path, | |
4386 | &last_extent, start_slot, | |
4387 | ins_nr, 1, 0); | |
4388 | if (ret < 0) | |
4389 | goto out; | |
4390 | ins_nr = 0; | |
4391 | } | |
4392 | ret = btrfs_next_leaf(root, path); | |
4393 | if (ret < 0) | |
4394 | goto out; | |
4395 | if (ret > 0) { | |
4396 | ret = 0; | |
4397 | break; | |
4398 | } | |
4399 | continue; | |
4400 | } | |
4401 | ||
4402 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4403 | if (key.objectid > ino) | |
4404 | break; | |
4405 | if (WARN_ON_ONCE(key.objectid < ino) || | |
4406 | key.type < BTRFS_EXTENT_DATA_KEY || | |
4407 | key.offset < i_size) { | |
4408 | path->slots[0]++; | |
4409 | continue; | |
4410 | } | |
4411 | if (last_extent == (u64)-1) { | |
4412 | last_extent = key.offset; | |
4413 | /* | |
4414 | * Avoid logging extent items logged in past fsync calls | |
4415 | * and leading to duplicate keys in the log tree. | |
4416 | */ | |
4417 | do { | |
4418 | ret = btrfs_truncate_inode_items(trans, | |
4419 | root->log_root, | |
4420 | &inode->vfs_inode, | |
4421 | i_size, | |
4422 | BTRFS_EXTENT_DATA_KEY); | |
4423 | } while (ret == -EAGAIN); | |
4424 | if (ret) | |
4425 | goto out; | |
4426 | } | |
4427 | if (ins_nr == 0) | |
4428 | start_slot = slot; | |
4429 | ins_nr++; | |
4430 | path->slots[0]++; | |
4431 | if (!dst_path) { | |
4432 | dst_path = btrfs_alloc_path(); | |
4433 | if (!dst_path) { | |
4434 | ret = -ENOMEM; | |
4435 | goto out; | |
4436 | } | |
4437 | } | |
4438 | } | |
4439 | if (ins_nr > 0) { | |
4440 | ret = copy_items(trans, inode, dst_path, path, &last_extent, | |
4441 | start_slot, ins_nr, 1, 0); | |
4442 | if (ret > 0) | |
4443 | ret = 0; | |
4444 | } | |
4445 | out: | |
4446 | btrfs_release_path(path); | |
4447 | btrfs_free_path(dst_path); | |
4448 | return ret; | |
4449 | } | |
4450 | ||
5dc562c5 JB |
4451 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, |
4452 | struct btrfs_root *root, | |
9d122629 | 4453 | struct btrfs_inode *inode, |
827463c4 | 4454 | struct btrfs_path *path, |
de0ee0ed FM |
4455 | struct btrfs_log_ctx *ctx, |
4456 | const u64 start, | |
4457 | const u64 end) | |
5dc562c5 | 4458 | { |
5dc562c5 JB |
4459 | struct extent_map *em, *n; |
4460 | struct list_head extents; | |
9d122629 | 4461 | struct extent_map_tree *tree = &inode->extent_tree; |
5dc562c5 JB |
4462 | u64 test_gen; |
4463 | int ret = 0; | |
2ab28f32 | 4464 | int num = 0; |
5dc562c5 JB |
4465 | |
4466 | INIT_LIST_HEAD(&extents); | |
4467 | ||
5dc562c5 JB |
4468 | write_lock(&tree->lock); |
4469 | test_gen = root->fs_info->last_trans_committed; | |
4470 | ||
4471 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
008c6753 FM |
4472 | /* |
4473 | * Skip extents outside our logging range. It's important to do | |
4474 | * it for correctness because if we don't ignore them, we may | |
4475 | * log them before their ordered extent completes, and therefore | |
4476 | * we could log them without logging their respective checksums | |
4477 | * (the checksum items are added to the csum tree at the very | |
4478 | * end of btrfs_finish_ordered_io()). Also leave such extents | |
4479 | * outside of our range in the list, since we may have another | |
4480 | * ranged fsync in the near future that needs them. If an extent | |
4481 | * outside our range corresponds to a hole, log it to avoid | |
4482 | * leaving gaps between extents (fsck will complain when we are | |
4483 | * not using the NO_HOLES feature). | |
4484 | */ | |
4485 | if ((em->start > end || em->start + em->len <= start) && | |
4486 | em->block_start != EXTENT_MAP_HOLE) | |
4487 | continue; | |
4488 | ||
5dc562c5 | 4489 | list_del_init(&em->list); |
2ab28f32 JB |
4490 | /* |
4491 | * Just an arbitrary number, this can be really CPU intensive | |
4492 | * once we start getting a lot of extents, and really once we | |
4493 | * have a bunch of extents we just want to commit since it will | |
4494 | * be faster. | |
4495 | */ | |
4496 | if (++num > 32768) { | |
4497 | list_del_init(&tree->modified_extents); | |
4498 | ret = -EFBIG; | |
4499 | goto process; | |
4500 | } | |
4501 | ||
5dc562c5 JB |
4502 | if (em->generation <= test_gen) |
4503 | continue; | |
8c6c5928 | 4504 | |
31d11b83 FM |
4505 | /* We log prealloc extents beyond eof later. */ |
4506 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && | |
4507 | em->start >= i_size_read(&inode->vfs_inode)) | |
4508 | continue; | |
4509 | ||
ff44c6e3 | 4510 | /* Need a ref to keep it from getting evicted from cache */ |
490b54d6 | 4511 | refcount_inc(&em->refs); |
ff44c6e3 | 4512 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); |
5dc562c5 | 4513 | list_add_tail(&em->list, &extents); |
2ab28f32 | 4514 | num++; |
5dc562c5 JB |
4515 | } |
4516 | ||
4517 | list_sort(NULL, &extents, extent_cmp); | |
2ab28f32 | 4518 | process: |
5dc562c5 JB |
4519 | while (!list_empty(&extents)) { |
4520 | em = list_entry(extents.next, struct extent_map, list); | |
4521 | ||
4522 | list_del_init(&em->list); | |
4523 | ||
4524 | /* | |
4525 | * If we had an error we just need to delete everybody from our | |
4526 | * private list. | |
4527 | */ | |
ff44c6e3 | 4528 | if (ret) { |
201a9038 | 4529 | clear_em_logging(tree, em); |
ff44c6e3 | 4530 | free_extent_map(em); |
5dc562c5 | 4531 | continue; |
ff44c6e3 JB |
4532 | } |
4533 | ||
4534 | write_unlock(&tree->lock); | |
5dc562c5 | 4535 | |
a2120a47 | 4536 | ret = log_one_extent(trans, inode, root, em, path, ctx); |
ff44c6e3 | 4537 | write_lock(&tree->lock); |
201a9038 JB |
4538 | clear_em_logging(tree, em); |
4539 | free_extent_map(em); | |
5dc562c5 | 4540 | } |
ff44c6e3 JB |
4541 | WARN_ON(!list_empty(&extents)); |
4542 | write_unlock(&tree->lock); | |
5dc562c5 | 4543 | |
5dc562c5 | 4544 | btrfs_release_path(path); |
31d11b83 FM |
4545 | if (!ret) |
4546 | ret = btrfs_log_prealloc_extents(trans, inode, path); | |
4547 | ||
5dc562c5 JB |
4548 | return ret; |
4549 | } | |
4550 | ||
481b01c0 | 4551 | static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode, |
1a4bcf47 FM |
4552 | struct btrfs_path *path, u64 *size_ret) |
4553 | { | |
4554 | struct btrfs_key key; | |
4555 | int ret; | |
4556 | ||
481b01c0 | 4557 | key.objectid = btrfs_ino(inode); |
1a4bcf47 FM |
4558 | key.type = BTRFS_INODE_ITEM_KEY; |
4559 | key.offset = 0; | |
4560 | ||
4561 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4562 | if (ret < 0) { | |
4563 | return ret; | |
4564 | } else if (ret > 0) { | |
2f2ff0ee | 4565 | *size_ret = 0; |
1a4bcf47 FM |
4566 | } else { |
4567 | struct btrfs_inode_item *item; | |
4568 | ||
4569 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4570 | struct btrfs_inode_item); | |
4571 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
bf504110 FM |
4572 | /* |
4573 | * If the in-memory inode's i_size is smaller then the inode | |
4574 | * size stored in the btree, return the inode's i_size, so | |
4575 | * that we get a correct inode size after replaying the log | |
4576 | * when before a power failure we had a shrinking truncate | |
4577 | * followed by addition of a new name (rename / new hard link). | |
4578 | * Otherwise return the inode size from the btree, to avoid | |
4579 | * data loss when replaying a log due to previously doing a | |
4580 | * write that expands the inode's size and logging a new name | |
4581 | * immediately after. | |
4582 | */ | |
4583 | if (*size_ret > inode->vfs_inode.i_size) | |
4584 | *size_ret = inode->vfs_inode.i_size; | |
1a4bcf47 FM |
4585 | } |
4586 | ||
4587 | btrfs_release_path(path); | |
4588 | return 0; | |
4589 | } | |
4590 | ||
36283bf7 FM |
4591 | /* |
4592 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4593 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4594 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4595 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4596 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4597 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4598 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4599 | */ | |
4600 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4601 | struct btrfs_root *root, | |
1a93c36a | 4602 | struct btrfs_inode *inode, |
36283bf7 FM |
4603 | struct btrfs_path *path, |
4604 | struct btrfs_path *dst_path) | |
4605 | { | |
4606 | int ret; | |
4607 | struct btrfs_key key; | |
1a93c36a | 4608 | const u64 ino = btrfs_ino(inode); |
36283bf7 FM |
4609 | int ins_nr = 0; |
4610 | int start_slot = 0; | |
4611 | ||
4612 | key.objectid = ino; | |
4613 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4614 | key.offset = 0; | |
4615 | ||
4616 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4617 | if (ret < 0) | |
4618 | return ret; | |
4619 | ||
4620 | while (true) { | |
4621 | int slot = path->slots[0]; | |
4622 | struct extent_buffer *leaf = path->nodes[0]; | |
4623 | int nritems = btrfs_header_nritems(leaf); | |
4624 | ||
4625 | if (slot >= nritems) { | |
4626 | if (ins_nr > 0) { | |
4627 | u64 last_extent = 0; | |
4628 | ||
1a93c36a | 4629 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
4630 | &last_extent, start_slot, |
4631 | ins_nr, 1, 0); | |
4632 | /* can't be 1, extent items aren't processed */ | |
4633 | ASSERT(ret <= 0); | |
4634 | if (ret < 0) | |
4635 | return ret; | |
4636 | ins_nr = 0; | |
4637 | } | |
4638 | ret = btrfs_next_leaf(root, path); | |
4639 | if (ret < 0) | |
4640 | return ret; | |
4641 | else if (ret > 0) | |
4642 | break; | |
4643 | continue; | |
4644 | } | |
4645 | ||
4646 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4647 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4648 | break; | |
4649 | ||
4650 | if (ins_nr == 0) | |
4651 | start_slot = slot; | |
4652 | ins_nr++; | |
4653 | path->slots[0]++; | |
4654 | cond_resched(); | |
4655 | } | |
4656 | if (ins_nr > 0) { | |
4657 | u64 last_extent = 0; | |
4658 | ||
1a93c36a | 4659 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
4660 | &last_extent, start_slot, |
4661 | ins_nr, 1, 0); | |
4662 | /* can't be 1, extent items aren't processed */ | |
4663 | ASSERT(ret <= 0); | |
4664 | if (ret < 0) | |
4665 | return ret; | |
