1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2008 Oracle. All rights reserved.
4 */
5
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/blkdev.h>
9 #include <linux/list_sort.h>
10 #include <linux/iversion.h>
11 #include "misc.h"
12 #include "ctree.h"
13 #include "tree-log.h"
14 #include "disk-io.h"
15 #include "locking.h"
16 #include "print-tree.h"
17 #include "backref.h"
18 #include "compression.h"
19 #include "qgroup.h"
20 #include "inode-map.h"
21 #include "block-group.h"
22 #include "space-info.h"
23
24 /* magic values for the inode_only field in btrfs_log_inode:
25 *
26 * LOG_INODE_ALL means to log everything
27 * LOG_INODE_EXISTS means to log just enough to recreate the inode
28 * during log replay
29 */
30 enum {
31 LOG_INODE_ALL,
32 LOG_INODE_EXISTS,
33 LOG_OTHER_INODE,
34 LOG_OTHER_INODE_ALL,
35 };
36
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
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 enum {
90 LOG_WALK_PIN_ONLY,
91 LOG_WALK_REPLAY_INODES,
92 LOG_WALK_REPLAY_DIR_INDEX,
93 LOG_WALK_REPLAY_ALL,
94 };
95
96 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root, struct btrfs_inode *inode,
98 int inode_only,
99 struct btrfs_log_ctx *ctx);
100 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root,
102 struct btrfs_path *path, u64 objectid);
103 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
104 struct btrfs_root *root,
105 struct btrfs_root *log,
106 struct btrfs_path *path,
107 u64 dirid, int del_all);
108
109 /*
110 * tree logging is a special write ahead log used to make sure that
111 * fsyncs and O_SYNCs can happen without doing full tree commits.
112 *
113 * Full tree commits are expensive because they require commonly
114 * modified blocks to be recowed, creating many dirty pages in the
115 * extent tree an 4x-6x higher write load than ext3.
116 *
117 * Instead of doing a tree commit on every fsync, we use the
118 * key ranges and transaction ids to find items for a given file or directory
119 * that have changed in this transaction. Those items are copied into
120 * a special tree (one per subvolume root), that tree is written to disk
121 * and then the fsync is considered complete.
122 *
123 * After a crash, items are copied out of the log-tree back into the
124 * subvolume tree. Any file data extents found are recorded in the extent
125 * allocation tree, and the log-tree freed.
126 *
127 * The log tree is read three times, once to pin down all the extents it is
128 * using in ram and once, once to create all the inodes logged in the tree
129 * and once to do all the other items.
130 */
131
132 /*
133 * start a sub transaction and setup the log tree
134 * this increments the log tree writer count to make the people
135 * syncing the tree wait for us to finish
136 */
start_log_trans(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_log_ctx * ctx)137 static int start_log_trans(struct btrfs_trans_handle *trans,
138 struct btrfs_root *root,
139 struct btrfs_log_ctx *ctx)
140 {
141 struct btrfs_fs_info *fs_info = root->fs_info;
142 int ret = 0;
143
144 mutex_lock(&root->log_mutex);
145
146 if (root->log_root) {
147 if (btrfs_need_log_full_commit(trans)) {
148 ret = -EAGAIN;
149 goto out;
150 }
151
152 if (!root->log_start_pid) {
153 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
154 root->log_start_pid = current->pid;
155 } else if (root->log_start_pid != current->pid) {
156 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
157 }
158 } else {
159 mutex_lock(&fs_info->tree_log_mutex);
160 if (!fs_info->log_root_tree)
161 ret = btrfs_init_log_root_tree(trans, fs_info);
162 mutex_unlock(&fs_info->tree_log_mutex);
163 if (ret)
164 goto out;
165
166 ret = btrfs_add_log_tree(trans, root);
167 if (ret)
168 goto out;
169
170 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
171 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
172 root->log_start_pid = current->pid;
173 }
174
175 atomic_inc(&root->log_batch);
176 atomic_inc(&root->log_writers);
177 if (ctx && !ctx->logging_new_name) {
178 int index = root->log_transid % 2;
179 list_add_tail(&ctx->list, &root->log_ctxs[index]);
180 ctx->log_transid = root->log_transid;
181 }
182
183 out:
184 mutex_unlock(&root->log_mutex);
185 return ret;
186 }
187
188 /*
189 * returns 0 if there was a log transaction running and we were able
190 * to join, or returns -ENOENT if there were not transactions
191 * in progress
192 */
join_running_log_trans(struct btrfs_root * root)193 static int join_running_log_trans(struct btrfs_root *root)
194 {
195 int ret = -ENOENT;
196
197 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state))
198 return ret;
199
200 mutex_lock(&root->log_mutex);
201 if (root->log_root) {
202 ret = 0;
203 atomic_inc(&root->log_writers);
204 }
205 mutex_unlock(&root->log_mutex);
206 return ret;
207 }
208
209 /*
210 * This either makes the current running log transaction wait
211 * until you call btrfs_end_log_trans() or it makes any future
212 * log transactions wait until you call btrfs_end_log_trans()
213 */
btrfs_pin_log_trans(struct btrfs_root * root)214 void btrfs_pin_log_trans(struct btrfs_root *root)
215 {
216 atomic_inc(&root->log_writers);
217 }
218
219 /*
220 * indicate we're done making changes to the log tree
221 * and wake up anyone waiting to do a sync
222 */
btrfs_end_log_trans(struct btrfs_root * root)223 void btrfs_end_log_trans(struct btrfs_root *root)
224 {
225 if (atomic_dec_and_test(&root->log_writers)) {
226 /* atomic_dec_and_test implies a barrier */
227 cond_wake_up_nomb(&root->log_writer_wait);
228 }
229 }
230
btrfs_write_tree_block(struct extent_buffer * buf)231 static int btrfs_write_tree_block(struct extent_buffer *buf)
232 {
233 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
234 buf->start + buf->len - 1);
235 }
236
btrfs_wait_tree_block_writeback(struct extent_buffer * buf)237 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
238 {
239 filemap_fdatawait_range(buf->pages[0]->mapping,
240 buf->start, buf->start + buf->len - 1);
241 }
242
243 /*
244 * the walk control struct is used to pass state down the chain when
245 * processing the log tree. The stage field tells us which part
246 * of the log tree processing we are currently doing. The others
247 * are state fields used for that specific part
248 */
249 struct walk_control {
250 /* should we free the extent on disk when done? This is used
251 * at transaction commit time while freeing a log tree
252 */
253 int free;
254
255 /* should we write out the extent buffer? This is used
256 * while flushing the log tree to disk during a sync
257 */
258 int write;
259
260 /* should we wait for the extent buffer io to finish? Also used
261 * while flushing the log tree to disk for a sync
262 */
263 int wait;
264
265 /* pin only walk, we record which extents on disk belong to the
266 * log trees
267 */
268 int pin;
269
270 /* what stage of the replay code we're currently in */
271 int stage;
272
273 /*
274 * Ignore any items from the inode currently being processed. Needs
275 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
276 * the LOG_WALK_REPLAY_INODES stage.
277 */
278 bool ignore_cur_inode;
279
280 /* the root we are currently replaying */
281 struct btrfs_root *replay_dest;
282
283 /* the trans handle for the current replay */
284 struct btrfs_trans_handle *trans;
285
286 /* the function that gets used to process blocks we find in the
287 * tree. Note the extent_buffer might not be up to date when it is
288 * passed in, and it must be checked or read if you need the data
289 * inside it
290 */
291 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
292 struct walk_control *wc, u64 gen, int level);
293 };
294
295 /*
296 * process_func used to pin down extents, write them or wait on them
297 */
process_one_buffer(struct btrfs_root * log,struct extent_buffer * eb,struct walk_control * wc,u64 gen,int level)298 static int process_one_buffer(struct btrfs_root *log,
299 struct extent_buffer *eb,
300 struct walk_control *wc, u64 gen, int level)
301 {
302 struct btrfs_fs_info *fs_info = log->fs_info;
303 int ret = 0;
304
305 /*
306 * If this fs is mixed then we need to be able to process the leaves to
307 * pin down any logged extents, so we have to read the block.
308 */
309 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
310 ret = btrfs_read_buffer(eb, gen, level, NULL);
311 if (ret)
312 return ret;
313 }
314
315 if (wc->pin)
316 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start,
317 eb->len);
318
319 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
320 if (wc->pin && btrfs_header_level(eb) == 0)
321 ret = btrfs_exclude_logged_extents(eb);
322 if (wc->write)
323 btrfs_write_tree_block(eb);
324 if (wc->wait)
325 btrfs_wait_tree_block_writeback(eb);
326 }
327 return ret;
328 }
329
330 /*
331 * Item overwrite used by replay and tree logging. eb, slot and key all refer
332 * to the src data we are copying out.
333 *
334 * root is the tree we are copying into, and path is a scratch
335 * path for use in this function (it should be released on entry and
336 * will be released on exit).
337 *
338 * If the key is already in the destination tree the existing item is
339 * overwritten. If the existing item isn't big enough, it is extended.
340 * If it is too large, it is truncated.
341 *
342 * If the key isn't in the destination yet, a new item is inserted.
343 */
overwrite_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct extent_buffer * eb,int slot,struct btrfs_key * key)344 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
345 struct btrfs_root *root,
346 struct btrfs_path *path,
347 struct extent_buffer *eb, int slot,
348 struct btrfs_key *key)
349 {
350 int ret;
351 u32 item_size;
352 u64 saved_i_size = 0;
353 int save_old_i_size = 0;
354 unsigned long src_ptr;
355 unsigned long dst_ptr;
356 int overwrite_root = 0;
357 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
358
359 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
360 overwrite_root = 1;
361
362 item_size = btrfs_item_size_nr(eb, slot);
363 src_ptr = btrfs_item_ptr_offset(eb, slot);
364
365 /* look for the key in the destination tree */
366 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
367 if (ret < 0)
368 return ret;
369
370 if (ret == 0) {
371 char *src_copy;
372 char *dst_copy;
373 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
374 path->slots[0]);
375 if (dst_size != item_size)
376 goto insert;
377
378 if (item_size == 0) {
379 btrfs_release_path(path);
380 return 0;
381 }
382 dst_copy = kmalloc(item_size, GFP_NOFS);
383 src_copy = kmalloc(item_size, GFP_NOFS);
384 if (!dst_copy || !src_copy) {
385 btrfs_release_path(path);
386 kfree(dst_copy);
387 kfree(src_copy);
388 return -ENOMEM;
389 }
390
391 read_extent_buffer(eb, src_copy, src_ptr, item_size);
392
393 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
394 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
395 item_size);
396 ret = memcmp(dst_copy, src_copy, item_size);
397
398 kfree(dst_copy);
399 kfree(src_copy);
400 /*
401 * they have the same contents, just return, this saves
402 * us from cowing blocks in the destination tree and doing
403 * extra writes that may not have been done by a previous
404 * sync
405 */
406 if (ret == 0) {
407 btrfs_release_path(path);
408 return 0;
409 }
410
411 /*
412 * We need to load the old nbytes into the inode so when we
413 * replay the extents we've logged we get the right nbytes.
414 */
415 if (inode_item) {
416 struct btrfs_inode_item *item;
417 u64 nbytes;
418 u32 mode;
419
420 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
421 struct btrfs_inode_item);
422 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
423 item = btrfs_item_ptr(eb, slot,
424 struct btrfs_inode_item);
425 btrfs_set_inode_nbytes(eb, item, nbytes);
426
427 /*
428 * If this is a directory we need to reset the i_size to
429 * 0 so that we can set it up properly when replaying
430 * the rest of the items in this log.
431 */
432 mode = btrfs_inode_mode(eb, item);
433 if (S_ISDIR(mode))
434 btrfs_set_inode_size(eb, item, 0);
435 }
436 } else if (inode_item) {
437 struct btrfs_inode_item *item;
438 u32 mode;
439
440 /*
441 * New inode, set nbytes to 0 so that the nbytes comes out
442 * properly when we replay the extents.
443 */
444 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
445 btrfs_set_inode_nbytes(eb, item, 0);
446
447 /*
448 * If this is a directory we need to reset the i_size to 0 so
449 * that we can set it up properly when replaying the rest of
450 * the items in this log.
451 */
452 mode = btrfs_inode_mode(eb, item);
453 if (S_ISDIR(mode))
454 btrfs_set_inode_size(eb, item, 0);
455 }
456 insert:
457 btrfs_release_path(path);
458 /* try to insert the key into the destination tree */
459 path->skip_release_on_error = 1;
460 ret = btrfs_insert_empty_item(trans, root, path,
461 key, item_size);
462 path->skip_release_on_error = 0;
463
464 /* make sure any existing item is the correct size */
465 if (ret == -EEXIST || ret == -EOVERFLOW) {
466 u32 found_size;
467 found_size = btrfs_item_size_nr(path->nodes[0],
468 path->slots[0]);
469 if (found_size > item_size)
470 btrfs_truncate_item(path, item_size, 1);
471 else if (found_size < item_size)
472 btrfs_extend_item(path, item_size - found_size);
473 } else if (ret) {
474 return ret;
475 }
476 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
477 path->slots[0]);
478
479 /* don't overwrite an existing inode if the generation number
480 * was logged as zero. This is done when the tree logging code
481 * is just logging an inode to make sure it exists after recovery.
482 *
483 * Also, don't overwrite i_size on directories during replay.
484 * log replay inserts and removes directory items based on the
485 * state of the tree found in the subvolume, and i_size is modified
486 * as it goes
487 */
488 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
489 struct btrfs_inode_item *src_item;
490 struct btrfs_inode_item *dst_item;
491
492 src_item = (struct btrfs_inode_item *)src_ptr;
493 dst_item = (struct btrfs_inode_item *)dst_ptr;
494
495 if (btrfs_inode_generation(eb, src_item) == 0) {
496 struct extent_buffer *dst_eb = path->nodes[0];
497 const u64 ino_size = btrfs_inode_size(eb, src_item);
498
499 /*
500 * For regular files an ino_size == 0 is used only when
501 * logging that an inode exists, as part of a directory
502 * fsync, and the inode wasn't fsynced before. In this
503 * case don't set the size of the inode in the fs/subvol
504 * tree, otherwise we would be throwing valid data away.
505 */
506 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
507 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
508 ino_size != 0)
509 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
510 goto no_copy;
511 }
512
513 if (overwrite_root &&
514 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
515 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
516 save_old_i_size = 1;
517 saved_i_size = btrfs_inode_size(path->nodes[0],
518 dst_item);
519 }
520 }
521
522 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
523 src_ptr, item_size);
524
525 if (save_old_i_size) {
526 struct btrfs_inode_item *dst_item;
527 dst_item = (struct btrfs_inode_item *)dst_ptr;
528 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
529 }
530
531 /* make sure the generation is filled in */
532 if (key->type == BTRFS_INODE_ITEM_KEY) {
533 struct btrfs_inode_item *dst_item;
534 dst_item = (struct btrfs_inode_item *)dst_ptr;
535 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
536 btrfs_set_inode_generation(path->nodes[0], dst_item,
537 trans->transid);
538 }
539 }
540 no_copy:
541 btrfs_mark_buffer_dirty(path->nodes[0]);
542 btrfs_release_path(path);
543 return 0;
544 }
545
546 /*
547 * simple helper to read an inode off the disk from a given root
548 * This can only be called for subvolume roots and not for the log
549 */
read_one_inode(struct btrfs_root * root,u64 objectid)550 static noinline struct inode *read_one_inode(struct btrfs_root *root,
551 u64 objectid)
552 {
553 struct inode *inode;
554
555 inode = btrfs_iget(root->fs_info->sb, objectid, root);
556 if (IS_ERR(inode))
557 inode = NULL;
558 return inode;
559 }
560
561 /* replays a single extent in 'eb' at 'slot' with 'key' into the
562 * subvolume 'root'. path is released on entry and should be released
563 * on exit.
564 *
565 * extents in the log tree have not been allocated out of the extent
566 * tree yet. So, this completes the allocation, taking a reference
567 * as required if the extent already exists or creating a new extent
568 * if it isn't in the extent allocation tree yet.
569 *
570 * The extent is inserted into the file, dropping any existing extents
571 * from the file that overlap the new one.
572 */
replay_one_extent(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct extent_buffer * eb,int slot,struct btrfs_key * key)573 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
574 struct btrfs_root *root,
575 struct btrfs_path *path,
576 struct extent_buffer *eb, int slot,
577 struct btrfs_key *key)
578 {
579 struct btrfs_fs_info *fs_info = root->fs_info;
580 int found_type;
581 u64 extent_end;
582 u64 start = key->offset;
583 u64 nbytes = 0;
584 struct btrfs_file_extent_item *item;
585 struct inode *inode = NULL;
586 unsigned long size;
587 int ret = 0;
588
589 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
590 found_type = btrfs_file_extent_type(eb, item);
591
592 if (found_type == BTRFS_FILE_EXTENT_REG ||
593 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
594 nbytes = btrfs_file_extent_num_bytes(eb, item);
595 extent_end = start + nbytes;
596
597 /*
598 * We don't add to the inodes nbytes if we are prealloc or a
599 * hole.
600 */
601 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
602 nbytes = 0;
603 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
604 size = btrfs_file_extent_ram_bytes(eb, item);
605 nbytes = btrfs_file_extent_ram_bytes(eb, item);
606 extent_end = ALIGN(start + size,
607 fs_info->sectorsize);
608 } else {
609 ret = 0;
610 goto out;
611 }
612
613 inode = read_one_inode(root, key->objectid);
614 if (!inode) {
615 ret = -EIO;
616 goto out;
617 }
618
619 /*
620 * first check to see if we already have this extent in the
621 * file. This must be done before the btrfs_drop_extents run
622 * so we don't try to drop this extent.
623 */
624 ret = btrfs_lookup_file_extent(trans, root, path,
625 btrfs_ino(BTRFS_I(inode)), start, 0);
626
627 if (ret == 0 &&
628 (found_type == BTRFS_FILE_EXTENT_REG ||
629 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
630 struct btrfs_file_extent_item cmp1;
631 struct btrfs_file_extent_item cmp2;
632 struct btrfs_file_extent_item *existing;
633 struct extent_buffer *leaf;
634
635 leaf = path->nodes[0];
636 existing = btrfs_item_ptr(leaf, path->slots[0],
637 struct btrfs_file_extent_item);
638
639 read_extent_buffer(eb, &cmp1, (unsigned long)item,
640 sizeof(cmp1));
641 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
642 sizeof(cmp2));
643
644 /*
645 * we already have a pointer to this exact extent,
646 * we don't have to do anything
647 */
648 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
649 btrfs_release_path(path);
650 goto out;
651 }
652 }
653 btrfs_release_path(path);
654
655 /* drop any overlapping extents */
656 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
657 if (ret)
658 goto out;
659
660 if (found_type == BTRFS_FILE_EXTENT_REG ||
661 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
662 u64 offset;
663 unsigned long dest_offset;
664 struct btrfs_key ins;
665
666 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
667 btrfs_fs_incompat(fs_info, NO_HOLES))
668 goto update_inode;
669
670 ret = btrfs_insert_empty_item(trans, root, path, key,
671 sizeof(*item));
672 if (ret)
673 goto out;
674 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
675 path->slots[0]);
676 copy_extent_buffer(path->nodes[0], eb, dest_offset,
677 (unsigned long)item, sizeof(*item));
678
679 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
680 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
681 ins.type = BTRFS_EXTENT_ITEM_KEY;
682 offset = key->offset - btrfs_file_extent_offset(eb, item);
683
684 /*
685 * Manually record dirty extent, as here we did a shallow
686 * file extent item copy and skip normal backref update,
687 * but modifying extent tree all by ourselves.
688 * So need to manually record dirty extent for qgroup,
689 * as the owner of the file extent changed from log tree
690 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
691 */
692 ret = btrfs_qgroup_trace_extent(trans,
693 btrfs_file_extent_disk_bytenr(eb, item),
694 btrfs_file_extent_disk_num_bytes(eb, item),
695 GFP_NOFS);
696 if (ret < 0)
697 goto out;
698
699 if (ins.objectid > 0) {
700 struct btrfs_ref ref = { 0 };
701 u64 csum_start;
702 u64 csum_end;
703 LIST_HEAD(ordered_sums);
704
705 /*
706 * is this extent already allocated in the extent
707 * allocation tree? If so, just add a reference
708 */
709 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
710 ins.offset);
711 if (ret < 0) {
712 goto out;
713 } else if (ret == 0) {
714 btrfs_init_generic_ref(&ref,
715 BTRFS_ADD_DELAYED_REF,
716 ins.objectid, ins.offset, 0);
717 btrfs_init_data_ref(&ref,
718 root->root_key.objectid,
719 key->objectid, offset);
720 ret = btrfs_inc_extent_ref(trans, &ref);
721 if (ret)
722 goto out;
723 } else {
724 /*
725 * insert the extent pointer in the extent
726 * allocation tree
727 */
728 ret = btrfs_alloc_logged_file_extent(trans,
729 root->root_key.objectid,
730 key->objectid, offset, &ins);
731 if (ret)
732 goto out;
733 }
734 btrfs_release_path(path);
735
736 if (btrfs_file_extent_compression(eb, item)) {
737 csum_start = ins.objectid;
738 csum_end = csum_start + ins.offset;
739 } else {
740 csum_start = ins.objectid +
741 btrfs_file_extent_offset(eb, item);
742 csum_end = csum_start +
743 btrfs_file_extent_num_bytes(eb, item);
744 }
745
746 ret = btrfs_lookup_csums_range(root->log_root,
747 csum_start, csum_end - 1,
748 &ordered_sums, 0);
749 if (ret)
750 goto out;
751 /*
752 * Now delete all existing cums in the csum root that
753 * cover our range. We do this because we can have an
754 * extent that is completely referenced by one file
755 * extent item and partially referenced by another
756 * file extent item (like after using the clone or
757 * extent_same ioctls). In this case if we end up doing
758 * the replay of the one that partially references the
759 * extent first, and we do not do the csum deletion
760 * below, we can get 2 csum items in the csum tree that
761 * overlap each other. For example, imagine our log has
762 * the two following file extent items:
763 *
764 * key (257 EXTENT_DATA 409600)
765 * extent data disk byte 12845056 nr 102400
766 * extent data offset 20480 nr 20480 ram 102400
767 *
768 * key (257 EXTENT_DATA 819200)
769 * extent data disk byte 12845056 nr 102400
770 * extent data offset 0 nr 102400 ram 102400
771 *
772 * Where the second one fully references the 100K extent
773 * that starts at disk byte 12845056, and the log tree
774 * has a single csum item that covers the entire range
775 * of the extent:
776 *
777 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
778 *
779 * After the first file extent item is replayed, the
780 * csum tree gets the following csum item:
781 *
782 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
783 *
784 * Which covers the 20K sub-range starting at offset 20K
785 * of our extent. Now when we replay the second file
786 * extent item, if we do not delete existing csum items
787 * that cover any of its blocks, we end up getting two
788 * csum items in our csum tree that overlap each other:
789 *
790 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
791 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
792 *
793 * Which is a problem, because after this anyone trying
794 * to lookup up for the checksum of any block of our
795 * extent starting at an offset of 40K or higher, will
796 * end up looking at the second csum item only, which
797 * does not contain the checksum for any block starting
798 * at offset 40K or higher of our extent.
799 */
800 while (!list_empty(&ordered_sums)) {
801 struct btrfs_ordered_sum *sums;
802 sums = list_entry(ordered_sums.next,
803 struct btrfs_ordered_sum,
804 list);
805 if (!ret)
806 ret = btrfs_del_csums(trans,
807 fs_info->csum_root,
808 sums->bytenr,
809 sums->len);
810 if (!ret)
811 ret = btrfs_csum_file_blocks(trans,
812 fs_info->csum_root, sums);
813 list_del(&sums->list);
814 kfree(sums);
815 }
816 if (ret)
817 goto out;
818 } else {
819 btrfs_release_path(path);
820 }
821 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
822 /* inline extents are easy, we just overwrite them */
823 ret = overwrite_item(trans, root, path, eb, slot, key);
824 if (ret)
825 goto out;
826 }
827
828 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
829 extent_end - start);
830 if (ret)
831 goto out;
832
833 inode_add_bytes(inode, nbytes);
834 update_inode:
835 ret = btrfs_update_inode(trans, root, inode);
836 out:
837 if (inode)
838 iput(inode);
839 return ret;
840 }
841
842 /*
843 * when cleaning up conflicts between the directory names in the
844 * subvolume, directory names in the log and directory names in the
845 * inode back references, we may have to unlink inodes from directories.
846 *
847 * This is a helper function to do the unlink of a specific directory
848 * item
849 */
drop_one_dir_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_inode * dir,struct btrfs_dir_item * di)850 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
851 struct btrfs_root *root,
852 struct btrfs_path *path,
853 struct btrfs_inode *dir,
854 struct btrfs_dir_item *di)
855 {
856 struct inode *inode;
857 char *name;
858 int name_len;
859 struct extent_buffer *leaf;
860 struct btrfs_key location;
861 int ret;
862
863 leaf = path->nodes[0];
864
865 btrfs_dir_item_key_to_cpu(leaf, di, &location);
866 name_len = btrfs_dir_name_len(leaf, di);
867 name = kmalloc(name_len, GFP_NOFS);
868 if (!name)
869 return -ENOMEM;
870
871 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
872 btrfs_release_path(path);
873
874 inode = read_one_inode(root, location.objectid);
875 if (!inode) {
876 ret = -EIO;
877 goto out;
878 }
879
880 ret = link_to_fixup_dir(trans, root, path, location.objectid);
881 if (ret)
882 goto out;
883
884 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
885 name_len);
886 if (ret)
887 goto out;
888 else
889 ret = btrfs_run_delayed_items(trans);
890 out:
891 kfree(name);
892 iput(inode);
893 return ret;
894 }
895
896 /*
897 * See if a given name and sequence number found in an inode back reference are
898 * already in a directory and correctly point to this inode.
