]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) 2008 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
19 | #include <linux/sched.h> | |
20 | #include <linux/slab.h> | |
21 | #include <linux/blkdev.h> | |
22 | #include <linux/list_sort.h> | |
23 | #include "tree-log.h" | |
24 | #include "disk-io.h" | |
25 | #include "locking.h" | |
26 | #include "print-tree.h" | |
27 | #include "backref.h" | |
28 | #include "hash.h" | |
29 | #include "compression.h" | |
30 | #include "qgroup.h" | |
31 | ||
32 | /* magic values for the inode_only field in btrfs_log_inode: | |
33 | * | |
34 | * LOG_INODE_ALL means to log everything | |
35 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
36 | * during log replay | |
37 | */ | |
38 | #define LOG_INODE_ALL 0 | |
39 | #define LOG_INODE_EXISTS 1 | |
40 | ||
41 | /* | |
42 | * directory trouble cases | |
43 | * | |
44 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
45 | * log, we must force a full commit before doing an fsync of the directory | |
46 | * where the unlink was done. | |
47 | * ---> record transid of last unlink/rename per directory | |
48 | * | |
49 | * mkdir foo/some_dir | |
50 | * normal commit | |
51 | * rename foo/some_dir foo2/some_dir | |
52 | * mkdir foo/some_dir | |
53 | * fsync foo/some_dir/some_file | |
54 | * | |
55 | * The fsync above will unlink the original some_dir without recording | |
56 | * it in its new location (foo2). After a crash, some_dir will be gone | |
57 | * unless the fsync of some_file forces a full commit | |
58 | * | |
59 | * 2) we must log any new names for any file or dir that is in the fsync | |
60 | * log. ---> check inode while renaming/linking. | |
61 | * | |
62 | * 2a) we must log any new names for any file or dir during rename | |
63 | * when the directory they are being removed from was logged. | |
64 | * ---> check inode and old parent dir during rename | |
65 | * | |
66 | * 2a is actually the more important variant. With the extra logging | |
67 | * a crash might unlink the old name without recreating the new one | |
68 | * | |
69 | * 3) after a crash, we must go through any directories with a link count | |
70 | * of zero and redo the rm -rf | |
71 | * | |
72 | * mkdir f1/foo | |
73 | * normal commit | |
74 | * rm -rf f1/foo | |
75 | * fsync(f1) | |
76 | * | |
77 | * The directory f1 was fully removed from the FS, but fsync was never | |
78 | * called on f1, only its parent dir. After a crash the rm -rf must | |
79 | * be replayed. This must be able to recurse down the entire | |
80 | * directory tree. The inode link count fixup code takes care of the | |
81 | * ugly details. | |
82 | */ | |
83 | ||
84 | /* | |
85 | * stages for the tree walking. The first | |
86 | * stage (0) is to only pin down the blocks we find | |
87 | * the second stage (1) is to make sure that all the inodes | |
88 | * we find in the log are created in the subvolume. | |
89 | * | |
90 | * The last stage is to deal with directories and links and extents | |
91 | * and all the other fun semantics | |
92 | */ | |
93 | #define LOG_WALK_PIN_ONLY 0 | |
94 | #define LOG_WALK_REPLAY_INODES 1 | |
95 | #define LOG_WALK_REPLAY_DIR_INDEX 2 | |
96 | #define LOG_WALK_REPLAY_ALL 3 | |
97 | ||
98 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, | |
99 | struct btrfs_root *root, struct inode *inode, | |
100 | int inode_only, | |
101 | const loff_t start, | |
102 | const loff_t end, | |
103 | struct btrfs_log_ctx *ctx); | |
104 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
105 | struct btrfs_root *root, | |
106 | struct btrfs_path *path, u64 objectid); | |
107 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
108 | struct btrfs_root *root, | |
109 | struct btrfs_root *log, | |
110 | struct btrfs_path *path, | |
111 | u64 dirid, int del_all); | |
112 | ||
113 | /* | |
114 | * tree logging is a special write ahead log used to make sure that | |
115 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
116 | * | |
117 | * Full tree commits are expensive because they require commonly | |
118 | * modified blocks to be recowed, creating many dirty pages in the | |
119 | * extent tree an 4x-6x higher write load than ext3. | |
120 | * | |
121 | * Instead of doing a tree commit on every fsync, we use the | |
122 | * key ranges and transaction ids to find items for a given file or directory | |
123 | * that have changed in this transaction. Those items are copied into | |
124 | * a special tree (one per subvolume root), that tree is written to disk | |
125 | * and then the fsync is considered complete. | |
126 | * | |
127 | * After a crash, items are copied out of the log-tree back into the | |
128 | * subvolume tree. Any file data extents found are recorded in the extent | |
129 | * allocation tree, and the log-tree freed. | |
130 | * | |
131 | * The log tree is read three times, once to pin down all the extents it is | |
132 | * using in ram and once, once to create all the inodes logged in the tree | |
133 | * and once to do all the other items. | |
134 | */ | |
135 | ||
136 | /* | |
137 | * start a sub transaction and setup the log tree | |
138 | * this increments the log tree writer count to make the people | |
139 | * syncing the tree wait for us to finish | |
140 | */ | |
141 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
142 | struct btrfs_root *root, | |
143 | struct btrfs_log_ctx *ctx) | |
144 | { | |
145 | int ret = 0; | |
146 | ||
147 | mutex_lock(&root->log_mutex); | |
148 | ||
149 | if (root->log_root) { | |
150 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { | |
151 | ret = -EAGAIN; | |
152 | goto out; | |
153 | } | |
154 | ||
155 | if (!root->log_start_pid) { | |
156 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); | |
157 | root->log_start_pid = current->pid; | |
158 | } else if (root->log_start_pid != current->pid) { | |
159 | set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); | |
160 | } | |
161 | } else { | |
162 | mutex_lock(&root->fs_info->tree_log_mutex); | |
163 | if (!root->fs_info->log_root_tree) | |
164 | ret = btrfs_init_log_root_tree(trans, root->fs_info); | |
165 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
166 | if (ret) | |
167 | goto out; | |
168 | ||
169 | ret = btrfs_add_log_tree(trans, root); | |
170 | if (ret) | |
171 | goto out; | |
172 | ||
173 | clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state); | |
174 | root->log_start_pid = current->pid; | |
175 | } | |
176 | ||
177 | atomic_inc(&root->log_batch); | |
178 | atomic_inc(&root->log_writers); | |
179 | if (ctx) { | |
180 | int index = root->log_transid % 2; | |
181 | list_add_tail(&ctx->list, &root->log_ctxs[index]); | |
182 | ctx->log_transid = root->log_transid; | |
183 | } | |
184 | ||
185 | out: | |
186 | mutex_unlock(&root->log_mutex); | |
187 | return ret; | |
188 | } | |
189 | ||
190 | /* | |
191 | * returns 0 if there was a log transaction running and we were able | |
192 | * to join, or returns -ENOENT if there were not transactions | |
193 | * in progress | |
194 | */ | |
195 | static int join_running_log_trans(struct btrfs_root *root) | |
196 | { | |
197 | int ret = -ENOENT; | |
198 | ||
199 | smp_mb(); | |
200 | if (!root->log_root) | |
201 | return -ENOENT; | |
202 | ||
203 | mutex_lock(&root->log_mutex); | |
204 | if (root->log_root) { | |
205 | ret = 0; | |
206 | atomic_inc(&root->log_writers); | |
207 | } | |
208 | mutex_unlock(&root->log_mutex); | |
209 | return ret; | |
210 | } | |
211 | ||
212 | /* | |
213 | * This either makes the current running log transaction wait | |
214 | * until you call btrfs_end_log_trans() or it makes any future | |
215 | * log transactions wait until you call btrfs_end_log_trans() | |
216 | */ | |
217 | int btrfs_pin_log_trans(struct btrfs_root *root) | |
218 | { | |
219 | int ret = -ENOENT; | |
220 | ||
221 | mutex_lock(&root->log_mutex); | |
222 | atomic_inc(&root->log_writers); | |
223 | mutex_unlock(&root->log_mutex); | |
224 | return ret; | |
225 | } | |
226 | ||
227 | /* | |
228 | * indicate we're done making changes to the log tree | |
229 | * and wake up anyone waiting to do a sync | |
230 | */ | |
231 | void btrfs_end_log_trans(struct btrfs_root *root) | |
232 | { | |
233 | if (atomic_dec_and_test(&root->log_writers)) { | |
234 | /* | |
235 | * Implicit memory barrier after atomic_dec_and_test | |
236 | */ | |
237 | if (waitqueue_active(&root->log_writer_wait)) | |
238 | wake_up(&root->log_writer_wait); | |
239 | } | |
240 | } | |
241 | ||
242 | ||
243 | /* | |
244 | * the walk control struct is used to pass state down the chain when | |
245 | * processing the log tree. The stage field tells us which part | |
246 | * of the log tree processing we are currently doing. The others | |
247 | * are state fields used for that specific part | |
248 | */ | |
249 | struct walk_control { | |
250 | /* should we free the extent on disk when done? This is used | |
251 | * at transaction commit time while freeing a log tree | |
252 | */ | |
253 | int free; | |
254 | ||
255 | /* should we write out the extent buffer? This is used | |
256 | * while flushing the log tree to disk during a sync | |
257 | */ | |
258 | int write; | |
259 | ||
260 | /* should we wait for the extent buffer io to finish? Also used | |
261 | * while flushing the log tree to disk for a sync | |
262 | */ | |
263 | int wait; | |
264 | ||
265 | /* pin only walk, we record which extents on disk belong to the | |
266 | * log trees | |
267 | */ | |
268 | int pin; | |
269 | ||
270 | /* what stage of the replay code we're currently in */ | |
271 | int stage; | |
272 | ||
273 | /* the root we are currently replaying */ | |
274 | struct btrfs_root *replay_dest; | |
275 | ||
276 | /* the trans handle for the current replay */ | |
277 | struct btrfs_trans_handle *trans; | |
278 | ||
279 | /* the function that gets used to process blocks we find in the | |
280 | * tree. Note the extent_buffer might not be up to date when it is | |
281 | * passed in, and it must be checked or read if you need the data | |
282 | * inside it | |
283 | */ | |
284 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
285 | struct walk_control *wc, u64 gen); | |
286 | }; | |
287 | ||
288 | /* | |
289 | * process_func used to pin down extents, write them or wait on them | |
290 | */ | |
291 | static int process_one_buffer(struct btrfs_root *log, | |
292 | struct extent_buffer *eb, | |
293 | struct walk_control *wc, u64 gen) | |
294 | { | |
295 | int ret = 0; | |
296 | ||
297 | /* | |
298 | * If this fs is mixed then we need to be able to process the leaves to | |
299 | * pin down any logged extents, so we have to read the block. | |
300 | */ | |
301 | if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) { | |
302 | ret = btrfs_read_buffer(eb, gen); | |
303 | if (ret) | |
304 | return ret; | |
305 | } | |
306 | ||
307 | if (wc->pin) | |
308 | ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root, | |
309 | eb->start, eb->len); | |
310 | ||
311 | if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) { | |
312 | if (wc->pin && btrfs_header_level(eb) == 0) | |
313 | ret = btrfs_exclude_logged_extents(log, eb); | |
314 | if (wc->write) | |
315 | btrfs_write_tree_block(eb); | |
316 | if (wc->wait) | |
317 | btrfs_wait_tree_block_writeback(eb); | |
318 | } | |
319 | return ret; | |
320 | } | |
321 | ||
322 | /* | |
323 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
324 | * to the src data we are copying out. | |
325 | * | |
326 | * root is the tree we are copying into, and path is a scratch | |
327 | * path for use in this function (it should be released on entry and | |
328 | * will be released on exit). | |
329 | * | |
330 | * If the key is already in the destination tree the existing item is | |
331 | * overwritten. If the existing item isn't big enough, it is extended. | |
332 | * If it is too large, it is truncated. | |
333 | * | |
334 | * If the key isn't in the destination yet, a new item is inserted. | |
335 | */ | |
336 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
337 | struct btrfs_root *root, | |
338 | struct btrfs_path *path, | |
339 | struct extent_buffer *eb, int slot, | |
340 | struct btrfs_key *key) | |
341 | { | |
342 | int ret; | |
343 | u32 item_size; | |
344 | u64 saved_i_size = 0; | |
345 | int save_old_i_size = 0; | |
346 | unsigned long src_ptr; | |
347 | unsigned long dst_ptr; | |
348 | int overwrite_root = 0; | |
349 | bool inode_item = key->type == BTRFS_INODE_ITEM_KEY; | |
350 | ||
351 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
352 | overwrite_root = 1; | |
353 | ||
354 | item_size = btrfs_item_size_nr(eb, slot); | |
355 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
356 | ||
357 | /* look for the key in the destination tree */ | |
358 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
359 | if (ret < 0) | |
360 | return ret; | |
361 | ||
362 | if (ret == 0) { | |
363 | char *src_copy; | |
364 | char *dst_copy; | |
365 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
366 | path->slots[0]); | |
367 | if (dst_size != item_size) | |
368 | goto insert; | |
369 | ||
370 | if (item_size == 0) { | |
371 | btrfs_release_path(path); | |
372 | return 0; | |
373 | } | |
374 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
375 | src_copy = kmalloc(item_size, GFP_NOFS); | |
376 | if (!dst_copy || !src_copy) { | |
377 | btrfs_release_path(path); | |
378 | kfree(dst_copy); | |
379 | kfree(src_copy); | |
380 | return -ENOMEM; | |
381 | } | |
382 | ||
383 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
384 | ||
385 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
386 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
387 | item_size); | |
388 | ret = memcmp(dst_copy, src_copy, item_size); | |
389 | ||
390 | kfree(dst_copy); | |
391 | kfree(src_copy); | |
392 | /* | |
393 | * they have the same contents, just return, this saves | |
394 | * us from cowing blocks in the destination tree and doing | |
395 | * extra writes that may not have been done by a previous | |
396 | * sync | |
397 | */ | |
398 | if (ret == 0) { | |
399 | btrfs_release_path(path); | |
400 | return 0; | |
401 | } | |
402 | ||
403 | /* | |
404 | * We need to load the old nbytes into the inode so when we | |
405 | * replay the extents we've logged we get the right nbytes. | |
406 | */ | |
407 | if (inode_item) { | |
408 | struct btrfs_inode_item *item; | |
409 | u64 nbytes; | |
410 | u32 mode; | |
411 | ||
412 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
413 | struct btrfs_inode_item); | |
414 | nbytes = btrfs_inode_nbytes(path->nodes[0], item); | |
415 | item = btrfs_item_ptr(eb, slot, | |
416 | struct btrfs_inode_item); | |
417 | btrfs_set_inode_nbytes(eb, item, nbytes); | |
418 | ||
419 | /* | |
420 | * If this is a directory we need to reset the i_size to | |
421 | * 0 so that we can set it up properly when replaying | |
422 | * the rest of the items in this log. | |
423 | */ | |
424 | mode = btrfs_inode_mode(eb, item); | |
425 | if (S_ISDIR(mode)) | |
426 | btrfs_set_inode_size(eb, item, 0); | |
427 | } | |
428 | } else if (inode_item) { | |
429 | struct btrfs_inode_item *item; | |
430 | u32 mode; | |
431 | ||
432 | /* | |
433 | * New inode, set nbytes to 0 so that the nbytes comes out | |
434 | * properly when we replay the extents. | |
435 | */ | |
436 | item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
437 | btrfs_set_inode_nbytes(eb, item, 0); | |
438 | ||
439 | /* | |
440 | * If this is a directory we need to reset the i_size to 0 so | |
441 | * that we can set it up properly when replaying the rest of | |
442 | * the items in this log. | |
443 | */ | |
444 | mode = btrfs_inode_mode(eb, item); | |
445 | if (S_ISDIR(mode)) | |
446 | btrfs_set_inode_size(eb, item, 0); | |
447 | } | |
448 | insert: | |
449 | btrfs_release_path(path); | |
450 | /* try to insert the key into the destination tree */ | |
451 | path->skip_release_on_error = 1; | |
452 | ret = btrfs_insert_empty_item(trans, root, path, | |
453 | key, item_size); | |
454 | path->skip_release_on_error = 0; | |
455 | ||
456 | /* make sure any existing item is the correct size */ | |
457 | if (ret == -EEXIST || ret == -EOVERFLOW) { | |
458 | u32 found_size; | |
459 | found_size = btrfs_item_size_nr(path->nodes[0], | |
460 | path->slots[0]); | |
461 | if (found_size > item_size) | |
462 | btrfs_truncate_item(root, path, item_size, 1); | |
463 | else if (found_size < item_size) | |
464 | btrfs_extend_item(root, path, | |
465 | item_size - found_size); | |
466 | } else if (ret) { | |
467 | return ret; | |
468 | } | |
469 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
470 | path->slots[0]); | |
471 | ||
472 | /* don't overwrite an existing inode if the generation number | |
473 | * was logged as zero. This is done when the tree logging code | |
474 | * is just logging an inode to make sure it exists after recovery. | |
475 | * | |
476 | * Also, don't overwrite i_size on directories during replay. | |
477 | * log replay inserts and removes directory items based on the | |
478 | * state of the tree found in the subvolume, and i_size is modified | |
479 | * as it goes | |
480 | */ | |
481 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
482 | struct btrfs_inode_item *src_item; | |
483 | struct btrfs_inode_item *dst_item; | |
484 | ||
485 | src_item = (struct btrfs_inode_item *)src_ptr; | |
486 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
487 | ||
488 | if (btrfs_inode_generation(eb, src_item) == 0) { | |
489 | struct extent_buffer *dst_eb = path->nodes[0]; | |
490 | const u64 ino_size = btrfs_inode_size(eb, src_item); | |
491 | ||
492 | /* | |
493 | * For regular files an ino_size == 0 is used only when | |
494 | * logging that an inode exists, as part of a directory | |
495 | * fsync, and the inode wasn't fsynced before. In this | |
496 | * case don't set the size of the inode in the fs/subvol | |
497 | * tree, otherwise we would be throwing valid data away. | |
498 | */ | |
499 | if (S_ISREG(btrfs_inode_mode(eb, src_item)) && | |
500 | S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) && | |
501 | ino_size != 0) { | |
502 | struct btrfs_map_token token; | |
503 | ||
504 | btrfs_init_map_token(&token); | |
505 | btrfs_set_token_inode_size(dst_eb, dst_item, | |
506 | ino_size, &token); | |
507 | } | |
508 | goto no_copy; | |
509 | } | |
510 | ||
511 | if (overwrite_root && | |
512 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
513 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
514 | save_old_i_size = 1; | |
515 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
516 | dst_item); | |
517 | } | |
518 | } | |
519 | ||
520 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
521 | src_ptr, item_size); | |
522 | ||
523 | if (save_old_i_size) { | |
524 | struct btrfs_inode_item *dst_item; | |
525 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
526 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
527 | } | |
528 | ||
529 | /* make sure the generation is filled in */ | |
530 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
531 | struct btrfs_inode_item *dst_item; | |
532 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
533 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
534 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
535 | trans->transid); | |
536 | } | |
537 | } | |
538 | no_copy: | |
539 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
540 | btrfs_release_path(path); | |
541 | return 0; | |
542 | } | |
543 | ||
544 | /* | |
545 | * simple helper to read an inode off the disk from a given root | |
546 | * This can only be called for subvolume roots and not for the log | |
547 | */ | |
548 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
549 | u64 objectid) | |
550 | { | |
551 | struct btrfs_key key; | |
552 | struct inode *inode; | |
553 | ||
554 | key.objectid = objectid; | |
555 | key.type = BTRFS_INODE_ITEM_KEY; | |
556 | key.offset = 0; | |
557 | inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); | |
558 | if (IS_ERR(inode)) { | |
559 | inode = NULL; | |
560 | } else if (is_bad_inode(inode)) { | |
561 | iput(inode); | |
562 | inode = NULL; | |
563 | } | |
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 | { | |
585 | int found_type; | |
586 | u64 extent_end; | |
587 | u64 start = key->offset; | |
588 | u64 nbytes = 0; | |
589 | struct btrfs_file_extent_item *item; | |
590 | struct inode *inode = NULL; | |
591 | unsigned long size; | |
592 | int ret = 0; | |
593 | ||
594 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
595 | found_type = btrfs_file_extent_type(eb, item); | |
596 | ||
597 | if (found_type == BTRFS_FILE_EXTENT_REG || | |
598 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
599 | nbytes = btrfs_file_extent_num_bytes(eb, item); | |
600 | extent_end = start + nbytes; | |
601 | ||
602 | /* | |
603 | * We don't add to the inodes nbytes if we are prealloc or a | |
604 | * hole. | |
605 | */ | |
606 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
607 | nbytes = 0; | |
608 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
609 | size = btrfs_file_extent_inline_len(eb, slot, item); | |
610 | nbytes = btrfs_file_extent_ram_bytes(eb, item); | |
611 | extent_end = ALIGN(start + size, root->sectorsize); | |
612 | } else { | |
613 | ret = 0; | |
614 | goto out; | |
615 | } | |
616 | ||
617 | inode = read_one_inode(root, key->objectid); | |
618 | if (!inode) { | |
619 | ret = -EIO; | |
620 | goto out; | |
621 | } | |
622 | ||
623 | /* | |
624 | * first check to see if we already have this extent in the | |
625 | * file. This must be done before the btrfs_drop_extents run | |
626 | * so we don't try to drop this extent. | |
627 | */ | |
628 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), | |
629 | start, 0); | |
630 | ||
631 | if (ret == 0 && | |
632 | (found_type == BTRFS_FILE_EXTENT_REG || | |
633 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
634 | struct btrfs_file_extent_item cmp1; | |
635 | struct btrfs_file_extent_item cmp2; | |
636 | struct btrfs_file_extent_item *existing; | |
637 | struct extent_buffer *leaf; | |
638 | ||
639 | leaf = path->nodes[0]; | |
640 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
641 | struct btrfs_file_extent_item); | |
642 | ||
643 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
644 | sizeof(cmp1)); | |
645 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
646 | sizeof(cmp2)); | |
647 | ||
648 | /* | |
649 | * we already have a pointer to this exact extent, | |
650 | * we don't have to do anything | |
651 | */ | |
652 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
653 | btrfs_release_path(path); | |
654 | goto out; | |
655 | } | |
656 | } | |
657 | btrfs_release_path(path); | |
658 | ||
659 | /* drop any overlapping extents */ | |
660 | ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1); | |
661 | if (ret) | |
662 | goto out; | |
663 | ||
664 | if (found_type == BTRFS_FILE_EXTENT_REG || | |
665 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
666 | u64 offset; | |
667 | unsigned long dest_offset; | |
668 | struct btrfs_key ins; | |
669 | ||
670 | ret = btrfs_insert_empty_item(trans, root, path, key, | |
671 | sizeof(*item)); | |
672 | if (ret) | |
673 | goto out; | |
674 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], | |
675 | path->slots[0]); | |
676 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
677 | (unsigned long)item, sizeof(*item)); | |
678 | ||
679 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
680 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
681 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
682 | offset = key->offset - btrfs_file_extent_offset(eb, item); | |
683 | ||
684 | /* | |
685 | * Manually record dirty extent, as here we did a shallow | |
686 | * file extent item copy and skip normal backref update, | |
687 | * but modifying extent tree all by ourselves. | |
688 | * So need to manually record dirty extent for qgroup, | |
689 | * as the owner of the file extent changed from log tree | |
690 | * (doesn't affect qgroup) to fs/file tree(affects qgroup) | |
691 | */ | |
692 | ret = btrfs_qgroup_insert_dirty_extent(trans, root->fs_info, | |
693 | btrfs_file_extent_disk_bytenr(eb, item), | |
694 | btrfs_file_extent_disk_num_bytes(eb, item), | |
695 | GFP_NOFS); | |
696 | if (ret < 0) | |
697 | goto out; | |
698 | ||
699 | if (ins.objectid > 0) { | |
700 | u64 csum_start; | |
701 | u64 csum_end; | |
702 | LIST_HEAD(ordered_sums); | |
703 | /* | |
704 | * is this extent already allocated in the extent | |
705 | * allocation tree? If so, just add a reference | |
706 | */ | |
707 | ret = btrfs_lookup_data_extent(root, ins.objectid, | |
708 | ins.offset); | |
709 | if (ret == 0) { | |
710 | ret = btrfs_inc_extent_ref(trans, root, | |
711 | ins.objectid, ins.offset, | |
712 | 0, root->root_key.objectid, | |
713 | key->objectid, offset); | |
714 | if (ret) | |
715 | goto out; | |
716 | } else { | |
717 | /* | |
718 | * insert the extent pointer in the extent | |
719 | * allocation tree | |
720 | */ | |
721 | ret = btrfs_alloc_logged_file_extent(trans, | |
722 | root, root->root_key.objectid, | |
723 | key->objectid, offset, &ins); | |
724 | if (ret) | |
725 | goto out; | |
726 | } | |
727 | btrfs_release_path(path); | |
728 | ||
729 | if (btrfs_file_extent_compression(eb, item)) { | |
730 | csum_start = ins.objectid; | |
731 | csum_end = csum_start + ins.offset; | |
