]>
Commit | Line | Data |
---|---|---|
e02119d5 CM |
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 "ctree.h" | |
21 | #include "transaction.h" | |
22 | #include "disk-io.h" | |
23 | #include "locking.h" | |
24 | #include "print-tree.h" | |
25 | #include "compat.h" | |
b2950863 | 26 | #include "tree-log.h" |
e02119d5 CM |
27 | |
28 | /* magic values for the inode_only field in btrfs_log_inode: | |
29 | * | |
30 | * LOG_INODE_ALL means to log everything | |
31 | * LOG_INODE_EXISTS means to log just enough to recreate the inode | |
32 | * during log replay | |
33 | */ | |
34 | #define LOG_INODE_ALL 0 | |
35 | #define LOG_INODE_EXISTS 1 | |
36 | ||
12fcfd22 CM |
37 | /* |
38 | * directory trouble cases | |
39 | * | |
40 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync | |
41 | * log, we must force a full commit before doing an fsync of the directory | |
42 | * where the unlink was done. | |
43 | * ---> record transid of last unlink/rename per directory | |
44 | * | |
45 | * mkdir foo/some_dir | |
46 | * normal commit | |
47 | * rename foo/some_dir foo2/some_dir | |
48 | * mkdir foo/some_dir | |
49 | * fsync foo/some_dir/some_file | |
50 | * | |
51 | * The fsync above will unlink the original some_dir without recording | |
52 | * it in its new location (foo2). After a crash, some_dir will be gone | |
53 | * unless the fsync of some_file forces a full commit | |
54 | * | |
55 | * 2) we must log any new names for any file or dir that is in the fsync | |
56 | * log. ---> check inode while renaming/linking. | |
57 | * | |
58 | * 2a) we must log any new names for any file or dir during rename | |
59 | * when the directory they are being removed from was logged. | |
60 | * ---> check inode and old parent dir during rename | |
61 | * | |
62 | * 2a is actually the more important variant. With the extra logging | |
63 | * a crash might unlink the old name without recreating the new one | |
64 | * | |
65 | * 3) after a crash, we must go through any directories with a link count | |
66 | * of zero and redo the rm -rf | |
67 | * | |
68 | * mkdir f1/foo | |
69 | * normal commit | |
70 | * rm -rf f1/foo | |
71 | * fsync(f1) | |
72 | * | |
73 | * The directory f1 was fully removed from the FS, but fsync was never | |
74 | * called on f1, only its parent dir. After a crash the rm -rf must | |
75 | * be replayed. This must be able to recurse down the entire | |
76 | * directory tree. The inode link count fixup code takes care of the | |
77 | * ugly details. | |
78 | */ | |
79 | ||
e02119d5 CM |
80 | /* |
81 | * stages for the tree walking. The first | |
82 | * stage (0) is to only pin down the blocks we find | |
83 | * the second stage (1) is to make sure that all the inodes | |
84 | * we find in the log are created in the subvolume. | |
85 | * | |
86 | * The last stage is to deal with directories and links and extents | |
87 | * and all the other fun semantics | |
88 | */ | |
89 | #define LOG_WALK_PIN_ONLY 0 | |
90 | #define LOG_WALK_REPLAY_INODES 1 | |
91 | #define LOG_WALK_REPLAY_ALL 2 | |
92 | ||
12fcfd22 | 93 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
e02119d5 CM |
94 | struct btrfs_root *root, struct inode *inode, |
95 | int inode_only); | |
ec051c0f YZ |
96 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
97 | struct btrfs_root *root, | |
98 | struct btrfs_path *path, u64 objectid); | |
12fcfd22 CM |
99 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
100 | struct btrfs_root *root, | |
101 | struct btrfs_root *log, | |
102 | struct btrfs_path *path, | |
103 | u64 dirid, int del_all); | |
e02119d5 CM |
104 | |
105 | /* | |
106 | * tree logging is a special write ahead log used to make sure that | |
107 | * fsyncs and O_SYNCs can happen without doing full tree commits. | |
108 | * | |
109 | * Full tree commits are expensive because they require commonly | |
110 | * modified blocks to be recowed, creating many dirty pages in the | |
111 | * extent tree an 4x-6x higher write load than ext3. | |
112 | * | |
113 | * Instead of doing a tree commit on every fsync, we use the | |
114 | * key ranges and transaction ids to find items for a given file or directory | |
115 | * that have changed in this transaction. Those items are copied into | |
116 | * a special tree (one per subvolume root), that tree is written to disk | |
117 | * and then the fsync is considered complete. | |
118 | * | |
119 | * After a crash, items are copied out of the log-tree back into the | |
120 | * subvolume tree. Any file data extents found are recorded in the extent | |
121 | * allocation tree, and the log-tree freed. | |
122 | * | |
123 | * The log tree is read three times, once to pin down all the extents it is | |
124 | * using in ram and once, once to create all the inodes logged in the tree | |
125 | * and once to do all the other items. | |
126 | */ | |
127 | ||
e02119d5 CM |
128 | /* |
129 | * start a sub transaction and setup the log tree | |
130 | * this increments the log tree writer count to make the people | |
131 | * syncing the tree wait for us to finish | |
132 | */ | |
133 | static int start_log_trans(struct btrfs_trans_handle *trans, | |
134 | struct btrfs_root *root) | |
135 | { | |
136 | int ret; | |
7237f183 YZ |
137 | |
138 | mutex_lock(&root->log_mutex); | |
139 | if (root->log_root) { | |
ff782e0a JB |
140 | if (!root->log_start_pid) { |
141 | root->log_start_pid = current->pid; | |
142 | root->log_multiple_pids = false; | |
143 | } else if (root->log_start_pid != current->pid) { | |
144 | root->log_multiple_pids = true; | |
145 | } | |
146 | ||
7237f183 YZ |
147 | root->log_batch++; |
148 | atomic_inc(&root->log_writers); | |
149 | mutex_unlock(&root->log_mutex); | |
150 | return 0; | |
151 | } | |
ff782e0a JB |
152 | root->log_multiple_pids = false; |
153 | root->log_start_pid = current->pid; | |
e02119d5 CM |
154 | mutex_lock(&root->fs_info->tree_log_mutex); |
155 | if (!root->fs_info->log_root_tree) { | |
156 | ret = btrfs_init_log_root_tree(trans, root->fs_info); | |
157 | BUG_ON(ret); | |
158 | } | |
159 | if (!root->log_root) { | |
160 | ret = btrfs_add_log_tree(trans, root); | |
161 | BUG_ON(ret); | |
162 | } | |
e02119d5 | 163 | mutex_unlock(&root->fs_info->tree_log_mutex); |
7237f183 YZ |
164 | root->log_batch++; |
165 | atomic_inc(&root->log_writers); | |
166 | mutex_unlock(&root->log_mutex); | |
e02119d5 CM |
167 | return 0; |
168 | } | |
169 | ||
170 | /* | |
171 | * returns 0 if there was a log transaction running and we were able | |
172 | * to join, or returns -ENOENT if there were not transactions | |
173 | * in progress | |
174 | */ | |
175 | static int join_running_log_trans(struct btrfs_root *root) | |
176 | { | |
177 | int ret = -ENOENT; | |
178 | ||
179 | smp_mb(); | |
180 | if (!root->log_root) | |
181 | return -ENOENT; | |
182 | ||
7237f183 | 183 | mutex_lock(&root->log_mutex); |
e02119d5 CM |
184 | if (root->log_root) { |
185 | ret = 0; | |
7237f183 | 186 | atomic_inc(&root->log_writers); |
e02119d5 | 187 | } |
7237f183 | 188 | mutex_unlock(&root->log_mutex); |
e02119d5 CM |
189 | return ret; |
190 | } | |
191 | ||
12fcfd22 CM |
192 | /* |
193 | * This either makes the current running log transaction wait | |
194 | * until you call btrfs_end_log_trans() or it makes any future | |
195 | * log transactions wait until you call btrfs_end_log_trans() | |
196 | */ | |
197 | int btrfs_pin_log_trans(struct btrfs_root *root) | |
198 | { | |
199 | int ret = -ENOENT; | |
200 | ||
201 | mutex_lock(&root->log_mutex); | |
202 | atomic_inc(&root->log_writers); | |
203 | mutex_unlock(&root->log_mutex); | |
204 | return ret; | |
205 | } | |
206 | ||
e02119d5 CM |
207 | /* |
208 | * indicate we're done making changes to the log tree | |
209 | * and wake up anyone waiting to do a sync | |
210 | */ | |
12fcfd22 | 211 | int btrfs_end_log_trans(struct btrfs_root *root) |
e02119d5 | 212 | { |
7237f183 YZ |
213 | if (atomic_dec_and_test(&root->log_writers)) { |
214 | smp_mb(); | |
215 | if (waitqueue_active(&root->log_writer_wait)) | |
216 | wake_up(&root->log_writer_wait); | |
217 | } | |
e02119d5 CM |
218 | return 0; |
219 | } | |
220 | ||
221 | ||
222 | /* | |
223 | * the walk control struct is used to pass state down the chain when | |
224 | * processing the log tree. The stage field tells us which part | |
225 | * of the log tree processing we are currently doing. The others | |
226 | * are state fields used for that specific part | |
227 | */ | |
228 | struct walk_control { | |
229 | /* should we free the extent on disk when done? This is used | |
230 | * at transaction commit time while freeing a log tree | |
231 | */ | |
232 | int free; | |
233 | ||
234 | /* should we write out the extent buffer? This is used | |
235 | * while flushing the log tree to disk during a sync | |
236 | */ | |
237 | int write; | |
238 | ||
239 | /* should we wait for the extent buffer io to finish? Also used | |
240 | * while flushing the log tree to disk for a sync | |
241 | */ | |
242 | int wait; | |
243 | ||
244 | /* pin only walk, we record which extents on disk belong to the | |
245 | * log trees | |
246 | */ | |
247 | int pin; | |
248 | ||
249 | /* what stage of the replay code we're currently in */ | |
250 | int stage; | |
251 | ||
252 | /* the root we are currently replaying */ | |
253 | struct btrfs_root *replay_dest; | |
254 | ||
255 | /* the trans handle for the current replay */ | |
256 | struct btrfs_trans_handle *trans; | |
257 | ||
258 | /* the function that gets used to process blocks we find in the | |
259 | * tree. Note the extent_buffer might not be up to date when it is | |
260 | * passed in, and it must be checked or read if you need the data | |
261 | * inside it | |
262 | */ | |
263 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, | |
264 | struct walk_control *wc, u64 gen); | |
265 | }; | |
266 | ||
267 | /* | |
268 | * process_func used to pin down extents, write them or wait on them | |
269 | */ | |
270 | static int process_one_buffer(struct btrfs_root *log, | |
271 | struct extent_buffer *eb, | |
272 | struct walk_control *wc, u64 gen) | |
273 | { | |
04018de5 | 274 | if (wc->pin) |
11833d66 YZ |
275 | btrfs_pin_extent(log->fs_info->extent_root, |
276 | eb->start, eb->len, 0); | |
e02119d5 CM |
277 | |
278 | if (btrfs_buffer_uptodate(eb, gen)) { | |
279 | if (wc->write) | |
280 | btrfs_write_tree_block(eb); | |
281 | if (wc->wait) | |
282 | btrfs_wait_tree_block_writeback(eb); | |
283 | } | |
284 | return 0; | |
285 | } | |
286 | ||
287 | /* | |
288 | * Item overwrite used by replay and tree logging. eb, slot and key all refer | |
289 | * to the src data we are copying out. | |
290 | * | |
291 | * root is the tree we are copying into, and path is a scratch | |
292 | * path for use in this function (it should be released on entry and | |
293 | * will be released on exit). | |
294 | * | |
295 | * If the key is already in the destination tree the existing item is | |
296 | * overwritten. If the existing item isn't big enough, it is extended. | |
297 | * If it is too large, it is truncated. | |
298 | * | |
299 | * If the key isn't in the destination yet, a new item is inserted. | |
300 | */ | |
301 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, | |
302 | struct btrfs_root *root, | |
303 | struct btrfs_path *path, | |
304 | struct extent_buffer *eb, int slot, | |
305 | struct btrfs_key *key) | |
306 | { | |
307 | int ret; | |
308 | u32 item_size; | |
309 | u64 saved_i_size = 0; | |
310 | int save_old_i_size = 0; | |
311 | unsigned long src_ptr; | |
312 | unsigned long dst_ptr; | |
313 | int overwrite_root = 0; | |
314 | ||
315 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
316 | overwrite_root = 1; | |
317 | ||
318 | item_size = btrfs_item_size_nr(eb, slot); | |
319 | src_ptr = btrfs_item_ptr_offset(eb, slot); | |
320 | ||
321 | /* look for the key in the destination tree */ | |
322 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
323 | if (ret == 0) { | |
324 | char *src_copy; | |
325 | char *dst_copy; | |
326 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], | |
327 | path->slots[0]); | |
328 | if (dst_size != item_size) | |
329 | goto insert; | |
330 | ||
331 | if (item_size == 0) { | |
332 | btrfs_release_path(root, path); | |
333 | return 0; | |
334 | } | |
335 | dst_copy = kmalloc(item_size, GFP_NOFS); | |
336 | src_copy = kmalloc(item_size, GFP_NOFS); | |
337 | ||
338 | read_extent_buffer(eb, src_copy, src_ptr, item_size); | |
339 | ||
340 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
341 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, | |
342 | item_size); | |
343 | ret = memcmp(dst_copy, src_copy, item_size); | |
344 | ||
345 | kfree(dst_copy); | |
346 | kfree(src_copy); | |
347 | /* | |
348 | * they have the same contents, just return, this saves | |
349 | * us from cowing blocks in the destination tree and doing | |
350 | * extra writes that may not have been done by a previous | |
351 | * sync | |
352 | */ | |
353 | if (ret == 0) { | |
354 | btrfs_release_path(root, path); | |
355 | return 0; | |
356 | } | |
357 | ||
358 | } | |
359 | insert: | |
360 | btrfs_release_path(root, path); | |
361 | /* try to insert the key into the destination tree */ | |
362 | ret = btrfs_insert_empty_item(trans, root, path, | |
363 | key, item_size); | |
364 | ||
365 | /* make sure any existing item is the correct size */ | |
366 | if (ret == -EEXIST) { | |
367 | u32 found_size; | |
368 | found_size = btrfs_item_size_nr(path->nodes[0], | |
369 | path->slots[0]); | |
370 | if (found_size > item_size) { | |
371 | btrfs_truncate_item(trans, root, path, item_size, 1); | |
372 | } else if (found_size < item_size) { | |
87b29b20 YZ |
373 | ret = btrfs_extend_item(trans, root, path, |
374 | item_size - found_size); | |
e02119d5 CM |
375 | BUG_ON(ret); |
376 | } | |
377 | } else if (ret) { | |
378 | BUG(); | |
379 | } | |
380 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], | |
381 | path->slots[0]); | |
382 | ||
383 | /* don't overwrite an existing inode if the generation number | |
384 | * was logged as zero. This is done when the tree logging code | |
385 | * is just logging an inode to make sure it exists after recovery. | |
386 | * | |
387 | * Also, don't overwrite i_size on directories during replay. | |
388 | * log replay inserts and removes directory items based on the | |
389 | * state of the tree found in the subvolume, and i_size is modified | |