899 *
900 * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it
901 * exists.
902 */
inode_in_dir(struct btrfs_root * root,struct btrfs_path * path,u64 dirid,u64 objectid,u64 index,const char * name,int name_len)903 static noinline int inode_in_dir(struct btrfs_root *root,
904 struct btrfs_path *path,
905 u64 dirid, u64 objectid, u64 index,
906 const char *name, int name_len)
907 {
908 struct btrfs_dir_item *di;
909 struct btrfs_key location;
910 int ret = 0;
911
912 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
913 index, name, name_len, 0);
914 if (IS_ERR(di)) {
915 if (PTR_ERR(di) != -ENOENT)
916 ret = PTR_ERR(di);
917 goto out;
918 } else if (di) {
919 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
920 if (location.objectid != objectid)
921 goto out;
922 } else {
923 goto out;
924 }
925
926 btrfs_release_path(path);
927 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
928 if (IS_ERR(di)) {
929 ret = PTR_ERR(di);
930 goto out;
931 } else if (di) {
932 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
933 if (location.objectid == objectid)
934 ret = 1;
935 }
936 out:
937 btrfs_release_path(path);
938 return ret;
939 }
940
941 /*
942 * helper function to check a log tree for a named back reference in
943 * an inode. This is used to decide if a back reference that is
944 * found in the subvolume conflicts with what we find in the log.
945 *
946 * inode backreferences may have multiple refs in a single item,
947 * during replay we process one reference at a time, and we don't
948 * want to delete valid links to a file from the subvolume if that
949 * link is also in the log.
950 */
backref_in_log(struct btrfs_root * log,struct btrfs_key * key,u64 ref_objectid,const char * name,int namelen)951 static noinline int backref_in_log(struct btrfs_root *log,
952 struct btrfs_key *key,
953 u64 ref_objectid,
954 const char *name, int namelen)
955 {
956 struct btrfs_path *path;
957 int ret;
958
959 path = btrfs_alloc_path();
960 if (!path)
961 return -ENOMEM;
962
963 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
964 if (ret < 0) {
965 goto out;
966 } else if (ret == 1) {
967 ret = 0;
968 goto out;
969 }
970
971 if (key->type == BTRFS_INODE_EXTREF_KEY)
972 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
973 path->slots[0],
974 ref_objectid,
975 name, namelen);
976 else
977 ret = !!btrfs_find_name_in_backref(path->nodes[0],
978 path->slots[0],
979 name, namelen);
980 out:
981 btrfs_free_path(path);
982 return ret;
983 }
984
__add_inode_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_root * log_root,struct btrfs_inode * dir,struct btrfs_inode * inode,u64 inode_objectid,u64 parent_objectid,u64 ref_index,char * name,int namelen,int * search_done)985 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
986 struct btrfs_root *root,
987 struct btrfs_path *path,
988 struct btrfs_root *log_root,
989 struct btrfs_inode *dir,
990 struct btrfs_inode *inode,
991 u64 inode_objectid, u64 parent_objectid,
992 u64 ref_index, char *name, int namelen,
993 int *search_done)
994 {
995 int ret;
996 char *victim_name;
997 int victim_name_len;
998 struct extent_buffer *leaf;
999 struct btrfs_dir_item *di;
1000 struct btrfs_key search_key;
1001 struct btrfs_inode_extref *extref;
1002
1003 again:
1004 /* Search old style refs */
1005 search_key.objectid = inode_objectid;
1006 search_key.type = BTRFS_INODE_REF_KEY;
1007 search_key.offset = parent_objectid;
1008 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1009 if (ret == 0) {
1010 struct btrfs_inode_ref *victim_ref;
1011 unsigned long ptr;
1012 unsigned long ptr_end;
1013
1014 leaf = path->nodes[0];
1015
1016 /* are we trying to overwrite a back ref for the root directory
1017 * if so, just jump out, we're done
1018 */
1019 if (search_key.objectid == search_key.offset)
1020 return 1;
1021
1022 /* check all the names in this back reference to see
1023 * if they are in the log. if so, we allow them to stay
1024 * otherwise they must be unlinked as a conflict
1025 */
1026 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1027 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1028 while (ptr < ptr_end) {
1029 victim_ref = (struct btrfs_inode_ref *)ptr;
1030 victim_name_len = btrfs_inode_ref_name_len(leaf,
1031 victim_ref);
1032 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1033 if (!victim_name)
1034 return -ENOMEM;
1035
1036 read_extent_buffer(leaf, victim_name,
1037 (unsigned long)(victim_ref + 1),
1038 victim_name_len);
1039
1040 ret = backref_in_log(log_root, &search_key,
1041 parent_objectid, victim_name,
1042 victim_name_len);
1043 if (ret < 0) {
1044 kfree(victim_name);
1045 return ret;
1046 } else if (!ret) {
1047 inc_nlink(&inode->vfs_inode);
1048 btrfs_release_path(path);
1049
1050 ret = btrfs_unlink_inode(trans, root, dir, inode,
1051 victim_name, victim_name_len);
1052 kfree(victim_name);
1053 if (ret)
1054 return ret;
1055 ret = btrfs_run_delayed_items(trans);
1056 if (ret)
1057 return ret;
1058 *search_done = 1;
1059 goto again;
1060 }
1061 kfree(victim_name);
1062
1063 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1064 }
1065
1066 /*
1067 * NOTE: we have searched root tree and checked the
1068 * corresponding ref, it does not need to check again.
1069 */
1070 *search_done = 1;
1071 }
1072 btrfs_release_path(path);
1073
1074 /* Same search but for extended refs */
1075 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1076 inode_objectid, parent_objectid, 0,
1077 0);
1078 if (IS_ERR(extref)) {
1079 return PTR_ERR(extref);
1080 } else if (extref) {
1081 u32 item_size;
1082 u32 cur_offset = 0;
1083 unsigned long base;
1084 struct inode *victim_parent;
1085
1086 leaf = path->nodes[0];
1087
1088 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1089 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1090
1091 while (cur_offset < item_size) {
1092 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1093
1094 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1095
1096 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1097 goto next;
1098
1099 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1100 if (!victim_name)
1101 return -ENOMEM;
1102 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1103 victim_name_len);
1104
1105 search_key.objectid = inode_objectid;
1106 search_key.type = BTRFS_INODE_EXTREF_KEY;
1107 search_key.offset = btrfs_extref_hash(parent_objectid,
1108 victim_name,
1109 victim_name_len);
1110 ret = backref_in_log(log_root, &search_key,
1111 parent_objectid, victim_name,
1112 victim_name_len);
1113 if (ret < 0) {
1114 kfree(victim_name);
1115 return ret;
1116 } else if (!ret) {
1117 ret = -ENOENT;
1118 victim_parent = read_one_inode(root,
1119 parent_objectid);
1120 if (victim_parent) {
1121 inc_nlink(&inode->vfs_inode);
1122 btrfs_release_path(path);
1123
1124 ret = btrfs_unlink_inode(trans, root,
1125 BTRFS_I(victim_parent),
1126 inode,
1127 victim_name,
1128 victim_name_len);
1129 if (!ret)
1130 ret = btrfs_run_delayed_items(
1131 trans);
1132 }
1133 iput(victim_parent);
1134 kfree(victim_name);
1135 if (ret)
1136 return ret;
1137 *search_done = 1;
1138 goto again;
1139 }
1140 kfree(victim_name);
1141 next:
1142 cur_offset += victim_name_len + sizeof(*extref);
1143 }
1144 *search_done = 1;
1145 }
1146 btrfs_release_path(path);
1147
1148 /* look for a conflicting sequence number */
1149 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1150 ref_index, name, namelen, 0);
1151 if (IS_ERR(di)) {
1152 if (PTR_ERR(di) != -ENOENT)
1153 return PTR_ERR(di);
1154 } else if (di) {
1155 ret = drop_one_dir_item(trans, root, path, dir, di);
1156 if (ret)
1157 return ret;
1158 }
1159 btrfs_release_path(path);
1160
1161 /* look for a conflicting name */
1162 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1163 name, namelen, 0);
1164 if (IS_ERR(di)) {
1165 return PTR_ERR(di);
1166 } else if (di) {
1167 ret = drop_one_dir_item(trans, root, path, dir, di);
1168 if (ret)
1169 return ret;
1170 }
1171 btrfs_release_path(path);
1172
1173 return 0;
1174 }
1175
extref_get_fields(struct extent_buffer * eb,unsigned long ref_ptr,u32 * namelen,char ** name,u64 * index,u64 * parent_objectid)1176 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1177 u32 *namelen, char **name, u64 *index,
1178 u64 *parent_objectid)
1179 {
1180 struct btrfs_inode_extref *extref;
1181
1182 extref = (struct btrfs_inode_extref *)ref_ptr;
1183
1184 *namelen = btrfs_inode_extref_name_len(eb, extref);
1185 *name = kmalloc(*namelen, GFP_NOFS);
1186 if (*name == NULL)
1187 return -ENOMEM;
1188
1189 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1190 *namelen);
1191
1192 if (index)
1193 *index = btrfs_inode_extref_index(eb, extref);
1194 if (parent_objectid)
1195 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1196
1197 return 0;
1198 }
1199
ref_get_fields(struct extent_buffer * eb,unsigned long ref_ptr,u32 * namelen,char ** name,u64 * index)1200 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1201 u32 *namelen, char **name, u64 *index)
1202 {
1203 struct btrfs_inode_ref *ref;
1204
1205 ref = (struct btrfs_inode_ref *)ref_ptr;
1206
1207 *namelen = btrfs_inode_ref_name_len(eb, ref);
1208 *name = kmalloc(*namelen, GFP_NOFS);
1209 if (*name == NULL)
1210 return -ENOMEM;
1211
1212 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1213
1214 if (index)
1215 *index = btrfs_inode_ref_index(eb, ref);
1216
1217 return 0;
1218 }
1219
1220 /*
1221 * Take an inode reference item from the log tree and iterate all names from the
1222 * inode reference item in the subvolume tree with the same key (if it exists).
1223 * For any name that is not in the inode reference item from the log tree, do a
1224 * proper unlink of that name (that is, remove its entry from the inode
1225 * reference item and both dir index keys).
1226 */
unlink_old_inode_refs(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_inode * inode,struct extent_buffer * log_eb,int log_slot,struct btrfs_key * key)1227 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1228 struct btrfs_root *root,
1229 struct btrfs_path *path,
1230 struct btrfs_inode *inode,
1231 struct extent_buffer *log_eb,
1232 int log_slot,
1233 struct btrfs_key *key)
1234 {
1235 int ret;
1236 unsigned long ref_ptr;
1237 unsigned long ref_end;
1238 struct extent_buffer *eb;
1239
1240 again:
1241 btrfs_release_path(path);
1242 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1243 if (ret > 0) {
1244 ret = 0;
1245 goto out;
1246 }
1247 if (ret < 0)
1248 goto out;
1249
1250 eb = path->nodes[0];
1251 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1252 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1253 while (ref_ptr < ref_end) {
1254 char *name = NULL;
1255 int namelen;
1256 u64 parent_id;
1257
1258 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1259 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1260 NULL, &parent_id);
1261 } else {
1262 parent_id = key->offset;
1263 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1264 NULL);
1265 }
1266 if (ret)
1267 goto out;
1268
1269 if (key->type == BTRFS_INODE_EXTREF_KEY)
1270 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1271 parent_id, name,
1272 namelen);
1273 else
1274 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1275 name, namelen);
1276
1277 if (!ret) {
1278 struct inode *dir;
1279
1280 btrfs_release_path(path);
1281 dir = read_one_inode(root, parent_id);
1282 if (!dir) {
1283 ret = -ENOENT;
1284 kfree(name);
1285 goto out;
1286 }
1287 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1288 inode, name, namelen);
1289 kfree(name);
1290 iput(dir);
1291 /*
1292 * Whenever we need to check if a name exists or not, we
1293 * check the subvolume tree. So after an unlink we must
1294 * run delayed items, so that future checks for a name
1295 * during log replay see that the name does not exists
1296 * anymore.
1297 */
1298 if (!ret)
1299 ret = btrfs_run_delayed_items(trans);
1300 if (ret)
1301 goto out;
1302 goto again;
1303 }
1304
1305 kfree(name);
1306 ref_ptr += namelen;
1307 if (key->type == BTRFS_INODE_EXTREF_KEY)
1308 ref_ptr += sizeof(struct btrfs_inode_extref);
1309 else
1310 ref_ptr += sizeof(struct btrfs_inode_ref);
1311 }
1312 ret = 0;
1313 out:
1314 btrfs_release_path(path);
1315 return ret;
1316 }
1317
btrfs_inode_ref_exists(struct inode * inode,struct inode * dir,const u8 ref_type,const char * name,const int namelen)1318 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1319 const u8 ref_type, const char *name,
1320 const int namelen)
1321 {
1322 struct btrfs_key key;
1323 struct btrfs_path *path;
1324 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1325 int ret;
1326
1327 path = btrfs_alloc_path();
1328 if (!path)
1329 return -ENOMEM;
1330
1331 key.objectid = btrfs_ino(BTRFS_I(inode));
1332 key.type = ref_type;
1333 if (key.type == BTRFS_INODE_REF_KEY)
1334 key.offset = parent_id;
1335 else
1336 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1337
1338 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1339 if (ret < 0)
1340 goto out;
1341 if (ret > 0) {
1342 ret = 0;
1343 goto out;
1344 }
1345 if (key.type == BTRFS_INODE_EXTREF_KEY)
1346 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1347 path->slots[0], parent_id, name, namelen);
1348 else
1349 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1350 name, namelen);
1351
1352 out:
1353 btrfs_free_path(path);
1354 return ret;
1355 }
1356
add_link(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct inode * dir,struct inode * inode,const char * name,int namelen,u64 ref_index)1357 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1358 struct inode *dir, struct inode *inode, const char *name,
1359 int namelen, u64 ref_index)
1360 {
1361 struct btrfs_dir_item *dir_item;
1362 struct btrfs_key key;
1363 struct btrfs_path *path;
1364 struct inode *other_inode = NULL;
1365 int ret;
1366
1367 path = btrfs_alloc_path();
1368 if (!path)
1369 return -ENOMEM;
1370
1371 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1372 btrfs_ino(BTRFS_I(dir)),
1373 name, namelen, 0);
1374 if (!dir_item) {
1375 btrfs_release_path(path);
1376 goto add_link;
1377 } else if (IS_ERR(dir_item)) {
1378 ret = PTR_ERR(dir_item);
1379 goto out;
1380 }
1381
1382 /*
1383 * Our inode's dentry collides with the dentry of another inode which is
1384 * in the log but not yet processed since it has a higher inode number.
1385 * So delete that other dentry.
1386 */
1387 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1388 btrfs_release_path(path);
1389 other_inode = read_one_inode(root, key.objectid);
1390 if (!other_inode) {
1391 ret = -ENOENT;
1392 goto out;
1393 }
1394 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1395 name, namelen);
1396 if (ret)
1397 goto out;
1398 /*
1399 * If we dropped the link count to 0, bump it so that later the iput()
1400 * on the inode will not free it. We will fixup the link count later.
1401 */
1402 if (other_inode->i_nlink == 0)
1403 inc_nlink(other_inode);
1404
1405 ret = btrfs_run_delayed_items(trans);
1406 if (ret)
1407 goto out;
1408 add_link:
1409 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1410 name, namelen, 0, ref_index);
1411 out:
1412 iput(other_inode);
1413 btrfs_free_path(path);
1414
1415 return ret;
1416 }
1417
1418 /*
1419 * replay one inode back reference item found in the log tree.
1420 * eb, slot and key refer to the buffer and key found in the log tree.
1421 * root is the destination we are replaying into, and path is for temp
1422 * use by this function. (it should be released on return).
1423 */
add_inode_ref(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_root * log,struct btrfs_path * path,struct extent_buffer * eb,int slot,struct btrfs_key * key)1424 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1425 struct btrfs_root *root,
1426 struct btrfs_root *log,
1427 struct btrfs_path *path,
1428 struct extent_buffer *eb, int slot,
1429 struct btrfs_key *key)
1430 {
1431 struct inode *dir = NULL;
1432 struct inode *inode = NULL;
1433 unsigned long ref_ptr;
1434 unsigned long ref_end;
1435 char *name = NULL;
1436 int namelen;
1437 int ret;
1438 int search_done = 0;
1439 int log_ref_ver = 0;
1440 u64 parent_objectid;
1441 u64 inode_objectid;
1442 u64 ref_index = 0;
1443 int ref_struct_size;
1444
1445 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1446 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1447
1448 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1449 struct btrfs_inode_extref *r;
1450
1451 ref_struct_size = sizeof(struct btrfs_inode_extref);
1452 log_ref_ver = 1;
1453 r = (struct btrfs_inode_extref *)ref_ptr;
1454 parent_objectid = btrfs_inode_extref_parent(eb, r);
1455 } else {
1456 ref_struct_size = sizeof(struct btrfs_inode_ref);
1457 parent_objectid = key->offset;
1458 }
1459 inode_objectid = key->objectid;
1460
1461 /*
1462 * it is possible that we didn't log all the parent directories
1463 * for a given inode. If we don't find the dir, just don't
1464 * copy the back ref in. The link count fixup code will take
1465 * care of the rest
1466 */
1467 dir = read_one_inode(root, parent_objectid);
1468 if (!dir) {
1469 ret = -ENOENT;
1470 goto out;
1471 }
1472
1473 inode = read_one_inode(root, inode_objectid);
1474 if (!inode) {
1475 ret = -EIO;
1476 goto out;
1477 }
1478
1479 while (ref_ptr < ref_end) {
1480 if (log_ref_ver) {
1481 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1482 &ref_index, &parent_objectid);
1483 /*
1484 * parent object can change from one array
1485 * item to another.
1486 */
1487 if (!dir)
1488 dir = read_one_inode(root, parent_objectid);
1489 if (!dir) {
1490 ret = -ENOENT;
1491 goto out;
1492 }
1493 } else {
1494 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1495 &ref_index);
1496 }
1497 if (ret)
1498 goto out;
1499
1500 ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1501 btrfs_ino(BTRFS_I(inode)), ref_index,
1502 name, namelen);
1503 if (ret < 0) {
1504 goto out;
1505 } else if (ret == 0) {
1506 /*
1507 * look for a conflicting back reference in the
1508 * metadata. if we find one we have to unlink that name
1509 * of the file before we add our new link. Later on, we
1510 * overwrite any existing back reference, and we don't
1511 * want to create dangling pointers in the directory.
1512 */
1513
1514 if (!search_done) {
1515 ret = __add_inode_ref(trans, root, path, log,
1516 BTRFS_I(dir),
1517 BTRFS_I(inode),
1518 inode_objectid,
1519 parent_objectid,
1520 ref_index, name, namelen,
1521 &search_done);
1522 if (ret) {
1523 if (ret == 1)
1524 ret = 0;
1525 goto out;
1526 }
1527 }
1528
1529 /*
1530 * If a reference item already exists for this inode
1531 * with the same parent and name, but different index,
1532 * drop it and the corresponding directory index entries
1533 * from the parent before adding the new reference item
1534 * and dir index entries, otherwise we would fail with
1535 * -EEXIST returned from btrfs_add_link() below.
1536 */
1537 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1538 name, namelen);
1539 if (ret > 0) {
1540 ret = btrfs_unlink_inode(trans, root,
1541 BTRFS_I(dir),
1542 BTRFS_I(inode),
1543 name, namelen);
1544 /*
1545 * If we dropped the link count to 0, bump it so
1546 * that later the iput() on the inode will not
1547 * free it. We will fixup the link count later.
1548 */
1549 if (!ret && inode->i_nlink == 0)
1550 inc_nlink(inode);
1551 /*
1552 * Whenever we need to check if a name exists or
1553 * not, we check the subvolume tree. So after an
1554 * unlink we must run delayed items, so that future
1555 * checks for a name during log replay see that the
1556 * name does not exists anymore.
1557 */
1558 if (!ret)
1559 ret = btrfs_run_delayed_items(trans);
1560 }
1561 if (ret < 0)
1562 goto out;
1563
1564 /* insert our name */
1565 ret = add_link(trans, root, dir, inode, name, namelen,
1566 ref_index);
1567 if (ret)
1568 goto out;
1569
1570 btrfs_update_inode(trans, root, inode);
1571 }
1572 /* Else, ret == 1, we already have a perfect match, we're done. */
1573
1574 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1575 kfree(name);
1576 name = NULL;
1577 if (log_ref_ver) {
1578 iput(dir);
1579 dir = NULL;
1580 }
1581 }
1582
1583 /*
1584 * Before we overwrite the inode reference item in the subvolume tree
1585 * with the item from the log tree, we must unlink all names from the
1586 * parent directory that are in the subvolume's tree inode reference
1587 * item, otherwise we end up with an inconsistent subvolume tree where
1588 * dir index entries exist for a name but there is no inode reference
1589 * item with the same name.
1590 */
1591 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1592 key);
1593 if (ret)
1594 goto out;
1595
1596 /* finally write the back reference in the inode */
1597 ret = overwrite_item(trans, root, path, eb, slot, key);
1598 out:
1599 btrfs_release_path(path);
1600 kfree(name);
1601 iput(dir);
1602 iput(inode);
1603 return ret;
1604 }
1605
insert_orphan_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 ino)1606 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1607 struct btrfs_root *root, u64 ino)
1608 {
1609 int ret;
1610
1611 ret = btrfs_insert_orphan_item(trans, root, ino);
1612 if (ret == -EEXIST)
1613 ret = 0;
1614
1615 return ret;
1616 }
1617
count_inode_extrefs(struct btrfs_root * root,struct btrfs_inode * inode,struct btrfs_path * path)1618 static int count_inode_extrefs(struct btrfs_root *root,
1619 struct btrfs_inode *inode, struct btrfs_path *path)
1620 {
1621 int ret = 0;
1622 int name_len;
1623 unsigned int nlink = 0;
1624 u32 item_size;
1625 u32 cur_offset = 0;
1626 u64 inode_objectid = btrfs_ino(inode);
1627 u64 offset = 0;
1628 unsigned long ptr;
1629 struct btrfs_inode_extref *extref;
1630 struct extent_buffer *leaf;
1631
1632 while (1) {
1633 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1634 &extref, &offset);
1635 if (ret)
1636 break;
1637
1638 leaf = path->nodes[0];
1639 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1640 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1641 cur_offset = 0;
1642
1643 while (cur_offset < item_size) {
1644 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1645 name_len = btrfs_inode_extref_name_len(leaf, extref);
1646
1647 nlink++;
1648
1649 cur_offset += name_len + sizeof(*extref);
1650 }
1651
1652 offset++;
1653 btrfs_release_path(path);
1654 }
1655 btrfs_release_path(path);
1656
1657 if (ret < 0 && ret != -ENOENT)
1658 return ret;
1659 return nlink;
1660 }
1661
count_inode_refs(struct btrfs_root * root,struct btrfs_inode * inode,struct btrfs_path * path)1662 static int count_inode_refs(struct btrfs_root *root,
1663 struct btrfs_inode *inode, struct btrfs_path *path)
1664 {
1665 int ret;
1666 struct btrfs_key key;
1667 unsigned int nlink = 0;
1668 unsigned long ptr;
1669 unsigned long ptr_end;
1670 int name_len;
1671 u64 ino = btrfs_ino(inode);
1672
1673 key.objectid = ino;
1674 key.type = BTRFS_INODE_REF_KEY;
1675 key.offset = (u64)-1;
1676
1677 while (1) {
1678 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1679 if (ret < 0)
1680 break;
1681 if (ret > 0) {
1682 if (path->slots[0] == 0)
1683 break;
1684 path->slots[0]--;
1685 }
1686 process_slot:
1687 btrfs_item_key_to_cpu(path->nodes[0], &key,
1688 path->slots[0]);
1689 if (key.objectid != ino ||
1690 key.type != BTRFS_INODE_REF_KEY)
1691 break;
1692 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1693 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1694 path->slots[0]);
1695 while (ptr < ptr_end) {
1696 struct btrfs_inode_ref *ref;
1697
1698 ref = (struct btrfs_inode_ref *)ptr;
1699 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1700 ref);
1701 ptr = (unsigned long)(ref + 1) + name_len;
1702 nlink++;
1703 }
1704
1705 if (key.offset == 0)
1706 break;
1707 if (path->slots[0] > 0) {
1708 path->slots[0]--;
1709 goto process_slot;
1710 }
1711 key.offset--;
1712 btrfs_release_path(path);
1713 }
1714 btrfs_release_path(path);
1715
1716 return nlink;
1717 }
1718
1719 /*
1720 * There are a few corners where the link count of the file can't
1721 * be properly maintained during replay. So, instead of adding
1722 * lots of complexity to the log code, we just scan the backrefs
1723 * for any file that has been through replay.
1724 *
1725 * The scan will update the link count on the inode to reflect the
1726 * number of back refs found. If it goes down to zero, the iput
1727 * will free the inode.
