2 * Copyright (C) 2008 Oracle. All rights reserved.
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.
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.
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.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/blkdev.h>
22 #include <linux/list_sort.h>
26 #include "print-tree.h"
29 #include "compression.h"
31 /* magic values for the inode_only field in btrfs_log_inode:
33 * LOG_INODE_ALL means to log everything
34 * LOG_INODE_EXISTS means to log just enough to recreate the inode
37 #define LOG_INODE_ALL 0
38 #define LOG_INODE_EXISTS 1
41 * directory trouble cases
43 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
44 * log, we must force a full commit before doing an fsync of the directory
45 * where the unlink was done.
46 * ---> record transid of last unlink/rename per directory
50 * rename foo/some_dir foo2/some_dir
52 * fsync foo/some_dir/some_file
54 * The fsync above will unlink the original some_dir without recording
55 * it in its new location (foo2). After a crash, some_dir will be gone
56 * unless the fsync of some_file forces a full commit
58 * 2) we must log any new names for any file or dir that is in the fsync
59 * log. ---> check inode while renaming/linking.
61 * 2a) we must log any new names for any file or dir during rename
62 * when the directory they are being removed from was logged.
63 * ---> check inode and old parent dir during rename
65 * 2a is actually the more important variant. With the extra logging
66 * a crash might unlink the old name without recreating the new one
68 * 3) after a crash, we must go through any directories with a link count
69 * of zero and redo the rm -rf
76 * The directory f1 was fully removed from the FS, but fsync was never
77 * called on f1, only its parent dir. After a crash the rm -rf must
78 * be replayed. This must be able to recurse down the entire
79 * directory tree. The inode link count fixup code takes care of the
84 * stages for the tree walking. The first
85 * stage (0) is to only pin down the blocks we find
86 * the second stage (1) is to make sure that all the inodes
87 * we find in the log are created in the subvolume.
89 * The last stage is to deal with directories and links and extents
90 * and all the other fun semantics
92 #define LOG_WALK_PIN_ONLY 0
93 #define LOG_WALK_REPLAY_INODES 1
94 #define LOG_WALK_REPLAY_DIR_INDEX 2
95 #define LOG_WALK_REPLAY_ALL 3
97 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode,
102 struct btrfs_log_ctx *ctx);
103 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
104 struct btrfs_root *root,
105 struct btrfs_path *path, u64 objectid);
106 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
107 struct btrfs_root *root,
108 struct btrfs_root *log,
109 struct btrfs_path *path,
110 u64 dirid, int del_all);
113 * tree logging is a special write ahead log used to make sure that
114 * fsyncs and O_SYNCs can happen without doing full tree commits.
116 * Full tree commits are expensive because they require commonly
117 * modified blocks to be recowed, creating many dirty pages in the
118 * extent tree an 4x-6x higher write load than ext3.
120 * Instead of doing a tree commit on every fsync, we use the
121 * key ranges and transaction ids to find items for a given file or directory
122 * that have changed in this transaction. Those items are copied into
123 * a special tree (one per subvolume root), that tree is written to disk
124 * and then the fsync is considered complete.
126 * After a crash, items are copied out of the log-tree back into the
127 * subvolume tree. Any file data extents found are recorded in the extent
128 * allocation tree, and the log-tree freed.
130 * The log tree is read three times, once to pin down all the extents it is
131 * using in ram and once, once to create all the inodes logged in the tree
132 * and once to do all the other items.
136 * start a sub transaction and setup the log tree
137 * this increments the log tree writer count to make the people
138 * syncing the tree wait for us to finish
140 static int start_log_trans(struct btrfs_trans_handle *trans,
141 struct btrfs_root *root,
142 struct btrfs_log_ctx *ctx)
146 mutex_lock(&root->log_mutex);
148 if (root->log_root) {
149 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
154 if (!root->log_start_pid) {
155 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
156 root->log_start_pid = current->pid;
157 } else if (root->log_start_pid != current->pid) {
158 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
161 mutex_lock(&root->fs_info->tree_log_mutex);
162 if (!root->fs_info->log_root_tree)
163 ret = btrfs_init_log_root_tree(trans, root->fs_info);
164 mutex_unlock(&root->fs_info->tree_log_mutex);
168 ret = btrfs_add_log_tree(trans, root);
172 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
173 root->log_start_pid = current->pid;
176 atomic_inc(&root->log_batch);
177 atomic_inc(&root->log_writers);
179 int index = root->log_transid % 2;
180 list_add_tail(&ctx->list, &root->log_ctxs[index]);
181 ctx->log_transid = root->log_transid;
185 mutex_unlock(&root->log_mutex);
190 * returns 0 if there was a log transaction running and we were able
191 * to join, or returns -ENOENT if there were not transactions
194 static int join_running_log_trans(struct btrfs_root *root)
202 mutex_lock(&root->log_mutex);
203 if (root->log_root) {
205 atomic_inc(&root->log_writers);
207 mutex_unlock(&root->log_mutex);
212 * This either makes the current running log transaction wait
213 * until you call btrfs_end_log_trans() or it makes any future
214 * log transactions wait until you call btrfs_end_log_trans()
216 int btrfs_pin_log_trans(struct btrfs_root *root)
220 mutex_lock(&root->log_mutex);
221 atomic_inc(&root->log_writers);
222 mutex_unlock(&root->log_mutex);
227 * indicate we're done making changes to the log tree
228 * and wake up anyone waiting to do a sync
230 void btrfs_end_log_trans(struct btrfs_root *root)
232 if (atomic_dec_and_test(&root->log_writers)) {
234 * Implicit memory barrier after atomic_dec_and_test
236 if (waitqueue_active(&root->log_writer_wait))
237 wake_up(&root->log_writer_wait);
243 * the walk control struct is used to pass state down the chain when
244 * processing the log tree. The stage field tells us which part
245 * of the log tree processing we are currently doing. The others
246 * are state fields used for that specific part
248 struct walk_control {
249 /* should we free the extent on disk when done? This is used
250 * at transaction commit time while freeing a log tree
254 /* should we write out the extent buffer? This is used
255 * while flushing the log tree to disk during a sync
259 /* should we wait for the extent buffer io to finish? Also used
260 * while flushing the log tree to disk for a sync
264 /* pin only walk, we record which extents on disk belong to the
269 /* what stage of the replay code we're currently in */
272 /* the root we are currently replaying */
273 struct btrfs_root *replay_dest;
275 /* the trans handle for the current replay */
276 struct btrfs_trans_handle *trans;
278 /* the function that gets used to process blocks we find in the
279 * tree. Note the extent_buffer might not be up to date when it is
280 * passed in, and it must be checked or read if you need the data
283 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
284 struct walk_control *wc, u64 gen);
288 * process_func used to pin down extents, write them or wait on them
290 static int process_one_buffer(struct btrfs_root *log,
291 struct extent_buffer *eb,
292 struct walk_control *wc, u64 gen)
297 * If this fs is mixed then we need to be able to process the leaves to
298 * pin down any logged extents, so we have to read the block.
300 if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) {
301 ret = btrfs_read_buffer(eb, gen);
307 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
310 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
311 if (wc->pin && btrfs_header_level(eb) == 0)
312 ret = btrfs_exclude_logged_extents(log, eb);
314 btrfs_write_tree_block(eb);
316 btrfs_wait_tree_block_writeback(eb);
322 * Item overwrite used by replay and tree logging. eb, slot and key all refer
323 * to the src data we are copying out.
325 * root is the tree we are copying into, and path is a scratch
326 * path for use in this function (it should be released on entry and
327 * will be released on exit).
329 * If the key is already in the destination tree the existing item is
330 * overwritten. If the existing item isn't big enough, it is extended.
331 * If it is too large, it is truncated.
333 * If the key isn't in the destination yet, a new item is inserted.
335 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
336 struct btrfs_root *root,
337 struct btrfs_path *path,
338 struct extent_buffer *eb, int slot,
339 struct btrfs_key *key)
343 u64 saved_i_size = 0;
344 int save_old_i_size = 0;
345 unsigned long src_ptr;
346 unsigned long dst_ptr;
347 int overwrite_root = 0;
348 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
350 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
353 item_size = btrfs_item_size_nr(eb, slot);
354 src_ptr = btrfs_item_ptr_offset(eb, slot);
356 /* look for the key in the destination tree */
357 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
364 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
366 if (dst_size != item_size)
369 if (item_size == 0) {
370 btrfs_release_path(path);
373 dst_copy = kmalloc(item_size, GFP_NOFS);
374 src_copy = kmalloc(item_size, GFP_NOFS);
375 if (!dst_copy || !src_copy) {
376 btrfs_release_path(path);
382 read_extent_buffer(eb, src_copy, src_ptr, item_size);
384 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
385 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
387 ret = memcmp(dst_copy, src_copy, item_size);
392 * they have the same contents, just return, this saves
393 * us from cowing blocks in the destination tree and doing
394 * extra writes that may not have been done by a previous
398 btrfs_release_path(path);
403 * We need to load the old nbytes into the inode so when we
404 * replay the extents we've logged we get the right nbytes.