390 | * as it goes | |
391 | */ | |
392 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { | |
393 | struct btrfs_inode_item *src_item; | |
394 | struct btrfs_inode_item *dst_item; | |
395 | ||
396 | src_item = (struct btrfs_inode_item *)src_ptr; | |
397 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
398 | ||
399 | if (btrfs_inode_generation(eb, src_item) == 0) | |
400 | goto no_copy; | |
401 | ||
402 | if (overwrite_root && | |
403 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && | |
404 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { | |
405 | save_old_i_size = 1; | |
406 | saved_i_size = btrfs_inode_size(path->nodes[0], | |
407 | dst_item); | |
408 | } | |
409 | } | |
410 | ||
411 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, | |
412 | src_ptr, item_size); | |
413 | ||
414 | if (save_old_i_size) { | |
415 | struct btrfs_inode_item *dst_item; | |
416 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
417 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); | |
418 | } | |
419 | ||
420 | /* make sure the generation is filled in */ | |
421 | if (key->type == BTRFS_INODE_ITEM_KEY) { | |
422 | struct btrfs_inode_item *dst_item; | |
423 | dst_item = (struct btrfs_inode_item *)dst_ptr; | |
424 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { | |
425 | btrfs_set_inode_generation(path->nodes[0], dst_item, | |
426 | trans->transid); | |
427 | } | |
428 | } | |
429 | no_copy: | |
430 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
431 | btrfs_release_path(root, path); | |
432 | return 0; | |
433 | } | |
434 | ||
435 | /* | |
436 | * simple helper to read an inode off the disk from a given root | |
437 | * This can only be called for subvolume roots and not for the log | |
438 | */ | |
439 | static noinline struct inode *read_one_inode(struct btrfs_root *root, | |
440 | u64 objectid) | |
441 | { | |
5d4f98a2 | 442 | struct btrfs_key key; |
e02119d5 | 443 | struct inode *inode; |
e02119d5 | 444 | |
5d4f98a2 YZ |
445 | key.objectid = objectid; |
446 | key.type = BTRFS_INODE_ITEM_KEY; | |
447 | key.offset = 0; | |
448 | inode = btrfs_iget(root->fs_info->sb, &key, root); | |
449 | if (IS_ERR(inode)) { | |
450 | inode = NULL; | |
451 | } else if (is_bad_inode(inode)) { | |
e02119d5 CM |
452 | iput(inode); |
453 | inode = NULL; | |
454 | } | |
455 | return inode; | |
456 | } | |
457 | ||
458 | /* replays a single extent in 'eb' at 'slot' with 'key' into the | |
459 | * subvolume 'root'. path is released on entry and should be released | |
460 | * on exit. | |
461 | * | |
462 | * extents in the log tree have not been allocated out of the extent | |
463 | * tree yet. So, this completes the allocation, taking a reference | |
464 | * as required if the extent already exists or creating a new extent | |
465 | * if it isn't in the extent allocation tree yet. | |
466 | * | |
467 | * The extent is inserted into the file, dropping any existing extents | |
468 | * from the file that overlap the new one. | |
469 | */ | |
470 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, | |
471 | struct btrfs_root *root, | |
472 | struct btrfs_path *path, | |
473 | struct extent_buffer *eb, int slot, | |
474 | struct btrfs_key *key) | |
475 | { | |
476 | int found_type; | |
477 | u64 mask = root->sectorsize - 1; | |
478 | u64 extent_end; | |
479 | u64 alloc_hint; | |
480 | u64 start = key->offset; | |
07d400a6 | 481 | u64 saved_nbytes; |
e02119d5 CM |
482 | struct btrfs_file_extent_item *item; |
483 | struct inode *inode = NULL; | |
484 | unsigned long size; | |
485 | int ret = 0; | |
486 | ||
487 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
488 | found_type = btrfs_file_extent_type(eb, item); | |
489 | ||
d899e052 YZ |
490 | if (found_type == BTRFS_FILE_EXTENT_REG || |
491 | found_type == BTRFS_FILE_EXTENT_PREALLOC) | |
e02119d5 CM |
492 | extent_end = start + btrfs_file_extent_num_bytes(eb, item); |
493 | else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
c8b97818 | 494 | size = btrfs_file_extent_inline_len(eb, item); |
e02119d5 CM |
495 | extent_end = (start + size + mask) & ~mask; |
496 | } else { | |
497 | ret = 0; | |
498 | goto out; | |
499 | } | |
500 | ||
501 | inode = read_one_inode(root, key->objectid); | |
502 | if (!inode) { | |
503 | ret = -EIO; | |
504 | goto out; | |
505 | } | |
506 | ||
507 | /* | |
508 | * first check to see if we already have this extent in the | |
509 | * file. This must be done before the btrfs_drop_extents run | |
510 | * so we don't try to drop this extent. | |
511 | */ | |
512 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
513 | start, 0); | |
514 | ||
d899e052 YZ |
515 | if (ret == 0 && |
516 | (found_type == BTRFS_FILE_EXTENT_REG || | |
517 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { | |
e02119d5 CM |
518 | struct btrfs_file_extent_item cmp1; |
519 | struct btrfs_file_extent_item cmp2; | |
520 | struct btrfs_file_extent_item *existing; | |
521 | struct extent_buffer *leaf; | |
522 | ||
523 | leaf = path->nodes[0]; | |
524 | existing = btrfs_item_ptr(leaf, path->slots[0], | |
525 | struct btrfs_file_extent_item); | |
526 | ||
527 | read_extent_buffer(eb, &cmp1, (unsigned long)item, | |
528 | sizeof(cmp1)); | |
529 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, | |
530 | sizeof(cmp2)); | |
531 | ||
532 | /* | |
533 | * we already have a pointer to this exact extent, | |
534 | * we don't have to do anything | |
535 | */ | |
536 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { | |
537 | btrfs_release_path(root, path); | |
538 | goto out; | |
539 | } | |
540 | } | |
541 | btrfs_release_path(root, path); | |
542 | ||
07d400a6 | 543 | saved_nbytes = inode_get_bytes(inode); |
e02119d5 CM |
544 | /* drop any overlapping extents */ |
545 | ret = btrfs_drop_extents(trans, root, inode, | |
a1ed835e | 546 | start, extent_end, extent_end, start, &alloc_hint, 1); |
e02119d5 CM |
547 | BUG_ON(ret); |
548 | ||
07d400a6 YZ |
549 | if (found_type == BTRFS_FILE_EXTENT_REG || |
550 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 | 551 | u64 offset; |
07d400a6 YZ |
552 | unsigned long dest_offset; |
553 | struct btrfs_key ins; | |
554 | ||
555 | ret = btrfs_insert_empty_item(trans, root, path, key, | |
556 | sizeof(*item)); | |
557 | BUG_ON(ret); | |
558 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], | |
559 | path->slots[0]); | |
560 | copy_extent_buffer(path->nodes[0], eb, dest_offset, | |
561 | (unsigned long)item, sizeof(*item)); | |
562 | ||
563 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
564 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
565 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 | 566 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
07d400a6 YZ |
567 | |
568 | if (ins.objectid > 0) { | |
569 | u64 csum_start; | |
570 | u64 csum_end; | |
571 | LIST_HEAD(ordered_sums); | |
572 | /* | |
573 | * is this extent already allocated in the extent | |
574 | * allocation tree? If so, just add a reference | |
575 | */ | |
576 | ret = btrfs_lookup_extent(root, ins.objectid, | |
577 | ins.offset); | |
578 | if (ret == 0) { | |
579 | ret = btrfs_inc_extent_ref(trans, root, | |
580 | ins.objectid, ins.offset, | |
5d4f98a2 YZ |
581 | 0, root->root_key.objectid, |
582 | key->objectid, offset); | |
07d400a6 YZ |
583 | } else { |
584 | /* | |
585 | * insert the extent pointer in the extent | |
586 | * allocation tree | |
587 | */ | |
5d4f98a2 YZ |
588 | ret = btrfs_alloc_logged_file_extent(trans, |
589 | root, root->root_key.objectid, | |
590 | key->objectid, offset, &ins); | |
07d400a6 YZ |
591 | BUG_ON(ret); |
592 | } | |
593 | btrfs_release_path(root, path); | |
594 | ||
595 | if (btrfs_file_extent_compression(eb, item)) { | |
596 | csum_start = ins.objectid; | |
597 | csum_end = csum_start + ins.offset; | |
598 | } else { | |
599 | csum_start = ins.objectid + | |
600 | btrfs_file_extent_offset(eb, item); | |
601 | csum_end = csum_start + | |
602 | btrfs_file_extent_num_bytes(eb, item); | |
603 | } | |
604 | ||
605 | ret = btrfs_lookup_csums_range(root->log_root, | |
606 | csum_start, csum_end - 1, | |
607 | &ordered_sums); | |
608 | BUG_ON(ret); | |
609 | while (!list_empty(&ordered_sums)) { | |
610 | struct btrfs_ordered_sum *sums; | |
611 | sums = list_entry(ordered_sums.next, | |
612 | struct btrfs_ordered_sum, | |
613 | list); | |
614 | ret = btrfs_csum_file_blocks(trans, | |
615 | root->fs_info->csum_root, | |
616 | sums); | |
617 | BUG_ON(ret); | |
618 | list_del(&sums->list); | |
619 | kfree(sums); | |
620 | } | |
621 | } else { | |
622 | btrfs_release_path(root, path); | |
623 | } | |
624 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
625 | /* inline extents are easy, we just overwrite them */ | |
626 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
627 | BUG_ON(ret); | |
628 | } | |
e02119d5 | 629 | |
07d400a6 | 630 | inode_set_bytes(inode, saved_nbytes); |
e02119d5 CM |
631 | btrfs_update_inode(trans, root, inode); |
632 | out: | |
633 | if (inode) | |
634 | iput(inode); | |
635 | return ret; | |
636 | } | |
637 | ||
638 | /* | |
639 | * when cleaning up conflicts between the directory names in the | |
640 | * subvolume, directory names in the log and directory names in the | |
641 | * inode back references, we may have to unlink inodes from directories. | |
642 | * | |
643 | * This is a helper function to do the unlink of a specific directory | |
644 | * item | |
645 | */ | |
646 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, | |
647 | struct btrfs_root *root, | |
648 | struct btrfs_path *path, | |
649 | struct inode *dir, | |
650 | struct btrfs_dir_item *di) | |
651 | { | |
652 | struct inode *inode; | |
653 | char *name; | |
654 | int name_len; | |
655 | struct extent_buffer *leaf; | |
656 | struct btrfs_key location; | |
657 | int ret; | |
658 | ||
659 | leaf = path->nodes[0]; | |
660 | ||
661 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
662 | name_len = btrfs_dir_name_len(leaf, di); | |
663 | name = kmalloc(name_len, GFP_NOFS); | |
664 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); | |
665 | btrfs_release_path(root, path); | |
666 | ||
667 | inode = read_one_inode(root, location.objectid); | |
668 | BUG_ON(!inode); | |
669 | ||
ec051c0f YZ |
670 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
671 | BUG_ON(ret); | |
12fcfd22 | 672 | |
e02119d5 | 673 | ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len); |
ec051c0f | 674 | BUG_ON(ret); |
e02119d5 CM |
675 | kfree(name); |
676 | ||
677 | iput(inode); | |
678 | return ret; | |
679 | } | |
680 | ||
681 | /* | |
682 | * helper function to see if a given name and sequence number found | |
683 | * in an inode back reference are already in a directory and correctly | |
684 | * point to this inode | |
685 | */ | |
686 | static noinline int inode_in_dir(struct btrfs_root *root, | |
687 | struct btrfs_path *path, | |
688 | u64 dirid, u64 objectid, u64 index, | |
689 | const char *name, int name_len) | |
690 | { | |
691 | struct btrfs_dir_item *di; | |
692 | struct btrfs_key location; | |
693 | int match = 0; | |
694 | ||
695 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, | |
696 | index, name, name_len, 0); | |
697 | if (di && !IS_ERR(di)) { | |
698 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
699 | if (location.objectid != objectid) | |
700 | goto out; | |
701 | } else | |
702 | goto out; | |
703 | btrfs_release_path(root, path); | |
704 | ||
705 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); | |
706 | if (di && !IS_ERR(di)) { | |
707 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); | |
708 | if (location.objectid != objectid) | |
709 | goto out; | |
710 | } else | |
711 | goto out; | |
712 | match = 1; | |
713 | out: | |
714 | btrfs_release_path(root, path); | |
715 | return match; | |
716 | } | |
717 | ||
718 | /* | |
719 | * helper function to check a log tree for a named back reference in | |
720 | * an inode. This is used to decide if a back reference that is | |
721 | * found in the subvolume conflicts with what we find in the log. | |
722 | * | |
723 | * inode backreferences may have multiple refs in a single item, | |
724 | * during replay we process one reference at a time, and we don't | |
725 | * want to delete valid links to a file from the subvolume if that | |
726 | * link is also in the log. | |
727 | */ | |
728 | static noinline int backref_in_log(struct btrfs_root *log, | |
729 | struct btrfs_key *key, | |
730 | char *name, int namelen) | |
731 | { | |
732 | struct btrfs_path *path; | |
733 | struct btrfs_inode_ref *ref; | |
734 | unsigned long ptr; | |
735 | unsigned long ptr_end; | |
736 | unsigned long name_ptr; | |
737 | int found_name_len; | |
738 | int item_size; | |
739 | int ret; | |
740 | int match = 0; | |
741 | ||
742 | path = btrfs_alloc_path(); | |
743 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); | |
744 | if (ret != 0) | |
745 | goto out; | |
746 | ||
747 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); | |
748 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
749 | ptr_end = ptr + item_size; | |
750 | while (ptr < ptr_end) { | |
751 | ref = (struct btrfs_inode_ref *)ptr; | |
752 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); | |
753 | if (found_name_len == namelen) { | |
754 | name_ptr = (unsigned long)(ref + 1); | |
755 | ret = memcmp_extent_buffer(path->nodes[0], name, | |
756 | name_ptr, namelen); | |
757 | if (ret == 0) { | |
758 | match = 1; | |
759 | goto out; | |
760 | } | |
761 | } | |
762 | ptr = (unsigned long)(ref + 1) + found_name_len; | |
763 | } | |
764 | out: | |
765 | btrfs_free_path(path); | |
766 | return match; | |
767 | } | |
768 | ||
769 | ||
770 | /* | |
771 | * replay one inode back reference item found in the log tree. | |
772 | * eb, slot and key refer to the buffer and key found in the log tree. | |
773 | * root is the destination we are replaying into, and path is for temp | |
774 | * use by this function. (it should be released on return). | |
775 | */ | |
776 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, | |
777 | struct btrfs_root *root, | |
778 | struct btrfs_root *log, | |
779 | struct btrfs_path *path, | |
780 | struct extent_buffer *eb, int slot, | |
781 | struct btrfs_key *key) | |
782 | { | |
783 | struct inode *dir; | |
784 | int ret; | |
785 | struct btrfs_key location; | |
786 | struct btrfs_inode_ref *ref; | |
