aboutsummaryrefslogtreecommitdiffstats
path: root/fs/btrfs/tree-log.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/btrfs/tree-log.c')
-rw-r--r--fs/btrfs/tree-log.c2896
1 files changed, 2896 insertions, 0 deletions
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
new file mode 100644
index 000000000000..be4fc30a30e4
--- /dev/null
+++ b/fs/btrfs/tree-log.c
@@ -0,0 +1,2896 @@
1/*
2 * Copyright (C) 2008 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/sched.h>
20#include "ctree.h"
21#include "transaction.h"
22#include "disk-io.h"
23#include "locking.h"
24#include "print-tree.h"
25#include "compat.h"
26
27/* magic values for the inode_only field in btrfs_log_inode:
28 *
29 * LOG_INODE_ALL means to log everything
30 * LOG_INODE_EXISTS means to log just enough to recreate the inode
31 * during log replay
32 */
33#define LOG_INODE_ALL 0
34#define LOG_INODE_EXISTS 1
35
36/*
37 * stages for the tree walking. The first
38 * stage (0) is to only pin down the blocks we find
39 * the second stage (1) is to make sure that all the inodes
40 * we find in the log are created in the subvolume.
41 *
42 * The last stage is to deal with directories and links and extents
43 * and all the other fun semantics
44 */
45#define LOG_WALK_PIN_ONLY 0
46#define LOG_WALK_REPLAY_INODES 1
47#define LOG_WALK_REPLAY_ALL 2
48
49static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
50 struct btrfs_root *root, struct inode *inode,
51 int inode_only);
52
53/*
54 * tree logging is a special write ahead log used to make sure that
55 * fsyncs and O_SYNCs can happen without doing full tree commits.
56 *
57 * Full tree commits are expensive because they require commonly
58 * modified blocks to be recowed, creating many dirty pages in the
59 * extent tree an 4x-6x higher write load than ext3.
60 *
61 * Instead of doing a tree commit on every fsync, we use the
62 * key ranges and transaction ids to find items for a given file or directory
63 * that have changed in this transaction. Those items are copied into
64 * a special tree (one per subvolume root), that tree is written to disk
65 * and then the fsync is considered complete.
66 *
67 * After a crash, items are copied out of the log-tree back into the
68 * subvolume tree. Any file data extents found are recorded in the extent
69 * allocation tree, and the log-tree freed.
70 *
71 * The log tree is read three times, once to pin down all the extents it is
72 * using in ram and once, once to create all the inodes logged in the tree
73 * and once to do all the other items.
74 */
75
76/*
77 * btrfs_add_log_tree adds a new per-subvolume log tree into the
78 * tree of log tree roots. This must be called with a tree log transaction
79 * running (see start_log_trans).
80 */
81int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
82 struct btrfs_root *root)
83{
84 struct btrfs_key key;
85 struct btrfs_root_item root_item;
86 struct btrfs_inode_item *inode_item;
87 struct extent_buffer *leaf;
88 struct btrfs_root *new_root = root;
89 int ret;
90 u64 objectid = root->root_key.objectid;
91
92 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
93 BTRFS_TREE_LOG_OBJECTID,
94 trans->transid, 0, 0, 0);
95 if (IS_ERR(leaf)) {
96 ret = PTR_ERR(leaf);
97 return ret;
98 }
99
100 btrfs_set_header_nritems(leaf, 0);
101 btrfs_set_header_level(leaf, 0);
102 btrfs_set_header_bytenr(leaf, leaf->start);
103 btrfs_set_header_generation(leaf, trans->transid);
104 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
105
106 write_extent_buffer(leaf, root->fs_info->fsid,
107 (unsigned long)btrfs_header_fsid(leaf),
108 BTRFS_FSID_SIZE);
109 btrfs_mark_buffer_dirty(leaf);
110
111 inode_item = &root_item.inode;
112 memset(inode_item, 0, sizeof(*inode_item));
113 inode_item->generation = cpu_to_le64(1);
114 inode_item->size = cpu_to_le64(3);
115 inode_item->nlink = cpu_to_le32(1);
116 inode_item->nbytes = cpu_to_le64(root->leafsize);
117 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
118
119 btrfs_set_root_bytenr(&root_item, leaf->start);
120 btrfs_set_root_generation(&root_item, trans->transid);
121 btrfs_set_root_level(&root_item, 0);
122 btrfs_set_root_refs(&root_item, 0);
123 btrfs_set_root_used(&root_item, 0);
124
125 memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
126 root_item.drop_level = 0;
127
128 btrfs_tree_unlock(leaf);
129 free_extent_buffer(leaf);
130 leaf = NULL;
131
132 btrfs_set_root_dirid(&root_item, 0);
133
134 key.objectid = BTRFS_TREE_LOG_OBJECTID;
135 key.offset = objectid;
136 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
137 ret = btrfs_insert_root(trans, root->fs_info->log_root_tree, &key,
138 &root_item);
139 if (ret)
140 goto fail;
141
142 new_root = btrfs_read_fs_root_no_radix(root->fs_info->log_root_tree,
143 &key);
144 BUG_ON(!new_root);
145
146 WARN_ON(root->log_root);
147 root->log_root = new_root;
148
149 /*
150 * log trees do not get reference counted because they go away
151 * before a real commit is actually done. They do store pointers
152 * to file data extents, and those reference counts still get
153 * updated (along with back refs to the log tree).
154 */
155 new_root->ref_cows = 0;
156 new_root->last_trans = trans->transid;
157fail:
158 return ret;
159}
160
161/*
162 * start a sub transaction and setup the log tree
163 * this increments the log tree writer count to make the people
164 * syncing the tree wait for us to finish
165 */
166static int start_log_trans(struct btrfs_trans_handle *trans,
167 struct btrfs_root *root)
168{
169 int ret;
170 mutex_lock(&root->fs_info->tree_log_mutex);
171 if (!root->fs_info->log_root_tree) {
172 ret = btrfs_init_log_root_tree(trans, root->fs_info);
173 BUG_ON(ret);
174 }
175 if (!root->log_root) {
176 ret = btrfs_add_log_tree(trans, root);
177 BUG_ON(ret);
178 }
179 atomic_inc(&root->fs_info->tree_log_writers);
180 root->fs_info->tree_log_batch++;
181 mutex_unlock(&root->fs_info->tree_log_mutex);
182 return 0;
183}
184
185/*
186 * returns 0 if there was a log transaction running and we were able
187 * to join, or returns -ENOENT if there were not transactions
188 * in progress
189 */
190static int join_running_log_trans(struct btrfs_root *root)
191{
192 int ret = -ENOENT;
193
194 smp_mb();
195 if (!root->log_root)
196 return -ENOENT;
197
198 mutex_lock(&root->fs_info->tree_log_mutex);
199 if (root->log_root) {
200 ret = 0;
201 atomic_inc(&root->fs_info->tree_log_writers);
202 root->fs_info->tree_log_batch++;
203 }
204 mutex_unlock(&root->fs_info->tree_log_mutex);
205 return ret;
206}
207
208/*
209 * indicate we're done making changes to the log tree
210 * and wake up anyone waiting to do a sync
211 */
212static int end_log_trans(struct btrfs_root *root)
213{
214 atomic_dec(&root->fs_info->tree_log_writers);
215 smp_mb();
216 if (waitqueue_active(&root->fs_info->tree_log_wait))
217 wake_up(&root->fs_info->tree_log_wait);
218 return 0;
219}
220
221
222/*
223 * the walk control struct is used to pass state down the chain when
224 * processing the log tree. The stage field tells us which part
225 * of the log tree processing we are currently doing. The others
226 * are state fields used for that specific part
227 */
228struct walk_control {
229 /* should we free the extent on disk when done? This is used
230 * at transaction commit time while freeing a log tree
231 */
232 int free;
233
234 /* should we write out the extent buffer? This is used
235 * while flushing the log tree to disk during a sync
236 */
237 int write;
238
239 /* should we wait for the extent buffer io to finish? Also used
240 * while flushing the log tree to disk for a sync
241 */
242 int wait;
243
244 /* pin only walk, we record which extents on disk belong to the
245 * log trees
246 */
247 int pin;
248
249 /* what stage of the replay code we're currently in */
250 int stage;
251
252 /* the root we are currently replaying */
253 struct btrfs_root *replay_dest;
254
255 /* the trans handle for the current replay */
256 struct btrfs_trans_handle *trans;
257
258 /* the function that gets used to process blocks we find in the
259 * tree. Note the extent_buffer might not be up to date when it is
260 * passed in, and it must be checked or read if you need the data
261 * inside it
262 */
263 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
264 struct walk_control *wc, u64 gen);
265};
266
267/*
268 * process_func used to pin down extents, write them or wait on them
269 */
270static int process_one_buffer(struct btrfs_root *log,
271 struct extent_buffer *eb,
272 struct walk_control *wc, u64 gen)
273{
274 if (wc->pin) {
275 mutex_lock(&log->fs_info->pinned_mutex);
276 btrfs_update_pinned_extents(log->fs_info->extent_root,
277 eb->start, eb->len, 1);
278 mutex_unlock(&log->fs_info->pinned_mutex);
279 }
280
281 if (btrfs_buffer_uptodate(eb, gen)) {
282 if (wc->write)
283 btrfs_write_tree_block(eb);
284 if (wc->wait)
285 btrfs_wait_tree_block_writeback(eb);
286 }
287 return 0;
288}
289
290/*
291 * Item overwrite used by replay and tree logging. eb, slot and key all refer
292 * to the src data we are copying out.
293 *
294 * root is the tree we are copying into, and path is a scratch
295 * path for use in this function (it should be released on entry and
296 * will be released on exit).
297 *
298 * If the key is already in the destination tree the existing item is
299 * overwritten. If the existing item isn't big enough, it is extended.
300 * If it is too large, it is truncated.
301 *
302 * If the key isn't in the destination yet, a new item is inserted.
303 */
304static noinline int overwrite_item(struct btrfs_trans_handle *trans,
305 struct btrfs_root *root,
306 struct btrfs_path *path,
307 struct extent_buffer *eb, int slot,
308 struct btrfs_key *key)
309{
310 int ret;
311 u32 item_size;
312 u64 saved_i_size = 0;
313 int save_old_i_size = 0;
314 unsigned long src_ptr;
315 unsigned long dst_ptr;
316 int overwrite_root = 0;
317
318 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
319 overwrite_root = 1;
320
321 item_size = btrfs_item_size_nr(eb, slot);
322 src_ptr = btrfs_item_ptr_offset(eb, slot);
323
324 /* look for the key in the destination tree */
325 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
326 if (ret == 0) {
327 char *src_copy;
328 char *dst_copy;
329 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
330 path->slots[0]);
331 if (dst_size != item_size)
332 goto insert;
333
334 if (item_size == 0) {
335 btrfs_release_path(root, path);
336 return 0;
337 }
338 dst_copy = kmalloc(item_size, GFP_NOFS);
339 src_copy = kmalloc(item_size, GFP_NOFS);
340
341 read_extent_buffer(eb, src_copy, src_ptr, item_size);
342
343 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
344 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
345 item_size);
346 ret = memcmp(dst_copy, src_copy, item_size);
347
348 kfree(dst_copy);
349 kfree(src_copy);
350 /*
351 * they have the same contents, just return, this saves
352 * us from cowing blocks in the destination tree and doing
353 * extra writes that may not have been done by a previous
354 * sync
355 */
356 if (ret == 0) {
357 btrfs_release_path(root, path);
358 return 0;
359 }
360
361 }
362insert:
363 btrfs_release_path(root, path);
364 /* try to insert the key into the destination tree */
365 ret = btrfs_insert_empty_item(trans, root, path,
366 key, item_size);
367
368 /* make sure any existing item is the correct size */
369 if (ret == -EEXIST) {
370 u32 found_size;
371 found_size = btrfs_item_size_nr(path->nodes[0],
372 path->slots[0]);
373 if (found_size > item_size) {
374 btrfs_truncate_item(trans, root, path, item_size, 1);
375 } else if (found_size < item_size) {
376 ret = btrfs_del_item(trans, root,
377 path);
378 BUG_ON(ret);
379
380 btrfs_release_path(root, path);
381 ret = btrfs_insert_empty_item(trans,
382 root, path, key, item_size);
383 BUG_ON(ret);
384 }
385 } else if (ret) {
386 BUG();
387 }
388 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
389 path->slots[0]);
390
391 /* don't overwrite an existing inode if the generation number
392 * was logged as zero. This is done when the tree logging code
393 * is just logging an inode to make sure it exists after recovery.