1728 */
fixup_inode_link_count(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct inode * inode)1729 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1730 struct btrfs_root *root,
1731 struct inode *inode)
1732 {
1733 struct btrfs_path *path;
1734 int ret;
1735 u64 nlink = 0;
1736 u64 ino = btrfs_ino(BTRFS_I(inode));
1737
1738 path = btrfs_alloc_path();
1739 if (!path)
1740 return -ENOMEM;
1741
1742 ret = count_inode_refs(root, BTRFS_I(inode), path);
1743 if (ret < 0)
1744 goto out;
1745
1746 nlink = ret;
1747
1748 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1749 if (ret < 0)
1750 goto out;
1751
1752 nlink += ret;
1753
1754 ret = 0;
1755
1756 if (nlink != inode->i_nlink) {
1757 set_nlink(inode, nlink);
1758 btrfs_update_inode(trans, root, inode);
1759 }
1760 BTRFS_I(inode)->index_cnt = (u64)-1;
1761
1762 if (inode->i_nlink == 0) {
1763 if (S_ISDIR(inode->i_mode)) {
1764 ret = replay_dir_deletes(trans, root, NULL, path,
1765 ino, 1);
1766 if (ret)
1767 goto out;
1768 }
1769 ret = insert_orphan_item(trans, root, ino);
1770 }
1771
1772 out:
1773 btrfs_free_path(path);
1774 return ret;
1775 }
1776
fixup_inode_link_counts(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path)1777 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1778 struct btrfs_root *root,
1779 struct btrfs_path *path)
1780 {
1781 int ret;
1782 struct btrfs_key key;
1783 struct inode *inode;
1784
1785 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1786 key.type = BTRFS_ORPHAN_ITEM_KEY;
1787 key.offset = (u64)-1;
1788 while (1) {
1789 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1790 if (ret < 0)
1791 break;
1792
1793 if (ret == 1) {
1794 ret = 0;
1795 if (path->slots[0] == 0)
1796 break;
1797 path->slots[0]--;
1798 }
1799
1800 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1801 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1802 key.type != BTRFS_ORPHAN_ITEM_KEY)
1803 break;
1804
1805 ret = btrfs_del_item(trans, root, path);
1806 if (ret)
1807 break;
1808
1809 btrfs_release_path(path);
1810 inode = read_one_inode(root, key.offset);
1811 if (!inode) {
1812 ret = -EIO;
1813 break;
1814 }
1815
1816 ret = fixup_inode_link_count(trans, root, inode);
1817 iput(inode);
1818 if (ret)
1819 break;
1820
1821 /*
1822 * fixup on a directory may create new entries,
1823 * make sure we always look for the highset possible
1824 * offset
1825 */
1826 key.offset = (u64)-1;
1827 }
1828 btrfs_release_path(path);
1829 return ret;
1830 }
1831
1832
1833 /*
1834 * record a given inode in the fixup dir so we can check its link
1835 * count when replay is done. The link count is incremented here
1836 * so the inode won't go away until we check it
1837 */
link_to_fixup_dir(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,u64 objectid)1838 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1839 struct btrfs_root *root,
1840 struct btrfs_path *path,
1841 u64 objectid)
1842 {
1843 struct btrfs_key key;
1844 int ret = 0;
1845 struct inode *inode;
1846
1847 inode = read_one_inode(root, objectid);
1848 if (!inode)
1849 return -EIO;
1850
1851 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1852 key.type = BTRFS_ORPHAN_ITEM_KEY;
1853 key.offset = objectid;
1854
1855 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1856
1857 btrfs_release_path(path);
1858 if (ret == 0) {
1859 if (!inode->i_nlink)
1860 set_nlink(inode, 1);
1861 else
1862 inc_nlink(inode);
1863 ret = btrfs_update_inode(trans, root, inode);
1864 } else if (ret == -EEXIST) {
1865 ret = 0;
1866 }
1867 iput(inode);
1868
1869 return ret;
1870 }
1871
1872 /*
1873 * when replaying the log for a directory, we only insert names
1874 * for inodes that actually exist. This means an fsync on a directory
1875 * does not implicitly fsync all the new files in it
1876 */
insert_one_name(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 dirid,u64 index,char * name,int name_len,struct btrfs_key * location)1877 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1878 struct btrfs_root *root,
1879 u64 dirid, u64 index,
1880 char *name, int name_len,
1881 struct btrfs_key *location)
1882 {
1883 struct inode *inode;
1884 struct inode *dir;
1885 int ret;
1886
1887 inode = read_one_inode(root, location->objectid);
1888 if (!inode)
1889 return -ENOENT;
1890
1891 dir = read_one_inode(root, dirid);
1892 if (!dir) {
1893 iput(inode);
1894 return -EIO;
1895 }
1896
1897 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1898 name_len, 1, index);
1899
1900 /* FIXME, put inode into FIXUP list */
1901
1902 iput(inode);
1903 iput(dir);
1904 return ret;
1905 }
1906
1907 /*
1908 * take a single entry in a log directory item and replay it into
1909 * the subvolume.
1910 *
1911 * if a conflicting item exists in the subdirectory already,
1912 * the inode it points to is unlinked and put into the link count
1913 * fix up tree.
1914 *
1915 * If a name from the log points to a file or directory that does
1916 * not exist in the FS, it is skipped. fsyncs on directories
1917 * do not force down inodes inside that directory, just changes to the
1918 * names or unlinks in a directory.
1919 *
1920 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1921 * non-existing inode) and 1 if the name was replayed.
1922 */
replay_one_name(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct extent_buffer * eb,struct btrfs_dir_item * di,struct btrfs_key * key)1923 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1924 struct btrfs_root *root,
1925 struct btrfs_path *path,
1926 struct extent_buffer *eb,
1927 struct btrfs_dir_item *di,
1928 struct btrfs_key *key)
1929 {
1930 char *name;
1931 int name_len;
1932 struct btrfs_dir_item *dst_di;
1933 struct btrfs_key found_key;
1934 struct btrfs_key log_key;
1935 struct inode *dir;
1936 u8 log_type;
1937 bool exists;
1938 int ret;
1939 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1940 bool name_added = false;
1941
1942 dir = read_one_inode(root, key->objectid);
1943 if (!dir)
1944 return -EIO;
1945
1946 name_len = btrfs_dir_name_len(eb, di);
1947 name = kmalloc(name_len, GFP_NOFS);
1948 if (!name) {
1949 ret = -ENOMEM;
1950 goto out;
1951 }
1952
1953 log_type = btrfs_dir_type(eb, di);
1954 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1955 name_len);
1956
1957 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1958 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1959 btrfs_release_path(path);
1960 if (ret < 0)
1961 goto out;
1962 exists = (ret == 0);
1963 ret = 0;
1964
1965 if (key->type == BTRFS_DIR_ITEM_KEY) {
1966 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1967 name, name_len, 1);
1968 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1969 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1970 key->objectid,
1971 key->offset, name,
1972 name_len, 1);
1973 } else {
1974 /* Corruption */
1975 ret = -EINVAL;
1976 goto out;
1977 }
1978
1979 if (dst_di == ERR_PTR(-ENOENT))
1980 dst_di = NULL;
1981
1982 if (IS_ERR(dst_di)) {
1983 ret = PTR_ERR(dst_di);
1984 goto out;
1985 } else if (!dst_di) {
1986 /* we need a sequence number to insert, so we only
1987 * do inserts for the BTRFS_DIR_INDEX_KEY types
1988 */
1989 if (key->type != BTRFS_DIR_INDEX_KEY)
1990 goto out;
1991 goto insert;
1992 }
1993
1994 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1995 /* the existing item matches the logged item */
1996 if (found_key.objectid == log_key.objectid &&
1997 found_key.type == log_key.type &&
1998 found_key.offset == log_key.offset &&
1999 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
2000 update_size = false;
2001 goto out;
2002 }
2003
2004 /*
2005 * don't drop the conflicting directory entry if the inode
2006 * for the new entry doesn't exist
2007 */
2008 if (!exists)
2009 goto out;
2010
2011 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
2012 if (ret)
2013 goto out;
2014
2015 if (key->type == BTRFS_DIR_INDEX_KEY)
2016 goto insert;
2017 out:
2018 btrfs_release_path(path);
2019 if (!ret && update_size) {
2020 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2021 ret = btrfs_update_inode(trans, root, dir);
2022 }
2023 kfree(name);
2024 iput(dir);
2025 if (!ret && name_added)
2026 ret = 1;
2027 return ret;
2028
2029 insert:
2030 /*
2031 * Check if the inode reference exists in the log for the given name,
2032 * inode and parent inode
2033 */
2034 found_key.objectid = log_key.objectid;
2035 found_key.type = BTRFS_INODE_REF_KEY;
2036 found_key.offset = key->objectid;
2037 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
2038 if (ret < 0) {
2039 goto out;
2040 } else if (ret) {
2041 /* The dentry will be added later. */
2042 ret = 0;
2043 update_size = false;
2044 goto out;
2045 }
2046
2047 found_key.objectid = log_key.objectid;
2048 found_key.type = BTRFS_INODE_EXTREF_KEY;
2049 found_key.offset = key->objectid;
2050 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2051 name_len);
2052 if (ret < 0) {
2053 goto out;
2054 } else if (ret) {
2055 /* The dentry will be added later. */
2056 ret = 0;
2057 update_size = false;
2058 goto out;
2059 }
2060 btrfs_release_path(path);
2061 ret = insert_one_name(trans, root, key->objectid, key->offset,
2062 name, name_len, &log_key);
2063 if (ret && ret != -ENOENT && ret != -EEXIST)
2064 goto out;
2065 if (!ret)
2066 name_added = true;
2067 update_size = false;
2068 ret = 0;
2069 goto out;
2070 }
2071
2072 /*
2073 * find all the names in a directory item and reconcile them into
2074 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2075 * one name in a directory item, but the same code gets used for
2076 * both directory index types
2077 */
replay_one_dir_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct extent_buffer * eb,int slot,struct btrfs_key * key)2078 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2079 struct btrfs_root *root,
2080 struct btrfs_path *path,
2081 struct extent_buffer *eb, int slot,
2082 struct btrfs_key *key)
2083 {
2084 int ret = 0;
2085 u32 item_size = btrfs_item_size_nr(eb, slot);
2086 struct btrfs_dir_item *di;
2087 int name_len;
2088 unsigned long ptr;
2089 unsigned long ptr_end;
2090 struct btrfs_path *fixup_path = NULL;
2091
2092 ptr = btrfs_item_ptr_offset(eb, slot);
2093 ptr_end = ptr + item_size;
2094 while (ptr < ptr_end) {
2095 di = (struct btrfs_dir_item *)ptr;
2096 name_len = btrfs_dir_name_len(eb, di);
2097 ret = replay_one_name(trans, root, path, eb, di, key);
2098 if (ret < 0)
2099 break;
2100 ptr = (unsigned long)(di + 1);
2101 ptr += name_len;
2102
2103 /*
2104 * If this entry refers to a non-directory (directories can not
2105 * have a link count > 1) and it was added in the transaction
2106 * that was not committed, make sure we fixup the link count of
2107 * the inode it the entry points to. Otherwise something like
2108 * the following would result in a directory pointing to an
2109 * inode with a wrong link that does not account for this dir
2110 * entry:
2111 *
2112 * mkdir testdir
2113 * touch testdir/foo
2114 * touch testdir/bar
2115 * sync
2116 *
2117 * ln testdir/bar testdir/bar_link
2118 * ln testdir/foo testdir/foo_link
2119 * xfs_io -c "fsync" testdir/bar
2120 *
2121 * <power failure>
2122 *
2123 * mount fs, log replay happens
2124 *
2125 * File foo would remain with a link count of 1 when it has two
2126 * entries pointing to it in the directory testdir. This would
2127 * make it impossible to ever delete the parent directory has
2128 * it would result in stale dentries that can never be deleted.
2129 */
2130 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2131 struct btrfs_key di_key;
2132
2133 if (!fixup_path) {
2134 fixup_path = btrfs_alloc_path();
2135 if (!fixup_path) {
2136 ret = -ENOMEM;
2137 break;
2138 }
2139 }
2140
2141 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2142 ret = link_to_fixup_dir(trans, root, fixup_path,
2143 di_key.objectid);
2144 if (ret)
2145 break;
2146 }
2147 ret = 0;
2148 }
2149 btrfs_free_path(fixup_path);
2150 return ret;
2151 }
2152
2153 /*
2154 * directory replay has two parts. There are the standard directory
2155 * items in the log copied from the subvolume, and range items
2156 * created in the log while the subvolume was logged.
2157 *
2158 * The range items tell us which parts of the key space the log
2159 * is authoritative for. During replay, if a key in the subvolume
2160 * directory is in a logged range item, but not actually in the log
2161 * that means it was deleted from the directory before the fsync
2162 * and should be removed.
2163 */
find_dir_range(struct btrfs_root * root,struct btrfs_path * path,u64 dirid,int key_type,u64 * start_ret,u64 * end_ret)2164 static noinline int find_dir_range(struct btrfs_root *root,
2165 struct btrfs_path *path,
2166 u64 dirid, int key_type,
2167 u64 *start_ret, u64 *end_ret)
2168 {
2169 struct btrfs_key key;
2170 u64 found_end;
2171 struct btrfs_dir_log_item *item;
2172 int ret;
2173 int nritems;
2174
2175 if (*start_ret == (u64)-1)
2176 return 1;
2177
2178 key.objectid = dirid;
2179 key.type = key_type;
2180 key.offset = *start_ret;
2181
2182 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2183 if (ret < 0)
2184 goto out;
2185 if (ret > 0) {
2186 if (path->slots[0] == 0)
2187 goto out;
2188 path->slots[0]--;
2189 }
2190 if (ret != 0)
2191 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2192
2193 if (key.type != key_type || key.objectid != dirid) {
2194 ret = 1;
2195 goto next;
2196 }
2197 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2198 struct btrfs_dir_log_item);
2199 found_end = btrfs_dir_log_end(path->nodes[0], item);
2200
2201 if (*start_ret >= key.offset && *start_ret <= found_end) {
2202 ret = 0;
2203 *start_ret = key.offset;
2204 *end_ret = found_end;
2205 goto out;
2206 }
2207 ret = 1;
2208 next:
2209 /* check the next slot in the tree to see if it is a valid item */
2210 nritems = btrfs_header_nritems(path->nodes[0]);
2211 path->slots[0]++;
2212 if (path->slots[0] >= nritems) {
2213 ret = btrfs_next_leaf(root, path);
2214 if (ret)
2215 goto out;
2216 }
2217
2218 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2219
2220 if (key.type != key_type || key.objectid != dirid) {
2221 ret = 1;
2222 goto out;
2223 }
2224 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2225 struct btrfs_dir_log_item);
2226 found_end = btrfs_dir_log_end(path->nodes[0], item);
2227 *start_ret = key.offset;
2228 *end_ret = found_end;
2229 ret = 0;
2230 out:
2231 btrfs_release_path(path);
2232 return ret;
2233 }
2234
2235 /*
2236 * this looks for a given directory item in the log. If the directory
2237 * item is not in the log, the item is removed and the inode it points
2238 * to is unlinked
2239 */
check_item_in_log(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_root * log,struct btrfs_path * path,struct btrfs_path * log_path,struct inode * dir,struct btrfs_key * dir_key)2240 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2241 struct btrfs_root *root,
2242 struct btrfs_root *log,
2243 struct btrfs_path *path,
2244 struct btrfs_path *log_path,
2245 struct inode *dir,
2246 struct btrfs_key *dir_key)
2247 {
2248 int ret;
2249 struct extent_buffer *eb;
2250 int slot;
2251 u32 item_size;
2252 struct btrfs_dir_item *di;
2253 struct btrfs_dir_item *log_di;
2254 int name_len;
2255 unsigned long ptr;
2256 unsigned long ptr_end;
2257 char *name;
2258 struct inode *inode;
2259 struct btrfs_key location;
2260
2261 again:
2262 eb = path->nodes[0];
2263 slot = path->slots[0];
2264 item_size = btrfs_item_size_nr(eb, slot);
2265 ptr = btrfs_item_ptr_offset(eb, slot);
2266 ptr_end = ptr + item_size;
2267 while (ptr < ptr_end) {
2268 di = (struct btrfs_dir_item *)ptr;
2269 name_len = btrfs_dir_name_len(eb, di);
2270 name = kmalloc(name_len, GFP_NOFS);
2271 if (!name) {
2272 ret = -ENOMEM;
2273 goto out;
2274 }
2275 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2276 name_len);
2277 log_di = NULL;
2278 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2279 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2280 dir_key->objectid,
2281 name, name_len, 0);
2282 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2283 log_di = btrfs_lookup_dir_index_item(trans, log,
2284 log_path,
2285 dir_key->objectid,
2286 dir_key->offset,
2287 name, name_len, 0);
2288 }
2289 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2290 btrfs_dir_item_key_to_cpu(eb, di, &location);
2291 btrfs_release_path(path);
2292 btrfs_release_path(log_path);
2293 inode = read_one_inode(root, location.objectid);
2294 if (!inode) {
2295 kfree(name);
2296 return -EIO;
2297 }
2298
2299 ret = link_to_fixup_dir(trans, root,
2300 path, location.objectid);
2301 if (ret) {
2302 kfree(name);
2303 iput(inode);
2304 goto out;
2305 }
2306
2307 inc_nlink(inode);
2308 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2309 BTRFS_I(inode), name, name_len);
2310 if (!ret)
2311 ret = btrfs_run_delayed_items(trans);
2312 kfree(name);
2313 iput(inode);
2314 if (ret)
2315 goto out;
2316
2317 /* there might still be more names under this key
2318 * check and repeat if required
2319 */
2320 ret = btrfs_search_slot(NULL, root, dir_key, path,
2321 0, 0);
2322 if (ret == 0)
2323 goto again;
2324 ret = 0;
2325 goto out;
2326 } else if (IS_ERR(log_di)) {
2327 kfree(name);
2328 return PTR_ERR(log_di);
2329 }
2330 btrfs_release_path(log_path);
2331 kfree(name);
2332
2333 ptr = (unsigned long)(di + 1);
2334 ptr += name_len;
2335 }
2336 ret = 0;
2337 out:
2338 btrfs_release_path(path);
2339 btrfs_release_path(log_path);
2340 return ret;
2341 }
2342
replay_xattr_deletes(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_root * log,struct btrfs_path * path,const u64 ino)2343 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2344 struct btrfs_root *root,
2345 struct btrfs_root *log,
2346 struct btrfs_path *path,
2347 const u64 ino)
2348 {
2349 struct btrfs_key search_key;
2350 struct btrfs_path *log_path;
2351 int i;
2352 int nritems;
2353 int ret;
2354
2355 log_path = btrfs_alloc_path();
2356 if (!log_path)
2357 return -ENOMEM;
2358
2359 search_key.objectid = ino;
2360 search_key.type = BTRFS_XATTR_ITEM_KEY;
2361 search_key.offset = 0;
2362 again:
2363 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2364 if (ret < 0)
2365 goto out;
2366 process_leaf:
2367 nritems = btrfs_header_nritems(path->nodes[0]);
2368 for (i = path->slots[0]; i < nritems; i++) {
2369 struct btrfs_key key;
2370 struct btrfs_dir_item *di;
2371 struct btrfs_dir_item *log_di;
2372 u32 total_size;
2373 u32 cur;
2374
2375 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2376 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2377 ret = 0;
2378 goto out;
2379 }
2380
2381 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2382 total_size = btrfs_item_size_nr(path->nodes[0], i);
2383 cur = 0;
2384 while (cur < total_size) {
2385 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2386 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2387 u32 this_len = sizeof(*di) + name_len + data_len;
2388 char *name;
2389
2390 name = kmalloc(name_len, GFP_NOFS);
2391 if (!name) {
2392 ret = -ENOMEM;
2393 goto out;
2394 }
2395 read_extent_buffer(path->nodes[0], name,
2396 (unsigned long)(di + 1), name_len);
2397
2398 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2399 name, name_len, 0);
2400 btrfs_release_path(log_path);
2401 if (!log_di) {
2402 /* Doesn't exist in log tree, so delete it. */
2403 btrfs_release_path(path);
2404 di = btrfs_lookup_xattr(trans, root, path, ino,
2405 name, name_len, -1);
2406 kfree(name);
2407 if (IS_ERR(di)) {
2408 ret = PTR_ERR(di);
2409 goto out;
2410 }
2411 ASSERT(di);
2412 ret = btrfs_delete_one_dir_name(trans, root,
2413 path, di);
2414 if (ret)
2415 goto out;
2416 btrfs_release_path(path);
2417 search_key = key;
2418 goto again;
2419 }
2420 kfree(name);
2421 if (IS_ERR(log_di)) {
2422 ret = PTR_ERR(log_di);
2423 goto out;
2424 }
2425 cur += this_len;
2426 di = (struct btrfs_dir_item *)((char *)di + this_len);
2427 }
2428 }
2429 ret = btrfs_next_leaf(root, path);
2430 if (ret > 0)
2431 ret = 0;
2432 else if (ret == 0)
2433 goto process_leaf;
2434 out:
2435 btrfs_free_path(log_path);
2436 btrfs_release_path(path);
2437 return ret;
2438 }
2439
2440
2441 /*
2442 * deletion replay happens before we copy any new directory items
2443 * out of the log or out of backreferences from inodes. It
2444 * scans the log to find ranges of keys that log is authoritative for,
2445 * and then scans the directory to find items in those ranges that are
2446 * not present in the log.
2447 *
2448 * Anything we don't find in the log is unlinked and removed from the
2449 * directory.
2450 */
replay_dir_deletes(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_root * log,struct btrfs_path * path,u64 dirid,int del_all)2451 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2452 struct btrfs_root *root,
2453 struct btrfs_root *log,
2454 struct btrfs_path *path,
2455 u64 dirid, int del_all)
2456 {
2457 u64 range_start;
2458 u64 range_end;
2459 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2460 int ret = 0;
2461 struct btrfs_key dir_key;
2462 struct btrfs_key found_key;
2463 struct btrfs_path *log_path;
2464 struct inode *dir;
2465
2466 dir_key.objectid = dirid;
2467 dir_key.type = BTRFS_DIR_ITEM_KEY;
2468 log_path = btrfs_alloc_path();
2469 if (!log_path)
2470 return -ENOMEM;
2471
2472 dir = read_one_inode(root, dirid);
2473 /* it isn't an error if the inode isn't there, that can happen
2474 * because we replay the deletes before we copy in the inode item
2475 * from the log
2476 */
2477 if (!dir) {
2478 btrfs_free_path(log_path);
2479 return 0;
2480 }
2481 again:
2482 range_start = 0;
2483 range_end = 0;
2484 while (1) {
2485 if (del_all)
2486 range_end = (u64)-1;
2487 else {
2488 ret = find_dir_range(log, path, dirid, key_type,
2489 &range_start, &range_end);
2490 if (ret < 0)
2491 goto out;
2492 else if (ret > 0)
2493 break;
2494 }
2495
2496 dir_key.offset = range_start;
2497 while (1) {
2498 int nritems;
2499 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2500 0, 0);
2501 if (ret < 0)
2502 goto out;
2503
2504 nritems = btrfs_header_nritems(path->nodes[0]);
2505 if (path->slots[0] >= nritems) {
2506 ret = btrfs_next_leaf(root, path);
2507 if (ret == 1)
2508 break;
2509 else if (ret < 0)
2510 goto out;
2511 }
2512 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2513 path->slots[0]);
2514 if (found_key.objectid != dirid ||
2515 found_key.type != dir_key.type)
2516 goto next_type;
2517
2518 if (found_key.offset > range_end)
2519 break;
2520
2521 ret = check_item_in_log(trans, root, log, path,
2522 log_path, dir,
2523 &found_key);
2524 if (ret)
2525 goto out;
2526 if (found_key.offset == (u64)-1)
2527 break;
2528 dir_key.offset = found_key.offset + 1;
2529 }
2530 btrfs_release_path(path);
2531 if (range_end == (u64)-1)
2532 break;
2533 range_start = range_end + 1;
2534 }
2535
2536 next_type:
2537 ret = 0;
2538 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2539 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2540 dir_key.type = BTRFS_DIR_INDEX_KEY;
2541 btrfs_release_path(path);
2542 goto again;
2543 }
2544 out:
2545 btrfs_release_path(path);
2546 btrfs_free_path(log_path);
2547 iput(dir);
2548 return ret;
2549 }
2550
2551 /*
2552 * the process_func used to replay items from the log tree. This
2553 * gets called in two different stages. The first stage just looks
2554 * for inodes and makes sure they are all copied into the subvolume.
2555 *
2556 * The second stage copies all the other item types from the log into
2557 * the subvolume. The two stage approach is slower, but gets rid of
2558 * lots of complexity around inodes referencing other inodes that exist
2559 * only in the log (references come from either directory items or inode
2560 * back refs).
2561 */
replay_one_buffer(struct btrfs_root * log,struct extent_buffer * eb,struct walk_control * wc,u64 gen,int level)2562 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2563 struct walk_control *wc, u64 gen, int level)
2564 {
2565 int nritems;
2566 struct btrfs_path *path;
2567 struct btrfs_root *root = wc->replay_dest;
2568 struct btrfs_key key;
2569 int i;
2570 int ret;
2571
2572 ret = btrfs_read_buffer(eb, gen, level, NULL);
2573 if (ret)
2574 return ret;
2575
2576 level = btrfs_header_level(eb);
2577
2578 if (level != 0)
2579 return 0;
2580
2581 path = btrfs_alloc_path();
2582 if (!path)
2583 return -ENOMEM;
2584
2585 nritems = btrfs_header_nritems(eb);
2586 for (i = 0; i < nritems; i++) {
2587 btrfs_item_key_to_cpu(eb, &key, i);
2588
2589 /* inode keys are done during the first stage */
2590 if (key.type == BTRFS_INODE_ITEM_KEY &&
2591 wc->stage == LOG_WALK_REPLAY_INODES) {
2592 struct btrfs_inode_item *inode_item;
2593 u32 mode;
2594
2595 inode_item = btrfs_item_ptr(eb, i,
2596 struct btrfs_inode_item);
2597 /*
2598 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2599 * and never got linked before the fsync, skip it, as
2600 * replaying it is pointless since it would be deleted
2601 * later. We skip logging tmpfiles, but it's always
2602 * possible we are replaying a log created with a kernel
2603 * that used to log tmpfiles.