4666 | } | |
4667 | ||
4668 | return 0; | |
4669 | } | |
4670 | ||
a89ca6f2 FM |
4671 | /* |
4672 | * If the no holes feature is enabled we need to make sure any hole between the | |
4673 | * last extent and the i_size of our inode is explicitly marked in the log. This | |
4674 | * is to make sure that doing something like: | |
4675 | * | |
4676 | * 1) create file with 128Kb of data | |
4677 | * 2) truncate file to 64Kb | |
4678 | * 3) truncate file to 256Kb | |
4679 | * 4) fsync file | |
4680 | * 5) <crash/power failure> | |
4681 | * 6) mount fs and trigger log replay | |
4682 | * | |
4683 | * Will give us a file with a size of 256Kb, the first 64Kb of data match what | |
4684 | * the file had in its first 64Kb of data at step 1 and the last 192Kb of the | |
4685 | * file correspond to a hole. The presence of explicit holes in a log tree is | |
4686 | * what guarantees that log replay will remove/adjust file extent items in the | |
4687 | * fs/subvol tree. | |
4688 | * | |
4689 | * Here we do not need to care about holes between extents, that is already done | |
4690 | * by copy_items(). We also only need to do this in the full sync path, where we | |
4691 | * lookup for extents from the fs/subvol tree only. In the fast path case, we | |
4692 | * lookup the list of modified extent maps and if any represents a hole, we | |
4693 | * insert a corresponding extent representing a hole in the log tree. | |
4694 | */ | |
4695 | static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans, | |
4696 | struct btrfs_root *root, | |
a0308dd7 | 4697 | struct btrfs_inode *inode, |
a89ca6f2 FM |
4698 | struct btrfs_path *path) |
4699 | { | |
0b246afa | 4700 | struct btrfs_fs_info *fs_info = root->fs_info; |
a89ca6f2 FM |
4701 | int ret; |
4702 | struct btrfs_key key; | |
4703 | u64 hole_start; | |
4704 | u64 hole_size; | |
4705 | struct extent_buffer *leaf; | |
4706 | struct btrfs_root *log = root->log_root; | |
a0308dd7 NB |
4707 | const u64 ino = btrfs_ino(inode); |
4708 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
a89ca6f2 | 4709 | |
0b246afa | 4710 | if (!btrfs_fs_incompat(fs_info, NO_HOLES)) |
a89ca6f2 FM |
4711 | return 0; |
4712 | ||
4713 | key.objectid = ino; | |
4714 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4715 | key.offset = (u64)-1; | |
4716 | ||
4717 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4718 | ASSERT(ret != 0); | |
4719 | if (ret < 0) | |
4720 | return ret; | |
4721 | ||
4722 | ASSERT(path->slots[0] > 0); | |
4723 | path->slots[0]--; | |
4724 | leaf = path->nodes[0]; | |
4725 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4726 | ||
4727 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) { | |
4728 | /* inode does not have any extents */ | |
4729 | hole_start = 0; | |
4730 | hole_size = i_size; | |
4731 | } else { | |
4732 | struct btrfs_file_extent_item *extent; | |
4733 | u64 len; | |
4734 | ||
4735 | /* | |
4736 | * If there's an extent beyond i_size, an explicit hole was | |
4737 | * already inserted by copy_items(). | |
4738 | */ | |
4739 | if (key.offset >= i_size) | |
4740 | return 0; | |
4741 | ||
4742 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
4743 | struct btrfs_file_extent_item); | |
4744 | ||
4745 | if (btrfs_file_extent_type(leaf, extent) == | |
0ccc3876 | 4746 | BTRFS_FILE_EXTENT_INLINE) |
a89ca6f2 | 4747 | return 0; |
a89ca6f2 FM |
4748 | |
4749 | len = btrfs_file_extent_num_bytes(leaf, extent); | |
4750 | /* Last extent goes beyond i_size, no need to log a hole. */ | |
4751 | if (key.offset + len > i_size) | |
4752 | return 0; | |
4753 | hole_start = key.offset + len; | |
4754 | hole_size = i_size - hole_start; | |
4755 | } | |
4756 | btrfs_release_path(path); | |
4757 | ||
4758 | /* Last extent ends at i_size. */ | |
4759 | if (hole_size == 0) | |
4760 | return 0; | |
4761 | ||
0b246afa | 4762 | hole_size = ALIGN(hole_size, fs_info->sectorsize); |
a89ca6f2 FM |
4763 | ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0, |
4764 | hole_size, 0, hole_size, 0, 0, 0); | |
4765 | return ret; | |
4766 | } | |
4767 | ||
56f23fdb FM |
4768 | /* |
4769 | * When we are logging a new inode X, check if it doesn't have a reference that | |
4770 | * matches the reference from some other inode Y created in a past transaction | |
4771 | * and that was renamed in the current transaction. If we don't do this, then at | |
4772 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
4773 | * | |
4774 | * mkdir /mnt/x | |
4775 | * echo "hello world" > /mnt/x/foobar | |
4776 | * sync | |
4777 | * mv /mnt/x /mnt/y | |
4778 | * mkdir /mnt/x # or touch /mnt/x | |
4779 | * xfs_io -c fsync /mnt/x | |
4780 | * <power fail> | |
4781 | * mount fs, trigger log replay | |
4782 | * | |
4783 | * After the log replay procedure, we would lose the first directory and all its | |
4784 | * files (file foobar). | |
4785 | * For the case where inode Y is not a directory we simply end up losing it: | |
4786 | * | |
4787 | * echo "123" > /mnt/foo | |
4788 | * sync | |
4789 | * mv /mnt/foo /mnt/bar | |
4790 | * echo "abc" > /mnt/foo | |
4791 | * xfs_io -c fsync /mnt/foo | |
4792 | * <power fail> | |
4793 | * | |
4794 | * We also need this for cases where a snapshot entry is replaced by some other | |
4795 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
4796 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
4797 | * if it were a regular entry: | |
4798 | * | |
4799 | * mkdir /mnt/x | |
4800 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
4801 | * btrfs subvolume delete /mnt/x/snap | |
4802 | * rmdir /mnt/x | |
4803 | * mkdir /mnt/x | |
4804 | * fsync /mnt/x or fsync some new file inside it | |
4805 | * <power fail> | |
4806 | * | |
4807 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
4808 | * the same transaction. | |
4809 | */ | |
4810 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
4811 | const int slot, | |
4812 | const struct btrfs_key *key, | |
4791c8f1 | 4813 | struct btrfs_inode *inode, |
a3baaf0d | 4814 | u64 *other_ino, u64 *other_parent) |
56f23fdb FM |
4815 | { |
4816 | int ret; | |
4817 | struct btrfs_path *search_path; | |
4818 | char *name = NULL; | |
4819 | u32 name_len = 0; | |
4820 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
4821 | u32 cur_offset = 0; | |
4822 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
4823 | ||
4824 | search_path = btrfs_alloc_path(); | |
4825 | if (!search_path) | |
4826 | return -ENOMEM; | |
4827 | search_path->search_commit_root = 1; | |
4828 | search_path->skip_locking = 1; | |
4829 | ||
4830 | while (cur_offset < item_size) { | |
4831 | u64 parent; | |
4832 | u32 this_name_len; | |
4833 | u32 this_len; | |
4834 | unsigned long name_ptr; | |
4835 | struct btrfs_dir_item *di; | |
4836 | ||
4837 | if (key->type == BTRFS_INODE_REF_KEY) { | |
4838 | struct btrfs_inode_ref *iref; | |
4839 | ||
4840 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
4841 | parent = key->offset; | |
4842 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
4843 | name_ptr = (unsigned long)(iref + 1); | |
4844 | this_len = sizeof(*iref) + this_name_len; | |
4845 | } else { | |
4846 | struct btrfs_inode_extref *extref; | |
4847 | ||
4848 | extref = (struct btrfs_inode_extref *)(ptr + | |
4849 | cur_offset); | |
4850 | parent = btrfs_inode_extref_parent(eb, extref); | |
4851 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
4852 | name_ptr = (unsigned long)&extref->name; | |
4853 | this_len = sizeof(*extref) + this_name_len; | |
4854 | } | |
4855 | ||
4856 | if (this_name_len > name_len) { | |
4857 | char *new_name; | |
4858 | ||
4859 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
4860 | if (!new_name) { | |
4861 | ret = -ENOMEM; | |
4862 | goto out; | |
4863 | } | |
4864 | name_len = this_name_len; | |
4865 | name = new_name; | |
4866 | } | |
4867 | ||
4868 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4791c8f1 NB |
4869 | di = btrfs_lookup_dir_item(NULL, inode->root, search_path, |
4870 | parent, name, this_name_len, 0); | |
56f23fdb | 4871 | if (di && !IS_ERR(di)) { |
44f714da FM |
4872 | struct btrfs_key di_key; |
4873 | ||
4874 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
4875 | di, &di_key); | |
4876 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
6b5fc433 FM |
4877 | if (di_key.objectid != key->objectid) { |
4878 | ret = 1; | |
4879 | *other_ino = di_key.objectid; | |
a3baaf0d | 4880 | *other_parent = parent; |
6b5fc433 FM |
4881 | } else { |
4882 | ret = 0; | |
4883 | } | |
44f714da FM |
4884 | } else { |
4885 | ret = -EAGAIN; | |
4886 | } | |
56f23fdb FM |
4887 | goto out; |
4888 | } else if (IS_ERR(di)) { | |
4889 | ret = PTR_ERR(di); | |
4890 | goto out; | |
4891 | } | |
4892 | btrfs_release_path(search_path); | |
4893 | ||
4894 | cur_offset += this_len; | |
4895 | } | |
4896 | ret = 0; | |
4897 | out: | |
4898 | btrfs_free_path(search_path); | |
4899 | kfree(name); | |
4900 | return ret; | |
4901 | } | |
4902 | ||
6b5fc433 FM |
4903 | struct btrfs_ino_list { |
4904 | u64 ino; | |
a3baaf0d | 4905 | u64 parent; |
6b5fc433 FM |
4906 | struct list_head list; |
4907 | }; | |
4908 | ||
4909 | static int log_conflicting_inodes(struct btrfs_trans_handle *trans, | |
4910 | struct btrfs_root *root, | |
4911 | struct btrfs_path *path, | |
4912 | struct btrfs_log_ctx *ctx, | |
a3baaf0d | 4913 | u64 ino, u64 parent) |
6b5fc433 FM |
4914 | { |
4915 | struct btrfs_ino_list *ino_elem; | |
4916 | LIST_HEAD(inode_list); | |
4917 | int ret = 0; | |
4918 | ||
4919 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
4920 | if (!ino_elem) | |
4921 | return -ENOMEM; | |
4922 | ino_elem->ino = ino; | |
a3baaf0d | 4923 | ino_elem->parent = parent; |
6b5fc433 FM |
4924 | list_add_tail(&ino_elem->list, &inode_list); |
4925 | ||
4926 | while (!list_empty(&inode_list)) { | |
4927 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4928 | struct btrfs_key key; | |
4929 | struct inode *inode; | |
4930 | ||
4931 | ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list, | |
4932 | list); | |
4933 | ino = ino_elem->ino; | |