407 struct btrfs_inode_item *item;
411 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
412 struct btrfs_inode_item);
413 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
414 item = btrfs_item_ptr(eb, slot,
415 struct btrfs_inode_item);
416 btrfs_set_inode_nbytes(eb, item, nbytes);
419 * If this is a directory we need to reset the i_size to
420 * 0 so that we can set it up properly when replaying
421 * the rest of the items in this log.
423 mode = btrfs_inode_mode(eb, item);
425 btrfs_set_inode_size(eb, item, 0);
427 } else if (inode_item) {
428 struct btrfs_inode_item *item;
432 * New inode, set nbytes to 0 so that the nbytes comes out
433 * properly when we replay the extents.
435 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
436 btrfs_set_inode_nbytes(eb, item, 0);
439 * If this is a directory we need to reset the i_size to 0 so
440 * that we can set it up properly when replaying the rest of
441 * the items in this log.
443 mode = btrfs_inode_mode(eb, item);
445 btrfs_set_inode_size(eb, item, 0);
448 btrfs_release_path(path);
449 /* try to insert the key into the destination tree */
450 path->skip_release_on_error = 1;
451 ret = btrfs_insert_empty_item(trans, root, path,
453 path->skip_release_on_error = 0;
455 /* make sure any existing item is the correct size */
456 if (ret == -EEXIST || ret == -EOVERFLOW) {
458 found_size = btrfs_item_size_nr(path->nodes[0],
460 if (found_size > item_size)
461 btrfs_truncate_item(root, path, item_size, 1);
462 else if (found_size < item_size)
463 btrfs_extend_item(root, path,
464 item_size - found_size);
468 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
471 /* don't overwrite an existing inode if the generation number
472 * was logged as zero. This is done when the tree logging code
473 * is just logging an inode to make sure it exists after recovery.
475 * Also, don't overwrite i_size on directories during replay.
476 * log replay inserts and removes directory items based on the
477 * state of the tree found in the subvolume, and i_size is modified
480 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
481 struct btrfs_inode_item *src_item;
482 struct btrfs_inode_item *dst_item;
484 src_item = (struct btrfs_inode_item *)src_ptr;
485 dst_item = (struct btrfs_inode_item *)dst_ptr;
487 if (btrfs_inode_generation(eb, src_item) == 0) {
488 struct extent_buffer *dst_eb = path->nodes[0];
489 const u64 ino_size = btrfs_inode_size(eb, src_item);
492 * For regular files an ino_size == 0 is used only when
493 * logging that an inode exists, as part of a directory
494 * fsync, and the inode wasn't fsynced before. In this
495 * case don't set the size of the inode in the fs/subvol
496 * tree, otherwise we would be throwing valid data away.
498 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
499 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
501 struct btrfs_map_token token;
503 btrfs_init_map_token(&token);
504 btrfs_set_token_inode_size(dst_eb, dst_item,
510 if (overwrite_root &&
511 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
512 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
514 saved_i_size = btrfs_inode_size(path->nodes[0],
519 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
522 if (save_old_i_size) {
523 struct btrfs_inode_item *dst_item;
524 dst_item = (struct btrfs_inode_item *)dst_ptr;
525 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
528 /* make sure the generation is filled in */
529 if (key->type == BTRFS_INODE_ITEM_KEY) {
530 struct btrfs_inode_item *dst_item;
531 dst_item = (struct btrfs_inode_item *)dst_ptr;
532 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
533 btrfs_set_inode_generation(path->nodes[0], dst_item,
538 btrfs_mark_buffer_dirty(path->nodes[0]);
539 btrfs_release_path(path);
544 * simple helper to read an inode off the disk from a given root
545 * This can only be called for subvolume roots and not for the log
547 static noinline struct inode *read_one_inode(struct btrfs_root *root,
550 struct btrfs_key key;
553 key.objectid = objectid;
554 key.type = BTRFS_INODE_ITEM_KEY;
556 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
559 } else if (is_bad_inode(inode)) {
566 /* replays a single extent in 'eb' at 'slot' with 'key' into the
567 * subvolume 'root'. path is released on entry and should be released
570 * extents in the log tree have not been allocated out of the extent
571 * tree yet. So, this completes the allocation, taking a reference
572 * as required if the extent already exists or creating a new extent
573 * if it isn't in the extent allocation tree yet.
575 * The extent is inserted into the file, dropping any existing extents
576 * from the file that overlap the new one.
578 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
579 struct btrfs_root *root,
580 struct btrfs_path *path,
581 struct extent_buffer *eb, int slot,
582 struct btrfs_key *key)
586 u64 start = key->offset;
588 struct btrfs_file_extent_item *item;
589 struct inode *inode = NULL;
593 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
594 found_type = btrfs_file_extent_type(eb, item);
596 if (found_type == BTRFS_FILE_EXTENT_REG ||
597 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
598 nbytes = btrfs_file_extent_num_bytes(eb, item);
599 extent_end = start + nbytes;
602 * We don't add to the inodes nbytes if we are prealloc or a
605 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
607 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
608 size = btrfs_file_extent_inline_len(eb, slot, item);
609 nbytes = btrfs_file_extent_ram_bytes(eb, item);
610 extent_end = ALIGN(start + size, root->sectorsize);
616 inode = read_one_inode(root, key->objectid);
623 * first check to see if we already have this extent in the
624 * file. This must be done before the btrfs_drop_extents run
625 * so we don't try to drop this extent.
627 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
631 (found_type == BTRFS_FILE_EXTENT_REG ||
632 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
633 struct btrfs_file_extent_item cmp1;
634 struct btrfs_file_extent_item cmp2;
635 struct btrfs_file_extent_item *existing;
636 struct extent_buffer *leaf;
638 leaf = path->nodes[0];
639 existing = btrfs_item_ptr(leaf, path->slots[0],
640 struct btrfs_file_extent_item);
642 read_extent_buffer(eb, &cmp1, (unsigned long)item,
644 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
648 * we already have a pointer to this exact extent,
649 * we don't have to do anything
651 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
652 btrfs_release_path(path);
656 btrfs_release_path(path);
658 /* drop any overlapping extents */
659 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
663 if (found_type == BTRFS_FILE_EXTENT_REG ||
664 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
666 unsigned long dest_offset;
667 struct btrfs_key ins;
669 ret = btrfs_insert_empty_item(trans, root, path, key,
673 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
675 copy_extent_buffer(path->nodes[0], eb, dest_offset,
676 (unsigned long)item, sizeof(*item));
678 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
679 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
680 ins.type = BTRFS_EXTENT_ITEM_KEY;
681 offset = key->offset - btrfs_file_extent_offset(eb, item);
683 if (ins.objectid > 0) {
686 LIST_HEAD(ordered_sums);
688 * is this extent already allocated in the extent
689 * allocation tree? If so, just add a reference
691 ret = btrfs_lookup_data_extent(root, ins.objectid,
694 ret = btrfs_inc_extent_ref(trans, root,
695 ins.objectid, ins.offset,
696 0, root->root_key.objectid,
697 key->objectid, offset);
702 * insert the extent pointer in the extent
705 ret = btrfs_alloc_logged_file_extent(trans,
706 root, root->root_key.objectid,
707 key->objectid, offset, &ins);
711 btrfs_release_path(path);
713 if (btrfs_file_extent_compression(eb, item)) {
714 csum_start = ins.objectid;
715 csum_end = csum_start + ins.offset;
717 csum_start = ins.objectid +
718 btrfs_file_extent_offset(eb, item);
719 csum_end = csum_start +
720 btrfs_file_extent_num_bytes(eb, item);
723 ret = btrfs_lookup_csums_range(root->log_root,
724 csum_start, csum_end - 1,
729 * Now delete all existing cums in the csum root that
730 * cover our range. We do this because we can have an
731 * extent that is completely referenced by one file
732 * extent item and partially referenced by another
733 * file extent item (like after using the clone or
734 * extent_same ioctls). In this case if we end up doing
735 * the replay of the one that partially references the
736 * extent first, and we do not do the csum deletion
737 * below, we can get 2 csum items in the csum tree that
738 * overlap each other. For example, imagine our log has
739 * the two following file extent items:
741 * key (257 EXTENT_DATA 409600)
742 * extent data disk byte 12845056 nr 102400
743 * extent data offset 20480 nr 20480 ram 102400
745 * key (257 EXTENT_DATA 819200)
746 * extent data disk byte 12845056 nr 102400
747 * extent data offset 0 nr 102400 ram 102400
749 * Where the second one fully references the 100K extent
750 * that starts at disk byte 12845056, and the log tree
751 * has a single csum item that covers the entire range
754 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
756 * After the first file extent item is replayed, the
757 * csum tree gets the following csum item:
759 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
761 * Which covers the 20K sub-range starting at offset 20K
762 * of our extent. Now when we replay the second file
763 * extent item, if we do not delete existing csum items
764 * that cover any of its blocks, we end up getting two
765 * csum items in our csum tree that overlap each other:
767 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
768 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
770 * Which is a problem, because after this anyone trying
771 * to lookup up for the checksum of any block of our
772 * extent starting at an offset of 40K or higher, will
773 * end up looking at the second csum item only, which
774 * does not contain the checksum for any block starting
775 * at offset 40K or higher of our extent.