2604 */
2605 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2606 wc->ignore_cur_inode = true;
2607 continue;
2608 } else {
2609 wc->ignore_cur_inode = false;
2610 }
2611 ret = replay_xattr_deletes(wc->trans, root, log,
2612 path, key.objectid);
2613 if (ret)
2614 break;
2615 mode = btrfs_inode_mode(eb, inode_item);
2616 if (S_ISDIR(mode)) {
2617 ret = replay_dir_deletes(wc->trans,
2618 root, log, path, key.objectid, 0);
2619 if (ret)
2620 break;
2621 }
2622 ret = overwrite_item(wc->trans, root, path,
2623 eb, i, &key);
2624 if (ret)
2625 break;
2626
2627 /*
2628 * Before replaying extents, truncate the inode to its
2629 * size. We need to do it now and not after log replay
2630 * because before an fsync we can have prealloc extents
2631 * added beyond the inode's i_size. If we did it after,
2632 * through orphan cleanup for example, we would drop
2633 * those prealloc extents just after replaying them.
2634 */
2635 if (S_ISREG(mode)) {
2636 struct inode *inode;
2637 u64 from;
2638
2639 inode = read_one_inode(root, key.objectid);
2640 if (!inode) {
2641 ret = -EIO;
2642 break;
2643 }
2644 from = ALIGN(i_size_read(inode),
2645 root->fs_info->sectorsize);
2646 ret = btrfs_drop_extents(wc->trans, root, inode,
2647 from, (u64)-1, 1);
2648 if (!ret) {
2649 /* Update the inode's nbytes. */
2650 ret = btrfs_update_inode(wc->trans,
2651 root, inode);
2652 }
2653 iput(inode);
2654 if (ret)
2655 break;
2656 }
2657
2658 ret = link_to_fixup_dir(wc->trans, root,
2659 path, key.objectid);
2660 if (ret)
2661 break;
2662 }
2663
2664 if (wc->ignore_cur_inode)
2665 continue;
2666
2667 if (key.type == BTRFS_DIR_INDEX_KEY &&
2668 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2669 ret = replay_one_dir_item(wc->trans, root, path,
2670 eb, i, &key);
2671 if (ret)
2672 break;
2673 }
2674
2675 if (wc->stage < LOG_WALK_REPLAY_ALL)
2676 continue;
2677
2678 /* these keys are simply copied */
2679 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2680 ret = overwrite_item(wc->trans, root, path,
2681 eb, i, &key);
2682 if (ret)
2683 break;
2684 } else if (key.type == BTRFS_INODE_REF_KEY ||
2685 key.type == BTRFS_INODE_EXTREF_KEY) {
2686 ret = add_inode_ref(wc->trans, root, log, path,
2687 eb, i, &key);
2688 if (ret && ret != -ENOENT)
2689 break;
2690 ret = 0;
2691 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2692 ret = replay_one_extent(wc->trans, root, path,
2693 eb, i, &key);
2694 if (ret)
2695 break;
2696 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2697 ret = replay_one_dir_item(wc->trans, root, path,
2698 eb, i, &key);
2699 if (ret)
2700 break;
2701 }
2702 }
2703 btrfs_free_path(path);
2704 return ret;
2705 }
2706
2707 /*
2708 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2709 */
unaccount_log_buffer(struct btrfs_fs_info * fs_info,u64 start)2710 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2711 {
2712 struct btrfs_block_group *cache;
2713
2714 cache = btrfs_lookup_block_group(fs_info, start);
2715 if (!cache) {
2716 btrfs_err(fs_info, "unable to find block group for %llu", start);
2717 return;
2718 }
2719
2720 spin_lock(&cache->space_info->lock);
2721 spin_lock(&cache->lock);
2722 cache->reserved -= fs_info->nodesize;
2723 cache->space_info->bytes_reserved -= fs_info->nodesize;
2724 spin_unlock(&cache->lock);
2725 spin_unlock(&cache->space_info->lock);
2726
2727 btrfs_put_block_group(cache);
2728 }
2729
walk_down_log_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int * level,struct walk_control * wc)2730 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2731 struct btrfs_root *root,
2732 struct btrfs_path *path, int *level,
2733 struct walk_control *wc)
2734 {
2735 struct btrfs_fs_info *fs_info = root->fs_info;
2736 u64 bytenr;
2737 u64 ptr_gen;
2738 struct extent_buffer *next;
2739 struct extent_buffer *cur;
2740 u32 blocksize;
2741 int ret = 0;
2742
2743 while (*level > 0) {
2744 struct btrfs_key first_key;
2745
2746 cur = path->nodes[*level];
2747
2748 WARN_ON(btrfs_header_level(cur) != *level);
2749
2750 if (path->slots[*level] >=
2751 btrfs_header_nritems(cur))
2752 break;
2753
2754 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2755 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2756 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2757 blocksize = fs_info->nodesize;
2758
2759 next = btrfs_find_create_tree_block(fs_info, bytenr);
2760 if (IS_ERR(next))
2761 return PTR_ERR(next);
2762
2763 if (*level == 1) {
2764 ret = wc->process_func(root, next, wc, ptr_gen,
2765 *level - 1);
2766 if (ret) {
2767 free_extent_buffer(next);
2768 return ret;
2769 }
2770
2771 path->slots[*level]++;
2772 if (wc->free) {
2773 ret = btrfs_read_buffer(next, ptr_gen,
2774 *level - 1, &first_key);
2775 if (ret) {
2776 free_extent_buffer(next);
2777 return ret;
2778 }
2779
2780 if (trans) {
2781 btrfs_tree_lock(next);
2782 btrfs_set_lock_blocking_write(next);
2783 btrfs_clean_tree_block(next);
2784 btrfs_wait_tree_block_writeback(next);
2785 btrfs_tree_unlock(next);
2786 ret = btrfs_pin_reserved_extent(trans,
2787 bytenr, blocksize);
2788 if (ret) {
2789 free_extent_buffer(next);
2790 return ret;
2791 }
2792 } else {
2793 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2794 clear_extent_buffer_dirty(next);
2795 unaccount_log_buffer(fs_info, bytenr);
2796 }
2797 }
2798 free_extent_buffer(next);
2799 continue;
2800 }
2801 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2802 if (ret) {
2803 free_extent_buffer(next);
2804 return ret;
2805 }
2806
2807 if (path->nodes[*level-1])
2808 free_extent_buffer(path->nodes[*level-1]);
2809 path->nodes[*level-1] = next;
2810 *level = btrfs_header_level(next);
2811 path->slots[*level] = 0;
2812 cond_resched();
2813 }
2814 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2815
2816 cond_resched();
2817 return 0;
2818 }
2819
walk_up_log_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int * level,struct walk_control * wc)2820 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2821 struct btrfs_root *root,
2822 struct btrfs_path *path, int *level,
2823 struct walk_control *wc)
2824 {
2825 struct btrfs_fs_info *fs_info = root->fs_info;
2826 int i;
2827 int slot;
2828 int ret;
2829
2830 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2831 slot = path->slots[i];
2832 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2833 path->slots[i]++;
2834 *level = i;
2835 WARN_ON(*level == 0);
2836 return 0;
2837 } else {
2838 ret = wc->process_func(root, path->nodes[*level], wc,
2839 btrfs_header_generation(path->nodes[*level]),
2840 *level);
2841 if (ret)
2842 return ret;
2843
2844 if (wc->free) {
2845 struct extent_buffer *next;
2846
2847 next = path->nodes[*level];
2848
2849 if (trans) {
2850 btrfs_tree_lock(next);
2851 btrfs_set_lock_blocking_write(next);
2852 btrfs_clean_tree_block(next);
2853 btrfs_wait_tree_block_writeback(next);
2854 btrfs_tree_unlock(next);
2855 ret = btrfs_pin_reserved_extent(trans,
2856 path->nodes[*level]->start,
2857 path->nodes[*level]->len);
2858 if (ret)
2859 return ret;
2860 } else {
2861 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2862 clear_extent_buffer_dirty(next);
2863
2864 unaccount_log_buffer(fs_info,
2865 path->nodes[*level]->start);
2866 }
2867 }
2868 free_extent_buffer(path->nodes[*level]);
2869 path->nodes[*level] = NULL;
2870 *level = i + 1;
2871 }
2872 }
2873 return 1;
2874 }
2875
2876 /*
2877 * drop the reference count on the tree rooted at 'snap'. This traverses
2878 * the tree freeing any blocks that have a ref count of zero after being
2879 * decremented.
2880 */
walk_log_tree(struct btrfs_trans_handle * trans,struct btrfs_root * log,struct walk_control * wc)2881 static int walk_log_tree(struct btrfs_trans_handle *trans,
2882 struct btrfs_root *log, struct walk_control *wc)
2883 {
2884 struct btrfs_fs_info *fs_info = log->fs_info;
2885 int ret = 0;
2886 int wret;
2887 int level;
2888 struct btrfs_path *path;
2889 int orig_level;
2890
2891 path = btrfs_alloc_path();
2892 if (!path)
2893 return -ENOMEM;
2894
2895 level = btrfs_header_level(log->node);
2896 orig_level = level;
2897 path->nodes[level] = log->node;
2898 atomic_inc(&log->node->refs);
2899 path->slots[level] = 0;
2900
2901 while (1) {
2902 wret = walk_down_log_tree(trans, log, path, &level, wc);
2903 if (wret > 0)
2904 break;
2905 if (wret < 0) {
2906 ret = wret;
2907 goto out;
2908 }
2909
2910 wret = walk_up_log_tree(trans, log, path, &level, wc);
2911 if (wret > 0)
2912 break;
2913 if (wret < 0) {
2914 ret = wret;
2915 goto out;
2916 }
2917 }
2918
2919 /* was the root node processed? if not, catch it here */
2920 if (path->nodes[orig_level]) {
2921 ret = wc->process_func(log, path->nodes[orig_level], wc,
2922 btrfs_header_generation(path->nodes[orig_level]),
2923 orig_level);
2924 if (ret)
2925 goto out;
2926 if (wc->free) {
2927 struct extent_buffer *next;
2928
2929 next = path->nodes[orig_level];
2930
2931 if (trans) {
2932 btrfs_tree_lock(next);
2933 btrfs_set_lock_blocking_write(next);
2934 btrfs_clean_tree_block(next);
2935 btrfs_wait_tree_block_writeback(next);
2936 btrfs_tree_unlock(next);
2937 ret = btrfs_pin_reserved_extent(trans,
2938 next->start, next->len);
2939 if (ret)
2940 goto out;
2941 } else {
2942 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2943 clear_extent_buffer_dirty(next);
2944 unaccount_log_buffer(fs_info, next->start);
2945 }
2946 }
2947 }
2948
2949 out:
2950 btrfs_free_path(path);
2951 return ret;
2952 }
2953
2954 /*
2955 * helper function to update the item for a given subvolumes log root
2956 * in the tree of log roots
2957 */
update_log_root(struct btrfs_trans_handle * trans,struct btrfs_root * log,struct btrfs_root_item * root_item)2958 static int update_log_root(struct btrfs_trans_handle *trans,
2959 struct btrfs_root *log,
2960 struct btrfs_root_item *root_item)
2961 {
2962 struct btrfs_fs_info *fs_info = log->fs_info;
2963 int ret;
2964
2965 if (log->log_transid == 1) {
2966 /* insert root item on the first sync */
2967 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2968 &log->root_key, root_item);
2969 } else {
2970 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2971 &log->root_key, root_item);
2972 }
2973 return ret;
2974 }
2975
wait_log_commit(struct btrfs_root * root,int transid)2976 static void wait_log_commit(struct btrfs_root *root, int transid)
2977 {
2978 DEFINE_WAIT(wait);
2979 int index = transid % 2;
2980
2981 /*
2982 * we only allow two pending log transactions at a time,
2983 * so we know that if ours is more than 2 older than the
2984 * current transaction, we're done
2985 */
2986 for (;;) {
2987 prepare_to_wait(&root->log_commit_wait[index],
2988 &wait, TASK_UNINTERRUPTIBLE);
2989
2990 if (!(root->log_transid_committed < transid &&
2991 atomic_read(&root->log_commit[index])))
2992 break;
2993
2994 mutex_unlock(&root->log_mutex);
2995 schedule();
2996 mutex_lock(&root->log_mutex);
2997 }
2998 finish_wait(&root->log_commit_wait[index], &wait);
2999 }
3000
wait_for_writer(struct btrfs_root * root)3001 static void wait_for_writer(struct btrfs_root *root)
3002 {
3003 DEFINE_WAIT(wait);
3004
3005 for (;;) {
3006 prepare_to_wait(&root->log_writer_wait, &wait,
3007 TASK_UNINTERRUPTIBLE);
3008 if (!atomic_read(&root->log_writers))
3009 break;
3010
3011 mutex_unlock(&root->log_mutex);
3012 schedule();
3013 mutex_lock(&root->log_mutex);
3014 }
3015 finish_wait(&root->log_writer_wait, &wait);
3016 }
3017
btrfs_remove_log_ctx(struct btrfs_root * root,struct btrfs_log_ctx * ctx)3018 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
3019 struct btrfs_log_ctx *ctx)
3020 {
3021 if (!ctx)
3022 return;
3023
3024 mutex_lock(&root->log_mutex);
3025 list_del_init(&ctx->list);
3026 mutex_unlock(&root->log_mutex);
3027 }
3028
3029 /*
3030 * Invoked in log mutex context, or be sure there is no other task which
3031 * can access the list.
3032 */
btrfs_remove_all_log_ctxs(struct btrfs_root * root,int index,int error)3033 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3034 int index, int error)
3035 {
3036 struct btrfs_log_ctx *ctx;
3037 struct btrfs_log_ctx *safe;
3038
3039 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3040 list_del_init(&ctx->list);
3041 ctx->log_ret = error;
3042 }
3043
3044 INIT_LIST_HEAD(&root->log_ctxs[index]);
3045 }
3046
3047 /*
3048 * btrfs_sync_log does sends a given tree log down to the disk and
3049 * updates the super blocks to record it. When this call is done,
3050 * you know that any inodes previously logged are safely on disk only
3051 * if it returns 0.
3052 *
3053 * Any other return value means you need to call btrfs_commit_transaction.
3054 * Some of the edge cases for fsyncing directories that have had unlinks
3055 * or renames done in the past mean that sometimes the only safe
3056 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3057 * that has happened.
3058 */
btrfs_sync_log(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_log_ctx * ctx)3059 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3060 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3061 {
3062 int index1;
3063 int index2;
3064 int mark;
3065 int ret;
3066 struct btrfs_fs_info *fs_info = root->fs_info;
3067 struct btrfs_root *log = root->log_root;
3068 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3069 struct btrfs_root_item new_root_item;
3070 int log_transid = 0;
3071 struct btrfs_log_ctx root_log_ctx;
3072 struct blk_plug plug;
3073
3074 mutex_lock(&root->log_mutex);
3075 log_transid = ctx->log_transid;
3076 if (root->log_transid_committed >= log_transid) {
3077 mutex_unlock(&root->log_mutex);
3078 return ctx->log_ret;
3079 }
3080
3081 index1 = log_transid % 2;
3082 if (atomic_read(&root->log_commit[index1])) {
3083 wait_log_commit(root, log_transid);
3084 mutex_unlock(&root->log_mutex);
3085 return ctx->log_ret;
3086 }
3087 ASSERT(log_transid == root->log_transid);
3088 atomic_set(&root->log_commit[index1], 1);
3089
3090 /* wait for previous tree log sync to complete */
3091 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3092 wait_log_commit(root, log_transid - 1);
3093
3094 while (1) {
3095 int batch = atomic_read(&root->log_batch);
3096 /* when we're on an ssd, just kick the log commit out */
3097 if (!btrfs_test_opt(fs_info, SSD) &&
3098 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3099 mutex_unlock(&root->log_mutex);
3100 schedule_timeout_uninterruptible(1);
3101 mutex_lock(&root->log_mutex);
3102 }
3103 wait_for_writer(root);
3104 if (batch == atomic_read(&root->log_batch))
3105 break;
3106 }
3107
3108 /* bail out if we need to do a full commit */
3109 if (btrfs_need_log_full_commit(trans)) {
3110 ret = -EAGAIN;
3111 mutex_unlock(&root->log_mutex);
3112 goto out;
3113 }
3114
3115 if (log_transid % 2 == 0)
3116 mark = EXTENT_DIRTY;
3117 else
3118 mark = EXTENT_NEW;
3119
3120 /* we start IO on all the marked extents here, but we don't actually
3121 * wait for them until later.
3122 */
3123 blk_start_plug(&plug);
3124 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3125 if (ret) {
3126 blk_finish_plug(&plug);
3127 btrfs_abort_transaction(trans, ret);
3128 btrfs_set_log_full_commit(trans);
3129 mutex_unlock(&root->log_mutex);
3130 goto out;
3131 }
3132
3133 /*
3134 * We _must_ update under the root->log_mutex in order to make sure we
3135 * have a consistent view of the log root we are trying to commit at
3136 * this moment.
3137 *
3138 * We _must_ copy this into a local copy, because we are not holding the
3139 * log_root_tree->log_mutex yet. This is important because when we
3140 * commit the log_root_tree we must have a consistent view of the
3141 * log_root_tree when we update the super block to point at the
3142 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3143 * with the commit and possibly point at the new block which we may not
3144 * have written out.
3145 */
3146 btrfs_set_root_node(&log->root_item, log->node);
3147 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3148
3149 root->log_transid++;
3150 log->log_transid = root->log_transid;
3151 root->log_start_pid = 0;
3152 /*
3153 * IO has been started, blocks of the log tree have WRITTEN flag set
3154 * in their headers. new modifications of the log will be written to
3155 * new positions. so it's safe to allow log writers to go in.
3156 */
3157 mutex_unlock(&root->log_mutex);
3158
3159 btrfs_init_log_ctx(&root_log_ctx, NULL);
3160
3161 mutex_lock(&log_root_tree->log_mutex);
3162
3163 index2 = log_root_tree->log_transid % 2;
3164 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3165 root_log_ctx.log_transid = log_root_tree->log_transid;
3166
3167 /*
3168 * Now we are safe to update the log_root_tree because we're under the
3169 * log_mutex, and we're a current writer so we're holding the commit
3170 * open until we drop the log_mutex.
3171 */
3172 ret = update_log_root(trans, log, &new_root_item);
3173 if (ret) {
3174 if (!list_empty(&root_log_ctx.list))
3175 list_del_init(&root_log_ctx.list);
3176
3177 blk_finish_plug(&plug);
3178 btrfs_set_log_full_commit(trans);
3179
3180 if (ret != -ENOSPC) {
3181 btrfs_abort_transaction(trans, ret);
3182 mutex_unlock(&log_root_tree->log_mutex);
3183 goto out;
3184 }
3185 btrfs_wait_tree_log_extents(log, mark);
3186 mutex_unlock(&log_root_tree->log_mutex);
3187 ret = -EAGAIN;
3188 goto out;
3189 }
3190
3191 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3192 blk_finish_plug(&plug);
3193 list_del_init(&root_log_ctx.list);
3194 mutex_unlock(&log_root_tree->log_mutex);
3195 ret = root_log_ctx.log_ret;
3196 goto out;
3197 }
3198
3199 index2 = root_log_ctx.log_transid % 2;
3200 if (atomic_read(&log_root_tree->log_commit[index2])) {
3201 blk_finish_plug(&plug);
3202 ret = btrfs_wait_tree_log_extents(log, mark);
3203 wait_log_commit(log_root_tree,
3204 root_log_ctx.log_transid);
3205 mutex_unlock(&log_root_tree->log_mutex);
3206 if (!ret)
3207 ret = root_log_ctx.log_ret;
3208 goto out;
3209 }
3210 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3211 atomic_set(&log_root_tree->log_commit[index2], 1);
3212
3213 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3214 wait_log_commit(log_root_tree,
3215 root_log_ctx.log_transid - 1);
3216 }
3217
3218 /*
3219 * now that we've moved on to the tree of log tree roots,
3220 * check the full commit flag again
3221 */
3222 if (btrfs_need_log_full_commit(trans)) {
3223 blk_finish_plug(&plug);
3224 btrfs_wait_tree_log_extents(log, mark);
3225 mutex_unlock(&log_root_tree->log_mutex);
3226 ret = -EAGAIN;
3227 goto out_wake_log_root;
3228 }
3229
3230 ret = btrfs_write_marked_extents(fs_info,
3231 &log_root_tree->dirty_log_pages,
3232 EXTENT_DIRTY | EXTENT_NEW);
3233 blk_finish_plug(&plug);
3234 if (ret) {
3235 btrfs_set_log_full_commit(trans);
3236 btrfs_abort_transaction(trans, ret);
3237 mutex_unlock(&log_root_tree->log_mutex);
3238 goto out_wake_log_root;
3239 }
3240 ret = btrfs_wait_tree_log_extents(log, mark);
3241 if (!ret)
3242 ret = btrfs_wait_tree_log_extents(log_root_tree,
3243 EXTENT_NEW | EXTENT_DIRTY);
3244 if (ret) {
3245 btrfs_set_log_full_commit(trans);
3246 mutex_unlock(&log_root_tree->log_mutex);
3247 goto out_wake_log_root;
3248 }
3249
3250 btrfs_set_super_log_root(fs_info->super_for_commit,
3251 log_root_tree->node->start);
3252 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3253 btrfs_header_level(log_root_tree->node));
3254
3255 log_root_tree->log_transid++;
3256 mutex_unlock(&log_root_tree->log_mutex);
3257
3258 /*
3259 * Nobody else is going to jump in and write the ctree
3260 * super here because the log_commit atomic below is protecting
3261 * us. We must be called with a transaction handle pinning
3262 * the running transaction open, so a full commit can't hop
3263 * in and cause problems either.
3264 */
3265 ret = write_all_supers(fs_info, 1);
3266 if (ret) {
3267 btrfs_set_log_full_commit(trans);
3268 btrfs_abort_transaction(trans, ret);
3269 goto out_wake_log_root;
3270 }
3271
3272 mutex_lock(&root->log_mutex);
3273 if (root->last_log_commit < log_transid)
3274 root->last_log_commit = log_transid;
3275 mutex_unlock(&root->log_mutex);
3276
3277 out_wake_log_root:
3278 mutex_lock(&log_root_tree->log_mutex);
3279 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3280
3281 log_root_tree->log_transid_committed++;
3282 atomic_set(&log_root_tree->log_commit[index2], 0);
3283 mutex_unlock(&log_root_tree->log_mutex);
3284
3285 /*
3286 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3287 * all the updates above are seen by the woken threads. It might not be
3288 * necessary, but proving that seems to be hard.
3289 */
3290 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3291 out:
3292 mutex_lock(&root->log_mutex);
3293 btrfs_remove_all_log_ctxs(root, index1, ret);
3294 root->log_transid_committed++;
3295 atomic_set(&root->log_commit[index1], 0);
3296 mutex_unlock(&root->log_mutex);
3297
3298 /*
3299 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3300 * all the updates above are seen by the woken threads. It might not be
3301 * necessary, but proving that seems to be hard.
3302 */
3303 cond_wake_up(&root->log_commit_wait[index1]);
3304 return ret;
3305 }
3306
free_log_tree(struct btrfs_trans_handle * trans,struct btrfs_root * log)3307 static void free_log_tree(struct btrfs_trans_handle *trans,
3308 struct btrfs_root *log)
3309 {
3310 int ret;
3311 struct walk_control wc = {
3312 .free = 1,
3313 .process_func = process_one_buffer
3314 };
3315
3316 ret = walk_log_tree(trans, log, &wc);
3317 if (ret) {
3318 if (trans)
3319 btrfs_abort_transaction(trans, ret);
3320 else
3321 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3322 }
3323
3324 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3325 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3326 extent_io_tree_release(&log->log_csum_range);
3327 btrfs_put_root(log);
3328 }
3329
3330 /*
3331 * free all the extents used by the tree log. This should be called
3332 * at commit time of the full transaction
3333 */
btrfs_free_log(struct btrfs_trans_handle * trans,struct btrfs_root * root)3334 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3335 {
3336 if (root->log_root) {
3337 free_log_tree(trans, root->log_root);
3338 root->log_root = NULL;
3339 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3340 }
3341 return 0;
3342 }
3343
btrfs_free_log_root_tree(struct btrfs_trans_handle * trans,struct btrfs_fs_info * fs_info)3344 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3345 struct btrfs_fs_info *fs_info)
3346 {
3347 if (fs_info->log_root_tree) {
3348 free_log_tree(trans, fs_info->log_root_tree);
3349 fs_info->log_root_tree = NULL;
3350 }
3351 return 0;
3352 }
3353
3354 /*
3355 * Check if an inode was logged in the current transaction. We can't always rely
3356 * on an inode's logged_trans value, because it's an in-memory only field and
3357 * therefore not persisted. This means that its value is lost if the inode gets
3358 * evicted and loaded again from disk (in which case it has a value of 0, and
3359 * certainly it is smaller then any possible transaction ID), when that happens
3360 * the full_sync flag is set in the inode's runtime flags, so on that case we
3361 * assume eviction happened and ignore the logged_trans value, assuming the
3362 * worst case, that the inode was logged before in the current transaction.
3363 */
inode_logged(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)3364 static bool inode_logged(struct btrfs_trans_handle *trans,
3365 struct btrfs_inode *inode)
3366 {
3367 if (inode->logged_trans == trans->transid)
3368 return true;
3369
3370 if (inode->last_trans == trans->transid &&
3371 test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
3372 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3373 return true;
3374
3375 return false;
3376 }
3377
3378 /*
3379 * If both a file and directory are logged, and unlinks or renames are
3380 * mixed in, we have a few interesting corners:
3381 *
3382 * create file X in dir Y
3383 * link file X to X.link in dir Y
3384 * fsync file X
3385 * unlink file X but leave X.link
3386 * fsync dir Y
3387 *
3388 * After a crash we would expect only X.link to exist. But file X
3389 * didn't get fsync'd again so the log has back refs for X and X.link.
3390 *
3391 * We solve this by removing directory entries and inode backrefs from the
3392 * log when a file that was logged in the current transaction is
3393 * unlinked. Any later fsync will include the updated log entries, and
3394 * we'll be able to reconstruct the proper directory items from backrefs.