787 | struct btrfs_dir_item *di; | |
788 | struct inode *inode; | |
789 | char *name; | |
790 | int namelen; | |
791 | unsigned long ref_ptr; | |
792 | unsigned long ref_end; | |
793 | ||
794 | location.objectid = key->objectid; | |
795 | location.type = BTRFS_INODE_ITEM_KEY; | |
796 | location.offset = 0; | |
797 | ||
798 | /* | |
799 | * it is possible that we didn't log all the parent directories | |
800 | * for a given inode. If we don't find the dir, just don't | |
801 | * copy the back ref in. The link count fixup code will take | |
802 | * care of the rest | |
803 | */ | |
804 | dir = read_one_inode(root, key->offset); | |
805 | if (!dir) | |
806 | return -ENOENT; | |
807 | ||
808 | inode = read_one_inode(root, key->objectid); | |
631c07c8 | 809 | BUG_ON(!inode); |
e02119d5 CM |
810 | |
811 | ref_ptr = btrfs_item_ptr_offset(eb, slot); | |
812 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); | |
813 | ||
814 | again: | |
815 | ref = (struct btrfs_inode_ref *)ref_ptr; | |
816 | ||
817 | namelen = btrfs_inode_ref_name_len(eb, ref); | |
818 | name = kmalloc(namelen, GFP_NOFS); | |
819 | BUG_ON(!name); | |
820 | ||
821 | read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen); | |
822 | ||
823 | /* if we already have a perfect match, we're done */ | |
824 | if (inode_in_dir(root, path, dir->i_ino, inode->i_ino, | |
825 | btrfs_inode_ref_index(eb, ref), | |
826 | name, namelen)) { | |
827 | goto out; | |
828 | } | |
829 | ||
830 | /* | |
831 | * look for a conflicting back reference in the metadata. | |
832 | * if we find one we have to unlink that name of the file | |
833 | * before we add our new link. Later on, we overwrite any | |
834 | * existing back reference, and we don't want to create | |
835 | * dangling pointers in the directory. | |
836 | */ | |
837 | conflict_again: | |
838 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
839 | if (ret == 0) { | |
840 | char *victim_name; | |
841 | int victim_name_len; | |
842 | struct btrfs_inode_ref *victim_ref; | |
843 | unsigned long ptr; | |
844 | unsigned long ptr_end; | |
845 | struct extent_buffer *leaf = path->nodes[0]; | |
846 | ||
847 | /* are we trying to overwrite a back ref for the root directory | |
848 | * if so, just jump out, we're done | |
849 | */ | |
850 | if (key->objectid == key->offset) | |
851 | goto out_nowrite; | |
852 | ||
853 | /* check all the names in this back reference to see | |
854 | * if they are in the log. if so, we allow them to stay | |
855 | * otherwise they must be unlinked as a conflict | |
856 | */ | |
857 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
858 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); | |
d397712b | 859 | while (ptr < ptr_end) { |
e02119d5 CM |
860 | victim_ref = (struct btrfs_inode_ref *)ptr; |
861 | victim_name_len = btrfs_inode_ref_name_len(leaf, | |
862 | victim_ref); | |
863 | victim_name = kmalloc(victim_name_len, GFP_NOFS); | |
864 | BUG_ON(!victim_name); | |
865 | ||
866 | read_extent_buffer(leaf, victim_name, | |
867 | (unsigned long)(victim_ref + 1), | |
868 | victim_name_len); | |
869 | ||
870 | if (!backref_in_log(log, key, victim_name, | |
871 | victim_name_len)) { | |
872 | btrfs_inc_nlink(inode); | |
873 | btrfs_release_path(root, path); | |
12fcfd22 | 874 | |
e02119d5 CM |
875 | ret = btrfs_unlink_inode(trans, root, dir, |
876 | inode, victim_name, | |
877 | victim_name_len); | |
878 | kfree(victim_name); | |
879 | btrfs_release_path(root, path); | |
880 | goto conflict_again; | |
881 | } | |
882 | kfree(victim_name); | |
883 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; | |
884 | } | |
885 | BUG_ON(ret); | |
886 | } | |
887 | btrfs_release_path(root, path); | |
888 | ||
889 | /* look for a conflicting sequence number */ | |
890 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
891 | btrfs_inode_ref_index(eb, ref), | |
892 | name, namelen, 0); | |
893 | if (di && !IS_ERR(di)) { | |
894 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
895 | BUG_ON(ret); | |
896 | } | |
897 | btrfs_release_path(root, path); | |
898 | ||
899 | ||
900 | /* look for a conflicting name */ | |
901 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
902 | name, namelen, 0); | |
903 | if (di && !IS_ERR(di)) { | |
904 | ret = drop_one_dir_item(trans, root, path, dir, di); | |
905 | BUG_ON(ret); | |
906 | } | |
907 | btrfs_release_path(root, path); | |
908 | ||
909 | /* insert our name */ | |
910 | ret = btrfs_add_link(trans, dir, inode, name, namelen, 0, | |
911 | btrfs_inode_ref_index(eb, ref)); | |
912 | BUG_ON(ret); | |
913 | ||
914 | btrfs_update_inode(trans, root, inode); | |
915 | ||
916 | out: | |
917 | ref_ptr = (unsigned long)(ref + 1) + namelen; | |
918 | kfree(name); | |
919 | if (ref_ptr < ref_end) | |
920 | goto again; | |
921 | ||
922 | /* finally write the back reference in the inode */ | |
923 | ret = overwrite_item(trans, root, path, eb, slot, key); | |
924 | BUG_ON(ret); | |
925 | ||
926 | out_nowrite: | |
927 | btrfs_release_path(root, path); | |
928 | iput(dir); | |
929 | iput(inode); | |
930 | return 0; | |
931 | } | |
932 | ||
e02119d5 CM |
933 | /* |
934 | * There are a few corners where the link count of the file can't | |
935 | * be properly maintained during replay. So, instead of adding | |
936 | * lots of complexity to the log code, we just scan the backrefs | |
937 | * for any file that has been through replay. | |
938 | * | |
939 | * The scan will update the link count on the inode to reflect the | |
940 | * number of back refs found. If it goes down to zero, the iput | |
941 | * will free the inode. | |
942 | */ | |
943 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, | |
944 | struct btrfs_root *root, | |
945 | struct inode *inode) | |
946 | { | |
947 | struct btrfs_path *path; | |
948 | int ret; | |
949 | struct btrfs_key key; | |
950 | u64 nlink = 0; | |
951 | unsigned long ptr; | |
952 | unsigned long ptr_end; | |
953 | int name_len; | |
954 | ||
955 | key.objectid = inode->i_ino; | |
956 | key.type = BTRFS_INODE_REF_KEY; | |
957 | key.offset = (u64)-1; | |
958 | ||
959 | path = btrfs_alloc_path(); | |
960 | ||
d397712b | 961 | while (1) { |
e02119d5 CM |
962 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
963 | if (ret < 0) | |
964 | break; | |
965 | if (ret > 0) { | |
966 | if (path->slots[0] == 0) | |
967 | break; | |
968 | path->slots[0]--; | |
969 | } | |
970 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
971 | path->slots[0]); | |
972 | if (key.objectid != inode->i_ino || | |
973 | key.type != BTRFS_INODE_REF_KEY) | |
974 | break; | |
975 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); | |
976 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], | |
977 | path->slots[0]); | |
d397712b | 978 | while (ptr < ptr_end) { |
e02119d5 CM |
979 | struct btrfs_inode_ref *ref; |
980 | ||
981 | ref = (struct btrfs_inode_ref *)ptr; | |
982 | name_len = btrfs_inode_ref_name_len(path->nodes[0], | |
983 | ref); | |
984 | ptr = (unsigned long)(ref + 1) + name_len; | |
985 | nlink++; | |
986 | } | |
987 | ||
988 | if (key.offset == 0) | |
989 | break; | |
990 | key.offset--; | |
991 | btrfs_release_path(root, path); | |
992 | } | |
12fcfd22 | 993 | btrfs_release_path(root, path); |
e02119d5 CM |
994 | if (nlink != inode->i_nlink) { |
995 | inode->i_nlink = nlink; | |
996 | btrfs_update_inode(trans, root, inode); | |
997 | } | |
8d5bf1cb | 998 | BTRFS_I(inode)->index_cnt = (u64)-1; |
e02119d5 | 999 | |
12fcfd22 CM |
1000 | if (inode->i_nlink == 0 && S_ISDIR(inode->i_mode)) { |
1001 | ret = replay_dir_deletes(trans, root, NULL, path, | |
1002 | inode->i_ino, 1); | |
1003 | BUG_ON(ret); | |
1004 | } | |
1005 | btrfs_free_path(path); | |
1006 | ||
e02119d5 CM |
1007 | return 0; |
1008 | } | |
1009 | ||
1010 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, | |
1011 | struct btrfs_root *root, | |
1012 | struct btrfs_path *path) | |
1013 | { | |
1014 | int ret; | |
1015 | struct btrfs_key key; | |
1016 | struct inode *inode; | |
1017 | ||
1018 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1019 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1020 | key.offset = (u64)-1; | |
d397712b | 1021 | while (1) { |
e02119d5 CM |
1022 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1023 | if (ret < 0) | |
1024 | break; | |
1025 | ||
1026 | if (ret == 1) { | |
1027 | if (path->slots[0] == 0) | |
1028 | break; | |
1029 | path->slots[0]--; | |
1030 | } | |
1031 | ||
1032 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1033 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || | |
1034 | key.type != BTRFS_ORPHAN_ITEM_KEY) | |
1035 | break; | |
1036 | ||
1037 | ret = btrfs_del_item(trans, root, path); | |
1038 | BUG_ON(ret); | |
1039 | ||
1040 | btrfs_release_path(root, path); | |
1041 | inode = read_one_inode(root, key.offset); | |
1042 | BUG_ON(!inode); | |
1043 | ||
1044 | ret = fixup_inode_link_count(trans, root, inode); | |
1045 | BUG_ON(ret); | |
1046 | ||
1047 | iput(inode); | |
1048 | ||
12fcfd22 CM |
1049 | /* |
1050 | * fixup on a directory may create new entries, | |
1051 | * make sure we always look for the highset possible | |
1052 | * offset | |
1053 | */ | |
1054 | key.offset = (u64)-1; | |
e02119d5 CM |
1055 | } |
1056 | btrfs_release_path(root, path); | |
1057 | return 0; | |
1058 | } | |
1059 | ||
1060 | ||
1061 | /* | |
1062 | * record a given inode in the fixup dir so we can check its link | |
1063 | * count when replay is done. The link count is incremented here | |
1064 | * so the inode won't go away until we check it | |
1065 | */ | |
1066 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |
1067 | struct btrfs_root *root, | |
1068 | struct btrfs_path *path, | |
1069 | u64 objectid) | |
1070 | { | |
1071 | struct btrfs_key key; | |
1072 | int ret = 0; | |
1073 | struct inode *inode; | |
1074 | ||
1075 | inode = read_one_inode(root, objectid); | |
1076 | BUG_ON(!inode); | |
1077 | ||
1078 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; | |
1079 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
1080 | key.offset = objectid; | |
1081 | ||
1082 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1083 | ||
1084 | btrfs_release_path(root, path); | |
1085 | if (ret == 0) { | |
1086 | btrfs_inc_nlink(inode); | |
1087 | btrfs_update_inode(trans, root, inode); | |
1088 | } else if (ret == -EEXIST) { | |
1089 | ret = 0; | |
1090 | } else { | |
1091 | BUG(); | |
1092 | } | |
1093 | iput(inode); | |
1094 | ||
1095 | return ret; | |
1096 | } | |
1097 | ||
1098 | /* | |
1099 | * when replaying the log for a directory, we only insert names | |
1100 | * for inodes that actually exist. This means an fsync on a directory | |
1101 | * does not implicitly fsync all the new files in it | |
1102 | */ | |
1103 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, | |
1104 | struct btrfs_root *root, | |
1105 | struct btrfs_path *path, | |
1106 | u64 dirid, u64 index, | |
1107 | char *name, int name_len, u8 type, | |
1108 | struct btrfs_key *location) | |
1109 | { | |
1110 | struct inode *inode; | |
1111 | struct inode *dir; | |
1112 | int ret; | |
1113 | ||
1114 | inode = read_one_inode(root, location->objectid); | |
1115 | if (!inode) | |
1116 | return -ENOENT; | |
1117 | ||
1118 | dir = read_one_inode(root, dirid); | |
1119 | if (!dir) { | |
1120 | iput(inode); | |
1121 | return -EIO; | |
1122 | } | |
1123 | ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index); | |
1124 | ||
1125 | /* FIXME, put inode into FIXUP list */ | |
1126 | ||
1127 | iput(inode); | |
1128 | iput(dir); | |
1129 | return ret; | |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * take a single entry in a log directory item and replay it into | |
1134 | * the subvolume. | |
1135 | * | |
1136 | * if a conflicting item exists in the subdirectory already, | |
1137 | * the inode it points to is unlinked and put into the link count | |
1138 | * fix up tree. | |
1139 | * | |
1140 | * If a name from the log points to a file or directory that does | |
1141 | * not exist in the FS, it is skipped. fsyncs on directories | |
1142 | * do not force down inodes inside that directory, just changes to the | |
1143 | * names or unlinks in a directory. | |
1144 | */ | |
1145 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, | |
1146 | struct btrfs_root *root, | |
1147 | struct btrfs_path *path, | |
1148 | struct extent_buffer *eb, | |
1149 | struct btrfs_dir_item *di, | |
1150 | struct btrfs_key *key) | |
1151 | { | |
1152 | char *name; | |
1153 | int name_len; | |
1154 | struct btrfs_dir_item *dst_di; | |
1155 | struct btrfs_key found_key; | |
1156 | struct btrfs_key log_key; | |
1157 | struct inode *dir; | |
e02119d5 | 1158 | u8 log_type; |
4bef0848 | 1159 | int exists; |
e02119d5 CM |
1160 | int ret; |
1161 | ||
1162 | dir = read_one_inode(root, key->objectid); | |
1163 | BUG_ON(!dir); | |
1164 | ||
1165 | name_len = btrfs_dir_name_len(eb, di); | |
1166 | name = kmalloc(name_len, GFP_NOFS); | |
1167 | log_type = btrfs_dir_type(eb, di); | |
1168 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1169 | name_len); | |
1170 | ||
1171 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); | |
4bef0848 CM |
1172 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
1173 | if (exists == 0) | |
1174 | exists = 1; | |
1175 | else | |
1176 | exists = 0; | |
1177 | btrfs_release_path(root, path); | |
1178 | ||
e02119d5 CM |
1179 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
1180 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, | |
1181 | name, name_len, 1); | |
d397712b | 1182 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1183 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
1184 | key->objectid, | |
1185 | key->offset, name, | |
1186 | name_len, 1); | |
1187 | } else { | |
1188 | BUG(); | |
1189 | } | |
1190 | if (!dst_di || IS_ERR(dst_di)) { | |
1191 | /* we need a sequence number to insert, so we only | |
1192 | * do inserts for the BTRFS_DIR_INDEX_KEY types | |
1193 | */ | |
1194 | if (key->type != BTRFS_DIR_INDEX_KEY) | |
1195 | goto out; | |
1196 | goto insert; | |
1197 | } | |
1198 | ||
1199 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); | |
1200 | /* the existing item matches the logged item */ | |
1201 | if (found_key.objectid == log_key.objectid && | |
1202 | found_key.type == log_key.type && | |
1203 | found_key.offset == log_key.offset && | |