777 while (!list_empty(&ordered_sums)) {
778 struct btrfs_ordered_sum *sums;
779 sums = list_entry(ordered_sums.next,
780 struct btrfs_ordered_sum,
783 ret = btrfs_del_csums(trans,
784 root->fs_info->csum_root,
788 ret = btrfs_csum_file_blocks(trans,
789 root->fs_info->csum_root,
791 list_del(&sums->list);
797 btrfs_release_path(path);
799 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
800 /* inline extents are easy, we just overwrite them */
801 ret = overwrite_item(trans, root, path, eb, slot, key);
806 inode_add_bytes(inode, nbytes);
807 ret = btrfs_update_inode(trans, root, inode);
815 * when cleaning up conflicts between the directory names in the
816 * subvolume, directory names in the log and directory names in the
817 * inode back references, we may have to unlink inodes from directories.
819 * This is a helper function to do the unlink of a specific directory
822 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
823 struct btrfs_root *root,
824 struct btrfs_path *path,
826 struct btrfs_dir_item *di)
831 struct extent_buffer *leaf;
832 struct btrfs_key location;
835 leaf = path->nodes[0];
837 btrfs_dir_item_key_to_cpu(leaf, di, &location);
838 name_len = btrfs_dir_name_len(leaf, di);
839 name = kmalloc(name_len, GFP_NOFS);
843 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
844 btrfs_release_path(path);
846 inode = read_one_inode(root, location.objectid);
852 ret = link_to_fixup_dir(trans, root, path, location.objectid);
856 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
860 ret = btrfs_run_delayed_items(trans, root);
868 * helper function to see if a given name and sequence number found
869 * in an inode back reference are already in a directory and correctly
870 * point to this inode
872 static noinline int inode_in_dir(struct btrfs_root *root,
873 struct btrfs_path *path,
874 u64 dirid, u64 objectid, u64 index,
875 const char *name, int name_len)
877 struct btrfs_dir_item *di;
878 struct btrfs_key location;
881 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
882 index, name, name_len, 0);
883 if (di && !IS_ERR(di)) {
884 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
885 if (location.objectid != objectid)
889 btrfs_release_path(path);
891 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
892 if (di && !IS_ERR(di)) {
893 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
894 if (location.objectid != objectid)
900 btrfs_release_path(path);
905 * helper function to check a log tree for a named back reference in
906 * an inode. This is used to decide if a back reference that is
907 * found in the subvolume conflicts with what we find in the log.
909 * inode backreferences may have multiple refs in a single item,
910 * during replay we process one reference at a time, and we don't
911 * want to delete valid links to a file from the subvolume if that
912 * link is also in the log.
914 static noinline int backref_in_log(struct btrfs_root *log,
915 struct btrfs_key *key,
917 const char *name, int namelen)
919 struct btrfs_path *path;
920 struct btrfs_inode_ref *ref;
922 unsigned long ptr_end;
923 unsigned long name_ptr;
929 path = btrfs_alloc_path();
933 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
937 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
939 if (key->type == BTRFS_INODE_EXTREF_KEY) {
940 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
941 name, namelen, NULL))
947 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
948 ptr_end = ptr + item_size;
949 while (ptr < ptr_end) {
950 ref = (struct btrfs_inode_ref *)ptr;
951 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
952 if (found_name_len == namelen) {
953 name_ptr = (unsigned long)(ref + 1);
954 ret = memcmp_extent_buffer(path->nodes[0], name,
961 ptr = (unsigned long)(ref + 1) + found_name_len;
964 btrfs_free_path(path);
968 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
969 struct btrfs_root *root,
970 struct btrfs_path *path,
971 struct btrfs_root *log_root,
972 struct inode *dir, struct inode *inode,
973 struct extent_buffer *eb,
974 u64 inode_objectid, u64 parent_objectid,
975 u64 ref_index, char *name, int namelen,
981 struct extent_buffer *leaf;
982 struct btrfs_dir_item *di;
983 struct btrfs_key search_key;
984 struct btrfs_inode_extref *extref;
987 /* Search old style refs */
988 search_key.objectid = inode_objectid;
989 search_key.type = BTRFS_INODE_REF_KEY;
990 search_key.offset = parent_objectid;
991 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
993 struct btrfs_inode_ref *victim_ref;
995 unsigned long ptr_end;
997 leaf = path->nodes[0];
999 /* are we trying to overwrite a back ref for the root directory
1000 * if so, just jump out, we're done
1002 if (search_key.objectid == search_key.offset)
1005 /* check all the names in this back reference to see
1006 * if they are in the log. if so, we allow them to stay
1007 * otherwise they must be unlinked as a conflict
1009 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1010 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1011 while (ptr < ptr_end) {
1012 victim_ref = (struct btrfs_inode_ref *)ptr;
1013 victim_name_len = btrfs_inode_ref_name_len(leaf,
1015 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1019 read_extent_buffer(leaf, victim_name,
1020 (unsigned long)(victim_ref + 1),
1023 if (!backref_in_log(log_root, &search_key,
1028 btrfs_release_path(path);
1030 ret = btrfs_unlink_inode(trans, root, dir,
1036 ret = btrfs_run_delayed_items(trans, root);
1044 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1048 * NOTE: we have searched root tree and checked the
1049 * corresponding ref, it does not need to check again.
1053 btrfs_release_path(path);
1055 /* Same search but for extended refs */
1056 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1057 inode_objectid, parent_objectid, 0,
1059 if (!IS_ERR_OR_NULL(extref)) {
1063 struct inode *victim_parent;
1065 leaf = path->nodes[0];
1067 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1068 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1070 while (cur_offset < item_size) {
1071 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1073 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1075 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1078 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1081 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1084 search_key.objectid = inode_objectid;
1085 search_key.type = BTRFS_INODE_EXTREF_KEY;
1086 search_key.offset = btrfs_extref_hash(parent_objectid,
1090 if (!backref_in_log(log_root, &search_key,
1091 parent_objectid, victim_name,
1094 victim_parent = read_one_inode(root,
1096 if (victim_parent) {
1098 btrfs_release_path(path);
1100 ret = btrfs_unlink_inode(trans, root,
1106 ret = btrfs_run_delayed_items(
1109 iput(victim_parent);
1120 cur_offset += victim_name_len + sizeof(*extref);
1124 btrfs_release_path(path);
1126 /* look for a conflicting sequence number */
1127 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1128 ref_index, name, namelen, 0);
1129 if (di && !IS_ERR(di)) {
1130 ret = drop_one_dir_item(trans, root, path, dir, di);
1134 btrfs_release_path(path);
1136 /* look for a conflicing name */
1137 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1139 if (di && !IS_ERR(di)) {
1140 ret = drop_one_dir_item(trans, root, path, dir, di);
1144 btrfs_release_path(path);
1149 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1150 u32 *namelen, char **name, u64 *index,
1151 u64 *parent_objectid)
1153 struct btrfs_inode_extref *extref;
1155 extref = (struct btrfs_inode_extref *)ref_ptr;
1157 *namelen = btrfs_inode_extref_name_len(eb, extref);
1158 *name = kmalloc(*namelen, GFP_NOFS);
1162 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1165 *index = btrfs_inode_extref_index(eb, extref);
1166 if (parent_objectid)
1167 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1172 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1173 u32 *namelen, char **name, u64 *index)
1175 struct btrfs_inode_ref *ref;
1177 ref = (struct btrfs_inode_ref *)ref_ptr;
1179 *namelen = btrfs_inode_ref_name_len(eb, ref);
1180 *name = kmalloc(*namelen, GFP_NOFS);
1184 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1186 *index = btrfs_inode_ref_index(eb, ref);
1192 * replay one inode back reference item found in the log tree.