732 | } else { | |
733 | csum_start = ins.objectid + | |
734 | btrfs_file_extent_offset(eb, item); | |
735 | csum_end = csum_start + | |
736 | btrfs_file_extent_num_bytes(eb, item); | |
737 | } | |
738 | ||
739 | ret = btrfs_lookup_csums_range(root->log_root, | |
740 | csum_start, csum_end - 1, | |
741 | &ordered_sums, 0); | |
742 | if (ret) | |
743 | goto out; | |
744 | /* | |
745 | * Now delete all existing cums in the csum root that | |
746 | * cover our range. We do this because we can have an | |
747 | * extent that is completely referenced by one file | |
748 | * extent item and partially referenced by another | |
749 | * file extent item (like after using the clone or | |
750 | * extent_same ioctls). In this case if we end up doing | |
751 | * the replay of the one that partially references the | |
752 | * extent first, and we do not do the csum deletion | |
753 | * below, we can get 2 csum items in the csum tree that | |
754 | * overlap each other. For example, imagine our log has | |
755 | * the two following file extent items: | |
756 | * | |
757 | * key (257 EXTENT_DATA 409600) | |
758 | * extent data disk byte 12845056 nr 102400 | |
759 | * extent data offset 20480 nr 20480 ram 102400 | |
760 | * | |
761 | * key (257 EXTENT_DATA 819200) | |
762 | * extent data disk byte 12845056 nr 102400 | |
763 | * extent data offset 0 nr 102400 ram 102400 | |
764 | * | |
765 | * Where the second one fully references the 100K extent | |
766 | * that starts at disk byte 12845056, and the log tree | |
767 | * has a single csum item that covers the entire range | |
768 | * of the extent: | |
769 | * | |
770 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
771 | * | |
772 | * After the first file extent item is replayed, the | |
773 | * csum tree gets the following csum item: | |
774 | * | |
775 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
776 | * | |
777 | * Which covers the 20K sub-range starting at offset 20K | |
778 | * of our extent. Now when we replay the second file | |
779 | * extent item, if we do not delete existing csum items | |
780 | * that cover any of its blocks, we end up getting two | |
781 | * csum items in our csum tree that overlap each other: | |
782 | * | |
783 | * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100 | |
784 | * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20 | |
785 | * | |
786 | * Which is a problem, because after this anyone trying | |
787 | * to lookup up for the checksum of any block of our | |
788 | * extent starting at an offset of 40K or higher, will | |
789 | * end up looking at the second csum item only, which | |
790 | * does not contain the checksum for any block starting | |
791 | * at offset 40K or higher of our extent. | |
792 | */ | |
793 | while (!list_empty(&ordered_sums)) { | |
794 | struct btrfs_ordered_sum *sums; | |
795 | sums = list_entry(ordered_sums.next, | |
796 | struct btrfs_ordered_sum, | |
797 | list); | |
798 | if (!ret) | |
799 | ret = btrfs_del_csums(trans, | |
800 | root->fs_info->csum_root, | |
801 | sums->bytenr, | |
802 | sums->len); | |
803 | if (!ret) | |
804 | ret = btrfs_csum_file_blocks(trans, | |
805 | root->fs_info->csum_root, | |
806 | sums); | |
807 | list_del(&sums->list); | |
808 | kfree(sums); | |
809 | } | |
810 | if (ret) | |
811 | goto out; | |
812 | } else { | |
813 | btrfs_release_path(path); | |
814 | } | |
815 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
816 | /* inline extents are easy, we just overwrite them */ | |
817 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
818 | if (ret) | |
819 | goto out; | |
820 | } | |
821 | ||
822 | inode_add_bytes(inode, nbytes); | |
823 | ret = btrfs_update_inode(trans, root, inode); | |
824 | out: | |
825 | if (inode) | |
826 | iput(inode); | |
827 | return ret; | |
828 | } | |
829 | ||
830 | /* | |
831 | * when cleaning up conflicts between the directory names in the | |
832 | * subvolume, directory names in the log and directory names in the | |
833 | * inode back references, we may have to unlink inodes from directories. | |
834 | * | |
835 | * This is a helper function to do the unlink of a specific directory | |
836 | * item | |
837 | */ | |
838 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
839 | struct btrfs_root *root, | |
840 | struct btrfs_path *path, | |
841 | struct inode *dir, | |
842 | struct btrfs_dir_item *di) | |
843 | { | |
844 | struct inode *inode; | |
845 | char *name; | |
846 | int name_len; | |
847 | struct extent_buffer *leaf; | |
848 | struct btrfs_key location; | |
849 | int ret; | |
850 | ||
851 | leaf = path->nodes[0]; | |
852 | ||
853 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
854 | name_len = btrfs_dir_name_len(leaf, di); | |
855 | name = kmalloc(name_len, GFP_NOFS); | |
856 | if (!name) | |
857 | return -ENOMEM; | |
858 | ||
859 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); | |
860 | btrfs_release_path(path); | |
861 | ||
862 | inode = read_one_inode(root, location.objectid); | |
863 | if (!inode) { | |
864 | ret = -EIO; | |
865 | goto out; | |
866 | } | |
867 | ||
868 | ret = link_to_fixup_dir(trans, root, path, location.objectid); | |
869 | if (ret) | |
870 | goto out; | |
871 | ||
872 | ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
873 | if (ret) | |
874 | goto out; | |
875 | else | |
876 | ret = btrfs_run_delayed_items(trans, root); | |
877 | out: | |
878 | kfree(name); | |
879 | iput(inode); | |
880 | return ret; | |
881 | } | |
882 | ||
883 | /* | |
884 | * helper function to see if a given name and sequence number found | |
885 | * in an inode back reference are already in a directory and correctly | |
886 | * point to this inode | |
887 | */ | |
888 | static noinline int inode_in_dir(struct btrfs_root *root, | |
889 | struct btrfs_path *path, | |
890 | u64 dirid, u64 objectid, u64 index, | |
891 | const char *name, int name_len) | |
892 | { | |
893 | struct btrfs_dir_item *di; | |
894 | struct btrfs_key location; | |
895 | int match = 0; | |
896 | ||
897 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
898 | index, name, name_len, 0); | |
899 | if (di && !IS_ERR(di)) { | |
900 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
901 | if (location.objectid != objectid) | |
902 | goto out; | |
903 | } else | |
904 | goto out; | |
905 | btrfs_release_path(path); | |
906 | ||
907 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
908 | if (di && !IS_ERR(di)) { | |
909 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
910 | if (location.objectid != objectid) | |
911 | goto out; | |
912 | } else | |
913 | goto out; | |
914 | match = 1; | |
915 | out: | |
916 | btrfs_release_path(path); | |
917 | return match; | |
918 | } | |
919 | ||
920 | /* | |
921 | * helper function to check a log tree for a named back reference in | |
922 | * an inode. This is used to decide if a back reference that is | |
923 | * found in the subvolume conflicts with what we find in the log. | |
924 | * | |
925 | * inode backreferences may have multiple refs in a single item, | |
926 | * during replay we process one reference at a time, and we don't | |
927 | * want to delete valid links to a file from the subvolume if that | |
928 | * link is also in the log. | |
929 | */ | |
930 | static noinline int backref_in_log(struct btrfs_root *log, | |
931 | struct btrfs_key *key, | |
932 | u64 ref_objectid, | |
933 | const char *name, int namelen) | |
934 | { | |
935 | struct btrfs_path *path; | |
936 | struct btrfs_inode_ref *ref; | |
937 | unsigned long ptr; | |
938 | unsigned long ptr_end; | |
939 | unsigned long name_ptr; | |
940 | int found_name_len; | |
941 | int item_size; | |
942 | int ret; | |
943 | int match = 0; | |
944 | ||
945 | path = btrfs_alloc_path(); | |
946 | if (!path) | |
947 | return -ENOMEM; | |
948 | ||
949 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); | |
950 | if (ret != 0) | |
951 | goto out; | |
952 | ||
953 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
954 | ||
955 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
956 | if (btrfs_find_name_in_ext_backref(path, ref_objectid, | |
957 | name, namelen, NULL)) | |
958 | match = 1; | |
959 | ||
960 | goto out; | |
961 | } | |
962 | ||
963 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); | |
964 | ptr_end = ptr + item_size; | |
965 | while (ptr < ptr_end) { | |
966 | ref = (struct btrfs_inode_ref *)ptr; | |
967 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); | |
968 | if (found_name_len == namelen) { | |
969 | name_ptr = (unsigned long)(ref + 1); | |
970 | ret = memcmp_extent_buffer(path->nodes[0], name, | |
971 | name_ptr, namelen); | |
972 | if (ret == 0) { | |
973 | match = 1; | |
974 | goto out; | |
975 | } | |
976 | } | |
977 | ptr = (unsigned long)(ref + 1) + found_name_len; | |
978 | } | |
979 | out: | |
980 | btrfs_free_path(path); | |
981 | return match; | |
982 | } | |
983 | ||
984 | static inline int __add_inode_ref(struct btrfs_trans_handle *trans, | |
985 | struct btrfs_root *root, | |
986 | struct btrfs_path *path, | |
987 | struct btrfs_root *log_root, | |
988 | struct inode *dir, struct inode *inode, | |
989 | struct extent_buffer *eb, | |
990 | u64 inode_objectid, u64 parent_objectid, | |
991 | u64 ref_index, char *name, int namelen, | |
992 | int *search_done) | |
993 | { | |
994 | int ret; | |
995 | char *victim_name; | |
996 | int victim_name_len; | |
997 | struct extent_buffer *leaf; | |
998 | struct btrfs_dir_item *di; | |
999 | struct btrfs_key search_key; | |
1000 | struct btrfs_inode_extref *extref; | |
1001 | ||
1002 | again: | |
1003 | /* Search old style refs */ | |
1004 | search_key.objectid = inode_objectid; | |
1005 | search_key.type = BTRFS_INODE_REF_KEY; | |
1006 | search_key.offset = parent_objectid; | |
1007 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
1008 | if (ret == 0) { | |
1009 | struct btrfs_inode_ref *victim_ref; | |
1010 | unsigned long ptr; | |
1011 | unsigned long ptr_end; | |
1012 | ||
1013 | leaf = path->nodes[0]; | |
1014 | ||
1015 | /* are we trying to overwrite a back ref for the root directory | |
1016 | * if so, just jump out, we're done | |
1017 | */ | |
1018 | if (search_key.objectid == search_key.offset) | |
1019 | return 1; | |
1020 | ||
1021 | /* check all the names in this back reference to see | |
1022 | * if they are in the log. if so, we allow them to stay | |
1023 | * otherwise they must be unlinked as a conflict | |
1024 | */ | |
1025 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1026 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
1027 | while (ptr < ptr_end) { | |
1028 | victim_ref = (struct btrfs_inode_ref *)ptr; | |
1029 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
1030 | victim_ref); | |
1031 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
1032 | if (!victim_name) | |
1033 | return -ENOMEM; | |
1034 | ||
1035 | read_extent_buffer(leaf, victim_name, | |
1036 | (unsigned long)(victim_ref + 1), | |
1037 | victim_name_len); | |
1038 | ||
1039 | if (!backref_in_log(log_root, &search_key, | |
1040 | parent_objectid, | |
1041 | victim_name, | |
1042 | victim_name_len)) { | |
1043 | inc_nlink(inode); | |
1044 | btrfs_release_path(path); | |
1045 | ||
1046 | ret = btrfs_unlink_inode(trans, root, dir, | |
1047 | inode, victim_name, | |
1048 | victim_name_len); | |
1049 | kfree(victim_name); | |
1050 | if (ret) | |
1051 | return ret; | |
1052 | ret = btrfs_run_delayed_items(trans, root); | |
1053 | if (ret) | |
1054 | return ret; | |
1055 | *search_done = 1; | |
1056 | goto again; | |
1057 | } | |
1058 | kfree(victim_name); | |
1059 | ||
1060 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; | |
1061 | } | |
1062 | ||
1063 | /* | |
1064 | * NOTE: we have searched root tree and checked the | |
1065 | * corresponding ref, it does not need to check again. | |
1066 | */ | |
1067 | *search_done = 1; | |
1068 | } | |
1069 | btrfs_release_path(path); | |
1070 | ||
1071 | /* Same search but for extended refs */ | |
1072 | extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen, | |
1073 | inode_objectid, parent_objectid, 0, | |
1074 | 0); | |
1075 | if (!IS_ERR_OR_NULL(extref)) { | |
1076 | u32 item_size; | |
1077 | u32 cur_offset = 0; | |
1078 | unsigned long base; | |
1079 | struct inode *victim_parent; | |
1080 | ||
1081 | leaf = path->nodes[0]; | |
1082 | ||
1083 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1084 | base = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1085 | ||
1086 | while (cur_offset < item_size) { | |
1087 | extref = (struct btrfs_inode_extref *)(base + cur_offset); | |
1088 | ||
1089 | victim_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1090 | ||
1091 | if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid) | |
1092 | goto next; | |
1093 | ||
1094 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
1095 | if (!victim_name) | |
1096 | return -ENOMEM; | |
1097 | read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name, | |
1098 | victim_name_len); | |
1099 | ||
1100 | search_key.objectid = inode_objectid; | |
1101 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1102 | search_key.offset = btrfs_extref_hash(parent_objectid, | |
1103 | victim_name, | |
1104 | victim_name_len); | |
1105 | ret = 0; | |
1106 | if (!backref_in_log(log_root, &search_key, | |
1107 | parent_objectid, victim_name, | |
1108 | victim_name_len)) { | |
1109 | ret = -ENOENT; | |
1110 | victim_parent = read_one_inode(root, | |
1111 | parent_objectid); | |
1112 | if (victim_parent) { | |
1113 | inc_nlink(inode); | |
1114 | btrfs_release_path(path); | |
1115 | ||
1116 | ret = btrfs_unlink_inode(trans, root, | |
1117 | victim_parent, | |
1118 | inode, | |
1119 | victim_name, | |
1120 | victim_name_len); | |
1121 | if (!ret) | |
1122 | ret = btrfs_run_delayed_items( | |
1123 | trans, root); | |
1124 | } | |
1125 | iput(victim_parent); | |
1126 | kfree(victim_name); | |
1127 | if (ret) | |
1128 | return ret; | |
1129 | *search_done = 1; | |
1130 | goto again; | |
1131 | } | |
1132 | kfree(victim_name); | |
1133 | if (ret) | |
1134 | return ret; | |
1135 | next: | |
1136 | cur_offset += victim_name_len + sizeof(*extref); | |
1137 | } | |
1138 | *search_done = 1; | |
1139 | } | |
1140 | btrfs_release_path(path); | |
1141 | ||
1142 | /* look for a conflicting sequence number */ | |
1143 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), | |
1144 | ref_index, name, namelen, 0); | |
1145 | if (di && !IS_ERR(di)) { | |
1146 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
1147 | if (ret) | |
1148 | return ret; | |
1149 | } | |
1150 | btrfs_release_path(path); | |
1151 | ||
1152 | /* look for a conflicing name */ | |
1153 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), | |
1154 | name, namelen, 0); | |
1155 | if (di && !IS_ERR(di)) { | |
1156 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
1157 | if (ret) | |
1158 | return ret; | |
1159 | } | |
1160 | btrfs_release_path(path); | |
1161 | ||
1162 | return 0; | |
1163 | } | |
1164 | ||
1165 | static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, | |
1166 | u32 *namelen, char **name, u64 *index, | |
1167 | u64 *parent_objectid) | |
1168 | { | |
1169 | struct btrfs_inode_extref *extref; | |
1170 | ||
1171 | extref = (struct btrfs_inode_extref *)ref_ptr; | |
1172 | ||
1173 | *namelen = btrfs_inode_extref_name_len(eb, extref); | |
1174 | *name = kmalloc(*namelen, GFP_NOFS); | |
1175 | if (*name == NULL) | |
1176 | return -ENOMEM; | |
1177 | ||
1178 | read_extent_buffer(eb, *name, (unsigned long)&extref->name, | |
1179 | *namelen); | |
1180 | ||
1181 | *index = btrfs_inode_extref_index(eb, extref); | |
1182 | if (parent_objectid) | |
1183 | *parent_objectid = btrfs_inode_extref_parent(eb, extref); | |
1184 | ||
1185 | return 0; | |
1186 | } | |
1187 | ||
1188 | static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr, | |
1189 | u32 *namelen, char **name, u64 *index) | |
1190 | { | |
1191 | struct btrfs_inode_ref *ref; | |
1192 | ||
1193 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
1194 | ||
1195 | *namelen = btrfs_inode_ref_name_len(eb, ref); | |
1196 | *name = kmalloc(*namelen, GFP_NOFS); | |
1197 | if (*name == NULL) | |
1198 | return -ENOMEM; | |
1199 | ||
1200 | read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen); | |
1201 | ||
1202 | *index = btrfs_inode_ref_index(eb, ref); | |
1203 | ||
1204 | return 0; | |
1205 | } | |
1206 | ||
1207 | /* | |
1208 | * replay one inode back reference item found in the log tree. | |
1209 | * eb, slot and key refer to the buffer and key found in the log tree. | |
1210 | * root is the destination we are replaying into, and path is for temp | |
1211 | * use by this function. (it should be released on return). | |
1212 | */ | |
1213 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
1214 | struct btrfs_root *root, | |
1215 | struct btrfs_root *log, | |
1216 | struct btrfs_path *path, | |
1217 | struct extent_buffer *eb, int slot, | |
1218 | struct btrfs_key *key) | |
1219 | { | |
1220 | struct inode *dir = NULL; | |
1221 | struct inode *inode = NULL; | |
1222 | unsigned long ref_ptr; | |
1223 | unsigned long ref_end; | |
1224 | char *name = NULL; | |
1225 | int namelen; | |
1226 | int ret; | |
1227 | int search_done = 0; | |
1228 | int log_ref_ver = 0; | |
1229 | u64 parent_objectid; | |
1230 | u64 inode_objectid; | |
1231 | u64 ref_index = 0; | |
1232 | int ref_struct_size; | |
1233 | ||
1234 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
1235 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
1236 | ||
1237 | if (key->type == BTRFS_INODE_EXTREF_KEY) { | |
1238 | struct btrfs_inode_extref *r; | |
1239 | ||
1240 | ref_struct_size = sizeof(struct btrfs_inode_extref); | |
1241 | log_ref_ver = 1; | |
1242 | r = (struct btrfs_inode_extref *)ref_ptr; | |
1243 | parent_objectid = btrfs_inode_extref_parent(eb, r); | |
1244 | } else { | |
1245 | ref_struct_size = sizeof(struct btrfs_inode_ref); | |
1246 | parent_objectid = key->offset; | |
1247 | } | |
1248 | inode_objectid = key->objectid; | |
1249 | ||
1250 | /* | |
1251 | * it is possible that we didn't log all the parent directories | |
1252 | * for a given inode. If we don't find the dir, just don't | |
1253 | * copy the back ref in. The link count fixup code will take | |
1254 | * care of the rest | |
1255 | */ | |
1256 | dir = read_one_inode(root, parent_objectid); | |
1257 | if (!dir) { | |
1258 | ret = -ENOENT; | |
1259 | goto out; | |
1260 | } | |
1261 | ||
1262 | inode = read_one_inode(root, inode_objectid); | |
1263 | if (!inode) { | |
1264 | ret = -EIO; | |
1265 | goto out; | |
1266 | } | |
1267 | ||
1268 | while (ref_ptr < ref_end) { | |
1269 | if (log_ref_ver) { | |
1270 | ret = extref_get_fields(eb, ref_ptr, &namelen, &name, | |
1271 | &ref_index, &parent_objectid); | |
1272 | /* | |
1273 | * parent object can change from one array | |
1274 | * item to another. | |
1275 | */ | |
1276 | if (!dir) | |
1277 | dir = read_one_inode(root, parent_objectid); | |
1278 | if (!dir) { | |
1279 | ret = -ENOENT; | |
1280 | goto out; | |
1281 | } | |
1282 | } else { | |
1283 | ret = ref_get_fields(eb, ref_ptr, &namelen, &name, | |
1284 | &ref_index); | |
1285 | } | |
1286 | if (ret) | |
1287 | goto out; | |
1288 | ||
1289 | /* if we already have a perfect match, we're done */ | |
1290 | if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode), | |
1291 | ref_index, name, namelen)) { | |
1292 | /* | |
1293 | * look for a conflicting back reference in the | |
1294 | * metadata. if we find one we have to unlink that name | |
1295 | * of the file before we add our new link. Later on, we | |
1296 | * overwrite any existing back reference, and we don't | |
1297 | * want to create dangling pointers in the directory. | |
1298 | */ | |
1299 | ||
1300 | if (!search_done) { | |
1301 | ret = __add_inode_ref(trans, root, path, log, | |
1302 | dir, inode, eb, | |
1303 | inode_objectid, | |
1304 | parent_objectid, | |
1305 | ref_index, name, namelen, | |
1306 | &search_done); | |
1307 | if (ret) { | |
1308 | if (ret == 1) | |
1309 | ret = 0; | |
1310 | goto out; | |
1311 | } | |
1312 | } | |
1313 | ||
1314 | /* insert our name */ | |
1315 | ret = btrfs_add_link(trans, dir, inode, name, namelen, | |
1316 | 0, ref_index); | |
1317 | if (ret) | |
1318 | goto out; | |
1319 | ||
1320 | btrfs_update_inode(trans, root, inode); | |
1321 | } | |
1322 | ||
1323 | ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen; | |
1324 | kfree(name); | |
1325 | name = NULL; | |
1326 | if (log_ref_ver) { | |
1327 | iput(dir); | |
1328 | dir = NULL; | |
1329 | } | |
1330 | } | |
1331 | ||
1332 | /* finally write the back reference in the inode */ | |
1333 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
1334 | out: | |
1335 | btrfs_release_path(path); | |
1336 | kfree(name); | |
1337 | iput(dir); | |
1338 | iput(inode); | |
1339 | return ret; | |
1340 | } | |
1341 | ||
1342 | static int insert_orphan_item(struct btrfs_trans_handle *trans, | |
1343 | struct btrfs_root *root, u64 ino) | |
1344 | { | |
1345 | int ret; | |
1346 | ||
1347 | ret = btrfs_insert_orphan_item(trans, root, ino); | |
1348 | if (ret == -EEXIST) | |
1349 | ret = 0; | |
1350 | ||
1351 | return ret; | |
1352 | } | |
1353 | ||
1354 | static int count_inode_extrefs(struct btrfs_root *root, | |
1355 | struct inode *inode, struct btrfs_path *path) | |
1356 | { | |
1357 | int ret = 0; | |
1358 | int name_len; | |
1359 | unsigned int nlink = 0; | |
1360 | u32 item_size; | |
1361 | u32 cur_offset = 0; | |
1362 | u64 inode_objectid = btrfs_ino(inode); | |
1363 | u64 offset = 0; | |
1364 | unsigned long ptr; | |
1365 | struct btrfs_inode_extref *extref; | |
1366 | struct extent_buffer *leaf; | |
1367 | ||
1368 | while (1) { | |
1369 | ret = btrfs_find_one_extref(root, inode_objectid, offset, path, | |
1370 | &extref, &offset); | |
1371 | if (ret) | |
1372 | break; | |
1373 | ||
1374 | leaf = path->nodes[0]; | |
1375 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1376 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1377 | cur_offset = 0; | |
1378 | ||
1379 | while (cur_offset < item_size) { | |
1380 | extref = (struct btrfs_inode_extref *) (ptr + cur_offset); | |
1381 | name_len = btrfs_inode_extref_name_len(leaf, extref); | |
1382 | ||
1383 | nlink++; | |
1384 | ||
1385 | cur_offset += name_len + sizeof(*extref); | |
1386 | } | |
1387 | ||
1388 | offset++; | |
1389 | btrfs_release_path(path); | |
1390 | } | |
1391 | btrfs_release_path(path); | |
1392 | ||
1393 | if (ret < 0 && ret != -ENOENT) | |
1394 | return ret; | |
1395 | return nlink; | |
1396 | } | |
1397 | ||
1398 | static int count_inode_refs(struct btrfs_root *root, | |
1399 | struct inode *inode, struct btrfs_path *path) | |
1400 | { | |
1401 | int ret; | |
1402 | struct btrfs_key key; | |
1403 | unsigned int nlink = 0; | |
1404 | unsigned long ptr; | |
1405 | unsigned long ptr_end; | |
1406 | int name_len; | |
1407 | u64 ino = btrfs_ino(inode); | |
1408 | ||
1409 | key.objectid = ino; | |
1410 | key.type = BTRFS_INODE_REF_KEY; | |
1411 | key.offset = (u64)-1; | |
1412 | ||
1413 | while (1) { | |
1414 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1415 | if (ret < 0) | |
1416 | break; | |
1417 | if (ret > 0) { | |
1418 | if (path->slots[0] == 0) | |
1419 | break; | |
1420 | path->slots[0]--; | |
1421 | } | |
1422 | process_slot: | |
1423 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1424 | path->slots[0]); | |
1425 | if (key.objectid != ino || | |
1426 | key.type != BTRFS_INODE_REF_KEY) | |
1427 | break; | |
1428 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
1429 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
1430 | path->slots[0]); | |
1431 | while (ptr < ptr_end) { | |
1432 | struct btrfs_inode_ref *ref; | |
1433 | ||
1434 | ref = (struct btrfs_inode_ref *)ptr; | |
1435 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
1436 | ref); | |
1437 | ptr = (unsigned long)(ref + 1) + name_len; | |
1438 | nlink++; | |
1439 | } | |
1440 | ||
1441 | if (key.offset == 0) | |
1442 | break; | |
1443 | if (path->slots[0] > 0) { | |
1444 | path->slots[0]--; | |
1445 | goto process_slot; | |
1446 | } | |
1447 | key.offset--; | |
1448 | btrfs_release_path(path); | |
1449 | } | |
1450 | btrfs_release_path(path); | |
1451 | ||
1452 | return nlink; | |
1453 | } | |
1454 | ||
1455 | /* | |
1456 | * There are a few corners where the link count of the file can't | |
1457 | * be properly maintained during replay. So, instead of adding | |
1458 | * lots of complexity to the log code, we just scan the backrefs | |
1459 | * for any file that has been through replay. | |
1460 | * | |
1461 | * The scan will update the link count on the inode to reflect the | |
1462 | * number of back refs found. If it goes down to zero, the iput | |
1463 | * will free the inode. | |
1464 | */ | |
1465 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
1466 | struct btrfs_root *root, | |
1467 | struct inode *inode) | |
1468 | { | |
1469 | struct btrfs_path *path; | |
1470 | int ret; | |
1471 | u64 nlink = 0; | |
1472 | u64 ino = btrfs_ino(inode); | |
1473 | ||
1474 | path = btrfs_alloc_path(); | |
1475 | if (!path) | |
1476 | return -ENOMEM; | |
1477 | ||
1478 | ret = count_inode_refs(root, inode, path); | |
1479 | if (ret < 0) | |
1480 | goto out; | |
1481 | ||
1482 | nlink = ret; | |
1483 | ||
1484 | ret = count_inode_extrefs(root, inode, path); | |
1485 | if (ret < 0) | |
1486 | goto out; | |