394 *
395 * Also, don't overwrite i_size on directories during replay.
396 * log replay inserts and removes directory items based on the
397 * state of the tree found in the subvolume, and i_size is modified
398 * as it goes
399 */
400 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
401 struct btrfs_inode_item *src_item;
402 struct btrfs_inode_item *dst_item;
403
404 src_item = (struct btrfs_inode_item *)src_ptr;
405 dst_item = (struct btrfs_inode_item *)dst_ptr;
406
407 if (btrfs_inode_generation(eb, src_item) == 0)
408 goto no_copy;
409
410 if (overwrite_root &&
411 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
412 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
413 save_old_i_size = 1;
414 saved_i_size = btrfs_inode_size(path->nodes[0],
415 dst_item);
416 }
417 }
418
419 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
420 src_ptr, item_size);
421
422 if (save_old_i_size) {
423 struct btrfs_inode_item *dst_item;
424 dst_item = (struct btrfs_inode_item *)dst_ptr;
425 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
426 }
427
428 /* make sure the generation is filled in */
429 if (key->type == BTRFS_INODE_ITEM_KEY) {
430 struct btrfs_inode_item *dst_item;
431 dst_item = (struct btrfs_inode_item *)dst_ptr;
432 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
433 btrfs_set_inode_generation(path->nodes[0], dst_item,
434 trans->transid);
435 }
436 }
437
438 if (overwrite_root &&
439 key->type == BTRFS_EXTENT_DATA_KEY) {
440 int extent_type;
441 struct btrfs_file_extent_item *fi;
442
443 fi = (struct btrfs_file_extent_item *)dst_ptr;
444 extent_type = btrfs_file_extent_type(path->nodes[0], fi);
445 if (extent_type == BTRFS_FILE_EXTENT_REG ||
446 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
447 struct btrfs_key ins;
448 ins.objectid = btrfs_file_extent_disk_bytenr(
449 path->nodes[0], fi);
450 ins.offset = btrfs_file_extent_disk_num_bytes(
451 path->nodes[0], fi);
452 ins.type = BTRFS_EXTENT_ITEM_KEY;
453
454 /*
455 * is this extent already allocated in the extent
456 * allocation tree? If so, just add a reference
457 */
458 ret = btrfs_lookup_extent(root, ins.objectid,
459 ins.offset);
460 if (ret == 0) {
461 ret = btrfs_inc_extent_ref(trans, root,
462 ins.objectid, ins.offset,
463 path->nodes[0]->start,
464 root->root_key.objectid,
465 trans->transid, key->objectid);
466 } else {
467 /*
468 * insert the extent pointer in the extent
469 * allocation tree
470 */
471 ret = btrfs_alloc_logged_extent(trans, root,
472 path->nodes[0]->start,
473 root->root_key.objectid,
474 trans->transid, key->objectid,
475 &ins);
476 BUG_ON(ret);
477 }
478 }
479 }
480no_copy:
481 btrfs_mark_buffer_dirty(path->nodes[0]);
482 btrfs_release_path(root, path);
483 return 0;
484}
485
486/*
487 * simple helper to read an inode off the disk from a given root
488 * This can only be called for subvolume roots and not for the log
489 */
490static noinline struct inode *read_one_inode(struct btrfs_root *root,
491 u64 objectid)
492{
493 struct inode *inode;
494 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
495 if (inode->i_state & I_NEW) {
496 BTRFS_I(inode)->root = root;
497 BTRFS_I(inode)->location.objectid = objectid;
498 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
499 BTRFS_I(inode)->location.offset = 0;
500 btrfs_read_locked_inode(inode);
501 unlock_new_inode(inode);
502
503 }
504 if (is_bad_inode(inode)) {
505 iput(inode);
506 inode = NULL;
507 }
508 return inode;
509}
510
511/* replays a single extent in 'eb' at 'slot' with 'key' into the
512 * subvolume 'root'. path is released on entry and should be released
513 * on exit.
514 *
515 * extents in the log tree have not been allocated out of the extent
516 * tree yet. So, this completes the allocation, taking a reference
517 * as required if the extent already exists or creating a new extent
518 * if it isn't in the extent allocation tree yet.
519 *
520 * The extent is inserted into the file, dropping any existing extents
521 * from the file that overlap the new one.
522 */
523static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
524 struct btrfs_root *root,
525 struct btrfs_path *path,
526 struct extent_buffer *eb, int slot,
527 struct btrfs_key *key)
528{
529 int found_type;
530 u64 mask = root->sectorsize - 1;
531 u64 extent_end;
532 u64 alloc_hint;
533 u64 start = key->offset;
534 struct btrfs_file_extent_item *item;
535 struct inode *inode = NULL;
536 unsigned long size;
537 int ret = 0;
538
539 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
540 found_type = btrfs_file_extent_type(eb, item);
541
542 if (found_type == BTRFS_FILE_EXTENT_REG ||
543 found_type == BTRFS_FILE_EXTENT_PREALLOC)
544 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
545 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
546 size = btrfs_file_extent_inline_len(eb, item);
547 extent_end = (start + size + mask) & ~mask;
548 } else {
549 ret = 0;
550 goto out;
551 }
552
553 inode = read_one_inode(root, key->objectid);
554 if (!inode) {
555 ret = -EIO;
556 goto out;
557 }
558
559 /*
560 * first check to see if we already have this extent in the
561 * file. This must be done before the btrfs_drop_extents run
562 * so we don't try to drop this extent.
563 */
564 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
565 start, 0);
566
567 if (ret == 0 &&
568 (found_type == BTRFS_FILE_EXTENT_REG ||
569 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
570 struct btrfs_file_extent_item cmp1;
571 struct btrfs_file_extent_item cmp2;
572 struct btrfs_file_extent_item *existing;
573 struct extent_buffer *leaf;
574
575 leaf = path->nodes[0];
576 existing = btrfs_item_ptr(leaf, path->slots[0],
577 struct btrfs_file_extent_item);
578
579 read_extent_buffer(eb, &cmp1, (unsigned long)item,
580 sizeof(cmp1));
581 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
582 sizeof(cmp2));
583
584 /*
585 * we already have a pointer to this exact extent,
586 * we don't have to do anything
587 */
588 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
589 btrfs_release_path(root, path);
590 goto out;
591 }
592 }
593 btrfs_release_path(root, path);
594
595 /* drop any overlapping extents */
596 ret = btrfs_drop_extents(trans, root, inode,
597 start, extent_end, start, &alloc_hint);
598 BUG_ON(ret);
599
600 /* insert the extent */
601 ret = overwrite_item(trans, root, path, eb, slot, key);
602 BUG_ON(ret);
603
604 /* btrfs_drop_extents changes i_bytes & i_blocks, update it here */
605 inode_add_bytes(inode, extent_end - start);
606 btrfs_update_inode(trans, root, inode);
607out:
608 if (inode)
609 iput(inode);
610 return ret;
611}
612
613/*
614 * when cleaning up conflicts between the directory names in the
615 * subvolume, directory names in the log and directory names in the
616 * inode back references, we may have to unlink inodes from directories.
617 *
618 * This is a helper function to do the unlink of a specific directory
619 * item
620 */
621static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
622 struct btrfs_root *root,
623 struct btrfs_path *path,
624 struct inode *dir,
625 struct btrfs_dir_item *di)
626{
627 struct inode *inode;
628 char *name;
629 int name_len;
630 struct extent_buffer *leaf;
631 struct btrfs_key location;
632 int ret;
633
634 leaf = path->nodes[0];
635
636 btrfs_dir_item_key_to_cpu(leaf, di, &location);
637 name_len = btrfs_dir_name_len(leaf, di);
638 name = kmalloc(name_len, GFP_NOFS);
639 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
640 btrfs_release_path(root, path);
641
642 inode = read_one_inode(root, location.objectid);
643 BUG_ON(!inode);
644
645 btrfs_inc_nlink(inode);
646 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
647 kfree(name);
648
649 iput(inode);
650 return ret;
651}
652
653/*
654 * helper function to see if a given name and sequence number found
655 * in an inode back reference are already in a directory and correctly
656 * point to this inode
657 */
658static noinline int inode_in_dir(struct btrfs_root *root,
659 struct btrfs_path *path,
660 u64 dirid, u64 objectid, u64 index,
661 const char *name, int name_len)
662{
663 struct btrfs_dir_item *di;
664 struct btrfs_key location;
665 int match = 0;
666
667 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
668 index, name, name_len, 0);
669 if (di && !IS_ERR(di)) {
670 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
671 if (location.objectid != objectid)
672 goto out;
673 } else
674 goto out;
675 btrfs_release_path(root, path);
676
677 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
678 if (di && !IS_ERR(di)) {
679 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
680 if (location.objectid != objectid)
681 goto out;
682 } else
683 goto out;
684 match = 1;
685out:
686 btrfs_release_path(root, path);
687 return match;
688}
689
690/*
691 * helper function to check a log tree for a named back reference in
692 * an inode. This is used to decide if a back reference that is
693 * found in the subvolume conflicts with what we find in the log.
694 *
695 * inode backreferences may have multiple refs in a single item,
696 * during replay we process one reference at a time, and we don't
697 * want to delete valid links to a file from the subvolume if that
698 * link is also in the log.
699 */
700static noinline int backref_in_log(struct btrfs_root *log,
701 struct btrfs_key *key,
702 char *name, int namelen)
703{
704 struct btrfs_path *path;
705 struct btrfs_inode_ref *ref;
706 unsigned long ptr;
707 unsigned long ptr_end;
708 unsigned long name_ptr;
709 int found_name_len;
710 int item_size;
711 int ret;
712 int match = 0;
713
714 path = btrfs_alloc_path();
715 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
716 if (ret != 0)
717 goto out;
718
719 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
720 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
721 ptr_end = ptr + item_size;
722 while (ptr < ptr_end) {
723 ref = (struct btrfs_inode_ref *)ptr;
724 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
725 if (found_name_len == namelen) {
726 name_ptr = (unsigned long)(ref + 1);
727 ret = memcmp_extent_buffer(path->nodes[0], name,
728 name_ptr, namelen);
729 if (ret == 0) {
730 match = 1;
731 goto out;
732 }
733 }
734 ptr = (unsigned long)(ref + 1) + found_name_len;
735 }
736out:
737 btrfs_free_path(path);
738 return match;
739}
740
741
742/*
743 * replay one inode back reference item found in the log tree.