1204 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { | |
1205 | goto out; | |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * don't drop the conflicting directory entry if the inode | |
1210 | * for the new entry doesn't exist | |
1211 | */ | |
4bef0848 | 1212 | if (!exists) |
e02119d5 CM |
1213 | goto out; |
1214 | ||
e02119d5 CM |
1215 | ret = drop_one_dir_item(trans, root, path, dir, dst_di); |
1216 | BUG_ON(ret); | |
1217 | ||
1218 | if (key->type == BTRFS_DIR_INDEX_KEY) | |
1219 | goto insert; | |
1220 | out: | |
1221 | btrfs_release_path(root, path); | |
1222 | kfree(name); | |
1223 | iput(dir); | |
1224 | return 0; | |
1225 | ||
1226 | insert: | |
1227 | btrfs_release_path(root, path); | |
1228 | ret = insert_one_name(trans, root, path, key->objectid, key->offset, | |
1229 | name, name_len, log_type, &log_key); | |
1230 | ||
c293498b | 1231 | BUG_ON(ret && ret != -ENOENT); |
e02119d5 CM |
1232 | goto out; |
1233 | } | |
1234 | ||
1235 | /* | |
1236 | * find all the names in a directory item and reconcile them into | |
1237 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than | |
1238 | * one name in a directory item, but the same code gets used for | |
1239 | * both directory index types | |
1240 | */ | |
1241 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, | |
1242 | struct btrfs_root *root, | |
1243 | struct btrfs_path *path, | |
1244 | struct extent_buffer *eb, int slot, | |
1245 | struct btrfs_key *key) | |
1246 | { | |
1247 | int ret; | |
1248 | u32 item_size = btrfs_item_size_nr(eb, slot); | |
1249 | struct btrfs_dir_item *di; | |
1250 | int name_len; | |
1251 | unsigned long ptr; | |
1252 | unsigned long ptr_end; | |
1253 | ||
1254 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1255 | ptr_end = ptr + item_size; | |
d397712b | 1256 | while (ptr < ptr_end) { |
e02119d5 CM |
1257 | di = (struct btrfs_dir_item *)ptr; |
1258 | name_len = btrfs_dir_name_len(eb, di); | |
1259 | ret = replay_one_name(trans, root, path, eb, di, key); | |
1260 | BUG_ON(ret); | |
1261 | ptr = (unsigned long)(di + 1); | |
1262 | ptr += name_len; | |
1263 | } | |
1264 | return 0; | |
1265 | } | |
1266 | ||
1267 | /* | |
1268 | * directory replay has two parts. There are the standard directory | |
1269 | * items in the log copied from the subvolume, and range items | |
1270 | * created in the log while the subvolume was logged. | |
1271 | * | |
1272 | * The range items tell us which parts of the key space the log | |
1273 | * is authoritative for. During replay, if a key in the subvolume | |
1274 | * directory is in a logged range item, but not actually in the log | |
1275 | * that means it was deleted from the directory before the fsync | |
1276 | * and should be removed. | |
1277 | */ | |
1278 | static noinline int find_dir_range(struct btrfs_root *root, | |
1279 | struct btrfs_path *path, | |
1280 | u64 dirid, int key_type, | |
1281 | u64 *start_ret, u64 *end_ret) | |
1282 | { | |
1283 | struct btrfs_key key; | |
1284 | u64 found_end; | |
1285 | struct btrfs_dir_log_item *item; | |
1286 | int ret; | |
1287 | int nritems; | |
1288 | ||
1289 | if (*start_ret == (u64)-1) | |
1290 | return 1; | |
1291 | ||
1292 | key.objectid = dirid; | |
1293 | key.type = key_type; | |
1294 | key.offset = *start_ret; | |
1295 | ||
1296 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1297 | if (ret < 0) | |
1298 | goto out; | |
1299 | if (ret > 0) { | |
1300 | if (path->slots[0] == 0) | |
1301 | goto out; | |
1302 | path->slots[0]--; | |
1303 | } | |
1304 | if (ret != 0) | |
1305 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1306 | ||
1307 | if (key.type != key_type || key.objectid != dirid) { | |
1308 | ret = 1; | |
1309 | goto next; | |
1310 | } | |
1311 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1312 | struct btrfs_dir_log_item); | |
1313 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1314 | ||
1315 | if (*start_ret >= key.offset && *start_ret <= found_end) { | |
1316 | ret = 0; | |
1317 | *start_ret = key.offset; | |
1318 | *end_ret = found_end; | |
1319 | goto out; | |
1320 | } | |
1321 | ret = 1; | |
1322 | next: | |
1323 | /* check the next slot in the tree to see if it is a valid item */ | |
1324 | nritems = btrfs_header_nritems(path->nodes[0]); | |
1325 | if (path->slots[0] >= nritems) { | |
1326 | ret = btrfs_next_leaf(root, path); | |
1327 | if (ret) | |
1328 | goto out; | |
1329 | } else { | |
1330 | path->slots[0]++; | |
1331 | } | |
1332 | ||
1333 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
1334 | ||
1335 | if (key.type != key_type || key.objectid != dirid) { | |
1336 | ret = 1; | |
1337 | goto out; | |
1338 | } | |
1339 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1340 | struct btrfs_dir_log_item); | |
1341 | found_end = btrfs_dir_log_end(path->nodes[0], item); | |
1342 | *start_ret = key.offset; | |
1343 | *end_ret = found_end; | |
1344 | ret = 0; | |
1345 | out: | |
1346 | btrfs_release_path(root, path); | |
1347 | return ret; | |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * this looks for a given directory item in the log. If the directory | |
1352 | * item is not in the log, the item is removed and the inode it points | |
1353 | * to is unlinked | |
1354 | */ | |
1355 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, | |
1356 | struct btrfs_root *root, | |
1357 | struct btrfs_root *log, | |
1358 | struct btrfs_path *path, | |
1359 | struct btrfs_path *log_path, | |
1360 | struct inode *dir, | |
1361 | struct btrfs_key *dir_key) | |
1362 | { | |
1363 | int ret; | |
1364 | struct extent_buffer *eb; | |
1365 | int slot; | |
1366 | u32 item_size; | |
1367 | struct btrfs_dir_item *di; | |
1368 | struct btrfs_dir_item *log_di; | |
1369 | int name_len; | |
1370 | unsigned long ptr; | |
1371 | unsigned long ptr_end; | |
1372 | char *name; | |
1373 | struct inode *inode; | |
1374 | struct btrfs_key location; | |
1375 | ||
1376 | again: | |
1377 | eb = path->nodes[0]; | |
1378 | slot = path->slots[0]; | |
1379 | item_size = btrfs_item_size_nr(eb, slot); | |
1380 | ptr = btrfs_item_ptr_offset(eb, slot); | |
1381 | ptr_end = ptr + item_size; | |
d397712b | 1382 | while (ptr < ptr_end) { |
e02119d5 CM |
1383 | di = (struct btrfs_dir_item *)ptr; |
1384 | name_len = btrfs_dir_name_len(eb, di); | |
1385 | name = kmalloc(name_len, GFP_NOFS); | |
1386 | if (!name) { | |
1387 | ret = -ENOMEM; | |
1388 | goto out; | |
1389 | } | |
1390 | read_extent_buffer(eb, name, (unsigned long)(di + 1), | |
1391 | name_len); | |
1392 | log_di = NULL; | |
12fcfd22 | 1393 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
e02119d5 CM |
1394 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
1395 | dir_key->objectid, | |
1396 | name, name_len, 0); | |
12fcfd22 | 1397 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
e02119d5 CM |
1398 | log_di = btrfs_lookup_dir_index_item(trans, log, |
1399 | log_path, | |
1400 | dir_key->objectid, | |
1401 | dir_key->offset, | |
1402 | name, name_len, 0); | |
1403 | } | |
1404 | if (!log_di || IS_ERR(log_di)) { | |
1405 | btrfs_dir_item_key_to_cpu(eb, di, &location); | |
1406 | btrfs_release_path(root, path); | |
1407 | btrfs_release_path(log, log_path); | |
1408 | inode = read_one_inode(root, location.objectid); | |
1409 | BUG_ON(!inode); | |
1410 | ||
1411 | ret = link_to_fixup_dir(trans, root, | |
1412 | path, location.objectid); | |
1413 | BUG_ON(ret); | |
1414 | btrfs_inc_nlink(inode); | |
1415 | ret = btrfs_unlink_inode(trans, root, dir, inode, | |
1416 | name, name_len); | |
1417 | BUG_ON(ret); | |
1418 | kfree(name); | |
1419 | iput(inode); | |
1420 | ||
1421 | /* there might still be more names under this key | |
1422 | * check and repeat if required | |
1423 | */ | |
1424 | ret = btrfs_search_slot(NULL, root, dir_key, path, | |
1425 | 0, 0); | |
1426 | if (ret == 0) | |
1427 | goto again; | |
1428 | ret = 0; | |
1429 | goto out; | |
1430 | } | |
1431 | btrfs_release_path(log, log_path); | |
1432 | kfree(name); | |
1433 | ||
1434 | ptr = (unsigned long)(di + 1); | |
1435 | ptr += name_len; | |
1436 | } | |
1437 | ret = 0; | |
1438 | out: | |
1439 | btrfs_release_path(root, path); | |
1440 | btrfs_release_path(log, log_path); | |
1441 | return ret; | |
1442 | } | |
1443 | ||
1444 | /* | |
1445 | * deletion replay happens before we copy any new directory items | |
1446 | * out of the log or out of backreferences from inodes. It | |
1447 | * scans the log to find ranges of keys that log is authoritative for, | |
1448 | * and then scans the directory to find items in those ranges that are | |
1449 | * not present in the log. | |
1450 | * | |
1451 | * Anything we don't find in the log is unlinked and removed from the | |
1452 | * directory. | |
1453 | */ | |
1454 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, | |
1455 | struct btrfs_root *root, | |
1456 | struct btrfs_root *log, | |
1457 | struct btrfs_path *path, | |
12fcfd22 | 1458 | u64 dirid, int del_all) |
e02119d5 CM |
1459 | { |
1460 | u64 range_start; | |
1461 | u64 range_end; | |
1462 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; | |
1463 | int ret = 0; | |
1464 | struct btrfs_key dir_key; | |
1465 | struct btrfs_key found_key; | |
1466 | struct btrfs_path *log_path; | |
1467 | struct inode *dir; | |
1468 | ||
1469 | dir_key.objectid = dirid; | |
1470 | dir_key.type = BTRFS_DIR_ITEM_KEY; | |
1471 | log_path = btrfs_alloc_path(); | |
1472 | if (!log_path) | |
1473 | return -ENOMEM; | |
1474 | ||
1475 | dir = read_one_inode(root, dirid); | |
1476 | /* it isn't an error if the inode isn't there, that can happen | |
1477 | * because we replay the deletes before we copy in the inode item | |
1478 | * from the log | |
1479 | */ | |
1480 | if (!dir) { | |
1481 | btrfs_free_path(log_path); | |
1482 | return 0; | |
1483 | } | |
1484 | again: | |
1485 | range_start = 0; | |
1486 | range_end = 0; | |
d397712b | 1487 | while (1) { |
12fcfd22 CM |
1488 | if (del_all) |
1489 | range_end = (u64)-1; | |
1490 | else { | |
1491 | ret = find_dir_range(log, path, dirid, key_type, | |
1492 | &range_start, &range_end); | |
1493 | if (ret != 0) | |
1494 | break; | |
1495 | } | |
e02119d5 CM |
1496 | |
1497 | dir_key.offset = range_start; | |
d397712b | 1498 | while (1) { |
e02119d5 CM |
1499 | int nritems; |
1500 | ret = btrfs_search_slot(NULL, root, &dir_key, path, | |
1501 | 0, 0); | |
1502 | if (ret < 0) | |
1503 | goto out; | |
1504 | ||
1505 | nritems = btrfs_header_nritems(path->nodes[0]); | |
1506 | if (path->slots[0] >= nritems) { | |
1507 | ret = btrfs_next_leaf(root, path); | |
1508 | if (ret) | |
1509 | break; | |
1510 | } | |
1511 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1512 | path->slots[0]); | |
1513 | if (found_key.objectid != dirid || | |
1514 | found_key.type != dir_key.type) | |
1515 | goto next_type; | |
1516 | ||
1517 | if (found_key.offset > range_end) | |
1518 | break; | |
1519 | ||
1520 | ret = check_item_in_log(trans, root, log, path, | |
12fcfd22 CM |
1521 | log_path, dir, |
1522 | &found_key); | |
e02119d5 CM |
1523 | BUG_ON(ret); |
1524 | if (found_key.offset == (u64)-1) | |
1525 | break; | |
1526 | dir_key.offset = found_key.offset + 1; | |
1527 | } | |
1528 | btrfs_release_path(root, path); | |
1529 | if (range_end == (u64)-1) | |
1530 | break; | |
1531 | range_start = range_end + 1; | |
1532 | } | |
1533 | ||
1534 | next_type: | |
1535 | ret = 0; | |
1536 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { | |
1537 | key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
1538 | dir_key.type = BTRFS_DIR_INDEX_KEY; | |
1539 | btrfs_release_path(root, path); | |
1540 | goto again; | |
1541 | } | |
1542 | out: | |
1543 | btrfs_release_path(root, path); | |
1544 | btrfs_free_path(log_path); | |
1545 | iput(dir); | |
1546 | return ret; | |
1547 | } | |
1548 | ||
1549 | /* | |
1550 | * the process_func used to replay items from the log tree. This | |
1551 | * gets called in two different stages. The first stage just looks | |
1552 | * for inodes and makes sure they are all copied into the subvolume. | |
1553 | * | |
1554 | * The second stage copies all the other item types from the log into | |
1555 | * the subvolume. The two stage approach is slower, but gets rid of | |
1556 | * lots of complexity around inodes referencing other inodes that exist | |
1557 | * only in the log (references come from either directory items or inode | |
1558 | * back refs). | |
1559 | */ | |
1560 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, | |
1561 | struct walk_control *wc, u64 gen) | |
1562 | { | |
1563 | int nritems; | |
1564 | struct btrfs_path *path; | |
1565 | struct btrfs_root *root = wc->replay_dest; | |
1566 | struct btrfs_key key; | |
1567 | u32 item_size; | |
1568 | int level; | |
1569 | int i; | |
1570 | int ret; | |
1571 | ||
1572 | btrfs_read_buffer(eb, gen); | |
1573 | ||
1574 | level = btrfs_header_level(eb); | |
1575 | ||
1576 | if (level != 0) | |
1577 | return 0; | |
1578 | ||
1579 | path = btrfs_alloc_path(); | |
1580 | BUG_ON(!path); | |
1581 | ||
1582 | nritems = btrfs_header_nritems(eb); | |
1583 | for (i = 0; i < nritems; i++) { | |
1584 | btrfs_item_key_to_cpu(eb, &key, i); | |
1585 | item_size = btrfs_item_size_nr(eb, i); | |
1586 | ||
1587 | /* inode keys are done during the first stage */ | |
1588 | if (key.type == BTRFS_INODE_ITEM_KEY && | |
1589 | wc->stage == LOG_WALK_REPLAY_INODES) { | |
1590 | struct inode *inode; | |
1591 | struct btrfs_inode_item *inode_item; | |
1592 | u32 mode; | |
1593 | ||
1594 | inode_item = btrfs_item_ptr(eb, i, | |
1595 | struct btrfs_inode_item); | |
1596 | mode = btrfs_inode_mode(eb, inode_item); | |
1597 | if (S_ISDIR(mode)) { | |
1598 | ret = replay_dir_deletes(wc->trans, | |
12fcfd22 | 1599 | root, log, path, key.objectid, 0); |
e02119d5 CM |
1600 | BUG_ON(ret); |
1601 | } | |
1602 | ret = overwrite_item(wc->trans, root, path, | |
1603 | eb, i, &key); | |
1604 | BUG_ON(ret); | |
1605 | ||
1606 | /* for regular files, truncate away | |
1607 | * extents past the new EOF | |
1608 | */ | |
1609 | if (S_ISREG(mode)) { | |
1610 | inode = read_one_inode(root, | |
1611 | key.objectid); | |
1612 | BUG_ON(!inode); | |
1613 | ||
1614 | ret = btrfs_truncate_inode_items(wc->trans, | |
1615 | root, inode, inode->i_size, | |
1616 | BTRFS_EXTENT_DATA_KEY); | |
1617 | BUG_ON(ret); | |
a74ac322 CM |
1618 | |
1619 | /* if the nlink count is zero here, the iput | |