1193 * eb, slot and key refer to the buffer and key found in the log tree.
1194 * root is the destination we are replaying into, and path is for temp
1195 * use by this function. (it should be released on return).
1197 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1198 struct btrfs_root *root,
1199 struct btrfs_root *log,
1200 struct btrfs_path *path,
1201 struct extent_buffer *eb, int slot,
1202 struct btrfs_key *key)
1204 struct inode *dir = NULL;
1205 struct inode *inode = NULL;
1206 unsigned long ref_ptr;
1207 unsigned long ref_end;
1211 int search_done = 0;
1212 int log_ref_ver = 0;
1213 u64 parent_objectid;
1216 int ref_struct_size;
1218 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1219 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1221 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1222 struct btrfs_inode_extref *r;
1224 ref_struct_size = sizeof(struct btrfs_inode_extref);
1226 r = (struct btrfs_inode_extref *)ref_ptr;
1227 parent_objectid = btrfs_inode_extref_parent(eb, r);
1229 ref_struct_size = sizeof(struct btrfs_inode_ref);
1230 parent_objectid = key->offset;
1232 inode_objectid = key->objectid;
1235 * it is possible that we didn't log all the parent directories
1236 * for a given inode. If we don't find the dir, just don't
1237 * copy the back ref in. The link count fixup code will take
1240 dir = read_one_inode(root, parent_objectid);
1246 inode = read_one_inode(root, inode_objectid);
1252 while (ref_ptr < ref_end) {
1254 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1255 &ref_index, &parent_objectid);
1257 * parent object can change from one array
1261 dir = read_one_inode(root, parent_objectid);
1267 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1273 /* if we already have a perfect match, we're done */
1274 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
1275 ref_index, name, namelen)) {
1277 * look for a conflicting back reference in the
1278 * metadata. if we find one we have to unlink that name
1279 * of the file before we add our new link. Later on, we
1280 * overwrite any existing back reference, and we don't
1281 * want to create dangling pointers in the directory.
1285 ret = __add_inode_ref(trans, root, path, log,
1289 ref_index, name, namelen,
1298 /* insert our name */
1299 ret = btrfs_add_link(trans, dir, inode, name, namelen,
1304 btrfs_update_inode(trans, root, inode);
1307 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1316 /* finally write the back reference in the inode */
1317 ret = overwrite_item(trans, root, path, eb, slot, key);
1319 btrfs_release_path(path);
1326 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1327 struct btrfs_root *root, u64 ino)
1331 ret = btrfs_insert_orphan_item(trans, root, ino);
1338 static int count_inode_extrefs(struct btrfs_root *root,
1339 struct inode *inode, struct btrfs_path *path)
1343 unsigned int nlink = 0;
1346 u64 inode_objectid = btrfs_ino(inode);
1349 struct btrfs_inode_extref *extref;
1350 struct extent_buffer *leaf;
1353 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1358 leaf = path->nodes[0];
1359 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1360 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1363 while (cur_offset < item_size) {
1364 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1365 name_len = btrfs_inode_extref_name_len(leaf, extref);
1369 cur_offset += name_len + sizeof(*extref);
1373 btrfs_release_path(path);
1375 btrfs_release_path(path);
1377 if (ret < 0 && ret != -ENOENT)
1382 static int count_inode_refs(struct btrfs_root *root,
1383 struct inode *inode, struct btrfs_path *path)
1386 struct btrfs_key key;
1387 unsigned int nlink = 0;
1389 unsigned long ptr_end;
1391 u64 ino = btrfs_ino(inode);
1394 key.type = BTRFS_INODE_REF_KEY;
1395 key.offset = (u64)-1;
1398 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1402 if (path->slots[0] == 0)
1407 btrfs_item_key_to_cpu(path->nodes[0], &key,
1409 if (key.objectid != ino ||
1410 key.type != BTRFS_INODE_REF_KEY)
1412 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1413 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1415 while (ptr < ptr_end) {
1416 struct btrfs_inode_ref *ref;
1418 ref = (struct btrfs_inode_ref *)ptr;
1419 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1421 ptr = (unsigned long)(ref + 1) + name_len;
1425 if (key.offset == 0)
1427 if (path->slots[0] > 0) {
1432 btrfs_release_path(path);
1434 btrfs_release_path(path);
1440 * There are a few corners where the link count of the file can't
1441 * be properly maintained during replay. So, instead of adding
1442 * lots of complexity to the log code, we just scan the backrefs
1443 * for any file that has been through replay.
1445 * The scan will update the link count on the inode to reflect the
1446 * number of back refs found. If it goes down to zero, the iput
1447 * will free the inode.
1449 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1450 struct btrfs_root *root,
1451 struct inode *inode)
1453 struct btrfs_path *path;
1456 u64 ino = btrfs_ino(inode);
1458 path = btrfs_alloc_path();
1462 ret = count_inode_refs(root, inode, path);
1468 ret = count_inode_extrefs(root, inode, path);
1476 if (nlink != inode->i_nlink) {
1477 set_nlink(inode, nlink);
1478 btrfs_update_inode(trans, root, inode);
1480 BTRFS_I(inode)->index_cnt = (u64)-1;
1482 if (inode->i_nlink == 0) {
1483 if (S_ISDIR(inode->i_mode)) {
1484 ret = replay_dir_deletes(trans, root, NULL, path,
1489 ret = insert_orphan_item(trans, root, ino);
1493 btrfs_free_path(path);
1497 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1498 struct btrfs_root *root,
1499 struct btrfs_path *path)
1502 struct btrfs_key key;
1503 struct inode *inode;
1505 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1506 key.type = BTRFS_ORPHAN_ITEM_KEY;
1507 key.offset = (u64)-1;
1509 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1514 if (path->slots[0] == 0)
1519 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1520 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1521 key.type != BTRFS_ORPHAN_ITEM_KEY)
1524 ret = btrfs_del_item(trans, root, path);
1528 btrfs_release_path(path);
1529 inode = read_one_inode(root, key.offset);
1533 ret = fixup_inode_link_count(trans, root, inode);
1539 * fixup on a directory may create new entries,
1540 * make sure we always look for the highset possible
1543 key.offset = (u64)-1;
1547 btrfs_release_path(path);
1553 * record a given inode in the fixup dir so we can check its link
1554 * count when replay is done. The link count is incremented here
1555 * so the inode won't go away until we check it
1557 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1558 struct btrfs_root *root,
1559 struct btrfs_path *path,
1562 struct btrfs_key key;
1564 struct inode *inode;
1566 inode = read_one_inode(root, objectid);
1570 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1571 key.type = BTRFS_ORPHAN_ITEM_KEY;
1572 key.offset = objectid;
1574 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1576 btrfs_release_path(path);
1578 if (!inode->i_nlink)
1579 set_nlink(inode, 1);
1582 ret = btrfs_update_inode(trans, root, inode);
1583 } else if (ret == -EEXIST) {
1586 BUG(); /* Logic Error */
1594 * when replaying the log for a directory, we only insert names
1595 * for inodes that actually exist. This means an fsync on a directory
1596 * does not implicitly fsync all the new files in it
1598 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1599 struct btrfs_root *root,
1600 u64 dirid, u64 index,
1601 char *name, int name_len,
1602 struct btrfs_key *location)
1604 struct inode *inode;
1608 inode = read_one_inode(root, location->objectid);
1612 dir = read_one_inode(root, dirid);
1618 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1620 /* FIXME, put inode into FIXUP list */
1628 * Return true if an inode reference exists in the log for the given name,
1629 * inode and parent inode.
1631 static bool name_in_log_ref(struct btrfs_root *log_root,
1632 const char *name, const int name_len,
1633 const u64 dirid, const u64 ino)
1635 struct btrfs_key search_key;
1637 search_key.objectid = ino;
1638 search_key.type = BTRFS_INODE_REF_KEY;
1639 search_key.offset = dirid;
1640 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1643 search_key.type = BTRFS_INODE_EXTREF_KEY;
1644 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1645 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1652 * take a single entry in a log directory item and replay it into
1655 * if a conflicting item exists in the subdirectory already,
1656 * the inode it points to is unlinked and put into the link count
1659 * If a name from the log points to a file or directory that does
1660 * not exist in the FS, it is skipped. fsyncs on directories
1661 * do not force down inodes inside that directory, just changes to the
1662 * names or unlinks in a directory.
1664 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1665 * non-existing inode) and 1 if the name was replayed.