a3baaf0d | 4934 | parent = ino_elem->parent; |
6b5fc433 FM |
4935 | list_del(&ino_elem->list); |
4936 | kfree(ino_elem); | |
4937 | if (ret) | |
4938 | continue; | |
4939 | ||
4940 | btrfs_release_path(path); | |
4941 | ||
4942 | key.objectid = ino; | |
4943 | key.type = BTRFS_INODE_ITEM_KEY; | |
4944 | key.offset = 0; | |
4945 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
4946 | /* | |
4947 | * If the other inode that had a conflicting dir entry was | |
a3baaf0d FM |
4948 | * deleted in the current transaction, we need to log its parent |
4949 | * directory. | |
6b5fc433 FM |
4950 | */ |
4951 | if (IS_ERR(inode)) { | |
4952 | ret = PTR_ERR(inode); | |
a3baaf0d FM |
4953 | if (ret == -ENOENT) { |
4954 | key.objectid = parent; | |
4955 | inode = btrfs_iget(fs_info->sb, &key, root, | |
4956 | NULL); | |
4957 | if (IS_ERR(inode)) { | |
4958 | ret = PTR_ERR(inode); | |
4959 | } else { | |
4960 | ret = btrfs_log_inode(trans, root, | |
4961 | BTRFS_I(inode), | |
4962 | LOG_OTHER_INODE_ALL, | |
4963 | 0, LLONG_MAX, ctx); | |
4964 | iput(inode); | |
4965 | } | |
4966 | } | |
6b5fc433 FM |
4967 | continue; |
4968 | } | |
4969 | /* | |
4970 | * We are safe logging the other inode without acquiring its | |
4971 | * lock as long as we log with the LOG_INODE_EXISTS mode. We | |
4972 | * are safe against concurrent renames of the other inode as | |
4973 | * well because during a rename we pin the log and update the | |
4974 | * log with the new name before we unpin it. | |
4975 | */ | |
4976 | ret = btrfs_log_inode(trans, root, BTRFS_I(inode), | |
4977 | LOG_OTHER_INODE, 0, LLONG_MAX, ctx); | |
4978 | if (ret) { | |
4979 | iput(inode); | |
4980 | continue; | |
4981 | } | |
4982 | ||
4983 | key.objectid = ino; | |
4984 | key.type = BTRFS_INODE_REF_KEY; | |
4985 | key.offset = 0; | |
4986 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4987 | if (ret < 0) { | |
4988 | iput(inode); | |
4989 | continue; | |
4990 | } | |
4991 | ||
4992 | while (true) { | |
4993 | struct extent_buffer *leaf = path->nodes[0]; | |
4994 | int slot = path->slots[0]; | |
4995 | u64 other_ino = 0; | |
a3baaf0d | 4996 | u64 other_parent = 0; |
6b5fc433 FM |
4997 | |
4998 | if (slot >= btrfs_header_nritems(leaf)) { | |
4999 | ret = btrfs_next_leaf(root, path); | |
5000 | if (ret < 0) { | |
5001 | break; | |
5002 | } else if (ret > 0) { | |
5003 | ret = 0; | |
5004 | break; | |
5005 | } | |
5006 | continue; | |
5007 | } | |
5008 | ||
5009 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5010 | if (key.objectid != ino || | |
5011 | (key.type != BTRFS_INODE_REF_KEY && | |
5012 | key.type != BTRFS_INODE_EXTREF_KEY)) { | |
5013 | ret = 0; | |
5014 | break; | |
5015 | } | |
5016 | ||
5017 | ret = btrfs_check_ref_name_override(leaf, slot, &key, | |
a3baaf0d FM |
5018 | BTRFS_I(inode), &other_ino, |
5019 | &other_parent); | |
6b5fc433 FM |
5020 | if (ret < 0) |
5021 | break; | |
5022 | if (ret > 0) { | |
5023 | ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS); | |
5024 | if (!ino_elem) { | |
5025 | ret = -ENOMEM; | |
5026 | break; | |
5027 | } | |
5028 | ino_elem->ino = other_ino; | |
a3baaf0d | 5029 | ino_elem->parent = other_parent; |
6b5fc433 FM |
5030 | list_add_tail(&ino_elem->list, &inode_list); |
5031 | ret = 0; | |
5032 | } | |
5033 | path->slots[0]++; | |
5034 | } | |
5035 | iput(inode); | |
5036 | } | |
5037 | ||
5038 | return ret; | |
5039 | } | |
5040 | ||
e02119d5 CM |
5041 | /* log a single inode in the tree log. |
5042 | * At least one parent directory for this inode must exist in the tree | |
5043 | * or be logged already. | |
5044 | * | |
5045 | * Any items from this inode changed by the current transaction are copied | |
5046 | * to the log tree. An extra reference is taken on any extents in this | |
5047 | * file, allowing us to avoid a whole pile of corner cases around logging | |
5048 | * blocks that have been removed from the tree. | |
5049 | * | |
5050 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
5051 | * does. | |
5052 | * | |
5053 | * This handles both files and directories. | |
5054 | */ | |
12fcfd22 | 5055 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
a59108a7 | 5056 | struct btrfs_root *root, struct btrfs_inode *inode, |
49dae1bc FM |
5057 | int inode_only, |
5058 | const loff_t start, | |
8407f553 FM |
5059 | const loff_t end, |
5060 | struct btrfs_log_ctx *ctx) | |
e02119d5 | 5061 | { |
0b246afa | 5062 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 CM |
5063 | struct btrfs_path *path; |
5064 | struct btrfs_path *dst_path; | |
5065 | struct btrfs_key min_key; | |
5066 | struct btrfs_key max_key; | |
5067 | struct btrfs_root *log = root->log_root; | |
16e7549f | 5068 | u64 last_extent = 0; |
4a500fd1 | 5069 | int err = 0; |
e02119d5 | 5070 | int ret; |
3a5f1d45 | 5071 | int nritems; |
31ff1cd2 CM |
5072 | int ins_start_slot = 0; |
5073 | int ins_nr; | |
5dc562c5 | 5074 | bool fast_search = false; |
a59108a7 NB |
5075 | u64 ino = btrfs_ino(inode); |
5076 | struct extent_map_tree *em_tree = &inode->extent_tree; | |
1a4bcf47 | 5077 | u64 logged_isize = 0; |
e4545de5 | 5078 | bool need_log_inode_item = true; |
9a8fca62 | 5079 | bool xattrs_logged = false; |
a3baaf0d | 5080 | bool recursive_logging = false; |
e02119d5 | 5081 | |
e02119d5 | 5082 | path = btrfs_alloc_path(); |
5df67083 TI |
5083 | if (!path) |
5084 | return -ENOMEM; | |
e02119d5 | 5085 | dst_path = btrfs_alloc_path(); |
5df67083 TI |
5086 | if (!dst_path) { |
5087 | btrfs_free_path(path); | |
5088 | return -ENOMEM; | |
5089 | } | |
e02119d5 | 5090 | |
33345d01 | 5091 | min_key.objectid = ino; |
e02119d5 CM |
5092 | min_key.type = BTRFS_INODE_ITEM_KEY; |
5093 | min_key.offset = 0; | |
5094 | ||
33345d01 | 5095 | max_key.objectid = ino; |
12fcfd22 | 5096 | |
12fcfd22 | 5097 | |
5dc562c5 | 5098 | /* today the code can only do partial logging of directories */ |
a59108a7 | 5099 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5269b67e | 5100 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5101 | &inode->runtime_flags) && |
781feef7 | 5102 | inode_only >= LOG_INODE_EXISTS)) |
e02119d5 CM |
5103 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5104 | else | |
5105 | max_key.type = (u8)-1; | |
5106 | max_key.offset = (u64)-1; | |
5107 | ||
2c2c452b FM |
5108 | /* |
5109 | * Only run delayed items if we are a dir or a new file. | |
5110 | * Otherwise commit the delayed inode only, which is needed in | |
5111 | * order for the log replay code to mark inodes for link count | |
5112 | * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items). | |
5113 | */ | |
a59108a7 NB |
5114 | if (S_ISDIR(inode->vfs_inode.i_mode) || |
5115 | inode->generation > fs_info->last_trans_committed) | |
5116 | ret = btrfs_commit_inode_delayed_items(trans, inode); | |
2c2c452b | 5117 | else |
a59108a7 | 5118 | ret = btrfs_commit_inode_delayed_inode(inode); |
2c2c452b FM |
5119 | |
5120 | if (ret) { | |
5121 | btrfs_free_path(path); | |
5122 | btrfs_free_path(dst_path); | |
5123 | return ret; | |
16cdcec7 MX |
5124 | } |
5125 | ||
a3baaf0d FM |
5126 | if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) { |
5127 | recursive_logging = true; | |
5128 | if (inode_only == LOG_OTHER_INODE) | |
5129 | inode_only = LOG_INODE_EXISTS; | |
5130 | else | |
5131 | inode_only = LOG_INODE_ALL; | |
a59108a7 | 5132 | mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING); |
781feef7 | 5133 | } else { |
a59108a7 | 5134 | mutex_lock(&inode->log_mutex); |
781feef7 | 5135 | } |
e02119d5 CM |
5136 | |
5137 | /* | |
5138 | * a brute force approach to making sure we get the most uptodate | |
5139 | * copies of everything. | |
5140 | */ | |
a59108a7 | 5141 | if (S_ISDIR(inode->vfs_inode.i_mode)) { |
e02119d5 CM |
5142 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; |
5143 | ||
4f764e51 FM |
5144 | if (inode_only == LOG_INODE_EXISTS) |
5145 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
33345d01 | 5146 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
e02119d5 | 5147 | } else { |
1a4bcf47 FM |
5148 | if (inode_only == LOG_INODE_EXISTS) { |
5149 | /* | |
5150 | * Make sure the new inode item we write to the log has | |
5151 | * the same isize as the current one (if it exists). | |
5152 | * This is necessary to prevent data loss after log | |
5153 | * replay, and also to prevent doing a wrong expanding | |
5154 | * truncate - for e.g. create file, write 4K into offset | |
5155 | * 0, fsync, write 4K into offset 4096, add hard link, | |
5156 | * fsync some other file (to sync log), power fail - if | |
5157 | * we use the inode's current i_size, after log replay | |
5158 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
5159 | * (zeroes), as if an expanding truncate happened, | |
5160 | * instead of getting a file of 4Kb only. | |
5161 | */ | |
a59108a7 | 5162 | err = logged_inode_size(log, inode, path, &logged_isize); |
1a4bcf47 FM |
5163 | if (err) |
5164 | goto out_unlock; | |
5165 | } | |
a742994a | 5166 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
a59108a7 | 5167 | &inode->runtime_flags)) { |
a742994a | 5168 | if (inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5169 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
a742994a FM |
5170 | ret = drop_objectid_items(trans, log, path, ino, |
5171 | max_key.type); | |
5172 | } else { | |
5173 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
a59108a7 | 5174 | &inode->runtime_flags); |
a742994a | 5175 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5176 | &inode->runtime_flags); |
28ed1345 CM |
5177 | while(1) { |
5178 | ret = btrfs_truncate_inode_items(trans, | |
a59108a7 | 5179 | log, &inode->vfs_inode, 0, 0); |
28ed1345 CM |
5180 | if (ret != -EAGAIN) |
5181 | break; | |
5182 | } | |
a742994a | 5183 | } |
4f764e51 | 5184 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, |
a59108a7 | 5185 | &inode->runtime_flags) || |
6cfab851 | 5186 | inode_only == LOG_INODE_EXISTS) { |
4f764e51 | 5187 | if (inode_only == LOG_INODE_ALL) |
183f37fa | 5188 | fast_search = true; |
4f764e51 | 5189 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