1620 | * will free the inode. We bump it to make | |
1621 | * sure it doesn't get freed until the link | |
1622 | * count fixup is done | |
1623 | */ | |
1624 | if (inode->i_nlink == 0) { | |
1625 | btrfs_inc_nlink(inode); | |
1626 | btrfs_update_inode(wc->trans, | |
1627 | root, inode); | |
1628 | } | |
e02119d5 CM |
1629 | iput(inode); |
1630 | } | |
1631 | ret = link_to_fixup_dir(wc->trans, root, | |
1632 | path, key.objectid); | |
1633 | BUG_ON(ret); | |
1634 | } | |
1635 | if (wc->stage < LOG_WALK_REPLAY_ALL) | |
1636 | continue; | |
1637 | ||
1638 | /* these keys are simply copied */ | |
1639 | if (key.type == BTRFS_XATTR_ITEM_KEY) { | |
1640 | ret = overwrite_item(wc->trans, root, path, | |
1641 | eb, i, &key); | |
1642 | BUG_ON(ret); | |
1643 | } else if (key.type == BTRFS_INODE_REF_KEY) { | |
1644 | ret = add_inode_ref(wc->trans, root, log, path, | |
1645 | eb, i, &key); | |
1646 | BUG_ON(ret && ret != -ENOENT); | |
1647 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { | |
1648 | ret = replay_one_extent(wc->trans, root, path, | |
1649 | eb, i, &key); | |
1650 | BUG_ON(ret); | |
e02119d5 CM |
1651 | } else if (key.type == BTRFS_DIR_ITEM_KEY || |
1652 | key.type == BTRFS_DIR_INDEX_KEY) { | |
1653 | ret = replay_one_dir_item(wc->trans, root, path, | |
1654 | eb, i, &key); | |
1655 | BUG_ON(ret); | |
1656 | } | |
1657 | } | |
1658 | btrfs_free_path(path); | |
1659 | return 0; | |
1660 | } | |
1661 | ||
d397712b | 1662 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
1663 | struct btrfs_root *root, |
1664 | struct btrfs_path *path, int *level, | |
1665 | struct walk_control *wc) | |
1666 | { | |
1667 | u64 root_owner; | |
1668 | u64 root_gen; | |
1669 | u64 bytenr; | |
1670 | u64 ptr_gen; | |
1671 | struct extent_buffer *next; | |
1672 | struct extent_buffer *cur; | |
1673 | struct extent_buffer *parent; | |
1674 | u32 blocksize; | |
1675 | int ret = 0; | |
1676 | ||
1677 | WARN_ON(*level < 0); | |
1678 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
1679 | ||
d397712b | 1680 | while (*level > 0) { |
e02119d5 CM |
1681 | WARN_ON(*level < 0); |
1682 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
1683 | cur = path->nodes[*level]; | |
1684 | ||
1685 | if (btrfs_header_level(cur) != *level) | |
1686 | WARN_ON(1); | |
1687 | ||
1688 | if (path->slots[*level] >= | |
1689 | btrfs_header_nritems(cur)) | |
1690 | break; | |
1691 | ||
1692 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
1693 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
1694 | blocksize = btrfs_level_size(root, *level - 1); | |
1695 | ||
1696 | parent = path->nodes[*level]; | |
1697 | root_owner = btrfs_header_owner(parent); | |
1698 | root_gen = btrfs_header_generation(parent); | |
1699 | ||
1700 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
1701 | ||
1702 | wc->process_func(root, next, wc, ptr_gen); | |
1703 | ||
1704 | if (*level == 1) { | |
1705 | path->slots[*level]++; | |
1706 | if (wc->free) { | |
1707 | btrfs_read_buffer(next, ptr_gen); | |
1708 | ||
1709 | btrfs_tree_lock(next); | |
1710 | clean_tree_block(trans, root, next); | |
b4ce94de | 1711 | btrfs_set_lock_blocking(next); |
e02119d5 CM |
1712 | btrfs_wait_tree_block_writeback(next); |
1713 | btrfs_tree_unlock(next); | |
1714 | ||
e02119d5 CM |
1715 | WARN_ON(root_owner != |
1716 | BTRFS_TREE_LOG_OBJECTID); | |
d00aff00 CM |
1717 | ret = btrfs_free_reserved_extent(root, |
1718 | bytenr, blocksize); | |
e02119d5 CM |
1719 | BUG_ON(ret); |
1720 | } | |
1721 | free_extent_buffer(next); | |
1722 | continue; | |
1723 | } | |
1724 | btrfs_read_buffer(next, ptr_gen); | |
1725 | ||
1726 | WARN_ON(*level <= 0); | |
1727 | if (path->nodes[*level-1]) | |
1728 | free_extent_buffer(path->nodes[*level-1]); | |
1729 | path->nodes[*level-1] = next; | |
1730 | *level = btrfs_header_level(next); | |
1731 | path->slots[*level] = 0; | |
1732 | cond_resched(); | |
1733 | } | |
1734 | WARN_ON(*level < 0); | |
1735 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
1736 | ||
d397712b | 1737 | if (path->nodes[*level] == root->node) |
e02119d5 | 1738 | parent = path->nodes[*level]; |
d397712b | 1739 | else |
e02119d5 | 1740 | parent = path->nodes[*level + 1]; |
d397712b | 1741 | |
e02119d5 CM |
1742 | bytenr = path->nodes[*level]->start; |
1743 | ||
1744 | blocksize = btrfs_level_size(root, *level); | |
1745 | root_owner = btrfs_header_owner(parent); | |
1746 | root_gen = btrfs_header_generation(parent); | |
1747 | ||
1748 | wc->process_func(root, path->nodes[*level], wc, | |
1749 | btrfs_header_generation(path->nodes[*level])); | |
1750 | ||
1751 | if (wc->free) { | |
1752 | next = path->nodes[*level]; | |
1753 | btrfs_tree_lock(next); | |
1754 | clean_tree_block(trans, root, next); | |
b4ce94de | 1755 | btrfs_set_lock_blocking(next); |
e02119d5 CM |
1756 | btrfs_wait_tree_block_writeback(next); |
1757 | btrfs_tree_unlock(next); | |
1758 | ||
e02119d5 | 1759 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
d00aff00 | 1760 | ret = btrfs_free_reserved_extent(root, bytenr, blocksize); |
e02119d5 CM |
1761 | BUG_ON(ret); |
1762 | } | |
1763 | free_extent_buffer(path->nodes[*level]); | |
1764 | path->nodes[*level] = NULL; | |
1765 | *level += 1; | |
1766 | ||
1767 | cond_resched(); | |
1768 | return 0; | |
1769 | } | |
1770 | ||
d397712b | 1771 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
e02119d5 CM |
1772 | struct btrfs_root *root, |
1773 | struct btrfs_path *path, int *level, | |
1774 | struct walk_control *wc) | |
1775 | { | |
1776 | u64 root_owner; | |
1777 | u64 root_gen; | |
1778 | int i; | |
1779 | int slot; | |
1780 | int ret; | |
1781 | ||
d397712b | 1782 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
e02119d5 CM |
1783 | slot = path->slots[i]; |
1784 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { | |
1785 | struct extent_buffer *node; | |
1786 | node = path->nodes[i]; | |
1787 | path->slots[i]++; | |
1788 | *level = i; | |
1789 | WARN_ON(*level == 0); | |
1790 | return 0; | |
1791 | } else { | |
31840ae1 ZY |
1792 | struct extent_buffer *parent; |
1793 | if (path->nodes[*level] == root->node) | |
1794 | parent = path->nodes[*level]; | |
1795 | else | |
1796 | parent = path->nodes[*level + 1]; | |
1797 | ||
1798 | root_owner = btrfs_header_owner(parent); | |
1799 | root_gen = btrfs_header_generation(parent); | |
e02119d5 CM |
1800 | wc->process_func(root, path->nodes[*level], wc, |
1801 | btrfs_header_generation(path->nodes[*level])); | |
1802 | if (wc->free) { | |
1803 | struct extent_buffer *next; | |
1804 | ||
1805 | next = path->nodes[*level]; | |
1806 | ||
1807 | btrfs_tree_lock(next); | |
1808 | clean_tree_block(trans, root, next); | |
b4ce94de | 1809 | btrfs_set_lock_blocking(next); |
e02119d5 CM |
1810 | btrfs_wait_tree_block_writeback(next); |
1811 | btrfs_tree_unlock(next); | |
1812 | ||
e02119d5 | 1813 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
d00aff00 | 1814 | ret = btrfs_free_reserved_extent(root, |
e02119d5 | 1815 | path->nodes[*level]->start, |
d00aff00 | 1816 | path->nodes[*level]->len); |
e02119d5 CM |
1817 | BUG_ON(ret); |
1818 | } | |
1819 | free_extent_buffer(path->nodes[*level]); | |
1820 | path->nodes[*level] = NULL; | |
1821 | *level = i + 1; | |
1822 | } | |
1823 | } | |
1824 | return 1; | |
1825 | } | |
1826 | ||
1827 | /* | |
1828 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
1829 | * the tree freeing any blocks that have a ref count of zero after being | |
1830 | * decremented. | |
1831 | */ | |
1832 | static int walk_log_tree(struct btrfs_trans_handle *trans, | |
1833 | struct btrfs_root *log, struct walk_control *wc) | |
1834 | { | |
1835 | int ret = 0; | |
1836 | int wret; | |
1837 | int level; | |
1838 | struct btrfs_path *path; | |
1839 | int i; | |
1840 | int orig_level; | |
1841 | ||
1842 | path = btrfs_alloc_path(); | |
1843 | BUG_ON(!path); | |
1844 | ||
1845 | level = btrfs_header_level(log->node); | |
1846 | orig_level = level; | |
1847 | path->nodes[level] = log->node; | |
1848 | extent_buffer_get(log->node); | |
1849 | path->slots[level] = 0; | |
1850 | ||
d397712b | 1851 | while (1) { |
e02119d5 CM |
1852 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
1853 | if (wret > 0) | |
1854 | break; | |
1855 | if (wret < 0) | |
1856 | ret = wret; | |
1857 | ||
1858 | wret = walk_up_log_tree(trans, log, path, &level, wc); | |
1859 | if (wret > 0) | |
1860 | break; | |
1861 | if (wret < 0) | |
1862 | ret = wret; | |
1863 | } | |
1864 | ||
1865 | /* was the root node processed? if not, catch it here */ | |
1866 | if (path->nodes[orig_level]) { | |
1867 | wc->process_func(log, path->nodes[orig_level], wc, | |
1868 | btrfs_header_generation(path->nodes[orig_level])); | |
1869 | if (wc->free) { | |
1870 | struct extent_buffer *next; | |
1871 | ||
1872 | next = path->nodes[orig_level]; | |
1873 | ||
1874 | btrfs_tree_lock(next); | |
1875 | clean_tree_block(trans, log, next); | |
b4ce94de | 1876 | btrfs_set_lock_blocking(next); |
e02119d5 CM |
1877 | btrfs_wait_tree_block_writeback(next); |
1878 | btrfs_tree_unlock(next); | |
1879 | ||
e02119d5 CM |
1880 | WARN_ON(log->root_key.objectid != |
1881 | BTRFS_TREE_LOG_OBJECTID); | |
d00aff00 CM |
1882 | ret = btrfs_free_reserved_extent(log, next->start, |
1883 | next->len); | |
e02119d5 CM |
1884 | BUG_ON(ret); |
1885 | } | |
1886 | } | |
1887 | ||
1888 | for (i = 0; i <= orig_level; i++) { | |
1889 | if (path->nodes[i]) { | |
1890 | free_extent_buffer(path->nodes[i]); | |
1891 | path->nodes[i] = NULL; | |
1892 | } | |
1893 | } | |
1894 | btrfs_free_path(path); | |
e02119d5 CM |
1895 | return ret; |
1896 | } | |
1897 | ||
7237f183 YZ |
1898 | /* |
1899 | * helper function to update the item for a given subvolumes log root | |
1900 | * in the tree of log roots | |
1901 | */ | |
1902 | static int update_log_root(struct btrfs_trans_handle *trans, | |
1903 | struct btrfs_root *log) | |
1904 | { | |
1905 | int ret; | |
1906 | ||
1907 | if (log->log_transid == 1) { | |
1908 | /* insert root item on the first sync */ | |
1909 | ret = btrfs_insert_root(trans, log->fs_info->log_root_tree, | |
1910 | &log->root_key, &log->root_item); | |
1911 | } else { | |
1912 | ret = btrfs_update_root(trans, log->fs_info->log_root_tree, | |
1913 | &log->root_key, &log->root_item); | |
1914 | } | |
1915 | return ret; | |
1916 | } | |
1917 | ||
12fcfd22 CM |
1918 | static int wait_log_commit(struct btrfs_trans_handle *trans, |
1919 | struct btrfs_root *root, unsigned long transid) | |
e02119d5 CM |
1920 | { |
1921 | DEFINE_WAIT(wait); | |
7237f183 | 1922 | int index = transid % 2; |
e02119d5 | 1923 | |
7237f183 YZ |
1924 | /* |
1925 | * we only allow two pending log transactions at a time, | |
1926 | * so we know that if ours is more than 2 older than the | |
1927 | * current transaction, we're done | |
1928 | */ | |
e02119d5 | 1929 | do { |
7237f183 YZ |
1930 | prepare_to_wait(&root->log_commit_wait[index], |
1931 | &wait, TASK_UNINTERRUPTIBLE); | |
1932 | mutex_unlock(&root->log_mutex); | |
12fcfd22 CM |
1933 | |
1934 | if (root->fs_info->last_trans_log_full_commit != | |
1935 | trans->transid && root->log_transid < transid + 2 && | |
7237f183 YZ |
1936 | atomic_read(&root->log_commit[index])) |
1937 | schedule(); | |
12fcfd22 | 1938 | |
7237f183 YZ |
1939 | finish_wait(&root->log_commit_wait[index], &wait); |
1940 | mutex_lock(&root->log_mutex); | |
1941 | } while (root->log_transid < transid + 2 && | |
1942 | atomic_read(&root->log_commit[index])); | |
1943 | return 0; | |
1944 | } | |
1945 | ||
12fcfd22 CM |
1946 | static int wait_for_writer(struct btrfs_trans_handle *trans, |
1947 | struct btrfs_root *root) | |
7237f183 YZ |
1948 | { |
1949 | DEFINE_WAIT(wait); | |
1950 | while (atomic_read(&root->log_writers)) { | |
1951 | prepare_to_wait(&root->log_writer_wait, | |
1952 | &wait, TASK_UNINTERRUPTIBLE); | |
1953 | mutex_unlock(&root->log_mutex); | |
12fcfd22 CM |
1954 | if (root->fs_info->last_trans_log_full_commit != |
1955 | trans->transid && atomic_read(&root->log_writers)) | |
e02119d5 | 1956 | schedule(); |
7237f183 YZ |
1957 | mutex_lock(&root->log_mutex); |
1958 | finish_wait(&root->log_writer_wait, &wait); | |
1959 | } | |
e02119d5 CM |
1960 | return 0; |
1961 | } | |
1962 | ||
1963 | /* | |
1964 | * btrfs_sync_log does sends a given tree log down to the disk and | |
1965 | * updates the super blocks to record it. When this call is done, | |
12fcfd22 CM |
1966 | * you know that any inodes previously logged are safely on disk only |
1967 | * if it returns 0. | |
1968 | * | |
1969 | * Any other return value means you need to call btrfs_commit_transaction. | |
1970 | * Some of the edge cases for fsyncing directories that have had unlinks | |
1971 | * or renames done in the past mean that sometimes the only safe | |
1972 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, | |
1973 | * that has happened. | |
e02119d5 CM |
1974 | */ |
1975 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | |
1976 | struct btrfs_root *root) | |
1977 | { | |
7237f183 YZ |
1978 | int index1; |
1979 | int index2; | |
e02119d5 | 1980 | int ret; |
e02119d5 | 1981 | struct btrfs_root *log = root->log_root; |
7237f183 | 1982 | struct btrfs_root *log_root_tree = root->fs_info->log_root_tree; |
257c62e1 | 1983 | u64 log_transid = 0; |
e02119d5 | 1984 | |
7237f183 YZ |
1985 | mutex_lock(&root->log_mutex); |
1986 | index1 = root->log_transid % 2; | |
1987 | if (atomic_read(&root->log_commit[index1])) { | |
12fcfd22 | 1988 | wait_log_commit(trans, root, root->log_transid); |
7237f183 YZ |
1989 | mutex_unlock(&root->log_mutex); |
1990 | return 0; | |
e02119d5 | 1991 | } |
7237f183 YZ |
1992 | atomic_set(&root->log_commit[index1], 1); |
1993 | ||
1994 | /* wait for previous tree log sync to complete */ | |
1995 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | |
12fcfd22 | 1996 | wait_log_commit(trans, root, root->log_transid - 1); |
e02119d5 | 1997 | |
ff782e0a | 1998 | while (root->log_multiple_pids) { |
7237f183 YZ |
1999 | unsigned long batch = root->log_batch; |
2000 | mutex_unlock(&root->log_mutex); | |
e02119d5 | 2001 | schedule_timeout_uninterruptible(1); |
7237f183 | 2002 | mutex_lock(&root->log_mutex); |
12fcfd22 CM |
2003 | |
2004 | wait_for_writer(trans, root); | |
7237f183 | 2005 | if (batch == root->log_batch) |
e02119d5 CM |
2006 | break; |
2007 | } | |
e02119d5 | 2008 | |
12fcfd22 CM |
2009 | /* bail out if we need to do a full commit */ |