3395 *
3396 * This optimizations allows us to avoid relogging the entire inode
3397 * or the entire directory.
3398 */
btrfs_del_dir_entries_in_log(struct btrfs_trans_handle * trans,struct btrfs_root * root,const char * name,int name_len,struct btrfs_inode * dir,u64 index)3399 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3400 struct btrfs_root *root,
3401 const char *name, int name_len,
3402 struct btrfs_inode *dir, u64 index)
3403 {
3404 struct btrfs_root *log;
3405 struct btrfs_dir_item *di;
3406 struct btrfs_path *path;
3407 int ret;
3408 int err = 0;
3409 int bytes_del = 0;
3410 u64 dir_ino = btrfs_ino(dir);
3411
3412 if (!inode_logged(trans, dir))
3413 return 0;
3414
3415 ret = join_running_log_trans(root);
3416 if (ret)
3417 return 0;
3418
3419 mutex_lock(&dir->log_mutex);
3420
3421 log = root->log_root;
3422 path = btrfs_alloc_path();
3423 if (!path) {
3424 err = -ENOMEM;
3425 goto out_unlock;
3426 }
3427
3428 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3429 name, name_len, -1);
3430 if (IS_ERR(di)) {
3431 err = PTR_ERR(di);
3432 goto fail;
3433 }
3434 if (di) {
3435 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3436 bytes_del += name_len;
3437 if (ret) {
3438 err = ret;
3439 goto fail;
3440 }
3441 }
3442 btrfs_release_path(path);
3443 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3444 index, name, name_len, -1);
3445 if (IS_ERR(di)) {
3446 err = PTR_ERR(di);
3447 goto fail;
3448 }
3449 if (di) {
3450 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3451 bytes_del += name_len;
3452 if (ret) {
3453 err = ret;
3454 goto fail;
3455 }
3456 }
3457
3458 /* update the directory size in the log to reflect the names
3459 * we have removed
3460 */
3461 if (bytes_del) {
3462 struct btrfs_key key;
3463
3464 key.objectid = dir_ino;
3465 key.offset = 0;
3466 key.type = BTRFS_INODE_ITEM_KEY;
3467 btrfs_release_path(path);
3468
3469 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3470 if (ret < 0) {
3471 err = ret;
3472 goto fail;
3473 }
3474 if (ret == 0) {
3475 struct btrfs_inode_item *item;
3476 u64 i_size;
3477
3478 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3479 struct btrfs_inode_item);
3480 i_size = btrfs_inode_size(path->nodes[0], item);
3481 if (i_size > bytes_del)
3482 i_size -= bytes_del;
3483 else
3484 i_size = 0;
3485 btrfs_set_inode_size(path->nodes[0], item, i_size);
3486 btrfs_mark_buffer_dirty(path->nodes[0]);
3487 } else
3488 ret = 0;
3489 btrfs_release_path(path);
3490 }
3491 fail:
3492 btrfs_free_path(path);
3493 out_unlock:
3494 mutex_unlock(&dir->log_mutex);
3495 if (err == -ENOSPC) {
3496 btrfs_set_log_full_commit(trans);
3497 err = 0;
3498 } else if (err < 0 && err != -ENOENT) {
3499 /* ENOENT can be returned if the entry hasn't been fsynced yet */
3500 btrfs_abort_transaction(trans, err);
3501 }
3502
3503 btrfs_end_log_trans(root);
3504
3505 return err;
3506 }
3507
3508 /* see comments for btrfs_del_dir_entries_in_log */
btrfs_del_inode_ref_in_log(struct btrfs_trans_handle * trans,struct btrfs_root * root,const char * name,int name_len,struct btrfs_inode * inode,u64 dirid)3509 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3510 struct btrfs_root *root,
3511 const char *name, int name_len,
3512 struct btrfs_inode *inode, u64 dirid)
3513 {
3514 struct btrfs_root *log;
3515 u64 index;
3516 int ret;
3517
3518 if (!inode_logged(trans, inode))
3519 return 0;
3520
3521 ret = join_running_log_trans(root);
3522 if (ret)
3523 return 0;
3524 log = root->log_root;
3525 mutex_lock(&inode->log_mutex);
3526
3527 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3528 dirid, &index);
3529 mutex_unlock(&inode->log_mutex);
3530 if (ret == -ENOSPC) {
3531 btrfs_set_log_full_commit(trans);
3532 ret = 0;
3533 } else if (ret < 0 && ret != -ENOENT)
3534 btrfs_abort_transaction(trans, ret);
3535 btrfs_end_log_trans(root);
3536
3537 return ret;
3538 }
3539
3540 /*
3541 * creates a range item in the log for 'dirid'. first_offset and
3542 * last_offset tell us which parts of the key space the log should
3543 * be considered authoritative for.
3544 */
insert_dir_log_key(struct btrfs_trans_handle * trans,struct btrfs_root * log,struct btrfs_path * path,int key_type,u64 dirid,u64 first_offset,u64 last_offset)3545 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3546 struct btrfs_root *log,
3547 struct btrfs_path *path,
3548 int key_type, u64 dirid,
3549 u64 first_offset, u64 last_offset)
3550 {
3551 int ret;
3552 struct btrfs_key key;
3553 struct btrfs_dir_log_item *item;
3554
3555 key.objectid = dirid;
3556 key.offset = first_offset;
3557 if (key_type == BTRFS_DIR_ITEM_KEY)
3558 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3559 else
3560 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3561 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3562 if (ret)
3563 return ret;
3564
3565 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3566 struct btrfs_dir_log_item);
3567 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3568 btrfs_mark_buffer_dirty(path->nodes[0]);
3569 btrfs_release_path(path);
3570 return 0;
3571 }
3572
3573 /*
3574 * log all the items included in the current transaction for a given
3575 * directory. This also creates the range items in the log tree required
3576 * to replay anything deleted before the fsync
3577 */
log_dir_items(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode,struct btrfs_path * path,struct btrfs_path * dst_path,int key_type,struct btrfs_log_ctx * ctx,u64 min_offset,u64 * last_offset_ret)3578 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3579 struct btrfs_root *root, struct btrfs_inode *inode,
3580 struct btrfs_path *path,
3581 struct btrfs_path *dst_path, int key_type,
3582 struct btrfs_log_ctx *ctx,
3583 u64 min_offset, u64 *last_offset_ret)
3584 {
3585 struct btrfs_key min_key;
3586 struct btrfs_root *log = root->log_root;
3587 struct extent_buffer *src;
3588 int err = 0;
3589 int ret;
3590 int i;
3591 int nritems;
3592 u64 first_offset = min_offset;
3593 u64 last_offset = (u64)-1;
3594 u64 ino = btrfs_ino(inode);
3595
3596 log = root->log_root;
3597
3598 min_key.objectid = ino;
3599 min_key.type = key_type;
3600 min_key.offset = min_offset;
3601
3602 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3603
3604 /*
3605 * we didn't find anything from this transaction, see if there
3606 * is anything at all
3607 */
3608 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3609 min_key.objectid = ino;
3610 min_key.type = key_type;
3611 min_key.offset = (u64)-1;
3612 btrfs_release_path(path);
3613 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3614 if (ret < 0) {
3615 btrfs_release_path(path);
3616 return ret;
3617 }
3618 ret = btrfs_previous_item(root, path, ino, key_type);
3619
3620 /* if ret == 0 there are items for this type,
3621 * create a range to tell us the last key of this type.
3622 * otherwise, there are no items in this directory after
3623 * *min_offset, and we create a range to indicate that.
3624 */
3625 if (ret == 0) {
3626 struct btrfs_key tmp;
3627 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3628 path->slots[0]);
3629 if (key_type == tmp.type)
3630 first_offset = max(min_offset, tmp.offset) + 1;
3631 }
3632 goto done;
3633 }
3634
3635 /* go backward to find any previous key */
3636 ret = btrfs_previous_item(root, path, ino, key_type);
3637 if (ret == 0) {
3638 struct btrfs_key tmp;
3639 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3640 if (key_type == tmp.type) {
3641 first_offset = tmp.offset;
3642 ret = overwrite_item(trans, log, dst_path,
3643 path->nodes[0], path->slots[0],
3644 &tmp);
3645 if (ret) {
3646 err = ret;
3647 goto done;
3648 }
3649 }
3650 }
3651 btrfs_release_path(path);
3652
3653 /*
3654 * Find the first key from this transaction again. See the note for
3655 * log_new_dir_dentries, if we're logging a directory recursively we
3656 * won't be holding its i_mutex, which means we can modify the directory
3657 * while we're logging it. If we remove an entry between our first
3658 * search and this search we'll not find the key again and can just
3659 * bail.
3660 */
3661 search:
3662 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3663 if (ret != 0)
3664 goto done;
3665
3666 /*
3667 * we have a block from this transaction, log every item in it
3668 * from our directory
3669 */
3670 while (1) {
3671 struct btrfs_key tmp;
3672 src = path->nodes[0];
3673 nritems = btrfs_header_nritems(src);
3674 for (i = path->slots[0]; i < nritems; i++) {
3675 struct btrfs_dir_item *di;
3676
3677 btrfs_item_key_to_cpu(src, &min_key, i);
3678
3679 if (min_key.objectid != ino || min_key.type != key_type)
3680 goto done;
3681
3682 if (need_resched()) {
3683 btrfs_release_path(path);
3684 cond_resched();
3685 goto search;
3686 }
3687
3688 ret = overwrite_item(trans, log, dst_path, src, i,
3689 &min_key);
3690 if (ret) {
3691 err = ret;
3692 goto done;
3693 }
3694
3695 /*
3696 * We must make sure that when we log a directory entry,
3697 * the corresponding inode, after log replay, has a
3698 * matching link count. For example:
3699 *
3700 * touch foo
3701 * mkdir mydir
3702 * sync
3703 * ln foo mydir/bar
3704 * xfs_io -c "fsync" mydir
3705 * <crash>
3706 * <mount fs and log replay>
3707 *
3708 * Would result in a fsync log that when replayed, our
3709 * file inode would have a link count of 1, but we get
3710 * two directory entries pointing to the same inode.
3711 * After removing one of the names, it would not be
3712 * possible to remove the other name, which resulted
3713 * always in stale file handle errors, and would not
3714 * be possible to rmdir the parent directory, since
3715 * its i_size could never decrement to the value
3716 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3717 */
3718 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3719 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3720 if (ctx &&
3721 (btrfs_dir_transid(src, di) == trans->transid ||
3722 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3723 tmp.type != BTRFS_ROOT_ITEM_KEY)
3724 ctx->log_new_dentries = true;
3725 }
3726 path->slots[0] = nritems;
3727
3728 /*
3729 * look ahead to the next item and see if it is also
3730 * from this directory and from this transaction
3731 */
3732 ret = btrfs_next_leaf(root, path);
3733 if (ret) {
3734 if (ret == 1)
3735 last_offset = (u64)-1;
3736 else
3737 err = ret;
3738 goto done;
3739 }
3740 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3741 if (tmp.objectid != ino || tmp.type != key_type) {
3742 last_offset = (u64)-1;
3743 goto done;
3744 }
3745 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3746 ret = overwrite_item(trans, log, dst_path,
3747 path->nodes[0], path->slots[0],
3748 &tmp);
3749 if (ret)
3750 err = ret;
3751 else
3752 last_offset = tmp.offset;
3753 goto done;
3754 }
3755 }
3756 done:
3757 btrfs_release_path(path);
3758 btrfs_release_path(dst_path);
3759
3760 if (err == 0) {
3761 *last_offset_ret = last_offset;
3762 /*
3763 * insert the log range keys to indicate where the log
3764 * is valid
3765 */
3766 ret = insert_dir_log_key(trans, log, path, key_type,
3767 ino, first_offset, last_offset);
3768 if (ret)
3769 err = ret;
3770 }
3771 return err;
3772 }
3773
3774 /*
3775 * logging directories is very similar to logging inodes, We find all the items
3776 * from the current transaction and write them to the log.
3777 *
3778 * The recovery code scans the directory in the subvolume, and if it finds a
3779 * key in the range logged that is not present in the log tree, then it means
3780 * that dir entry was unlinked during the transaction.
3781 *
3782 * In order for that scan to work, we must include one key smaller than
3783 * the smallest logged by this transaction and one key larger than the largest
3784 * key logged by this transaction.
3785 */
log_directory_changes(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode,struct btrfs_path * path,struct btrfs_path * dst_path,struct btrfs_log_ctx * ctx)3786 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3787 struct btrfs_root *root, struct btrfs_inode *inode,
3788 struct btrfs_path *path,
3789 struct btrfs_path *dst_path,
3790 struct btrfs_log_ctx *ctx)
3791 {
3792 u64 min_key;
3793 u64 max_key;
3794 int ret;
3795 int key_type = BTRFS_DIR_ITEM_KEY;
3796
3797 again:
3798 min_key = 0;
3799 max_key = 0;
3800 while (1) {
3801 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3802 ctx, min_key, &max_key);
3803 if (ret)
3804 return ret;
3805 if (max_key == (u64)-1)
3806 break;
3807 min_key = max_key + 1;
3808 }
3809
3810 if (key_type == BTRFS_DIR_ITEM_KEY) {
3811 key_type = BTRFS_DIR_INDEX_KEY;
3812 goto again;
3813 }
3814 return 0;
3815 }
3816
3817 /*
3818 * a helper function to drop items from the log before we relog an
3819 * inode. max_key_type indicates the highest item type to remove.
3820 * This cannot be run for file data extents because it does not
3821 * free the extents they point to.
3822 */
drop_objectid_items(struct btrfs_trans_handle * trans,struct btrfs_root * log,struct btrfs_path * path,u64 objectid,int max_key_type)3823 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3824 struct btrfs_root *log,
3825 struct btrfs_path *path,
3826 u64 objectid, int max_key_type)
3827 {
3828 int ret;
3829 struct btrfs_key key;
3830 struct btrfs_key found_key;
3831 int start_slot;
3832
3833 key.objectid = objectid;
3834 key.type = max_key_type;
3835 key.offset = (u64)-1;
3836
3837 while (1) {
3838 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3839 BUG_ON(ret == 0); /* Logic error */
3840 if (ret < 0)
3841 break;
3842
3843 if (path->slots[0] == 0)
3844 break;
3845
3846 path->slots[0]--;
3847 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3848 path->slots[0]);
3849
3850 if (found_key.objectid != objectid)
3851 break;
3852
3853 found_key.offset = 0;
3854 found_key.type = 0;
3855 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
3856 if (ret < 0)
3857 break;
3858
3859 ret = btrfs_del_items(trans, log, path, start_slot,
3860 path->slots[0] - start_slot + 1);
3861 /*
3862 * If start slot isn't 0 then we don't need to re-search, we've
3863 * found the last guy with the objectid in this tree.
3864 */
3865 if (ret || start_slot != 0)
3866 break;
3867 btrfs_release_path(path);
3868 }
3869 btrfs_release_path(path);
3870 if (ret > 0)
3871 ret = 0;
3872 return ret;
3873 }
3874
fill_inode_item(struct btrfs_trans_handle * trans,struct extent_buffer * leaf,struct btrfs_inode_item * item,struct inode * inode,int log_inode_only,u64 logged_isize)3875 static void fill_inode_item(struct btrfs_trans_handle *trans,
3876 struct extent_buffer *leaf,
3877 struct btrfs_inode_item *item,
3878 struct inode *inode, int log_inode_only,
3879 u64 logged_isize)
3880 {
3881 struct btrfs_map_token token;
3882
3883 btrfs_init_map_token(&token, leaf);
3884
3885 if (log_inode_only) {
3886 /* set the generation to zero so the recover code
3887 * can tell the difference between an logging
3888 * just to say 'this inode exists' and a logging
3889 * to say 'update this inode with these values'
3890 */
3891 btrfs_set_token_inode_generation(&token, item, 0);
3892 btrfs_set_token_inode_size(&token, item, logged_isize);
3893 } else {
3894 btrfs_set_token_inode_generation(&token, item,
3895 BTRFS_I(inode)->generation);
3896 btrfs_set_token_inode_size(&token, item, inode->i_size);
3897 }
3898
3899 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3900 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3901 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3902 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3903
3904 btrfs_set_token_timespec_sec(&token, &item->atime,
3905 inode->i_atime.tv_sec);
3906 btrfs_set_token_timespec_nsec(&token, &item->atime,
3907 inode->i_atime.tv_nsec);
3908
3909 btrfs_set_token_timespec_sec(&token, &item->mtime,
3910 inode->i_mtime.tv_sec);
3911 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3912 inode->i_mtime.tv_nsec);
3913
3914 btrfs_set_token_timespec_sec(&token, &item->ctime,
3915 inode->i_ctime.tv_sec);
3916 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3917 inode->i_ctime.tv_nsec);
3918
3919 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3920
3921 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3922 btrfs_set_token_inode_transid(&token, item, trans->transid);
3923 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3924 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3925 btrfs_set_token_inode_block_group(&token, item, 0);
3926 }
3927
log_inode_item(struct btrfs_trans_handle * trans,struct btrfs_root * log,struct btrfs_path * path,struct btrfs_inode * inode)3928 static int log_inode_item(struct btrfs_trans_handle *trans,
3929 struct btrfs_root *log, struct btrfs_path *path,
3930 struct btrfs_inode *inode)
3931 {
3932 struct btrfs_inode_item *inode_item;
3933 int ret;
3934
3935 ret = btrfs_insert_empty_item(trans, log, path,
3936 &inode->location, sizeof(*inode_item));
3937 if (ret && ret != -EEXIST)
3938 return ret;
3939 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3940 struct btrfs_inode_item);
3941 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3942 0, 0);
3943 btrfs_release_path(path);
3944 return 0;
3945 }
3946
log_csums(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct btrfs_root * log_root,struct btrfs_ordered_sum * sums)3947 static int log_csums(struct btrfs_trans_handle *trans,
3948 struct btrfs_inode *inode,
3949 struct btrfs_root *log_root,
3950 struct btrfs_ordered_sum *sums)
3951 {
3952 const u64 lock_end = sums->bytenr + sums->len - 1;
3953 struct extent_state *cached_state = NULL;
3954 int ret;
3955
3956 /*
3957 * If this inode was not used for reflink operations in the current
3958 * transaction with new extents, then do the fast path, no need to
3959 * worry about logging checksum items with overlapping ranges.
3960 */
3961 if (inode->last_reflink_trans < trans->transid)
3962 return btrfs_csum_file_blocks(trans, log_root, sums);
3963
3964 /*
3965 * Serialize logging for checksums. This is to avoid racing with the
3966 * same checksum being logged by another task that is logging another
3967 * file which happens to refer to the same extent as well. Such races
3968 * can leave checksum items in the log with overlapping ranges.
3969 */
3970 ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr,
3971 lock_end, &cached_state);
3972 if (ret)
3973 return ret;
3974 /*
3975 * Due to extent cloning, we might have logged a csum item that covers a
3976 * subrange of a cloned extent, and later we can end up logging a csum
3977 * item for a larger subrange of the same extent or the entire range.
3978 * This would leave csum items in the log tree that cover the same range
3979 * and break the searches for checksums in the log tree, resulting in
3980 * some checksums missing in the fs/subvolume tree. So just delete (or
3981 * trim and adjust) any existing csum items in the log for this range.
3982 */
3983 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
3984 if (!ret)
3985 ret = btrfs_csum_file_blocks(trans, log_root, sums);
3986
3987 unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end,
3988 &cached_state);
3989
3990 return ret;
3991 }
3992
copy_items(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct btrfs_path * dst_path,struct btrfs_path * src_path,int start_slot,int nr,int inode_only,u64 logged_isize)3993 static noinline int copy_items(struct btrfs_trans_handle *trans,
3994 struct btrfs_inode *inode,
3995 struct btrfs_path *dst_path,
3996 struct btrfs_path *src_path,
3997 int start_slot, int nr, int inode_only,
3998 u64 logged_isize)
3999 {
4000 struct btrfs_fs_info *fs_info = trans->fs_info;
4001 unsigned long src_offset;
4002 unsigned long dst_offset;
4003 struct btrfs_root *log = inode->root->log_root;
4004 struct btrfs_file_extent_item *extent;
4005 struct btrfs_inode_item *inode_item;
4006 struct extent_buffer *src = src_path->nodes[0];
4007 int ret;
4008 struct btrfs_key *ins_keys;
4009 u32 *ins_sizes;
4010 char *ins_data;
4011 int i;
4012 struct list_head ordered_sums;
4013 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
4014
4015 INIT_LIST_HEAD(&ordered_sums);
4016
4017 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4018 nr * sizeof(u32), GFP_NOFS);
4019 if (!ins_data)
4020 return -ENOMEM;
4021
4022 ins_sizes = (u32 *)ins_data;
4023 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4024
4025 for (i = 0; i < nr; i++) {
4026 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
4027 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
4028 }
4029 ret = btrfs_insert_empty_items(trans, log, dst_path,
4030 ins_keys, ins_sizes, nr);
4031 if (ret) {
4032 kfree(ins_data);
4033 return ret;
4034 }
4035
4036 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
4037 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
4038 dst_path->slots[0]);
4039
4040 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
4041
4042 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
4043 inode_item = btrfs_item_ptr(dst_path->nodes[0],
4044 dst_path->slots[0],
4045 struct btrfs_inode_item);
4046 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4047 &inode->vfs_inode,
4048 inode_only == LOG_INODE_EXISTS,
4049 logged_isize);
4050 } else {
4051 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4052 src_offset, ins_sizes[i]);
4053 }
4054
4055 /* take a reference on file data extents so that truncates
4056 * or deletes of this inode don't have to relog the inode
4057 * again
4058 */
4059 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4060 !skip_csum) {
4061 int found_type;
4062 extent = btrfs_item_ptr(src, start_slot + i,
4063 struct btrfs_file_extent_item);
4064
4065 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4066 continue;
4067
4068 found_type = btrfs_file_extent_type(src, extent);
4069 if (found_type == BTRFS_FILE_EXTENT_REG) {
4070 u64 ds, dl, cs, cl;
4071 ds = btrfs_file_extent_disk_bytenr(src,
4072 extent);
4073 /* ds == 0 is a hole */
4074 if (ds == 0)
4075 continue;
4076
4077 dl = btrfs_file_extent_disk_num_bytes(src,
4078 extent);
4079 cs = btrfs_file_extent_offset(src, extent);
4080 cl = btrfs_file_extent_num_bytes(src,
4081 extent);
4082 if (btrfs_file_extent_compression(src,
4083 extent)) {
4084 cs = 0;
4085 cl = dl;
4086 }
4087
4088 ret = btrfs_lookup_csums_range(
4089 fs_info->csum_root,
4090 ds + cs, ds + cs + cl - 1,
4091 &ordered_sums, 0);
4092 if (ret)
4093 break;
4094 }
4095 }
4096 }
4097
4098 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4099 btrfs_release_path(dst_path);
4100 kfree(ins_data);
4101
4102 /*
4103 * we have to do this after the loop above to avoid changing the
4104 * log tree while trying to change the log tree.
4105 */
4106 while (!list_empty(&ordered_sums)) {
4107 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4108 struct btrfs_ordered_sum,
4109 list);
4110 if (!ret)
4111 ret = log_csums(trans, inode, log, sums);
4112 list_del(&sums->list);
4113 kfree(sums);
4114 }
4115
4116 return ret;
4117 }
4118
extent_cmp(void * priv,struct list_head * a,struct list_head * b)4119 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4120 {
4121 struct extent_map *em1, *em2;
4122
4123 em1 = list_entry(a, struct extent_map, list);
4124 em2 = list_entry(b, struct extent_map, list);
4125
4126 if (em1->start < em2->start)
4127 return -1;
4128 else if (em1->start > em2->start)
4129 return 1;
4130 return 0;
4131 }
4132
log_extent_csums(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct btrfs_root * log_root,const struct extent_map * em,struct btrfs_log_ctx * ctx)4133 static int log_extent_csums(struct btrfs_trans_handle *trans,
4134 struct btrfs_inode *inode,
4135 struct btrfs_root *log_root,
4136 const struct extent_map *em,
4137 struct btrfs_log_ctx *ctx)
4138 {
4139 struct btrfs_ordered_extent *ordered;
4140 u64 csum_offset;
4141 u64 csum_len;
4142 u64 mod_start = em->mod_start;
4143 u64 mod_len = em->mod_len;
4144 LIST_HEAD(ordered_sums);
4145 int ret = 0;
4146
4147 if (inode->flags & BTRFS_INODE_NODATASUM ||
4148 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4149 em->block_start == EXTENT_MAP_HOLE)
4150 return 0;
4151
4152 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4153 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4154 const u64 mod_end = mod_start + mod_len;
4155 struct btrfs_ordered_sum *sums;
4156
4157 if (mod_len == 0)
4158 break;
4159
4160 if (ordered_end <= mod_start)
4161 continue;
4162 if (mod_end <= ordered->file_offset)
4163 break;
4164
4165 /*
4166 * We are going to copy all the csums on this ordered extent, so
4167 * go ahead and adjust mod_start and mod_len in case this ordered
4168 * extent has already been logged.
4169 */
4170 if (ordered->file_offset > mod_start) {
4171 if (ordered_end >= mod_end)
4172 mod_len = ordered->file_offset - mod_start;
4173 /*
4174 * If we have this case
4175 *
4176 * |--------- logged extent ---------|
4177 * |----- ordered extent ----|
4178 *
4179 * Just don't mess with mod_start and mod_len, we'll
4180 * just end up logging more csums than we need and it
4181 * will be ok.
4182 */
4183 } else {
4184 if (ordered_end < mod_end) {
4185 mod_len = mod_end - ordered_end;
4186 mod_start = ordered_end;
4187 } else {
4188 mod_len = 0;
4189 }
4190 }
4191
4192 /*
4193 * To keep us from looping for the above case of an ordered
4194 * extent that falls inside of the logged extent.