744 * eb, slot and key refer to the buffer and key found in the log tree.
745 * root is the destination we are replaying into, and path is for temp
746 * use by this function. (it should be released on return).
747 */
748static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
749 struct btrfs_root *root,
750 struct btrfs_root *log,
751 struct btrfs_path *path,
752 struct extent_buffer *eb, int slot,
753 struct btrfs_key *key)
754{
755 struct inode *dir;
756 int ret;
757 struct btrfs_key location;
758 struct btrfs_inode_ref *ref;
759 struct btrfs_dir_item *di;
760 struct inode *inode;
761 char *name;
762 int namelen;
763 unsigned long ref_ptr;
764 unsigned long ref_end;
765
766 location.objectid = key->objectid;
767 location.type = BTRFS_INODE_ITEM_KEY;
768 location.offset = 0;
769
770 /*
771 * it is possible that we didn't log all the parent directories
772 * for a given inode. If we don't find the dir, just don't
773 * copy the back ref in. The link count fixup code will take
774 * care of the rest
775 */
776 dir = read_one_inode(root, key->offset);
777 if (!dir)
778 return -ENOENT;
779
780 inode = read_one_inode(root, key->objectid);
781 BUG_ON(!dir);
782
783 ref_ptr = btrfs_item_ptr_offset(eb, slot);
784 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
785
786again:
787 ref = (struct btrfs_inode_ref *)ref_ptr;
788
789 namelen = btrfs_inode_ref_name_len(eb, ref);
790 name = kmalloc(namelen, GFP_NOFS);
791 BUG_ON(!name);
792
793 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
794
795 /* if we already have a perfect match, we're done */
796 if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
797 btrfs_inode_ref_index(eb, ref),
798 name, namelen)) {
799 goto out;
800 }
801
802 /*
803 * look for a conflicting back reference in the metadata.
804 * if we find one we have to unlink that name of the file
805 * before we add our new link. Later on, we overwrite any
806 * existing back reference, and we don't want to create
807 * dangling pointers in the directory.
808 */
809conflict_again:
810 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
811 if (ret == 0) {
812 char *victim_name;
813 int victim_name_len;
814 struct btrfs_inode_ref *victim_ref;
815 unsigned long ptr;
816 unsigned long ptr_end;
817 struct extent_buffer *leaf = path->nodes[0];
818
819 /* are we trying to overwrite a back ref for the root directory
820 * if so, just jump out, we're done
821 */
822 if (key->objectid == key->offset)
823 goto out_nowrite;
824
825 /* check all the names in this back reference to see
826 * if they are in the log. if so, we allow them to stay
827 * otherwise they must be unlinked as a conflict
828 */
829 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
830 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
831 while(ptr < ptr_end) {
832 victim_ref = (struct btrfs_inode_ref *)ptr;
833 victim_name_len = btrfs_inode_ref_name_len(leaf,
834 victim_ref);
835 victim_name = kmalloc(victim_name_len, GFP_NOFS);
836 BUG_ON(!victim_name);
837
838 read_extent_buffer(leaf, victim_name,
839 (unsigned long)(victim_ref + 1),
840 victim_name_len);
841
842 if (!backref_in_log(log, key, victim_name,
843 victim_name_len)) {
844 btrfs_inc_nlink(inode);
845 btrfs_release_path(root, path);
846 ret = btrfs_unlink_inode(trans, root, dir,
847 inode, victim_name,
848 victim_name_len);
849 kfree(victim_name);
850 btrfs_release_path(root, path);
851 goto conflict_again;
852 }
853 kfree(victim_name);
854 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
855 }
856 BUG_ON(ret);
857 }
858 btrfs_release_path(root, path);
859
860 /* look for a conflicting sequence number */
861 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
862 btrfs_inode_ref_index(eb, ref),
863 name, namelen, 0);
864 if (di && !IS_ERR(di)) {
865 ret = drop_one_dir_item(trans, root, path, dir, di);
866 BUG_ON(ret);
867 }
868 btrfs_release_path(root, path);
869
870
871 /* look for a conflicting name */
872 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
873 name, namelen, 0);
874 if (di && !IS_ERR(di)) {
875 ret = drop_one_dir_item(trans, root, path, dir, di);
876 BUG_ON(ret);
877 }
878 btrfs_release_path(root, path);
879
880 /* insert our name */
881 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
882 btrfs_inode_ref_index(eb, ref));
883 BUG_ON(ret);
884
885 btrfs_update_inode(trans, root, inode);
886
887out:
888 ref_ptr = (unsigned long)(ref + 1) + namelen;
889 kfree(name);
890 if (ref_ptr < ref_end)
891 goto again;
892
893 /* finally write the back reference in the inode */
894 ret = overwrite_item(trans, root, path, eb, slot, key);
895 BUG_ON(ret);
896
897out_nowrite:
898 btrfs_release_path(root, path);
899 iput(dir);
900 iput(inode);
901 return 0;
902}
903
904/*
905 * replay one csum item from the log tree into the subvolume 'root'
906 * eb, slot and key all refer to the log tree
907 * path is for temp use by this function and should be released on return
908 *
909 * This copies the checksums out of the log tree and inserts them into
910 * the subvolume. Any existing checksums for this range in the file
911 * are overwritten, and new items are added where required.
912 *
913 * We keep this simple by reusing the btrfs_ordered_sum code from
914 * the data=ordered mode. This basically means making a copy
915 * of all the checksums in ram, which we have to do anyway for kmap
916 * rules.
917 *
918 * The copy is then sent down to btrfs_csum_file_blocks, which
919 * does all the hard work of finding existing items in the file
920 * or adding new ones.
921 */
922static noinline int replay_one_csum(struct btrfs_trans_handle *trans,
923 struct btrfs_root *root,
924 struct btrfs_path *path,
925 struct extent_buffer *eb, int slot,
926 struct btrfs_key *key)
927{
928 int ret;
929 u32 item_size = btrfs_item_size_nr(eb, slot);
930 u64 cur_offset;
931 unsigned long file_bytes;
932 struct btrfs_ordered_sum *sums;
933 struct btrfs_sector_sum *sector_sum;
934 struct inode *inode;
935 unsigned long ptr;
936
937 file_bytes = (item_size / BTRFS_CRC32_SIZE) * root->sectorsize;
938 inode = read_one_inode(root, key->objectid);
939 if (!inode) {
940 return -EIO;
941 }
942
943 sums = kzalloc(btrfs_ordered_sum_size(root, file_bytes), GFP_NOFS);
944 if (!sums) {
945 iput(inode);
946 return -ENOMEM;
947 }
948
949 INIT_LIST_HEAD(&sums->list);
950 sums->len = file_bytes;
951 sums->file_offset = key->offset;
952
953 /*
954 * copy all the sums into the ordered sum struct
955 */
956 sector_sum = sums->sums;
957 cur_offset = key->offset;
958 ptr = btrfs_item_ptr_offset(eb, slot);
959 while(item_size > 0) {
960 sector_sum->offset = cur_offset;
961 read_extent_buffer(eb, &sector_sum->sum, ptr, BTRFS_CRC32_SIZE);
962 sector_sum++;
963 item_size -= BTRFS_CRC32_SIZE;
964 ptr += BTRFS_CRC32_SIZE;
965 cur_offset += root->sectorsize;
966 }
967
968 /* let btrfs_csum_file_blocks add them into the file */
969 ret = btrfs_csum_file_blocks(trans, root, inode, sums);
970 BUG_ON(ret);
971 kfree(sums);
972 iput(inode);
973
974 return 0;
975}
976/*
977 * There are a few corners where the link count of the file can't
978 * be properly maintained during replay. So, instead of adding
979 * lots of complexity to the log code, we just scan the backrefs
980 * for any file that has been through replay.
981 *
982 * The scan will update the link count on the inode to reflect the
983 * number of back refs found. If it goes down to zero, the iput
984 * will free the inode.
985 */
986static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
987 struct btrfs_root *root,
988 struct inode *inode)
989{
990 struct btrfs_path *path;
991 int ret;
992 struct btrfs_key key;
993 u64 nlink = 0;
994 unsigned long ptr;
995 unsigned long ptr_end;
996 int name_len;
997
998 key.objectid = inode->i_ino;
999 key.type = BTRFS_INODE_REF_KEY;
1000 key.offset = (u64)-1;
1001
1002 path = btrfs_alloc_path();
1003
1004 while(1) {
1005 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1006 if (ret < 0)
1007 break;
1008 if (ret > 0) {
1009 if (path->slots[0] == 0)
1010 break;
1011 path->slots[0]--;
1012 }
1013 btrfs_item_key_to_cpu(path->nodes[0], &key,
1014 path->slots[0]);
1015 if (key.objectid != inode->i_ino ||
1016 key.type != BTRFS_INODE_REF_KEY)
1017 break;
1018 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1019 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1020 path->slots[0]);
1021 while(ptr < ptr_end) {
1022 struct btrfs_inode_ref *ref;
1023
1024 ref = (struct btrfs_inode_ref *)ptr;
1025 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1026 ref);
1027 ptr = (unsigned long)(ref + 1) + name_len;
1028 nlink++;
1029 }
1030
1031 if (key.offset == 0)
1032 break;
1033 key.offset--;
1034 btrfs_release_path(root, path);
1035 }
1036 btrfs_free_path(path);
1037 if (nlink != inode->i_nlink) {
1038 inode->i_nlink = nlink;
1039 btrfs_update_inode(trans, root, inode);
1040 }
1041 BTRFS_I(inode)->index_cnt = (u64)-1;
1042
1043 return 0;
1044}
1045
1046static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1047 struct btrfs_root *root,
1048 struct btrfs_path *path)
1049{
1050 int ret;
1051 struct btrfs_key key;
1052 struct inode *inode;
1053
1054 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1055 key.type = BTRFS_ORPHAN_ITEM_KEY;
1056 key.offset = (u64)-1;
1057 while(1) {
1058 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1059 if (ret < 0)
1060 break;
1061
1062 if (ret == 1) {
1063 if (path->slots[0] == 0)
1064 break;
1065 path->slots[0]--;
1066 }
1067
1068 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1069 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1070 key.type != BTRFS_ORPHAN_ITEM_KEY)
1071 break;
1072
1073 ret = btrfs_del_item(trans, root, path);
1074 BUG_ON(ret);
1075
1076 btrfs_release_path(root, path);
1077 inode = read_one_inode(root, key.offset);
1078 BUG_ON(!inode);
1079
1080 ret = fixup_inode_link_count(trans, root, inode);
1081 BUG_ON(ret);
1082
1083 iput(inode);
1084
1085 if (key.offset == 0)
1086 break;
1087 key.offset--;
1088 }
1089 btrfs_release_path(root, path);
1090 return 0;
1091}
1092
1093
1094/*
1095 * record a given inode in the fixup dir so we can check its link
1096 * count when replay is done. The link count is incremented here
1097 * so the inode won't go away until we check it
1098 */
1099static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1100 struct btrfs_root *root,
1101 struct btrfs_path *path,
1102 u64 objectid)
1103{
1104 struct btrfs_key key;
1105 int ret = 0;
1106 struct inode *inode;
1107
1108 inode = read_one_inode(root, objectid);
1109 BUG_ON(!inode);
1110
1111 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1112 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1113 key.offset = objectid;
1114
1115 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1116
1117 btrfs_release_path(root, path);
1118 if (ret == 0) {
1119 btrfs_inc_nlink(inode);
1120 btrfs_update_inode(trans, root, inode);
1121 } else if (ret == -EEXIST) {
1122 ret = 0;
1123 } else {
1124 BUG();
1125 }
1126 iput(inode);
1127
1128 return ret;
1129}
1130
1131/*
1132 * when replaying the log for a directory, we only insert names
1133 * for inodes that actually exist. This means an fsync on a directory
1134 * does not implicitly fsync all the new files in it
1135 */
1136static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1137 struct btrfs_root *root,
1138 struct btrfs_path *path,
1139 u64 dirid, u64 index,
1140 char *name, int name_len, u8 type,
1141 struct btrfs_key *location)
1142{
1143 struct inode *inode;
1144 struct inode *dir;
1145 int ret;
1146
1147 inode = read_one_inode(root, location->objectid);
1148 if (!inode)
1149 return -ENOENT;
1150
1151 dir = read_one_inode(root, dirid);
1152 if (!dir) {
1153 iput(inode);
1154 return -EIO;
1155 }
1156 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1157
1158 /* FIXME, put inode into FIXUP list */
1159
1160 iput(inode);
1161 iput(dir);
1162 return ret;
1163}
1164
1165/*
1166 * take a single entry in a log directory item and replay it into
1167 * the subvolume.