1487 | ||
1488 | nlink += ret; | |
1489 | ||
1490 | ret = 0; | |
1491 | ||
1492 | if (nlink != inode->i_nlink) { | |
1493 | set_nlink(inode, nlink); | |
1494 | btrfs_update_inode(trans, root, inode); | |
1495 | } | |
1496 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
1497 | ||
1498 | if (inode->i_nlink == 0) { | |
1499 | if (S_ISDIR(inode->i_mode)) { | |
1500 | ret = replay_dir_deletes(trans, root, NULL, path, | |
1501 | ino, 1); | |
1502 | if (ret) | |
1503 | goto out; | |
1504 | } | |
1505 | ret = insert_orphan_item(trans, root, ino); | |
1506 | } | |
1507 | ||
1508 | out: | |
1509 | btrfs_free_path(path); | |
1510 | return ret; | |
1511 | } | |
1512 | ||
1513 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1514 | struct btrfs_root *root, | |
1515 | struct btrfs_path *path) | |
1516 | { | |
1517 | int ret; | |
1518 | struct btrfs_key key; | |
1519 | struct inode *inode; | |
1520 | ||
1521 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1522 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1523 | key.offset = (u64)-1; | |
1524 | while (1) { | |
1525 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1526 | if (ret < 0) | |
1527 | break; | |
1528 | ||
1529 | if (ret == 1) { | |
1530 | if (path->slots[0] == 0) | |
1531 | break; | |
1532 | path->slots[0]--; | |
1533 | } | |
1534 | ||
1535 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1536 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1537 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1538 | break; | |
1539 | ||
1540 | ret = btrfs_del_item(trans, root, path); | |
1541 | if (ret) | |
1542 | goto out; | |
1543 | ||
1544 | btrfs_release_path(path); | |
1545 | inode = read_one_inode(root, key.offset); | |
1546 | if (!inode) | |
1547 | return -EIO; | |
1548 | ||
1549 | ret = fixup_inode_link_count(trans, root, inode); | |
1550 | iput(inode); | |
1551 | if (ret) | |
1552 | goto out; | |
1553 | ||
1554 | /* | |
1555 | * fixup on a directory may create new entries, | |
1556 | * make sure we always look for the highset possible | |
1557 | * offset | |
1558 | */ | |
1559 | key.offset = (u64)-1; | |
1560 | } | |
1561 | ret = 0; | |
1562 | out: | |
1563 | btrfs_release_path(path); | |
1564 | return ret; | |
1565 | } | |
1566 | ||
1567 | ||
1568 | /* | |
1569 | * record a given inode in the fixup dir so we can check its link | |
1570 | * count when replay is done. The link count is incremented here | |
1571 | * so the inode won't go away until we check it | |
1572 | */ | |
1573 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1574 | struct btrfs_root *root, | |
1575 | struct btrfs_path *path, | |
1576 | u64 objectid) | |
1577 | { | |
1578 | struct btrfs_key key; | |
1579 | int ret = 0; | |
1580 | struct inode *inode; | |
1581 | ||
1582 | inode = read_one_inode(root, objectid); | |
1583 | if (!inode) | |
1584 | return -EIO; | |
1585 | ||
1586 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1587 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1588 | key.offset = objectid; | |
1589 | ||
1590 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1591 | ||
1592 | btrfs_release_path(path); | |
1593 | if (ret == 0) { | |
1594 | if (!inode->i_nlink) | |
1595 | set_nlink(inode, 1); | |
1596 | else | |
1597 | inc_nlink(inode); | |
1598 | ret = btrfs_update_inode(trans, root, inode); | |
1599 | } else if (ret == -EEXIST) { | |
1600 | ret = 0; | |
1601 | } else { | |
1602 | BUG(); /* Logic Error */ | |
1603 | } | |
1604 | iput(inode); | |
1605 | ||
1606 | return ret; | |
1607 | } | |
1608 | ||
1609 | /* | |
1610 | * when replaying the log for a directory, we only insert names | |
1611 | * for inodes that actually exist. This means an fsync on a directory | |
1612 | * does not implicitly fsync all the new files in it | |
1613 | */ | |
1614 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1615 | struct btrfs_root *root, | |
1616 | u64 dirid, u64 index, | |
1617 | char *name, int name_len, | |
1618 | struct btrfs_key *location) | |
1619 | { | |
1620 | struct inode *inode; | |
1621 | struct inode *dir; | |
1622 | int ret; | |
1623 | ||
1624 | inode = read_one_inode(root, location->objectid); | |
1625 | if (!inode) | |
1626 | return -ENOENT; | |
1627 | ||
1628 | dir = read_one_inode(root, dirid); | |
1629 | if (!dir) { | |
1630 | iput(inode); | |
1631 | return -EIO; | |
1632 | } | |
1633 | ||
1634 | ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index); | |
1635 | ||
1636 | /* FIXME, put inode into FIXUP list */ | |
1637 | ||
1638 | iput(inode); | |
1639 | iput(dir); | |
1640 | return ret; | |
1641 | } | |
1642 | ||
1643 | /* | |
1644 | * Return true if an inode reference exists in the log for the given name, | |
1645 | * inode and parent inode. | |
1646 | */ | |
1647 | static bool name_in_log_ref(struct btrfs_root *log_root, | |
1648 | const char *name, const int name_len, | |
1649 | const u64 dirid, const u64 ino) | |
1650 | { | |
1651 | struct btrfs_key search_key; | |
1652 | ||
1653 | search_key.objectid = ino; | |
1654 | search_key.type = BTRFS_INODE_REF_KEY; | |
1655 | search_key.offset = dirid; | |
1656 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1657 | return true; | |
1658 | ||
1659 | search_key.type = BTRFS_INODE_EXTREF_KEY; | |
1660 | search_key.offset = btrfs_extref_hash(dirid, name, name_len); | |
1661 | if (backref_in_log(log_root, &search_key, dirid, name, name_len)) | |
1662 | return true; | |
1663 | ||
1664 | return false; | |
1665 | } | |
1666 | ||
1667 | /* | |
1668 | * take a single entry in a log directory item and replay it into | |
1669 | * the subvolume. | |
1670 | * | |
1671 | * if a conflicting item exists in the subdirectory already, | |
1672 | * the inode it points to is unlinked and put into the link count | |
1673 | * fix up tree. | |
1674 | * | |
1675 | * If a name from the log points to a file or directory that does | |
1676 | * not exist in the FS, it is skipped. fsyncs on directories | |
1677 | * do not force down inodes inside that directory, just changes to the | |
1678 | * names or unlinks in a directory. | |
1679 | * | |
1680 | * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a | |
1681 | * non-existing inode) and 1 if the name was replayed. | |
1682 | */ | |
1683 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1684 | struct btrfs_root *root, | |
1685 | struct btrfs_path *path, | |
1686 | struct extent_buffer *eb, | |
1687 | struct btrfs_dir_item *di, | |
1688 | struct btrfs_key *key) | |
1689 | { | |
1690 | char *name; | |
1691 | int name_len; | |
1692 | struct btrfs_dir_item *dst_di; | |
1693 | struct btrfs_key found_key; | |
1694 | struct btrfs_key log_key; | |
1695 | struct inode *dir; | |
1696 | u8 log_type; | |
1697 | int exists; | |
1698 | int ret = 0; | |
1699 | bool update_size = (key->type == BTRFS_DIR_INDEX_KEY); | |
1700 | bool name_added = false; | |
1701 | ||
1702 | dir = read_one_inode(root, key->objectid); | |
1703 | if (!dir) | |
1704 | return -EIO; | |
1705 | ||
1706 | name_len = btrfs_dir_name_len(eb, di); | |
1707 | name = kmalloc(name_len, GFP_NOFS); | |
1708 | if (!name) { | |
1709 | ret = -ENOMEM; | |
1710 | goto out; | |
1711 | } | |
1712 | ||
1713 | log_type = btrfs_dir_type(eb, di); | |
1714 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1715 | name_len); | |
1716 | ||
1717 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
1718 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); | |
1719 | if (exists == 0) | |
1720 | exists = 1; | |
1721 | else | |
1722 | exists = 0; | |
1723 | btrfs_release_path(path); | |
1724 | ||
1725 | if (key->type == BTRFS_DIR_ITEM_KEY) { | |
1726 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1727 | name, name_len, 1); | |
1728 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { | |
1729 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, | |
1730 | key->objectid, | |
1731 | key->offset, name, | |
1732 | name_len, 1); | |
1733 | } else { | |
1734 | /* Corruption */ | |
1735 | ret = -EINVAL; | |
1736 | goto out; | |
1737 | } | |
1738 | if (IS_ERR_OR_NULL(dst_di)) { | |
1739 | /* we need a sequence number to insert, so we only | |
1740 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1741 | */ | |
1742 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1743 | goto out; | |
1744 | goto insert; | |
1745 | } | |
1746 | ||
1747 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1748 | /* the existing item matches the logged item */ | |
1749 | if (found_key.objectid == log_key.objectid && | |
1750 | found_key.type == log_key.type && | |
1751 | found_key.offset == log_key.offset && | |
1752 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
1753 | update_size = false; | |
1754 | goto out; | |
1755 | } | |
1756 | ||
1757 | /* | |
1758 | * don't drop the conflicting directory entry if the inode | |
1759 | * for the new entry doesn't exist | |
1760 | */ | |
1761 | if (!exists) | |
1762 | goto out; | |
1763 | ||
1764 | ret = drop_one_dir_item(trans, root, path, dir, dst_di); | |
1765 | if (ret) | |
1766 | goto out; | |
1767 | ||
1768 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1769 | goto insert; | |
1770 | out: | |
1771 | btrfs_release_path(path); | |
1772 | if (!ret && update_size) { | |
1773 | btrfs_i_size_write(dir, dir->i_size + name_len * 2); | |
1774 | ret = btrfs_update_inode(trans, root, dir); | |
1775 | } | |
1776 | kfree(name); | |
1777 | iput(dir); | |
1778 | if (!ret && name_added) | |
1779 | ret = 1; | |
1780 | return ret; | |
1781 | ||
1782 | insert: | |
1783 | if (name_in_log_ref(root->log_root, name, name_len, | |
1784 | key->objectid, log_key.objectid)) { | |
1785 | /* The dentry will be added later. */ | |
1786 | ret = 0; | |
1787 | update_size = false; | |
1788 | goto out; | |
1789 | } | |
1790 | btrfs_release_path(path); | |
1791 | ret = insert_one_name(trans, root, key->objectid, key->offset, | |
1792 | name, name_len, &log_key); | |
1793 | if (ret && ret != -ENOENT && ret != -EEXIST) | |
1794 | goto out; | |
1795 | if (!ret) | |
1796 | name_added = true; | |
1797 | update_size = false; | |
1798 | ret = 0; | |
1799 | goto out; | |
1800 | } | |
1801 | ||
1802 | /* | |
1803 | * find all the names in a directory item and reconcile them into | |
1804 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
1805 | * one name in a directory item, but the same code gets used for | |
1806 | * both directory index types | |
1807 | */ | |
1808 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
1809 | struct btrfs_root *root, | |
1810 | struct btrfs_path *path, | |
1811 | struct extent_buffer *eb, int slot, | |
1812 | struct btrfs_key *key) | |
1813 | { | |
1814 | int ret = 0; | |
1815 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
1816 | struct btrfs_dir_item *di; | |
1817 | int name_len; | |
1818 | unsigned long ptr; | |
1819 | unsigned long ptr_end; | |
1820 | struct btrfs_path *fixup_path = NULL; | |
1821 | ||
1822 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1823 | ptr_end = ptr + item_size; | |
1824 | while (ptr < ptr_end) { | |
1825 | di = (struct btrfs_dir_item *)ptr; | |
1826 | if (verify_dir_item(root, eb, di)) | |
1827 | return -EIO; | |
1828 | name_len = btrfs_dir_name_len(eb, di); | |
1829 | ret = replay_one_name(trans, root, path, eb, di, key); | |
1830 | if (ret < 0) | |
1831 | break; | |
1832 | ptr = (unsigned long)(di + 1); | |
1833 | ptr += name_len; | |
1834 | ||
1835 | /* | |
1836 | * If this entry refers to a non-directory (directories can not | |
1837 | * have a link count > 1) and it was added in the transaction | |
1838 | * that was not committed, make sure we fixup the link count of | |
1839 | * the inode it the entry points to. Otherwise something like | |
1840 | * the following would result in a directory pointing to an | |
1841 | * inode with a wrong link that does not account for this dir | |
1842 | * entry: | |
1843 | * | |
1844 | * mkdir testdir | |
1845 | * touch testdir/foo | |
1846 | * touch testdir/bar | |
1847 | * sync | |
1848 | * | |
1849 | * ln testdir/bar testdir/bar_link | |
1850 | * ln testdir/foo testdir/foo_link | |
1851 | * xfs_io -c "fsync" testdir/bar | |
1852 | * | |
1853 | * <power failure> | |
1854 | * | |
1855 | * mount fs, log replay happens | |
1856 | * | |
1857 | * File foo would remain with a link count of 1 when it has two | |
1858 | * entries pointing to it in the directory testdir. This would | |
1859 | * make it impossible to ever delete the parent directory has | |
1860 | * it would result in stale dentries that can never be deleted. | |
1861 | */ | |
1862 | if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) { | |
1863 | struct btrfs_key di_key; | |
1864 | ||
1865 | if (!fixup_path) { | |
1866 | fixup_path = btrfs_alloc_path(); | |
1867 | if (!fixup_path) { | |
1868 | ret = -ENOMEM; | |
1869 | break; | |
1870 | } | |
1871 | } | |
1872 | ||
1873 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1874 | ret = link_to_fixup_dir(trans, root, fixup_path, | |
1875 | di_key.objectid); | |
1876 | if (ret) | |
1877 | break; | |
1878 | } | |
1879 | ret = 0; | |
1880 | } | |
1881 | btrfs_free_path(fixup_path); | |
1882 | return ret; | |
1883 | } | |
1884 | ||
1885 | /* | |
1886 | * directory replay has two parts. There are the standard directory | |
1887 | * items in the log copied from the subvolume, and range items | |
1888 | * created in the log while the subvolume was logged. | |
1889 | * | |
1890 | * The range items tell us which parts of the key space the log | |
1891 | * is authoritative for. During replay, if a key in the subvolume | |
1892 | * directory is in a logged range item, but not actually in the log | |
1893 | * that means it was deleted from the directory before the fsync | |
1894 | * and should be removed. | |
1895 | */ | |
1896 | static noinline int find_dir_range(struct btrfs_root *root, | |
1897 | struct btrfs_path *path, | |
1898 | u64 dirid, int key_type, | |
1899 | u64 *start_ret, u64 *end_ret) | |
1900 | { | |
1901 | struct btrfs_key key; | |
1902 | u64 found_end; | |
1903 | struct btrfs_dir_log_item *item; | |
1904 | int ret; | |
1905 | int nritems; | |
1906 | ||
1907 | if (*start_ret == (u64)-1) | |
1908 | return 1; | |
1909 | ||
1910 | key.objectid = dirid; | |
1911 | key.type = key_type; | |
1912 | key.offset = *start_ret; | |
1913 | ||
1914 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1915 | if (ret < 0) | |
1916 | goto out; | |
1917 | if (ret > 0) { | |
1918 | if (path->slots[0] == 0) | |
1919 | goto out; | |
1920 | path->slots[0]--; | |
1921 | } | |
1922 | if (ret != 0) | |
1923 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1924 | ||
1925 | if (key.type != key_type || key.objectid != dirid) { | |
1926 | ret = 1; | |
1927 | goto next; | |
1928 | } | |
1929 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1930 | struct btrfs_dir_log_item); | |
1931 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1932 | ||
1933 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
1934 | ret = 0; | |
1935 | *start_ret = key.offset; | |
1936 | *end_ret = found_end; | |
1937 | goto out; | |
1938 | } | |
1939 | ret = 1; | |
1940 | next: | |
1941 | /* check the next slot in the tree to see if it is a valid item */ | |
1942 | nritems = btrfs_header_nritems(path->nodes[0]); | |
1943 | if (path->slots[0] >= nritems) { | |
1944 | ret = btrfs_next_leaf(root, path); | |
1945 | if (ret) | |
1946 | goto out; | |
1947 | } else { | |
1948 | path->slots[0]++; | |
1949 | } | |
1950 | ||
1951 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1952 | ||
1953 | if (key.type != key_type || key.objectid != dirid) { | |
1954 | ret = 1; | |
1955 | goto out; | |
1956 | } | |
1957 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1958 | struct btrfs_dir_log_item); | |
1959 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1960 | *start_ret = key.offset; | |
1961 | *end_ret = found_end; | |
1962 | ret = 0; | |
1963 | out: | |
1964 | btrfs_release_path(path); | |
1965 | return ret; | |
1966 | } | |
1967 | ||
1968 | /* | |
1969 | * this looks for a given directory item in the log. If the directory | |
1970 | * item is not in the log, the item is removed and the inode it points | |
1971 | * to is unlinked | |
1972 | */ | |
1973 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
1974 | struct btrfs_root *root, | |
1975 | struct btrfs_root *log, | |
1976 | struct btrfs_path *path, | |
1977 | struct btrfs_path *log_path, | |
1978 | struct inode *dir, | |
1979 | struct btrfs_key *dir_key) | |
1980 | { | |
1981 | int ret; | |
1982 | struct extent_buffer *eb; | |
1983 | int slot; | |
1984 | u32 item_size; | |
1985 | struct btrfs_dir_item *di; | |
1986 | struct btrfs_dir_item *log_di; | |
1987 | int name_len; | |
1988 | unsigned long ptr; | |
1989 | unsigned long ptr_end; | |
1990 | char *name; | |
1991 | struct inode *inode; | |
1992 | struct btrfs_key location; | |
1993 | ||
1994 | again: | |
1995 | eb = path->nodes[0]; | |
1996 | slot = path->slots[0]; | |
1997 | item_size = btrfs_item_size_nr(eb, slot); | |
1998 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1999 | ptr_end = ptr + item_size; | |
2000 | while (ptr < ptr_end) { | |
2001 | di = (struct btrfs_dir_item *)ptr; | |
2002 | if (verify_dir_item(root, eb, di)) { | |
2003 | ret = -EIO; | |
2004 | goto out; | |
2005 | } | |
2006 | ||
2007 | name_len = btrfs_dir_name_len(eb, di); | |
2008 | name = kmalloc(name_len, GFP_NOFS); | |
2009 | if (!name) { | |
2010 | ret = -ENOMEM; | |
2011 | goto out; | |
2012 | } | |
2013 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
2014 | name_len); | |
2015 | log_di = NULL; | |
2016 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { | |
2017 | log_di = btrfs_lookup_dir_item(trans, log, log_path, | |
2018 | dir_key->objectid, | |
2019 | name, name_len, 0); | |
2020 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { | |
2021 | log_di = btrfs_lookup_dir_index_item(trans, log, | |
2022 | log_path, | |
2023 | dir_key->objectid, | |
2024 | dir_key->offset, | |
2025 | name, name_len, 0); | |
2026 | } | |
2027 | if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) { | |
2028 | btrfs_dir_item_key_to_cpu(eb, di, &location); | |
2029 | btrfs_release_path(path); | |
2030 | btrfs_release_path(log_path); | |
2031 | inode = read_one_inode(root, location.objectid); | |
2032 | if (!inode) { | |
2033 | kfree(name); | |
2034 | return -EIO; | |
2035 | } | |
2036 | ||
2037 | ret = link_to_fixup_dir(trans, root, | |
2038 | path, location.objectid); | |
2039 | if (ret) { | |
2040 | kfree(name); | |
2041 | iput(inode); | |
2042 | goto out; | |
2043 | } | |
2044 | ||
2045 | inc_nlink(inode); | |
2046 | ret = btrfs_unlink_inode(trans, root, dir, inode, | |
2047 | name, name_len); | |
2048 | if (!ret) | |
2049 | ret = btrfs_run_delayed_items(trans, root); | |
2050 | kfree(name); | |
2051 | iput(inode); | |
2052 | if (ret) | |
2053 | goto out; | |
2054 | ||
2055 | /* there might still be more names under this key | |
2056 | * check and repeat if required | |
2057 | */ | |
2058 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
2059 | 0, 0); | |
2060 | if (ret == 0) | |
2061 | goto again; | |
2062 | ret = 0; | |
2063 | goto out; | |
2064 | } else if (IS_ERR(log_di)) { | |
2065 | kfree(name); | |
2066 | return PTR_ERR(log_di); | |
2067 | } | |
2068 | btrfs_release_path(log_path); | |
2069 | kfree(name); | |
2070 | ||
2071 | ptr = (unsigned long)(di + 1); | |
2072 | ptr += name_len; | |
2073 | } | |
2074 | ret = 0; | |
2075 | out: | |
2076 | btrfs_release_path(path); | |
2077 | btrfs_release_path(log_path); | |
2078 | return ret; | |
2079 | } | |
2080 | ||
2081 | static int replay_xattr_deletes(struct btrfs_trans_handle *trans, | |
2082 | struct btrfs_root *root, | |
2083 | struct btrfs_root *log, | |
2084 | struct btrfs_path *path, | |
2085 | const u64 ino) | |
2086 | { | |
2087 | struct btrfs_key search_key; | |
2088 | struct btrfs_path *log_path; | |
2089 | int i; | |
2090 | int nritems; | |
2091 | int ret; | |
2092 | ||
2093 | log_path = btrfs_alloc_path(); | |
2094 | if (!log_path) | |
2095 | return -ENOMEM; | |
2096 | ||
2097 | search_key.objectid = ino; | |
2098 | search_key.type = BTRFS_XATTR_ITEM_KEY; | |
2099 | search_key.offset = 0; | |
2100 | again: | |
2101 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
2102 | if (ret < 0) | |
2103 | goto out; | |
2104 | process_leaf: | |
2105 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2106 | for (i = path->slots[0]; i < nritems; i++) { | |
2107 | struct btrfs_key key; | |
2108 | struct btrfs_dir_item *di; | |
2109 | struct btrfs_dir_item *log_di; | |
2110 | u32 total_size; | |
2111 | u32 cur; | |
2112 | ||
2113 | btrfs_item_key_to_cpu(path->nodes[0], &key, i); | |
2114 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) { | |
2115 | ret = 0; | |
2116 | goto out; | |
2117 | } | |
2118 | ||
2119 | di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item); | |
2120 | total_size = btrfs_item_size_nr(path->nodes[0], i); | |
2121 | cur = 0; | |
2122 | while (cur < total_size) { | |
2123 | u16 name_len = btrfs_dir_name_len(path->nodes[0], di); | |
2124 | u16 data_len = btrfs_dir_data_len(path->nodes[0], di); | |
2125 | u32 this_len = sizeof(*di) + name_len + data_len; | |
2126 | char *name; | |
2127 | ||
2128 | name = kmalloc(name_len, GFP_NOFS); | |
2129 | if (!name) { | |
2130 | ret = -ENOMEM; | |
2131 | goto out; | |
2132 | } | |
2133 | read_extent_buffer(path->nodes[0], name, | |
2134 | (unsigned long)(di + 1), name_len); | |
2135 | ||
2136 | log_di = btrfs_lookup_xattr(NULL, log, log_path, ino, | |
2137 | name, name_len, 0); | |
2138 | btrfs_release_path(log_path); | |
2139 | if (!log_di) { | |
2140 | /* Doesn't exist in log tree, so delete it. */ | |
2141 | btrfs_release_path(path); | |
2142 | di = btrfs_lookup_xattr(trans, root, path, ino, | |
2143 | name, name_len, -1); | |
2144 | kfree(name); | |
2145 | if (IS_ERR(di)) { | |
2146 | ret = PTR_ERR(di); | |
2147 | goto out; | |
2148 | } | |
2149 | ASSERT(di); | |
2150 | ret = btrfs_delete_one_dir_name(trans, root, | |
2151 | path, di); | |
2152 | if (ret) | |
2153 | goto out; | |
2154 | btrfs_release_path(path); | |
2155 | search_key = key; | |
2156 | goto again; | |
2157 | } | |
2158 | kfree(name); | |
2159 | if (IS_ERR(log_di)) { | |
2160 | ret = PTR_ERR(log_di); | |
2161 | goto out; | |
2162 | } | |
2163 | cur += this_len; | |
2164 | di = (struct btrfs_dir_item *)((char *)di + this_len); | |
2165 | } | |
2166 | } | |
2167 | ret = btrfs_next_leaf(root, path); | |
2168 | if (ret > 0) | |
2169 | ret = 0; | |
2170 | else if (ret == 0) | |
2171 | goto process_leaf; | |
2172 | out: | |
2173 | btrfs_free_path(log_path); | |
2174 | btrfs_release_path(path); | |
2175 | return ret; | |
2176 | } | |
2177 | ||
2178 | ||
2179 | /* | |
2180 | * deletion replay happens before we copy any new directory items | |
2181 | * out of the log or out of backreferences from inodes. It | |
2182 | * scans the log to find ranges of keys that log is authoritative for, | |
2183 | * and then scans the directory to find items in those ranges that are | |
2184 | * not present in the log. | |
2185 | * | |
2186 | * Anything we don't find in the log is unlinked and removed from the | |
2187 | * directory. | |
2188 | */ | |
2189 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
2190 | struct btrfs_root *root, | |
2191 | struct btrfs_root *log, | |
2192 | struct btrfs_path *path, | |
2193 | u64 dirid, int del_all) | |
2194 | { | |
2195 | u64 range_start; | |
2196 | u64 range_end; | |
2197 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
2198 | int ret = 0; | |
2199 | struct btrfs_key dir_key; | |
2200 | struct btrfs_key found_key; | |
2201 | struct btrfs_path *log_path; | |
2202 | struct inode *dir; | |
2203 | ||
2204 | dir_key.objectid = dirid; | |
2205 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
2206 | log_path = btrfs_alloc_path(); | |
2207 | if (!log_path) | |
2208 | return -ENOMEM; | |
2209 | ||
2210 | dir = read_one_inode(root, dirid); | |
2211 | /* it isn't an error if the inode isn't there, that can happen | |
2212 | * because we replay the deletes before we copy in the inode item | |
2213 | * from the log | |
2214 | */ | |
2215 | if (!dir) { | |
2216 | btrfs_free_path(log_path); | |
2217 | return 0; | |
2218 | } | |
2219 | again: | |
2220 | range_start = 0; | |
2221 | range_end = 0; | |
2222 | while (1) { | |
2223 | if (del_all) | |
2224 | range_end = (u64)-1; | |
2225 | else { | |
2226 | ret = find_dir_range(log, path, dirid, key_type, | |
2227 | &range_start, &range_end); | |
2228 | if (ret != 0) | |
2229 | break; | |
2230 | } | |
2231 | ||
2232 | dir_key.offset = range_start; | |
2233 | while (1) { | |
2234 | int nritems; | |
2235 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
2236 | 0, 0); | |
2237 | if (ret < 0) | |
2238 | goto out; | |
2239 | ||
2240 | nritems = btrfs_header_nritems(path->nodes[0]); | |
2241 | if (path->slots[0] >= nritems) { | |
2242 | ret = btrfs_next_leaf(root, path); | |
2243 | if (ret) | |
2244 | break; | |
2245 | } | |
2246 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2247 | path->slots[0]); | |
2248 | if (found_key.objectid != dirid || | |
2249 | found_key.type != dir_key.type) | |
2250 | goto next_type; | |
2251 | ||
2252 | if (found_key.offset > range_end) | |
2253 | break; | |
2254 | ||
2255 | ret = check_item_in_log(trans, root, log, path, | |
2256 | log_path, dir, | |
2257 | &found_key); | |
2258 | if (ret) | |
2259 | goto out; | |