5dc562c5 | 5190 | ret = drop_objectid_items(trans, log, path, ino, |
e9976151 | 5191 | max_key.type); |
a95249b3 JB |
5192 | } else { |
5193 | if (inode_only == LOG_INODE_ALL) | |
5194 | fast_search = true; | |
a95249b3 | 5195 | goto log_extents; |
5dc562c5 | 5196 | } |
a95249b3 | 5197 | |
e02119d5 | 5198 | } |
4a500fd1 YZ |
5199 | if (ret) { |
5200 | err = ret; | |
5201 | goto out_unlock; | |
5202 | } | |
e02119d5 | 5203 | |
d397712b | 5204 | while (1) { |
31ff1cd2 | 5205 | ins_nr = 0; |
6174d3cb | 5206 | ret = btrfs_search_forward(root, &min_key, |
de78b51a | 5207 | path, trans->transid); |
fb770ae4 LB |
5208 | if (ret < 0) { |
5209 | err = ret; | |
5210 | goto out_unlock; | |
5211 | } | |
e02119d5 CM |
5212 | if (ret != 0) |
5213 | break; | |
3a5f1d45 | 5214 | again: |
31ff1cd2 | 5215 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ |
33345d01 | 5216 | if (min_key.objectid != ino) |
e02119d5 CM |
5217 | break; |
5218 | if (min_key.type > max_key.type) | |
5219 | break; | |
31ff1cd2 | 5220 | |
e4545de5 FM |
5221 | if (min_key.type == BTRFS_INODE_ITEM_KEY) |
5222 | need_log_inode_item = false; | |
5223 | ||
56f23fdb FM |
5224 | if ((min_key.type == BTRFS_INODE_REF_KEY || |
5225 | min_key.type == BTRFS_INODE_EXTREF_KEY) && | |
6b5fc433 FM |
5226 | inode->generation == trans->transid && |
5227 | !recursive_logging) { | |
44f714da | 5228 | u64 other_ino = 0; |
a3baaf0d | 5229 | u64 other_parent = 0; |
44f714da | 5230 | |
56f23fdb | 5231 | ret = btrfs_check_ref_name_override(path->nodes[0], |
a59108a7 | 5232 | path->slots[0], &min_key, inode, |
a3baaf0d | 5233 | &other_ino, &other_parent); |
56f23fdb FM |
5234 | if (ret < 0) { |
5235 | err = ret; | |
5236 | goto out_unlock; | |
28a23593 | 5237 | } else if (ret > 0 && ctx && |
4a0cc7ca | 5238 | other_ino != btrfs_ino(BTRFS_I(ctx->inode))) { |
44f714da FM |
5239 | if (ins_nr > 0) { |
5240 | ins_nr++; | |
5241 | } else { | |
5242 | ins_nr = 1; | |
5243 | ins_start_slot = path->slots[0]; | |
5244 | } | |
a59108a7 | 5245 | ret = copy_items(trans, inode, dst_path, path, |
44f714da FM |
5246 | &last_extent, ins_start_slot, |
5247 | ins_nr, inode_only, | |
5248 | logged_isize); | |
5249 | if (ret < 0) { | |
5250 | err = ret; | |
5251 | goto out_unlock; | |
5252 | } | |
5253 | ins_nr = 0; | |
6b5fc433 FM |
5254 | |
5255 | err = log_conflicting_inodes(trans, root, path, | |
a3baaf0d | 5256 | ctx, other_ino, other_parent); |
44f714da FM |
5257 | if (err) |
5258 | goto out_unlock; | |
6b5fc433 FM |
5259 | btrfs_release_path(path); |
5260 | goto next_key; | |
56f23fdb FM |
5261 | } |
5262 | } | |
5263 | ||
36283bf7 FM |
5264 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ |
5265 | if (min_key.type == BTRFS_XATTR_ITEM_KEY) { | |
5266 | if (ins_nr == 0) | |
5267 | goto next_slot; | |
a59108a7 | 5268 | ret = copy_items(trans, inode, dst_path, path, |
36283bf7 FM |
5269 | &last_extent, ins_start_slot, |
5270 | ins_nr, inode_only, logged_isize); | |
5271 | if (ret < 0) { | |
5272 | err = ret; | |
5273 | goto out_unlock; | |
5274 | } | |
5275 | ins_nr = 0; | |
5276 | if (ret) { | |
5277 | btrfs_release_path(path); | |
5278 | continue; | |
5279 | } | |
5280 | goto next_slot; | |
5281 | } | |
5282 | ||
31ff1cd2 CM |
5283 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { |
5284 | ins_nr++; | |
5285 | goto next_slot; | |
5286 | } else if (!ins_nr) { | |
5287 | ins_start_slot = path->slots[0]; | |
5288 | ins_nr = 1; | |
5289 | goto next_slot; | |
e02119d5 CM |
5290 | } |
5291 | ||
a59108a7 | 5292 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
5293 | ins_start_slot, ins_nr, inode_only, |
5294 | logged_isize); | |
16e7549f | 5295 | if (ret < 0) { |
4a500fd1 YZ |
5296 | err = ret; |
5297 | goto out_unlock; | |
a71db86e RV |
5298 | } |
5299 | if (ret) { | |
16e7549f JB |
5300 | ins_nr = 0; |
5301 | btrfs_release_path(path); | |
5302 | continue; | |
4a500fd1 | 5303 | } |
31ff1cd2 CM |
5304 | ins_nr = 1; |
5305 | ins_start_slot = path->slots[0]; | |
5306 | next_slot: | |
e02119d5 | 5307 | |
3a5f1d45 CM |
5308 | nritems = btrfs_header_nritems(path->nodes[0]); |
5309 | path->slots[0]++; | |
5310 | if (path->slots[0] < nritems) { | |
5311 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, | |
5312 | path->slots[0]); | |
5313 | goto again; | |
5314 | } | |
31ff1cd2 | 5315 | if (ins_nr) { |
a59108a7 | 5316 | ret = copy_items(trans, inode, dst_path, path, |
16e7549f | 5317 | &last_extent, ins_start_slot, |
1a4bcf47 | 5318 | ins_nr, inode_only, logged_isize); |
16e7549f | 5319 | if (ret < 0) { |
4a500fd1 YZ |
5320 | err = ret; |
5321 | goto out_unlock; | |
5322 | } | |
16e7549f | 5323 | ret = 0; |
31ff1cd2 CM |
5324 | ins_nr = 0; |
5325 | } | |
b3b4aa74 | 5326 | btrfs_release_path(path); |
44f714da | 5327 | next_key: |
3d41d702 | 5328 | if (min_key.offset < (u64)-1) { |
e02119d5 | 5329 | min_key.offset++; |
3d41d702 | 5330 | } else if (min_key.type < max_key.type) { |
e02119d5 | 5331 | min_key.type++; |
3d41d702 FDBM |
5332 | min_key.offset = 0; |
5333 | } else { | |
e02119d5 | 5334 | break; |
3d41d702 | 5335 | } |
e02119d5 | 5336 | } |
31ff1cd2 | 5337 | if (ins_nr) { |
a59108a7 | 5338 | ret = copy_items(trans, inode, dst_path, path, &last_extent, |
1a4bcf47 FM |
5339 | ins_start_slot, ins_nr, inode_only, |
5340 | logged_isize); | |
16e7549f | 5341 | if (ret < 0) { |
4a500fd1 YZ |
5342 | err = ret; |
5343 | goto out_unlock; | |
5344 | } | |
16e7549f | 5345 | ret = 0; |
31ff1cd2 CM |
5346 | ins_nr = 0; |
5347 | } | |
5dc562c5 | 5348 | |
36283bf7 FM |
5349 | btrfs_release_path(path); |
5350 | btrfs_release_path(dst_path); | |
a59108a7 | 5351 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); |
36283bf7 FM |
5352 | if (err) |
5353 | goto out_unlock; | |
9a8fca62 | 5354 | xattrs_logged = true; |
a89ca6f2 FM |
5355 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { |
5356 | btrfs_release_path(path); | |
5357 | btrfs_release_path(dst_path); | |
a59108a7 | 5358 | err = btrfs_log_trailing_hole(trans, root, inode, path); |
a89ca6f2 FM |
5359 | if (err) |
5360 | goto out_unlock; | |
5361 | } | |
a95249b3 | 5362 | log_extents: |
f3b15ccd JB |
5363 | btrfs_release_path(path); |
5364 | btrfs_release_path(dst_path); | |
e4545de5 | 5365 | if (need_log_inode_item) { |
a59108a7 | 5366 | err = log_inode_item(trans, log, dst_path, inode); |
9a8fca62 FM |
5367 | if (!err && !xattrs_logged) { |
5368 | err = btrfs_log_all_xattrs(trans, root, inode, path, | |
5369 | dst_path); | |
5370 | btrfs_release_path(path); | |
5371 | } | |
e4545de5 FM |
5372 | if (err) |
5373 | goto out_unlock; | |
5374 | } | |
5dc562c5 | 5375 | if (fast_search) { |
a59108a7 | 5376 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, |
a2120a47 | 5377 | ctx, start, end); |
5dc562c5 JB |
5378 | if (ret) { |
5379 | err = ret; | |
5380 | goto out_unlock; | |
5381 | } | |
d006a048 | 5382 | } else if (inode_only == LOG_INODE_ALL) { |
06d3d22b LB |
5383 | struct extent_map *em, *n; |
5384 | ||
49dae1bc FM |
5385 | write_lock(&em_tree->lock); |
5386 | /* | |
5387 | * We can't just remove every em if we're called for a ranged | |
5388 | * fsync - that is, one that doesn't cover the whole possible | |
5389 | * file range (0 to LLONG_MAX). This is because we can have | |
5390 | * em's that fall outside the range we're logging and therefore | |
5391 | * their ordered operations haven't completed yet | |
5392 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
5393 | * didn't get their respective file extent item in the fs/subvol | |
5394 | * tree yet, and need to let the next fast fsync (one which | |
5395 | * consults the list of modified extent maps) find the em so | |
5396 | * that it logs a matching file extent item and waits for the | |
5397 | * respective ordered operation to complete (if it's still | |
5398 | * running). | |
5399 | * | |
5400 | * Removing every em outside the range we're logging would make | |
5401 | * the next fast fsync not log their matching file extent items, | |
5402 | * therefore making us lose data after a log replay. | |
5403 | */ | |
5404 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
5405 | list) { | |
5406 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
5407 | ||
5408 | if (em->mod_start >= start && mod_end <= end) | |
5409 | list_del_init(&em->list); | |
5410 | } | |
5411 | write_unlock(&em_tree->lock); | |
5dc562c5 JB |
5412 | } |
5413 | ||
a59108a7 NB |
5414 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) { |
5415 | ret = log_directory_changes(trans, root, inode, path, dst_path, | |
5416 | ctx); | |
4a500fd1 YZ |
5417 | if (ret) { |
5418 | err = ret; | |
5419 | goto out_unlock; | |
5420 | } | |
e02119d5 | 5421 | } |
49dae1bc | 5422 | |
a59108a7 NB |
5423 | spin_lock(&inode->lock); |
5424 | inode->logged_trans = trans->transid; | |
5425 | inode->last_log_commit = inode->last_sub_trans; | |
5426 | spin_unlock(&inode->lock); | |
4a500fd1 | 5427 | out_unlock: |
a59108a7 | 5428 | mutex_unlock(&inode->log_mutex); |
e02119d5 CM |
5429 | |
5430 | btrfs_free_path(path); | |
5431 | btrfs_free_path(dst_path); | |
4a500fd1 | 5432 | return err; |
e02119d5 CM |
5433 | } |
5434 | ||
2be63d5c FM |
5435 | /* |
5436 | * Check if we must fallback to a transaction commit when logging an inode. | |
5437 | * This must be called after logging the inode and is used only in the context | |
5438 | * when fsyncing an inode requires the need to log some other inode - in which | |
5439 | * case we can't lock the i_mutex of each other inode we need to log as that | |
5440 | * can lead to deadlocks with concurrent fsync against other inodes (as we can | |
5441 | * log inodes up or down in the hierarchy) or rename operations for example. So | |
5442 | * we take the log_mutex of the inode after we have logged it and then check for | |
5443 | * its last_unlink_trans value - this is safe because any task setting | |
5444 | * last_unlink_trans must take the log_mutex and it must do this before it does | |
5445 | * the actual unlink operation, so if we do this check before a concurrent task | |