4195 */
4196 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4197 continue;
4198
4199 list_for_each_entry(sums, &ordered->list, list) {
4200 ret = log_csums(trans, inode, log_root, sums);
4201 if (ret)
4202 return ret;
4203 }
4204 }
4205
4206 /* We're done, found all csums in the ordered extents. */
4207 if (mod_len == 0)
4208 return 0;
4209
4210 /* If we're compressed we have to save the entire range of csums. */
4211 if (em->compress_type) {
4212 csum_offset = 0;
4213 csum_len = max(em->block_len, em->orig_block_len);
4214 } else {
4215 csum_offset = mod_start - em->start;
4216 csum_len = mod_len;
4217 }
4218
4219 /* block start is already adjusted for the file extent offset. */
4220 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4221 em->block_start + csum_offset,
4222 em->block_start + csum_offset +
4223 csum_len - 1, &ordered_sums, 0);
4224 if (ret)
4225 return ret;
4226
4227 while (!list_empty(&ordered_sums)) {
4228 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4229 struct btrfs_ordered_sum,
4230 list);
4231 if (!ret)
4232 ret = log_csums(trans, inode, log_root, sums);
4233 list_del(&sums->list);
4234 kfree(sums);
4235 }
4236
4237 return ret;
4238 }
4239
log_one_extent(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct btrfs_root * root,const struct extent_map * em,struct btrfs_path * path,struct btrfs_log_ctx * ctx)4240 static int log_one_extent(struct btrfs_trans_handle *trans,
4241 struct btrfs_inode *inode, struct btrfs_root *root,
4242 const struct extent_map *em,
4243 struct btrfs_path *path,
4244 struct btrfs_log_ctx *ctx)
4245 {
4246 struct btrfs_root *log = root->log_root;
4247 struct btrfs_file_extent_item *fi;
4248 struct extent_buffer *leaf;
4249 struct btrfs_map_token token;
4250 struct btrfs_key key;
4251 u64 extent_offset = em->start - em->orig_start;
4252 u64 block_len;
4253 int ret;
4254 int extent_inserted = 0;
4255
4256 ret = log_extent_csums(trans, inode, log, em, ctx);
4257 if (ret)
4258 return ret;
4259
4260 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
4261 em->start + em->len, NULL, 0, 1,
4262 sizeof(*fi), &extent_inserted);
4263 if (ret)
4264 return ret;
4265
4266 if (!extent_inserted) {
4267 key.objectid = btrfs_ino(inode);
4268 key.type = BTRFS_EXTENT_DATA_KEY;
4269 key.offset = em->start;
4270
4271 ret = btrfs_insert_empty_item(trans, log, path, &key,
4272 sizeof(*fi));
4273 if (ret)
4274 return ret;
4275 }
4276 leaf = path->nodes[0];
4277 btrfs_init_map_token(&token, leaf);
4278 fi = btrfs_item_ptr(leaf, path->slots[0],
4279 struct btrfs_file_extent_item);
4280
4281 btrfs_set_token_file_extent_generation(&token, fi, trans->transid);
4282 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4283 btrfs_set_token_file_extent_type(&token, fi,
4284 BTRFS_FILE_EXTENT_PREALLOC);
4285 else
4286 btrfs_set_token_file_extent_type(&token, fi,
4287 BTRFS_FILE_EXTENT_REG);
4288
4289 block_len = max(em->block_len, em->orig_block_len);
4290 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4291 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4292 em->block_start);
4293 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4294 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4295 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4296 em->block_start -
4297 extent_offset);
4298 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4299 } else {
4300 btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0);
4301 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0);
4302 }
4303
4304 btrfs_set_token_file_extent_offset(&token, fi, extent_offset);
4305 btrfs_set_token_file_extent_num_bytes(&token, fi, em->len);
4306 btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes);
4307 btrfs_set_token_file_extent_compression(&token, fi, em->compress_type);
4308 btrfs_set_token_file_extent_encryption(&token, fi, 0);
4309 btrfs_set_token_file_extent_other_encoding(&token, fi, 0);
4310 btrfs_mark_buffer_dirty(leaf);
4311
4312 btrfs_release_path(path);
4313
4314 return ret;
4315 }
4316
4317 /*
4318 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4319 * lose them after doing a full/fast fsync and replaying the log. We scan the
4320 * subvolume's root instead of iterating the inode's extent map tree because
4321 * otherwise we can log incorrect extent items based on extent map conversion.
4322 * That can happen due to the fact that extent maps are merged when they
4323 * are not in the extent map tree's list of modified extents.
4324 */
btrfs_log_prealloc_extents(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct btrfs_path * path)4325 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4326 struct btrfs_inode *inode,
4327 struct btrfs_path *path)
4328 {
4329 struct btrfs_root *root = inode->root;
4330 struct btrfs_key key;
4331 const u64 i_size = i_size_read(&inode->vfs_inode);
4332 const u64 ino = btrfs_ino(inode);
4333 struct btrfs_path *dst_path = NULL;
4334 bool dropped_extents = false;
4335 u64 truncate_offset = i_size;
4336 struct extent_buffer *leaf;
4337 int slot;
4338 int ins_nr = 0;
4339 int start_slot;
4340 int ret;
4341
4342 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4343 return 0;
4344
4345 key.objectid = ino;
4346 key.type = BTRFS_EXTENT_DATA_KEY;
4347 key.offset = i_size;
4348 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4349 if (ret < 0)
4350 goto out;
4351
4352 /*
4353 * We must check if there is a prealloc extent that starts before the
4354 * i_size and crosses the i_size boundary. This is to ensure later we
4355 * truncate down to the end of that extent and not to the i_size, as
4356 * otherwise we end up losing part of the prealloc extent after a log
4357 * replay and with an implicit hole if there is another prealloc extent
4358 * that starts at an offset beyond i_size.
4359 */
4360 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4361 if (ret < 0)
4362 goto out;
4363
4364 if (ret == 0) {
4365 struct btrfs_file_extent_item *ei;
4366
4367 leaf = path->nodes[0];
4368 slot = path->slots[0];
4369 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4370
4371 if (btrfs_file_extent_type(leaf, ei) ==
4372 BTRFS_FILE_EXTENT_PREALLOC) {
4373 u64 extent_end;
4374
4375 btrfs_item_key_to_cpu(leaf, &key, slot);
4376 extent_end = key.offset +
4377 btrfs_file_extent_num_bytes(leaf, ei);
4378
4379 if (extent_end > i_size)
4380 truncate_offset = extent_end;
4381 }
4382 } else {
4383 ret = 0;
4384 }
4385
4386 while (true) {
4387 leaf = path->nodes[0];
4388 slot = path->slots[0];
4389
4390 if (slot >= btrfs_header_nritems(leaf)) {
4391 if (ins_nr > 0) {
4392 ret = copy_items(trans, inode, dst_path, path,
4393 start_slot, ins_nr, 1, 0);
4394 if (ret < 0)
4395 goto out;
4396 ins_nr = 0;
4397 }
4398 ret = btrfs_next_leaf(root, path);
4399 if (ret < 0)
4400 goto out;
4401 if (ret > 0) {
4402 ret = 0;
4403 break;
4404 }
4405 continue;
4406 }
4407
4408 btrfs_item_key_to_cpu(leaf, &key, slot);
4409 if (key.objectid > ino)
4410 break;
4411 if (WARN_ON_ONCE(key.objectid < ino) ||
4412 key.type < BTRFS_EXTENT_DATA_KEY ||
4413 key.offset < i_size) {
4414 path->slots[0]++;
4415 continue;
4416 }
4417 if (!dropped_extents) {
4418 /*
4419 * Avoid logging extent items logged in past fsync calls
4420 * and leading to duplicate keys in the log tree.
4421 */
4422 do {
4423 ret = btrfs_truncate_inode_items(trans,
4424 root->log_root,
4425 &inode->vfs_inode,
4426 truncate_offset,
4427 BTRFS_EXTENT_DATA_KEY);
4428 } while (ret == -EAGAIN);
4429 if (ret)
4430 goto out;
4431 dropped_extents = true;
4432 }
4433 if (ins_nr == 0)
4434 start_slot = slot;
4435 ins_nr++;
4436 path->slots[0]++;
4437 if (!dst_path) {
4438 dst_path = btrfs_alloc_path();
4439 if (!dst_path) {
4440 ret = -ENOMEM;
4441 goto out;
4442 }
4443 }
4444 }
4445 if (ins_nr > 0)
4446 ret = copy_items(trans, inode, dst_path, path,
4447 start_slot, ins_nr, 1, 0);
4448 out:
4449 btrfs_release_path(path);
4450 btrfs_free_path(dst_path);
4451 return ret;
4452 }
4453
btrfs_log_changed_extents(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode,struct btrfs_path * path,struct btrfs_log_ctx * ctx)4454 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4455 struct btrfs_root *root,
4456 struct btrfs_inode *inode,
4457 struct btrfs_path *path,
4458 struct btrfs_log_ctx *ctx)
4459 {
4460 struct btrfs_ordered_extent *ordered;
4461 struct btrfs_ordered_extent *tmp;
4462 struct extent_map *em, *n;
4463 struct list_head extents;
4464 struct extent_map_tree *tree = &inode->extent_tree;
4465 u64 test_gen;
4466 int ret = 0;
4467 int num = 0;
4468
4469 INIT_LIST_HEAD(&extents);
4470
4471 write_lock(&tree->lock);
4472 test_gen = root->fs_info->last_trans_committed;
4473
4474 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4475 list_del_init(&em->list);
4476 /*
4477 * Just an arbitrary number, this can be really CPU intensive
4478 * once we start getting a lot of extents, and really once we
4479 * have a bunch of extents we just want to commit since it will
4480 * be faster.
4481 */
4482 if (++num > 32768) {
4483 list_del_init(&tree->modified_extents);
4484 ret = -EFBIG;
4485 goto process;
4486 }
4487
4488 if (em->generation <= test_gen)
4489 continue;
4490
4491 /* We log prealloc extents beyond eof later. */
4492 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4493 em->start >= i_size_read(&inode->vfs_inode))
4494 continue;
4495
4496 /* Need a ref to keep it from getting evicted from cache */
4497 refcount_inc(&em->refs);
4498 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4499 list_add_tail(&em->list, &extents);
4500 num++;
4501 }
4502
4503 list_sort(NULL, &extents, extent_cmp);
4504 process:
4505 while (!list_empty(&extents)) {
4506 em = list_entry(extents.next, struct extent_map, list);
4507
4508 list_del_init(&em->list);
4509
4510 /*
4511 * If we had an error we just need to delete everybody from our
4512 * private list.
4513 */
4514 if (ret) {
4515 clear_em_logging(tree, em);
4516 free_extent_map(em);
4517 continue;
4518 }
4519
4520 write_unlock(&tree->lock);
4521
4522 ret = log_one_extent(trans, inode, root, em, path, ctx);
4523 write_lock(&tree->lock);
4524 clear_em_logging(tree, em);
4525 free_extent_map(em);
4526 }
4527 WARN_ON(!list_empty(&extents));
4528 write_unlock(&tree->lock);
4529
4530 btrfs_release_path(path);
4531 if (!ret)
4532 ret = btrfs_log_prealloc_extents(trans, inode, path);
4533 if (ret)
4534 return ret;
4535
4536 /*
4537 * We have logged all extents successfully, now make sure the commit of
4538 * the current transaction waits for the ordered extents to complete
4539 * before it commits and wipes out the log trees, otherwise we would
4540 * lose data if an ordered extents completes after the transaction
4541 * commits and a power failure happens after the transaction commit.
4542 */
4543 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4544 list_del_init(&ordered->log_list);
4545 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4546
4547 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4548 spin_lock_irq(&inode->ordered_tree.lock);
4549 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4550 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4551 atomic_inc(&trans->transaction->pending_ordered);
4552 }
4553 spin_unlock_irq(&inode->ordered_tree.lock);
4554 }
4555 btrfs_put_ordered_extent(ordered);
4556 }
4557
4558 return 0;
4559 }
4560
logged_inode_size(struct btrfs_root * log,struct btrfs_inode * inode,struct btrfs_path * path,u64 * size_ret)4561 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4562 struct btrfs_path *path, u64 *size_ret)
4563 {
4564 struct btrfs_key key;
4565 int ret;
4566
4567 key.objectid = btrfs_ino(inode);
4568 key.type = BTRFS_INODE_ITEM_KEY;
4569 key.offset = 0;
4570
4571 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4572 if (ret < 0) {
4573 return ret;
4574 } else if (ret > 0) {
4575 *size_ret = 0;
4576 } else {
4577 struct btrfs_inode_item *item;
4578
4579 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4580 struct btrfs_inode_item);
4581 *size_ret = btrfs_inode_size(path->nodes[0], item);
4582 /*
4583 * If the in-memory inode's i_size is smaller then the inode
4584 * size stored in the btree, return the inode's i_size, so
4585 * that we get a correct inode size after replaying the log
4586 * when before a power failure we had a shrinking truncate
4587 * followed by addition of a new name (rename / new hard link).
4588 * Otherwise return the inode size from the btree, to avoid
4589 * data loss when replaying a log due to previously doing a
4590 * write that expands the inode's size and logging a new name
4591 * immediately after.
4592 */
4593 if (*size_ret > inode->vfs_inode.i_size)
4594 *size_ret = inode->vfs_inode.i_size;
4595 }
4596
4597 btrfs_release_path(path);
4598 return 0;
4599 }
4600
4601 /*
4602 * At the moment we always log all xattrs. This is to figure out at log replay
4603 * time which xattrs must have their deletion replayed. If a xattr is missing
4604 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4605 * because if a xattr is deleted, the inode is fsynced and a power failure
4606 * happens, causing the log to be replayed the next time the fs is mounted,
4607 * we want the xattr to not exist anymore (same behaviour as other filesystems
4608 * with a journal, ext3/4, xfs, f2fs, etc).
4609 */
btrfs_log_all_xattrs(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode,struct btrfs_path * path,struct btrfs_path * dst_path)4610 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4611 struct btrfs_root *root,
4612 struct btrfs_inode *inode,
4613 struct btrfs_path *path,
4614 struct btrfs_path *dst_path)
4615 {
4616 int ret;
4617 struct btrfs_key key;
4618 const u64 ino = btrfs_ino(inode);
4619 int ins_nr = 0;
4620 int start_slot = 0;
4621 bool found_xattrs = false;
4622
4623 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
4624 return 0;
4625
4626 key.objectid = ino;
4627 key.type = BTRFS_XATTR_ITEM_KEY;
4628 key.offset = 0;
4629
4630 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4631 if (ret < 0)
4632 return ret;
4633
4634 while (true) {
4635 int slot = path->slots[0];
4636 struct extent_buffer *leaf = path->nodes[0];
4637 int nritems = btrfs_header_nritems(leaf);
4638
4639 if (slot >= nritems) {
4640 if (ins_nr > 0) {
4641 ret = copy_items(trans, inode, dst_path, path,
4642 start_slot, ins_nr, 1, 0);
4643 if (ret < 0)
4644 return ret;
4645 ins_nr = 0;
4646 }
4647 ret = btrfs_next_leaf(root, path);
4648 if (ret < 0)
4649 return ret;
4650 else if (ret > 0)
4651 break;
4652 continue;
4653 }
4654
4655 btrfs_item_key_to_cpu(leaf, &key, slot);
4656 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4657 break;
4658
4659 if (ins_nr == 0)
4660 start_slot = slot;
4661 ins_nr++;
4662 path->slots[0]++;
4663 found_xattrs = true;
4664 cond_resched();
4665 }
4666 if (ins_nr > 0) {
4667 ret = copy_items(trans, inode, dst_path, path,
4668 start_slot, ins_nr, 1, 0);
4669 if (ret < 0)
4670 return ret;
4671 }
4672
4673 if (!found_xattrs)
4674 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
4675
4676 return 0;
4677 }
4678
4679 /*
4680 * When using the NO_HOLES feature if we punched a hole that causes the
4681 * deletion of entire leafs or all the extent items of the first leaf (the one
4682 * that contains the inode item and references) we may end up not processing
4683 * any extents, because there are no leafs with a generation matching the
4684 * current transaction that have extent items for our inode. So we need to find
4685 * if any holes exist and then log them. We also need to log holes after any
4686 * truncate operation that changes the inode's size.
4687 */
btrfs_log_holes(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode,struct btrfs_path * path)4688 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4689 struct btrfs_root *root,
4690 struct btrfs_inode *inode,
4691 struct btrfs_path *path)
4692 {
4693 struct btrfs_fs_info *fs_info = root->fs_info;
4694 struct btrfs_key key;
4695 const u64 ino = btrfs_ino(inode);
4696 const u64 i_size = i_size_read(&inode->vfs_inode);
4697 u64 prev_extent_end = 0;
4698 int ret;
4699
4700 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4701 return 0;
4702
4703 key.objectid = ino;
4704 key.type = BTRFS_EXTENT_DATA_KEY;
4705 key.offset = 0;
4706
4707 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4708 if (ret < 0)
4709 return ret;
4710
4711 while (true) {
4712 struct extent_buffer *leaf = path->nodes[0];
4713
4714 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4715 ret = btrfs_next_leaf(root, path);
4716 if (ret < 0)
4717 return ret;
4718 if (ret > 0) {
4719 ret = 0;
4720 break;
4721 }
4722 leaf = path->nodes[0];
4723 }
4724
4725 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4726 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4727 break;
4728
4729 /* We have a hole, log it. */
4730 if (prev_extent_end < key.offset) {
4731 const u64 hole_len = key.offset - prev_extent_end;
4732
4733 /*
4734 * Release the path to avoid deadlocks with other code
4735 * paths that search the root while holding locks on
4736 * leafs from the log root.
4737 */
4738 btrfs_release_path(path);
4739 ret = btrfs_insert_file_extent(trans, root->log_root,
4740 ino, prev_extent_end, 0,
4741 0, hole_len, 0, hole_len,
4742 0, 0, 0);
4743 if (ret < 0)
4744 return ret;
4745
4746 /*
4747 * Search for the same key again in the root. Since it's
4748 * an extent item and we are holding the inode lock, the
4749 * key must still exist. If it doesn't just emit warning
4750 * and return an error to fall back to a transaction
4751 * commit.
4752 */
4753 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4754 if (ret < 0)
4755 return ret;
4756 if (WARN_ON(ret > 0))
4757 return -ENOENT;
4758 leaf = path->nodes[0];
4759 }
4760
4761 prev_extent_end = btrfs_file_extent_end(path);
4762 path->slots[0]++;
4763 cond_resched();
4764 }
4765
4766 if (prev_extent_end < i_size) {
4767 u64 hole_len;
4768
4769 btrfs_release_path(path);
4770 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4771 ret = btrfs_insert_file_extent(trans, root->log_root,
4772 ino, prev_extent_end, 0, 0,
4773 hole_len, 0, hole_len,
4774 0, 0, 0);
4775 if (ret < 0)
4776 return ret;
4777 }
4778
4779 return 0;
4780 }
4781
4782 /*
4783 * When we are logging a new inode X, check if it doesn't have a reference that
4784 * matches the reference from some other inode Y created in a past transaction
4785 * and that was renamed in the current transaction. If we don't do this, then at
4786 * log replay time we can lose inode Y (and all its files if it's a directory):
4787 *
4788 * mkdir /mnt/x
4789 * echo "hello world" > /mnt/x/foobar
4790 * sync
4791 * mv /mnt/x /mnt/y
4792 * mkdir /mnt/x # or touch /mnt/x
4793 * xfs_io -c fsync /mnt/x
4794 * <power fail>
4795 * mount fs, trigger log replay
4796 *
4797 * After the log replay procedure, we would lose the first directory and all its
4798 * files (file foobar).
4799 * For the case where inode Y is not a directory we simply end up losing it:
4800 *
4801 * echo "123" > /mnt/foo
4802 * sync
4803 * mv /mnt/foo /mnt/bar
4804 * echo "abc" > /mnt/foo
4805 * xfs_io -c fsync /mnt/foo
4806 * <power fail>
4807 *
4808 * We also need this for cases where a snapshot entry is replaced by some other
4809 * entry (file or directory) otherwise we end up with an unreplayable log due to
4810 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4811 * if it were a regular entry:
4812 *
4813 * mkdir /mnt/x
4814 * btrfs subvolume snapshot /mnt /mnt/x/snap
4815 * btrfs subvolume delete /mnt/x/snap
4816 * rmdir /mnt/x
4817 * mkdir /mnt/x
4818 * fsync /mnt/x or fsync some new file inside it
4819 * <power fail>
4820 *
4821 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4822 * the same transaction.
4823 */
btrfs_check_ref_name_override(struct extent_buffer * eb,const int slot,const struct btrfs_key * key,struct btrfs_inode * inode,u64 * other_ino,u64 * other_parent)4824 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4825 const int slot,
4826 const struct btrfs_key *key,
4827 struct btrfs_inode *inode,
4828 u64 *other_ino, u64 *other_parent)
4829 {
4830 int ret;
4831 struct btrfs_path *search_path;
4832 char *name = NULL;
4833 u32 name_len = 0;
4834 u32 item_size = btrfs_item_size_nr(eb, slot);
4835 u32 cur_offset = 0;
4836 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4837
4838 search_path = btrfs_alloc_path();
4839 if (!search_path)
4840 return -ENOMEM;
4841 search_path->search_commit_root = 1;
4842 search_path->skip_locking = 1;
4843
4844 while (cur_offset < item_size) {
4845 u64 parent;
4846 u32 this_name_len;
4847 u32 this_len;
4848 unsigned long name_ptr;
4849 struct btrfs_dir_item *di;
4850
4851 if (key->type == BTRFS_INODE_REF_KEY) {
4852 struct btrfs_inode_ref *iref;
4853
4854 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4855 parent = key->offset;
4856 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4857 name_ptr = (unsigned long)(iref + 1);
4858 this_len = sizeof(*iref) + this_name_len;
4859 } else {
4860 struct btrfs_inode_extref *extref;
4861
4862 extref = (struct btrfs_inode_extref *)(ptr +
4863 cur_offset);
4864 parent = btrfs_inode_extref_parent(eb, extref);
4865 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4866 name_ptr = (unsigned long)&extref->name;
4867 this_len = sizeof(*extref) + this_name_len;
4868 }
4869
4870 if (this_name_len > name_len) {
4871 char *new_name;
4872
4873 new_name = krealloc(name, this_name_len, GFP_NOFS);
4874 if (!new_name) {
4875 ret = -ENOMEM;
4876 goto out;
4877 }
4878 name_len = this_name_len;
4879 name = new_name;
4880 }
4881
4882 read_extent_buffer(eb, name, name_ptr, this_name_len);
4883 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4884 parent, name, this_name_len, 0);
4885 if (di && !IS_ERR(di)) {
4886 struct btrfs_key di_key;
4887
4888 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4889 di, &di_key);
4890 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4891 if (di_key.objectid != key->objectid) {
4892 ret = 1;
4893 *other_ino = di_key.objectid;
4894 *other_parent = parent;
4895 } else {
4896 ret = 0;
4897 }
4898 } else {
4899 ret = -EAGAIN;
4900 }
4901 goto out;
4902 } else if (IS_ERR(di)) {
4903 ret = PTR_ERR(di);
4904 goto out;
4905 }
4906 btrfs_release_path(search_path);
4907
4908 cur_offset += this_len;
4909 }
4910 ret = 0;
4911 out:
4912 btrfs_free_path(search_path);
4913 kfree(name);
4914 return ret;
4915 }
4916
4917 struct btrfs_ino_list {
4918 u64 ino;
4919 u64 parent;
4920 struct list_head list;
4921 };
4922
log_conflicting_inodes(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_log_ctx * ctx,u64 ino,u64 parent)4923 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4924 struct btrfs_root *root,
4925 struct btrfs_path *path,
4926 struct btrfs_log_ctx *ctx,
4927 u64 ino, u64 parent)
4928 {
4929 struct btrfs_ino_list *ino_elem;
4930 LIST_HEAD(inode_list);
4931 int ret = 0;
4932
4933 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4934 if (!ino_elem)
4935 return -ENOMEM;
4936 ino_elem->ino = ino;
4937 ino_elem->parent = parent;
4938 list_add_tail(&ino_elem->list, &inode_list);
4939
4940 while (!list_empty(&inode_list)) {
4941 struct btrfs_fs_info *fs_info = root->fs_info;
4942 struct btrfs_key key;
4943 struct inode *inode;
4944
4945 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4946 list);
4947 ino = ino_elem->ino;
4948 parent = ino_elem->parent;
4949 list_del(&ino_elem->list);
4950 kfree(ino_elem);
4951 if (ret)
4952 continue;
4953
4954 btrfs_release_path(path);
4955
4956 inode = btrfs_iget(fs_info->sb, ino, root);
4957 /*
4958 * If the other inode that had a conflicting dir entry was
4959 * deleted in the current transaction, we need to log its parent
4960 * directory.