1667 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1668 struct btrfs_root *root,
1669 struct btrfs_path *path,
1670 struct extent_buffer *eb,
1671 struct btrfs_dir_item *di,
1672 struct btrfs_key *key)
1676 struct btrfs_dir_item *dst_di;
1677 struct btrfs_key found_key;
1678 struct btrfs_key log_key;
1683 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1684 bool name_added = false;
1686 dir = read_one_inode(root, key->objectid);
1690 name_len = btrfs_dir_name_len(eb, di);
1691 name = kmalloc(name_len, GFP_NOFS);
1697 log_type = btrfs_dir_type(eb, di);
1698 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1701 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1702 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1707 btrfs_release_path(path);
1709 if (key->type == BTRFS_DIR_ITEM_KEY) {
1710 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1712 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1713 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1722 if (IS_ERR_OR_NULL(dst_di)) {
1723 /* we need a sequence number to insert, so we only
1724 * do inserts for the BTRFS_DIR_INDEX_KEY types
1726 if (key->type != BTRFS_DIR_INDEX_KEY)
1731 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1732 /* the existing item matches the logged item */
1733 if (found_key.objectid == log_key.objectid &&
1734 found_key.type == log_key.type &&
1735 found_key.offset == log_key.offset &&
1736 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1737 update_size = false;
1742 * don't drop the conflicting directory entry if the inode
1743 * for the new entry doesn't exist
1748 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1752 if (key->type == BTRFS_DIR_INDEX_KEY)
1755 btrfs_release_path(path);
1756 if (!ret && update_size) {
1757 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1758 ret = btrfs_update_inode(trans, root, dir);
1762 if (!ret && name_added)
1767 if (name_in_log_ref(root->log_root, name, name_len,
1768 key->objectid, log_key.objectid)) {
1769 /* The dentry will be added later. */
1771 update_size = false;
1774 btrfs_release_path(path);
1775 ret = insert_one_name(trans, root, key->objectid, key->offset,
1776 name, name_len, &log_key);
1777 if (ret && ret != -ENOENT && ret != -EEXIST)
1781 update_size = false;
1787 * find all the names in a directory item and reconcile them into
1788 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1789 * one name in a directory item, but the same code gets used for
1790 * both directory index types
1792 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1793 struct btrfs_root *root,
1794 struct btrfs_path *path,
1795 struct extent_buffer *eb, int slot,
1796 struct btrfs_key *key)
1799 u32 item_size = btrfs_item_size_nr(eb, slot);
1800 struct btrfs_dir_item *di;
1803 unsigned long ptr_end;
1804 struct btrfs_path *fixup_path = NULL;
1806 ptr = btrfs_item_ptr_offset(eb, slot);
1807 ptr_end = ptr + item_size;
1808 while (ptr < ptr_end) {
1809 di = (struct btrfs_dir_item *)ptr;
1810 if (verify_dir_item(root, eb, di))
1812 name_len = btrfs_dir_name_len(eb, di);
1813 ret = replay_one_name(trans, root, path, eb, di, key);
1816 ptr = (unsigned long)(di + 1);
1820 * If this entry refers to a non-directory (directories can not
1821 * have a link count > 1) and it was added in the transaction
1822 * that was not committed, make sure we fixup the link count of
1823 * the inode it the entry points to. Otherwise something like
1824 * the following would result in a directory pointing to an
1825 * inode with a wrong link that does not account for this dir
1833 * ln testdir/bar testdir/bar_link
1834 * ln testdir/foo testdir/foo_link
1835 * xfs_io -c "fsync" testdir/bar
1839 * mount fs, log replay happens
1841 * File foo would remain with a link count of 1 when it has two
1842 * entries pointing to it in the directory testdir. This would
1843 * make it impossible to ever delete the parent directory has
1844 * it would result in stale dentries that can never be deleted.
1846 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1847 struct btrfs_key di_key;
1850 fixup_path = btrfs_alloc_path();
1857 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1858 ret = link_to_fixup_dir(trans, root, fixup_path,
1865 btrfs_free_path(fixup_path);
1870 * directory replay has two parts. There are the standard directory
1871 * items in the log copied from the subvolume, and range items
1872 * created in the log while the subvolume was logged.
1874 * The range items tell us which parts of the key space the log
1875 * is authoritative for. During replay, if a key in the subvolume
1876 * directory is in a logged range item, but not actually in the log
1877 * that means it was deleted from the directory before the fsync
1878 * and should be removed.
1880 static noinline int find_dir_range(struct btrfs_root *root,
1881 struct btrfs_path *path,
1882 u64 dirid, int key_type,
1883 u64 *start_ret, u64 *end_ret)
1885 struct btrfs_key key;
1887 struct btrfs_dir_log_item *item;
1891 if (*start_ret == (u64)-1)
1894 key.objectid = dirid;
1895 key.type = key_type;
1896 key.offset = *start_ret;
1898 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1902 if (path->slots[0] == 0)
1907 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1909 if (key.type != key_type || key.objectid != dirid) {
1913 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1914 struct btrfs_dir_log_item);
1915 found_end = btrfs_dir_log_end(path->nodes[0], item);
1917 if (*start_ret >= key.offset && *start_ret <= found_end) {
1919 *start_ret = key.offset;
1920 *end_ret = found_end;
1925 /* check the next slot in the tree to see if it is a valid item */
1926 nritems = btrfs_header_nritems(path->nodes[0]);
1927 if (path->slots[0] >= nritems) {
1928 ret = btrfs_next_leaf(root, path);
1935 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1937 if (key.type != key_type || key.objectid != dirid) {
1941 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1942 struct btrfs_dir_log_item);
1943 found_end = btrfs_dir_log_end(path->nodes[0], item);
1944 *start_ret = key.offset;
1945 *end_ret = found_end;
1948 btrfs_release_path(path);
1953 * this looks for a given directory item in the log. If the directory
1954 * item is not in the log, the item is removed and the inode it points
1957 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1958 struct btrfs_root *root,
1959 struct btrfs_root *log,
1960 struct btrfs_path *path,
1961 struct btrfs_path *log_path,
1963 struct btrfs_key *dir_key)
1966 struct extent_buffer *eb;
1969 struct btrfs_dir_item *di;
1970 struct btrfs_dir_item *log_di;
1973 unsigned long ptr_end;
1975 struct inode *inode;
1976 struct btrfs_key location;
1979 eb = path->nodes[0];
1980 slot = path->slots[0];
1981 item_size = btrfs_item_size_nr(eb, slot);
1982 ptr = btrfs_item_ptr_offset(eb, slot);
1983 ptr_end = ptr + item_size;
1984 while (ptr < ptr_end) {
1985 di = (struct btrfs_dir_item *)ptr;
1986 if (verify_dir_item(root, eb, di)) {
1991 name_len = btrfs_dir_name_len(eb, di);
1992 name = kmalloc(name_len, GFP_NOFS);
1997 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2000 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2001 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2004 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2005 log_di = btrfs_lookup_dir_index_item(trans, log,
2011 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
2012 btrfs_dir_item_key_to_cpu(eb, di, &location);
2013 btrfs_release_path(path);
2014 btrfs_release_path(log_path);
2015 inode = read_one_inode(root, location.objectid);
2021 ret = link_to_fixup_dir(trans, root,
2022 path, location.objectid);
2030 ret = btrfs_unlink_inode(trans, root, dir, inode,
2033 ret = btrfs_run_delayed_items(trans, root);
2039 /* there might still be more names under this key
2040 * check and repeat if required
2042 ret = btrfs_search_slot(NULL, root, dir_key, path,
2048 } else if (IS_ERR(log_di)) {
2050 return PTR_ERR(log_di);
2052 btrfs_release_path(log_path);
2055 ptr = (unsigned long)(di + 1);
2060 btrfs_release_path(path);
2061 btrfs_release_path(log_path);
2065 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2066 struct btrfs_root *root,
2067 struct btrfs_root *log,
2068 struct btrfs_path *path,
2071 struct btrfs_key search_key;
2072 struct btrfs_path *log_path;
2077 log_path = btrfs_alloc_path();
2081 search_key.objectid = ino;
2082 search_key.type = BTRFS_XATTR_ITEM_KEY;
2083 search_key.offset = 0;
2085 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2089 nritems = btrfs_header_nritems(path->nodes[0]);
2090 for (i = path->slots[0]; i < nritems; i++) {
2091 struct btrfs_key key;
2092 struct btrfs_dir_item *di;
2093 struct btrfs_dir_item *log_di;
2097 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2098 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2103 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2104 total_size = btrfs_item_size_nr(path->nodes[0], i);
2106 while (cur < total_size) {
2107 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2108 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2109 u32 this_len = sizeof(*di) + name_len + data_len;
2112 name = kmalloc(name_len, GFP_NOFS);
2117 read_extent_buffer(path->nodes[0], name,
2118 (unsigned long)(di + 1), name_len);
2120 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2122 btrfs_release_path(log_path);
2124 /* Doesn't exist in log tree, so delete it. */
2125 btrfs_release_path(path);
2126 di = btrfs_lookup_xattr(trans, root, path, ino,
2127 name, name_len, -1);
2134 ret = btrfs_delete_one_dir_name(trans, root,
2138 btrfs_release_path(path);
2143 if (IS_ERR(log_di)) {
2144 ret = PTR_ERR(log_di);
2148 di = (struct btrfs_dir_item *)((char *)di + this_len);
2151 ret = btrfs_next_leaf(root, path);
2157 btrfs_free_path(log_path);
2158 btrfs_release_path(path);
2164 * deletion replay happens before we copy any new directory items
2165 * out of the log or out of backreferences from inodes. It
2166 * scans the log to find ranges of keys that log is authoritative for,
2167 * and then scans the directory to find items in those ranges that are
2168 * not present in the log.