2260 | if (found_key.offset == (u64)-1) | |
2261 | break; | |
2262 | dir_key.offset = found_key.offset + 1; | |
2263 | } | |
2264 | btrfs_release_path(path); | |
2265 | if (range_end == (u64)-1) | |
2266 | break; | |
2267 | range_start = range_end + 1; | |
2268 | } | |
2269 | ||
2270 | next_type: | |
2271 | ret = 0; | |
2272 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
2273 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2274 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
2275 | btrfs_release_path(path); | |
2276 | goto again; | |
2277 | } | |
2278 | out: | |
2279 | btrfs_release_path(path); | |
2280 | btrfs_free_path(log_path); | |
2281 | iput(dir); | |
2282 | return ret; | |
2283 | } | |
2284 | ||
2285 | /* | |
2286 | * the process_func used to replay items from the log tree. This | |
2287 | * gets called in two different stages. The first stage just looks | |
2288 | * for inodes and makes sure they are all copied into the subvolume. | |
2289 | * | |
2290 | * The second stage copies all the other item types from the log into | |
2291 | * the subvolume. The two stage approach is slower, but gets rid of | |
2292 | * lots of complexity around inodes referencing other inodes that exist | |
2293 | * only in the log (references come from either directory items or inode | |
2294 | * back refs). | |
2295 | */ | |
2296 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
2297 | struct walk_control *wc, u64 gen) | |
2298 | { | |
2299 | int nritems; | |
2300 | struct btrfs_path *path; | |
2301 | struct btrfs_root *root = wc->replay_dest; | |
2302 | struct btrfs_key key; | |
2303 | int level; | |
2304 | int i; | |
2305 | int ret; | |
2306 | ||
2307 | ret = btrfs_read_buffer(eb, gen); | |
2308 | if (ret) | |
2309 | return ret; | |
2310 | ||
2311 | level = btrfs_header_level(eb); | |
2312 | ||
2313 | if (level != 0) | |
2314 | return 0; | |
2315 | ||
2316 | path = btrfs_alloc_path(); | |
2317 | if (!path) | |
2318 | return -ENOMEM; | |
2319 | ||
2320 | nritems = btrfs_header_nritems(eb); | |
2321 | for (i = 0; i < nritems; i++) { | |
2322 | btrfs_item_key_to_cpu(eb, &key, i); | |
2323 | ||
2324 | /* inode keys are done during the first stage */ | |
2325 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
2326 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
2327 | struct btrfs_inode_item *inode_item; | |
2328 | u32 mode; | |
2329 | ||
2330 | inode_item = btrfs_item_ptr(eb, i, | |
2331 | struct btrfs_inode_item); | |
2332 | ret = replay_xattr_deletes(wc->trans, root, log, | |
2333 | path, key.objectid); | |
2334 | if (ret) | |
2335 | break; | |
2336 | mode = btrfs_inode_mode(eb, inode_item); | |
2337 | if (S_ISDIR(mode)) { | |
2338 | ret = replay_dir_deletes(wc->trans, | |
2339 | root, log, path, key.objectid, 0); | |
2340 | if (ret) | |
2341 | break; | |
2342 | } | |
2343 | ret = overwrite_item(wc->trans, root, path, | |
2344 | eb, i, &key); | |
2345 | if (ret) | |
2346 | break; | |
2347 | ||
2348 | /* for regular files, make sure corresponding | |
2349 | * orphan item exist. extents past the new EOF | |
2350 | * will be truncated later by orphan cleanup. | |
2351 | */ | |
2352 | if (S_ISREG(mode)) { | |
2353 | ret = insert_orphan_item(wc->trans, root, | |
2354 | key.objectid); | |
2355 | if (ret) | |
2356 | break; | |
2357 | } | |
2358 | ||
2359 | ret = link_to_fixup_dir(wc->trans, root, | |
2360 | path, key.objectid); | |
2361 | if (ret) | |
2362 | break; | |
2363 | } | |
2364 | ||
2365 | if (key.type == BTRFS_DIR_INDEX_KEY && | |
2366 | wc->stage == LOG_WALK_REPLAY_DIR_INDEX) { | |
2367 | ret = replay_one_dir_item(wc->trans, root, path, | |
2368 | eb, i, &key); | |
2369 | if (ret) | |
2370 | break; | |
2371 | } | |
2372 | ||
2373 | if (wc->stage < LOG_WALK_REPLAY_ALL) | |
2374 | continue; | |
2375 | ||
2376 | /* these keys are simply copied */ | |
2377 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
2378 | ret = overwrite_item(wc->trans, root, path, | |
2379 | eb, i, &key); | |
2380 | if (ret) | |
2381 | break; | |
2382 | } else if (key.type == BTRFS_INODE_REF_KEY || | |
2383 | key.type == BTRFS_INODE_EXTREF_KEY) { | |
2384 | ret = add_inode_ref(wc->trans, root, log, path, | |
2385 | eb, i, &key); | |
2386 | if (ret && ret != -ENOENT) | |
2387 | break; | |
2388 | ret = 0; | |
2389 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { | |
2390 | ret = replay_one_extent(wc->trans, root, path, | |
2391 | eb, i, &key); | |
2392 | if (ret) | |
2393 | break; | |
2394 | } else if (key.type == BTRFS_DIR_ITEM_KEY) { | |
2395 | ret = replay_one_dir_item(wc->trans, root, path, | |
2396 | eb, i, &key); | |
2397 | if (ret) | |
2398 | break; | |
2399 | } | |
2400 | } | |
2401 | btrfs_free_path(path); | |
2402 | return ret; | |
2403 | } | |
2404 | ||
2405 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, | |
2406 | struct btrfs_root *root, | |
2407 | struct btrfs_path *path, int *level, | |
2408 | struct walk_control *wc) | |
2409 | { | |
2410 | u64 root_owner; | |
2411 | u64 bytenr; | |
2412 | u64 ptr_gen; | |
2413 | struct extent_buffer *next; | |
2414 | struct extent_buffer *cur; | |
2415 | struct extent_buffer *parent; | |
2416 | u32 blocksize; | |
2417 | int ret = 0; | |
2418 | ||
2419 | WARN_ON(*level < 0); | |
2420 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2421 | ||
2422 | while (*level > 0) { | |
2423 | WARN_ON(*level < 0); | |
2424 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2425 | cur = path->nodes[*level]; | |
2426 | ||
2427 | WARN_ON(btrfs_header_level(cur) != *level); | |
2428 | ||
2429 | if (path->slots[*level] >= | |
2430 | btrfs_header_nritems(cur)) | |
2431 | break; | |
2432 | ||
2433 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2434 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
2435 | blocksize = root->nodesize; | |
2436 | ||
2437 | parent = path->nodes[*level]; | |
2438 | root_owner = btrfs_header_owner(parent); | |
2439 | ||
2440 | next = btrfs_find_create_tree_block(root, bytenr); | |
2441 | if (IS_ERR(next)) | |
2442 | return PTR_ERR(next); | |
2443 | ||
2444 | if (*level == 1) { | |
2445 | ret = wc->process_func(root, next, wc, ptr_gen); | |
2446 | if (ret) { | |
2447 | free_extent_buffer(next); | |
2448 | return ret; | |
2449 | } | |
2450 | ||
2451 | path->slots[*level]++; | |
2452 | if (wc->free) { | |
2453 | ret = btrfs_read_buffer(next, ptr_gen); | |
2454 | if (ret) { | |
2455 | free_extent_buffer(next); | |
2456 | return ret; | |
2457 | } | |
2458 | ||
2459 | if (trans) { | |
2460 | btrfs_tree_lock(next); | |
2461 | btrfs_set_lock_blocking(next); | |
2462 | clean_tree_block(trans, root->fs_info, | |
2463 | next); | |
2464 | btrfs_wait_tree_block_writeback(next); | |
2465 | btrfs_tree_unlock(next); | |
2466 | } | |
2467 | ||
2468 | WARN_ON(root_owner != | |
2469 | BTRFS_TREE_LOG_OBJECTID); | |
2470 | ret = btrfs_free_and_pin_reserved_extent(root, | |
2471 | bytenr, blocksize); | |
2472 | if (ret) { | |
2473 | free_extent_buffer(next); | |
2474 | return ret; | |
2475 | } | |
2476 | } | |
2477 | free_extent_buffer(next); | |
2478 | continue; | |
2479 | } | |
2480 | ret = btrfs_read_buffer(next, ptr_gen); | |
2481 | if (ret) { | |
2482 | free_extent_buffer(next); | |
2483 | return ret; | |
2484 | } | |
2485 | ||
2486 | WARN_ON(*level <= 0); | |
2487 | if (path->nodes[*level-1]) | |
2488 | free_extent_buffer(path->nodes[*level-1]); | |
2489 | path->nodes[*level-1] = next; | |
2490 | *level = btrfs_header_level(next); | |
2491 | path->slots[*level] = 0; | |
2492 | cond_resched(); | |
2493 | } | |
2494 | WARN_ON(*level < 0); | |
2495 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2496 | ||
2497 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); | |
2498 | ||
2499 | cond_resched(); | |
2500 | return 0; | |
2501 | } | |
2502 | ||
2503 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, | |
2504 | struct btrfs_root *root, | |
2505 | struct btrfs_path *path, int *level, | |
2506 | struct walk_control *wc) | |
2507 | { | |
2508 | u64 root_owner; | |
2509 | int i; | |
2510 | int slot; | |
2511 | int ret; | |
2512 | ||
2513 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { | |
2514 | slot = path->slots[i]; | |
2515 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { | |
2516 | path->slots[i]++; | |
2517 | *level = i; | |
2518 | WARN_ON(*level == 0); | |
2519 | return 0; | |
2520 | } else { | |
2521 | struct extent_buffer *parent; | |
2522 | if (path->nodes[*level] == root->node) | |
2523 | parent = path->nodes[*level]; | |
2524 | else | |
2525 | parent = path->nodes[*level + 1]; | |
2526 | ||
2527 | root_owner = btrfs_header_owner(parent); | |
2528 | ret = wc->process_func(root, path->nodes[*level], wc, | |
2529 | btrfs_header_generation(path->nodes[*level])); | |
2530 | if (ret) | |
2531 | return ret; | |
2532 | ||
2533 | if (wc->free) { | |
2534 | struct extent_buffer *next; | |
2535 | ||
2536 | next = path->nodes[*level]; | |
2537 | ||
2538 | if (trans) { | |
2539 | btrfs_tree_lock(next); | |
2540 | btrfs_set_lock_blocking(next); | |
2541 | clean_tree_block(trans, root->fs_info, | |
2542 | next); | |
2543 | btrfs_wait_tree_block_writeback(next); | |
2544 | btrfs_tree_unlock(next); | |
2545 | } | |
2546 | ||
2547 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); | |
2548 | ret = btrfs_free_and_pin_reserved_extent(root, | |
2549 | path->nodes[*level]->start, | |
2550 | path->nodes[*level]->len); | |
2551 | if (ret) | |
2552 | return ret; | |
2553 | } | |
2554 | free_extent_buffer(path->nodes[*level]); | |
2555 | path->nodes[*level] = NULL; | |
2556 | *level = i + 1; | |
2557 | } | |
2558 | } | |
2559 | return 1; | |
2560 | } | |
2561 | ||
2562 | /* | |
2563 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2564 | * the tree freeing any blocks that have a ref count of zero after being | |
2565 | * decremented. | |
2566 | */ | |
2567 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
2568 | struct btrfs_root *log, struct walk_control *wc) | |
2569 | { | |
2570 | int ret = 0; | |
2571 | int wret; | |
2572 | int level; | |
2573 | struct btrfs_path *path; | |
2574 | int orig_level; | |
2575 | ||
2576 | path = btrfs_alloc_path(); | |
2577 | if (!path) | |
2578 | return -ENOMEM; | |
2579 | ||
2580 | level = btrfs_header_level(log->node); | |
2581 | orig_level = level; | |
2582 | path->nodes[level] = log->node; | |
2583 | extent_buffer_get(log->node); | |
2584 | path->slots[level] = 0; | |
2585 | ||
2586 | while (1) { | |
2587 | wret = walk_down_log_tree(trans, log, path, &level, wc); | |
2588 | if (wret > 0) | |
2589 | break; | |
2590 | if (wret < 0) { | |
2591 | ret = wret; | |
2592 | goto out; | |
2593 | } | |
2594 | ||
2595 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
2596 | if (wret > 0) | |
2597 | break; | |
2598 | if (wret < 0) { | |
2599 | ret = wret; | |
2600 | goto out; | |
2601 | } | |
2602 | } | |
2603 | ||
2604 | /* was the root node processed? if not, catch it here */ | |
2605 | if (path->nodes[orig_level]) { | |
2606 | ret = wc->process_func(log, path->nodes[orig_level], wc, | |
2607 | btrfs_header_generation(path->nodes[orig_level])); | |
2608 | if (ret) | |
2609 | goto out; | |
2610 | if (wc->free) { | |
2611 | struct extent_buffer *next; | |
2612 | ||
2613 | next = path->nodes[orig_level]; | |
2614 | ||
2615 | if (trans) { | |
2616 | btrfs_tree_lock(next); | |
2617 | btrfs_set_lock_blocking(next); | |
2618 | clean_tree_block(trans, log->fs_info, next); | |
2619 | btrfs_wait_tree_block_writeback(next); | |
2620 | btrfs_tree_unlock(next); | |
2621 | } | |
2622 | ||
2623 | WARN_ON(log->root_key.objectid != | |
2624 | BTRFS_TREE_LOG_OBJECTID); | |
2625 | ret = btrfs_free_and_pin_reserved_extent(log, next->start, | |
2626 | next->len); | |
2627 | if (ret) | |
2628 | goto out; | |
2629 | } | |
2630 | } | |
2631 | ||
2632 | out: | |
2633 | btrfs_free_path(path); | |
2634 | return ret; | |
2635 | } | |
2636 | ||
2637 | /* | |
2638 | * helper function to update the item for a given subvolumes log root | |
2639 | * in the tree of log roots | |
2640 | */ | |
2641 | static int update_log_root(struct btrfs_trans_handle *trans, | |
2642 | struct btrfs_root *log) | |
2643 | { | |
2644 | int ret; | |
2645 | ||
2646 | if (log->log_transid == 1) { | |
2647 | /* insert root item on the first sync */ | |
2648 | ret = btrfs_insert_root(trans, log->fs_info->log_root_tree, | |
2649 | &log->root_key, &log->root_item); | |
2650 | } else { | |
2651 | ret = btrfs_update_root(trans, log->fs_info->log_root_tree, | |
2652 | &log->root_key, &log->root_item); | |
2653 | } | |
2654 | return ret; | |
2655 | } | |
2656 | ||
2657 | static void wait_log_commit(struct btrfs_root *root, int transid) | |
2658 | { | |
2659 | DEFINE_WAIT(wait); | |
2660 | int index = transid % 2; | |
2661 | ||
2662 | /* | |
2663 | * we only allow two pending log transactions at a time, | |
2664 | * so we know that if ours is more than 2 older than the | |
2665 | * current transaction, we're done | |
2666 | */ | |
2667 | do { | |
2668 | prepare_to_wait(&root->log_commit_wait[index], | |
2669 | &wait, TASK_UNINTERRUPTIBLE); | |
2670 | mutex_unlock(&root->log_mutex); | |
2671 | ||
2672 | if (root->log_transid_committed < transid && | |
2673 | atomic_read(&root->log_commit[index])) | |
2674 | schedule(); | |
2675 | ||
2676 | finish_wait(&root->log_commit_wait[index], &wait); | |
2677 | mutex_lock(&root->log_mutex); | |
2678 | } while (root->log_transid_committed < transid && | |
2679 | atomic_read(&root->log_commit[index])); | |
2680 | } | |
2681 | ||
2682 | static void wait_for_writer(struct btrfs_root *root) | |
2683 | { | |
2684 | DEFINE_WAIT(wait); | |
2685 | ||
2686 | while (atomic_read(&root->log_writers)) { | |
2687 | prepare_to_wait(&root->log_writer_wait, | |
2688 | &wait, TASK_UNINTERRUPTIBLE); | |
2689 | mutex_unlock(&root->log_mutex); | |
2690 | if (atomic_read(&root->log_writers)) | |
2691 | schedule(); | |
2692 | finish_wait(&root->log_writer_wait, &wait); | |
2693 | mutex_lock(&root->log_mutex); | |
2694 | } | |
2695 | } | |
2696 | ||
2697 | static inline void btrfs_remove_log_ctx(struct btrfs_root *root, | |
2698 | struct btrfs_log_ctx *ctx) | |
2699 | { | |
2700 | if (!ctx) | |
2701 | return; | |
2702 | ||
2703 | mutex_lock(&root->log_mutex); | |
2704 | list_del_init(&ctx->list); | |
2705 | mutex_unlock(&root->log_mutex); | |
2706 | } | |
2707 | ||
2708 | /* | |
2709 | * Invoked in log mutex context, or be sure there is no other task which | |
2710 | * can access the list. | |
2711 | */ | |
2712 | static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root, | |
2713 | int index, int error) | |
2714 | { | |
2715 | struct btrfs_log_ctx *ctx; | |
2716 | ||
2717 | if (!error) { | |
2718 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2719 | return; | |
2720 | } | |
2721 | ||
2722 | list_for_each_entry(ctx, &root->log_ctxs[index], list) | |
2723 | ctx->log_ret = error; | |
2724 | ||
2725 | INIT_LIST_HEAD(&root->log_ctxs[index]); | |
2726 | } | |
2727 | ||
2728 | /* | |
2729 | * btrfs_sync_log does sends a given tree log down to the disk and | |
2730 | * updates the super blocks to record it. When this call is done, | |
2731 | * you know that any inodes previously logged are safely on disk only | |
2732 | * if it returns 0. | |
2733 | * | |
2734 | * Any other return value means you need to call btrfs_commit_transaction. | |
2735 | * Some of the edge cases for fsyncing directories that have had unlinks | |
2736 | * or renames done in the past mean that sometimes the only safe | |
2737 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
2738 | * that has happened. | |
2739 | */ | |
2740 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
2741 | struct btrfs_root *root, struct btrfs_log_ctx *ctx) | |
2742 | { | |
2743 | int index1; | |
2744 | int index2; | |
2745 | int mark; | |
2746 | int ret; | |
2747 | struct btrfs_root *log = root->log_root; | |
2748 | struct btrfs_root *log_root_tree = root->fs_info->log_root_tree; | |
2749 | int log_transid = 0; | |
2750 | struct btrfs_log_ctx root_log_ctx; | |
2751 | struct blk_plug plug; | |
2752 | ||
2753 | mutex_lock(&root->log_mutex); | |
2754 | log_transid = ctx->log_transid; | |
2755 | if (root->log_transid_committed >= log_transid) { | |
2756 | mutex_unlock(&root->log_mutex); | |
2757 | return ctx->log_ret; | |
2758 | } | |
2759 | ||
2760 | index1 = log_transid % 2; | |
2761 | if (atomic_read(&root->log_commit[index1])) { | |
2762 | wait_log_commit(root, log_transid); | |
2763 | mutex_unlock(&root->log_mutex); | |
2764 | return ctx->log_ret; | |
2765 | } | |
2766 | ASSERT(log_transid == root->log_transid); | |
2767 | atomic_set(&root->log_commit[index1], 1); | |
2768 | ||
2769 | /* wait for previous tree log sync to complete */ | |
2770 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
2771 | wait_log_commit(root, log_transid - 1); | |
2772 | ||
2773 | while (1) { | |
2774 | int batch = atomic_read(&root->log_batch); | |
2775 | /* when we're on an ssd, just kick the log commit out */ | |
2776 | if (!btrfs_test_opt(root->fs_info, SSD) && | |
2777 | test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) { | |
2778 | mutex_unlock(&root->log_mutex); | |
2779 | schedule_timeout_uninterruptible(1); | |
2780 | mutex_lock(&root->log_mutex); | |
2781 | } | |
2782 | wait_for_writer(root); | |
2783 | if (batch == atomic_read(&root->log_batch)) | |
2784 | break; | |
2785 | } | |
2786 | ||
2787 | /* bail out if we need to do a full commit */ | |
2788 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { | |
2789 | ret = -EAGAIN; | |
2790 | btrfs_free_logged_extents(log, log_transid); | |
2791 | mutex_unlock(&root->log_mutex); | |
2792 | goto out; | |
2793 | } | |
2794 | ||
2795 | if (log_transid % 2 == 0) | |
2796 | mark = EXTENT_DIRTY; | |
2797 | else | |
2798 | mark = EXTENT_NEW; | |
2799 | ||
2800 | /* we start IO on all the marked extents here, but we don't actually | |
2801 | * wait for them until later. | |
2802 | */ | |
2803 | blk_start_plug(&plug); | |
2804 | ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark); | |
2805 | if (ret) { | |
2806 | blk_finish_plug(&plug); | |
2807 | btrfs_abort_transaction(trans, ret); | |
2808 | btrfs_free_logged_extents(log, log_transid); | |
2809 | btrfs_set_log_full_commit(root->fs_info, trans); | |
2810 | mutex_unlock(&root->log_mutex); | |
2811 | goto out; | |
2812 | } | |
2813 | ||
2814 | btrfs_set_root_node(&log->root_item, log->node); | |
2815 | ||
2816 | root->log_transid++; | |
2817 | log->log_transid = root->log_transid; | |
2818 | root->log_start_pid = 0; | |
2819 | /* | |
2820 | * IO has been started, blocks of the log tree have WRITTEN flag set | |
2821 | * in their headers. new modifications of the log will be written to | |
2822 | * new positions. so it's safe to allow log writers to go in. | |
2823 | */ | |
2824 | mutex_unlock(&root->log_mutex); | |
2825 | ||
2826 | btrfs_init_log_ctx(&root_log_ctx, NULL); | |
2827 | ||
2828 | mutex_lock(&log_root_tree->log_mutex); | |
2829 | atomic_inc(&log_root_tree->log_batch); | |
2830 | atomic_inc(&log_root_tree->log_writers); | |
2831 | ||
2832 | index2 = log_root_tree->log_transid % 2; | |
2833 | list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]); | |
2834 | root_log_ctx.log_transid = log_root_tree->log_transid; | |
2835 | ||
2836 | mutex_unlock(&log_root_tree->log_mutex); | |
2837 | ||
2838 | ret = update_log_root(trans, log); | |
2839 | ||
2840 | mutex_lock(&log_root_tree->log_mutex); | |
2841 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | |
2842 | /* | |
2843 | * Implicit memory barrier after atomic_dec_and_test | |
2844 | */ | |
2845 | if (waitqueue_active(&log_root_tree->log_writer_wait)) | |
2846 | wake_up(&log_root_tree->log_writer_wait); | |
2847 | } | |
2848 | ||
2849 | if (ret) { | |
2850 | if (!list_empty(&root_log_ctx.list)) | |
2851 | list_del_init(&root_log_ctx.list); | |
2852 | ||
2853 | blk_finish_plug(&plug); | |
2854 | btrfs_set_log_full_commit(root->fs_info, trans); | |
2855 | ||
2856 | if (ret != -ENOSPC) { | |
2857 | btrfs_abort_transaction(trans, ret); | |
2858 | mutex_unlock(&log_root_tree->log_mutex); | |
2859 | goto out; | |
2860 | } | |
2861 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); | |
2862 | btrfs_free_logged_extents(log, log_transid); | |
2863 | mutex_unlock(&log_root_tree->log_mutex); | |
2864 | ret = -EAGAIN; | |
2865 | goto out; | |
2866 | } | |
2867 | ||
2868 | if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) { | |
2869 | blk_finish_plug(&plug); | |
2870 | list_del_init(&root_log_ctx.list); | |
2871 | mutex_unlock(&log_root_tree->log_mutex); | |
2872 | ret = root_log_ctx.log_ret; | |
2873 | goto out; | |
2874 | } | |
2875 | ||
2876 | index2 = root_log_ctx.log_transid % 2; | |
2877 | if (atomic_read(&log_root_tree->log_commit[index2])) { | |
2878 | blk_finish_plug(&plug); | |
2879 | ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, | |
2880 | mark); | |
2881 | btrfs_wait_logged_extents(trans, log, log_transid); | |
2882 | wait_log_commit(log_root_tree, | |
2883 | root_log_ctx.log_transid); | |
2884 | mutex_unlock(&log_root_tree->log_mutex); | |
2885 | if (!ret) | |
2886 | ret = root_log_ctx.log_ret; | |
2887 | goto out; | |
2888 | } | |
2889 | ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid); | |
2890 | atomic_set(&log_root_tree->log_commit[index2], 1); | |
2891 | ||
2892 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { | |
2893 | wait_log_commit(log_root_tree, | |
2894 | root_log_ctx.log_transid - 1); | |
2895 | } | |
2896 | ||
2897 | wait_for_writer(log_root_tree); | |
2898 | ||
2899 | /* | |
2900 | * now that we've moved on to the tree of log tree roots, | |
2901 | * check the full commit flag again | |
2902 | */ | |
2903 | if (btrfs_need_log_full_commit(root->fs_info, trans)) { | |
2904 | blk_finish_plug(&plug); | |
2905 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); | |
2906 | btrfs_free_logged_extents(log, log_transid); | |
2907 | mutex_unlock(&log_root_tree->log_mutex); | |
2908 | ret = -EAGAIN; | |
2909 | goto out_wake_log_root; | |
2910 | } | |
2911 | ||
2912 | ret = btrfs_write_marked_extents(log_root_tree, | |
2913 | &log_root_tree->dirty_log_pages, | |
2914 | EXTENT_DIRTY | EXTENT_NEW); | |
2915 | blk_finish_plug(&plug); | |
2916 | if (ret) { | |
2917 | btrfs_set_log_full_commit(root->fs_info, trans); | |
2918 | btrfs_abort_transaction(trans, ret); | |
2919 | btrfs_free_logged_extents(log, log_transid); | |
2920 | mutex_unlock(&log_root_tree->log_mutex); | |
2921 | goto out_wake_log_root; | |
2922 | } | |
2923 | ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); | |
2924 | if (!ret) | |
2925 | ret = btrfs_wait_marked_extents(log_root_tree, | |
2926 | &log_root_tree->dirty_log_pages, | |
2927 | EXTENT_NEW | EXTENT_DIRTY); | |
2928 | if (ret) { | |
2929 | btrfs_set_log_full_commit(root->fs_info, trans); | |
2930 | btrfs_free_logged_extents(log, log_transid); | |
2931 | mutex_unlock(&log_root_tree->log_mutex); | |
2932 | goto out_wake_log_root; | |
2933 | } | |
2934 | btrfs_wait_logged_extents(trans, log, log_transid); | |
2935 | ||
2936 | btrfs_set_super_log_root(root->fs_info->super_for_commit, | |
2937 | log_root_tree->node->start); | |
2938 | btrfs_set_super_log_root_level(root->fs_info->super_for_commit, | |
2939 | btrfs_header_level(log_root_tree->node)); | |
2940 | ||
2941 | log_root_tree->log_transid++; | |
2942 | mutex_unlock(&log_root_tree->log_mutex); | |
2943 | ||
2944 | /* | |
2945 | * nobody else is going to jump in and write the the ctree | |
2946 | * super here because the log_commit atomic below is protecting | |
2947 | * us. We must be called with a transaction handle pinning | |
2948 | * the running transaction open, so a full commit can't hop | |
2949 | * in and cause problems either. | |
2950 | */ | |
2951 | ret = write_ctree_super(trans, root->fs_info->tree_root, 1); | |
2952 | if (ret) { | |
2953 | btrfs_set_log_full_commit(root->fs_info, trans); | |
2954 | btrfs_abort_transaction(trans, ret); | |
2955 | goto out_wake_log_root; | |
2956 | } | |
2957 | ||
2958 | mutex_lock(&root->log_mutex); | |
2959 | if (root->last_log_commit < log_transid) | |
2960 | root->last_log_commit = log_transid; | |
2961 | mutex_unlock(&root->log_mutex); | |
2962 | ||
2963 | out_wake_log_root: | |
2964 | /* | |
2965 | * We needn't get log_mutex here because we are sure all | |
2966 | * the other tasks are blocked. | |
2967 | */ | |
2968 | btrfs_remove_all_log_ctxs(log_root_tree, index2, ret); | |
2969 | ||
2970 | mutex_lock(&log_root_tree->log_mutex); | |
2971 | log_root_tree->log_transid_committed++; | |
2972 | atomic_set(&log_root_tree->log_commit[index2], 0); | |
2973 | mutex_unlock(&log_root_tree->log_mutex); | |
2974 | ||
2975 | /* | |
2976 | * The barrier before waitqueue_active is implied by mutex_unlock | |
2977 | */ | |
2978 | if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) | |
2979 | wake_up(&log_root_tree->log_commit_wait[index2]); | |
2980 | out: | |
2981 | /* See above. */ | |
2982 | btrfs_remove_all_log_ctxs(root, index1, ret); | |
2983 | ||
2984 | mutex_lock(&root->log_mutex); | |
2985 | root->log_transid_committed++; | |
2986 | atomic_set(&root->log_commit[index1], 0); | |
2987 | mutex_unlock(&root->log_mutex); | |
2988 | ||
2989 | /* | |
2990 | * The barrier before waitqueue_active is implied by mutex_unlock | |
2991 | */ | |
2992 | if (waitqueue_active(&root->log_commit_wait[index1])) | |
2993 | wake_up(&root->log_commit_wait[index1]); | |