2010 | if (root->fs_info->last_trans_log_full_commit == trans->transid) { | |
2011 | ret = -EAGAIN; | |
2012 | mutex_unlock(&root->log_mutex); | |
2013 | goto out; | |
2014 | } | |
2015 | ||
d0c803c4 | 2016 | ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages); |
e02119d5 | 2017 | BUG_ON(ret); |
7237f183 | 2018 | |
5d4f98a2 | 2019 | btrfs_set_root_node(&log->root_item, log->node); |
7237f183 YZ |
2020 | |
2021 | root->log_batch = 0; | |
257c62e1 | 2022 | log_transid = root->log_transid; |
7237f183 YZ |
2023 | root->log_transid++; |
2024 | log->log_transid = root->log_transid; | |
ff782e0a | 2025 | root->log_start_pid = 0; |
7237f183 YZ |
2026 | smp_mb(); |
2027 | /* | |
2028 | * log tree has been flushed to disk, new modifications of | |
2029 | * the log will be written to new positions. so it's safe to | |
2030 | * allow log writers to go in. | |
2031 | */ | |
2032 | mutex_unlock(&root->log_mutex); | |
2033 | ||
2034 | mutex_lock(&log_root_tree->log_mutex); | |
2035 | log_root_tree->log_batch++; | |
2036 | atomic_inc(&log_root_tree->log_writers); | |
2037 | mutex_unlock(&log_root_tree->log_mutex); | |
2038 | ||
2039 | ret = update_log_root(trans, log); | |
2040 | BUG_ON(ret); | |
2041 | ||
2042 | mutex_lock(&log_root_tree->log_mutex); | |
2043 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | |
2044 | smp_mb(); | |
2045 | if (waitqueue_active(&log_root_tree->log_writer_wait)) | |
2046 | wake_up(&log_root_tree->log_writer_wait); | |
2047 | } | |
2048 | ||
2049 | index2 = log_root_tree->log_transid % 2; | |
2050 | if (atomic_read(&log_root_tree->log_commit[index2])) { | |
12fcfd22 CM |
2051 | wait_log_commit(trans, log_root_tree, |
2052 | log_root_tree->log_transid); | |
7237f183 YZ |
2053 | mutex_unlock(&log_root_tree->log_mutex); |
2054 | goto out; | |
2055 | } | |
2056 | atomic_set(&log_root_tree->log_commit[index2], 1); | |
2057 | ||
12fcfd22 CM |
2058 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
2059 | wait_log_commit(trans, log_root_tree, | |
2060 | log_root_tree->log_transid - 1); | |
2061 | } | |
2062 | ||
2063 | wait_for_writer(trans, log_root_tree); | |
7237f183 | 2064 | |
12fcfd22 CM |
2065 | /* |
2066 | * now that we've moved on to the tree of log tree roots, | |
2067 | * check the full commit flag again | |
2068 | */ | |
2069 | if (root->fs_info->last_trans_log_full_commit == trans->transid) { | |
2070 | mutex_unlock(&log_root_tree->log_mutex); | |
2071 | ret = -EAGAIN; | |
2072 | goto out_wake_log_root; | |
2073 | } | |
7237f183 YZ |
2074 | |
2075 | ret = btrfs_write_and_wait_marked_extents(log_root_tree, | |
2076 | &log_root_tree->dirty_log_pages); | |
e02119d5 CM |
2077 | BUG_ON(ret); |
2078 | ||
2079 | btrfs_set_super_log_root(&root->fs_info->super_for_commit, | |
7237f183 | 2080 | log_root_tree->node->start); |
e02119d5 | 2081 | btrfs_set_super_log_root_level(&root->fs_info->super_for_commit, |
7237f183 | 2082 | btrfs_header_level(log_root_tree->node)); |
e02119d5 | 2083 | |
7237f183 YZ |
2084 | log_root_tree->log_batch = 0; |
2085 | log_root_tree->log_transid++; | |
e02119d5 | 2086 | smp_mb(); |
7237f183 YZ |
2087 | |
2088 | mutex_unlock(&log_root_tree->log_mutex); | |
2089 | ||
2090 | /* | |
2091 | * nobody else is going to jump in and write the the ctree | |
2092 | * super here because the log_commit atomic below is protecting | |
2093 | * us. We must be called with a transaction handle pinning | |
2094 | * the running transaction open, so a full commit can't hop | |
2095 | * in and cause problems either. | |
2096 | */ | |
4722607d | 2097 | write_ctree_super(trans, root->fs_info->tree_root, 1); |
12fcfd22 | 2098 | ret = 0; |
7237f183 | 2099 | |
257c62e1 CM |
2100 | mutex_lock(&root->log_mutex); |
2101 | if (root->last_log_commit < log_transid) | |
2102 | root->last_log_commit = log_transid; | |
2103 | mutex_unlock(&root->log_mutex); | |
2104 | ||
12fcfd22 | 2105 | out_wake_log_root: |
7237f183 YZ |
2106 | atomic_set(&log_root_tree->log_commit[index2], 0); |
2107 | smp_mb(); | |
2108 | if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) | |
2109 | wake_up(&log_root_tree->log_commit_wait[index2]); | |
e02119d5 | 2110 | out: |
7237f183 YZ |
2111 | atomic_set(&root->log_commit[index1], 0); |
2112 | smp_mb(); | |
2113 | if (waitqueue_active(&root->log_commit_wait[index1])) | |
2114 | wake_up(&root->log_commit_wait[index1]); | |
e02119d5 | 2115 | return 0; |
e02119d5 CM |
2116 | } |
2117 | ||
12fcfd22 CM |
2118 | /* |
2119 | * free all the extents used by the tree log. This should be called | |
e02119d5 CM |
2120 | * at commit time of the full transaction |
2121 | */ | |
2122 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |
2123 | { | |
2124 | int ret; | |
2125 | struct btrfs_root *log; | |
2126 | struct key; | |
d0c803c4 CM |
2127 | u64 start; |
2128 | u64 end; | |
e02119d5 CM |
2129 | struct walk_control wc = { |
2130 | .free = 1, | |
2131 | .process_func = process_one_buffer | |
2132 | }; | |
2133 | ||
07d400a6 | 2134 | if (!root->log_root || root->fs_info->log_root_recovering) |
e02119d5 CM |
2135 | return 0; |
2136 | ||
2137 | log = root->log_root; | |
2138 | ret = walk_log_tree(trans, log, &wc); | |
2139 | BUG_ON(ret); | |
2140 | ||
d397712b | 2141 | while (1) { |
d0c803c4 CM |
2142 | ret = find_first_extent_bit(&log->dirty_log_pages, |
2143 | 0, &start, &end, EXTENT_DIRTY); | |
2144 | if (ret) | |
2145 | break; | |
2146 | ||
2147 | clear_extent_dirty(&log->dirty_log_pages, | |
2148 | start, end, GFP_NOFS); | |
2149 | } | |
2150 | ||
7237f183 YZ |
2151 | if (log->log_transid > 0) { |
2152 | ret = btrfs_del_root(trans, root->fs_info->log_root_tree, | |
2153 | &log->root_key); | |
2154 | BUG_ON(ret); | |
2155 | } | |
e02119d5 | 2156 | root->log_root = NULL; |
7237f183 YZ |
2157 | free_extent_buffer(log->node); |
2158 | kfree(log); | |
e02119d5 CM |
2159 | return 0; |
2160 | } | |
2161 | ||
e02119d5 CM |
2162 | /* |
2163 | * If both a file and directory are logged, and unlinks or renames are | |
2164 | * mixed in, we have a few interesting corners: | |
2165 | * | |
2166 | * create file X in dir Y | |
2167 | * link file X to X.link in dir Y | |
2168 | * fsync file X | |
2169 | * unlink file X but leave X.link | |
2170 | * fsync dir Y | |
2171 | * | |
2172 | * After a crash we would expect only X.link to exist. But file X | |
2173 | * didn't get fsync'd again so the log has back refs for X and X.link. | |
2174 | * | |
2175 | * We solve this by removing directory entries and inode backrefs from the | |
2176 | * log when a file that was logged in the current transaction is | |
2177 | * unlinked. Any later fsync will include the updated log entries, and | |
2178 | * we'll be able to reconstruct the proper directory items from backrefs. | |
2179 | * | |
2180 | * This optimizations allows us to avoid relogging the entire inode | |
2181 | * or the entire directory. | |
2182 | */ | |
2183 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, | |
2184 | struct btrfs_root *root, | |
2185 | const char *name, int name_len, | |
2186 | struct inode *dir, u64 index) | |
2187 | { | |
2188 | struct btrfs_root *log; | |
2189 | struct btrfs_dir_item *di; | |
2190 | struct btrfs_path *path; | |
2191 | int ret; | |
2192 | int bytes_del = 0; | |
2193 | ||
3a5f1d45 CM |
2194 | if (BTRFS_I(dir)->logged_trans < trans->transid) |
2195 | return 0; | |
2196 | ||
e02119d5 CM |
2197 | ret = join_running_log_trans(root); |
2198 | if (ret) | |
2199 | return 0; | |
2200 | ||
2201 | mutex_lock(&BTRFS_I(dir)->log_mutex); | |
2202 | ||
2203 | log = root->log_root; | |
2204 | path = btrfs_alloc_path(); | |
2205 | di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino, | |
2206 | name, name_len, -1); | |
2207 | if (di && !IS_ERR(di)) { | |
2208 | ret = btrfs_delete_one_dir_name(trans, log, path, di); | |
2209 | bytes_del += name_len; | |
2210 | BUG_ON(ret); | |
2211 | } | |
2212 | btrfs_release_path(log, path); | |
2213 | di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino, | |
2214 | index, name, name_len, -1); | |
2215 | if (di && !IS_ERR(di)) { | |
2216 | ret = btrfs_delete_one_dir_name(trans, log, path, di); | |
2217 | bytes_del += name_len; | |
2218 | BUG_ON(ret); | |
2219 | } | |
2220 | ||
2221 | /* update the directory size in the log to reflect the names | |
2222 | * we have removed | |
2223 | */ | |
2224 | if (bytes_del) { | |
2225 | struct btrfs_key key; | |
2226 | ||
2227 | key.objectid = dir->i_ino; | |
2228 | key.offset = 0; | |
2229 | key.type = BTRFS_INODE_ITEM_KEY; | |
2230 | btrfs_release_path(log, path); | |
2231 | ||
2232 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); | |
2233 | if (ret == 0) { | |
2234 | struct btrfs_inode_item *item; | |
2235 | u64 i_size; | |
2236 | ||
2237 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2238 | struct btrfs_inode_item); | |
2239 | i_size = btrfs_inode_size(path->nodes[0], item); | |
2240 | if (i_size > bytes_del) | |
2241 | i_size -= bytes_del; | |
2242 | else | |
2243 | i_size = 0; | |
2244 | btrfs_set_inode_size(path->nodes[0], item, i_size); | |
2245 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
2246 | } else | |
2247 | ret = 0; | |
2248 | btrfs_release_path(log, path); | |
2249 | } | |
2250 | ||
2251 | btrfs_free_path(path); | |
2252 | mutex_unlock(&BTRFS_I(dir)->log_mutex); | |
12fcfd22 | 2253 | btrfs_end_log_trans(root); |
e02119d5 CM |
2254 | |
2255 | return 0; | |
2256 | } | |
2257 | ||
2258 | /* see comments for btrfs_del_dir_entries_in_log */ | |
2259 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, | |
2260 | struct btrfs_root *root, | |
2261 | const char *name, int name_len, | |
2262 | struct inode *inode, u64 dirid) | |
2263 | { | |
2264 | struct btrfs_root *log; | |
2265 | u64 index; | |
2266 | int ret; | |
2267 | ||
3a5f1d45 CM |
2268 | if (BTRFS_I(inode)->logged_trans < trans->transid) |
2269 | return 0; | |
2270 | ||
e02119d5 CM |
2271 | ret = join_running_log_trans(root); |
2272 | if (ret) | |
2273 | return 0; | |
2274 | log = root->log_root; | |
2275 | mutex_lock(&BTRFS_I(inode)->log_mutex); | |
2276 | ||
2277 | ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino, | |
2278 | dirid, &index); | |
2279 | mutex_unlock(&BTRFS_I(inode)->log_mutex); | |
12fcfd22 | 2280 | btrfs_end_log_trans(root); |
e02119d5 | 2281 | |
e02119d5 CM |
2282 | return ret; |
2283 | } | |
2284 | ||
2285 | /* | |
2286 | * creates a range item in the log for 'dirid'. first_offset and | |
2287 | * last_offset tell us which parts of the key space the log should | |
2288 | * be considered authoritative for. | |
2289 | */ | |
2290 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, | |
2291 | struct btrfs_root *log, | |
2292 | struct btrfs_path *path, | |
2293 | int key_type, u64 dirid, | |
2294 | u64 first_offset, u64 last_offset) | |
2295 | { | |
2296 | int ret; | |
2297 | struct btrfs_key key; | |
2298 | struct btrfs_dir_log_item *item; | |
2299 | ||
2300 | key.objectid = dirid; | |
2301 | key.offset = first_offset; | |
2302 | if (key_type == BTRFS_DIR_ITEM_KEY) | |
2303 | key.type = BTRFS_DIR_LOG_ITEM_KEY; | |
2304 | else | |
2305 | key.type = BTRFS_DIR_LOG_INDEX_KEY; | |
2306 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); | |
2307 | BUG_ON(ret); | |
2308 | ||
2309 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2310 | struct btrfs_dir_log_item); | |
2311 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); | |
2312 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
2313 | btrfs_release_path(log, path); | |
2314 | return 0; | |
2315 | } | |
2316 | ||
2317 | /* | |
2318 | * log all the items included in the current transaction for a given | |
2319 | * directory. This also creates the range items in the log tree required | |
2320 | * to replay anything deleted before the fsync | |
2321 | */ | |
2322 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, | |
2323 | struct btrfs_root *root, struct inode *inode, | |
2324 | struct btrfs_path *path, | |
2325 | struct btrfs_path *dst_path, int key_type, | |
2326 | u64 min_offset, u64 *last_offset_ret) | |
2327 | { | |
2328 | struct btrfs_key min_key; | |
2329 | struct btrfs_key max_key; | |
2330 | struct btrfs_root *log = root->log_root; | |
2331 | struct extent_buffer *src; | |
2332 | int ret; | |
2333 | int i; | |
2334 | int nritems; | |
2335 | u64 first_offset = min_offset; | |
2336 | u64 last_offset = (u64)-1; | |
2337 | ||
2338 | log = root->log_root; | |
2339 | max_key.objectid = inode->i_ino; | |
2340 | max_key.offset = (u64)-1; | |
2341 | max_key.type = key_type; | |
2342 | ||
2343 | min_key.objectid = inode->i_ino; | |
2344 | min_key.type = key_type; | |
2345 | min_key.offset = min_offset; | |
2346 | ||
2347 | path->keep_locks = 1; | |
2348 | ||
2349 | ret = btrfs_search_forward(root, &min_key, &max_key, | |
2350 | path, 0, trans->transid); | |
2351 | ||
2352 | /* | |
2353 | * we didn't find anything from this transaction, see if there | |
2354 | * is anything at all | |
2355 | */ | |
2356 | if (ret != 0 || min_key.objectid != inode->i_ino || | |
2357 | min_key.type != key_type) { | |
2358 | min_key.objectid = inode->i_ino; | |
2359 | min_key.type = key_type; | |
2360 | min_key.offset = (u64)-1; | |
2361 | btrfs_release_path(root, path); | |
2362 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); | |
2363 | if (ret < 0) { | |
2364 | btrfs_release_path(root, path); | |
2365 | return ret; | |
2366 | } | |
2367 | ret = btrfs_previous_item(root, path, inode->i_ino, key_type); | |
2368 | ||
2369 | /* if ret == 0 there are items for this type, | |
2370 | * create a range to tell us the last key of this type. | |
2371 | * otherwise, there are no items in this directory after | |
2372 | * *min_offset, and we create a range to indicate that. | |
2373 | */ | |
2374 | if (ret == 0) { | |
2375 | struct btrfs_key tmp; | |
2376 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, | |
2377 | path->slots[0]); | |
d397712b | 2378 | if (key_type == tmp.type) |
e02119d5 | 2379 | first_offset = max(min_offset, tmp.offset) + 1; |
e02119d5 CM |
2380 | } |
2381 | goto done; | |
2382 | } | |
2383 | ||
2384 | /* go backward to find any previous key */ | |
2385 | ret = btrfs_previous_item(root, path, inode->i_ino, key_type); | |
2386 | if (ret == 0) { | |
2387 | struct btrfs_key tmp; | |
2388 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
2389 | if (key_type == tmp.type) { | |
2390 | first_offset = tmp.offset; | |
2391 | ret = overwrite_item(trans, log, dst_path, | |