1168 *
1169 * if a conflicting item exists in the subdirectory already,
1170 * the inode it points to is unlinked and put into the link count
1171 * fix up tree.
1172 *
1173 * If a name from the log points to a file or directory that does
1174 * not exist in the FS, it is skipped. fsyncs on directories
1175 * do not force down inodes inside that directory, just changes to the
1176 * names or unlinks in a directory.
1177 */
1178static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1179 struct btrfs_root *root,
1180 struct btrfs_path *path,
1181 struct extent_buffer *eb,
1182 struct btrfs_dir_item *di,
1183 struct btrfs_key *key)
1184{
1185 char *name;
1186 int name_len;
1187 struct btrfs_dir_item *dst_di;
1188 struct btrfs_key found_key;
1189 struct btrfs_key log_key;
1190 struct inode *dir;
1191 u8 log_type;
1192 int exists;
1193 int ret;
1194
1195 dir = read_one_inode(root, key->objectid);
1196 BUG_ON(!dir);
1197
1198 name_len = btrfs_dir_name_len(eb, di);
1199 name = kmalloc(name_len, GFP_NOFS);
1200 log_type = btrfs_dir_type(eb, di);
1201 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1202 name_len);
1203
1204 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1205 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1206 if (exists == 0)
1207 exists = 1;
1208 else
1209 exists = 0;
1210 btrfs_release_path(root, path);
1211
1212 if (key->type == BTRFS_DIR_ITEM_KEY) {
1213 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1214 name, name_len, 1);
1215 }
1216 else if (key->type == BTRFS_DIR_INDEX_KEY) {
1217 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1218 key->objectid,
1219 key->offset, name,
1220 name_len, 1);
1221 } else {
1222 BUG();
1223 }
1224 if (!dst_di || IS_ERR(dst_di)) {
1225 /* we need a sequence number to insert, so we only
1226 * do inserts for the BTRFS_DIR_INDEX_KEY types
1227 */
1228 if (key->type != BTRFS_DIR_INDEX_KEY)
1229 goto out;
1230 goto insert;
1231 }
1232
1233 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1234 /* the existing item matches the logged item */
1235 if (found_key.objectid == log_key.objectid &&
1236 found_key.type == log_key.type &&
1237 found_key.offset == log_key.offset &&
1238 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1239 goto out;
1240 }
1241
1242 /*
1243 * don't drop the conflicting directory entry if the inode
1244 * for the new entry doesn't exist
1245 */
1246 if (!exists)
1247 goto out;
1248
1249 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1250 BUG_ON(ret);
1251
1252 if (key->type == BTRFS_DIR_INDEX_KEY)
1253 goto insert;
1254out:
1255 btrfs_release_path(root, path);
1256 kfree(name);
1257 iput(dir);
1258 return 0;
1259
1260insert:
1261 btrfs_release_path(root, path);
1262 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1263 name, name_len, log_type, &log_key);
1264
1265 if (ret && ret != -ENOENT)
1266 BUG();
1267 goto out;
1268}
1269
1270/*
1271 * find all the names in a directory item and reconcile them into
1272 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1273 * one name in a directory item, but the same code gets used for
1274 * both directory index types
1275 */
1276static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1277 struct btrfs_root *root,
1278 struct btrfs_path *path,
1279 struct extent_buffer *eb, int slot,
1280 struct btrfs_key *key)
1281{
1282 int ret;
1283 u32 item_size = btrfs_item_size_nr(eb, slot);
1284 struct btrfs_dir_item *di;
1285 int name_len;
1286 unsigned long ptr;
1287 unsigned long ptr_end;
1288
1289 ptr = btrfs_item_ptr_offset(eb, slot);
1290 ptr_end = ptr + item_size;
1291 while(ptr < ptr_end) {
1292 di = (struct btrfs_dir_item *)ptr;
1293 name_len = btrfs_dir_name_len(eb, di);
1294 ret = replay_one_name(trans, root, path, eb, di, key);
1295 BUG_ON(ret);
1296 ptr = (unsigned long)(di + 1);
1297 ptr += name_len;
1298 }
1299 return 0;
1300}
1301
1302/*
1303 * directory replay has two parts. There are the standard directory
1304 * items in the log copied from the subvolume, and range items
1305 * created in the log while the subvolume was logged.
1306 *
1307 * The range items tell us which parts of the key space the log
1308 * is authoritative for. During replay, if a key in the subvolume
1309 * directory is in a logged range item, but not actually in the log
1310 * that means it was deleted from the directory before the fsync
1311 * and should be removed.
1312 */
1313static noinline int find_dir_range(struct btrfs_root *root,
1314 struct btrfs_path *path,
1315 u64 dirid, int key_type,
1316 u64 *start_ret, u64 *end_ret)
1317{
1318 struct btrfs_key key;
1319 u64 found_end;
1320 struct btrfs_dir_log_item *item;
1321 int ret;
1322 int nritems;
1323
1324 if (*start_ret == (u64)-1)
1325 return 1;
1326
1327 key.objectid = dirid;
1328 key.type = key_type;
1329 key.offset = *start_ret;
1330
1331 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1332 if (ret < 0)
1333 goto out;
1334 if (ret > 0) {
1335 if (path->slots[0] == 0)
1336 goto out;
1337 path->slots[0]--;
1338 }
1339 if (ret != 0)
1340 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1341
1342 if (key.type != key_type || key.objectid != dirid) {
1343 ret = 1;
1344 goto next;
1345 }
1346 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1347 struct btrfs_dir_log_item);
1348 found_end = btrfs_dir_log_end(path->nodes[0], item);
1349
1350 if (*start_ret >= key.offset && *start_ret <= found_end) {
1351 ret = 0;
1352 *start_ret = key.offset;
1353 *end_ret = found_end;
1354 goto out;
1355 }
1356 ret = 1;
1357next:
1358 /* check the next slot in the tree to see if it is a valid item */
1359 nritems = btrfs_header_nritems(path->nodes[0]);
1360 if (path->slots[0] >= nritems) {
1361 ret = btrfs_next_leaf(root, path);
1362 if (ret)
1363 goto out;
1364 } else {
1365 path->slots[0]++;
1366 }
1367
1368 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1369
1370 if (key.type != key_type || key.objectid != dirid) {
1371 ret = 1;
1372 goto out;
1373 }
1374 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1375 struct btrfs_dir_log_item);
1376 found_end = btrfs_dir_log_end(path->nodes[0], item);
1377 *start_ret = key.offset;
1378 *end_ret = found_end;
1379 ret = 0;
1380out:
1381 btrfs_release_path(root, path);
1382 return ret;
1383}
1384
1385/*
1386 * this looks for a given directory item in the log. If the directory
1387 * item is not in the log, the item is removed and the inode it points
1388 * to is unlinked
1389 */
1390static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1391 struct btrfs_root *root,
1392 struct btrfs_root *log,
1393 struct btrfs_path *path,
1394 struct btrfs_path *log_path,
1395 struct inode *dir,
1396 struct btrfs_key *dir_key)
1397{
1398 int ret;
1399 struct extent_buffer *eb;
1400 int slot;
1401 u32 item_size;
1402 struct btrfs_dir_item *di;
1403 struct btrfs_dir_item *log_di;
1404 int name_len;
1405 unsigned long ptr;
1406 unsigned long ptr_end;
1407 char *name;
1408 struct inode *inode;
1409 struct btrfs_key location;
1410
1411again:
1412 eb = path->nodes[0];
1413 slot = path->slots[0];
1414 item_size = btrfs_item_size_nr(eb, slot);
1415 ptr = btrfs_item_ptr_offset(eb, slot);
1416 ptr_end = ptr + item_size;
1417 while(ptr < ptr_end) {
1418 di = (struct btrfs_dir_item *)ptr;
1419 name_len = btrfs_dir_name_len(eb, di);
1420 name = kmalloc(name_len, GFP_NOFS);
1421 if (!name) {
1422 ret = -ENOMEM;
1423 goto out;
1424 }
1425 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1426 name_len);
1427 log_di = NULL;
1428 if (dir_key->type == BTRFS_DIR_ITEM_KEY) {
1429 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1430 dir_key->objectid,
1431 name, name_len, 0);
1432 } else if (dir_key->type == BTRFS_DIR_INDEX_KEY) {
1433 log_di = btrfs_lookup_dir_index_item(trans, log,
1434 log_path,
1435 dir_key->objectid,
1436 dir_key->offset,
1437 name, name_len, 0);
1438 }
1439 if (!log_di || IS_ERR(log_di)) {
1440 btrfs_dir_item_key_to_cpu(eb, di, &location);
1441 btrfs_release_path(root, path);
1442 btrfs_release_path(log, log_path);
1443 inode = read_one_inode(root, location.objectid);
1444 BUG_ON(!inode);
1445
1446 ret = link_to_fixup_dir(trans, root,
1447 path, location.objectid);
1448 BUG_ON(ret);
1449 btrfs_inc_nlink(inode);
1450 ret = btrfs_unlink_inode(trans, root, dir, inode,
1451 name, name_len);
1452 BUG_ON(ret);
1453 kfree(name);
1454 iput(inode);
1455
1456 /* there might still be more names under this key
1457 * check and repeat if required
1458 */
1459 ret = btrfs_search_slot(NULL, root, dir_key, path,
1460 0, 0);
1461 if (ret == 0)
1462 goto again;
1463 ret = 0;
1464 goto out;
1465 }
1466 btrfs_release_path(log, log_path);
1467 kfree(name);
1468
1469 ptr = (unsigned long)(di + 1);
1470 ptr += name_len;
1471 }
1472 ret = 0;
1473out:
1474 btrfs_release_path(root, path);
1475 btrfs_release_path(log, log_path);
1476 return ret;
1477}
1478
1479/*
1480 * deletion replay happens before we copy any new directory items
1481 * out of the log or out of backreferences from inodes. It
1482 * scans the log to find ranges of keys that log is authoritative for,
1483 * and then scans the directory to find items in those ranges that are
1484 * not present in the log.