2994 | return ret; | |
2995 | } | |
2996 | ||
2997 | static void free_log_tree(struct btrfs_trans_handle *trans, | |
2998 | struct btrfs_root *log) | |
2999 | { | |
3000 | int ret; | |
3001 | u64 start; | |
3002 | u64 end; | |
3003 | struct walk_control wc = { | |
3004 | .free = 1, | |
3005 | .process_func = process_one_buffer | |
3006 | }; | |
3007 | ||
3008 | ret = walk_log_tree(trans, log, &wc); | |
3009 | /* I don't think this can happen but just in case */ | |
3010 | if (ret) | |
3011 | btrfs_abort_transaction(trans, ret); | |
3012 | ||
3013 | while (1) { | |
3014 | ret = find_first_extent_bit(&log->dirty_log_pages, | |
3015 | 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW, | |
3016 | NULL); | |
3017 | if (ret) | |
3018 | break; | |
3019 | ||
3020 | clear_extent_bits(&log->dirty_log_pages, start, end, | |
3021 | EXTENT_DIRTY | EXTENT_NEW); | |
3022 | } | |
3023 | ||
3024 | /* | |
3025 | * We may have short-circuited the log tree with the full commit logic | |
3026 | * and left ordered extents on our list, so clear these out to keep us | |
3027 | * from leaking inodes and memory. | |
3028 | */ | |
3029 | btrfs_free_logged_extents(log, 0); | |
3030 | btrfs_free_logged_extents(log, 1); | |
3031 | ||
3032 | free_extent_buffer(log->node); | |
3033 | kfree(log); | |
3034 | } | |
3035 | ||
3036 | /* | |
3037 | * free all the extents used by the tree log. This should be called | |
3038 | * at commit time of the full transaction | |
3039 | */ | |
3040 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
3041 | { | |
3042 | if (root->log_root) { | |
3043 | free_log_tree(trans, root->log_root); | |
3044 | root->log_root = NULL; | |
3045 | } | |
3046 | return 0; | |
3047 | } | |
3048 | ||
3049 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |
3050 | struct btrfs_fs_info *fs_info) | |
3051 | { | |
3052 | if (fs_info->log_root_tree) { | |
3053 | free_log_tree(trans, fs_info->log_root_tree); | |
3054 | fs_info->log_root_tree = NULL; | |
3055 | } | |
3056 | return 0; | |
3057 | } | |
3058 | ||
3059 | /* | |
3060 | * If both a file and directory are logged, and unlinks or renames are | |
3061 | * mixed in, we have a few interesting corners: | |
3062 | * | |
3063 | * create file X in dir Y | |
3064 | * link file X to X.link in dir Y | |
3065 | * fsync file X | |
3066 | * unlink file X but leave X.link | |
3067 | * fsync dir Y | |
3068 | * | |
3069 | * After a crash we would expect only X.link to exist. But file X | |
3070 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
3071 | * | |
3072 | * We solve this by removing directory entries and inode backrefs from the | |
3073 | * log when a file that was logged in the current transaction is | |
3074 | * unlinked. Any later fsync will include the updated log entries, and | |
3075 | * we'll be able to reconstruct the proper directory items from backrefs. | |
3076 | * | |
3077 | * This optimizations allows us to avoid relogging the entire inode | |
3078 | * or the entire directory. | |
3079 | */ | |
3080 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
3081 | struct btrfs_root *root, | |
3082 | const char *name, int name_len, | |
3083 | struct inode *dir, u64 index) | |
3084 | { | |
3085 | struct btrfs_root *log; | |
3086 | struct btrfs_dir_item *di; | |
3087 | struct btrfs_path *path; | |
3088 | int ret; | |
3089 | int err = 0; | |
3090 | int bytes_del = 0; | |
3091 | u64 dir_ino = btrfs_ino(dir); | |
3092 | ||
3093 | if (BTRFS_I(dir)->logged_trans < trans->transid) | |
3094 | return 0; | |
3095 | ||
3096 | ret = join_running_log_trans(root); | |
3097 | if (ret) | |
3098 | return 0; | |
3099 | ||
3100 | mutex_lock(&BTRFS_I(dir)->log_mutex); | |
3101 | ||
3102 | log = root->log_root; | |
3103 | path = btrfs_alloc_path(); | |
3104 | if (!path) { | |
3105 | err = -ENOMEM; | |
3106 | goto out_unlock; | |
3107 | } | |
3108 | ||
3109 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, | |
3110 | name, name_len, -1); | |
3111 | if (IS_ERR(di)) { | |
3112 | err = PTR_ERR(di); | |
3113 | goto fail; | |
3114 | } | |
3115 | if (di) { | |
3116 | ret = btrfs_delete_one_dir_name(trans, log, path, di); | |
3117 | bytes_del += name_len; | |
3118 | if (ret) { | |
3119 | err = ret; | |
3120 | goto fail; | |
3121 | } | |
3122 | } | |
3123 | btrfs_release_path(path); | |
3124 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, | |
3125 | index, name, name_len, -1); | |
3126 | if (IS_ERR(di)) { | |
3127 | err = PTR_ERR(di); | |
3128 | goto fail; | |
3129 | } | |
3130 | if (di) { | |
3131 | ret = btrfs_delete_one_dir_name(trans, log, path, di); | |
3132 | bytes_del += name_len; | |
3133 | if (ret) { | |
3134 | err = ret; | |
3135 | goto fail; | |
3136 | } | |
3137 | } | |
3138 | ||
3139 | /* update the directory size in the log to reflect the names | |
3140 | * we have removed | |
3141 | */ | |
3142 | if (bytes_del) { | |
3143 | struct btrfs_key key; | |
3144 | ||
3145 | key.objectid = dir_ino; | |
3146 | key.offset = 0; | |
3147 | key.type = BTRFS_INODE_ITEM_KEY; | |
3148 | btrfs_release_path(path); | |
3149 | ||
3150 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
3151 | if (ret < 0) { | |
3152 | err = ret; | |
3153 | goto fail; | |
3154 | } | |
3155 | if (ret == 0) { | |
3156 | struct btrfs_inode_item *item; | |
3157 | u64 i_size; | |
3158 | ||
3159 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3160 | struct btrfs_inode_item); | |
3161 | i_size = btrfs_inode_size(path->nodes[0], item); | |
3162 | if (i_size > bytes_del) | |
3163 | i_size -= bytes_del; | |
3164 | else | |
3165 | i_size = 0; | |
3166 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
3167 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3168 | } else | |
3169 | ret = 0; | |
3170 | btrfs_release_path(path); | |
3171 | } | |
3172 | fail: | |
3173 | btrfs_free_path(path); | |
3174 | out_unlock: | |
3175 | mutex_unlock(&BTRFS_I(dir)->log_mutex); | |
3176 | if (ret == -ENOSPC) { | |
3177 | btrfs_set_log_full_commit(root->fs_info, trans); | |
3178 | ret = 0; | |
3179 | } else if (ret < 0) | |
3180 | btrfs_abort_transaction(trans, ret); | |
3181 | ||
3182 | btrfs_end_log_trans(root); | |
3183 | ||
3184 | return err; | |
3185 | } | |
3186 | ||
3187 | /* see comments for btrfs_del_dir_entries_in_log */ | |
3188 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
3189 | struct btrfs_root *root, | |
3190 | const char *name, int name_len, | |
3191 | struct inode *inode, u64 dirid) | |
3192 | { | |
3193 | struct btrfs_root *log; | |
3194 | u64 index; | |
3195 | int ret; | |
3196 | ||
3197 | if (BTRFS_I(inode)->logged_trans < trans->transid) | |
3198 | return 0; | |
3199 | ||
3200 | ret = join_running_log_trans(root); | |
3201 | if (ret) | |
3202 | return 0; | |
3203 | log = root->log_root; | |
3204 | mutex_lock(&BTRFS_I(inode)->log_mutex); | |
3205 | ||
3206 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), | |
3207 | dirid, &index); | |
3208 | mutex_unlock(&BTRFS_I(inode)->log_mutex); | |
3209 | if (ret == -ENOSPC) { | |
3210 | btrfs_set_log_full_commit(root->fs_info, trans); | |
3211 | ret = 0; | |
3212 | } else if (ret < 0 && ret != -ENOENT) | |
3213 | btrfs_abort_transaction(trans, ret); | |
3214 | btrfs_end_log_trans(root); | |
3215 | ||
3216 | return ret; | |
3217 | } | |
3218 | ||
3219 | /* | |
3220 | * creates a range item in the log for 'dirid'. first_offset and | |
3221 | * last_offset tell us which parts of the key space the log should | |
3222 | * be considered authoritative for. | |
3223 | */ | |
3224 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
3225 | struct btrfs_root *log, | |
3226 | struct btrfs_path *path, | |
3227 | int key_type, u64 dirid, | |
3228 | u64 first_offset, u64 last_offset) | |
3229 | { | |
3230 | int ret; | |
3231 | struct btrfs_key key; | |
3232 | struct btrfs_dir_log_item *item; | |
3233 | ||
3234 | key.objectid = dirid; | |
3235 | key.offset = first_offset; | |
3236 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
3237 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
3238 | else | |
3239 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
3240 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
3241 | if (ret) | |
3242 | return ret; | |
3243 | ||
3244 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3245 | struct btrfs_dir_log_item); | |
3246 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
3247 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3248 | btrfs_release_path(path); | |
3249 | return 0; | |
3250 | } | |
3251 | ||
3252 | /* | |
3253 | * log all the items included in the current transaction for a given | |
3254 | * directory. This also creates the range items in the log tree required | |
3255 | * to replay anything deleted before the fsync | |
3256 | */ | |
3257 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
3258 | struct btrfs_root *root, struct inode *inode, | |
3259 | struct btrfs_path *path, | |
3260 | struct btrfs_path *dst_path, int key_type, | |
3261 | struct btrfs_log_ctx *ctx, | |
3262 | u64 min_offset, u64 *last_offset_ret) | |
3263 | { | |
3264 | struct btrfs_key min_key; | |
3265 | struct btrfs_root *log = root->log_root; | |
3266 | struct extent_buffer *src; | |
3267 | int err = 0; | |
3268 | int ret; | |
3269 | int i; | |
3270 | int nritems; | |
3271 | u64 first_offset = min_offset; | |
3272 | u64 last_offset = (u64)-1; | |
3273 | u64 ino = btrfs_ino(inode); | |
3274 | ||
3275 | log = root->log_root; | |
3276 | ||
3277 | min_key.objectid = ino; | |
3278 | min_key.type = key_type; | |
3279 | min_key.offset = min_offset; | |
3280 | ||
3281 | ret = btrfs_search_forward(root, &min_key, path, trans->transid); | |
3282 | ||
3283 | /* | |
3284 | * we didn't find anything from this transaction, see if there | |
3285 | * is anything at all | |
3286 | */ | |
3287 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { | |
3288 | min_key.objectid = ino; | |
3289 | min_key.type = key_type; | |
3290 | min_key.offset = (u64)-1; | |
3291 | btrfs_release_path(path); | |
3292 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); | |
3293 | if (ret < 0) { | |
3294 | btrfs_release_path(path); | |
3295 | return ret; | |
3296 | } | |
3297 | ret = btrfs_previous_item(root, path, ino, key_type); | |
3298 | ||
3299 | /* if ret == 0 there are items for this type, | |
3300 | * create a range to tell us the last key of this type. | |
3301 | * otherwise, there are no items in this directory after | |
3302 | * *min_offset, and we create a range to indicate that. | |
3303 | */ | |
3304 | if (ret == 0) { | |
3305 | struct btrfs_key tmp; | |
3306 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
3307 | path->slots[0]); | |
3308 | if (key_type == tmp.type) | |
3309 | first_offset = max(min_offset, tmp.offset) + 1; | |
3310 | } | |
3311 | goto done; | |
3312 | } | |
3313 | ||
3314 | /* go backward to find any previous key */ | |
3315 | ret = btrfs_previous_item(root, path, ino, key_type); | |
3316 | if (ret == 0) { | |
3317 | struct btrfs_key tmp; | |
3318 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3319 | if (key_type == tmp.type) { | |
3320 | first_offset = tmp.offset; | |
3321 | ret = overwrite_item(trans, log, dst_path, | |
3322 | path->nodes[0], path->slots[0], | |
3323 | &tmp); | |
3324 | if (ret) { | |
3325 | err = ret; | |
3326 | goto done; | |
3327 | } | |
3328 | } | |
3329 | } | |
3330 | btrfs_release_path(path); | |
3331 | ||
3332 | /* find the first key from this transaction again */ | |
3333 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); | |
3334 | if (WARN_ON(ret != 0)) | |
3335 | goto done; | |
3336 | ||
3337 | /* | |
3338 | * we have a block from this transaction, log every item in it | |
3339 | * from our directory | |
3340 | */ | |
3341 | while (1) { | |
3342 | struct btrfs_key tmp; | |
3343 | src = path->nodes[0]; | |
3344 | nritems = btrfs_header_nritems(src); | |
3345 | for (i = path->slots[0]; i < nritems; i++) { | |
3346 | struct btrfs_dir_item *di; | |
3347 | ||
3348 | btrfs_item_key_to_cpu(src, &min_key, i); | |
3349 | ||
3350 | if (min_key.objectid != ino || min_key.type != key_type) | |
3351 | goto done; | |
3352 | ret = overwrite_item(trans, log, dst_path, src, i, | |
3353 | &min_key); | |
3354 | if (ret) { | |
3355 | err = ret; | |
3356 | goto done; | |
3357 | } | |
3358 | ||
3359 | /* | |
3360 | * We must make sure that when we log a directory entry, | |
3361 | * the corresponding inode, after log replay, has a | |
3362 | * matching link count. For example: | |
3363 | * | |
3364 | * touch foo | |
3365 | * mkdir mydir | |
3366 | * sync | |
3367 | * ln foo mydir/bar | |
3368 | * xfs_io -c "fsync" mydir | |
3369 | * <crash> | |
3370 | * <mount fs and log replay> | |
3371 | * | |
3372 | * Would result in a fsync log that when replayed, our | |
3373 | * file inode would have a link count of 1, but we get | |
3374 | * two directory entries pointing to the same inode. | |
3375 | * After removing one of the names, it would not be | |
3376 | * possible to remove the other name, which resulted | |
3377 | * always in stale file handle errors, and would not | |
3378 | * be possible to rmdir the parent directory, since | |
3379 | * its i_size could never decrement to the value | |
3380 | * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors. | |
3381 | */ | |
3382 | di = btrfs_item_ptr(src, i, struct btrfs_dir_item); | |
3383 | btrfs_dir_item_key_to_cpu(src, di, &tmp); | |
3384 | if (ctx && | |
3385 | (btrfs_dir_transid(src, di) == trans->transid || | |
3386 | btrfs_dir_type(src, di) == BTRFS_FT_DIR) && | |
3387 | tmp.type != BTRFS_ROOT_ITEM_KEY) | |
3388 | ctx->log_new_dentries = true; | |
3389 | } | |
3390 | path->slots[0] = nritems; | |
3391 | ||
3392 | /* | |
3393 | * look ahead to the next item and see if it is also | |
3394 | * from this directory and from this transaction | |
3395 | */ | |
3396 | ret = btrfs_next_leaf(root, path); | |
3397 | if (ret == 1) { | |
3398 | last_offset = (u64)-1; | |
3399 | goto done; | |
3400 | } | |
3401 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
3402 | if (tmp.objectid != ino || tmp.type != key_type) { | |
3403 | last_offset = (u64)-1; | |
3404 | goto done; | |
3405 | } | |
3406 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
3407 | ret = overwrite_item(trans, log, dst_path, | |
3408 | path->nodes[0], path->slots[0], | |
3409 | &tmp); | |
3410 | if (ret) | |
3411 | err = ret; | |
3412 | else | |
3413 | last_offset = tmp.offset; | |
3414 | goto done; | |
3415 | } | |
3416 | } | |
3417 | done: | |
3418 | btrfs_release_path(path); | |
3419 | btrfs_release_path(dst_path); | |
3420 | ||
3421 | if (err == 0) { | |
3422 | *last_offset_ret = last_offset; | |
3423 | /* | |
3424 | * insert the log range keys to indicate where the log | |
3425 | * is valid | |
3426 | */ | |
3427 | ret = insert_dir_log_key(trans, log, path, key_type, | |
3428 | ino, first_offset, last_offset); | |
3429 | if (ret) | |
3430 | err = ret; | |
3431 | } | |
3432 | return err; | |
3433 | } | |
3434 | ||
3435 | /* | |
3436 | * logging directories is very similar to logging inodes, We find all the items | |
3437 | * from the current transaction and write them to the log. | |
3438 | * | |
3439 | * The recovery code scans the directory in the subvolume, and if it finds a | |
3440 | * key in the range logged that is not present in the log tree, then it means | |
3441 | * that dir entry was unlinked during the transaction. | |
3442 | * | |
3443 | * In order for that scan to work, we must include one key smaller than | |
3444 | * the smallest logged by this transaction and one key larger than the largest | |
3445 | * key logged by this transaction. | |
3446 | */ | |
3447 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
3448 | struct btrfs_root *root, struct inode *inode, | |
3449 | struct btrfs_path *path, | |
3450 | struct btrfs_path *dst_path, | |
3451 | struct btrfs_log_ctx *ctx) | |
3452 | { | |
3453 | u64 min_key; | |
3454 | u64 max_key; | |
3455 | int ret; | |
3456 | int key_type = BTRFS_DIR_ITEM_KEY; | |
3457 | ||
3458 | again: | |
3459 | min_key = 0; | |
3460 | max_key = 0; | |
3461 | while (1) { | |
3462 | ret = log_dir_items(trans, root, inode, path, | |
3463 | dst_path, key_type, ctx, min_key, | |
3464 | &max_key); | |
3465 | if (ret) | |
3466 | return ret; | |
3467 | if (max_key == (u64)-1) | |
3468 | break; | |
3469 | min_key = max_key + 1; | |
3470 | } | |
3471 | ||
3472 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
3473 | key_type = BTRFS_DIR_INDEX_KEY; | |
3474 | goto again; | |
3475 | } | |
3476 | return 0; | |
3477 | } | |
3478 | ||
3479 | /* | |
3480 | * a helper function to drop items from the log before we relog an | |
3481 | * inode. max_key_type indicates the highest item type to remove. | |
3482 | * This cannot be run for file data extents because it does not | |
3483 | * free the extents they point to. | |
3484 | */ | |
3485 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
3486 | struct btrfs_root *log, | |
3487 | struct btrfs_path *path, | |
3488 | u64 objectid, int max_key_type) | |
3489 | { | |
3490 | int ret; | |
3491 | struct btrfs_key key; | |
3492 | struct btrfs_key found_key; | |
3493 | int start_slot; | |
3494 | ||
3495 | key.objectid = objectid; | |
3496 | key.type = max_key_type; | |
3497 | key.offset = (u64)-1; | |
3498 | ||
3499 | while (1) { | |
3500 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); | |
3501 | BUG_ON(ret == 0); /* Logic error */ | |
3502 | if (ret < 0) | |
3503 | break; | |
3504 | ||
3505 | if (path->slots[0] == 0) | |
3506 | break; | |
3507 | ||
3508 | path->slots[0]--; | |
3509 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3510 | path->slots[0]); | |
3511 | ||
3512 | if (found_key.objectid != objectid) | |
3513 | break; | |
3514 | ||
3515 | found_key.offset = 0; | |
3516 | found_key.type = 0; | |
3517 | ret = btrfs_bin_search(path->nodes[0], &found_key, 0, | |
3518 | &start_slot); | |
3519 | ||
3520 | ret = btrfs_del_items(trans, log, path, start_slot, | |
3521 | path->slots[0] - start_slot + 1); | |
3522 | /* | |
3523 | * If start slot isn't 0 then we don't need to re-search, we've | |
3524 | * found the last guy with the objectid in this tree. | |
3525 | */ | |
3526 | if (ret || start_slot != 0) | |
3527 | break; | |
3528 | btrfs_release_path(path); | |
3529 | } | |
3530 | btrfs_release_path(path); | |
3531 | if (ret > 0) | |
3532 | ret = 0; | |
3533 | return ret; | |
3534 | } | |
3535 | ||
3536 | static void fill_inode_item(struct btrfs_trans_handle *trans, | |
3537 | struct extent_buffer *leaf, | |
3538 | struct btrfs_inode_item *item, | |
3539 | struct inode *inode, int log_inode_only, | |
3540 | u64 logged_isize) | |
3541 | { | |
3542 | struct btrfs_map_token token; | |
3543 | ||
3544 | btrfs_init_map_token(&token); | |
3545 | ||
3546 | if (log_inode_only) { | |
3547 | /* set the generation to zero so the recover code | |
3548 | * can tell the difference between an logging | |
3549 | * just to say 'this inode exists' and a logging | |
3550 | * to say 'update this inode with these values' | |
3551 | */ | |
3552 | btrfs_set_token_inode_generation(leaf, item, 0, &token); | |
3553 | btrfs_set_token_inode_size(leaf, item, logged_isize, &token); | |
3554 | } else { | |
3555 | btrfs_set_token_inode_generation(leaf, item, | |
3556 | BTRFS_I(inode)->generation, | |
3557 | &token); | |
3558 | btrfs_set_token_inode_size(leaf, item, inode->i_size, &token); | |
3559 | } | |
3560 | ||
3561 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
3562 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3563 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3564 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
3565 | ||
3566 | btrfs_set_token_timespec_sec(leaf, &item->atime, | |
3567 | inode->i_atime.tv_sec, &token); | |
3568 | btrfs_set_token_timespec_nsec(leaf, &item->atime, | |
3569 | inode->i_atime.tv_nsec, &token); | |
3570 | ||
3571 | btrfs_set_token_timespec_sec(leaf, &item->mtime, | |
3572 | inode->i_mtime.tv_sec, &token); | |
3573 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, | |
3574 | inode->i_mtime.tv_nsec, &token); | |
3575 | ||
3576 | btrfs_set_token_timespec_sec(leaf, &item->ctime, | |
3577 | inode->i_ctime.tv_sec, &token); | |
3578 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, | |
3579 | inode->i_ctime.tv_nsec, &token); | |
3580 | ||
3581 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
3582 | &token); | |
3583 | ||
3584 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3585 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3586 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3587 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3588 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
3589 | } | |
3590 | ||
3591 | static int log_inode_item(struct btrfs_trans_handle *trans, | |
3592 | struct btrfs_root *log, struct btrfs_path *path, | |
3593 | struct inode *inode) | |
3594 | { | |
3595 | struct btrfs_inode_item *inode_item; | |
3596 | int ret; | |
3597 | ||
3598 | ret = btrfs_insert_empty_item(trans, log, path, | |
3599 | &BTRFS_I(inode)->location, | |
3600 | sizeof(*inode_item)); | |
3601 | if (ret && ret != -EEXIST) | |
3602 | return ret; | |
3603 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3604 | struct btrfs_inode_item); | |
3605 | fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0); | |
3606 | btrfs_release_path(path); | |
3607 | return 0; | |
3608 | } | |
3609 | ||
3610 | static noinline int copy_items(struct btrfs_trans_handle *trans, | |
3611 | struct inode *inode, | |
3612 | struct btrfs_path *dst_path, | |
3613 | struct btrfs_path *src_path, u64 *last_extent, | |
3614 | int start_slot, int nr, int inode_only, | |
3615 | u64 logged_isize) | |
3616 | { | |
3617 | unsigned long src_offset; | |
3618 | unsigned long dst_offset; | |
3619 | struct btrfs_root *log = BTRFS_I(inode)->root->log_root; | |
3620 | struct btrfs_file_extent_item *extent; | |
3621 | struct btrfs_inode_item *inode_item; | |
3622 | struct extent_buffer *src = src_path->nodes[0]; | |
3623 | struct btrfs_key first_key, last_key, key; | |
3624 | int ret; | |
3625 | struct btrfs_key *ins_keys; | |
3626 | u32 *ins_sizes; | |
3627 | char *ins_data; | |
3628 | int i; | |
3629 | struct list_head ordered_sums; | |
3630 | int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
3631 | bool has_extents = false; | |
3632 | bool need_find_last_extent = true; | |
3633 | bool done = false; | |
3634 | ||
3635 | INIT_LIST_HEAD(&ordered_sums); | |
3636 | ||
3637 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
3638 | nr * sizeof(u32), GFP_NOFS); | |
3639 | if (!ins_data) | |
3640 | return -ENOMEM; | |
3641 | ||
3642 | first_key.objectid = (u64)-1; | |
3643 | ||
3644 | ins_sizes = (u32 *)ins_data; | |
3645 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
3646 | ||
3647 | for (i = 0; i < nr; i++) { | |
3648 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
3649 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
3650 | } | |
3651 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
3652 | ins_keys, ins_sizes, nr); | |
3653 | if (ret) { | |
3654 | kfree(ins_data); | |
3655 | return ret; | |
3656 | } | |
3657 | ||
3658 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { | |
3659 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], | |
3660 | dst_path->slots[0]); | |
3661 | ||
3662 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
3663 | ||
3664 | if ((i == (nr - 1))) | |
3665 | last_key = ins_keys[i]; | |
3666 | ||
3667 | if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { | |
3668 | inode_item = btrfs_item_ptr(dst_path->nodes[0], | |
3669 | dst_path->slots[0], | |
3670 | struct btrfs_inode_item); | |
3671 | fill_inode_item(trans, dst_path->nodes[0], inode_item, | |
3672 | inode, inode_only == LOG_INODE_EXISTS, | |
3673 | logged_isize); | |
3674 | } else { | |
3675 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
3676 | src_offset, ins_sizes[i]); | |
3677 | } | |
3678 | ||
3679 | /* | |
3680 | * We set need_find_last_extent here in case we know we were | |
3681 | * processing other items and then walk into the first extent in | |
3682 | * the inode. If we don't hit an extent then nothing changes, | |
3683 | * we'll do the last search the next time around. | |
3684 | */ | |
3685 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) { | |
3686 | has_extents = true; | |
3687 | if (first_key.objectid == (u64)-1) | |
3688 | first_key = ins_keys[i]; | |
3689 | } else { | |
3690 | need_find_last_extent = false; | |
3691 | } | |
3692 | ||
3693 | /* take a reference on file data extents so that truncates | |
3694 | * or deletes of this inode don't have to relog the inode | |
3695 | * again | |
3696 | */ | |
3697 | if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY && | |
3698 | !skip_csum) { | |
3699 | int found_type; | |
3700 | extent = btrfs_item_ptr(src, start_slot + i, | |
3701 | struct btrfs_file_extent_item); | |
3702 | ||
3703 | if (btrfs_file_extent_generation(src, extent) < trans->transid) | |
3704 | continue; | |
3705 | ||
3706 | found_type = btrfs_file_extent_type(src, extent); | |
3707 | if (found_type == BTRFS_FILE_EXTENT_REG) { | |
3708 | u64 ds, dl, cs, cl; | |
3709 | ds = btrfs_file_extent_disk_bytenr(src, | |
3710 | extent); | |
3711 | /* ds == 0 is a hole */ | |
3712 | if (ds == 0) | |
3713 | continue; | |
3714 | ||
3715 | dl = btrfs_file_extent_disk_num_bytes(src, | |
3716 | extent); | |