2170 * Anything we don't find in the log is unlinked and removed from the
2173 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2174 struct btrfs_root *root,
2175 struct btrfs_root *log,
2176 struct btrfs_path *path,
2177 u64 dirid, int del_all)
2181 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2183 struct btrfs_key dir_key;
2184 struct btrfs_key found_key;
2185 struct btrfs_path *log_path;
2188 dir_key.objectid = dirid;
2189 dir_key.type = BTRFS_DIR_ITEM_KEY;
2190 log_path = btrfs_alloc_path();
2194 dir = read_one_inode(root, dirid);
2195 /* it isn't an error if the inode isn't there, that can happen
2196 * because we replay the deletes before we copy in the inode item
2200 btrfs_free_path(log_path);
2208 range_end = (u64)-1;
2210 ret = find_dir_range(log, path, dirid, key_type,
2211 &range_start, &range_end);
2216 dir_key.offset = range_start;
2219 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2224 nritems = btrfs_header_nritems(path->nodes[0]);
2225 if (path->slots[0] >= nritems) {
2226 ret = btrfs_next_leaf(root, path);
2230 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2232 if (found_key.objectid != dirid ||
2233 found_key.type != dir_key.type)
2236 if (found_key.offset > range_end)
2239 ret = check_item_in_log(trans, root, log, path,
2244 if (found_key.offset == (u64)-1)
2246 dir_key.offset = found_key.offset + 1;
2248 btrfs_release_path(path);
2249 if (range_end == (u64)-1)
2251 range_start = range_end + 1;
2256 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2257 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2258 dir_key.type = BTRFS_DIR_INDEX_KEY;
2259 btrfs_release_path(path);
2263 btrfs_release_path(path);
2264 btrfs_free_path(log_path);
2270 * the process_func used to replay items from the log tree. This
2271 * gets called in two different stages. The first stage just looks
2272 * for inodes and makes sure they are all copied into the subvolume.
2274 * The second stage copies all the other item types from the log into
2275 * the subvolume. The two stage approach is slower, but gets rid of
2276 * lots of complexity around inodes referencing other inodes that exist
2277 * only in the log (references come from either directory items or inode
2280 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2281 struct walk_control *wc, u64 gen)
2284 struct btrfs_path *path;
2285 struct btrfs_root *root = wc->replay_dest;
2286 struct btrfs_key key;
2291 ret = btrfs_read_buffer(eb, gen);
2295 level = btrfs_header_level(eb);
2300 path = btrfs_alloc_path();
2304 nritems = btrfs_header_nritems(eb);
2305 for (i = 0; i < nritems; i++) {
2306 btrfs_item_key_to_cpu(eb, &key, i);
2308 /* inode keys are done during the first stage */
2309 if (key.type == BTRFS_INODE_ITEM_KEY &&
2310 wc->stage == LOG_WALK_REPLAY_INODES) {
2311 struct btrfs_inode_item *inode_item;
2314 inode_item = btrfs_item_ptr(eb, i,
2315 struct btrfs_inode_item);
2316 ret = replay_xattr_deletes(wc->trans, root, log,
2317 path, key.objectid);
2320 mode = btrfs_inode_mode(eb, inode_item);
2321 if (S_ISDIR(mode)) {
2322 ret = replay_dir_deletes(wc->trans,
2323 root, log, path, key.objectid, 0);
2327 ret = overwrite_item(wc->trans, root, path,
2332 /* for regular files, make sure corresponding
2333 * orphan item exist. extents past the new EOF
2334 * will be truncated later by orphan cleanup.
2336 if (S_ISREG(mode)) {
2337 ret = insert_orphan_item(wc->trans, root,
2343 ret = link_to_fixup_dir(wc->trans, root,
2344 path, key.objectid);
2349 if (key.type == BTRFS_DIR_INDEX_KEY &&
2350 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2351 ret = replay_one_dir_item(wc->trans, root, path,
2357 if (wc->stage < LOG_WALK_REPLAY_ALL)
2360 /* these keys are simply copied */
2361 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2362 ret = overwrite_item(wc->trans, root, path,
2366 } else if (key.type == BTRFS_INODE_REF_KEY ||
2367 key.type == BTRFS_INODE_EXTREF_KEY) {
2368 ret = add_inode_ref(wc->trans, root, log, path,
2370 if (ret && ret != -ENOENT)
2373 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2374 ret = replay_one_extent(wc->trans, root, path,
2378 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2379 ret = replay_one_dir_item(wc->trans, root, path,
2385 btrfs_free_path(path);
2389 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2390 struct btrfs_root *root,
2391 struct btrfs_path *path, int *level,
2392 struct walk_control *wc)
2397 struct extent_buffer *next;
2398 struct extent_buffer *cur;
2399 struct extent_buffer *parent;
2403 WARN_ON(*level < 0);
2404 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2406 while (*level > 0) {
2407 WARN_ON(*level < 0);
2408 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2409 cur = path->nodes[*level];
2411 WARN_ON(btrfs_header_level(cur) != *level);
2413 if (path->slots[*level] >=
2414 btrfs_header_nritems(cur))
2417 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2418 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2419 blocksize = root->nodesize;
2421 parent = path->nodes[*level];
2422 root_owner = btrfs_header_owner(parent);
2424 next = btrfs_find_create_tree_block(root, bytenr);
2426 return PTR_ERR(next);
2429 ret = wc->process_func(root, next, wc, ptr_gen);
2431 free_extent_buffer(next);
2435 path->slots[*level]++;
2437 ret = btrfs_read_buffer(next, ptr_gen);
2439 free_extent_buffer(next);
2444 btrfs_tree_lock(next);
2445 btrfs_set_lock_blocking(next);
2446 clean_tree_block(trans, root->fs_info,
2448 btrfs_wait_tree_block_writeback(next);
2449 btrfs_tree_unlock(next);
2452 WARN_ON(root_owner !=
2453 BTRFS_TREE_LOG_OBJECTID);
2454 ret = btrfs_free_and_pin_reserved_extent(root,
2457 free_extent_buffer(next);
2461 free_extent_buffer(next);
2464 ret = btrfs_read_buffer(next, ptr_gen);
2466 free_extent_buffer(next);
2470 WARN_ON(*level <= 0);
2471 if (path->nodes[*level-1])
2472 free_extent_buffer(path->nodes[*level-1]);
2473 path->nodes[*level-1] = next;
2474 *level = btrfs_header_level(next);
2475 path->slots[*level] = 0;
2478 WARN_ON(*level < 0);
2479 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2481 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2487 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2488 struct btrfs_root *root,
2489 struct btrfs_path *path, int *level,
2490 struct walk_control *wc)
2497 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2498 slot = path->slots[i];
2499 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2502 WARN_ON(*level == 0);
2505 struct extent_buffer *parent;
2506 if (path->nodes[*level] == root->node)
2507 parent = path->nodes[*level];
2509 parent = path->nodes[*level + 1];
2511 root_owner = btrfs_header_owner(parent);
2512 ret = wc->process_func(root, path->nodes[*level], wc,
2513 btrfs_header_generation(path->nodes[*level]));
2518 struct extent_buffer *next;
2520 next = path->nodes[*level];
2523 btrfs_tree_lock(next);
2524 btrfs_set_lock_blocking(next);
2525 clean_tree_block(trans, root->fs_info,
2527 btrfs_wait_tree_block_writeback(next);
2528 btrfs_tree_unlock(next);
2531 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2532 ret = btrfs_free_and_pin_reserved_extent(root,
2533 path->nodes[*level]->start,
2534 path->nodes[*level]->len);
2538 free_extent_buffer(path->nodes[*level]);
2539 path->nodes[*level] = NULL;
2547 * drop the reference count on the tree rooted at 'snap'. This traverses
2548 * the tree freeing any blocks that have a ref count of zero after being
2551 static int walk_log_tree(struct btrfs_trans_handle *trans,
2552 struct btrfs_root *log, struct walk_control *wc)
2557 struct btrfs_path *path;
2560 path = btrfs_alloc_path();
2564 level = btrfs_header_level(log->node);
2566 path->nodes[level] = log->node;
2567 extent_buffer_get(log->node);
2568 path->slots[level] = 0;
2571 wret = walk_down_log_tree(trans, log, path, &level, wc);
2579 wret = walk_up_log_tree(trans, log, path, &level, wc);
2588 /* was the root node processed? if not, catch it here */
2589 if (path->nodes[orig_level]) {
2590 ret = wc->process_func(log, path->nodes[orig_level], wc,
2591 btrfs_header_generation(path->nodes[orig_level]));
2595 struct extent_buffer *next;
2597 next = path->nodes[orig_level];
2600 btrfs_tree_lock(next);
2601 btrfs_set_lock_blocking(next);
2602 clean_tree_block(trans, log->fs_info, next);
2603 btrfs_wait_tree_block_writeback(next);
2604 btrfs_tree_unlock(next);
2607 WARN_ON(log->root_key.objectid !=
2608 BTRFS_TREE_LOG_OBJECTID);
2609 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2617 btrfs_free_path(path);
2622 * helper function to update the item for a given subvolumes log root
2623 * in the tree of log roots