1485 *
1486 * Anything we don't find in the log is unlinked and removed from the
1487 * directory.
1488 */
1489static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1490 struct btrfs_root *root,
1491 struct btrfs_root *log,
1492 struct btrfs_path *path,
1493 u64 dirid)
1494{
1495 u64 range_start;
1496 u64 range_end;
1497 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1498 int ret = 0;
1499 struct btrfs_key dir_key;
1500 struct btrfs_key found_key;
1501 struct btrfs_path *log_path;
1502 struct inode *dir;
1503
1504 dir_key.objectid = dirid;
1505 dir_key.type = BTRFS_DIR_ITEM_KEY;
1506 log_path = btrfs_alloc_path();
1507 if (!log_path)
1508 return -ENOMEM;
1509
1510 dir = read_one_inode(root, dirid);
1511 /* it isn't an error if the inode isn't there, that can happen
1512 * because we replay the deletes before we copy in the inode item
1513 * from the log
1514 */
1515 if (!dir) {
1516 btrfs_free_path(log_path);
1517 return 0;
1518 }
1519again:
1520 range_start = 0;
1521 range_end = 0;
1522 while(1) {
1523 ret = find_dir_range(log, path, dirid, key_type,
1524 &range_start, &range_end);
1525 if (ret != 0)
1526 break;
1527
1528 dir_key.offset = range_start;
1529 while(1) {
1530 int nritems;
1531 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1532 0, 0);
1533 if (ret < 0)
1534 goto out;
1535
1536 nritems = btrfs_header_nritems(path->nodes[0]);
1537 if (path->slots[0] >= nritems) {
1538 ret = btrfs_next_leaf(root, path);
1539 if (ret)
1540 break;
1541 }
1542 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1543 path->slots[0]);
1544 if (found_key.objectid != dirid ||
1545 found_key.type != dir_key.type)
1546 goto next_type;
1547
1548 if (found_key.offset > range_end)
1549 break;
1550
1551 ret = check_item_in_log(trans, root, log, path,
1552 log_path, dir, &found_key);
1553 BUG_ON(ret);
1554 if (found_key.offset == (u64)-1)
1555 break;
1556 dir_key.offset = found_key.offset + 1;
1557 }
1558 btrfs_release_path(root, path);
1559 if (range_end == (u64)-1)
1560 break;
1561 range_start = range_end + 1;
1562 }
1563
1564next_type:
1565 ret = 0;
1566 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1567 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1568 dir_key.type = BTRFS_DIR_INDEX_KEY;
1569 btrfs_release_path(root, path);
1570 goto again;
1571 }
1572out:
1573 btrfs_release_path(root, path);
1574 btrfs_free_path(log_path);
1575 iput(dir);
1576 return ret;
1577}
1578
1579/*
1580 * the process_func used to replay items from the log tree. This
1581 * gets called in two different stages. The first stage just looks
1582 * for inodes and makes sure they are all copied into the subvolume.
1583 *
1584 * The second stage copies all the other item types from the log into
1585 * the subvolume. The two stage approach is slower, but gets rid of
1586 * lots of complexity around inodes referencing other inodes that exist
1587 * only in the log (references come from either directory items or inode
1588 * back refs).
1589 */
1590static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1591 struct walk_control *wc, u64 gen)
1592{
1593 int nritems;
1594 struct btrfs_path *path;
1595 struct btrfs_root *root = wc->replay_dest;
1596 struct btrfs_key key;
1597 u32 item_size;
1598 int level;
1599 int i;
1600 int ret;
1601
1602 btrfs_read_buffer(eb, gen);
1603
1604 level = btrfs_header_level(eb);
1605
1606 if (level != 0)
1607 return 0;
1608
1609 path = btrfs_alloc_path();
1610 BUG_ON(!path);
1611
1612 nritems = btrfs_header_nritems(eb);
1613 for (i = 0; i < nritems; i++) {
1614 btrfs_item_key_to_cpu(eb, &key, i);
1615 item_size = btrfs_item_size_nr(eb, i);
1616
1617 /* inode keys are done during the first stage */
1618 if (key.type == BTRFS_INODE_ITEM_KEY &&
1619 wc->stage == LOG_WALK_REPLAY_INODES) {
1620 struct inode *inode;
1621 struct btrfs_inode_item *inode_item;
1622 u32 mode;
1623
1624 inode_item = btrfs_item_ptr(eb, i,
1625 struct btrfs_inode_item);
1626 mode = btrfs_inode_mode(eb, inode_item);
1627 if (S_ISDIR(mode)) {
1628 ret = replay_dir_deletes(wc->trans,
1629 root, log, path, key.objectid);
1630 BUG_ON(ret);
1631 }
1632 ret = overwrite_item(wc->trans, root, path,
1633 eb, i, &key);
1634 BUG_ON(ret);
1635
1636 /* for regular files, truncate away
1637 * extents past the new EOF
1638 */
1639 if (S_ISREG(mode)) {
1640 inode = read_one_inode(root,
1641 key.objectid);
1642 BUG_ON(!inode);
1643
1644 ret = btrfs_truncate_inode_items(wc->trans,
1645 root, inode, inode->i_size,
1646 BTRFS_EXTENT_DATA_KEY);
1647 BUG_ON(ret);
1648 iput(inode);
1649 }
1650 ret = link_to_fixup_dir(wc->trans, root,
1651 path, key.objectid);
1652 BUG_ON(ret);
1653 }
1654 if (wc->stage < LOG_WALK_REPLAY_ALL)
1655 continue;
1656
1657 /* these keys are simply copied */
1658 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1659 ret = overwrite_item(wc->trans, root, path,
1660 eb, i, &key);
1661 BUG_ON(ret);
1662 } else if (key.type == BTRFS_INODE_REF_KEY) {
1663 ret = add_inode_ref(wc->trans, root, log, path,
1664 eb, i, &key);
1665 BUG_ON(ret && ret != -ENOENT);
1666 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1667 ret = replay_one_extent(wc->trans, root, path,
1668 eb, i, &key);
1669 BUG_ON(ret);
1670 } else if (key.type == BTRFS_CSUM_ITEM_KEY) {
1671 ret = replay_one_csum(wc->trans, root, path,
1672 eb, i, &key);
1673 BUG_ON(ret);
1674 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1675 key.type == BTRFS_DIR_INDEX_KEY) {
1676 ret = replay_one_dir_item(wc->trans, root, path,
1677 eb, i, &key);
1678 BUG_ON(ret);
1679 }
1680 }
1681 btrfs_free_path(path);
1682 return 0;
1683}
1684
1685static int noinline walk_down_log_tree(struct btrfs_trans_handle *trans,
1686 struct btrfs_root *root,
1687 struct btrfs_path *path, int *level,
1688 struct walk_control *wc)
1689{
1690 u64 root_owner;
1691 u64 root_gen;
1692 u64 bytenr;
1693 u64 ptr_gen;
1694 struct extent_buffer *next;
1695 struct extent_buffer *cur;
1696 struct extent_buffer *parent;
1697 u32 blocksize;
1698 int ret = 0;
1699
1700 WARN_ON(*level < 0);
1701 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1702
1703 while(*level > 0) {
1704 WARN_ON(*level < 0);
1705 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1706 cur = path->nodes[*level];
1707
1708 if (btrfs_header_level(cur) != *level)
1709 WARN_ON(1);
1710
1711 if (path->slots[*level] >=
1712 btrfs_header_nritems(cur))
1713 break;
1714
1715 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1716 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1717 blocksize = btrfs_level_size(root, *level - 1);
1718
1719 parent = path->nodes[*level];
1720 root_owner = btrfs_header_owner(parent);
1721 root_gen = btrfs_header_generation(parent);
1722
1723 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1724
1725 wc->process_func(root, next, wc, ptr_gen);
1726
1727 if (*level == 1) {
1728 path->slots[*level]++;
1729 if (wc->free) {
1730 btrfs_read_buffer(next, ptr_gen);
1731
1732 btrfs_tree_lock(next);
1733 clean_tree_block(trans, root, next);
1734 btrfs_wait_tree_block_writeback(next);
1735 btrfs_tree_unlock(next);
1736
1737 ret = btrfs_drop_leaf_ref(trans, root, next);
1738 BUG_ON(ret);
1739
1740 WARN_ON(root_owner !=
1741 BTRFS_TREE_LOG_OBJECTID);
1742 ret = btrfs_free_reserved_extent(root,
1743 bytenr, blocksize);
1744 BUG_ON(ret);
1745 }
1746 free_extent_buffer(next);
1747 continue;
1748 }
1749 btrfs_read_buffer(next, ptr_gen);
1750
1751 WARN_ON(*level <= 0);
1752 if (path->nodes[*level-1])
1753 free_extent_buffer(path->nodes[*level-1]);
1754 path->nodes[*level-1] = next;
1755 *level = btrfs_header_level(next);
1756 path->slots[*level] = 0;
1757 cond_resched();
1758 }
1759 WARN_ON(*level < 0);
1760 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1761
1762 if (path->nodes[*level] == root->node) {
1763 parent = path->nodes[*level];
1764 } else {
1765 parent = path->nodes[*level + 1];
1766 }
1767 bytenr = path->nodes[*level]->start;
1768
1769 blocksize = btrfs_level_size(root, *level);
1770 root_owner = btrfs_header_owner(parent);
1771 root_gen = btrfs_header_generation(parent);
1772
1773 wc->process_func(root, path->nodes[*level], wc,
1774 btrfs_header_generation(path->nodes[*level]));
1775
1776 if (wc->free) {
1777 next = path->nodes[*level];
1778 btrfs_tree_lock(next);
1779 clean_tree_block(trans, root, next);
1780 btrfs_wait_tree_block_writeback(next);
1781 btrfs_tree_unlock(next);
1782
1783 if (*level == 0) {
1784 ret = btrfs_drop_leaf_ref(trans, root, next);
1785 BUG_ON(ret);
1786 }
1787 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1788 ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
1789 BUG_ON(ret);
1790 }
1791 free_extent_buffer(path->nodes[*level]);
1792 path->nodes[*level] = NULL;
1793 *level += 1;
1794
1795 cond_resched();
1796 return 0;
1797}
1798
1799static int noinline walk_up_log_tree(struct btrfs_trans_handle *trans,
1800 struct btrfs_root *root,
1801 struct btrfs_path *path, int *level,
1802 struct walk_control *wc)
1803{
1804 u64 root_owner;
1805 u64 root_gen;
1806 int i;
1807 int slot;
1808 int ret;
1809
1810 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1811 slot = path->slots[i];
1812 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1813 struct extent_buffer *node;
1814 node = path->nodes[i];
1815 path->slots[i]++;
1816 *level = i;
1817 WARN_ON(*level == 0);
1818 return 0;
1819 } else {
1820 struct extent_buffer *parent;
1821 if (path->nodes[*level] == root->node)
1822 parent = path->nodes[*level];
1823 else
1824 parent = path->nodes[*level + 1];
1825
1826 root_owner = btrfs_header_owner(parent);
1827 root_gen = btrfs_header_generation(parent);
1828 wc->process_func(root, path->nodes[*level], wc,
1829 btrfs_header_generation(path->nodes[*level]));
1830 if (wc->free) {
1831 struct extent_buffer *next;
1832
1833 next = path->nodes[*level];
1834
1835 btrfs_tree_lock(next);
1836 clean_tree_block(trans, root, next);
1837 btrfs_wait_tree_block_writeback(next);
1838 btrfs_tree_unlock(next);
1839
1840 if (*level == 0) {
1841 ret = btrfs_drop_leaf_ref(trans, root,
1842 next);
1843 BUG_ON(ret);
1844 }
1845
1846 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1847 ret = btrfs_free_reserved_extent(root,
1848 path->nodes[*level]->start,
1849 path->nodes[*level]->len);
1850 BUG_ON(ret);
1851 }
1852 free_extent_buffer(path->nodes[*level]);
1853 path->nodes[*level] = NULL;
1854 *level = i + 1;
1855 }
1856 }
1857 return 1;
1858}
1859
1860/*
1861 * drop the reference count on the tree rooted at 'snap'. This traverses
1862 * the tree freeing any blocks that have a ref count of zero after being
1863 * decremented.