5446 | * sets last_unlink_trans it means we've logged a consistent version/state of | |
5447 | * all the inode items, otherwise we are not sure and must do a transaction | |
01327610 | 5448 | * commit (the concurrent task might have only updated last_unlink_trans before |
2be63d5c FM |
5449 | * we logged the inode or it might have also done the unlink). |
5450 | */ | |
5451 | static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans, | |
ab1717b2 | 5452 | struct btrfs_inode *inode) |
2be63d5c | 5453 | { |
ab1717b2 | 5454 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2be63d5c FM |
5455 | bool ret = false; |
5456 | ||
ab1717b2 NB |
5457 | mutex_lock(&inode->log_mutex); |
5458 | if (inode->last_unlink_trans > fs_info->last_trans_committed) { | |
2be63d5c FM |
5459 | /* |
5460 | * Make sure any commits to the log are forced to be full | |
5461 | * commits. | |
5462 | */ | |
90787766 | 5463 | btrfs_set_log_full_commit(trans); |
2be63d5c FM |
5464 | ret = true; |
5465 | } | |
ab1717b2 | 5466 | mutex_unlock(&inode->log_mutex); |
2be63d5c FM |
5467 | |
5468 | return ret; | |
5469 | } | |
5470 | ||
12fcfd22 CM |
5471 | /* |
5472 | * follow the dentry parent pointers up the chain and see if any | |
5473 | * of the directories in it require a full commit before they can | |
5474 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
5475 | * a full commit is required. | |
5476 | */ | |
5477 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
aefa6115 | 5478 | struct btrfs_inode *inode, |
12fcfd22 CM |
5479 | struct dentry *parent, |
5480 | struct super_block *sb, | |
5481 | u64 last_committed) | |
e02119d5 | 5482 | { |
12fcfd22 | 5483 | int ret = 0; |
6a912213 | 5484 | struct dentry *old_parent = NULL; |
e02119d5 | 5485 | |
af4176b4 CM |
5486 | /* |
5487 | * for regular files, if its inode is already on disk, we don't | |
5488 | * have to worry about the parents at all. This is because | |
5489 | * we can use the last_unlink_trans field to record renames | |
5490 | * and other fun in this file. | |
5491 | */ | |
aefa6115 NB |
5492 | if (S_ISREG(inode->vfs_inode.i_mode) && |
5493 | inode->generation <= last_committed && | |
5494 | inode->last_unlink_trans <= last_committed) | |
5495 | goto out; | |
af4176b4 | 5496 | |
aefa6115 | 5497 | if (!S_ISDIR(inode->vfs_inode.i_mode)) { |
fc64005c | 5498 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 | 5499 | goto out; |
aefa6115 | 5500 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5501 | } |
5502 | ||
5503 | while (1) { | |
aefa6115 | 5504 | if (btrfs_must_commit_transaction(trans, inode)) { |
12fcfd22 CM |
5505 | ret = 1; |
5506 | break; | |
5507 | } | |
5508 | ||
fc64005c | 5509 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) |
12fcfd22 CM |
5510 | break; |
5511 | ||
44f714da | 5512 | if (IS_ROOT(parent)) { |
aefa6115 NB |
5513 | inode = BTRFS_I(d_inode(parent)); |
5514 | if (btrfs_must_commit_transaction(trans, inode)) | |
44f714da | 5515 | ret = 1; |
12fcfd22 | 5516 | break; |
44f714da | 5517 | } |
12fcfd22 | 5518 | |
6a912213 JB |
5519 | parent = dget_parent(parent); |
5520 | dput(old_parent); | |
5521 | old_parent = parent; | |
aefa6115 | 5522 | inode = BTRFS_I(d_inode(parent)); |
12fcfd22 CM |
5523 | |
5524 | } | |
6a912213 | 5525 | dput(old_parent); |
12fcfd22 | 5526 | out: |
e02119d5 CM |
5527 | return ret; |
5528 | } | |
5529 | ||
2f2ff0ee FM |
5530 | struct btrfs_dir_list { |
5531 | u64 ino; | |
5532 | struct list_head list; | |
5533 | }; | |
5534 | ||
5535 | /* | |
5536 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
5537 | * details about the why it is needed. | |
5538 | * This is a recursive operation - if an existing dentry corresponds to a | |
5539 | * directory, that directory's new entries are logged too (same behaviour as | |
5540 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5541 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
5542 | * complains about the following circular lock dependency / possible deadlock: | |
5543 | * | |
5544 | * CPU0 CPU1 | |
5545 | * ---- ---- | |
5546 | * lock(&type->i_mutex_dir_key#3/2); | |
5547 | * lock(sb_internal#2); | |
5548 | * lock(&type->i_mutex_dir_key#3/2); | |
5549 | * lock(&sb->s_type->i_mutex_key#14); | |
5550 | * | |
5551 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5552 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5553 | * Not locking i_mutex of the inodes is still safe because: | |
5554 | * | |
5555 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5556 | * that while logging the inode new references (names) are added or removed | |
5557 | * from the inode, leaving the logged inode item with a link count that does | |
5558 | * not match the number of logged inode reference items. This is fine because | |
5559 | * at log replay time we compute the real number of links and correct the | |
5560 | * link count in the inode item (see replay_one_buffer() and | |
5561 | * link_to_fixup_dir()); | |
5562 | * | |
5563 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5564 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
5565 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
5566 | * has a size that doesn't match the sum of the lengths of all the logged | |
5567 | * names. This does not result in a problem because if a dir_item key is | |
5568 | * logged but its matching dir_index key is not logged, at log replay time we | |
5569 | * don't use it to replay the respective name (see replay_one_name()). On the | |
5570 | * other hand if only the dir_index key ends up being logged, the respective | |
5571 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
5572 | * keys created (see replay_one_name()). | |
5573 | * The directory's inode item with a wrong i_size is not a problem as well, | |
5574 | * since we don't use it at log replay time to set the i_size in the inode | |
5575 | * item of the fs/subvol tree (see overwrite_item()). | |
5576 | */ | |
5577 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5578 | struct btrfs_root *root, | |
51cc0d32 | 5579 | struct btrfs_inode *start_inode, |
2f2ff0ee FM |
5580 | struct btrfs_log_ctx *ctx) |
5581 | { | |
0b246afa | 5582 | struct btrfs_fs_info *fs_info = root->fs_info; |
2f2ff0ee FM |
5583 | struct btrfs_root *log = root->log_root; |
5584 | struct btrfs_path *path; | |
5585 | LIST_HEAD(dir_list); | |
5586 | struct btrfs_dir_list *dir_elem; | |
5587 | int ret = 0; | |
5588 | ||
5589 | path = btrfs_alloc_path(); | |
5590 | if (!path) | |
5591 | return -ENOMEM; | |
5592 | ||
5593 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5594 | if (!dir_elem) { | |
5595 | btrfs_free_path(path); | |
5596 | return -ENOMEM; | |
5597 | } | |
51cc0d32 | 5598 | dir_elem->ino = btrfs_ino(start_inode); |
2f2ff0ee FM |
5599 | list_add_tail(&dir_elem->list, &dir_list); |
5600 | ||
5601 | while (!list_empty(&dir_list)) { | |
5602 | struct extent_buffer *leaf; | |
5603 | struct btrfs_key min_key; | |
5604 | int nritems; | |
5605 | int i; | |
5606 | ||
5607 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
5608 | list); | |
5609 | if (ret) | |
5610 | goto next_dir_inode; | |
5611 | ||
5612 | min_key.objectid = dir_elem->ino; | |
5613 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
5614 | min_key.offset = 0; | |
5615 | again: | |
5616 | btrfs_release_path(path); | |
5617 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
5618 | if (ret < 0) { | |
5619 | goto next_dir_inode; | |
5620 | } else if (ret > 0) { | |
5621 | ret = 0; | |
5622 | goto next_dir_inode; | |
5623 | } | |
5624 | ||
5625 | process_leaf: | |
5626 | leaf = path->nodes[0]; | |
5627 | nritems = btrfs_header_nritems(leaf); | |
5628 | for (i = path->slots[0]; i < nritems; i++) { | |
5629 | struct btrfs_dir_item *di; | |
5630 | struct btrfs_key di_key; | |
5631 | struct inode *di_inode; | |
5632 | struct btrfs_dir_list *new_dir_elem; | |
5633 | int log_mode = LOG_INODE_EXISTS; | |
5634 | int type; | |
5635 | ||
5636 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5637 | if (min_key.objectid != dir_elem->ino || | |
5638 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5639 | goto next_dir_inode; | |
5640 | ||
5641 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5642 | type = btrfs_dir_type(leaf, di); | |
5643 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5644 | type != BTRFS_FT_DIR) | |
5645 | continue; | |
5646 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5647 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5648 | continue; | |
5649 | ||
ec125cfb | 5650 | btrfs_release_path(path); |
0b246afa | 5651 | di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL); |
2f2ff0ee FM |
5652 | if (IS_ERR(di_inode)) { |
5653 | ret = PTR_ERR(di_inode); | |
5654 | goto next_dir_inode; | |
5655 | } | |
5656 | ||
0f8939b8 | 5657 | if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) { |
2f2ff0ee | 5658 | iput(di_inode); |
ec125cfb | 5659 | break; |
2f2ff0ee FM |
5660 | } |
5661 | ||
5662 | ctx->log_new_dentries = false; | |
3f9749f6 | 5663 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) |
2f2ff0ee | 5664 | log_mode = LOG_INODE_ALL; |
a59108a7 | 5665 | ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode), |
2f2ff0ee | 5666 | log_mode, 0, LLONG_MAX, ctx); |
2be63d5c | 5667 | if (!ret && |
ab1717b2 | 5668 | btrfs_must_commit_transaction(trans, BTRFS_I(di_inode))) |
2be63d5c | 5669 | ret = 1; |
2f2ff0ee FM |
5670 | iput(di_inode); |
5671 | if (ret) | |
5672 | goto next_dir_inode; | |
5673 | if (ctx->log_new_dentries) { | |
5674 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5675 | GFP_NOFS); | |
5676 | if (!new_dir_elem) { | |
5677 | ret = -ENOMEM; | |
5678 | goto next_dir_inode; | |
5679 | } | |
5680 | new_dir_elem->ino = di_key.objectid; | |
5681 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5682 | } | |
5683 | break; | |
5684 | } | |
5685 | if (i == nritems) { | |
5686 | ret = btrfs_next_leaf(log, path); | |
5687 | if (ret < 0) { | |
5688 | goto next_dir_inode; | |
5689 | } else if (ret > 0) { | |
5690 | ret = 0; | |
5691 | goto next_dir_inode; | |