2392 | path->nodes[0], path->slots[0], | |
2393 | &tmp); | |
2394 | } | |
2395 | } | |
2396 | btrfs_release_path(root, path); | |
2397 | ||
2398 | /* find the first key from this transaction again */ | |
2399 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); | |
2400 | if (ret != 0) { | |
2401 | WARN_ON(1); | |
2402 | goto done; | |
2403 | } | |
2404 | ||
2405 | /* | |
2406 | * we have a block from this transaction, log every item in it | |
2407 | * from our directory | |
2408 | */ | |
d397712b | 2409 | while (1) { |
e02119d5 CM |
2410 | struct btrfs_key tmp; |
2411 | src = path->nodes[0]; | |
2412 | nritems = btrfs_header_nritems(src); | |
2413 | for (i = path->slots[0]; i < nritems; i++) { | |
2414 | btrfs_item_key_to_cpu(src, &min_key, i); | |
2415 | ||
2416 | if (min_key.objectid != inode->i_ino || | |
2417 | min_key.type != key_type) | |
2418 | goto done; | |
2419 | ret = overwrite_item(trans, log, dst_path, src, i, | |
2420 | &min_key); | |
2421 | BUG_ON(ret); | |
2422 | } | |
2423 | path->slots[0] = nritems; | |
2424 | ||
2425 | /* | |
2426 | * look ahead to the next item and see if it is also | |
2427 | * from this directory and from this transaction | |
2428 | */ | |
2429 | ret = btrfs_next_leaf(root, path); | |
2430 | if (ret == 1) { | |
2431 | last_offset = (u64)-1; | |
2432 | goto done; | |
2433 | } | |
2434 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); | |
2435 | if (tmp.objectid != inode->i_ino || tmp.type != key_type) { | |
2436 | last_offset = (u64)-1; | |
2437 | goto done; | |
2438 | } | |
2439 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { | |
2440 | ret = overwrite_item(trans, log, dst_path, | |
2441 | path->nodes[0], path->slots[0], | |
2442 | &tmp); | |
2443 | ||
2444 | BUG_ON(ret); | |
2445 | last_offset = tmp.offset; | |
2446 | goto done; | |
2447 | } | |
2448 | } | |
2449 | done: | |
2450 | *last_offset_ret = last_offset; | |
2451 | btrfs_release_path(root, path); | |
2452 | btrfs_release_path(log, dst_path); | |
2453 | ||
2454 | /* insert the log range keys to indicate where the log is valid */ | |
2455 | ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino, | |
2456 | first_offset, last_offset); | |
2457 | BUG_ON(ret); | |
2458 | return 0; | |
2459 | } | |
2460 | ||
2461 | /* | |
2462 | * logging directories is very similar to logging inodes, We find all the items | |
2463 | * from the current transaction and write them to the log. | |
2464 | * | |
2465 | * The recovery code scans the directory in the subvolume, and if it finds a | |
2466 | * key in the range logged that is not present in the log tree, then it means | |
2467 | * that dir entry was unlinked during the transaction. | |
2468 | * | |
2469 | * In order for that scan to work, we must include one key smaller than | |
2470 | * the smallest logged by this transaction and one key larger than the largest | |
2471 | * key logged by this transaction. | |
2472 | */ | |
2473 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, | |
2474 | struct btrfs_root *root, struct inode *inode, | |
2475 | struct btrfs_path *path, | |
2476 | struct btrfs_path *dst_path) | |
2477 | { | |
2478 | u64 min_key; | |
2479 | u64 max_key; | |
2480 | int ret; | |
2481 | int key_type = BTRFS_DIR_ITEM_KEY; | |
2482 | ||
2483 | again: | |
2484 | min_key = 0; | |
2485 | max_key = 0; | |
d397712b | 2486 | while (1) { |
e02119d5 CM |
2487 | ret = log_dir_items(trans, root, inode, path, |
2488 | dst_path, key_type, min_key, | |
2489 | &max_key); | |
2490 | BUG_ON(ret); | |
2491 | if (max_key == (u64)-1) | |
2492 | break; | |
2493 | min_key = max_key + 1; | |
2494 | } | |
2495 | ||
2496 | if (key_type == BTRFS_DIR_ITEM_KEY) { | |
2497 | key_type = BTRFS_DIR_INDEX_KEY; | |
2498 | goto again; | |
2499 | } | |
2500 | return 0; | |
2501 | } | |
2502 | ||
2503 | /* | |
2504 | * a helper function to drop items from the log before we relog an | |
2505 | * inode. max_key_type indicates the highest item type to remove. | |
2506 | * This cannot be run for file data extents because it does not | |
2507 | * free the extents they point to. | |
2508 | */ | |
2509 | static int drop_objectid_items(struct btrfs_trans_handle *trans, | |
2510 | struct btrfs_root *log, | |
2511 | struct btrfs_path *path, | |
2512 | u64 objectid, int max_key_type) | |
2513 | { | |
2514 | int ret; | |
2515 | struct btrfs_key key; | |
2516 | struct btrfs_key found_key; | |
2517 | ||
2518 | key.objectid = objectid; | |
2519 | key.type = max_key_type; | |
2520 | key.offset = (u64)-1; | |
2521 | ||
d397712b | 2522 | while (1) { |
e02119d5 CM |
2523 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
2524 | ||
2525 | if (ret != 1) | |
2526 | break; | |
2527 | ||
2528 | if (path->slots[0] == 0) | |
2529 | break; | |
2530 | ||
2531 | path->slots[0]--; | |
2532 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2533 | path->slots[0]); | |
2534 | ||
2535 | if (found_key.objectid != objectid) | |
2536 | break; | |
2537 | ||
2538 | ret = btrfs_del_item(trans, log, path); | |
2539 | BUG_ON(ret); | |
2540 | btrfs_release_path(log, path); | |
2541 | } | |
2542 | btrfs_release_path(log, path); | |
2543 | return 0; | |
2544 | } | |
2545 | ||
31ff1cd2 CM |
2546 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
2547 | struct btrfs_root *log, | |
2548 | struct btrfs_path *dst_path, | |
2549 | struct extent_buffer *src, | |
2550 | int start_slot, int nr, int inode_only) | |
2551 | { | |
2552 | unsigned long src_offset; | |
2553 | unsigned long dst_offset; | |
2554 | struct btrfs_file_extent_item *extent; | |
2555 | struct btrfs_inode_item *inode_item; | |
2556 | int ret; | |
2557 | struct btrfs_key *ins_keys; | |
2558 | u32 *ins_sizes; | |
2559 | char *ins_data; | |
2560 | int i; | |
d20f7043 CM |
2561 | struct list_head ordered_sums; |
2562 | ||
2563 | INIT_LIST_HEAD(&ordered_sums); | |
31ff1cd2 CM |
2564 | |
2565 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + | |
2566 | nr * sizeof(u32), GFP_NOFS); | |
2567 | ins_sizes = (u32 *)ins_data; | |
2568 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); | |
2569 | ||
2570 | for (i = 0; i < nr; i++) { | |
2571 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); | |
2572 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); | |
2573 | } | |
2574 | ret = btrfs_insert_empty_items(trans, log, dst_path, | |
2575 | ins_keys, ins_sizes, nr); | |
2576 | BUG_ON(ret); | |
2577 | ||
5d4f98a2 | 2578 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
31ff1cd2 CM |
2579 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
2580 | dst_path->slots[0]); | |
2581 | ||
2582 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); | |
2583 | ||
2584 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, | |
2585 | src_offset, ins_sizes[i]); | |
2586 | ||
2587 | if (inode_only == LOG_INODE_EXISTS && | |
2588 | ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { | |
2589 | inode_item = btrfs_item_ptr(dst_path->nodes[0], | |
2590 | dst_path->slots[0], | |
2591 | struct btrfs_inode_item); | |
2592 | btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0); | |
2593 | ||
2594 | /* set the generation to zero so the recover code | |
2595 | * can tell the difference between an logging | |
2596 | * just to say 'this inode exists' and a logging | |
2597 | * to say 'update this inode with these values' | |
2598 | */ | |
2599 | btrfs_set_inode_generation(dst_path->nodes[0], | |
2600 | inode_item, 0); | |
2601 | } | |
2602 | /* take a reference on file data extents so that truncates | |
2603 | * or deletes of this inode don't have to relog the inode | |
2604 | * again | |
2605 | */ | |
2606 | if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) { | |
2607 | int found_type; | |
2608 | extent = btrfs_item_ptr(src, start_slot + i, | |
2609 | struct btrfs_file_extent_item); | |
2610 | ||
2611 | found_type = btrfs_file_extent_type(src, extent); | |
d899e052 YZ |
2612 | if (found_type == BTRFS_FILE_EXTENT_REG || |
2613 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5d4f98a2 YZ |
2614 | u64 ds, dl, cs, cl; |
2615 | ds = btrfs_file_extent_disk_bytenr(src, | |
2616 | extent); | |
2617 | /* ds == 0 is a hole */ | |
2618 | if (ds == 0) | |
2619 | continue; | |
2620 | ||
2621 | dl = btrfs_file_extent_disk_num_bytes(src, | |
2622 | extent); | |
2623 | cs = btrfs_file_extent_offset(src, extent); | |
2624 | cl = btrfs_file_extent_num_bytes(src, | |
2625 | extent);; | |
580afd76 CM |
2626 | if (btrfs_file_extent_compression(src, |
2627 | extent)) { | |
2628 | cs = 0; | |
2629 | cl = dl; | |
2630 | } | |
5d4f98a2 YZ |
2631 | |
2632 | ret = btrfs_lookup_csums_range( | |
2633 | log->fs_info->csum_root, | |
2634 | ds + cs, ds + cs + cl - 1, | |
2635 | &ordered_sums); | |
2636 | BUG_ON(ret); | |
31ff1cd2 CM |
2637 | } |
2638 | } | |
31ff1cd2 CM |
2639 | } |
2640 | ||
2641 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); | |
2642 | btrfs_release_path(log, dst_path); | |
2643 | kfree(ins_data); | |
d20f7043 CM |
2644 | |
2645 | /* | |
2646 | * we have to do this after the loop above to avoid changing the | |
2647 | * log tree while trying to change the log tree. | |
2648 | */ | |
d397712b | 2649 | while (!list_empty(&ordered_sums)) { |
d20f7043 CM |
2650 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
2651 | struct btrfs_ordered_sum, | |
2652 | list); | |
2653 | ret = btrfs_csum_file_blocks(trans, log, sums); | |
2654 | BUG_ON(ret); | |
2655 | list_del(&sums->list); | |
2656 | kfree(sums); | |
2657 | } | |
31ff1cd2 CM |
2658 | return 0; |
2659 | } | |
2660 | ||
e02119d5 CM |
2661 | /* log a single inode in the tree log. |
2662 | * At least one parent directory for this inode must exist in the tree | |
2663 | * or be logged already. | |
2664 | * | |
2665 | * Any items from this inode changed by the current transaction are copied | |
2666 | * to the log tree. An extra reference is taken on any extents in this | |
2667 | * file, allowing us to avoid a whole pile of corner cases around logging | |
2668 | * blocks that have been removed from the tree. | |
2669 | * | |
2670 | * See LOG_INODE_ALL and related defines for a description of what inode_only | |
2671 | * does. | |
2672 | * | |
2673 | * This handles both files and directories. | |
2674 | */ | |
12fcfd22 | 2675 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
e02119d5 CM |
2676 | struct btrfs_root *root, struct inode *inode, |
2677 | int inode_only) | |
2678 | { | |
2679 | struct btrfs_path *path; | |
2680 | struct btrfs_path *dst_path; | |
2681 | struct btrfs_key min_key; | |
2682 | struct btrfs_key max_key; | |
2683 | struct btrfs_root *log = root->log_root; | |
31ff1cd2 | 2684 | struct extent_buffer *src = NULL; |
e02119d5 CM |
2685 | u32 size; |
2686 | int ret; | |
3a5f1d45 | 2687 | int nritems; |
31ff1cd2 CM |
2688 | int ins_start_slot = 0; |
2689 | int ins_nr; | |
e02119d5 CM |
2690 | |
2691 | log = root->log_root; | |
2692 | ||
2693 | path = btrfs_alloc_path(); | |
2694 | dst_path = btrfs_alloc_path(); | |
2695 | ||
2696 | min_key.objectid = inode->i_ino; | |
2697 | min_key.type = BTRFS_INODE_ITEM_KEY; | |
2698 | min_key.offset = 0; | |
2699 | ||
2700 | max_key.objectid = inode->i_ino; | |
12fcfd22 CM |
2701 | |
2702 | /* today the code can only do partial logging of directories */ | |
2703 | if (!S_ISDIR(inode->i_mode)) | |
2704 | inode_only = LOG_INODE_ALL; | |
2705 | ||
e02119d5 CM |
2706 | if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode)) |
2707 | max_key.type = BTRFS_XATTR_ITEM_KEY; | |
2708 | else | |
2709 | max_key.type = (u8)-1; | |
2710 | max_key.offset = (u64)-1; | |
2711 | ||
e02119d5 CM |
2712 | mutex_lock(&BTRFS_I(inode)->log_mutex); |
2713 | ||
2714 | /* | |
2715 | * a brute force approach to making sure we get the most uptodate | |
2716 | * copies of everything. | |
2717 | */ | |
2718 | if (S_ISDIR(inode->i_mode)) { | |
2719 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; | |
2720 | ||
2721 | if (inode_only == LOG_INODE_EXISTS) | |
2722 | max_key_type = BTRFS_XATTR_ITEM_KEY; | |
2723 | ret = drop_objectid_items(trans, log, path, | |
2724 | inode->i_ino, max_key_type); | |
2725 | } else { | |
2726 | ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0); | |
2727 | } | |
2728 | BUG_ON(ret); | |
2729 | path->keep_locks = 1; | |
2730 | ||
d397712b | 2731 | while (1) { |
31ff1cd2 | 2732 | ins_nr = 0; |
e02119d5 CM |
2733 | ret = btrfs_search_forward(root, &min_key, &max_key, |
2734 | path, 0, trans->transid); | |
2735 | if (ret != 0) | |
2736 | break; | |
3a5f1d45 | 2737 | again: |
31ff1cd2 | 2738 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ |
e02119d5 CM |
2739 | if (min_key.objectid != inode->i_ino) |
2740 | break; | |
2741 | if (min_key.type > max_key.type) | |
2742 | break; | |
31ff1cd2 | 2743 | |
e02119d5 CM |
2744 | src = path->nodes[0]; |
2745 | size = btrfs_item_size_nr(src, path->slots[0]); | |
31ff1cd2 CM |
2746 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { |
2747 | ins_nr++; | |
2748 | goto next_slot; | |
2749 | } else if (!ins_nr) { | |
2750 | ins_start_slot = path->slots[0]; | |
2751 | ins_nr = 1; | |
2752 | goto next_slot; | |
e02119d5 CM |
2753 | } |
2754 | ||
31ff1cd2 CM |
2755 | ret = copy_items(trans, log, dst_path, src, ins_start_slot, |
2756 | ins_nr, inode_only); | |
2757 | BUG_ON(ret); | |
2758 | ins_nr = 1; | |
2759 | ins_start_slot = path->slots[0]; | |
2760 | next_slot: | |
e02119d5 | 2761 | |
3a5f1d45 CM |
2762 | nritems = btrfs_header_nritems(path->nodes[0]); |
2763 | path->slots[0]++; | |
2764 | if (path->slots[0] < nritems) { | |
2765 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, | |
2766 | path->slots[0]); | |
2767 | goto again; | |
2768 | } | |
31ff1cd2 CM |
2769 | if (ins_nr) { |
2770 | ret = copy_items(trans, log, dst_path, src, | |
2771 | ins_start_slot, | |
2772 | ins_nr, inode_only); | |
2773 | BUG_ON(ret); | |
2774 | ins_nr = 0; | |
2775 | } | |
3a5f1d45 CM |
2776 | btrfs_release_path(root, path); |
2777 | ||
e02119d5 CM |
2778 | if (min_key.offset < (u64)-1) |
2779 | min_key.offset++; | |
2780 | else if (min_key.type < (u8)-1) | |
2781 | min_key.type++; | |
2782 | else if (min_key.objectid < (u64)-1) | |
2783 | min_key.objectid++; | |
2784 | else | |
2785 | break; | |
2786 | } | |
31ff1cd2 CM |
2787 | if (ins_nr) { |
2788 | ret = copy_items(trans, log, dst_path, src, | |
2789 | ins_start_slot, | |
2790 | ins_nr, inode_only); | |
2791 | BUG_ON(ret); | |