4961 */
4962 if (IS_ERR(inode)) {
4963 ret = PTR_ERR(inode);
4964 if (ret == -ENOENT) {
4965 inode = btrfs_iget(fs_info->sb, parent, root);
4966 if (IS_ERR(inode)) {
4967 ret = PTR_ERR(inode);
4968 } else {
4969 ret = btrfs_log_inode(trans, root,
4970 BTRFS_I(inode),
4971 LOG_OTHER_INODE_ALL,
4972 ctx);
4973 btrfs_add_delayed_iput(inode);
4974 }
4975 }
4976 continue;
4977 }
4978 /*
4979 * If the inode was already logged skip it - otherwise we can
4980 * hit an infinite loop. Example:
4981 *
4982 * From the commit root (previous transaction) we have the
4983 * following inodes:
4984 *
4985 * inode 257 a directory
4986 * inode 258 with references "zz" and "zz_link" on inode 257
4987 * inode 259 with reference "a" on inode 257
4988 *
4989 * And in the current (uncommitted) transaction we have:
4990 *
4991 * inode 257 a directory, unchanged
4992 * inode 258 with references "a" and "a2" on inode 257
4993 * inode 259 with reference "zz_link" on inode 257
4994 * inode 261 with reference "zz" on inode 257
4995 *
4996 * When logging inode 261 the following infinite loop could
4997 * happen if we don't skip already logged inodes:
4998 *
4999 * - we detect inode 258 as a conflicting inode, with inode 261
5000 * on reference "zz", and log it;
5001 *
5002 * - we detect inode 259 as a conflicting inode, with inode 258
5003 * on reference "a", and log it;
5004 *
5005 * - we detect inode 258 as a conflicting inode, with inode 259
5006 * on reference "zz_link", and log it - again! After this we
5007 * repeat the above steps forever.
5008 */
5009 spin_lock(&BTRFS_I(inode)->lock);
5010 /*
5011 * Check the inode's logged_trans only instead of
5012 * btrfs_inode_in_log(). This is because the last_log_commit of
5013 * the inode is not updated when we only log that it exists and
5014 * it has the full sync bit set (see btrfs_log_inode()).
5015 */
5016 if (BTRFS_I(inode)->logged_trans == trans->transid) {
5017 spin_unlock(&BTRFS_I(inode)->lock);
5018 btrfs_add_delayed_iput(inode);
5019 continue;
5020 }
5021 spin_unlock(&BTRFS_I(inode)->lock);
5022 /*
5023 * We are safe logging the other inode without acquiring its
5024 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5025 * are safe against concurrent renames of the other inode as
5026 * well because during a rename we pin the log and update the
5027 * log with the new name before we unpin it.
5028 */
5029 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5030 LOG_OTHER_INODE, ctx);
5031 if (ret) {
5032 btrfs_add_delayed_iput(inode);
5033 continue;
5034 }
5035
5036 key.objectid = ino;
5037 key.type = BTRFS_INODE_REF_KEY;
5038 key.offset = 0;
5039 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5040 if (ret < 0) {
5041 btrfs_add_delayed_iput(inode);
5042 continue;
5043 }
5044
5045 while (true) {
5046 struct extent_buffer *leaf = path->nodes[0];
5047 int slot = path->slots[0];
5048 u64 other_ino = 0;
5049 u64 other_parent = 0;
5050
5051 if (slot >= btrfs_header_nritems(leaf)) {
5052 ret = btrfs_next_leaf(root, path);
5053 if (ret < 0) {
5054 break;
5055 } else if (ret > 0) {
5056 ret = 0;
5057 break;
5058 }
5059 continue;
5060 }
5061
5062 btrfs_item_key_to_cpu(leaf, &key, slot);
5063 if (key.objectid != ino ||
5064 (key.type != BTRFS_INODE_REF_KEY &&
5065 key.type != BTRFS_INODE_EXTREF_KEY)) {
5066 ret = 0;
5067 break;
5068 }
5069
5070 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5071 BTRFS_I(inode), &other_ino,
5072 &other_parent);
5073 if (ret < 0)
5074 break;
5075 if (ret > 0) {
5076 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5077 if (!ino_elem) {
5078 ret = -ENOMEM;
5079 break;
5080 }
5081 ino_elem->ino = other_ino;
5082 ino_elem->parent = other_parent;
5083 list_add_tail(&ino_elem->list, &inode_list);
5084 ret = 0;
5085 }
5086 path->slots[0]++;
5087 }
5088 btrfs_add_delayed_iput(inode);
5089 }
5090
5091 return ret;
5092 }
5093
copy_inode_items_to_log(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct btrfs_key * min_key,const struct btrfs_key * max_key,struct btrfs_path * path,struct btrfs_path * dst_path,const u64 logged_isize,const bool recursive_logging,const int inode_only,struct btrfs_log_ctx * ctx,bool * need_log_inode_item)5094 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5095 struct btrfs_inode *inode,
5096 struct btrfs_key *min_key,
5097 const struct btrfs_key *max_key,
5098 struct btrfs_path *path,
5099 struct btrfs_path *dst_path,
5100 const u64 logged_isize,
5101 const bool recursive_logging,
5102 const int inode_only,
5103 struct btrfs_log_ctx *ctx,
5104 bool *need_log_inode_item)
5105 {
5106 const u64 i_size = i_size_read(&inode->vfs_inode);
5107 struct btrfs_root *root = inode->root;
5108 int ins_start_slot = 0;
5109 int ins_nr = 0;
5110 int ret;
5111
5112 while (1) {
5113 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5114 if (ret < 0)
5115 return ret;
5116 if (ret > 0) {
5117 ret = 0;
5118 break;
5119 }
5120 again:
5121 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5122 if (min_key->objectid != max_key->objectid)
5123 break;
5124 if (min_key->type > max_key->type)
5125 break;
5126
5127 if (min_key->type == BTRFS_INODE_ITEM_KEY) {
5128 *need_log_inode_item = false;
5129 } else if (min_key->type == BTRFS_EXTENT_DATA_KEY &&
5130 min_key->offset >= i_size) {
5131 /*
5132 * Extents at and beyond eof are logged with
5133 * btrfs_log_prealloc_extents().
5134 * Only regular files have BTRFS_EXTENT_DATA_KEY keys,
5135 * and no keys greater than that, so bail out.
5136 */
5137 break;
5138 } else if ((min_key->type == BTRFS_INODE_REF_KEY ||
5139 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5140 inode->generation == trans->transid &&
5141 !recursive_logging) {
5142 u64 other_ino = 0;
5143 u64 other_parent = 0;
5144
5145 ret = btrfs_check_ref_name_override(path->nodes[0],
5146 path->slots[0], min_key, inode,
5147 &other_ino, &other_parent);
5148 if (ret < 0) {
5149 return ret;
5150 } else if (ret > 0 && ctx &&
5151 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5152 if (ins_nr > 0) {
5153 ins_nr++;
5154 } else {
5155 ins_nr = 1;
5156 ins_start_slot = path->slots[0];
5157 }
5158 ret = copy_items(trans, inode, dst_path, path,
5159 ins_start_slot, ins_nr,
5160 inode_only, logged_isize);
5161 if (ret < 0)
5162 return ret;
5163 ins_nr = 0;
5164
5165 ret = log_conflicting_inodes(trans, root, path,
5166 ctx, other_ino, other_parent);
5167 if (ret)
5168 return ret;
5169 btrfs_release_path(path);
5170 goto next_key;
5171 }
5172 } else if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5173 /* Skip xattrs, logged later with btrfs_log_all_xattrs() */
5174 if (ins_nr == 0)
5175 goto next_slot;
5176 ret = copy_items(trans, inode, dst_path, path,
5177 ins_start_slot,
5178 ins_nr, inode_only, logged_isize);
5179 if (ret < 0)
5180 return ret;
5181 ins_nr = 0;
5182 goto next_slot;
5183 }
5184
5185 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5186 ins_nr++;
5187 goto next_slot;
5188 } else if (!ins_nr) {
5189 ins_start_slot = path->slots[0];
5190 ins_nr = 1;
5191 goto next_slot;
5192 }
5193
5194 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5195 ins_nr, inode_only, logged_isize);
5196 if (ret < 0)
5197 return ret;
5198 ins_nr = 1;
5199 ins_start_slot = path->slots[0];
5200 next_slot:
5201 path->slots[0]++;
5202 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5203 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5204 path->slots[0]);
5205 goto again;
5206 }
5207 if (ins_nr) {
5208 ret = copy_items(trans, inode, dst_path, path,
5209 ins_start_slot, ins_nr, inode_only,
5210 logged_isize);
5211 if (ret < 0)
5212 return ret;
5213 ins_nr = 0;
5214 }
5215 btrfs_release_path(path);
5216 next_key:
5217 if (min_key->offset < (u64)-1) {
5218 min_key->offset++;
5219 } else if (min_key->type < max_key->type) {
5220 min_key->type++;
5221 min_key->offset = 0;
5222 } else {
5223 break;
5224 }
5225 }
5226 if (ins_nr) {
5227 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5228 ins_nr, inode_only, logged_isize);
5229 if (ret)
5230 return ret;
5231 }
5232
5233 if (inode_only == LOG_INODE_ALL && S_ISREG(inode->vfs_inode.i_mode)) {
5234 /*
5235 * Release the path because otherwise we might attempt to double
5236 * lock the same leaf with btrfs_log_prealloc_extents() below.
5237 */
5238 btrfs_release_path(path);
5239 ret = btrfs_log_prealloc_extents(trans, inode, dst_path);
5240 }
5241
5242 return ret;
5243 }
5244
5245 /* log a single inode in the tree log.
5246 * At least one parent directory for this inode must exist in the tree
5247 * or be logged already.
5248 *
5249 * Any items from this inode changed by the current transaction are copied
5250 * to the log tree. An extra reference is taken on any extents in this
5251 * file, allowing us to avoid a whole pile of corner cases around logging
5252 * blocks that have been removed from the tree.
5253 *
5254 * See LOG_INODE_ALL and related defines for a description of what inode_only
5255 * does.
5256 *
5257 * This handles both files and directories.
5258 */
btrfs_log_inode(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * inode,int inode_only,struct btrfs_log_ctx * ctx)5259 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5260 struct btrfs_root *root, struct btrfs_inode *inode,
5261 int inode_only,
5262 struct btrfs_log_ctx *ctx)
5263 {
5264 struct btrfs_path *path;
5265 struct btrfs_path *dst_path;
5266 struct btrfs_key min_key;
5267 struct btrfs_key max_key;
5268 struct btrfs_root *log = root->log_root;
5269 int err = 0;
5270 int ret = 0;
5271 bool fast_search = false;
5272 u64 ino = btrfs_ino(inode);
5273 struct extent_map_tree *em_tree = &inode->extent_tree;
5274 u64 logged_isize = 0;
5275 bool need_log_inode_item = true;
5276 bool xattrs_logged = false;
5277 bool recursive_logging = false;
5278
5279 path = btrfs_alloc_path();
5280 if (!path)
5281 return -ENOMEM;
5282 dst_path = btrfs_alloc_path();
5283 if (!dst_path) {
5284 btrfs_free_path(path);
5285 return -ENOMEM;
5286 }
5287
5288 min_key.objectid = ino;
5289 min_key.type = BTRFS_INODE_ITEM_KEY;
5290 min_key.offset = 0;
5291
5292 max_key.objectid = ino;
5293
5294
5295 /* today the code can only do partial logging of directories */
5296 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5297 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5298 &inode->runtime_flags) &&
5299 inode_only >= LOG_INODE_EXISTS))
5300 max_key.type = BTRFS_XATTR_ITEM_KEY;
5301 else
5302 max_key.type = (u8)-1;
5303 max_key.offset = (u64)-1;
5304
5305 /*
5306 * Only run delayed items if we are a directory. We want to make sure
5307 * all directory indexes hit the fs/subvolume tree so we can find them
5308 * and figure out which index ranges have to be logged.
5309 *
5310 * Otherwise commit the delayed inode only if the full sync flag is set,
5311 * as we want to make sure an up to date version is in the subvolume
5312 * tree so copy_inode_items_to_log() / copy_items() can find it and copy
5313 * it to the log tree. For a non full sync, we always log the inode item
5314 * based on the in-memory struct btrfs_inode which is always up to date.
5315 */
5316 if (S_ISDIR(inode->vfs_inode.i_mode))
5317 ret = btrfs_commit_inode_delayed_items(trans, inode);
5318 else if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5319 ret = btrfs_commit_inode_delayed_inode(inode);
5320
5321 if (ret) {
5322 btrfs_free_path(path);
5323 btrfs_free_path(dst_path);
5324 return ret;
5325 }
5326
5327 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5328 recursive_logging = true;
5329 if (inode_only == LOG_OTHER_INODE)
5330 inode_only = LOG_INODE_EXISTS;
5331 else
5332 inode_only = LOG_INODE_ALL;
5333 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5334 } else {
5335 mutex_lock(&inode->log_mutex);
5336 }
5337
5338 /*
5339 * For symlinks, we must always log their content, which is stored in an
5340 * inline extent, otherwise we could end up with an empty symlink after
5341 * log replay, which is invalid on linux (symlink(2) returns -ENOENT if
5342 * one attempts to create an empty symlink).
5343 * We don't need to worry about flushing delalloc, because when we create
5344 * the inline extent when the symlink is created (we never have delalloc
5345 * for symlinks).
5346 */
5347 if (S_ISLNK(inode->vfs_inode.i_mode))
5348 inode_only = LOG_INODE_ALL;
5349
5350 /*
5351 * a brute force approach to making sure we get the most uptodate
5352 * copies of everything.
5353 */
5354 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5355 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5356
5357 if (inode_only == LOG_INODE_EXISTS)
5358 max_key_type = BTRFS_XATTR_ITEM_KEY;
5359 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5360 } else {
5361 if (inode_only == LOG_INODE_EXISTS) {
5362 /*
5363 * Make sure the new inode item we write to the log has
5364 * the same isize as the current one (if it exists).
5365 * This is necessary to prevent data loss after log
5366 * replay, and also to prevent doing a wrong expanding
5367 * truncate - for e.g. create file, write 4K into offset
5368 * 0, fsync, write 4K into offset 4096, add hard link,
5369 * fsync some other file (to sync log), power fail - if
5370 * we use the inode's current i_size, after log replay
5371 * we get a 8Kb file, with the last 4Kb extent as a hole
5372 * (zeroes), as if an expanding truncate happened,
5373 * instead of getting a file of 4Kb only.
5374 */
5375 err = logged_inode_size(log, inode, path, &logged_isize);
5376 if (err)
5377 goto out_unlock;
5378 }
5379 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5380 &inode->runtime_flags)) {
5381 if (inode_only == LOG_INODE_EXISTS) {
5382 max_key.type = BTRFS_XATTR_ITEM_KEY;
5383 ret = drop_objectid_items(trans, log, path, ino,
5384 max_key.type);
5385 } else {
5386 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5387 &inode->runtime_flags);
5388 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5389 &inode->runtime_flags);
5390 while(1) {
5391 ret = btrfs_truncate_inode_items(trans,
5392 log, &inode->vfs_inode, 0, 0);
5393 if (ret != -EAGAIN)
5394 break;
5395 }
5396 }
5397 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5398 &inode->runtime_flags) ||
5399 inode_only == LOG_INODE_EXISTS) {
5400 if (inode_only == LOG_INODE_ALL)
5401 fast_search = true;
5402 max_key.type = BTRFS_XATTR_ITEM_KEY;
5403 ret = drop_objectid_items(trans, log, path, ino,
5404 max_key.type);
5405 } else {
5406 if (inode_only == LOG_INODE_ALL)
5407 fast_search = true;
5408 goto log_extents;
5409 }
5410
5411 }
5412 if (ret) {
5413 err = ret;
5414 goto out_unlock;
5415 }
5416
5417 err = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
5418 path, dst_path, logged_isize,
5419 recursive_logging, inode_only, ctx,
5420 &need_log_inode_item);
5421 if (err)
5422 goto out_unlock;
5423
5424 btrfs_release_path(path);
5425 btrfs_release_path(dst_path);
5426 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5427 if (err)
5428 goto out_unlock;
5429 xattrs_logged = true;
5430 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5431 btrfs_release_path(path);
5432 btrfs_release_path(dst_path);
5433 err = btrfs_log_holes(trans, root, inode, path);
5434 if (err)
5435 goto out_unlock;
5436 }
5437 log_extents:
5438 btrfs_release_path(path);
5439 btrfs_release_path(dst_path);
5440 if (need_log_inode_item) {
5441 err = log_inode_item(trans, log, dst_path, inode);
5442 if (!err && !xattrs_logged) {
5443 err = btrfs_log_all_xattrs(trans, root, inode, path,
5444 dst_path);
5445 btrfs_release_path(path);
5446 }
5447 if (err)
5448 goto out_unlock;
5449 }
5450 if (fast_search) {
5451 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5452 ctx);
5453 if (ret) {
5454 err = ret;
5455 goto out_unlock;
5456 }
5457 } else if (inode_only == LOG_INODE_ALL) {
5458 struct extent_map *em, *n;
5459
5460 write_lock(&em_tree->lock);
5461 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
5462 list_del_init(&em->list);
5463 write_unlock(&em_tree->lock);
5464 }
5465
5466 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5467 ret = log_directory_changes(trans, root, inode, path, dst_path,
5468 ctx);
5469 if (ret) {
5470 err = ret;
5471 goto out_unlock;
5472 }
5473 }
5474
5475 /*
5476 * If we are logging that an ancestor inode exists as part of logging a
5477 * new name from a link or rename operation, don't mark the inode as
5478 * logged - otherwise if an explicit fsync is made against an ancestor,
5479 * the fsync considers the inode in the log and doesn't sync the log,
5480 * resulting in the ancestor missing after a power failure unless the
5481 * log was synced as part of an fsync against any other unrelated inode.
5482 * So keep it simple for this case and just don't flag the ancestors as
5483 * logged.
5484 */
5485 if (!ctx ||
5486 !(S_ISDIR(inode->vfs_inode.i_mode) && ctx->logging_new_name &&
5487 &inode->vfs_inode != ctx->inode)) {
5488 spin_lock(&inode->lock);
5489 inode->logged_trans = trans->transid;
5490 /*
5491 * Don't update last_log_commit if we logged that an inode exists
5492 * after it was loaded to memory (full_sync bit set).
5493 * This is to prevent data loss when we do a write to the inode,
5494 * then the inode gets evicted after all delalloc was flushed,
5495 * then we log it exists (due to a rename for example) and then
5496 * fsync it. This last fsync would do nothing (not logging the
5497 * extents previously written).
5498 */
5499 if (inode_only != LOG_INODE_EXISTS ||
5500 !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5501 inode->last_log_commit = inode->last_sub_trans;
5502 spin_unlock(&inode->lock);
5503 }
5504 out_unlock:
5505 mutex_unlock(&inode->log_mutex);
5506
5507 btrfs_free_path(path);
5508 btrfs_free_path(dst_path);
5509 return err;
5510 }
5511
5512 /*
5513 * Check if we must fallback to a transaction commit when logging an inode.
5514 * This must be called after logging the inode and is used only in the context
5515 * when fsyncing an inode requires the need to log some other inode - in which
5516 * case we can't lock the i_mutex of each other inode we need to log as that
5517 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5518 * log inodes up or down in the hierarchy) or rename operations for example. So
5519 * we take the log_mutex of the inode after we have logged it and then check for
5520 * its last_unlink_trans value - this is safe because any task setting
5521 * last_unlink_trans must take the log_mutex and it must do this before it does
5522 * the actual unlink operation, so if we do this check before a concurrent task
5523 * sets last_unlink_trans it means we've logged a consistent version/state of
5524 * all the inode items, otherwise we are not sure and must do a transaction
5525 * commit (the concurrent task might have only updated last_unlink_trans before
5526 * we logged the inode or it might have also done the unlink).
5527 */
btrfs_must_commit_transaction(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)5528 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5529 struct btrfs_inode *inode)
5530 {
5531 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5532 bool ret = false;
5533
5534 mutex_lock(&inode->log_mutex);
5535 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5536 /*
5537 * Make sure any commits to the log are forced to be full
5538 * commits.
5539 */
5540 btrfs_set_log_full_commit(trans);
5541 ret = true;
5542 }
5543 mutex_unlock(&inode->log_mutex);
5544
5545 return ret;
5546 }
5547
5548 /*
5549 * follow the dentry parent pointers up the chain and see if any
5550 * of the directories in it require a full commit before they can
5551 * be logged. Returns zero if nothing special needs to be done or 1 if
5552 * a full commit is required.
5553 */
check_parent_dirs_for_sync(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct dentry * parent,struct super_block * sb,u64 last_committed)5554 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5555 struct btrfs_inode *inode,
5556 struct dentry *parent,
5557 struct super_block *sb,
5558 u64 last_committed)
5559 {
5560 int ret = 0;
5561 struct dentry *old_parent = NULL;
5562
5563 /*
5564 * for regular files, if its inode is already on disk, we don't
5565 * have to worry about the parents at all. This is because
5566 * we can use the last_unlink_trans field to record renames
5567 * and other fun in this file.
5568 */
5569 if (S_ISREG(inode->vfs_inode.i_mode) &&
5570 inode->generation <= last_committed &&
5571 inode->last_unlink_trans <= last_committed)
5572 goto out;
5573
5574 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5575 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5576 goto out;
5577 inode = BTRFS_I(d_inode(parent));
5578 }
5579
5580 while (1) {
5581 if (btrfs_must_commit_transaction(trans, inode)) {
5582 ret = 1;
5583 break;
5584 }
5585
5586 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5587 break;
5588
5589 if (IS_ROOT(parent)) {
5590 inode = BTRFS_I(d_inode(parent));
5591 if (btrfs_must_commit_transaction(trans, inode))
5592 ret = 1;
5593 break;
5594 }
5595
5596 parent = dget_parent(parent);
5597 dput(old_parent);
5598 old_parent = parent;
5599 inode = BTRFS_I(d_inode(parent));
5600
5601 }
5602 dput(old_parent);
5603 out:
5604 return ret;
5605 }
5606
5607 struct btrfs_dir_list {
5608 u64 ino;
5609 struct list_head list;
5610 };
5611
5612 /*
5613 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5614 * details about the why it is needed.
5615 * This is a recursive operation - if an existing dentry corresponds to a
5616 * directory, that directory's new entries are logged too (same behaviour as
5617 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5618 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5619 * complains about the following circular lock dependency / possible deadlock:
5620 *
5621 * CPU0 CPU1
5622 * ---- ----
5623 * lock(&type->i_mutex_dir_key#3/2);
5624 * lock(sb_internal#2);
5625 * lock(&type->i_mutex_dir_key#3/2);
5626 * lock(&sb->s_type->i_mutex_key#14);
5627 *
5628 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5629 * sb_start_intwrite() in btrfs_start_transaction().
5630 * Not locking i_mutex of the inodes is still safe because:
5631 *
5632 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5633 * that while logging the inode new references (names) are added or removed
5634 * from the inode, leaving the logged inode item with a link count that does
5635 * not match the number of logged inode reference items. This is fine because
5636 * at log replay time we compute the real number of links and correct the
5637 * link count in the inode item (see replay_one_buffer() and
5638 * link_to_fixup_dir());
5639 *
5640 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5641 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5642 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5643 * has a size that doesn't match the sum of the lengths of all the logged
5644 * names. This does not result in a problem because if a dir_item key is
5645 * logged but its matching dir_index key is not logged, at log replay time we
5646 * don't use it to replay the respective name (see replay_one_name()). On the
5647 * other hand if only the dir_index key ends up being logged, the respective
5648 * name is added to the fs/subvol tree with both the dir_item and dir_index
5649 * keys created (see replay_one_name()).
5650 * The directory's inode item with a wrong i_size is not a problem as well,
5651 * since we don't use it at log replay time to set the i_size in the inode
5652 * item of the fs/subvol tree (see overwrite_item()).