5692 | } | |
5693 | goto process_leaf; | |
5694 | } | |
5695 | if (min_key.offset < (u64)-1) { | |
5696 | min_key.offset++; | |
5697 | goto again; | |
5698 | } | |
5699 | next_dir_inode: | |
5700 | list_del(&dir_elem->list); | |
5701 | kfree(dir_elem); | |
5702 | } | |
5703 | ||
5704 | btrfs_free_path(path); | |
5705 | return ret; | |
5706 | } | |
5707 | ||
18aa0922 | 5708 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, |
d0a0b78d | 5709 | struct btrfs_inode *inode, |
18aa0922 FM |
5710 | struct btrfs_log_ctx *ctx) |
5711 | { | |
3ffbd68c | 5712 | struct btrfs_fs_info *fs_info = trans->fs_info; |
18aa0922 FM |
5713 | int ret; |
5714 | struct btrfs_path *path; | |
5715 | struct btrfs_key key; | |
d0a0b78d NB |
5716 | struct btrfs_root *root = inode->root; |
5717 | const u64 ino = btrfs_ino(inode); | |
18aa0922 FM |
5718 | |
5719 | path = btrfs_alloc_path(); | |
5720 | if (!path) | |
5721 | return -ENOMEM; | |
5722 | path->skip_locking = 1; | |
5723 | path->search_commit_root = 1; | |
5724 | ||
5725 | key.objectid = ino; | |
5726 | key.type = BTRFS_INODE_REF_KEY; | |
5727 | key.offset = 0; | |
5728 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5729 | if (ret < 0) | |
5730 | goto out; | |
5731 | ||
5732 | while (true) { | |
5733 | struct extent_buffer *leaf = path->nodes[0]; | |
5734 | int slot = path->slots[0]; | |
5735 | u32 cur_offset = 0; | |
5736 | u32 item_size; | |
5737 | unsigned long ptr; | |
5738 | ||
5739 | if (slot >= btrfs_header_nritems(leaf)) { | |
5740 | ret = btrfs_next_leaf(root, path); | |
5741 | if (ret < 0) | |
5742 | goto out; | |
5743 | else if (ret > 0) | |
5744 | break; | |
5745 | continue; | |
5746 | } | |
5747 | ||
5748 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5749 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5750 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5751 | break; | |
5752 | ||
5753 | item_size = btrfs_item_size_nr(leaf, slot); | |
5754 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5755 | while (cur_offset < item_size) { | |
5756 | struct btrfs_key inode_key; | |
5757 | struct inode *dir_inode; | |
5758 | ||
5759 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5760 | inode_key.offset = 0; | |
5761 | ||
5762 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5763 | struct btrfs_inode_extref *extref; | |
5764 | ||
5765 | extref = (struct btrfs_inode_extref *) | |
5766 | (ptr + cur_offset); | |
5767 | inode_key.objectid = btrfs_inode_extref_parent( | |
5768 | leaf, extref); | |
5769 | cur_offset += sizeof(*extref); | |
5770 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5771 | extref); | |
5772 | } else { | |
5773 | inode_key.objectid = key.offset; | |
5774 | cur_offset = item_size; | |
5775 | } | |
5776 | ||
0b246afa | 5777 | dir_inode = btrfs_iget(fs_info->sb, &inode_key, |
18aa0922 | 5778 | root, NULL); |
0f375eed FM |
5779 | /* |
5780 | * If the parent inode was deleted, return an error to | |
5781 | * fallback to a transaction commit. This is to prevent | |
5782 | * getting an inode that was moved from one parent A to | |
5783 | * a parent B, got its former parent A deleted and then | |
5784 | * it got fsync'ed, from existing at both parents after | |
5785 | * a log replay (and the old parent still existing). | |
5786 | * Example: | |
5787 | * | |
5788 | * mkdir /mnt/A | |
5789 | * mkdir /mnt/B | |
5790 | * touch /mnt/B/bar | |
5791 | * sync | |
5792 | * mv /mnt/B/bar /mnt/A/bar | |
5793 | * mv -T /mnt/A /mnt/B | |
5794 | * fsync /mnt/B/bar | |
5795 | * <power fail> | |
5796 | * | |
5797 | * If we ignore the old parent B which got deleted, | |
5798 | * after a log replay we would have file bar linked | |
5799 | * at both parents and the old parent B would still | |
5800 | * exist. | |
5801 | */ | |
5802 | if (IS_ERR(dir_inode)) { | |
5803 | ret = PTR_ERR(dir_inode); | |
5804 | goto out; | |
5805 | } | |
18aa0922 | 5806 | |
657ed1aa FM |
5807 | if (ctx) |
5808 | ctx->log_new_dentries = false; | |
a59108a7 | 5809 | ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode), |
18aa0922 | 5810 | LOG_INODE_ALL, 0, LLONG_MAX, ctx); |
2be63d5c | 5811 | if (!ret && |
ab1717b2 | 5812 | btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode))) |
2be63d5c | 5813 | ret = 1; |
657ed1aa FM |
5814 | if (!ret && ctx && ctx->log_new_dentries) |
5815 | ret = log_new_dir_dentries(trans, root, | |
f85b7379 | 5816 | BTRFS_I(dir_inode), ctx); |
18aa0922 FM |
5817 | iput(dir_inode); |
5818 | if (ret) | |
5819 | goto out; | |
5820 | } | |
5821 | path->slots[0]++; | |
5822 | } | |
5823 | ret = 0; | |
5824 | out: | |
5825 | btrfs_free_path(path); | |
5826 | return ret; | |
5827 | } | |
5828 | ||
b8aa330d FM |
5829 | static int log_new_ancestors(struct btrfs_trans_handle *trans, |
5830 | struct btrfs_root *root, | |
5831 | struct btrfs_path *path, | |
5832 | struct btrfs_log_ctx *ctx) | |
5833 | { | |
5834 | struct btrfs_key found_key; | |
5835 | ||
5836 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
5837 | ||
5838 | while (true) { | |
5839 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5840 | const u64 last_committed = fs_info->last_trans_committed; | |
5841 | struct extent_buffer *leaf = path->nodes[0]; | |
5842 | int slot = path->slots[0]; | |
5843 | struct btrfs_key search_key; | |
5844 | struct inode *inode; | |
5845 | int ret = 0; | |
5846 | ||
5847 | btrfs_release_path(path); | |
5848 | ||
5849 | search_key.objectid = found_key.offset; | |
5850 | search_key.type = BTRFS_INODE_ITEM_KEY; | |
5851 | search_key.offset = 0; | |
5852 | inode = btrfs_iget(fs_info->sb, &search_key, root, NULL); | |
5853 | if (IS_ERR(inode)) | |
5854 | return PTR_ERR(inode); | |
5855 | ||
5856 | if (BTRFS_I(inode)->generation > last_committed) | |
5857 | ret = btrfs_log_inode(trans, root, BTRFS_I(inode), | |
5858 | LOG_INODE_EXISTS, | |
5859 | 0, LLONG_MAX, ctx); | |
5860 | iput(inode); | |
5861 | if (ret) | |
5862 | return ret; | |
5863 | ||
5864 | if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID) | |
5865 | break; | |
5866 | ||
5867 | search_key.type = BTRFS_INODE_REF_KEY; | |
5868 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
5869 | if (ret < 0) | |
5870 | return ret; | |
5871 | ||
5872 | leaf = path->nodes[0]; | |
5873 | slot = path->slots[0]; | |
5874 | if (slot >= btrfs_header_nritems(leaf)) { | |
5875 | ret = btrfs_next_leaf(root, path); | |
5876 | if (ret < 0) | |
5877 | return ret; | |
5878 | else if (ret > 0) | |
5879 | return -ENOENT; | |
5880 | leaf = path->nodes[0]; | |
5881 | slot = path->slots[0]; | |
5882 | } | |
5883 | ||
5884 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5885 | if (found_key.objectid != search_key.objectid || | |
5886 | found_key.type != BTRFS_INODE_REF_KEY) | |
5887 | return -ENOENT; | |
5888 | } | |
5889 | return 0; | |
5890 | } | |
5891 | ||
5892 | static int log_new_ancestors_fast(struct btrfs_trans_handle *trans, | |
5893 | struct btrfs_inode *inode, | |
5894 | struct dentry *parent, | |
5895 | struct btrfs_log_ctx *ctx) | |
5896 | { | |
5897 | struct btrfs_root *root = inode->root; | |
5898 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5899 | struct dentry *old_parent = NULL; | |
5900 | struct super_block *sb = inode->vfs_inode.i_sb; | |
5901 | int ret = 0; | |
5902 | ||
5903 | while (true) { | |
5904 | if (!parent || d_really_is_negative(parent) || | |
5905 | sb != parent->d_sb) | |
5906 | break; | |
5907 | ||
5908 | inode = BTRFS_I(d_inode(parent)); | |
5909 | if (root != inode->root) | |
5910 | break; | |
5911 | ||
5912 | if (inode->generation > fs_info->last_trans_committed) { | |
5913 | ret = btrfs_log_inode(trans, root, inode, | |
5914 | LOG_INODE_EXISTS, 0, LLONG_MAX, ctx); | |
5915 | if (ret) | |
5916 | break; | |
5917 | } | |
5918 | if (IS_ROOT(parent)) | |
5919 | break; | |
5920 | ||
5921 | parent = dget_parent(parent); | |
5922 | dput(old_parent); | |
5923 | old_parent = parent; | |
5924 | } | |
5925 | dput(old_parent); | |
5926 | ||
5927 | return ret; | |
5928 | } | |
5929 | ||
5930 | static int log_all_new_ancestors(struct btrfs_trans_handle *trans, | |
5931 | struct btrfs_inode *inode, | |
5932 | struct dentry *parent, | |
5933 | struct btrfs_log_ctx *ctx) | |
5934 | { | |
5935 | struct btrfs_root *root = inode->root; | |
5936 | const u64 ino = btrfs_ino(inode); | |
5937 | struct btrfs_path *path; | |
5938 | struct btrfs_key search_key; | |
5939 | int ret; | |
5940 | ||
5941 | /* | |
5942 | * For a single hard link case, go through a fast path that does not | |
5943 | * need to iterate the fs/subvolume tree. | |
5944 | */ | |
5945 | if (inode->vfs_inode.i_nlink < 2) | |
5946 | return log_new_ancestors_fast(trans, inode, parent, ctx); | |
5947 | ||
5948 | path = btrfs_alloc_path(); | |
5949 | if (!path) | |
5950 | return -ENOMEM; | |
5951 | ||
5952 | search_key.objectid = ino; | |
5953 | search_key.type = BTRFS_INODE_REF_KEY; | |
5954 | search_key.offset = 0; | |
5955 | again: | |
5956 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
5957 | if (ret < 0) | |
5958 | goto out; | |
5959 | if (ret == 0) | |
5960 | path->slots[0]++; | |
5961 | ||
5962 | while (true) { | |
5963 | struct extent_buffer *leaf = path->nodes[0]; | |
5964 | int slot = path->slots[0]; | |
5965 | struct btrfs_key found_key; | |
5966 | ||
5967 | if (slot >= btrfs_header_nritems(leaf)) { | |
5968 | ret = btrfs_next_leaf(root, path); | |
5969 | if (ret < 0) | |
5970 | goto out; | |
5971 | else if (ret > 0) | |
5972 | break; | |
5973 | continue; | |
5974 | } | |
5975 | ||
5976 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5977 | if (found_key.objectid != ino || | |
5978 | found_key.type > BTRFS_INODE_EXTREF_KEY) | |
5979 | break; | |
5980 | ||
5981 | /* | |
5982 | * Don't deal with extended references because they are rare | |
5983 | * cases and too complex to deal with (we would need to keep | |
5984 | * track of which subitem we are processing for each item in | |
5985 | * this loop, etc). So just return some error to fallback to | |
5986 | * a transaction commit. | |
5987 | */ | |
5988 | if (found_key.type == BTRFS_INODE_EXTREF_KEY) { | |