2792 | ins_nr = 0; | |
2793 | } | |
2794 | WARN_ON(ins_nr); | |
9623f9a3 | 2795 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) { |
e02119d5 CM |
2796 | btrfs_release_path(root, path); |
2797 | btrfs_release_path(log, dst_path); | |
2798 | ret = log_directory_changes(trans, root, inode, path, dst_path); | |
2799 | BUG_ON(ret); | |
2800 | } | |
3a5f1d45 | 2801 | BTRFS_I(inode)->logged_trans = trans->transid; |
e02119d5 CM |
2802 | mutex_unlock(&BTRFS_I(inode)->log_mutex); |
2803 | ||
2804 | btrfs_free_path(path); | |
2805 | btrfs_free_path(dst_path); | |
e02119d5 CM |
2806 | return 0; |
2807 | } | |
2808 | ||
12fcfd22 CM |
2809 | /* |
2810 | * follow the dentry parent pointers up the chain and see if any | |
2811 | * of the directories in it require a full commit before they can | |
2812 | * be logged. Returns zero if nothing special needs to be done or 1 if | |
2813 | * a full commit is required. | |
2814 | */ | |
2815 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, | |
2816 | struct inode *inode, | |
2817 | struct dentry *parent, | |
2818 | struct super_block *sb, | |
2819 | u64 last_committed) | |
e02119d5 | 2820 | { |
12fcfd22 CM |
2821 | int ret = 0; |
2822 | struct btrfs_root *root; | |
e02119d5 | 2823 | |
af4176b4 CM |
2824 | /* |
2825 | * for regular files, if its inode is already on disk, we don't | |
2826 | * have to worry about the parents at all. This is because | |
2827 | * we can use the last_unlink_trans field to record renames | |
2828 | * and other fun in this file. | |
2829 | */ | |
2830 | if (S_ISREG(inode->i_mode) && | |
2831 | BTRFS_I(inode)->generation <= last_committed && | |
2832 | BTRFS_I(inode)->last_unlink_trans <= last_committed) | |
2833 | goto out; | |
2834 | ||
12fcfd22 CM |
2835 | if (!S_ISDIR(inode->i_mode)) { |
2836 | if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) | |
2837 | goto out; | |
2838 | inode = parent->d_inode; | |
2839 | } | |
2840 | ||
2841 | while (1) { | |
2842 | BTRFS_I(inode)->logged_trans = trans->transid; | |
2843 | smp_mb(); | |
2844 | ||
2845 | if (BTRFS_I(inode)->last_unlink_trans > last_committed) { | |
2846 | root = BTRFS_I(inode)->root; | |
2847 | ||
2848 | /* | |
2849 | * make sure any commits to the log are forced | |
2850 | * to be full commits | |
2851 | */ | |
2852 | root->fs_info->last_trans_log_full_commit = | |
2853 | trans->transid; | |
2854 | ret = 1; | |
2855 | break; | |
2856 | } | |
2857 | ||
2858 | if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) | |
2859 | break; | |
2860 | ||
76dda93c | 2861 | if (IS_ROOT(parent)) |
12fcfd22 CM |
2862 | break; |
2863 | ||
2864 | parent = parent->d_parent; | |
2865 | inode = parent->d_inode; | |
2866 | ||
2867 | } | |
2868 | out: | |
e02119d5 CM |
2869 | return ret; |
2870 | } | |
2871 | ||
257c62e1 CM |
2872 | static int inode_in_log(struct btrfs_trans_handle *trans, |
2873 | struct inode *inode) | |
2874 | { | |
2875 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2876 | int ret = 0; | |
2877 | ||
2878 | mutex_lock(&root->log_mutex); | |
2879 | if (BTRFS_I(inode)->logged_trans == trans->transid && | |
2880 | BTRFS_I(inode)->last_sub_trans <= root->last_log_commit) | |
2881 | ret = 1; | |
2882 | mutex_unlock(&root->log_mutex); | |
2883 | return ret; | |
2884 | } | |
2885 | ||
2886 | ||
e02119d5 CM |
2887 | /* |
2888 | * helper function around btrfs_log_inode to make sure newly created | |
2889 | * parent directories also end up in the log. A minimal inode and backref | |
2890 | * only logging is done of any parent directories that are older than | |
2891 | * the last committed transaction | |
2892 | */ | |
12fcfd22 CM |
2893 | int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
2894 | struct btrfs_root *root, struct inode *inode, | |
2895 | struct dentry *parent, int exists_only) | |
e02119d5 | 2896 | { |
12fcfd22 | 2897 | int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL; |
e02119d5 | 2898 | struct super_block *sb; |
12fcfd22 CM |
2899 | int ret = 0; |
2900 | u64 last_committed = root->fs_info->last_trans_committed; | |
2901 | ||
2902 | sb = inode->i_sb; | |
2903 | ||
3a5e1404 SW |
2904 | if (btrfs_test_opt(root, NOTREELOG)) { |
2905 | ret = 1; | |
2906 | goto end_no_trans; | |
2907 | } | |
2908 | ||
12fcfd22 CM |
2909 | if (root->fs_info->last_trans_log_full_commit > |
2910 | root->fs_info->last_trans_committed) { | |
2911 | ret = 1; | |
2912 | goto end_no_trans; | |
2913 | } | |
2914 | ||
76dda93c YZ |
2915 | if (root != BTRFS_I(inode)->root || |
2916 | btrfs_root_refs(&root->root_item) == 0) { | |
2917 | ret = 1; | |
2918 | goto end_no_trans; | |
2919 | } | |
2920 | ||
12fcfd22 CM |
2921 | ret = check_parent_dirs_for_sync(trans, inode, parent, |
2922 | sb, last_committed); | |
2923 | if (ret) | |
2924 | goto end_no_trans; | |
e02119d5 | 2925 | |
257c62e1 CM |
2926 | if (inode_in_log(trans, inode)) { |
2927 | ret = BTRFS_NO_LOG_SYNC; | |
2928 | goto end_no_trans; | |
2929 | } | |
2930 | ||
e02119d5 | 2931 | start_log_trans(trans, root); |
e02119d5 | 2932 | |
12fcfd22 CM |
2933 | ret = btrfs_log_inode(trans, root, inode, inode_only); |
2934 | BUG_ON(ret); | |
12fcfd22 | 2935 | |
af4176b4 CM |
2936 | /* |
2937 | * for regular files, if its inode is already on disk, we don't | |
2938 | * have to worry about the parents at all. This is because | |
2939 | * we can use the last_unlink_trans field to record renames | |
2940 | * and other fun in this file. | |
2941 | */ | |
2942 | if (S_ISREG(inode->i_mode) && | |
2943 | BTRFS_I(inode)->generation <= last_committed && | |
2944 | BTRFS_I(inode)->last_unlink_trans <= last_committed) | |
2945 | goto no_parent; | |
2946 | ||
2947 | inode_only = LOG_INODE_EXISTS; | |
12fcfd22 CM |
2948 | while (1) { |
2949 | if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) | |
e02119d5 CM |
2950 | break; |
2951 | ||
12fcfd22 | 2952 | inode = parent->d_inode; |
76dda93c YZ |
2953 | if (root != BTRFS_I(inode)->root) |
2954 | break; | |
2955 | ||
12fcfd22 CM |
2956 | if (BTRFS_I(inode)->generation > |
2957 | root->fs_info->last_trans_committed) { | |
2958 | ret = btrfs_log_inode(trans, root, inode, inode_only); | |
2959 | BUG_ON(ret); | |
2960 | } | |
76dda93c | 2961 | if (IS_ROOT(parent)) |
e02119d5 | 2962 | break; |
12fcfd22 CM |
2963 | |
2964 | parent = parent->d_parent; | |
e02119d5 | 2965 | } |
af4176b4 | 2966 | no_parent: |
12fcfd22 CM |
2967 | ret = 0; |
2968 | btrfs_end_log_trans(root); | |
2969 | end_no_trans: | |
2970 | return ret; | |
e02119d5 CM |
2971 | } |
2972 | ||
2973 | /* | |
2974 | * it is not safe to log dentry if the chunk root has added new | |
2975 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. | |
2976 | * If this returns 1, you must commit the transaction to safely get your | |
2977 | * data on disk. | |
2978 | */ | |
2979 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, | |
2980 | struct btrfs_root *root, struct dentry *dentry) | |
2981 | { | |
12fcfd22 CM |
2982 | return btrfs_log_inode_parent(trans, root, dentry->d_inode, |
2983 | dentry->d_parent, 0); | |
e02119d5 CM |
2984 | } |
2985 | ||
2986 | /* | |
2987 | * should be called during mount to recover any replay any log trees | |
2988 | * from the FS | |
2989 | */ | |
2990 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) | |
2991 | { | |
2992 | int ret; | |
2993 | struct btrfs_path *path; | |
2994 | struct btrfs_trans_handle *trans; | |
2995 | struct btrfs_key key; | |
2996 | struct btrfs_key found_key; | |
2997 | struct btrfs_key tmp_key; | |
2998 | struct btrfs_root *log; | |
2999 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; | |
3000 | struct walk_control wc = { | |
3001 | .process_func = process_one_buffer, | |
3002 | .stage = 0, | |
3003 | }; | |
3004 | ||
3005 | fs_info->log_root_recovering = 1; | |
3006 | path = btrfs_alloc_path(); | |
3007 | BUG_ON(!path); | |
3008 | ||
3009 | trans = btrfs_start_transaction(fs_info->tree_root, 1); | |
3010 | ||
3011 | wc.trans = trans; | |
3012 | wc.pin = 1; | |
3013 | ||
3014 | walk_log_tree(trans, log_root_tree, &wc); | |
3015 | ||
3016 | again: | |
3017 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
3018 | key.offset = (u64)-1; | |
3019 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); | |
3020 | ||
d397712b | 3021 | while (1) { |
e02119d5 CM |
3022 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
3023 | if (ret < 0) | |
3024 | break; | |
3025 | if (ret > 0) { | |
3026 | if (path->slots[0] == 0) | |
3027 | break; | |
3028 | path->slots[0]--; | |
3029 | } | |
3030 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
3031 | path->slots[0]); | |
3032 | btrfs_release_path(log_root_tree, path); | |
3033 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
3034 | break; | |
3035 | ||
3036 | log = btrfs_read_fs_root_no_radix(log_root_tree, | |
3037 | &found_key); | |
3038 | BUG_ON(!log); | |
3039 | ||
3040 | ||
3041 | tmp_key.objectid = found_key.offset; | |
3042 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; | |
3043 | tmp_key.offset = (u64)-1; | |
3044 | ||
3045 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); | |
e02119d5 CM |
3046 | BUG_ON(!wc.replay_dest); |
3047 | ||
07d400a6 | 3048 | wc.replay_dest->log_root = log; |
5d4f98a2 | 3049 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
e02119d5 CM |
3050 | ret = walk_log_tree(trans, log, &wc); |
3051 | BUG_ON(ret); | |
3052 | ||
3053 | if (wc.stage == LOG_WALK_REPLAY_ALL) { | |
3054 | ret = fixup_inode_link_counts(trans, wc.replay_dest, | |
3055 | path); | |
3056 | BUG_ON(ret); | |
3057 | } | |
3058 | ||
3059 | key.offset = found_key.offset - 1; | |
07d400a6 | 3060 | wc.replay_dest->log_root = NULL; |
e02119d5 | 3061 | free_extent_buffer(log->node); |
b263c2c8 | 3062 | free_extent_buffer(log->commit_root); |
e02119d5 CM |
3063 | kfree(log); |
3064 | ||
3065 | if (found_key.offset == 0) | |
3066 | break; | |
3067 | } | |
3068 | btrfs_release_path(log_root_tree, path); | |
3069 | ||
3070 | /* step one is to pin it all, step two is to replay just inodes */ | |
3071 | if (wc.pin) { | |
3072 | wc.pin = 0; | |
3073 | wc.process_func = replay_one_buffer; | |
3074 | wc.stage = LOG_WALK_REPLAY_INODES; | |
3075 | goto again; | |
3076 | } | |
3077 | /* step three is to replay everything */ | |
3078 | if (wc.stage < LOG_WALK_REPLAY_ALL) { | |
3079 | wc.stage++; | |
3080 | goto again; | |
3081 | } | |
3082 | ||
3083 | btrfs_free_path(path); | |
3084 | ||
3085 | free_extent_buffer(log_root_tree->node); | |
3086 | log_root_tree->log_root = NULL; | |
3087 | fs_info->log_root_recovering = 0; | |
3088 | ||
3089 | /* step 4: commit the transaction, which also unpins the blocks */ | |
3090 | btrfs_commit_transaction(trans, fs_info->tree_root); | |
3091 | ||
3092 | kfree(log_root_tree); | |
3093 | return 0; | |
3094 | } | |
12fcfd22 CM |
3095 | |
3096 | /* | |
3097 | * there are some corner cases where we want to force a full | |
3098 | * commit instead of allowing a directory to be logged. | |
3099 | * | |
3100 | * They revolve around files there were unlinked from the directory, and | |
3101 | * this function updates the parent directory so that a full commit is | |
3102 | * properly done if it is fsync'd later after the unlinks are done. | |
3103 | */ | |
3104 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, | |
3105 | struct inode *dir, struct inode *inode, | |
3106 | int for_rename) | |
3107 | { | |
af4176b4 CM |
3108 | /* |
3109 | * when we're logging a file, if it hasn't been renamed | |
3110 | * or unlinked, and its inode is fully committed on disk, | |
3111 | * we don't have to worry about walking up the directory chain | |
3112 | * to log its parents. | |
3113 | * | |
3114 | * So, we use the last_unlink_trans field to put this transid | |
3115 | * into the file. When the file is logged we check it and | |
3116 | * don't log the parents if the file is fully on disk. | |
3117 | */ | |
3118 | if (S_ISREG(inode->i_mode)) | |
3119 | BTRFS_I(inode)->last_unlink_trans = trans->transid; | |
3120 | ||
12fcfd22 CM |
3121 | /* |
3122 | * if this directory was already logged any new | |
3123 | * names for this file/dir will get recorded | |
3124 | */ | |
3125 | smp_mb(); | |
3126 | if (BTRFS_I(dir)->logged_trans == trans->transid) | |
3127 | return; | |
3128 | ||
3129 | /* | |
3130 | * if the inode we're about to unlink was logged, | |
3131 | * the log will be properly updated for any new names | |
3132 | */ | |
3133 | if (BTRFS_I(inode)->logged_trans == trans->transid) | |
3134 | return; | |
3135 | ||
3136 | /* | |
3137 | * when renaming files across directories, if the directory | |
3138 | * there we're unlinking from gets fsync'd later on, there's | |
3139 | * no way to find the destination directory later and fsync it | |
3140 | * properly. So, we have to be conservative and force commits | |
3141 | * so the new name gets discovered. | |
3142 | */ | |
3143 | if (for_rename) | |
3144 | goto record; | |
3145 | ||
3146 | /* we can safely do the unlink without any special recording */ | |
3147 | return; | |
3148 | ||
3149 | record: | |
3150 | BTRFS_I(dir)->last_unlink_trans = trans->transid; | |
3151 | } | |
3152 | ||
3153 | /* | |
3154 | * Call this after adding a new name for a file and it will properly | |
3155 | * update the log to reflect the new name. | |
3156 | * | |
3157 | * It will return zero if all goes well, and it will return 1 if a | |
3158 | * full transaction commit is required. | |
3159 | */ | |
3160 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, | |
3161 | struct inode *inode, struct inode *old_dir, | |
3162 | struct dentry *parent) | |
3163 | { | |
3164 | struct btrfs_root * root = BTRFS_I(inode)->root; | |
3165 | ||
af4176b4 CM |
3166 | /* |
3167 | * this will force the logging code to walk the dentry chain | |
3168 | * up for the file | |
3169 | */ | |
3170 | if (S_ISREG(inode->i_mode)) | |
3171 | BTRFS_I(inode)->last_unlink_trans = trans->transid; | |
3172 | ||
12fcfd22 CM |
3173 | /* |
3174 | * if this inode hasn't been logged and directory we're renaming it | |
3175 | * from hasn't been logged, we don't need to log it | |
3176 | */ | |
3177 | if (BTRFS_I(inode)->logged_trans <= | |
3178 | root->fs_info->last_trans_committed && | |
3179 | (!old_dir || BTRFS_I(old_dir)->logged_trans <= | |
3180 | root->fs_info->last_trans_committed)) | |
3181 | return 0; | |
3182 | ||
3183 | return btrfs_log_inode_parent(trans, root, inode, parent, 1); | |
3184 | } | |
3185 |