3717 | cs = btrfs_file_extent_offset(src, extent); | |
3718 | cl = btrfs_file_extent_num_bytes(src, | |
3719 | extent); | |
3720 | if (btrfs_file_extent_compression(src, | |
3721 | extent)) { | |
3722 | cs = 0; | |
3723 | cl = dl; | |
3724 | } | |
3725 | ||
3726 | ret = btrfs_lookup_csums_range( | |
3727 | log->fs_info->csum_root, | |
3728 | ds + cs, ds + cs + cl - 1, | |
3729 | &ordered_sums, 0); | |
3730 | if (ret) { | |
3731 | btrfs_release_path(dst_path); | |
3732 | kfree(ins_data); | |
3733 | return ret; | |
3734 | } | |
3735 | } | |
3736 | } | |
3737 | } | |
3738 | ||
3739 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
3740 | btrfs_release_path(dst_path); | |
3741 | kfree(ins_data); | |
3742 | ||
3743 | /* | |
3744 | * we have to do this after the loop above to avoid changing the | |
3745 | * log tree while trying to change the log tree. | |
3746 | */ | |
3747 | ret = 0; | |
3748 | while (!list_empty(&ordered_sums)) { | |
3749 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
3750 | struct btrfs_ordered_sum, | |
3751 | list); | |
3752 | if (!ret) | |
3753 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
3754 | list_del(&sums->list); | |
3755 | kfree(sums); | |
3756 | } | |
3757 | ||
3758 | if (!has_extents) | |
3759 | return ret; | |
3760 | ||
3761 | if (need_find_last_extent && *last_extent == first_key.offset) { | |
3762 | /* | |
3763 | * We don't have any leafs between our current one and the one | |
3764 | * we processed before that can have file extent items for our | |
3765 | * inode (and have a generation number smaller than our current | |
3766 | * transaction id). | |
3767 | */ | |
3768 | need_find_last_extent = false; | |
3769 | } | |
3770 | ||
3771 | /* | |
3772 | * Because we use btrfs_search_forward we could skip leaves that were | |
3773 | * not modified and then assume *last_extent is valid when it really | |
3774 | * isn't. So back up to the previous leaf and read the end of the last | |
3775 | * extent before we go and fill in holes. | |
3776 | */ | |
3777 | if (need_find_last_extent) { | |
3778 | u64 len; | |
3779 | ||
3780 | ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path); | |
3781 | if (ret < 0) | |
3782 | return ret; | |
3783 | if (ret) | |
3784 | goto fill_holes; | |
3785 | if (src_path->slots[0]) | |
3786 | src_path->slots[0]--; | |
3787 | src = src_path->nodes[0]; | |
3788 | btrfs_item_key_to_cpu(src, &key, src_path->slots[0]); | |
3789 | if (key.objectid != btrfs_ino(inode) || | |
3790 | key.type != BTRFS_EXTENT_DATA_KEY) | |
3791 | goto fill_holes; | |
3792 | extent = btrfs_item_ptr(src, src_path->slots[0], | |
3793 | struct btrfs_file_extent_item); | |
3794 | if (btrfs_file_extent_type(src, extent) == | |
3795 | BTRFS_FILE_EXTENT_INLINE) { | |
3796 | len = btrfs_file_extent_inline_len(src, | |
3797 | src_path->slots[0], | |
3798 | extent); | |
3799 | *last_extent = ALIGN(key.offset + len, | |
3800 | log->sectorsize); | |
3801 | } else { | |
3802 | len = btrfs_file_extent_num_bytes(src, extent); | |
3803 | *last_extent = key.offset + len; | |
3804 | } | |
3805 | } | |
3806 | fill_holes: | |
3807 | /* So we did prev_leaf, now we need to move to the next leaf, but a few | |
3808 | * things could have happened | |
3809 | * | |
3810 | * 1) A merge could have happened, so we could currently be on a leaf | |
3811 | * that holds what we were copying in the first place. | |
3812 | * 2) A split could have happened, and now not all of the items we want | |
3813 | * are on the same leaf. | |
3814 | * | |
3815 | * So we need to adjust how we search for holes, we need to drop the | |
3816 | * path and re-search for the first extent key we found, and then walk | |
3817 | * forward until we hit the last one we copied. | |
3818 | */ | |
3819 | if (need_find_last_extent) { | |
3820 | /* btrfs_prev_leaf could return 1 without releasing the path */ | |
3821 | btrfs_release_path(src_path); | |
3822 | ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key, | |
3823 | src_path, 0, 0); | |
3824 | if (ret < 0) | |
3825 | return ret; | |
3826 | ASSERT(ret == 0); | |
3827 | src = src_path->nodes[0]; | |
3828 | i = src_path->slots[0]; | |
3829 | } else { | |
3830 | i = start_slot; | |
3831 | } | |
3832 | ||
3833 | /* | |
3834 | * Ok so here we need to go through and fill in any holes we may have | |
3835 | * to make sure that holes are punched for those areas in case they had | |
3836 | * extents previously. | |
3837 | */ | |
3838 | while (!done) { | |
3839 | u64 offset, len; | |
3840 | u64 extent_end; | |
3841 | ||
3842 | if (i >= btrfs_header_nritems(src_path->nodes[0])) { | |
3843 | ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path); | |
3844 | if (ret < 0) | |
3845 | return ret; | |
3846 | ASSERT(ret == 0); | |
3847 | src = src_path->nodes[0]; | |
3848 | i = 0; | |
3849 | } | |
3850 | ||
3851 | btrfs_item_key_to_cpu(src, &key, i); | |
3852 | if (!btrfs_comp_cpu_keys(&key, &last_key)) | |
3853 | done = true; | |
3854 | if (key.objectid != btrfs_ino(inode) || | |
3855 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
3856 | i++; | |
3857 | continue; | |
3858 | } | |
3859 | extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item); | |
3860 | if (btrfs_file_extent_type(src, extent) == | |
3861 | BTRFS_FILE_EXTENT_INLINE) { | |
3862 | len = btrfs_file_extent_inline_len(src, i, extent); | |
3863 | extent_end = ALIGN(key.offset + len, log->sectorsize); | |
3864 | } else { | |
3865 | len = btrfs_file_extent_num_bytes(src, extent); | |
3866 | extent_end = key.offset + len; | |
3867 | } | |
3868 | i++; | |
3869 | ||
3870 | if (*last_extent == key.offset) { | |
3871 | *last_extent = extent_end; | |
3872 | continue; | |
3873 | } | |
3874 | offset = *last_extent; | |
3875 | len = key.offset - *last_extent; | |
3876 | ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode), | |
3877 | offset, 0, 0, len, 0, len, 0, | |
3878 | 0, 0); | |
3879 | if (ret) | |
3880 | break; | |
3881 | *last_extent = extent_end; | |
3882 | } | |
3883 | /* | |
3884 | * Need to let the callers know we dropped the path so they should | |
3885 | * re-search. | |
3886 | */ | |
3887 | if (!ret && need_find_last_extent) | |
3888 | ret = 1; | |
3889 | return ret; | |
3890 | } | |
3891 | ||
3892 | static int extent_cmp(void *priv, struct list_head *a, struct list_head *b) | |
3893 | { | |
3894 | struct extent_map *em1, *em2; | |
3895 | ||
3896 | em1 = list_entry(a, struct extent_map, list); | |
3897 | em2 = list_entry(b, struct extent_map, list); | |
3898 | ||
3899 | if (em1->start < em2->start) | |
3900 | return -1; | |
3901 | else if (em1->start > em2->start) | |
3902 | return 1; | |
3903 | return 0; | |
3904 | } | |
3905 | ||
3906 | static int wait_ordered_extents(struct btrfs_trans_handle *trans, | |
3907 | struct inode *inode, | |
3908 | struct btrfs_root *root, | |
3909 | const struct extent_map *em, | |
3910 | const struct list_head *logged_list, | |
3911 | bool *ordered_io_error) | |
3912 | { | |
3913 | struct btrfs_ordered_extent *ordered; | |
3914 | struct btrfs_root *log = root->log_root; | |
3915 | u64 mod_start = em->mod_start; | |
3916 | u64 mod_len = em->mod_len; | |
3917 | const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
3918 | u64 csum_offset; | |
3919 | u64 csum_len; | |
3920 | LIST_HEAD(ordered_sums); | |
3921 | int ret = 0; | |
3922 | ||
3923 | *ordered_io_error = false; | |
3924 | ||
3925 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
3926 | em->block_start == EXTENT_MAP_HOLE) | |
3927 | return 0; | |
3928 | ||
3929 | /* | |
3930 | * Wait far any ordered extent that covers our extent map. If it | |
3931 | * finishes without an error, first check and see if our csums are on | |
3932 | * our outstanding ordered extents. | |
3933 | */ | |
3934 | list_for_each_entry(ordered, logged_list, log_list) { | |
3935 | struct btrfs_ordered_sum *sum; | |
3936 | ||
3937 | if (!mod_len) | |
3938 | break; | |
3939 | ||
3940 | if (ordered->file_offset + ordered->len <= mod_start || | |
3941 | mod_start + mod_len <= ordered->file_offset) | |
3942 | continue; | |
3943 | ||
3944 | if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) && | |
3945 | !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) && | |
3946 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) { | |
3947 | const u64 start = ordered->file_offset; | |
3948 | const u64 end = ordered->file_offset + ordered->len - 1; | |
3949 | ||
3950 | WARN_ON(ordered->inode != inode); | |
3951 | filemap_fdatawrite_range(inode->i_mapping, start, end); | |
3952 | } | |
3953 | ||
3954 | wait_event(ordered->wait, | |
3955 | (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) || | |
3956 | test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))); | |
3957 | ||
3958 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) { | |
3959 | /* | |
3960 | * Clear the AS_EIO/AS_ENOSPC flags from the inode's | |
3961 | * i_mapping flags, so that the next fsync won't get | |
3962 | * an outdated io error too. | |
3963 | */ | |
3964 | btrfs_inode_check_errors(inode); | |
3965 | *ordered_io_error = true; | |
3966 | break; | |
3967 | } | |
3968 | /* | |
3969 | * We are going to copy all the csums on this ordered extent, so | |
3970 | * go ahead and adjust mod_start and mod_len in case this | |
3971 | * ordered extent has already been logged. | |
3972 | */ | |
3973 | if (ordered->file_offset > mod_start) { | |
3974 | if (ordered->file_offset + ordered->len >= | |
3975 | mod_start + mod_len) | |
3976 | mod_len = ordered->file_offset - mod_start; | |
3977 | /* | |
3978 | * If we have this case | |
3979 | * | |
3980 | * |--------- logged extent ---------| | |
3981 | * |----- ordered extent ----| | |
3982 | * | |
3983 | * Just don't mess with mod_start and mod_len, we'll | |
3984 | * just end up logging more csums than we need and it | |
3985 | * will be ok. | |
3986 | */ | |
3987 | } else { | |
3988 | if (ordered->file_offset + ordered->len < | |
3989 | mod_start + mod_len) { | |
3990 | mod_len = (mod_start + mod_len) - | |
3991 | (ordered->file_offset + ordered->len); | |
3992 | mod_start = ordered->file_offset + | |
3993 | ordered->len; | |
3994 | } else { | |
3995 | mod_len = 0; | |
3996 | } | |
3997 | } | |
3998 | ||
3999 | if (skip_csum) | |
4000 | continue; | |
4001 | ||
4002 | /* | |
4003 | * To keep us from looping for the above case of an ordered | |
4004 | * extent that falls inside of the logged extent. | |
4005 | */ | |
4006 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, | |
4007 | &ordered->flags)) | |
4008 | continue; | |
4009 | ||
4010 | list_for_each_entry(sum, &ordered->list, list) { | |
4011 | ret = btrfs_csum_file_blocks(trans, log, sum); | |
4012 | if (ret) | |
4013 | break; | |
4014 | } | |
4015 | } | |
4016 | ||
4017 | if (*ordered_io_error || !mod_len || ret || skip_csum) | |
4018 | return ret; | |
4019 | ||
4020 | if (em->compress_type) { | |
4021 | csum_offset = 0; | |
4022 | csum_len = max(em->block_len, em->orig_block_len); | |
4023 | } else { | |
4024 | csum_offset = mod_start - em->start; | |
4025 | csum_len = mod_len; | |
4026 | } | |
4027 | ||
4028 | /* block start is already adjusted for the file extent offset. */ | |
4029 | ret = btrfs_lookup_csums_range(log->fs_info->csum_root, | |
4030 | em->block_start + csum_offset, | |
4031 | em->block_start + csum_offset + | |
4032 | csum_len - 1, &ordered_sums, 0); | |
4033 | if (ret) | |
4034 | return ret; | |
4035 | ||
4036 | while (!list_empty(&ordered_sums)) { | |
4037 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, | |
4038 | struct btrfs_ordered_sum, | |
4039 | list); | |
4040 | if (!ret) | |
4041 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
4042 | list_del(&sums->list); | |
4043 | kfree(sums); | |
4044 | } | |
4045 | ||
4046 | return ret; | |
4047 | } | |
4048 | ||
4049 | static int log_one_extent(struct btrfs_trans_handle *trans, | |
4050 | struct inode *inode, struct btrfs_root *root, | |
4051 | const struct extent_map *em, | |
4052 | struct btrfs_path *path, | |
4053 | const struct list_head *logged_list, | |
4054 | struct btrfs_log_ctx *ctx) | |
4055 | { | |
4056 | struct btrfs_root *log = root->log_root; | |
4057 | struct btrfs_file_extent_item *fi; | |
4058 | struct extent_buffer *leaf; | |
4059 | struct btrfs_map_token token; | |
4060 | struct btrfs_key key; | |
4061 | u64 extent_offset = em->start - em->orig_start; | |
4062 | u64 block_len; | |
4063 | int ret; | |
4064 | int extent_inserted = 0; | |
4065 | bool ordered_io_err = false; | |
4066 | ||
4067 | ret = wait_ordered_extents(trans, inode, root, em, logged_list, | |
4068 | &ordered_io_err); | |
4069 | if (ret) | |
4070 | return ret; | |
4071 | ||
4072 | if (ordered_io_err) { | |
4073 | ctx->io_err = -EIO; | |
4074 | return 0; | |
4075 | } | |
4076 | ||
4077 | btrfs_init_map_token(&token); | |
4078 | ||
4079 | ret = __btrfs_drop_extents(trans, log, inode, path, em->start, | |
4080 | em->start + em->len, NULL, 0, 1, | |
4081 | sizeof(*fi), &extent_inserted); | |
4082 | if (ret) | |
4083 | return ret; | |
4084 | ||
4085 | if (!extent_inserted) { | |
4086 | key.objectid = btrfs_ino(inode); | |
4087 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4088 | key.offset = em->start; | |
4089 | ||
4090 | ret = btrfs_insert_empty_item(trans, log, path, &key, | |
4091 | sizeof(*fi)); | |
4092 | if (ret) | |
4093 | return ret; | |
4094 | } | |
4095 | leaf = path->nodes[0]; | |
4096 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4097 | struct btrfs_file_extent_item); | |
4098 | ||
4099 | btrfs_set_token_file_extent_generation(leaf, fi, trans->transid, | |
4100 | &token); | |
4101 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
4102 | btrfs_set_token_file_extent_type(leaf, fi, | |
4103 | BTRFS_FILE_EXTENT_PREALLOC, | |
4104 | &token); | |
4105 | else | |
4106 | btrfs_set_token_file_extent_type(leaf, fi, | |
4107 | BTRFS_FILE_EXTENT_REG, | |
4108 | &token); | |
4109 | ||
4110 | block_len = max(em->block_len, em->orig_block_len); | |
4111 | if (em->compress_type != BTRFS_COMPRESS_NONE) { | |
4112 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4113 | em->block_start, | |
4114 | &token); | |
4115 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4116 | &token); | |
4117 | } else if (em->block_start < EXTENT_MAP_LAST_BYTE) { | |
4118 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, | |
4119 | em->block_start - | |
4120 | extent_offset, &token); | |
4121 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len, | |
4122 | &token); | |
4123 | } else { | |
4124 | btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token); | |
4125 | btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0, | |
4126 | &token); | |
4127 | } | |
4128 | ||
4129 | btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token); | |
4130 | btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token); | |
4131 | btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token); | |
4132 | btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type, | |
4133 | &token); | |
4134 | btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token); | |
4135 | btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token); | |
4136 | btrfs_mark_buffer_dirty(leaf); | |
4137 | ||
4138 | btrfs_release_path(path); | |
4139 | ||
4140 | return ret; | |
4141 | } | |
4142 | ||
4143 | static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans, | |
4144 | struct btrfs_root *root, | |
4145 | struct inode *inode, | |
4146 | struct btrfs_path *path, | |
4147 | struct list_head *logged_list, | |
4148 | struct btrfs_log_ctx *ctx, | |
4149 | const u64 start, | |
4150 | const u64 end) | |
4151 | { | |
4152 | struct extent_map *em, *n; | |
4153 | struct list_head extents; | |
4154 | struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree; | |
4155 | u64 test_gen; | |
4156 | int ret = 0; | |
4157 | int num = 0; | |
4158 | ||
4159 | INIT_LIST_HEAD(&extents); | |
4160 | ||
4161 | down_write(&BTRFS_I(inode)->dio_sem); | |
4162 | write_lock(&tree->lock); | |
4163 | test_gen = root->fs_info->last_trans_committed; | |
4164 | ||
4165 | list_for_each_entry_safe(em, n, &tree->modified_extents, list) { | |
4166 | list_del_init(&em->list); | |
4167 | ||
4168 | /* | |
4169 | * Just an arbitrary number, this can be really CPU intensive | |
4170 | * once we start getting a lot of extents, and really once we | |
4171 | * have a bunch of extents we just want to commit since it will | |
4172 | * be faster. | |
4173 | */ | |
4174 | if (++num > 32768) { | |
4175 | list_del_init(&tree->modified_extents); | |
4176 | ret = -EFBIG; | |
4177 | goto process; | |
4178 | } | |
4179 | ||
4180 | if (em->generation <= test_gen) | |
4181 | continue; | |
4182 | /* Need a ref to keep it from getting evicted from cache */ | |
4183 | atomic_inc(&em->refs); | |
4184 | set_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
4185 | list_add_tail(&em->list, &extents); | |
4186 | num++; | |
4187 | } | |
4188 | ||
4189 | list_sort(NULL, &extents, extent_cmp); | |
4190 | btrfs_get_logged_extents(inode, logged_list, start, end); | |
4191 | /* | |
4192 | * Some ordered extents started by fsync might have completed | |
4193 | * before we could collect them into the list logged_list, which | |
4194 | * means they're gone, not in our logged_list nor in the inode's | |
4195 | * ordered tree. We want the application/user space to know an | |
4196 | * error happened while attempting to persist file data so that | |
4197 | * it can take proper action. If such error happened, we leave | |
4198 | * without writing to the log tree and the fsync must report the | |
4199 | * file data write error and not commit the current transaction. | |
4200 | */ | |
4201 | ret = btrfs_inode_check_errors(inode); | |
4202 | if (ret) | |
4203 | ctx->io_err = ret; | |
4204 | process: | |
4205 | while (!list_empty(&extents)) { | |
4206 | em = list_entry(extents.next, struct extent_map, list); | |
4207 | ||
4208 | list_del_init(&em->list); | |
4209 | ||
4210 | /* | |
4211 | * If we had an error we just need to delete everybody from our | |
4212 | * private list. | |
4213 | */ | |
4214 | if (ret) { | |
4215 | clear_em_logging(tree, em); | |
4216 | free_extent_map(em); | |
4217 | continue; | |
4218 | } | |
4219 | ||
4220 | write_unlock(&tree->lock); | |
4221 | ||
4222 | ret = log_one_extent(trans, inode, root, em, path, logged_list, | |
4223 | ctx); | |
4224 | write_lock(&tree->lock); | |
4225 | clear_em_logging(tree, em); | |
4226 | free_extent_map(em); | |
4227 | } | |
4228 | WARN_ON(!list_empty(&extents)); | |
4229 | write_unlock(&tree->lock); | |
4230 | up_write(&BTRFS_I(inode)->dio_sem); | |
4231 | ||
4232 | btrfs_release_path(path); | |
4233 | return ret; | |
4234 | } | |
4235 | ||
4236 | static int logged_inode_size(struct btrfs_root *log, struct inode *inode, | |
4237 | struct btrfs_path *path, u64 *size_ret) | |
4238 | { | |
4239 | struct btrfs_key key; | |
4240 | int ret; | |
4241 | ||
4242 | key.objectid = btrfs_ino(inode); | |
4243 | key.type = BTRFS_INODE_ITEM_KEY; | |
4244 | key.offset = 0; | |
4245 | ||
4246 | ret = btrfs_search_slot(NULL, log, &key, path, 0, 0); | |
4247 | if (ret < 0) { | |
4248 | return ret; | |
4249 | } else if (ret > 0) { | |
4250 | *size_ret = 0; | |
4251 | } else { | |
4252 | struct btrfs_inode_item *item; | |
4253 | ||
4254 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4255 | struct btrfs_inode_item); | |
4256 | *size_ret = btrfs_inode_size(path->nodes[0], item); | |
4257 | } | |
4258 | ||
4259 | btrfs_release_path(path); | |
4260 | return 0; | |
4261 | } | |
4262 | ||
4263 | /* | |
4264 | * At the moment we always log all xattrs. This is to figure out at log replay | |
4265 | * time which xattrs must have their deletion replayed. If a xattr is missing | |
4266 | * in the log tree and exists in the fs/subvol tree, we delete it. This is | |
4267 | * because if a xattr is deleted, the inode is fsynced and a power failure | |
4268 | * happens, causing the log to be replayed the next time the fs is mounted, | |
4269 | * we want the xattr to not exist anymore (same behaviour as other filesystems | |
4270 | * with a journal, ext3/4, xfs, f2fs, etc). | |
4271 | */ | |
4272 | static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans, | |
4273 | struct btrfs_root *root, | |
4274 | struct inode *inode, | |
4275 | struct btrfs_path *path, | |
4276 | struct btrfs_path *dst_path) | |
4277 | { | |
4278 | int ret; | |
4279 | struct btrfs_key key; | |
4280 | const u64 ino = btrfs_ino(inode); | |
4281 | int ins_nr = 0; | |
4282 | int start_slot = 0; | |
4283 | ||
4284 | key.objectid = ino; | |
4285 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4286 | key.offset = 0; | |
4287 | ||
4288 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4289 | if (ret < 0) | |
4290 | return ret; | |
4291 | ||
4292 | while (true) { | |
4293 | int slot = path->slots[0]; | |
4294 | struct extent_buffer *leaf = path->nodes[0]; | |
4295 | int nritems = btrfs_header_nritems(leaf); | |
4296 | ||
4297 | if (slot >= nritems) { | |
4298 | if (ins_nr > 0) { | |
4299 | u64 last_extent = 0; | |
4300 | ||
4301 | ret = copy_items(trans, inode, dst_path, path, | |
4302 | &last_extent, start_slot, | |
4303 | ins_nr, 1, 0); | |
4304 | /* can't be 1, extent items aren't processed */ | |
4305 | ASSERT(ret <= 0); | |
4306 | if (ret < 0) | |
4307 | return ret; | |
4308 | ins_nr = 0; | |
4309 | } | |
4310 | ret = btrfs_next_leaf(root, path); | |
4311 | if (ret < 0) | |
4312 | return ret; | |
4313 | else if (ret > 0) | |
4314 | break; | |
4315 | continue; | |
4316 | } | |
4317 | ||
4318 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4319 | if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) | |
4320 | break; | |
4321 | ||
4322 | if (ins_nr == 0) | |
4323 | start_slot = slot; | |
4324 | ins_nr++; | |
4325 | path->slots[0]++; | |
4326 | cond_resched(); | |
4327 | } | |
4328 | if (ins_nr > 0) { | |
4329 | u64 last_extent = 0; | |
4330 | ||
4331 | ret = copy_items(trans, inode, dst_path, path, | |
4332 | &last_extent, start_slot, | |
4333 | ins_nr, 1, 0); | |
4334 | /* can't be 1, extent items aren't processed */ | |
4335 | ASSERT(ret <= 0); | |
4336 | if (ret < 0) | |
4337 | return ret; | |
4338 | } | |
4339 | ||
4340 | return 0; | |
4341 | } | |
4342 | ||
4343 | /* | |
4344 | * If the no holes feature is enabled we need to make sure any hole between the | |
4345 | * last extent and the i_size of our inode is explicitly marked in the log. This | |
4346 | * is to make sure that doing something like: | |
4347 | * | |
4348 | * 1) create file with 128Kb of data | |
4349 | * 2) truncate file to 64Kb | |
4350 | * 3) truncate file to 256Kb | |
4351 | * 4) fsync file | |
4352 | * 5) <crash/power failure> | |
4353 | * 6) mount fs and trigger log replay | |
4354 | * | |
4355 | * Will give us a file with a size of 256Kb, the first 64Kb of data match what | |
4356 | * the file had in its first 64Kb of data at step 1 and the last 192Kb of the | |
4357 | * file correspond to a hole. The presence of explicit holes in a log tree is | |
4358 | * what guarantees that log replay will remove/adjust file extent items in the | |
4359 | * fs/subvol tree. | |
4360 | * | |
4361 | * Here we do not need to care about holes between extents, that is already done | |
4362 | * by copy_items(). We also only need to do this in the full sync path, where we | |
4363 | * lookup for extents from the fs/subvol tree only. In the fast path case, we | |
4364 | * lookup the list of modified extent maps and if any represents a hole, we | |
4365 | * insert a corresponding extent representing a hole in the log tree. | |
4366 | */ | |
4367 | static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans, | |
4368 | struct btrfs_root *root, | |
4369 | struct inode *inode, | |
4370 | struct btrfs_path *path) | |
4371 | { | |
4372 | int ret; | |
4373 | struct btrfs_key key; | |
4374 | u64 hole_start; | |
4375 | u64 hole_size; | |
4376 | struct extent_buffer *leaf; | |
4377 | struct btrfs_root *log = root->log_root; | |
4378 | const u64 ino = btrfs_ino(inode); | |
4379 | const u64 i_size = i_size_read(inode); | |
4380 | ||
4381 | if (!btrfs_fs_incompat(root->fs_info, NO_HOLES)) | |
4382 | return 0; | |
4383 | ||
4384 | key.objectid = ino; | |
4385 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4386 | key.offset = (u64)-1; | |
4387 | ||
4388 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4389 | ASSERT(ret != 0); | |
4390 | if (ret < 0) | |
4391 | return ret; | |
4392 | ||
4393 | ASSERT(path->slots[0] > 0); | |
4394 | path->slots[0]--; | |
4395 | leaf = path->nodes[0]; | |
4396 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4397 | ||
4398 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) { | |
4399 | /* inode does not have any extents */ | |
4400 | hole_start = 0; | |
4401 | hole_size = i_size; | |
4402 | } else { | |
4403 | struct btrfs_file_extent_item *extent; | |
4404 | u64 len; | |
4405 | ||
4406 | /* | |
4407 | * If there's an extent beyond i_size, an explicit hole was | |
4408 | * already inserted by copy_items(). | |
4409 | */ | |
4410 | if (key.offset >= i_size) | |
4411 | return 0; | |
4412 | ||
4413 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
4414 | struct btrfs_file_extent_item); | |
4415 | ||
4416 | if (btrfs_file_extent_type(leaf, extent) == | |
4417 | BTRFS_FILE_EXTENT_INLINE) { | |
4418 | len = btrfs_file_extent_inline_len(leaf, | |
4419 | path->slots[0], | |
4420 | extent); | |