1864 */
1865static int walk_log_tree(struct btrfs_trans_handle *trans,
1866 struct btrfs_root *log, struct walk_control *wc)
1867{
1868 int ret = 0;
1869 int wret;
1870 int level;
1871 struct btrfs_path *path;
1872 int i;
1873 int orig_level;
1874
1875 path = btrfs_alloc_path();
1876 BUG_ON(!path);
1877
1878 level = btrfs_header_level(log->node);
1879 orig_level = level;
1880 path->nodes[level] = log->node;
1881 extent_buffer_get(log->node);
1882 path->slots[level] = 0;
1883
1884 while(1) {
1885 wret = walk_down_log_tree(trans, log, path, &level, wc);
1886 if (wret > 0)
1887 break;
1888 if (wret < 0)
1889 ret = wret;
1890
1891 wret = walk_up_log_tree(trans, log, path, &level, wc);
1892 if (wret > 0)
1893 break;
1894 if (wret < 0)
1895 ret = wret;
1896 }
1897
1898 /* was the root node processed? if not, catch it here */
1899 if (path->nodes[orig_level]) {
1900 wc->process_func(log, path->nodes[orig_level], wc,
1901 btrfs_header_generation(path->nodes[orig_level]));
1902 if (wc->free) {
1903 struct extent_buffer *next;
1904
1905 next = path->nodes[orig_level];
1906
1907 btrfs_tree_lock(next);
1908 clean_tree_block(trans, log, next);
1909 btrfs_wait_tree_block_writeback(next);
1910 btrfs_tree_unlock(next);
1911
1912 if (orig_level == 0) {
1913 ret = btrfs_drop_leaf_ref(trans, log,
1914 next);
1915 BUG_ON(ret);
1916 }
1917 WARN_ON(log->root_key.objectid !=
1918 BTRFS_TREE_LOG_OBJECTID);
1919 ret = btrfs_free_reserved_extent(log, next->start,
1920 next->len);
1921 BUG_ON(ret);
1922 }
1923 }
1924
1925 for (i = 0; i <= orig_level; i++) {
1926 if (path->nodes[i]) {
1927 free_extent_buffer(path->nodes[i]);
1928 path->nodes[i] = NULL;
1929 }
1930 }
1931 btrfs_free_path(path);
1932 if (wc->free)
1933 free_extent_buffer(log->node);
1934 return ret;
1935}
1936
1937int wait_log_commit(struct btrfs_root *log)
1938{
1939 DEFINE_WAIT(wait);
1940 u64 transid = log->fs_info->tree_log_transid;
1941
1942 do {
1943 prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1944 TASK_UNINTERRUPTIBLE);
1945 mutex_unlock(&log->fs_info->tree_log_mutex);
1946 if (atomic_read(&log->fs_info->tree_log_commit))
1947 schedule();
1948 finish_wait(&log->fs_info->tree_log_wait, &wait);
1949 mutex_lock(&log->fs_info->tree_log_mutex);
1950 } while(transid == log->fs_info->tree_log_transid &&
1951 atomic_read(&log->fs_info->tree_log_commit));
1952 return 0;
1953}
1954
1955/*
1956 * btrfs_sync_log does sends a given tree log down to the disk and
1957 * updates the super blocks to record it. When this call is done,
1958 * you know that any inodes previously logged are safely on disk
1959 */
1960int btrfs_sync_log(struct btrfs_trans_handle *trans,
1961 struct btrfs_root *root)
1962{
1963 int ret;
1964 unsigned long batch;
1965 struct btrfs_root *log = root->log_root;
1966
1967 mutex_lock(&log->fs_info->tree_log_mutex);
1968 if (atomic_read(&log->fs_info->tree_log_commit)) {
1969 wait_log_commit(log);
1970 goto out;
1971 }
1972 atomic_set(&log->fs_info->tree_log_commit, 1);
1973
1974 while(1) {
1975 batch = log->fs_info->tree_log_batch;
1976 mutex_unlock(&log->fs_info->tree_log_mutex);
1977 schedule_timeout_uninterruptible(1);
1978 mutex_lock(&log->fs_info->tree_log_mutex);
1979
1980 while(atomic_read(&log->fs_info->tree_log_writers)) {
1981 DEFINE_WAIT(wait);
1982 prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1983 TASK_UNINTERRUPTIBLE);
1984 mutex_unlock(&log->fs_info->tree_log_mutex);
1985 if (atomic_read(&log->fs_info->tree_log_writers))
1986 schedule();
1987 mutex_lock(&log->fs_info->tree_log_mutex);
1988 finish_wait(&log->fs_info->tree_log_wait, &wait);
1989 }
1990 if (batch == log->fs_info->tree_log_batch)
1991 break;
1992 }
1993
1994 ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages);
1995 BUG_ON(ret);
1996 ret = btrfs_write_and_wait_marked_extents(root->fs_info->log_root_tree,
1997 &root->fs_info->log_root_tree->dirty_log_pages);
1998 BUG_ON(ret);
1999
2000 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
2001 log->fs_info->log_root_tree->node->start);
2002 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
2003 btrfs_header_level(log->fs_info->log_root_tree->node));
2004
2005 write_ctree_super(trans, log->fs_info->tree_root);
2006 log->fs_info->tree_log_transid++;
2007 log->fs_info->tree_log_batch = 0;
2008 atomic_set(&log->fs_info->tree_log_commit, 0);
2009 smp_mb();
2010 if (waitqueue_active(&log->fs_info->tree_log_wait))
2011 wake_up(&log->fs_info->tree_log_wait);
2012out:
2013 mutex_unlock(&log->fs_info->tree_log_mutex);
2014 return 0;
2015
2016}
2017
2018/* * free all the extents used by the tree log. This should be called
2019 * at commit time of the full transaction
2020 */
2021int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2022{
2023 int ret;
2024 struct btrfs_root *log;
2025 struct key;
2026 u64 start;
2027 u64 end;
2028 struct walk_control wc = {
2029 .free = 1,
2030 .process_func = process_one_buffer
2031 };
2032
2033 if (!root->log_root)
2034 return 0;
2035
2036 log = root->log_root;
2037 ret = walk_log_tree(trans, log, &wc);
2038 BUG_ON(ret);
2039
2040 while(1) {
2041 ret = find_first_extent_bit(&log->dirty_log_pages,
2042 0, &start, &end, EXTENT_DIRTY);
2043 if (ret)
2044 break;
2045
2046 clear_extent_dirty(&log->dirty_log_pages,
2047 start, end, GFP_NOFS);
2048 }
2049
2050 log = root->log_root;
2051 ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
2052 &log->root_key);
2053 BUG_ON(ret);
2054 root->log_root = NULL;
2055 kfree(root->log_root);
2056 return 0;
2057}
2058
2059/*
2060 * helper function to update the item for a given subvolumes log root
2061 * in the tree of log roots
2062 */
2063static int update_log_root(struct btrfs_trans_handle *trans,
2064 struct btrfs_root *log)
2065{
2066 u64 bytenr = btrfs_root_bytenr(&log->root_item);
2067 int ret;
2068
2069 if (log->node->start == bytenr)
2070 return 0;
2071
2072 btrfs_set_root_bytenr(&log->root_item, log->node->start);
2073 btrfs_set_root_generation(&log->root_item, trans->transid);
2074 btrfs_set_root_level(&log->root_item, btrfs_header_level(log->node));
2075 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2076 &log->root_key, &log->root_item);
2077 BUG_ON(ret);
2078 return ret;
2079}
2080
2081/*
2082 * If both a file and directory are logged, and unlinks or renames are
2083 * mixed in, we have a few interesting corners:
2084 *
2085 * create file X in dir Y
2086 * link file X to X.link in dir Y
2087 * fsync file X
2088 * unlink file X but leave X.link
2089 * fsync dir Y
2090 *
2091 * After a crash we would expect only X.link to exist. But file X
2092 * didn't get fsync'd again so the log has back refs for X and X.link.
2093 *
2094 * We solve this by removing directory entries and inode backrefs from the
2095 * log when a file that was logged in the current transaction is
2096 * unlinked. Any later fsync will include the updated log entries, and
2097 * we'll be able to reconstruct the proper directory items from backrefs.
2098 *
2099 * This optimizations allows us to avoid relogging the entire inode
2100 * or the entire directory.
2101 */
2102int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2103 struct btrfs_root *root,
2104 const char *name, int name_len,
2105 struct inode *dir, u64 index)
2106{
2107 struct btrfs_root *log;
2108 struct btrfs_dir_item *di;
2109 struct btrfs_path *path;
2110 int ret;
2111 int bytes_del = 0;
2112
2113 if (BTRFS_I(dir)->logged_trans < trans->transid)
2114 return 0;
2115
2116 ret = join_running_log_trans(root);
2117 if (ret)
2118 return 0;
2119
2120 mutex_lock(&BTRFS_I(dir)->log_mutex);
2121
2122 log = root->log_root;
2123 path = btrfs_alloc_path();
2124 di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2125 name, name_len, -1);
2126 if (di && !IS_ERR(di)) {
2127 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2128 bytes_del += name_len;
2129 BUG_ON(ret);
2130 }
2131 btrfs_release_path(log, path);
2132 di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2133 index, name, name_len, -1);
2134 if (di && !IS_ERR(di)) {
2135 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2136 bytes_del += name_len;
2137 BUG_ON(ret);
2138 }
2139
2140 /* update the directory size in the log to reflect the names
2141 * we have removed
2142 */
2143 if (bytes_del) {
2144 struct btrfs_key key;
2145
2146 key.objectid = dir->i_ino;
2147 key.offset = 0;
2148 key.type = BTRFS_INODE_ITEM_KEY;
2149 btrfs_release_path(log, path);
2150
2151 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2152 if (ret == 0) {
2153 struct btrfs_inode_item *item;
2154 u64 i_size;
2155
2156 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2157 struct btrfs_inode_item);
2158 i_size = btrfs_inode_size(path->nodes[0], item);
2159 if (i_size > bytes_del)
2160 i_size -= bytes_del;
2161 else
2162 i_size = 0;
2163 btrfs_set_inode_size(path->nodes[0], item, i_size);
2164 btrfs_mark_buffer_dirty(path->nodes[0]);
2165 } else
2166 ret = 0;
2167 btrfs_release_path(log, path);
2168 }
2169
2170 btrfs_free_path(path);
2171 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2172 end_log_trans(root);
2173
2174 return 0;
2175}
2176
2177/* see comments for btrfs_del_dir_entries_in_log */
2178int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2179 struct btrfs_root *root,
2180 const char *name, int name_len,
2181 struct inode *inode, u64 dirid)
2182{
2183 struct btrfs_root *log;
2184 u64 index;
2185 int ret;
2186
2187 if (BTRFS_I(inode)->logged_trans < trans->transid)
2188 return 0;
2189
2190 ret = join_running_log_trans(root);
2191 if (ret)
2192 return 0;
2193 log = root->log_root;
2194 mutex_lock(&BTRFS_I(inode)->log_mutex);
2195
2196 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2197 dirid, &index);
2198 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2199 end_log_trans(root);
2200
2201 return ret;
2202}
2203
2204/*
2205 * creates a range item in the log for 'dirid'. first_offset and
2206 * last_offset tell us which parts of the key space the log should
2207 * be considered authoritative for.