5989 | ret = -EMLINK; | |
5990 | goto out; | |
5991 | } | |
5992 | ||
5993 | /* | |
5994 | * Logging ancestors needs to do more searches on the fs/subvol | |
5995 | * tree, so it releases the path as needed to avoid deadlocks. | |
5996 | * Keep track of the last inode ref key and resume from that key | |
5997 | * after logging all new ancestors for the current hard link. | |
5998 | */ | |
5999 | memcpy(&search_key, &found_key, sizeof(search_key)); | |
6000 | ||
6001 | ret = log_new_ancestors(trans, root, path, ctx); | |
6002 | if (ret) | |
6003 | goto out; | |
6004 | btrfs_release_path(path); | |
6005 | goto again; | |
6006 | } | |
6007 | ret = 0; | |
6008 | out: | |
6009 | btrfs_free_path(path); | |
6010 | return ret; | |
6011 | } | |
6012 | ||
e02119d5 CM |
6013 | /* |
6014 | * helper function around btrfs_log_inode to make sure newly created | |
6015 | * parent directories also end up in the log. A minimal inode and backref | |
6016 | * only logging is done of any parent directories that are older than | |
6017 | * the last committed transaction | |
6018 | */ | |
48a3b636 | 6019 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
19df27a9 | 6020 | struct btrfs_inode *inode, |
49dae1bc FM |
6021 | struct dentry *parent, |
6022 | const loff_t start, | |
6023 | const loff_t end, | |
41a1eada | 6024 | int inode_only, |
8b050d35 | 6025 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6026 | { |
f882274b | 6027 | struct btrfs_root *root = inode->root; |
0b246afa | 6028 | struct btrfs_fs_info *fs_info = root->fs_info; |
e02119d5 | 6029 | struct super_block *sb; |
12fcfd22 | 6030 | int ret = 0; |
0b246afa | 6031 | u64 last_committed = fs_info->last_trans_committed; |
2f2ff0ee | 6032 | bool log_dentries = false; |
12fcfd22 | 6033 | |
19df27a9 | 6034 | sb = inode->vfs_inode.i_sb; |
12fcfd22 | 6035 | |
0b246afa | 6036 | if (btrfs_test_opt(fs_info, NOTREELOG)) { |
3a5e1404 SW |
6037 | ret = 1; |
6038 | goto end_no_trans; | |
6039 | } | |
6040 | ||
995946dd MX |
6041 | /* |
6042 | * The prev transaction commit doesn't complete, we need do | |
6043 | * full commit by ourselves. | |
6044 | */ | |
0b246afa JM |
6045 | if (fs_info->last_trans_log_full_commit > |
6046 | fs_info->last_trans_committed) { | |
12fcfd22 CM |
6047 | ret = 1; |
6048 | goto end_no_trans; | |
6049 | } | |
6050 | ||
f882274b | 6051 | if (btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
6052 | ret = 1; |
6053 | goto end_no_trans; | |
6054 | } | |
6055 | ||
19df27a9 NB |
6056 | ret = check_parent_dirs_for_sync(trans, inode, parent, sb, |
6057 | last_committed); | |
12fcfd22 CM |
6058 | if (ret) |
6059 | goto end_no_trans; | |
e02119d5 | 6060 | |
f2d72f42 FM |
6061 | /* |
6062 | * Skip already logged inodes or inodes corresponding to tmpfiles | |
6063 | * (since logging them is pointless, a link count of 0 means they | |
6064 | * will never be accessible). | |
6065 | */ | |
6066 | if (btrfs_inode_in_log(inode, trans->transid) || | |
6067 | inode->vfs_inode.i_nlink == 0) { | |
257c62e1 CM |
6068 | ret = BTRFS_NO_LOG_SYNC; |
6069 | goto end_no_trans; | |
6070 | } | |
6071 | ||
8b050d35 | 6072 | ret = start_log_trans(trans, root, ctx); |
4a500fd1 | 6073 | if (ret) |
e87ac136 | 6074 | goto end_no_trans; |
e02119d5 | 6075 | |
19df27a9 | 6076 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); |
4a500fd1 YZ |
6077 | if (ret) |
6078 | goto end_trans; | |
12fcfd22 | 6079 | |
af4176b4 CM |
6080 | /* |
6081 | * for regular files, if its inode is already on disk, we don't | |
6082 | * have to worry about the parents at all. This is because | |
6083 | * we can use the last_unlink_trans field to record renames | |
6084 | * and other fun in this file. | |
6085 | */ | |
19df27a9 NB |
6086 | if (S_ISREG(inode->vfs_inode.i_mode) && |
6087 | inode->generation <= last_committed && | |
6088 | inode->last_unlink_trans <= last_committed) { | |
4a500fd1 YZ |
6089 | ret = 0; |
6090 | goto end_trans; | |
6091 | } | |
af4176b4 | 6092 | |
19df27a9 | 6093 | if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries) |
2f2ff0ee FM |
6094 | log_dentries = true; |
6095 | ||
18aa0922 | 6096 | /* |
01327610 | 6097 | * On unlink we must make sure all our current and old parent directory |
18aa0922 FM |
6098 | * inodes are fully logged. This is to prevent leaving dangling |
6099 | * directory index entries in directories that were our parents but are | |
6100 | * not anymore. Not doing this results in old parent directory being | |
6101 | * impossible to delete after log replay (rmdir will always fail with | |
6102 | * error -ENOTEMPTY). | |
6103 | * | |
6104 | * Example 1: | |
6105 | * | |
6106 | * mkdir testdir | |
6107 | * touch testdir/foo | |
6108 | * ln testdir/foo testdir/bar | |
6109 | * sync | |
6110 | * unlink testdir/bar | |
6111 | * xfs_io -c fsync testdir/foo | |
6112 | * <power failure> | |
6113 | * mount fs, triggers log replay | |
6114 | * | |
6115 | * If we don't log the parent directory (testdir), after log replay the | |
6116 | * directory still has an entry pointing to the file inode using the bar | |
6117 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
6118 | * the file inode has a link count of 1. | |
6119 | * | |
6120 | * Example 2: | |
6121 | * | |
6122 | * mkdir testdir | |
6123 | * touch foo | |
6124 | * ln foo testdir/foo2 | |
6125 | * ln foo testdir/foo3 | |
6126 | * sync | |
6127 | * unlink testdir/foo3 | |
6128 | * xfs_io -c fsync foo | |
6129 | * <power failure> | |
6130 | * mount fs, triggers log replay | |
6131 | * | |
6132 | * Similar as the first example, after log replay the parent directory | |
6133 | * testdir still has an entry pointing to the inode file with name foo3 | |
6134 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
6135 | * and has a link count of 2. | |
6136 | */ | |
19df27a9 | 6137 | if (inode->last_unlink_trans > last_committed) { |
b8aa330d | 6138 | ret = btrfs_log_all_parents(trans, inode, ctx); |
18aa0922 FM |
6139 | if (ret) |
6140 | goto end_trans; | |
6141 | } | |
6142 | ||
b8aa330d FM |
6143 | ret = log_all_new_ancestors(trans, inode, parent, ctx); |
6144 | if (ret) | |
41bd6067 | 6145 | goto end_trans; |
76dda93c | 6146 | |
2f2ff0ee | 6147 | if (log_dentries) |
b8aa330d | 6148 | ret = log_new_dir_dentries(trans, root, inode, ctx); |
2f2ff0ee FM |
6149 | else |
6150 | ret = 0; | |
4a500fd1 YZ |
6151 | end_trans: |
6152 | if (ret < 0) { | |
90787766 | 6153 | btrfs_set_log_full_commit(trans); |
4a500fd1 YZ |
6154 | ret = 1; |
6155 | } | |
8b050d35 MX |
6156 | |
6157 | if (ret) | |
6158 | btrfs_remove_log_ctx(root, ctx); | |
12fcfd22 CM |
6159 | btrfs_end_log_trans(root); |
6160 | end_no_trans: | |
6161 | return ret; | |
e02119d5 CM |
6162 | } |
6163 | ||
6164 | /* | |
6165 | * it is not safe to log dentry if the chunk root has added new | |
6166 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
6167 | * If this returns 1, you must commit the transaction to safely get your | |
6168 | * data on disk. | |
6169 | */ | |
6170 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
e5b84f7a | 6171 | struct dentry *dentry, |
49dae1bc FM |
6172 | const loff_t start, |
6173 | const loff_t end, | |
8b050d35 | 6174 | struct btrfs_log_ctx *ctx) |
e02119d5 | 6175 | { |
6a912213 JB |
6176 | struct dentry *parent = dget_parent(dentry); |
6177 | int ret; | |
6178 | ||
f882274b NB |
6179 | ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent, |
6180 | start, end, LOG_INODE_ALL, ctx); | |
6a912213 JB |
6181 | dput(parent); |
6182 | ||
6183 | return ret; | |
e02119d5 CM |
6184 | } |
6185 | ||
6186 | /* | |
6187 | * should be called during mount to recover any replay any log trees | |
6188 | * from the FS | |
6189 | */ | |
6190 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
6191 | { | |
6192 | int ret; | |
6193 | struct btrfs_path *path; | |
6194 | struct btrfs_trans_handle *trans; | |
6195 | struct btrfs_key key; | |
6196 | struct btrfs_key found_key; | |
6197 | struct btrfs_key tmp_key; | |
6198 | struct btrfs_root *log; | |
6199 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
6200 | struct walk_control wc = { | |
6201 | .process_func = process_one_buffer, | |
6202 | .stage = 0, | |
6203 | }; | |
6204 | ||
e02119d5 | 6205 | path = btrfs_alloc_path(); |
db5b493a TI |
6206 | if (!path) |
6207 | return -ENOMEM; | |
6208 | ||
afcdd129 | 6209 | set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 6210 | |
4a500fd1 | 6211 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
79787eaa JM |
6212 | if (IS_ERR(trans)) { |
6213 | ret = PTR_ERR(trans); | |
6214 | goto error; | |
6215 | } | |
e02119d5 CM |
6216 | |
6217 | wc.trans = trans; | |
6218 | wc.pin = 1; | |
6219 | ||
db5b493a | 6220 | ret = walk_log_tree(trans, log_root_tree, &wc); |
79787eaa | 6221 | if (ret) { |
5d163e0e JM |
6222 | btrfs_handle_fs_error(fs_info, ret, |
6223 | "Failed to pin buffers while recovering log root tree."); | |
79787eaa JM |
6224 | goto error; |
6225 | } | |
e02119d5 CM |
6226 | |
6227 | again: | |
6228 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
6229 | key.offset = (u64)-1; | |
962a298f | 6230 | key.type = BTRFS_ROOT_ITEM_KEY; |
e02119d5 | 6231 | |
d397712b | 6232 | while (1) { |
e02119d5 | 6233 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
79787eaa JM |
6234 | |
6235 | if (ret < 0) { | |
34d97007 | 6236 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6237 | "Couldn't find tree log root."); |
6238 | goto error; | |
6239 | } | |
e02119d5 CM |
6240 | if (ret > 0) { |
6241 | if (path->slots[0] == 0) | |
6242 | break; | |
6243 | path->slots[0]--; | |
6244 | } | |
6245 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
6246 | path->slots[0]); | |
b3b4aa74 | 6247 | btrfs_release_path(path); |
e02119d5 CM |
6248 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
6249 | break; | |
6250 | ||
cb517eab | 6251 | log = btrfs_read_fs_root(log_root_tree, &found_key); |
79787eaa JM |
6252 | if (IS_ERR(log)) { |
6253 | ret = PTR_ERR(log); | |
34d97007 | 6254 | btrfs_handle_fs_error(fs_info, ret, |
79787eaa JM |