4421 | ASSERT(len == i_size); | |
4422 | return 0; | |
4423 | } | |
4424 | ||
4425 | len = btrfs_file_extent_num_bytes(leaf, extent); | |
4426 | /* Last extent goes beyond i_size, no need to log a hole. */ | |
4427 | if (key.offset + len > i_size) | |
4428 | return 0; | |
4429 | hole_start = key.offset + len; | |
4430 | hole_size = i_size - hole_start; | |
4431 | } | |
4432 | btrfs_release_path(path); | |
4433 | ||
4434 | /* Last extent ends at i_size. */ | |
4435 | if (hole_size == 0) | |
4436 | return 0; | |
4437 | ||
4438 | hole_size = ALIGN(hole_size, root->sectorsize); | |
4439 | ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0, | |
4440 | hole_size, 0, hole_size, 0, 0, 0); | |
4441 | return ret; | |
4442 | } | |
4443 | ||
4444 | /* | |
4445 | * When we are logging a new inode X, check if it doesn't have a reference that | |
4446 | * matches the reference from some other inode Y created in a past transaction | |
4447 | * and that was renamed in the current transaction. If we don't do this, then at | |
4448 | * log replay time we can lose inode Y (and all its files if it's a directory): | |
4449 | * | |
4450 | * mkdir /mnt/x | |
4451 | * echo "hello world" > /mnt/x/foobar | |
4452 | * sync | |
4453 | * mv /mnt/x /mnt/y | |
4454 | * mkdir /mnt/x # or touch /mnt/x | |
4455 | * xfs_io -c fsync /mnt/x | |
4456 | * <power fail> | |
4457 | * mount fs, trigger log replay | |
4458 | * | |
4459 | * After the log replay procedure, we would lose the first directory and all its | |
4460 | * files (file foobar). | |
4461 | * For the case where inode Y is not a directory we simply end up losing it: | |
4462 | * | |
4463 | * echo "123" > /mnt/foo | |
4464 | * sync | |
4465 | * mv /mnt/foo /mnt/bar | |
4466 | * echo "abc" > /mnt/foo | |
4467 | * xfs_io -c fsync /mnt/foo | |
4468 | * <power fail> | |
4469 | * | |
4470 | * We also need this for cases where a snapshot entry is replaced by some other | |
4471 | * entry (file or directory) otherwise we end up with an unreplayable log due to | |
4472 | * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as | |
4473 | * if it were a regular entry: | |
4474 | * | |
4475 | * mkdir /mnt/x | |
4476 | * btrfs subvolume snapshot /mnt /mnt/x/snap | |
4477 | * btrfs subvolume delete /mnt/x/snap | |
4478 | * rmdir /mnt/x | |
4479 | * mkdir /mnt/x | |
4480 | * fsync /mnt/x or fsync some new file inside it | |
4481 | * <power fail> | |
4482 | * | |
4483 | * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in | |
4484 | * the same transaction. | |
4485 | */ | |
4486 | static int btrfs_check_ref_name_override(struct extent_buffer *eb, | |
4487 | const int slot, | |
4488 | const struct btrfs_key *key, | |
4489 | struct inode *inode, | |
4490 | u64 *other_ino) | |
4491 | { | |
4492 | int ret; | |
4493 | struct btrfs_path *search_path; | |
4494 | char *name = NULL; | |
4495 | u32 name_len = 0; | |
4496 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
4497 | u32 cur_offset = 0; | |
4498 | unsigned long ptr = btrfs_item_ptr_offset(eb, slot); | |
4499 | ||
4500 | search_path = btrfs_alloc_path(); | |
4501 | if (!search_path) | |
4502 | return -ENOMEM; | |
4503 | search_path->search_commit_root = 1; | |
4504 | search_path->skip_locking = 1; | |
4505 | ||
4506 | while (cur_offset < item_size) { | |
4507 | u64 parent; | |
4508 | u32 this_name_len; | |
4509 | u32 this_len; | |
4510 | unsigned long name_ptr; | |
4511 | struct btrfs_dir_item *di; | |
4512 | ||
4513 | if (key->type == BTRFS_INODE_REF_KEY) { | |
4514 | struct btrfs_inode_ref *iref; | |
4515 | ||
4516 | iref = (struct btrfs_inode_ref *)(ptr + cur_offset); | |
4517 | parent = key->offset; | |
4518 | this_name_len = btrfs_inode_ref_name_len(eb, iref); | |
4519 | name_ptr = (unsigned long)(iref + 1); | |
4520 | this_len = sizeof(*iref) + this_name_len; | |
4521 | } else { | |
4522 | struct btrfs_inode_extref *extref; | |
4523 | ||
4524 | extref = (struct btrfs_inode_extref *)(ptr + | |
4525 | cur_offset); | |
4526 | parent = btrfs_inode_extref_parent(eb, extref); | |
4527 | this_name_len = btrfs_inode_extref_name_len(eb, extref); | |
4528 | name_ptr = (unsigned long)&extref->name; | |
4529 | this_len = sizeof(*extref) + this_name_len; | |
4530 | } | |
4531 | ||
4532 | if (this_name_len > name_len) { | |
4533 | char *new_name; | |
4534 | ||
4535 | new_name = krealloc(name, this_name_len, GFP_NOFS); | |
4536 | if (!new_name) { | |
4537 | ret = -ENOMEM; | |
4538 | goto out; | |
4539 | } | |
4540 | name_len = this_name_len; | |
4541 | name = new_name; | |
4542 | } | |
4543 | ||
4544 | read_extent_buffer(eb, name, name_ptr, this_name_len); | |
4545 | di = btrfs_lookup_dir_item(NULL, BTRFS_I(inode)->root, | |
4546 | search_path, parent, | |
4547 | name, this_name_len, 0); | |
4548 | if (di && !IS_ERR(di)) { | |
4549 | struct btrfs_key di_key; | |
4550 | ||
4551 | btrfs_dir_item_key_to_cpu(search_path->nodes[0], | |
4552 | di, &di_key); | |
4553 | if (di_key.type == BTRFS_INODE_ITEM_KEY) { | |
4554 | ret = 1; | |
4555 | *other_ino = di_key.objectid; | |
4556 | } else { | |
4557 | ret = -EAGAIN; | |
4558 | } | |
4559 | goto out; | |
4560 | } else if (IS_ERR(di)) { | |
4561 | ret = PTR_ERR(di); | |
4562 | goto out; | |
4563 | } | |
4564 | btrfs_release_path(search_path); | |
4565 | ||
4566 | cur_offset += this_len; | |
4567 | } | |
4568 | ret = 0; | |
4569 | out: | |
4570 | btrfs_free_path(search_path); | |
4571 | kfree(name); | |
4572 | return ret; | |
4573 | } | |
4574 | ||
4575 | /* log a single inode in the tree log. | |
4576 | * At least one parent directory for this inode must exist in the tree | |
4577 | * or be logged already. | |
4578 | * | |
4579 | * Any items from this inode changed by the current transaction are copied | |
4580 | * to the log tree. An extra reference is taken on any extents in this | |
4581 | * file, allowing us to avoid a whole pile of corner cases around logging | |
4582 | * blocks that have been removed from the tree. | |
4583 | * | |
4584 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
4585 | * does. | |
4586 | * | |
4587 | * This handles both files and directories. | |
4588 | */ | |
4589 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, | |
4590 | struct btrfs_root *root, struct inode *inode, | |
4591 | int inode_only, | |
4592 | const loff_t start, | |
4593 | const loff_t end, | |
4594 | struct btrfs_log_ctx *ctx) | |
4595 | { | |
4596 | struct btrfs_path *path; | |
4597 | struct btrfs_path *dst_path; | |
4598 | struct btrfs_key min_key; | |
4599 | struct btrfs_key max_key; | |
4600 | struct btrfs_root *log = root->log_root; | |
4601 | struct extent_buffer *src = NULL; | |
4602 | LIST_HEAD(logged_list); | |
4603 | u64 last_extent = 0; | |
4604 | int err = 0; | |
4605 | int ret; | |
4606 | int nritems; | |
4607 | int ins_start_slot = 0; | |
4608 | int ins_nr; | |
4609 | bool fast_search = false; | |
4610 | u64 ino = btrfs_ino(inode); | |
4611 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
4612 | u64 logged_isize = 0; | |
4613 | bool need_log_inode_item = true; | |
4614 | ||
4615 | path = btrfs_alloc_path(); | |
4616 | if (!path) | |
4617 | return -ENOMEM; | |
4618 | dst_path = btrfs_alloc_path(); | |
4619 | if (!dst_path) { | |
4620 | btrfs_free_path(path); | |
4621 | return -ENOMEM; | |
4622 | } | |
4623 | ||
4624 | min_key.objectid = ino; | |
4625 | min_key.type = BTRFS_INODE_ITEM_KEY; | |
4626 | min_key.offset = 0; | |
4627 | ||
4628 | max_key.objectid = ino; | |
4629 | ||
4630 | ||
4631 | /* today the code can only do partial logging of directories */ | |
4632 | if (S_ISDIR(inode->i_mode) || | |
4633 | (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4634 | &BTRFS_I(inode)->runtime_flags) && | |
4635 | inode_only == LOG_INODE_EXISTS)) | |
4636 | max_key.type = BTRFS_XATTR_ITEM_KEY; | |
4637 | else | |
4638 | max_key.type = (u8)-1; | |
4639 | max_key.offset = (u64)-1; | |
4640 | ||
4641 | /* | |
4642 | * Only run delayed items if we are a dir or a new file. | |
4643 | * Otherwise commit the delayed inode only, which is needed in | |
4644 | * order for the log replay code to mark inodes for link count | |
4645 | * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items). | |
4646 | */ | |
4647 | if (S_ISDIR(inode->i_mode) || | |
4648 | BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) | |
4649 | ret = btrfs_commit_inode_delayed_items(trans, inode); | |
4650 | else | |
4651 | ret = btrfs_commit_inode_delayed_inode(inode); | |
4652 | ||
4653 | if (ret) { | |
4654 | btrfs_free_path(path); | |
4655 | btrfs_free_path(dst_path); | |
4656 | return ret; | |
4657 | } | |
4658 | ||
4659 | mutex_lock(&BTRFS_I(inode)->log_mutex); | |
4660 | ||
4661 | /* | |
4662 | * a brute force approach to making sure we get the most uptodate | |
4663 | * copies of everything. | |
4664 | */ | |
4665 | if (S_ISDIR(inode->i_mode)) { | |
4666 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
4667 | ||
4668 | if (inode_only == LOG_INODE_EXISTS) | |
4669 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
4670 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); | |
4671 | } else { | |
4672 | if (inode_only == LOG_INODE_EXISTS) { | |
4673 | /* | |
4674 | * Make sure the new inode item we write to the log has | |
4675 | * the same isize as the current one (if it exists). | |
4676 | * This is necessary to prevent data loss after log | |
4677 | * replay, and also to prevent doing a wrong expanding | |
4678 | * truncate - for e.g. create file, write 4K into offset | |
4679 | * 0, fsync, write 4K into offset 4096, add hard link, | |
4680 | * fsync some other file (to sync log), power fail - if | |
4681 | * we use the inode's current i_size, after log replay | |
4682 | * we get a 8Kb file, with the last 4Kb extent as a hole | |
4683 | * (zeroes), as if an expanding truncate happened, | |
4684 | * instead of getting a file of 4Kb only. | |
4685 | */ | |
4686 | err = logged_inode_size(log, inode, path, | |
4687 | &logged_isize); | |
4688 | if (err) | |
4689 | goto out_unlock; | |
4690 | } | |
4691 | if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4692 | &BTRFS_I(inode)->runtime_flags)) { | |
4693 | if (inode_only == LOG_INODE_EXISTS) { | |
4694 | max_key.type = BTRFS_XATTR_ITEM_KEY; | |
4695 | ret = drop_objectid_items(trans, log, path, ino, | |
4696 | max_key.type); | |
4697 | } else { | |
4698 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4699 | &BTRFS_I(inode)->runtime_flags); | |
4700 | clear_bit(BTRFS_INODE_COPY_EVERYTHING, | |
4701 | &BTRFS_I(inode)->runtime_flags); | |
4702 | while(1) { | |
4703 | ret = btrfs_truncate_inode_items(trans, | |
4704 | log, inode, 0, 0); | |
4705 | if (ret != -EAGAIN) | |
4706 | break; | |
4707 | } | |
4708 | } | |
4709 | } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING, | |
4710 | &BTRFS_I(inode)->runtime_flags) || | |
4711 | inode_only == LOG_INODE_EXISTS) { | |
4712 | if (inode_only == LOG_INODE_ALL) | |
4713 | fast_search = true; | |
4714 | max_key.type = BTRFS_XATTR_ITEM_KEY; | |
4715 | ret = drop_objectid_items(trans, log, path, ino, | |
4716 | max_key.type); | |
4717 | } else { | |
4718 | if (inode_only == LOG_INODE_ALL) | |
4719 | fast_search = true; | |
4720 | goto log_extents; | |
4721 | } | |
4722 | ||
4723 | } | |
4724 | if (ret) { | |
4725 | err = ret; | |
4726 | goto out_unlock; | |
4727 | } | |
4728 | ||
4729 | while (1) { | |
4730 | ins_nr = 0; | |
4731 | ret = btrfs_search_forward(root, &min_key, | |
4732 | path, trans->transid); | |
4733 | if (ret < 0) { | |
4734 | err = ret; | |
4735 | goto out_unlock; | |
4736 | } | |
4737 | if (ret != 0) | |
4738 | break; | |
4739 | again: | |
4740 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ | |
4741 | if (min_key.objectid != ino) | |
4742 | break; | |
4743 | if (min_key.type > max_key.type) | |
4744 | break; | |
4745 | ||
4746 | if (min_key.type == BTRFS_INODE_ITEM_KEY) | |
4747 | need_log_inode_item = false; | |
4748 | ||
4749 | if ((min_key.type == BTRFS_INODE_REF_KEY || | |
4750 | min_key.type == BTRFS_INODE_EXTREF_KEY) && | |
4751 | BTRFS_I(inode)->generation == trans->transid) { | |
4752 | u64 other_ino = 0; | |
4753 | ||
4754 | ret = btrfs_check_ref_name_override(path->nodes[0], | |
4755 | path->slots[0], | |
4756 | &min_key, inode, | |
4757 | &other_ino); | |
4758 | if (ret < 0) { | |
4759 | err = ret; | |
4760 | goto out_unlock; | |
4761 | } else if (ret > 0 && ctx && | |
4762 | other_ino != btrfs_ino(ctx->inode)) { | |
4763 | struct btrfs_key inode_key; | |
4764 | struct inode *other_inode; | |
4765 | ||
4766 | if (ins_nr > 0) { | |
4767 | ins_nr++; | |
4768 | } else { | |
4769 | ins_nr = 1; | |
4770 | ins_start_slot = path->slots[0]; | |
4771 | } | |
4772 | ret = copy_items(trans, inode, dst_path, path, | |
4773 | &last_extent, ins_start_slot, | |
4774 | ins_nr, inode_only, | |
4775 | logged_isize); | |
4776 | if (ret < 0) { | |
4777 | err = ret; | |
4778 | goto out_unlock; | |
4779 | } | |
4780 | ins_nr = 0; | |
4781 | btrfs_release_path(path); | |
4782 | inode_key.objectid = other_ino; | |
4783 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
4784 | inode_key.offset = 0; | |
4785 | other_inode = btrfs_iget(root->fs_info->sb, | |
4786 | &inode_key, root, | |
4787 | NULL); | |
4788 | /* | |
4789 | * If the other inode that had a conflicting dir | |
4790 | * entry was deleted in the current transaction, | |
4791 | * we don't need to do more work nor fallback to | |
4792 | * a transaction commit. | |
4793 | */ | |
4794 | if (IS_ERR(other_inode) && | |
4795 | PTR_ERR(other_inode) == -ENOENT) { | |
4796 | goto next_key; | |
4797 | } else if (IS_ERR(other_inode)) { | |
4798 | err = PTR_ERR(other_inode); | |
4799 | goto out_unlock; | |
4800 | } | |
4801 | /* | |
4802 | * We are safe logging the other inode without | |
4803 | * acquiring its i_mutex as long as we log with | |
4804 | * the LOG_INODE_EXISTS mode. We're safe against | |
4805 | * concurrent renames of the other inode as well | |
4806 | * because during a rename we pin the log and | |
4807 | * update the log with the new name before we | |
4808 | * unpin it. | |
4809 | */ | |
4810 | err = btrfs_log_inode(trans, root, other_inode, | |
4811 | LOG_INODE_EXISTS, | |
4812 | 0, LLONG_MAX, ctx); | |
4813 | iput(other_inode); | |
4814 | if (err) | |
4815 | goto out_unlock; | |
4816 | else | |
4817 | goto next_key; | |
4818 | } | |
4819 | } | |
4820 | ||
4821 | /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */ | |
4822 | if (min_key.type == BTRFS_XATTR_ITEM_KEY) { | |
4823 | if (ins_nr == 0) | |
4824 | goto next_slot; | |
4825 | ret = copy_items(trans, inode, dst_path, path, | |
4826 | &last_extent, ins_start_slot, | |
4827 | ins_nr, inode_only, logged_isize); | |
4828 | if (ret < 0) { | |
4829 | err = ret; | |
4830 | goto out_unlock; | |
4831 | } | |
4832 | ins_nr = 0; | |
4833 | if (ret) { | |
4834 | btrfs_release_path(path); | |
4835 | continue; | |
4836 | } | |
4837 | goto next_slot; | |
4838 | } | |
4839 | ||
4840 | src = path->nodes[0]; | |
4841 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { | |
4842 | ins_nr++; | |
4843 | goto next_slot; | |
4844 | } else if (!ins_nr) { | |
4845 | ins_start_slot = path->slots[0]; | |
4846 | ins_nr = 1; | |
4847 | goto next_slot; | |
4848 | } | |
4849 | ||
4850 | ret = copy_items(trans, inode, dst_path, path, &last_extent, | |
4851 | ins_start_slot, ins_nr, inode_only, | |
4852 | logged_isize); | |
4853 | if (ret < 0) { | |
4854 | err = ret; | |
4855 | goto out_unlock; | |
4856 | } | |
4857 | if (ret) { | |
4858 | ins_nr = 0; | |
4859 | btrfs_release_path(path); | |
4860 | continue; | |
4861 | } | |
4862 | ins_nr = 1; | |
4863 | ins_start_slot = path->slots[0]; | |
4864 | next_slot: | |
4865 | ||
4866 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4867 | path->slots[0]++; | |
4868 | if (path->slots[0] < nritems) { | |
4869 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, | |
4870 | path->slots[0]); | |
4871 | goto again; | |
4872 | } | |
4873 | if (ins_nr) { | |
4874 | ret = copy_items(trans, inode, dst_path, path, | |
4875 | &last_extent, ins_start_slot, | |
4876 | ins_nr, inode_only, logged_isize); | |
4877 | if (ret < 0) { | |
4878 | err = ret; | |
4879 | goto out_unlock; | |
4880 | } | |
4881 | ret = 0; | |
4882 | ins_nr = 0; | |
4883 | } | |
4884 | btrfs_release_path(path); | |
4885 | next_key: | |
4886 | if (min_key.offset < (u64)-1) { | |
4887 | min_key.offset++; | |
4888 | } else if (min_key.type < max_key.type) { | |
4889 | min_key.type++; | |
4890 | min_key.offset = 0; | |
4891 | } else { | |
4892 | break; | |
4893 | } | |
4894 | } | |
4895 | if (ins_nr) { | |
4896 | ret = copy_items(trans, inode, dst_path, path, &last_extent, | |
4897 | ins_start_slot, ins_nr, inode_only, | |
4898 | logged_isize); | |
4899 | if (ret < 0) { | |
4900 | err = ret; | |
4901 | goto out_unlock; | |
4902 | } | |
4903 | ret = 0; | |
4904 | ins_nr = 0; | |
4905 | } | |
4906 | ||
4907 | btrfs_release_path(path); | |
4908 | btrfs_release_path(dst_path); | |
4909 | err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path); | |
4910 | if (err) | |
4911 | goto out_unlock; | |
4912 | if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) { | |
4913 | btrfs_release_path(path); | |
4914 | btrfs_release_path(dst_path); | |
4915 | err = btrfs_log_trailing_hole(trans, root, inode, path); | |
4916 | if (err) | |
4917 | goto out_unlock; | |
4918 | } | |
4919 | log_extents: | |
4920 | btrfs_release_path(path); | |
4921 | btrfs_release_path(dst_path); | |
4922 | if (need_log_inode_item) { | |
4923 | err = log_inode_item(trans, log, dst_path, inode); | |
4924 | if (err) | |
4925 | goto out_unlock; | |
4926 | } | |
4927 | if (fast_search) { | |
4928 | ret = btrfs_log_changed_extents(trans, root, inode, dst_path, | |
4929 | &logged_list, ctx, start, end); | |
4930 | if (ret) { | |
4931 | err = ret; | |
4932 | goto out_unlock; | |
4933 | } | |
4934 | } else if (inode_only == LOG_INODE_ALL) { | |
4935 | struct extent_map *em, *n; | |
4936 | ||
4937 | write_lock(&em_tree->lock); | |
4938 | /* | |
4939 | * We can't just remove every em if we're called for a ranged | |
4940 | * fsync - that is, one that doesn't cover the whole possible | |
4941 | * file range (0 to LLONG_MAX). This is because we can have | |
4942 | * em's that fall outside the range we're logging and therefore | |
4943 | * their ordered operations haven't completed yet | |
4944 | * (btrfs_finish_ordered_io() not invoked yet). This means we | |
4945 | * didn't get their respective file extent item in the fs/subvol | |
4946 | * tree yet, and need to let the next fast fsync (one which | |
4947 | * consults the list of modified extent maps) find the em so | |
4948 | * that it logs a matching file extent item and waits for the | |
4949 | * respective ordered operation to complete (if it's still | |
4950 | * running). | |
4951 | * | |
4952 | * Removing every em outside the range we're logging would make | |
4953 | * the next fast fsync not log their matching file extent items, | |
4954 | * therefore making us lose data after a log replay. | |
4955 | */ | |
4956 | list_for_each_entry_safe(em, n, &em_tree->modified_extents, | |
4957 | list) { | |
4958 | const u64 mod_end = em->mod_start + em->mod_len - 1; | |
4959 | ||
4960 | if (em->mod_start >= start && mod_end <= end) | |
4961 | list_del_init(&em->list); | |
4962 | } | |
4963 | write_unlock(&em_tree->lock); | |
4964 | } | |
4965 | ||
4966 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) { | |
4967 | ret = log_directory_changes(trans, root, inode, path, dst_path, | |
4968 | ctx); | |
4969 | if (ret) { | |
4970 | err = ret; | |
4971 | goto out_unlock; | |
4972 | } | |
4973 | } | |
4974 | ||
4975 | spin_lock(&BTRFS_I(inode)->lock); | |
4976 | BTRFS_I(inode)->logged_trans = trans->transid; | |
4977 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans; | |
4978 | spin_unlock(&BTRFS_I(inode)->lock); | |
4979 | out_unlock: | |
4980 | if (unlikely(err)) | |
4981 | btrfs_put_logged_extents(&logged_list); | |
4982 | else | |
4983 | btrfs_submit_logged_extents(&logged_list, log); | |
4984 | mutex_unlock(&BTRFS_I(inode)->log_mutex); | |
4985 | ||
4986 | btrfs_free_path(path); | |
4987 | btrfs_free_path(dst_path); | |
4988 | return err; | |
4989 | } | |
4990 | ||
4991 | /* | |
4992 | * Check if we must fallback to a transaction commit when logging an inode. | |
4993 | * This must be called after logging the inode and is used only in the context | |
4994 | * when fsyncing an inode requires the need to log some other inode - in which | |
4995 | * case we can't lock the i_mutex of each other inode we need to log as that | |
4996 | * can lead to deadlocks with concurrent fsync against other inodes (as we can | |
4997 | * log inodes up or down in the hierarchy) or rename operations for example. So | |
4998 | * we take the log_mutex of the inode after we have logged it and then check for | |
4999 | * its last_unlink_trans value - this is safe because any task setting | |
5000 | * last_unlink_trans must take the log_mutex and it must do this before it does | |
5001 | * the actual unlink operation, so if we do this check before a concurrent task | |
5002 | * sets last_unlink_trans it means we've logged a consistent version/state of | |
5003 | * all the inode items, otherwise we are not sure and must do a transaction | |
5004 | * commit (the concurrent task might have only updated last_unlink_trans before | |
5005 | * we logged the inode or it might have also done the unlink). | |
5006 | */ | |
5007 | static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans, | |
5008 | struct inode *inode) | |
5009 | { | |
5010 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
5011 | bool ret = false; | |
5012 | ||
5013 | mutex_lock(&BTRFS_I(inode)->log_mutex); | |
5014 | if (BTRFS_I(inode)->last_unlink_trans > fs_info->last_trans_committed) { | |
5015 | /* | |
5016 | * Make sure any commits to the log are forced to be full | |
5017 | * commits. | |
5018 | */ | |
5019 | btrfs_set_log_full_commit(fs_info, trans); | |
5020 | ret = true; | |
5021 | } | |
5022 | mutex_unlock(&BTRFS_I(inode)->log_mutex); | |
5023 | ||
5024 | return ret; | |
5025 | } | |
5026 | ||
5027 | /* | |
5028 | * follow the dentry parent pointers up the chain and see if any | |
5029 | * of the directories in it require a full commit before they can | |
5030 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
5031 | * a full commit is required. | |
5032 | */ | |
5033 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
5034 | struct inode *inode, | |
5035 | struct dentry *parent, | |
5036 | struct super_block *sb, | |
5037 | u64 last_committed) | |
5038 | { | |
5039 | int ret = 0; | |
5040 | struct dentry *old_parent = NULL; | |
5041 | struct inode *orig_inode = inode; | |
5042 | ||
5043 | /* | |
5044 | * for regular files, if its inode is already on disk, we don't | |
5045 | * have to worry about the parents at all. This is because | |
5046 | * we can use the last_unlink_trans field to record renames | |
5047 | * and other fun in this file. | |
5048 | */ | |
5049 | if (S_ISREG(inode->i_mode) && | |
5050 | BTRFS_I(inode)->generation <= last_committed && | |
5051 | BTRFS_I(inode)->last_unlink_trans <= last_committed) | |
5052 | goto out; | |
5053 | ||
5054 | if (!S_ISDIR(inode->i_mode)) { | |
5055 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) | |
5056 | goto out; | |
5057 | inode = d_inode(parent); | |
5058 | } | |
5059 | ||
5060 | while (1) { | |
5061 | /* | |
5062 | * If we are logging a directory then we start with our inode, | |
5063 | * not our parent's inode, so we need to skip setting the | |
5064 | * logged_trans so that further down in the log code we don't | |
5065 | * think this inode has already been logged. | |
5066 | */ | |
5067 | if (inode != orig_inode) | |
5068 | BTRFS_I(inode)->logged_trans = trans->transid; | |
5069 | smp_mb(); | |
5070 | ||
5071 | if (btrfs_must_commit_transaction(trans, inode)) { | |
5072 | ret = 1; | |
5073 | break; | |
5074 | } | |
5075 | ||
5076 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) | |
5077 | break; | |
5078 | ||
5079 | if (IS_ROOT(parent)) { | |
5080 | inode = d_inode(parent); | |
5081 | if (btrfs_must_commit_transaction(trans, inode)) | |
5082 | ret = 1; | |
5083 | break; | |
5084 | } | |
5085 | ||
5086 | parent = dget_parent(parent); | |
5087 | dput(old_parent); | |
5088 | old_parent = parent; | |
5089 | inode = d_inode(parent); | |
5090 | ||
5091 | } | |
5092 | dput(old_parent); | |
5093 | out: | |
5094 | return ret; | |
5095 | } | |
5096 | ||
5097 | struct btrfs_dir_list { | |
5098 | u64 ino; | |
5099 | struct list_head list; | |
5100 | }; | |
5101 | ||
5102 | /* | |
5103 | * Log the inodes of the new dentries of a directory. See log_dir_items() for | |
5104 | * details about the why it is needed. | |
5105 | * This is a recursive operation - if an existing dentry corresponds to a | |
5106 | * directory, that directory's new entries are logged too (same behaviour as | |
5107 | * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes | |
5108 | * the dentries point to we do not lock their i_mutex, otherwise lockdep | |
5109 | * complains about the following circular lock dependency / possible deadlock: | |
5110 | * | |
5111 | * CPU0 CPU1 | |
5112 | * ---- ---- | |
5113 | * lock(&type->i_mutex_dir_key#3/2); | |
5114 | * lock(sb_internal#2); | |
5115 | * lock(&type->i_mutex_dir_key#3/2); | |
5116 | * lock(&sb->s_type->i_mutex_key#14); | |
5117 | * | |