2208 */
2209static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2210 struct btrfs_root *log,
2211 struct btrfs_path *path,
2212 int key_type, u64 dirid,
2213 u64 first_offset, u64 last_offset)
2214{
2215 int ret;
2216 struct btrfs_key key;
2217 struct btrfs_dir_log_item *item;
2218
2219 key.objectid = dirid;
2220 key.offset = first_offset;
2221 if (key_type == BTRFS_DIR_ITEM_KEY)
2222 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2223 else
2224 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2225 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2226 BUG_ON(ret);
2227
2228 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2229 struct btrfs_dir_log_item);
2230 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2231 btrfs_mark_buffer_dirty(path->nodes[0]);
2232 btrfs_release_path(log, path);
2233 return 0;
2234}
2235
2236/*
2237 * log all the items included in the current transaction for a given
2238 * directory. This also creates the range items in the log tree required
2239 * to replay anything deleted before the fsync
2240 */
2241static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2242 struct btrfs_root *root, struct inode *inode,
2243 struct btrfs_path *path,
2244 struct btrfs_path *dst_path, int key_type,
2245 u64 min_offset, u64 *last_offset_ret)
2246{
2247 struct btrfs_key min_key;
2248 struct btrfs_key max_key;
2249 struct btrfs_root *log = root->log_root;
2250 struct extent_buffer *src;
2251 int ret;
2252 int i;
2253 int nritems;
2254 u64 first_offset = min_offset;
2255 u64 last_offset = (u64)-1;
2256
2257 log = root->log_root;
2258 max_key.objectid = inode->i_ino;
2259 max_key.offset = (u64)-1;
2260 max_key.type = key_type;
2261
2262 min_key.objectid = inode->i_ino;
2263 min_key.type = key_type;
2264 min_key.offset = min_offset;
2265
2266 path->keep_locks = 1;
2267
2268 ret = btrfs_search_forward(root, &min_key, &max_key,
2269 path, 0, trans->transid);
2270
2271 /*
2272 * we didn't find anything from this transaction, see if there
2273 * is anything at all
2274 */
2275 if (ret != 0 || min_key.objectid != inode->i_ino ||
2276 min_key.type != key_type) {
2277 min_key.objectid = inode->i_ino;
2278 min_key.type = key_type;
2279 min_key.offset = (u64)-1;
2280 btrfs_release_path(root, path);
2281 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2282 if (ret < 0) {
2283 btrfs_release_path(root, path);
2284 return ret;
2285 }
2286 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2287
2288 /* if ret == 0 there are items for this type,
2289 * create a range to tell us the last key of this type.
2290 * otherwise, there are no items in this directory after
2291 * *min_offset, and we create a range to indicate that.
2292 */
2293 if (ret == 0) {
2294 struct btrfs_key tmp;
2295 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2296 path->slots[0]);
2297 if (key_type == tmp.type) {
2298 first_offset = max(min_offset, tmp.offset) + 1;
2299 }
2300 }
2301 goto done;
2302 }
2303
2304 /* go backward to find any previous key */
2305 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2306 if (ret == 0) {
2307 struct btrfs_key tmp;
2308 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2309 if (key_type == tmp.type) {
2310 first_offset = tmp.offset;
2311 ret = overwrite_item(trans, log, dst_path,
2312 path->nodes[0], path->slots[0],
2313 &tmp);
2314 }
2315 }
2316 btrfs_release_path(root, path);
2317
2318 /* find the first key from this transaction again */
2319 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2320 if (ret != 0) {
2321 WARN_ON(1);
2322 goto done;
2323 }
2324
2325 /*
2326 * we have a block from this transaction, log every item in it
2327 * from our directory
2328 */
2329 while(1) {
2330 struct btrfs_key tmp;
2331 src = path->nodes[0];
2332 nritems = btrfs_header_nritems(src);
2333 for (i = path->slots[0]; i < nritems; i++) {
2334 btrfs_item_key_to_cpu(src, &min_key, i);
2335
2336 if (min_key.objectid != inode->i_ino ||
2337 min_key.type != key_type)
2338 goto done;
2339 ret = overwrite_item(trans, log, dst_path, src, i,
2340 &min_key);
2341 BUG_ON(ret);
2342 }
2343 path->slots[0] = nritems;
2344
2345 /*
2346 * look ahead to the next item and see if it is also
2347 * from this directory and from this transaction
2348 */
2349 ret = btrfs_next_leaf(root, path);
2350 if (ret == 1) {
2351 last_offset = (u64)-1;
2352 goto done;
2353 }
2354 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2355 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2356 last_offset = (u64)-1;
2357 goto done;
2358 }
2359 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2360 ret = overwrite_item(trans, log, dst_path,
2361 path->nodes[0], path->slots[0],
2362 &tmp);
2363
2364 BUG_ON(ret);
2365 last_offset = tmp.offset;
2366 goto done;
2367 }
2368 }
2369done:
2370 *last_offset_ret = last_offset;
2371 btrfs_release_path(root, path);
2372 btrfs_release_path(log, dst_path);
2373
2374 /* insert the log range keys to indicate where the log is valid */
2375 ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
2376 first_offset, last_offset);
2377 BUG_ON(ret);
2378 return 0;
2379}
2380
2381/*
2382 * logging directories is very similar to logging inodes, We find all the items
2383 * from the current transaction and write them to the log.
2384 *
2385 * The recovery code scans the directory in the subvolume, and if it finds a
2386 * key in the range logged that is not present in the log tree, then it means
2387 * that dir entry was unlinked during the transaction.
2388 *
2389 * In order for that scan to work, we must include one key smaller than
2390 * the smallest logged by this transaction and one key larger than the largest
2391 * key logged by this transaction.
2392 */
2393static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2394 struct btrfs_root *root, struct inode *inode,
2395 struct btrfs_path *path,
2396 struct btrfs_path *dst_path)
2397{
2398 u64 min_key;
2399 u64 max_key;
2400 int ret;
2401 int key_type = BTRFS_DIR_ITEM_KEY;
2402
2403again:
2404 min_key = 0;
2405 max_key = 0;
2406 while(1) {
2407 ret = log_dir_items(trans, root, inode, path,
2408 dst_path, key_type, min_key,
2409 &max_key);
2410 BUG_ON(ret);
2411 if (max_key == (u64)-1)
2412 break;
2413 min_key = max_key + 1;
2414 }
2415
2416 if (key_type == BTRFS_DIR_ITEM_KEY) {
2417 key_type = BTRFS_DIR_INDEX_KEY;
2418 goto again;
2419 }
2420 return 0;
2421}
2422
2423/*
2424 * a helper function to drop items from the log before we relog an
2425 * inode. max_key_type indicates the highest item type to remove.
2426 * This cannot be run for file data extents because it does not
2427 * free the extents they point to.
2428 */
2429static int drop_objectid_items(struct btrfs_trans_handle *trans,
2430 struct btrfs_root *log,
2431 struct btrfs_path *path,
2432 u64 objectid, int max_key_type)
2433{
2434 int ret;
2435 struct btrfs_key key;
2436 struct btrfs_key found_key;
2437
2438 key.objectid = objectid;
2439 key.type = max_key_type;
2440 key.offset = (u64)-1;
2441
2442 while(1) {
2443 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2444
2445 if (ret != 1)
2446 break;
2447
2448 if (path->slots[0] == 0)
2449 break;
2450
2451 path->slots[0]--;
2452 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2453 path->slots[0]);
2454
2455 if (found_key.objectid != objectid)
2456 break;
2457
2458 ret = btrfs_del_item(trans, log, path);
2459 BUG_ON(ret);
2460 btrfs_release_path(log, path);
2461 }
2462 btrfs_release_path(log, path);
2463 return 0;
2464}
2465
2466static noinline int copy_items(struct btrfs_trans_handle *trans,
2467 struct btrfs_root *log,
2468 struct btrfs_path *dst_path,
2469 struct extent_buffer *src,
2470 int start_slot, int nr, int inode_only)
2471{
2472 unsigned long src_offset;
2473 unsigned long dst_offset;
2474 struct btrfs_file_extent_item *extent;
2475 struct btrfs_inode_item *inode_item;
2476 int ret;
2477 struct btrfs_key *ins_keys;
2478 u32 *ins_sizes;
2479 char *ins_data;
2480 int i;
2481
2482 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2483 nr * sizeof(u32), GFP_NOFS);
2484 ins_sizes = (u32 *)ins_data;
2485 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2486
2487 for (i = 0; i < nr; i++) {
2488 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2489 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2490 }
2491 ret = btrfs_insert_empty_items(trans, log, dst_path,
2492 ins_keys, ins_sizes, nr);
2493 BUG_ON(ret);
2494
2495 for (i = 0; i < nr; i++) {
2496 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2497 dst_path->slots[0]);
2498
2499 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2500
2501 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2502 src_offset, ins_sizes[i]);
2503
2504 if (inode_only == LOG_INODE_EXISTS &&
2505 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2506 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2507 dst_path->slots[0],
2508 struct btrfs_inode_item);
2509 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2510
2511 /* set the generation to zero so the recover code
2512 * can tell the difference between an logging
2513 * just to say 'this inode exists' and a logging
2514 * to say 'update this inode with these values'
2515 */
2516 btrfs_set_inode_generation(dst_path->nodes[0],
2517 inode_item, 0);
2518 }
2519 /* take a reference on file data extents so that truncates
2520 * or deletes of this inode don't have to relog the inode
2521 * again
2522 */
2523 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2524 int found_type;
2525 extent = btrfs_item_ptr(src, start_slot + i,
2526 struct btrfs_file_extent_item);
2527
2528 found_type = btrfs_file_extent_type(src, extent);
2529 if (found_type == BTRFS_FILE_EXTENT_REG ||
2530 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
2531 u64 ds = btrfs_file_extent_disk_bytenr(src,
2532 extent);
2533 u64 dl = btrfs_file_extent_disk_num_bytes(src,
2534 extent);
2535 /* ds == 0 is a hole */
2536 if (ds != 0) {
2537 ret = btrfs_inc_extent_ref(trans, log,
2538 ds, dl,
2539 dst_path->nodes[0]->start,
2540 BTRFS_TREE_LOG_OBJECTID,
2541 trans->transid,
2542 ins_keys[i].objectid);
2543 BUG_ON(ret);
2544 }
2545 }
2546 }
2547 dst_path->slots[0]++;
2548 }
2549
2550 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2551 btrfs_release_path(log, dst_path);
2552 kfree(ins_data);
2553 return 0;
2554}
2555
2556/* log a single inode in the tree log.