6255 | "Couldn't read tree log root."); |
6256 | goto error; | |
6257 | } | |
e02119d5 CM |
6258 | |
6259 | tmp_key.objectid = found_key.offset; | |
6260 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; | |
6261 | tmp_key.offset = (u64)-1; | |
6262 | ||
6263 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); | |
79787eaa JM |
6264 | if (IS_ERR(wc.replay_dest)) { |
6265 | ret = PTR_ERR(wc.replay_dest); | |
b50c6e25 JB |
6266 | free_extent_buffer(log->node); |
6267 | free_extent_buffer(log->commit_root); | |
6268 | kfree(log); | |
5d163e0e JM |
6269 | btrfs_handle_fs_error(fs_info, ret, |
6270 | "Couldn't read target root for tree log recovery."); | |
79787eaa JM |
6271 | goto error; |
6272 | } | |
e02119d5 | 6273 | |
07d400a6 | 6274 | wc.replay_dest->log_root = log; |
5d4f98a2 | 6275 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 | 6276 | ret = walk_log_tree(trans, log, &wc); |
e02119d5 | 6277 | |
b50c6e25 | 6278 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
e02119d5 CM |
6279 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
6280 | path); | |
e02119d5 CM |
6281 | } |
6282 | ||
900c9981 LB |
6283 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { |
6284 | struct btrfs_root *root = wc.replay_dest; | |
6285 | ||
6286 | btrfs_release_path(path); | |
6287 | ||
6288 | /* | |
6289 | * We have just replayed everything, and the highest | |
6290 | * objectid of fs roots probably has changed in case | |
6291 | * some inode_item's got replayed. | |
6292 | * | |
6293 | * root->objectid_mutex is not acquired as log replay | |
6294 | * could only happen during mount. | |
6295 | */ | |
6296 | ret = btrfs_find_highest_objectid(root, | |
6297 | &root->highest_objectid); | |
6298 | } | |
6299 | ||
e02119d5 | 6300 | key.offset = found_key.offset - 1; |
07d400a6 | 6301 | wc.replay_dest->log_root = NULL; |
e02119d5 | 6302 | free_extent_buffer(log->node); |
b263c2c8 | 6303 | free_extent_buffer(log->commit_root); |
e02119d5 CM |
6304 | kfree(log); |
6305 | ||
b50c6e25 JB |
6306 | if (ret) |
6307 | goto error; | |
6308 | ||
e02119d5 CM |
6309 | if (found_key.offset == 0) |
6310 | break; | |
6311 | } | |
b3b4aa74 | 6312 | btrfs_release_path(path); |
e02119d5 CM |
6313 | |
6314 | /* step one is to pin it all, step two is to replay just inodes */ | |
6315 | if (wc.pin) { | |
6316 | wc.pin = 0; | |
6317 | wc.process_func = replay_one_buffer; | |
6318 | wc.stage = LOG_WALK_REPLAY_INODES; | |
6319 | goto again; | |
6320 | } | |
6321 | /* step three is to replay everything */ | |
6322 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
6323 | wc.stage++; | |
6324 | goto again; | |
6325 | } | |
6326 | ||
6327 | btrfs_free_path(path); | |
6328 | ||
abefa55a | 6329 | /* step 4: commit the transaction, which also unpins the blocks */ |
3a45bb20 | 6330 | ret = btrfs_commit_transaction(trans); |
abefa55a JB |
6331 | if (ret) |
6332 | return ret; | |
6333 | ||
e02119d5 CM |
6334 | free_extent_buffer(log_root_tree->node); |
6335 | log_root_tree->log_root = NULL; | |
afcdd129 | 6336 | clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags); |
e02119d5 | 6337 | kfree(log_root_tree); |
79787eaa | 6338 | |
abefa55a | 6339 | return 0; |
79787eaa | 6340 | error: |
b50c6e25 | 6341 | if (wc.trans) |
3a45bb20 | 6342 | btrfs_end_transaction(wc.trans); |
79787eaa JM |
6343 | btrfs_free_path(path); |
6344 | return ret; | |
e02119d5 | 6345 | } |
12fcfd22 CM |
6346 | |
6347 | /* | |
6348 | * there are some corner cases where we want to force a full | |
6349 | * commit instead of allowing a directory to be logged. | |
6350 | * | |
6351 | * They revolve around files there were unlinked from the directory, and | |
6352 | * this function updates the parent directory so that a full commit is | |
6353 | * properly done if it is fsync'd later after the unlinks are done. | |
2be63d5c FM |
6354 | * |
6355 | * Must be called before the unlink operations (updates to the subvolume tree, | |
6356 | * inodes, etc) are done. | |
12fcfd22 CM |
6357 | */ |
6358 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
4176bdbf | 6359 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
12fcfd22 CM |
6360 | int for_rename) |
6361 | { | |
af4176b4 CM |
6362 | /* |
6363 | * when we're logging a file, if it hasn't been renamed | |
6364 | * or unlinked, and its inode is fully committed on disk, | |
6365 | * we don't have to worry about walking up the directory chain | |
6366 | * to log its parents. | |
6367 | * | |
6368 | * So, we use the last_unlink_trans field to put this transid | |
6369 | * into the file. When the file is logged we check it and | |
6370 | * don't log the parents if the file is fully on disk. | |
6371 | */ | |
4176bdbf NB |
6372 | mutex_lock(&inode->log_mutex); |
6373 | inode->last_unlink_trans = trans->transid; | |
6374 | mutex_unlock(&inode->log_mutex); | |
af4176b4 | 6375 | |
12fcfd22 CM |
6376 | /* |
6377 | * if this directory was already logged any new | |
6378 | * names for this file/dir will get recorded | |
6379 | */ | |
4176bdbf | 6380 | if (dir->logged_trans == trans->transid) |
12fcfd22 CM |
6381 | return; |
6382 | ||
6383 | /* | |
6384 | * if the inode we're about to unlink was logged, | |
6385 | * the log will be properly updated for any new names | |
6386 | */ | |
4176bdbf | 6387 | if (inode->logged_trans == trans->transid) |
12fcfd22 CM |
6388 | return; |
6389 | ||
6390 | /* | |
6391 | * when renaming files across directories, if the directory | |
6392 | * there we're unlinking from gets fsync'd later on, there's | |
6393 | * no way to find the destination directory later and fsync it | |
6394 | * properly. So, we have to be conservative and force commits | |
6395 | * so the new name gets discovered. | |
6396 | */ | |
6397 | if (for_rename) | |
6398 | goto record; | |
6399 | ||
6400 | /* we can safely do the unlink without any special recording */ | |
6401 | return; | |
6402 | ||
6403 | record: | |
4176bdbf NB |
6404 | mutex_lock(&dir->log_mutex); |
6405 | dir->last_unlink_trans = trans->transid; | |
6406 | mutex_unlock(&dir->log_mutex); | |
1ec9a1ae FM |
6407 | } |
6408 | ||
6409 | /* | |
6410 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
6411 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
6412 | * that after replaying the log tree of the parent directory's root we will not | |
6413 | * see the snapshot anymore and at log replay time we will not see any log tree | |
6414 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
6415 | * it after replaying the log tree of the parent directory (which would replay | |
6416 | * the snapshot delete operation). | |
2be63d5c FM |
6417 | * |
6418 | * Must be called before the actual snapshot destroy operation (updates to the | |
6419 | * parent root and tree of tree roots trees, etc) are done. | |
1ec9a1ae FM |
6420 | */ |
6421 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
43663557 | 6422 | struct btrfs_inode *dir) |
1ec9a1ae | 6423 | { |
43663557 NB |
6424 | mutex_lock(&dir->log_mutex); |
6425 | dir->last_unlink_trans = trans->transid; | |
6426 | mutex_unlock(&dir->log_mutex); | |
12fcfd22 CM |
6427 | } |
6428 | ||
6429 | /* | |
6430 | * Call this after adding a new name for a file and it will properly | |
6431 | * update the log to reflect the new name. | |
6432 | * | |
d4682ba0 FM |
6433 | * @ctx can not be NULL when @sync_log is false, and should be NULL when it's |
6434 | * true (because it's not used). | |
6435 | * | |
6436 | * Return value depends on whether @sync_log is true or false. | |
6437 | * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be | |
6438 | * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT | |
6439 | * otherwise. | |
6440 | * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to | |
6441 | * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log, | |
6442 | * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be | |
6443 | * committed (without attempting to sync the log). | |
12fcfd22 CM |
6444 | */ |
6445 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
9ca5fbfb | 6446 | struct btrfs_inode *inode, struct btrfs_inode *old_dir, |
d4682ba0 FM |
6447 | struct dentry *parent, |
6448 | bool sync_log, struct btrfs_log_ctx *ctx) | |
12fcfd22 | 6449 | { |
3ffbd68c | 6450 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d4682ba0 | 6451 | int ret; |
12fcfd22 | 6452 | |
af4176b4 CM |
6453 | /* |
6454 | * this will force the logging code to walk the dentry chain | |
6455 | * up for the file | |
6456 | */ | |
9a6509c4 | 6457 | if (!S_ISDIR(inode->vfs_inode.i_mode)) |
9ca5fbfb | 6458 | inode->last_unlink_trans = trans->transid; |
af4176b4 | 6459 | |
12fcfd22 CM |
6460 | /* |
6461 | * if this inode hasn't been logged and directory we're renaming it | |
6462 | * from hasn't been logged, we don't need to log it | |
6463 | */ | |
9ca5fbfb NB |
6464 | if (inode->logged_trans <= fs_info->last_trans_committed && |
6465 | (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed)) | |
d4682ba0 FM |
6466 | return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT : |
6467 | BTRFS_DONT_NEED_LOG_SYNC; | |
6468 | ||
6469 | if (sync_log) { | |
6470 | struct btrfs_log_ctx ctx2; | |
6471 | ||
6472 | btrfs_init_log_ctx(&ctx2, &inode->vfs_inode); | |
6473 | ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX, | |
6474 | LOG_INODE_EXISTS, &ctx2); | |
6475 | if (ret == BTRFS_NO_LOG_SYNC) | |
6476 | return BTRFS_DONT_NEED_TRANS_COMMIT; | |
6477 | else if (ret) | |
6478 | return BTRFS_NEED_TRANS_COMMIT; | |
6479 | ||
6480 | ret = btrfs_sync_log(trans, inode->root, &ctx2); | |
6481 | if (ret) | |
6482 | return BTRFS_NEED_TRANS_COMMIT; | |
6483 | return BTRFS_DONT_NEED_TRANS_COMMIT; | |
6484 | } | |
6485 | ||
6486 | ASSERT(ctx); | |
6487 | ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX, | |
6488 | LOG_INODE_EXISTS, ctx); | |
6489 | if (ret == BTRFS_NO_LOG_SYNC) | |
6490 | return BTRFS_DONT_NEED_LOG_SYNC; | |
6491 | else if (ret) | |
6492 | return BTRFS_NEED_TRANS_COMMIT; | |
12fcfd22 | 6493 | |
d4682ba0 | 6494 | return BTRFS_NEED_LOG_SYNC; |
12fcfd22 CM |
6495 | } |
6496 |