5118 | * Where sb_internal is the lock (a counter that works as a lock) acquired by | |
5119 | * sb_start_intwrite() in btrfs_start_transaction(). | |
5120 | * Not locking i_mutex of the inodes is still safe because: | |
5121 | * | |
5122 | * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible | |
5123 | * that while logging the inode new references (names) are added or removed | |
5124 | * from the inode, leaving the logged inode item with a link count that does | |
5125 | * not match the number of logged inode reference items. This is fine because | |
5126 | * at log replay time we compute the real number of links and correct the | |
5127 | * link count in the inode item (see replay_one_buffer() and | |
5128 | * link_to_fixup_dir()); | |
5129 | * | |
5130 | * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that | |
5131 | * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and | |
5132 | * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item | |
5133 | * has a size that doesn't match the sum of the lengths of all the logged | |
5134 | * names. This does not result in a problem because if a dir_item key is | |
5135 | * logged but its matching dir_index key is not logged, at log replay time we | |
5136 | * don't use it to replay the respective name (see replay_one_name()). On the | |
5137 | * other hand if only the dir_index key ends up being logged, the respective | |
5138 | * name is added to the fs/subvol tree with both the dir_item and dir_index | |
5139 | * keys created (see replay_one_name()). | |
5140 | * The directory's inode item with a wrong i_size is not a problem as well, | |
5141 | * since we don't use it at log replay time to set the i_size in the inode | |
5142 | * item of the fs/subvol tree (see overwrite_item()). | |
5143 | */ | |
5144 | static int log_new_dir_dentries(struct btrfs_trans_handle *trans, | |
5145 | struct btrfs_root *root, | |
5146 | struct inode *start_inode, | |
5147 | struct btrfs_log_ctx *ctx) | |
5148 | { | |
5149 | struct btrfs_root *log = root->log_root; | |
5150 | struct btrfs_path *path; | |
5151 | LIST_HEAD(dir_list); | |
5152 | struct btrfs_dir_list *dir_elem; | |
5153 | int ret = 0; | |
5154 | ||
5155 | path = btrfs_alloc_path(); | |
5156 | if (!path) | |
5157 | return -ENOMEM; | |
5158 | ||
5159 | dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS); | |
5160 | if (!dir_elem) { | |
5161 | btrfs_free_path(path); | |
5162 | return -ENOMEM; | |
5163 | } | |
5164 | dir_elem->ino = btrfs_ino(start_inode); | |
5165 | list_add_tail(&dir_elem->list, &dir_list); | |
5166 | ||
5167 | while (!list_empty(&dir_list)) { | |
5168 | struct extent_buffer *leaf; | |
5169 | struct btrfs_key min_key; | |
5170 | int nritems; | |
5171 | int i; | |
5172 | ||
5173 | dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list, | |
5174 | list); | |
5175 | if (ret) | |
5176 | goto next_dir_inode; | |
5177 | ||
5178 | min_key.objectid = dir_elem->ino; | |
5179 | min_key.type = BTRFS_DIR_ITEM_KEY; | |
5180 | min_key.offset = 0; | |
5181 | again: | |
5182 | btrfs_release_path(path); | |
5183 | ret = btrfs_search_forward(log, &min_key, path, trans->transid); | |
5184 | if (ret < 0) { | |
5185 | goto next_dir_inode; | |
5186 | } else if (ret > 0) { | |
5187 | ret = 0; | |
5188 | goto next_dir_inode; | |
5189 | } | |
5190 | ||
5191 | process_leaf: | |
5192 | leaf = path->nodes[0]; | |
5193 | nritems = btrfs_header_nritems(leaf); | |
5194 | for (i = path->slots[0]; i < nritems; i++) { | |
5195 | struct btrfs_dir_item *di; | |
5196 | struct btrfs_key di_key; | |
5197 | struct inode *di_inode; | |
5198 | struct btrfs_dir_list *new_dir_elem; | |
5199 | int log_mode = LOG_INODE_EXISTS; | |
5200 | int type; | |
5201 | ||
5202 | btrfs_item_key_to_cpu(leaf, &min_key, i); | |
5203 | if (min_key.objectid != dir_elem->ino || | |
5204 | min_key.type != BTRFS_DIR_ITEM_KEY) | |
5205 | goto next_dir_inode; | |
5206 | ||
5207 | di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item); | |
5208 | type = btrfs_dir_type(leaf, di); | |
5209 | if (btrfs_dir_transid(leaf, di) < trans->transid && | |
5210 | type != BTRFS_FT_DIR) | |
5211 | continue; | |
5212 | btrfs_dir_item_key_to_cpu(leaf, di, &di_key); | |
5213 | if (di_key.type == BTRFS_ROOT_ITEM_KEY) | |
5214 | continue; | |
5215 | ||
5216 | di_inode = btrfs_iget(root->fs_info->sb, &di_key, | |
5217 | root, NULL); | |
5218 | if (IS_ERR(di_inode)) { | |
5219 | ret = PTR_ERR(di_inode); | |
5220 | goto next_dir_inode; | |
5221 | } | |
5222 | ||
5223 | if (btrfs_inode_in_log(di_inode, trans->transid)) { | |
5224 | iput(di_inode); | |
5225 | continue; | |
5226 | } | |
5227 | ||
5228 | ctx->log_new_dentries = false; | |
5229 | if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK) | |
5230 | log_mode = LOG_INODE_ALL; | |
5231 | btrfs_release_path(path); | |
5232 | ret = btrfs_log_inode(trans, root, di_inode, | |
5233 | log_mode, 0, LLONG_MAX, ctx); | |
5234 | if (!ret && | |
5235 | btrfs_must_commit_transaction(trans, di_inode)) | |
5236 | ret = 1; | |
5237 | iput(di_inode); | |
5238 | if (ret) | |
5239 | goto next_dir_inode; | |
5240 | if (ctx->log_new_dentries) { | |
5241 | new_dir_elem = kmalloc(sizeof(*new_dir_elem), | |
5242 | GFP_NOFS); | |
5243 | if (!new_dir_elem) { | |
5244 | ret = -ENOMEM; | |
5245 | goto next_dir_inode; | |
5246 | } | |
5247 | new_dir_elem->ino = di_key.objectid; | |
5248 | list_add_tail(&new_dir_elem->list, &dir_list); | |
5249 | } | |
5250 | break; | |
5251 | } | |
5252 | if (i == nritems) { | |
5253 | ret = btrfs_next_leaf(log, path); | |
5254 | if (ret < 0) { | |
5255 | goto next_dir_inode; | |
5256 | } else if (ret > 0) { | |
5257 | ret = 0; | |
5258 | goto next_dir_inode; | |
5259 | } | |
5260 | goto process_leaf; | |
5261 | } | |
5262 | if (min_key.offset < (u64)-1) { | |
5263 | min_key.offset++; | |
5264 | goto again; | |
5265 | } | |
5266 | next_dir_inode: | |
5267 | list_del(&dir_elem->list); | |
5268 | kfree(dir_elem); | |
5269 | } | |
5270 | ||
5271 | btrfs_free_path(path); | |
5272 | return ret; | |
5273 | } | |
5274 | ||
5275 | static int btrfs_log_all_parents(struct btrfs_trans_handle *trans, | |
5276 | struct inode *inode, | |
5277 | struct btrfs_log_ctx *ctx) | |
5278 | { | |
5279 | int ret; | |
5280 | struct btrfs_path *path; | |
5281 | struct btrfs_key key; | |
5282 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5283 | const u64 ino = btrfs_ino(inode); | |
5284 | ||
5285 | path = btrfs_alloc_path(); | |
5286 | if (!path) | |
5287 | return -ENOMEM; | |
5288 | path->skip_locking = 1; | |
5289 | path->search_commit_root = 1; | |
5290 | ||
5291 | key.objectid = ino; | |
5292 | key.type = BTRFS_INODE_REF_KEY; | |
5293 | key.offset = 0; | |
5294 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5295 | if (ret < 0) | |
5296 | goto out; | |
5297 | ||
5298 | while (true) { | |
5299 | struct extent_buffer *leaf = path->nodes[0]; | |
5300 | int slot = path->slots[0]; | |
5301 | u32 cur_offset = 0; | |
5302 | u32 item_size; | |
5303 | unsigned long ptr; | |
5304 | ||
5305 | if (slot >= btrfs_header_nritems(leaf)) { | |
5306 | ret = btrfs_next_leaf(root, path); | |
5307 | if (ret < 0) | |
5308 | goto out; | |
5309 | else if (ret > 0) | |
5310 | break; | |
5311 | continue; | |
5312 | } | |
5313 | ||
5314 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5315 | /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */ | |
5316 | if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY) | |
5317 | break; | |
5318 | ||
5319 | item_size = btrfs_item_size_nr(leaf, slot); | |
5320 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
5321 | while (cur_offset < item_size) { | |
5322 | struct btrfs_key inode_key; | |
5323 | struct inode *dir_inode; | |
5324 | ||
5325 | inode_key.type = BTRFS_INODE_ITEM_KEY; | |
5326 | inode_key.offset = 0; | |
5327 | ||
5328 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
5329 | struct btrfs_inode_extref *extref; | |
5330 | ||
5331 | extref = (struct btrfs_inode_extref *) | |
5332 | (ptr + cur_offset); | |
5333 | inode_key.objectid = btrfs_inode_extref_parent( | |
5334 | leaf, extref); | |
5335 | cur_offset += sizeof(*extref); | |
5336 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
5337 | extref); | |
5338 | } else { | |
5339 | inode_key.objectid = key.offset; | |
5340 | cur_offset = item_size; | |
5341 | } | |
5342 | ||
5343 | dir_inode = btrfs_iget(root->fs_info->sb, &inode_key, | |
5344 | root, NULL); | |
5345 | /* If parent inode was deleted, skip it. */ | |
5346 | if (IS_ERR(dir_inode)) | |
5347 | continue; | |
5348 | ||
5349 | if (ctx) | |
5350 | ctx->log_new_dentries = false; | |
5351 | ret = btrfs_log_inode(trans, root, dir_inode, | |
5352 | LOG_INODE_ALL, 0, LLONG_MAX, ctx); | |
5353 | if (!ret && | |
5354 | btrfs_must_commit_transaction(trans, dir_inode)) | |
5355 | ret = 1; | |
5356 | if (!ret && ctx && ctx->log_new_dentries) | |
5357 | ret = log_new_dir_dentries(trans, root, | |
5358 | dir_inode, ctx); | |
5359 | iput(dir_inode); | |
5360 | if (ret) | |
5361 | goto out; | |
5362 | } | |
5363 | path->slots[0]++; | |
5364 | } | |
5365 | ret = 0; | |
5366 | out: | |
5367 | btrfs_free_path(path); | |
5368 | return ret; | |
5369 | } | |
5370 | ||
5371 | /* | |
5372 | * helper function around btrfs_log_inode to make sure newly created | |
5373 | * parent directories also end up in the log. A minimal inode and backref | |
5374 | * only logging is done of any parent directories that are older than | |
5375 | * the last committed transaction | |
5376 | */ | |
5377 | static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, | |
5378 | struct btrfs_root *root, struct inode *inode, | |
5379 | struct dentry *parent, | |
5380 | const loff_t start, | |
5381 | const loff_t end, | |
5382 | int exists_only, | |
5383 | struct btrfs_log_ctx *ctx) | |
5384 | { | |
5385 | int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL; | |
5386 | struct super_block *sb; | |
5387 | struct dentry *old_parent = NULL; | |
5388 | int ret = 0; | |
5389 | u64 last_committed = root->fs_info->last_trans_committed; | |
5390 | bool log_dentries = false; | |
5391 | struct inode *orig_inode = inode; | |
5392 | ||
5393 | sb = inode->i_sb; | |
5394 | ||
5395 | if (btrfs_test_opt(root->fs_info, NOTREELOG)) { | |
5396 | ret = 1; | |
5397 | goto end_no_trans; | |
5398 | } | |
5399 | ||
5400 | /* | |
5401 | * The prev transaction commit doesn't complete, we need do | |
5402 | * full commit by ourselves. | |
5403 | */ | |
5404 | if (root->fs_info->last_trans_log_full_commit > | |
5405 | root->fs_info->last_trans_committed) { | |
5406 | ret = 1; | |
5407 | goto end_no_trans; | |
5408 | } | |
5409 | ||
5410 | if (root != BTRFS_I(inode)->root || | |
5411 | btrfs_root_refs(&root->root_item) == 0) { | |
5412 | ret = 1; | |
5413 | goto end_no_trans; | |
5414 | } | |
5415 | ||
5416 | ret = check_parent_dirs_for_sync(trans, inode, parent, | |
5417 | sb, last_committed); | |
5418 | if (ret) | |
5419 | goto end_no_trans; | |
5420 | ||
5421 | if (btrfs_inode_in_log(inode, trans->transid)) { | |
5422 | ret = BTRFS_NO_LOG_SYNC; | |
5423 | goto end_no_trans; | |
5424 | } | |
5425 | ||
5426 | ret = start_log_trans(trans, root, ctx); | |
5427 | if (ret) | |
5428 | goto end_no_trans; | |
5429 | ||
5430 | ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx); | |
5431 | if (ret) | |
5432 | goto end_trans; | |
5433 | ||
5434 | /* | |
5435 | * for regular files, if its inode is already on disk, we don't | |
5436 | * have to worry about the parents at all. This is because | |
5437 | * we can use the last_unlink_trans field to record renames | |
5438 | * and other fun in this file. | |
5439 | */ | |
5440 | if (S_ISREG(inode->i_mode) && | |
5441 | BTRFS_I(inode)->generation <= last_committed && | |
5442 | BTRFS_I(inode)->last_unlink_trans <= last_committed) { | |
5443 | ret = 0; | |
5444 | goto end_trans; | |
5445 | } | |
5446 | ||
5447 | if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries) | |
5448 | log_dentries = true; | |
5449 | ||
5450 | /* | |
5451 | * On unlink we must make sure all our current and old parent directory | |
5452 | * inodes are fully logged. This is to prevent leaving dangling | |
5453 | * directory index entries in directories that were our parents but are | |
5454 | * not anymore. Not doing this results in old parent directory being | |
5455 | * impossible to delete after log replay (rmdir will always fail with | |
5456 | * error -ENOTEMPTY). | |
5457 | * | |
5458 | * Example 1: | |
5459 | * | |
5460 | * mkdir testdir | |
5461 | * touch testdir/foo | |
5462 | * ln testdir/foo testdir/bar | |
5463 | * sync | |
5464 | * unlink testdir/bar | |
5465 | * xfs_io -c fsync testdir/foo | |
5466 | * <power failure> | |
5467 | * mount fs, triggers log replay | |
5468 | * | |
5469 | * If we don't log the parent directory (testdir), after log replay the | |
5470 | * directory still has an entry pointing to the file inode using the bar | |
5471 | * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and | |
5472 | * the file inode has a link count of 1. | |
5473 | * | |
5474 | * Example 2: | |
5475 | * | |
5476 | * mkdir testdir | |
5477 | * touch foo | |
5478 | * ln foo testdir/foo2 | |
5479 | * ln foo testdir/foo3 | |
5480 | * sync | |
5481 | * unlink testdir/foo3 | |
5482 | * xfs_io -c fsync foo | |
5483 | * <power failure> | |
5484 | * mount fs, triggers log replay | |
5485 | * | |
5486 | * Similar as the first example, after log replay the parent directory | |
5487 | * testdir still has an entry pointing to the inode file with name foo3 | |
5488 | * but the file inode does not have a matching BTRFS_INODE_REF_KEY item | |
5489 | * and has a link count of 2. | |
5490 | */ | |
5491 | if (BTRFS_I(inode)->last_unlink_trans > last_committed) { | |
5492 | ret = btrfs_log_all_parents(trans, orig_inode, ctx); | |
5493 | if (ret) | |
5494 | goto end_trans; | |
5495 | } | |
5496 | ||
5497 | while (1) { | |
5498 | if (!parent || d_really_is_negative(parent) || sb != parent->d_sb) | |
5499 | break; | |
5500 | ||
5501 | inode = d_inode(parent); | |
5502 | if (root != BTRFS_I(inode)->root) | |
5503 | break; | |
5504 | ||
5505 | if (BTRFS_I(inode)->generation > last_committed) { | |
5506 | ret = btrfs_log_inode(trans, root, inode, | |
5507 | LOG_INODE_EXISTS, | |
5508 | 0, LLONG_MAX, ctx); | |
5509 | if (ret) | |
5510 | goto end_trans; | |
5511 | } | |
5512 | if (IS_ROOT(parent)) | |
5513 | break; | |
5514 | ||
5515 | parent = dget_parent(parent); | |
5516 | dput(old_parent); | |
5517 | old_parent = parent; | |
5518 | } | |
5519 | if (log_dentries) | |
5520 | ret = log_new_dir_dentries(trans, root, orig_inode, ctx); | |
5521 | else | |
5522 | ret = 0; | |
5523 | end_trans: | |
5524 | dput(old_parent); | |
5525 | if (ret < 0) { | |
5526 | btrfs_set_log_full_commit(root->fs_info, trans); | |
5527 | ret = 1; | |
5528 | } | |
5529 | ||
5530 | if (ret) | |
5531 | btrfs_remove_log_ctx(root, ctx); | |
5532 | btrfs_end_log_trans(root); | |
5533 | end_no_trans: | |
5534 | return ret; | |
5535 | } | |
5536 | ||
5537 | /* | |
5538 | * it is not safe to log dentry if the chunk root has added new | |
5539 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
5540 | * If this returns 1, you must commit the transaction to safely get your | |
5541 | * data on disk. | |
5542 | */ | |
5543 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
5544 | struct btrfs_root *root, struct dentry *dentry, | |
5545 | const loff_t start, | |
5546 | const loff_t end, | |
5547 | struct btrfs_log_ctx *ctx) | |
5548 | { | |
5549 | struct dentry *parent = dget_parent(dentry); | |
5550 | int ret; | |
5551 | ||
5552 | ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent, | |
5553 | start, end, 0, ctx); | |
5554 | dput(parent); | |
5555 | ||
5556 | return ret; | |
5557 | } | |
5558 | ||
5559 | /* | |
5560 | * should be called during mount to recover any replay any log trees | |
5561 | * from the FS | |
5562 | */ | |
5563 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
5564 | { | |
5565 | int ret; | |
5566 | struct btrfs_path *path; | |
5567 | struct btrfs_trans_handle *trans; | |
5568 | struct btrfs_key key; | |
5569 | struct btrfs_key found_key; | |
5570 | struct btrfs_key tmp_key; | |
5571 | struct btrfs_root *log; | |
5572 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
5573 | struct walk_control wc = { | |
5574 | .process_func = process_one_buffer, | |
5575 | .stage = 0, | |
5576 | }; | |
5577 | ||
5578 | path = btrfs_alloc_path(); | |
5579 | if (!path) | |
5580 | return -ENOMEM; | |
5581 | ||
5582 | fs_info->log_root_recovering = 1; | |
5583 | ||
5584 | trans = btrfs_start_transaction(fs_info->tree_root, 0); | |
5585 | if (IS_ERR(trans)) { | |
5586 | ret = PTR_ERR(trans); | |
5587 | goto error; | |
5588 | } | |
5589 | ||
5590 | wc.trans = trans; | |
5591 | wc.pin = 1; | |
5592 | ||
5593 | ret = walk_log_tree(trans, log_root_tree, &wc); | |
5594 | if (ret) { | |
5595 | btrfs_handle_fs_error(fs_info, ret, "Failed to pin buffers while " | |
5596 | "recovering log root tree."); | |
5597 | goto error; | |
5598 | } | |
5599 | ||
5600 | again: | |
5601 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
5602 | key.offset = (u64)-1; | |
5603 | key.type = BTRFS_ROOT_ITEM_KEY; | |
5604 | ||
5605 | while (1) { | |
5606 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); | |
5607 | ||
5608 | if (ret < 0) { | |
5609 | btrfs_handle_fs_error(fs_info, ret, | |
5610 | "Couldn't find tree log root."); | |
5611 | goto error; | |
5612 | } | |
5613 | if (ret > 0) { | |
5614 | if (path->slots[0] == 0) | |
5615 | break; | |
5616 | path->slots[0]--; | |
5617 | } | |
5618 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
5619 | path->slots[0]); | |
5620 | btrfs_release_path(path); | |
5621 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
5622 | break; | |
5623 | ||
5624 | log = btrfs_read_fs_root(log_root_tree, &found_key); | |
5625 | if (IS_ERR(log)) { | |
5626 | ret = PTR_ERR(log); | |
5627 | btrfs_handle_fs_error(fs_info, ret, | |
5628 | "Couldn't read tree log root."); | |
5629 | goto error; | |
5630 | } | |
5631 | ||
5632 | tmp_key.objectid = found_key.offset; | |
5633 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; | |
5634 | tmp_key.offset = (u64)-1; | |
5635 | ||
5636 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); | |
5637 | if (IS_ERR(wc.replay_dest)) { | |
5638 | ret = PTR_ERR(wc.replay_dest); | |
5639 | free_extent_buffer(log->node); | |
5640 | free_extent_buffer(log->commit_root); | |
5641 | kfree(log); | |
5642 | btrfs_handle_fs_error(fs_info, ret, "Couldn't read target root " | |
5643 | "for tree log recovery."); | |
5644 | goto error; | |
5645 | } | |
5646 | ||
5647 | wc.replay_dest->log_root = log; | |
5648 | btrfs_record_root_in_trans(trans, wc.replay_dest); | |
5649 | ret = walk_log_tree(trans, log, &wc); | |
5650 | ||
5651 | if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) { | |
5652 | ret = fixup_inode_link_counts(trans, wc.replay_dest, | |
5653 | path); | |
5654 | } | |
5655 | ||
5656 | key.offset = found_key.offset - 1; | |
5657 | wc.replay_dest->log_root = NULL; | |
5658 | free_extent_buffer(log->node); | |
5659 | free_extent_buffer(log->commit_root); | |
5660 | kfree(log); | |
5661 | ||
5662 | if (ret) | |
5663 | goto error; | |
5664 | ||
5665 | if (found_key.offset == 0) | |
5666 | break; | |
5667 | } | |
5668 | btrfs_release_path(path); | |
5669 | ||
5670 | /* step one is to pin it all, step two is to replay just inodes */ | |
5671 | if (wc.pin) { | |
5672 | wc.pin = 0; | |
5673 | wc.process_func = replay_one_buffer; | |
5674 | wc.stage = LOG_WALK_REPLAY_INODES; | |
5675 | goto again; | |
5676 | } | |
5677 | /* step three is to replay everything */ | |
5678 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
5679 | wc.stage++; | |
5680 | goto again; | |
5681 | } | |
5682 | ||
5683 | btrfs_free_path(path); | |
5684 | ||
5685 | /* step 4: commit the transaction, which also unpins the blocks */ | |
5686 | ret = btrfs_commit_transaction(trans, fs_info->tree_root); | |
5687 | if (ret) | |
5688 | return ret; | |
5689 | ||
5690 | free_extent_buffer(log_root_tree->node); | |
5691 | log_root_tree->log_root = NULL; | |
5692 | fs_info->log_root_recovering = 0; | |
5693 | kfree(log_root_tree); | |
5694 | ||
5695 | return 0; | |
5696 | error: | |
5697 | if (wc.trans) | |
5698 | btrfs_end_transaction(wc.trans, fs_info->tree_root); | |
5699 | btrfs_free_path(path); | |
5700 | return ret; | |
5701 | } | |
5702 | ||
5703 | /* | |
5704 | * there are some corner cases where we want to force a full | |
5705 | * commit instead of allowing a directory to be logged. | |
5706 | * | |
5707 | * They revolve around files there were unlinked from the directory, and | |
5708 | * this function updates the parent directory so that a full commit is | |
5709 | * properly done if it is fsync'd later after the unlinks are done. | |
5710 | * | |
5711 | * Must be called before the unlink operations (updates to the subvolume tree, | |
5712 | * inodes, etc) are done. | |
5713 | */ | |
5714 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
5715 | struct inode *dir, struct inode *inode, | |
5716 | int for_rename) | |
5717 | { | |
5718 | /* | |
5719 | * when we're logging a file, if it hasn't been renamed | |
5720 | * or unlinked, and its inode is fully committed on disk, | |
5721 | * we don't have to worry about walking up the directory chain | |
5722 | * to log its parents. | |
5723 | * | |
5724 | * So, we use the last_unlink_trans field to put this transid | |
5725 | * into the file. When the file is logged we check it and | |
5726 | * don't log the parents if the file is fully on disk. | |
5727 | */ | |
5728 | mutex_lock(&BTRFS_I(inode)->log_mutex); | |
5729 | BTRFS_I(inode)->last_unlink_trans = trans->transid; | |
5730 | mutex_unlock(&BTRFS_I(inode)->log_mutex); | |
5731 | ||
5732 | /* | |
5733 | * if this directory was already logged any new | |
5734 | * names for this file/dir will get recorded | |
5735 | */ | |
5736 | smp_mb(); | |
5737 | if (BTRFS_I(dir)->logged_trans == trans->transid) | |
5738 | return; | |
5739 | ||
5740 | /* | |
5741 | * if the inode we're about to unlink was logged, | |
5742 | * the log will be properly updated for any new names | |
5743 | */ | |
5744 | if (BTRFS_I(inode)->logged_trans == trans->transid) | |
5745 | return; | |
5746 | ||
5747 | /* | |
5748 | * when renaming files across directories, if the directory | |
5749 | * there we're unlinking from gets fsync'd later on, there's | |
5750 | * no way to find the destination directory later and fsync it | |
5751 | * properly. So, we have to be conservative and force commits | |
5752 | * so the new name gets discovered. | |
5753 | */ | |
5754 | if (for_rename) | |
5755 | goto record; | |
5756 | ||
5757 | /* we can safely do the unlink without any special recording */ | |
5758 | return; | |
5759 | ||
5760 | record: | |
5761 | mutex_lock(&BTRFS_I(dir)->log_mutex); | |
5762 | BTRFS_I(dir)->last_unlink_trans = trans->transid; | |
5763 | mutex_unlock(&BTRFS_I(dir)->log_mutex); | |
5764 | } | |
5765 | ||
5766 | /* | |
5767 | * Make sure that if someone attempts to fsync the parent directory of a deleted | |
5768 | * snapshot, it ends up triggering a transaction commit. This is to guarantee | |
5769 | * that after replaying the log tree of the parent directory's root we will not | |
5770 | * see the snapshot anymore and at log replay time we will not see any log tree | |
5771 | * corresponding to the deleted snapshot's root, which could lead to replaying | |
5772 | * it after replaying the log tree of the parent directory (which would replay | |
5773 | * the snapshot delete operation). | |
5774 | * | |
5775 | * Must be called before the actual snapshot destroy operation (updates to the | |
5776 | * parent root and tree of tree roots trees, etc) are done. | |
5777 | */ | |
5778 | void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans, | |
5779 | struct inode *dir) | |
5780 | { | |
5781 | mutex_lock(&BTRFS_I(dir)->log_mutex); | |
5782 | BTRFS_I(dir)->last_unlink_trans = trans->transid; | |
5783 | mutex_unlock(&BTRFS_I(dir)->log_mutex); | |
5784 | } | |
5785 | ||
5786 | /* | |
5787 | * Call this after adding a new name for a file and it will properly | |
5788 | * update the log to reflect the new name. | |
5789 | * | |
5790 | * It will return zero if all goes well, and it will return 1 if a | |
5791 | * full transaction commit is required. | |
5792 | */ | |
5793 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
5794 | struct inode *inode, struct inode *old_dir, | |
5795 | struct dentry *parent) | |
5796 | { | |
5797 | struct btrfs_root * root = BTRFS_I(inode)->root; | |
5798 | ||
5799 | /* | |
5800 | * this will force the logging code to walk the dentry chain | |
5801 | * up for the file | |
5802 | */ | |
5803 | if (S_ISREG(inode->i_mode)) | |
5804 | BTRFS_I(inode)->last_unlink_trans = trans->transid; | |
5805 | ||
5806 | /* | |
5807 | * if this inode hasn't been logged and directory we're renaming it | |
5808 | * from hasn't been logged, we don't need to log it | |
5809 | */ | |
5810 | if (BTRFS_I(inode)->logged_trans <= | |
5811 | root->fs_info->last_trans_committed && | |
5812 | (!old_dir || BTRFS_I(old_dir)->logged_trans <= | |
5813 | root->fs_info->last_trans_committed)) | |
5814 | return 0; | |
5815 | ||
5816 | return btrfs_log_inode_parent(trans, root, inode, parent, 0, | |
5817 | LLONG_MAX, 1, NULL); | |
5818 | } | |
5819 |