2625 static int update_log_root(struct btrfs_trans_handle *trans,
2626 struct btrfs_root *log)
2630 if (log->log_transid == 1) {
2631 /* insert root item on the first sync */
2632 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2633 &log->root_key, &log->root_item);
2635 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2636 &log->root_key, &log->root_item);
2641 static void wait_log_commit(struct btrfs_root *root, int transid)
2644 int index = transid % 2;
2647 * we only allow two pending log transactions at a time,
2648 * so we know that if ours is more than 2 older than the
2649 * current transaction, we're done
2652 prepare_to_wait(&root->log_commit_wait[index],
2653 &wait, TASK_UNINTERRUPTIBLE);
2654 mutex_unlock(&root->log_mutex);
2656 if (root->log_transid_committed < transid &&
2657 atomic_read(&root->log_commit[index]))
2660 finish_wait(&root->log_commit_wait[index], &wait);
2661 mutex_lock(&root->log_mutex);
2662 } while (root->log_transid_committed < transid &&
2663 atomic_read(&root->log_commit[index]));
2666 static void wait_for_writer(struct btrfs_root *root)
2670 while (atomic_read(&root->log_writers)) {
2671 prepare_to_wait(&root->log_writer_wait,
2672 &wait, TASK_UNINTERRUPTIBLE);
2673 mutex_unlock(&root->log_mutex);
2674 if (atomic_read(&root->log_writers))
2676 finish_wait(&root->log_writer_wait, &wait);
2677 mutex_lock(&root->log_mutex);
2681 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2682 struct btrfs_log_ctx *ctx)
2687 mutex_lock(&root->log_mutex);
2688 list_del_init(&ctx->list);
2689 mutex_unlock(&root->log_mutex);
2693 * Invoked in log mutex context, or be sure there is no other task which
2694 * can access the list.
2696 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2697 int index, int error)
2699 struct btrfs_log_ctx *ctx;
2702 INIT_LIST_HEAD(&root->log_ctxs[index]);
2706 list_for_each_entry(ctx, &root->log_ctxs[index], list)
2707 ctx->log_ret = error;
2709 INIT_LIST_HEAD(&root->log_ctxs[index]);
2713 * btrfs_sync_log does sends a given tree log down to the disk and
2714 * updates the super blocks to record it. When this call is done,
2715 * you know that any inodes previously logged are safely on disk only
2718 * Any other return value means you need to call btrfs_commit_transaction.
2719 * Some of the edge cases for fsyncing directories that have had unlinks
2720 * or renames done in the past mean that sometimes the only safe
2721 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2722 * that has happened.
2724 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2725 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2731 struct btrfs_root *log = root->log_root;
2732 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2733 int log_transid = 0;
2734 struct btrfs_log_ctx root_log_ctx;
2735 struct blk_plug plug;
2737 mutex_lock(&root->log_mutex);
2738 log_transid = ctx->log_transid;
2739 if (root->log_transid_committed >= log_transid) {
2740 mutex_unlock(&root->log_mutex);
2741 return ctx->log_ret;
2744 index1 = log_transid % 2;
2745 if (atomic_read(&root->log_commit[index1])) {
2746 wait_log_commit(root, log_transid);
2747 mutex_unlock(&root->log_mutex);
2748 return ctx->log_ret;
2750 ASSERT(log_transid == root->log_transid);
2751 atomic_set(&root->log_commit[index1], 1);
2753 /* wait for previous tree log sync to complete */
2754 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2755 wait_log_commit(root, log_transid - 1);
2758 int batch = atomic_read(&root->log_batch);
2759 /* when we're on an ssd, just kick the log commit out */
2760 if (!btrfs_test_opt(root->fs_info, SSD) &&
2761 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2762 mutex_unlock(&root->log_mutex);
2763 schedule_timeout_uninterruptible(1);
2764 mutex_lock(&root->log_mutex);
2766 wait_for_writer(root);
2767 if (batch == atomic_read(&root->log_batch))
2771 /* bail out if we need to do a full commit */
2772 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2774 btrfs_free_logged_extents(log, log_transid);
2775 mutex_unlock(&root->log_mutex);
2779 if (log_transid % 2 == 0)
2780 mark = EXTENT_DIRTY;
2784 /* we start IO on all the marked extents here, but we don't actually
2785 * wait for them until later.
2787 blk_start_plug(&plug);
2788 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2790 blk_finish_plug(&plug);
2791 btrfs_abort_transaction(trans, ret);
2792 btrfs_free_logged_extents(log, log_transid);
2793 btrfs_set_log_full_commit(root->fs_info, trans);
2794 mutex_unlock(&root->log_mutex);
2798 btrfs_set_root_node(&log->root_item, log->node);
2800 root->log_transid++;
2801 log->log_transid = root->log_transid;
2802 root->log_start_pid = 0;
2804 * IO has been started, blocks of the log tree have WRITTEN flag set
2805 * in their headers. new modifications of the log will be written to
2806 * new positions. so it's safe to allow log writers to go in.
2808 mutex_unlock(&root->log_mutex);
2810 btrfs_init_log_ctx(&root_log_ctx);
2812 mutex_lock(&log_root_tree->log_mutex);
2813 atomic_inc(&log_root_tree->log_batch);
2814 atomic_inc(&log_root_tree->log_writers);
2816 index2 = log_root_tree->log_transid % 2;
2817 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2818 root_log_ctx.log_transid = log_root_tree->log_transid;
2820 mutex_unlock(&log_root_tree->log_mutex);
2822 ret = update_log_root(trans, log);
2824 mutex_lock(&log_root_tree->log_mutex);
2825 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2827 * Implicit memory barrier after atomic_dec_and_test
2829 if (waitqueue_active(&log_root_tree->log_writer_wait))
2830 wake_up(&log_root_tree->log_writer_wait);
2834 if (!list_empty(&root_log_ctx.list))
2835 list_del_init(&root_log_ctx.list);
2837 blk_finish_plug(&plug);
2838 btrfs_set_log_full_commit(root->fs_info, trans);
2840 if (ret != -ENOSPC) {
2841 btrfs_abort_transaction(trans, ret);
2842 mutex_unlock(&log_root_tree->log_mutex);
2845 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2846 btrfs_free_logged_extents(log, log_transid);
2847 mutex_unlock(&log_root_tree->log_mutex);
2852 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2853 blk_finish_plug(&plug);
2854 mutex_unlock(&log_root_tree->log_mutex);
2855 ret = root_log_ctx.log_ret;
2859 index2 = root_log_ctx.log_transid % 2;
2860 if (atomic_read(&log_root_tree->log_commit[index2])) {
2861 blk_finish_plug(&plug);
2862 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages,
2864 btrfs_wait_logged_extents(trans, log, log_transid);
2865 wait_log_commit(log_root_tree,
2866 root_log_ctx.log_transid);
2867 mutex_unlock(&log_root_tree->log_mutex);
2869 ret = root_log_ctx.log_ret;
2872 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
2873 atomic_set(&log_root_tree->log_commit[index2], 1);
2875 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2876 wait_log_commit(log_root_tree,
2877 root_log_ctx.log_transid - 1);
2880 wait_for_writer(log_root_tree);
2883 * now that we've moved on to the tree of log tree roots,
2884 * check the full commit flag again
2886 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2887 blk_finish_plug(&plug);
2888 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2889 btrfs_free_logged_extents(log, log_transid);
2890 mutex_unlock(&log_root_tree->log_mutex);
2892 goto out_wake_log_root;
2895 ret = btrfs_write_marked_extents(log_root_tree,
2896 &log_root_tree->dirty_log_pages,
2897 EXTENT_DIRTY | EXTENT_NEW);
2898 blk_finish_plug(&plug);
2900 btrfs_set_log_full_commit(root->fs_info, trans);
2901 btrfs_abort_transaction(trans, ret);
2902 btrfs_free_logged_extents(log, log_transid);
2903 mutex_unlock(&log_root_tree->log_mutex);
2904 goto out_wake_log_root;
2906 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2908 ret = btrfs_wait_marked_extents(log_root_tree,
2909 &log_root_tree->dirty_log_pages,
2910 EXTENT_NEW | EXTENT_DIRTY);
2912 btrfs_set_log_full_commit(root->fs_info, trans);
2913 btrfs_free_logged_extents(log, log_transid);
2914 mutex_unlock(&log_root_tree->log_mutex);
2915 goto out_wake_log_root;
2917 btrfs_wait_logged_extents(trans, log, log_transid);
2919 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2920 log_root_tree->node->start);
2921 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2922 btrfs_header_level(log_root_tree->node));
2924 log_root_tree->log_transid++;
2925 mutex_unlock(&log_root_tree->log_mutex);
2928 * nobody else is going to jump in and write the the ctree
2929 * super here because the log_commit atomic below is protecting
2930 * us. We must be called with a transaction handle pinning
2931 * the running transaction open, so a full commit can't hop
2932 * in and cause problems either.