2557 * At least one parent directory for this inode must exist in the tree
2558 * or be logged already.
2559 *
2560 * Any items from this inode changed by the current transaction are copied
2561 * to the log tree. An extra reference is taken on any extents in this
2562 * file, allowing us to avoid a whole pile of corner cases around logging
2563 * blocks that have been removed from the tree.
2564 *
2565 * See LOG_INODE_ALL and related defines for a description of what inode_only
2566 * does.
2567 *
2568 * This handles both files and directories.
2569 */
2570static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
2571 struct btrfs_root *root, struct inode *inode,
2572 int inode_only)
2573{
2574 struct btrfs_path *path;
2575 struct btrfs_path *dst_path;
2576 struct btrfs_key min_key;
2577 struct btrfs_key max_key;
2578 struct btrfs_root *log = root->log_root;
2579 struct extent_buffer *src = NULL;
2580 u32 size;
2581 int ret;
2582 int nritems;
2583 int ins_start_slot = 0;
2584 int ins_nr;
2585
2586 log = root->log_root;
2587
2588 path = btrfs_alloc_path();
2589 dst_path = btrfs_alloc_path();
2590
2591 min_key.objectid = inode->i_ino;
2592 min_key.type = BTRFS_INODE_ITEM_KEY;
2593 min_key.offset = 0;
2594
2595 max_key.objectid = inode->i_ino;
2596 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2597 max_key.type = BTRFS_XATTR_ITEM_KEY;
2598 else
2599 max_key.type = (u8)-1;
2600 max_key.offset = (u64)-1;
2601
2602 /*
2603 * if this inode has already been logged and we're in inode_only
2604 * mode, we don't want to delete the things that have already
2605 * been written to the log.
2606 *
2607 * But, if the inode has been through an inode_only log,
2608 * the logged_trans field is not set. This allows us to catch
2609 * any new names for this inode in the backrefs by logging it
2610 * again
2611 */
2612 if (inode_only == LOG_INODE_EXISTS &&
2613 BTRFS_I(inode)->logged_trans == trans->transid) {
2614 btrfs_free_path(path);
2615 btrfs_free_path(dst_path);
2616 goto out;
2617 }
2618 mutex_lock(&BTRFS_I(inode)->log_mutex);
2619
2620 /*
2621 * a brute force approach to making sure we get the most uptodate
2622 * copies of everything.
2623 */
2624 if (S_ISDIR(inode->i_mode)) {
2625 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2626
2627 if (inode_only == LOG_INODE_EXISTS)
2628 max_key_type = BTRFS_XATTR_ITEM_KEY;
2629 ret = drop_objectid_items(trans, log, path,
2630 inode->i_ino, max_key_type);
2631 } else {
2632 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2633 }
2634 BUG_ON(ret);
2635 path->keep_locks = 1;
2636
2637 while(1) {
2638 ins_nr = 0;
2639 ret = btrfs_search_forward(root, &min_key, &max_key,
2640 path, 0, trans->transid);
2641 if (ret != 0)
2642 break;
2643again:
2644 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2645 if (min_key.objectid != inode->i_ino)
2646 break;
2647 if (min_key.type > max_key.type)
2648 break;
2649
2650 src = path->nodes[0];
2651 size = btrfs_item_size_nr(src, path->slots[0]);
2652 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2653 ins_nr++;
2654 goto next_slot;
2655 } else if (!ins_nr) {
2656 ins_start_slot = path->slots[0];
2657 ins_nr = 1;
2658 goto next_slot;
2659 }
2660
2661 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2662 ins_nr, inode_only);
2663 BUG_ON(ret);
2664 ins_nr = 1;
2665 ins_start_slot = path->slots[0];
2666next_slot:
2667
2668 nritems = btrfs_header_nritems(path->nodes[0]);
2669 path->slots[0]++;
2670 if (path->slots[0] < nritems) {
2671 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2672 path->slots[0]);
2673 goto again;
2674 }
2675 if (ins_nr) {
2676 ret = copy_items(trans, log, dst_path, src,
2677 ins_start_slot,
2678 ins_nr, inode_only);
2679 BUG_ON(ret);
2680 ins_nr = 0;
2681 }
2682 btrfs_release_path(root, path);
2683
2684 if (min_key.offset < (u64)-1)
2685 min_key.offset++;
2686 else if (min_key.type < (u8)-1)
2687 min_key.type++;
2688 else if (min_key.objectid < (u64)-1)
2689 min_key.objectid++;
2690 else
2691 break;
2692 }
2693 if (ins_nr) {
2694 ret = copy_items(trans, log, dst_path, src,
2695 ins_start_slot,
2696 ins_nr, inode_only);
2697 BUG_ON(ret);
2698 ins_nr = 0;
2699 }
2700 WARN_ON(ins_nr);
2701 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2702 btrfs_release_path(root, path);
2703 btrfs_release_path(log, dst_path);
2704 BTRFS_I(inode)->log_dirty_trans = 0;
2705 ret = log_directory_changes(trans, root, inode, path, dst_path);
2706 BUG_ON(ret);
2707 }
2708 BTRFS_I(inode)->logged_trans = trans->transid;
2709 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2710
2711 btrfs_free_path(path);
2712 btrfs_free_path(dst_path);
2713
2714 mutex_lock(&root->fs_info->tree_log_mutex);
2715 ret = update_log_root(trans, log);
2716 BUG_ON(ret);
2717 mutex_unlock(&root->fs_info->tree_log_mutex);
2718out:
2719 return 0;
2720}
2721
2722int btrfs_log_inode(struct btrfs_trans_handle *trans,
2723 struct btrfs_root *root, struct inode *inode,
2724 int inode_only)
2725{
2726 int ret;
2727
2728 start_log_trans(trans, root);
2729 ret = __btrfs_log_inode(trans, root, inode, inode_only);
2730 end_log_trans(root);
2731 return ret;
2732}
2733
2734/*
2735 * helper function around btrfs_log_inode to make sure newly created
2736 * parent directories also end up in the log. A minimal inode and backref
2737 * only logging is done of any parent directories that are older than
2738 * the last committed transaction
2739 */
2740int btrfs_log_dentry(struct btrfs_trans_handle *trans,
2741 struct btrfs_root *root, struct dentry *dentry)
2742{
2743 int inode_only = LOG_INODE_ALL;
2744 struct super_block *sb;
2745 int ret;
2746
2747 start_log_trans(trans, root);
2748 sb = dentry->d_inode->i_sb;
2749 while(1) {
2750 ret = __btrfs_log_inode(trans, root, dentry->d_inode,
2751 inode_only);
2752 BUG_ON(ret);
2753 inode_only = LOG_INODE_EXISTS;
2754
2755 dentry = dentry->d_parent;
2756 if (!dentry || !dentry->d_inode || sb != dentry->d_inode->i_sb)
2757 break;
2758
2759 if (BTRFS_I(dentry->d_inode)->generation <=
2760 root->fs_info->last_trans_committed)
2761 break;
2762 }
2763 end_log_trans(root);
2764 return 0;
2765}
2766
2767/*
2768 * it is not safe to log dentry if the chunk root has added new
2769 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
2770 * If this returns 1, you must commit the transaction to safely get your
2771 * data on disk.
2772 */
2773int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
2774 struct btrfs_root *root, struct dentry *dentry)
2775{
2776 u64 gen;
2777 gen = root->fs_info->last_trans_new_blockgroup;
2778 if (gen > root->fs_info->last_trans_committed)
2779 return 1;
2780 else
2781 return btrfs_log_dentry(trans, root, dentry);
2782}
2783
2784/*
2785 * should be called during mount to recover any replay any log trees
2786 * from the FS
2787 */
2788int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
2789{
2790 int ret;
2791 struct btrfs_path *path;
2792 struct btrfs_trans_handle *trans;
2793 struct btrfs_key key;
2794 struct btrfs_key found_key;
2795 struct btrfs_key tmp_key;
2796 struct btrfs_root *log;
2797 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
2798 u64 highest_inode;
2799 struct walk_control wc = {
2800 .process_func = process_one_buffer,
2801 .stage = 0,
2802 };
2803
2804 fs_info->log_root_recovering = 1;
2805 path = btrfs_alloc_path();
2806 BUG_ON(!path);
2807
2808 trans = btrfs_start_transaction(fs_info->tree_root, 1);
2809
2810 wc.trans = trans;
2811 wc.pin = 1;
2812
2813 walk_log_tree(trans, log_root_tree, &wc);
2814
2815again:
2816 key.objectid = BTRFS_TREE_LOG_OBJECTID;
2817 key.offset = (u64)-1;
2818 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2819
2820 while(1) {
2821 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
2822 if (ret < 0)
2823 break;
2824 if (ret > 0) {
2825 if (path->slots[0] == 0)
2826 break;
2827 path->slots[0]--;
2828 }
2829 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2830 path->slots[0]);
2831 btrfs_release_path(log_root_tree, path);
2832 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
2833 break;
2834
2835 log = btrfs_read_fs_root_no_radix(log_root_tree,
2836 &found_key);
2837 BUG_ON(!log);
2838
2839
2840 tmp_key.objectid = found_key.offset;
2841 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
2842 tmp_key.offset = (u64)-1;
2843
2844 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
2845
2846 BUG_ON(!wc.replay_dest);
2847
2848 btrfs_record_root_in_trans(wc.replay_dest);
2849 ret = walk_log_tree(trans, log, &wc);
2850 BUG_ON(ret);
2851
2852 if (wc.stage == LOG_WALK_REPLAY_ALL) {
2853 ret = fixup_inode_link_counts(trans, wc.replay_dest,
2854 path);
2855 BUG_ON(ret);
2856 }
2857 ret = btrfs_find_highest_inode(wc.replay_dest, &highest_inode);
2858 if (ret == 0) {
2859 wc.replay_dest->highest_inode = highest_inode;
2860 wc.replay_dest->last_inode_alloc = highest_inode;
2861 }
2862
2863 key.offset = found_key.offset - 1;
2864 free_extent_buffer(log->node);
2865 kfree(log);
2866
2867 if (found_key.offset == 0)
2868 break;
2869 }
2870 btrfs_release_path(log_root_tree, path);
2871
2872 /* step one is to pin it all, step two is to replay just inodes */
2873 if (wc.pin) {
2874 wc.pin = 0;
2875 wc.process_func = replay_one_buffer;
2876 wc.stage = LOG_WALK_REPLAY_INODES;
2877 goto again;
2878 }
2879 /* step three is to replay everything */
2880 if (wc.stage < LOG_WALK_REPLAY_ALL) {
2881 wc.stage++;
2882 goto again;
2883 }
2884
2885 btrfs_free_path(path);
2886
2887 free_extent_buffer(log_root_tree->node);
2888 log_root_tree->log_root = NULL;
2889 fs_info->log_root_recovering = 0;
2890
2891 /* step 4: commit the transaction, which also unpins the blocks */
2892 btrfs_commit_transaction(trans, fs_info->tree_root);
2893
2894 kfree(log_root_tree);
2895 return 0;
2896}
generated by cgit v1.2.2 (git 2.25.0) at 2025-09-01 20:25:36 -0400