5653 */
log_new_dir_dentries(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_inode * start_inode,struct btrfs_log_ctx * ctx)5654 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5655 struct btrfs_root *root,
5656 struct btrfs_inode *start_inode,
5657 struct btrfs_log_ctx *ctx)
5658 {
5659 struct btrfs_fs_info *fs_info = root->fs_info;
5660 struct btrfs_root *log = root->log_root;
5661 struct btrfs_path *path;
5662 LIST_HEAD(dir_list);
5663 struct btrfs_dir_list *dir_elem;
5664 int ret = 0;
5665
5666 path = btrfs_alloc_path();
5667 if (!path)
5668 return -ENOMEM;
5669
5670 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5671 if (!dir_elem) {
5672 btrfs_free_path(path);
5673 return -ENOMEM;
5674 }
5675 dir_elem->ino = btrfs_ino(start_inode);
5676 list_add_tail(&dir_elem->list, &dir_list);
5677
5678 while (!list_empty(&dir_list)) {
5679 struct extent_buffer *leaf;
5680 struct btrfs_key min_key;
5681 int nritems;
5682 int i;
5683
5684 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5685 list);
5686 if (ret)
5687 goto next_dir_inode;
5688
5689 min_key.objectid = dir_elem->ino;
5690 min_key.type = BTRFS_DIR_ITEM_KEY;
5691 min_key.offset = 0;
5692 again:
5693 btrfs_release_path(path);
5694 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5695 if (ret < 0) {
5696 goto next_dir_inode;
5697 } else if (ret > 0) {
5698 ret = 0;
5699 goto next_dir_inode;
5700 }
5701
5702 process_leaf:
5703 leaf = path->nodes[0];
5704 nritems = btrfs_header_nritems(leaf);
5705 for (i = path->slots[0]; i < nritems; i++) {
5706 struct btrfs_dir_item *di;
5707 struct btrfs_key di_key;
5708 struct inode *di_inode;
5709 struct btrfs_dir_list *new_dir_elem;
5710 int log_mode = LOG_INODE_EXISTS;
5711 int type;
5712
5713 btrfs_item_key_to_cpu(leaf, &min_key, i);
5714 if (min_key.objectid != dir_elem->ino ||
5715 min_key.type != BTRFS_DIR_ITEM_KEY)
5716 goto next_dir_inode;
5717
5718 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5719 type = btrfs_dir_type(leaf, di);
5720 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5721 type != BTRFS_FT_DIR)
5722 continue;
5723 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5724 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5725 continue;
5726
5727 btrfs_release_path(path);
5728 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5729 if (IS_ERR(di_inode)) {
5730 ret = PTR_ERR(di_inode);
5731 goto next_dir_inode;
5732 }
5733
5734 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5735 btrfs_add_delayed_iput(di_inode);
5736 break;
5737 }
5738
5739 ctx->log_new_dentries = false;
5740 if (type == BTRFS_FT_DIR)
5741 log_mode = LOG_INODE_ALL;
5742 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5743 log_mode, ctx);
5744 if (!ret &&
5745 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5746 ret = 1;
5747 btrfs_add_delayed_iput(di_inode);
5748 if (ret)
5749 goto next_dir_inode;
5750 if (ctx->log_new_dentries) {
5751 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5752 GFP_NOFS);
5753 if (!new_dir_elem) {
5754 ret = -ENOMEM;
5755 goto next_dir_inode;
5756 }
5757 new_dir_elem->ino = di_key.objectid;
5758 list_add_tail(&new_dir_elem->list, &dir_list);
5759 }
5760 break;
5761 }
5762 if (i == nritems) {
5763 ret = btrfs_next_leaf(log, path);
5764 if (ret < 0) {
5765 goto next_dir_inode;
5766 } else if (ret > 0) {
5767 ret = 0;
5768 goto next_dir_inode;
5769 }
5770 goto process_leaf;
5771 }
5772 if (min_key.offset < (u64)-1) {
5773 min_key.offset++;
5774 goto again;
5775 }
5776 next_dir_inode:
5777 list_del(&dir_elem->list);
5778 kfree(dir_elem);
5779 }
5780
5781 btrfs_free_path(path);
5782 return ret;
5783 }
5784
btrfs_log_all_parents(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct btrfs_log_ctx * ctx)5785 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5786 struct btrfs_inode *inode,
5787 struct btrfs_log_ctx *ctx)
5788 {
5789 struct btrfs_fs_info *fs_info = trans->fs_info;
5790 int ret;
5791 struct btrfs_path *path;
5792 struct btrfs_key key;
5793 struct btrfs_root *root = inode->root;
5794 const u64 ino = btrfs_ino(inode);
5795
5796 path = btrfs_alloc_path();
5797 if (!path)
5798 return -ENOMEM;
5799 path->skip_locking = 1;
5800 path->search_commit_root = 1;
5801
5802 key.objectid = ino;
5803 key.type = BTRFS_INODE_REF_KEY;
5804 key.offset = 0;
5805 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5806 if (ret < 0)
5807 goto out;
5808
5809 while (true) {
5810 struct extent_buffer *leaf = path->nodes[0];
5811 int slot = path->slots[0];
5812 u32 cur_offset = 0;
5813 u32 item_size;
5814 unsigned long ptr;
5815
5816 if (slot >= btrfs_header_nritems(leaf)) {
5817 ret = btrfs_next_leaf(root, path);
5818 if (ret < 0)
5819 goto out;
5820 else if (ret > 0)
5821 break;
5822 continue;
5823 }
5824
5825 btrfs_item_key_to_cpu(leaf, &key, slot);
5826 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5827 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5828 break;
5829
5830 item_size = btrfs_item_size_nr(leaf, slot);
5831 ptr = btrfs_item_ptr_offset(leaf, slot);
5832 while (cur_offset < item_size) {
5833 struct btrfs_key inode_key;
5834 struct inode *dir_inode;
5835
5836 inode_key.type = BTRFS_INODE_ITEM_KEY;
5837 inode_key.offset = 0;
5838
5839 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5840 struct btrfs_inode_extref *extref;
5841
5842 extref = (struct btrfs_inode_extref *)
5843 (ptr + cur_offset);
5844 inode_key.objectid = btrfs_inode_extref_parent(
5845 leaf, extref);
5846 cur_offset += sizeof(*extref);
5847 cur_offset += btrfs_inode_extref_name_len(leaf,
5848 extref);
5849 } else {
5850 inode_key.objectid = key.offset;
5851 cur_offset = item_size;
5852 }
5853
5854 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
5855 root);
5856 /*
5857 * If the parent inode was deleted, return an error to
5858 * fallback to a transaction commit. This is to prevent
5859 * getting an inode that was moved from one parent A to
5860 * a parent B, got its former parent A deleted and then
5861 * it got fsync'ed, from existing at both parents after
5862 * a log replay (and the old parent still existing).
5863 * Example:
5864 *
5865 * mkdir /mnt/A
5866 * mkdir /mnt/B
5867 * touch /mnt/B/bar
5868 * sync
5869 * mv /mnt/B/bar /mnt/A/bar
5870 * mv -T /mnt/A /mnt/B
5871 * fsync /mnt/B/bar
5872 * <power fail>
5873 *
5874 * If we ignore the old parent B which got deleted,
5875 * after a log replay we would have file bar linked
5876 * at both parents and the old parent B would still
5877 * exist.
5878 */
5879 if (IS_ERR(dir_inode)) {
5880 ret = PTR_ERR(dir_inode);
5881 goto out;
5882 }
5883
5884 if (ctx)
5885 ctx->log_new_dentries = false;
5886 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5887 LOG_INODE_ALL, ctx);
5888 if (!ret &&
5889 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5890 ret = 1;
5891 if (!ret && ctx && ctx->log_new_dentries)
5892 ret = log_new_dir_dentries(trans, root,
5893 BTRFS_I(dir_inode), ctx);
5894 btrfs_add_delayed_iput(dir_inode);
5895 if (ret)
5896 goto out;
5897 }
5898 path->slots[0]++;
5899 }
5900 ret = 0;
5901 out:
5902 btrfs_free_path(path);
5903 return ret;
5904 }
5905
log_new_ancestors(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct btrfs_log_ctx * ctx)5906 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5907 struct btrfs_root *root,
5908 struct btrfs_path *path,
5909 struct btrfs_log_ctx *ctx)
5910 {
5911 struct btrfs_key found_key;
5912
5913 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5914
5915 while (true) {
5916 struct btrfs_fs_info *fs_info = root->fs_info;
5917 const u64 last_committed = fs_info->last_trans_committed;
5918 struct extent_buffer *leaf = path->nodes[0];
5919 int slot = path->slots[0];
5920 struct btrfs_key search_key;
5921 struct inode *inode;
5922 u64 ino;
5923 int ret = 0;
5924
5925 btrfs_release_path(path);
5926
5927 ino = found_key.offset;
5928
5929 search_key.objectid = found_key.offset;
5930 search_key.type = BTRFS_INODE_ITEM_KEY;
5931 search_key.offset = 0;
5932 inode = btrfs_iget(fs_info->sb, ino, root);
5933 if (IS_ERR(inode))
5934 return PTR_ERR(inode);
5935
5936 if (BTRFS_I(inode)->generation > last_committed)
5937 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5938 LOG_INODE_EXISTS, ctx);
5939 btrfs_add_delayed_iput(inode);
5940 if (ret)
5941 return ret;
5942
5943 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5944 break;
5945
5946 search_key.type = BTRFS_INODE_REF_KEY;
5947 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5948 if (ret < 0)
5949 return ret;
5950
5951 leaf = path->nodes[0];
5952 slot = path->slots[0];
5953 if (slot >= btrfs_header_nritems(leaf)) {
5954 ret = btrfs_next_leaf(root, path);
5955 if (ret < 0)
5956 return ret;
5957 else if (ret > 0)
5958 return -ENOENT;
5959 leaf = path->nodes[0];
5960 slot = path->slots[0];
5961 }
5962
5963 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5964 if (found_key.objectid != search_key.objectid ||
5965 found_key.type != BTRFS_INODE_REF_KEY)
5966 return -ENOENT;
5967 }
5968 return 0;
5969 }
5970
log_new_ancestors_fast(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct dentry * parent,struct btrfs_log_ctx * ctx)5971 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
5972 struct btrfs_inode *inode,
5973 struct dentry *parent,
5974 struct btrfs_log_ctx *ctx)
5975 {
5976 struct btrfs_root *root = inode->root;
5977 struct btrfs_fs_info *fs_info = root->fs_info;
5978 struct dentry *old_parent = NULL;
5979 struct super_block *sb = inode->vfs_inode.i_sb;
5980 int ret = 0;
5981
5982 while (true) {
5983 if (!parent || d_really_is_negative(parent) ||
5984 sb != parent->d_sb)
5985 break;
5986
5987 inode = BTRFS_I(d_inode(parent));
5988 if (root != inode->root)
5989 break;
5990
5991 if (inode->generation > fs_info->last_trans_committed) {
5992 ret = btrfs_log_inode(trans, root, inode,
5993 LOG_INODE_EXISTS, ctx);
5994 if (ret)
5995 break;
5996 }
5997 if (IS_ROOT(parent))
5998 break;
5999
6000 parent = dget_parent(parent);
6001 dput(old_parent);
6002 old_parent = parent;
6003 }
6004 dput(old_parent);
6005
6006 return ret;
6007 }
6008
log_all_new_ancestors(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct dentry * parent,struct btrfs_log_ctx * ctx)6009 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6010 struct btrfs_inode *inode,
6011 struct dentry *parent,
6012 struct btrfs_log_ctx *ctx)
6013 {
6014 struct btrfs_root *root = inode->root;
6015 const u64 ino = btrfs_ino(inode);
6016 struct btrfs_path *path;
6017 struct btrfs_key search_key;
6018 int ret;
6019
6020 /*
6021 * For a single hard link case, go through a fast path that does not
6022 * need to iterate the fs/subvolume tree.
6023 */
6024 if (inode->vfs_inode.i_nlink < 2)
6025 return log_new_ancestors_fast(trans, inode, parent, ctx);
6026
6027 path = btrfs_alloc_path();
6028 if (!path)
6029 return -ENOMEM;
6030
6031 search_key.objectid = ino;
6032 search_key.type = BTRFS_INODE_REF_KEY;
6033 search_key.offset = 0;
6034 again:
6035 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6036 if (ret < 0)
6037 goto out;
6038 if (ret == 0)
6039 path->slots[0]++;
6040
6041 while (true) {
6042 struct extent_buffer *leaf = path->nodes[0];
6043 int slot = path->slots[0];
6044 struct btrfs_key found_key;
6045
6046 if (slot >= btrfs_header_nritems(leaf)) {
6047 ret = btrfs_next_leaf(root, path);
6048 if (ret < 0)
6049 goto out;
6050 else if (ret > 0)
6051 break;
6052 continue;
6053 }
6054
6055 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6056 if (found_key.objectid != ino ||
6057 found_key.type > BTRFS_INODE_EXTREF_KEY)
6058 break;
6059
6060 /*
6061 * Don't deal with extended references because they are rare
6062 * cases and too complex to deal with (we would need to keep
6063 * track of which subitem we are processing for each item in
6064 * this loop, etc). So just return some error to fallback to
6065 * a transaction commit.
6066 */
6067 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6068 ret = -EMLINK;
6069 goto out;
6070 }
6071
6072 /*
6073 * Logging ancestors needs to do more searches on the fs/subvol
6074 * tree, so it releases the path as needed to avoid deadlocks.
6075 * Keep track of the last inode ref key and resume from that key
6076 * after logging all new ancestors for the current hard link.
6077 */
6078 memcpy(&search_key, &found_key, sizeof(search_key));
6079
6080 ret = log_new_ancestors(trans, root, path, ctx);
6081 if (ret)
6082 goto out;
6083 btrfs_release_path(path);
6084 goto again;
6085 }
6086 ret = 0;
6087 out:
6088 btrfs_free_path(path);
6089 return ret;
6090 }
6091
6092 /*
6093 * helper function around btrfs_log_inode to make sure newly created
6094 * parent directories also end up in the log. A minimal inode and backref
6095 * only logging is done of any parent directories that are older than
6096 * the last committed transaction
6097 */
btrfs_log_inode_parent(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct dentry * parent,int inode_only,struct btrfs_log_ctx * ctx)6098 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6099 struct btrfs_inode *inode,
6100 struct dentry *parent,
6101 int inode_only,
6102 struct btrfs_log_ctx *ctx)
6103 {
6104 struct btrfs_root *root = inode->root;
6105 struct btrfs_fs_info *fs_info = root->fs_info;
6106 struct super_block *sb;
6107 int ret = 0;
6108 u64 last_committed = fs_info->last_trans_committed;
6109 bool log_dentries = false;
6110
6111 sb = inode->vfs_inode.i_sb;
6112
6113 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6114 ret = 1;
6115 goto end_no_trans;
6116 }
6117
6118 /*
6119 * The prev transaction commit doesn't complete, we need do
6120 * full commit by ourselves.
6121 */
6122 if (fs_info->last_trans_log_full_commit >
6123 fs_info->last_trans_committed) {
6124 ret = 1;
6125 goto end_no_trans;
6126 }
6127
6128 if (btrfs_root_refs(&root->root_item) == 0) {
6129 ret = 1;
6130 goto end_no_trans;
6131 }
6132
6133 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
6134 last_committed);
6135 if (ret)
6136 goto end_no_trans;
6137
6138 /*
6139 * Skip already logged inodes or inodes corresponding to tmpfiles
6140 * (since logging them is pointless, a link count of 0 means they
6141 * will never be accessible).
6142 */
6143 if ((btrfs_inode_in_log(inode, trans->transid) &&
6144 list_empty(&ctx->ordered_extents)) ||
6145 inode->vfs_inode.i_nlink == 0) {
6146 ret = BTRFS_NO_LOG_SYNC;
6147 goto end_no_trans;
6148 }
6149
6150 ret = start_log_trans(trans, root, ctx);
6151 if (ret)
6152 goto end_no_trans;
6153
6154 ret = btrfs_log_inode(trans, root, inode, inode_only, ctx);
6155 if (ret)
6156 goto end_trans;
6157
6158 /*
6159 * for regular files, if its inode is already on disk, we don't
6160 * have to worry about the parents at all. This is because
6161 * we can use the last_unlink_trans field to record renames
6162 * and other fun in this file.
6163 */
6164 if (S_ISREG(inode->vfs_inode.i_mode) &&
6165 inode->generation <= last_committed &&
6166 inode->last_unlink_trans <= last_committed) {
6167 ret = 0;
6168 goto end_trans;
6169 }
6170
6171 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6172 log_dentries = true;
6173
6174 /*
6175 * On unlink we must make sure all our current and old parent directory
6176 * inodes are fully logged. This is to prevent leaving dangling
6177 * directory index entries in directories that were our parents but are
6178 * not anymore. Not doing this results in old parent directory being
6179 * impossible to delete after log replay (rmdir will always fail with
6180 * error -ENOTEMPTY).
6181 *
6182 * Example 1:
6183 *
6184 * mkdir testdir
6185 * touch testdir/foo
6186 * ln testdir/foo testdir/bar
6187 * sync
6188 * unlink testdir/bar
6189 * xfs_io -c fsync testdir/foo
6190 * <power failure>
6191 * mount fs, triggers log replay
6192 *
6193 * If we don't log the parent directory (testdir), after log replay the
6194 * directory still has an entry pointing to the file inode using the bar
6195 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6196 * the file inode has a link count of 1.
6197 *
6198 * Example 2:
6199 *
6200 * mkdir testdir
6201 * touch foo
6202 * ln foo testdir/foo2
6203 * ln foo testdir/foo3
6204 * sync
6205 * unlink testdir/foo3
6206 * xfs_io -c fsync foo
6207 * <power failure>
6208 * mount fs, triggers log replay
6209 *
6210 * Similar as the first example, after log replay the parent directory
6211 * testdir still has an entry pointing to the inode file with name foo3
6212 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6213 * and has a link count of 2.
6214 */
6215 if (inode->last_unlink_trans > last_committed) {
6216 ret = btrfs_log_all_parents(trans, inode, ctx);
6217 if (ret)
6218 goto end_trans;
6219 }
6220
6221 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6222 if (ret)
6223 goto end_trans;
6224
6225 if (log_dentries)
6226 ret = log_new_dir_dentries(trans, root, inode, ctx);
6227 else
6228 ret = 0;
6229 end_trans:
6230 if (ret < 0) {
6231 btrfs_set_log_full_commit(trans);
6232 ret = 1;
6233 }
6234
6235 if (ret)
6236 btrfs_remove_log_ctx(root, ctx);
6237 btrfs_end_log_trans(root);
6238 end_no_trans:
6239 return ret;
6240 }
6241
6242 /*
6243 * it is not safe to log dentry if the chunk root has added new
6244 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6245 * If this returns 1, you must commit the transaction to safely get your
6246 * data on disk.
6247 */
btrfs_log_dentry_safe(struct btrfs_trans_handle * trans,struct dentry * dentry,struct btrfs_log_ctx * ctx)6248 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6249 struct dentry *dentry,
6250 struct btrfs_log_ctx *ctx)
6251 {
6252 struct dentry *parent = dget_parent(dentry);
6253 int ret;
6254
6255 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6256 LOG_INODE_ALL, ctx);
6257 dput(parent);
6258
6259 return ret;
6260 }
6261
6262 /*
6263 * should be called during mount to recover any replay any log trees
6264 * from the FS
6265 */
btrfs_recover_log_trees(struct btrfs_root * log_root_tree)6266 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6267 {
6268 int ret;
6269 struct btrfs_path *path;
6270 struct btrfs_trans_handle *trans;
6271 struct btrfs_key key;
6272 struct btrfs_key found_key;
6273 struct btrfs_root *log;
6274 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6275 struct walk_control wc = {
6276 .process_func = process_one_buffer,
6277 .stage = LOG_WALK_PIN_ONLY,
6278 };
6279
6280 path = btrfs_alloc_path();
6281 if (!path)
6282 return -ENOMEM;
6283
6284 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6285
6286 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6287 if (IS_ERR(trans)) {
6288 ret = PTR_ERR(trans);
6289 goto error;
6290 }
6291
6292 wc.trans = trans;
6293 wc.pin = 1;
6294
6295 ret = walk_log_tree(trans, log_root_tree, &wc);
6296 if (ret) {
6297 btrfs_handle_fs_error(fs_info, ret,
6298 "Failed to pin buffers while recovering log root tree.");
6299 goto error;
6300 }
6301
6302 again:
6303 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6304 key.offset = (u64)-1;
6305 key.type = BTRFS_ROOT_ITEM_KEY;
6306
6307 while (1) {
6308 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6309
6310 if (ret < 0) {
6311 btrfs_handle_fs_error(fs_info, ret,
6312 "Couldn't find tree log root.");
6313 goto error;
6314 }
6315 if (ret > 0) {
6316 if (path->slots[0] == 0)
6317 break;
6318 path->slots[0]--;
6319 }
6320 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6321 path->slots[0]);
6322 btrfs_release_path(path);
6323 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6324 break;
6325
6326 log = btrfs_read_tree_root(log_root_tree, &found_key);
6327 if (IS_ERR(log)) {
6328 ret = PTR_ERR(log);
6329 btrfs_handle_fs_error(fs_info, ret,
6330 "Couldn't read tree log root.");
6331 goto error;
6332 }
6333
6334 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
6335 true);
6336 if (IS_ERR(wc.replay_dest)) {
6337 ret = PTR_ERR(wc.replay_dest);
6338
6339 /*
6340 * We didn't find the subvol, likely because it was
6341 * deleted. This is ok, simply skip this log and go to
6342 * the next one.
6343 *
6344 * We need to exclude the root because we can't have
6345 * other log replays overwriting this log as we'll read
6346 * it back in a few more times. This will keep our
6347 * block from being modified, and we'll just bail for
6348 * each subsequent pass.
6349 */
6350 if (ret == -ENOENT)
6351 ret = btrfs_pin_extent_for_log_replay(trans,
6352 log->node->start,
6353 log->node->len);
6354 btrfs_put_root(log);
6355
6356 if (!ret)
6357 goto next;
6358 btrfs_handle_fs_error(fs_info, ret,
6359 "Couldn't read target root for tree log recovery.");
6360 goto error;
6361 }
6362
6363 wc.replay_dest->log_root = log;
6364 btrfs_record_root_in_trans(trans, wc.replay_dest);
6365 ret = walk_log_tree(trans, log, &wc);
6366
6367 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6368 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6369 path);
6370 }
6371
6372 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6373 struct btrfs_root *root = wc.replay_dest;
6374
6375 btrfs_release_path(path);
6376
6377 /*
6378 * We have just replayed everything, and the highest
6379 * objectid of fs roots probably has changed in case
6380 * some inode_item's got replayed.
6381 *
6382 * root->objectid_mutex is not acquired as log replay
6383 * could only happen during mount.
6384 */
6385 ret = btrfs_find_highest_objectid(root,
6386 &root->highest_objectid);
6387 }
6388
6389 wc.replay_dest->log_root = NULL;
6390 btrfs_put_root(wc.replay_dest);
6391 btrfs_put_root(log);
6392
6393 if (ret)
6394 goto error;
6395 next:
6396 if (found_key.offset == 0)
6397 break;
6398 key.offset = found_key.offset - 1;
6399 }
6400 btrfs_release_path(path);
6401
6402 /* step one is to pin it all, step two is to replay just inodes */
6403 if (wc.pin) {
6404 wc.pin = 0;
6405 wc.process_func = replay_one_buffer;
6406 wc.stage = LOG_WALK_REPLAY_INODES;
6407 goto again;
6408 }
6409 /* step three is to replay everything */
6410 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6411 wc.stage++;
6412 goto again;
6413 }
6414
6415 btrfs_free_path(path);
6416
6417 /* step 4: commit the transaction, which also unpins the blocks */
6418 ret = btrfs_commit_transaction(trans);
6419 if (ret)
6420 return ret;
6421
6422 log_root_tree->log_root = NULL;
6423 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6424 btrfs_put_root(log_root_tree);
6425
6426 return 0;
6427 error:
6428 if (wc.trans)
6429 btrfs_end_transaction(wc.trans);
6430 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6431 btrfs_free_path(path);
6432 return ret;
6433 }
6434
6435 /*
6436 * there are some corner cases where we want to force a full
6437 * commit instead of allowing a directory to be logged.
6438 *
6439 * They revolve around files there were unlinked from the directory, and
6440 * this function updates the parent directory so that a full commit is
6441 * properly done if it is fsync'd later after the unlinks are done.
6442 *
6443 * Must be called before the unlink operations (updates to the subvolume tree,
6444 * inodes, etc) are done.
6445 */
btrfs_record_unlink_dir(struct btrfs_trans_handle * trans,struct btrfs_inode * dir,struct btrfs_inode * inode,int for_rename)6446 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6447 struct btrfs_inode *dir, struct btrfs_inode *inode,
6448 int for_rename)
6449 {
6450 /*
6451 * when we're logging a file, if it hasn't been renamed
6452 * or unlinked, and its inode is fully committed on disk,
6453 * we don't have to worry about walking up the directory chain
6454 * to log its parents.
6455 *
6456 * So, we use the last_unlink_trans field to put this transid
6457 * into the file. When the file is logged we check it and
6458 * don't log the parents if the file is fully on disk.
6459 */
6460 mutex_lock(&inode->log_mutex);
6461 inode->last_unlink_trans = trans->transid;
6462 mutex_unlock(&inode->log_mutex);
6463
6464 /*
6465 * if this directory was already logged any new
6466 * names for this file/dir will get recorded
6467 */
6468 if (dir->logged_trans == trans->transid)
6469 return;
6470
6471 /*
6472 * if the inode we're about to unlink was logged,
6473 * the log will be properly updated for any new names
6474 */
6475 if (inode->logged_trans == trans->transid)
6476 return;
6477
6478 /*
6479 * when renaming files across directories, if the directory
6480 * there we're unlinking from gets fsync'd later on, there's
6481 * no way to find the destination directory later and fsync it
6482 * properly. So, we have to be conservative and force commits
6483 * so the new name gets discovered.
6484 */
6485 if (for_rename)
6486 goto record;
6487
6488 /* we can safely do the unlink without any special recording */
6489 return;
6490
6491 record:
6492 mutex_lock(&dir->log_mutex);
6493 dir->last_unlink_trans = trans->transid;
6494 mutex_unlock(&dir->log_mutex);
6495 }
6496
6497 /*
6498 * Make sure that if someone attempts to fsync the parent directory of a deleted
6499 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6500 * that after replaying the log tree of the parent directory's root we will not
6501 * see the snapshot anymore and at log replay time we will not see any log tree
6502 * corresponding to the deleted snapshot's root, which could lead to replaying
6503 * it after replaying the log tree of the parent directory (which would replay
6504 * the snapshot delete operation).
6505 *
6506 * Must be called before the actual snapshot destroy operation (updates to the
6507 * parent root and tree of tree roots trees, etc) are done.
6508 */
btrfs_record_snapshot_destroy(struct btrfs_trans_handle * trans,struct btrfs_inode * dir)6509 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6510 struct btrfs_inode *dir)
6511 {
6512 mutex_lock(&dir->log_mutex);
6513 dir->last_unlink_trans = trans->transid;
6514 mutex_unlock(&dir->log_mutex);
6515 }
6516
6517 /*
6518 * Call this after adding a new name for a file and it will properly
6519 * update the log to reflect the new name.
6520 */
btrfs_log_new_name(struct btrfs_trans_handle * trans,struct btrfs_inode * inode,struct btrfs_inode * old_dir,struct dentry * parent)6521 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
6522 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6523 struct dentry *parent)
6524 {
6525 struct btrfs_log_ctx ctx;
6526
6527 /*
6528 * this will force the logging code to walk the dentry chain
6529 * up for the file
6530 */
6531 if (!S_ISDIR(inode->vfs_inode.i_mode))
6532 inode->last_unlink_trans = trans->transid;
6533
6534 /*
6535 * if this inode hasn't been logged and directory we're renaming it
6536 * from hasn't been logged, we don't need to log it
6537 */
6538 if (!inode_logged(trans, inode) &&
6539 (!old_dir || !inode_logged(trans, old_dir)))
6540 return;
6541
6542 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
6543 ctx.logging_new_name = true;
6544 /*
6545 * We don't care about the return value. If we fail to log the new name
6546 * then we know the next attempt to sync the log will fallback to a full
6547 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
6548 * we don't need to worry about getting a log committed that has an
6549 * inconsistent state after a rename operation.
6550 */
6551 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);
6552 }
6553
6554