2934 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2936 btrfs_set_log_full_commit(root->fs_info, trans);
2937 btrfs_abort_transaction(trans, ret);
2938 goto out_wake_log_root;
2941 mutex_lock(&root->log_mutex);
2942 if (root->last_log_commit < log_transid)
2943 root->last_log_commit = log_transid;
2944 mutex_unlock(&root->log_mutex);
2948 * We needn't get log_mutex here because we are sure all
2949 * the other tasks are blocked.
2951 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2953 mutex_lock(&log_root_tree->log_mutex);
2954 log_root_tree->log_transid_committed++;
2955 atomic_set(&log_root_tree->log_commit[index2], 0);
2956 mutex_unlock(&log_root_tree->log_mutex);
2959 * The barrier before waitqueue_active is implied by mutex_unlock
2961 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2962 wake_up(&log_root_tree->log_commit_wait[index2]);
2965 btrfs_remove_all_log_ctxs(root, index1, ret);
2967 mutex_lock(&root->log_mutex);
2968 root->log_transid_committed++;
2969 atomic_set(&root->log_commit[index1], 0);
2970 mutex_unlock(&root->log_mutex);
2973 * The barrier before waitqueue_active is implied by mutex_unlock
2975 if (waitqueue_active(&root->log_commit_wait[index1]))
2976 wake_up(&root->log_commit_wait[index1]);
2980 static void free_log_tree(struct btrfs_trans_handle *trans,
2981 struct btrfs_root *log)
2986 struct walk_control wc = {
2988 .process_func = process_one_buffer
2991 ret = walk_log_tree(trans, log, &wc);
2992 /* I don't think this can happen but just in case */
2994 btrfs_abort_transaction(trans, ret);
2997 ret = find_first_extent_bit(&log->dirty_log_pages,
2998 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
3003 clear_extent_bits(&log->dirty_log_pages, start, end,
3004 EXTENT_DIRTY | EXTENT_NEW);
3008 * We may have short-circuited the log tree with the full commit logic
3009 * and left ordered extents on our list, so clear these out to keep us
3010 * from leaking inodes and memory.
3012 btrfs_free_logged_extents(log, 0);
3013 btrfs_free_logged_extents(log, 1);
3015 free_extent_buffer(log->node);
3020 * free all the extents used by the tree log. This should be called
3021 * at commit time of the full transaction
3023 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3025 if (root->log_root) {
3026 free_log_tree(trans, root->log_root);
3027 root->log_root = NULL;
3032 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3033 struct btrfs_fs_info *fs_info)
3035 if (fs_info->log_root_tree) {
3036 free_log_tree(trans, fs_info->log_root_tree);
3037 fs_info->log_root_tree = NULL;
3043 * If both a file and directory are logged, and unlinks or renames are
3044 * mixed in, we have a few interesting corners:
3046 * create file X in dir Y
3047 * link file X to X.link in dir Y
3049 * unlink file X but leave X.link
3052 * After a crash we would expect only X.link to exist. But file X
3053 * didn't get fsync'd again so the log has back refs for X and X.link.
3055 * We solve this by removing directory entries and inode backrefs from the
3056 * log when a file that was logged in the current transaction is
3057 * unlinked. Any later fsync will include the updated log entries, and
3058 * we'll be able to reconstruct the proper directory items from backrefs.
3060 * This optimizations allows us to avoid relogging the entire inode
3061 * or the entire directory.
3063 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3064 struct btrfs_root *root,
3065 const char *name, int name_len,
3066 struct inode *dir, u64 index)
3068 struct btrfs_root *log;
3069 struct btrfs_dir_item *di;
3070 struct btrfs_path *path;
3074 u64 dir_ino = btrfs_ino(dir);
3076 if (BTRFS_I(dir)->logged_trans < trans->transid)
3079 ret = join_running_log_trans(root);
3083 mutex_lock(&BTRFS_I(dir)->log_mutex);
3085 log = root->log_root;
3086 path = btrfs_alloc_path();
3092 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3093 name, name_len, -1);
3099 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3100 bytes_del += name_len;
3106 btrfs_release_path(path);
3107 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3108 index, name, name_len, -1);
3114 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3115 bytes_del += name_len;
3122 /* update the directory size in the log to reflect the names
3126 struct btrfs_key key;
3128 key.objectid = dir_ino;
3130 key.type = BTRFS_INODE_ITEM_KEY;
3131 btrfs_release_path(path);
3133 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3139 struct btrfs_inode_item *item;
3142 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3143 struct btrfs_inode_item);
3144 i_size = btrfs_inode_size(path->nodes[0], item);
3145 if (i_size > bytes_del)
3146 i_size -= bytes_del;
3149 btrfs_set_inode_size(path->nodes[0], item, i_size);
3150 btrfs_mark_buffer_dirty(path->nodes[0]);
3153 btrfs_release_path(path);
3156 btrfs_free_path(path);
3158 mutex_unlock(&BTRFS_I(dir)->log_mutex);
3159 if (ret == -ENOSPC) {
3160 btrfs_set_log_full_commit(root->fs_info, trans);
3163 btrfs_abort_transaction(trans, ret);
3165 btrfs_end_log_trans(root);
3170 /* see comments for btrfs_del_dir_entries_in_log */
3171 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3172 struct btrfs_root *root,
3173 const char *name, int name_len,
3174 struct inode *inode, u64 dirid)
3176 struct btrfs_root *log;
3180 if (BTRFS_I(inode)->logged_trans < trans->transid)
3183 ret = join_running_log_trans(root);
3186 log = root->log_root;
3187 mutex_lock(&BTRFS_I(inode)->log_mutex);
3189 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3191 mutex_unlock(&BTRFS_I(inode)->log_mutex);
3192 if (ret == -ENOSPC) {
3193 btrfs_set_log_full_commit(root->fs_info, trans);
3195 } else if (ret < 0 && ret != -ENOENT)
3196 btrfs_abort_transaction(trans, ret);
3197 btrfs_end_log_trans(root);
3203 * creates a range item in the log for 'dirid'. first_offset and
3204 * last_offset tell us which parts of the key space the log should
3205 * be considered authoritative for.
3207 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3208 struct btrfs_root *log,
3209 struct btrfs_path *path,
3210 int key_type, u64 dirid,
3211 u64 first_offset, u64 last_offset)
3214 struct btrfs_key key;
3215 struct btrfs_dir_log_item *item;
3217 key.objectid = dirid;
3218 key.offset = first_offset;
3219 if (key_type == BTRFS_DIR_ITEM_KEY)
3220 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3222 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3223 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3227 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3228 struct btrfs_dir_log_item);
3229 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3230 btrfs_mark_buffer_dirty(path->nodes[0]);
3231 btrfs_release_path(path);
3236 * log all the items included in the current transaction for a given
3237 * directory. This also creates the range items in the log tree required
3238 * to replay anything deleted before the fsync
3240 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3241 struct btrfs_root *root, struct inode *inode,
3242 struct btrfs_path *path,
3243 struct btrfs_path *dst_path, int key_type,
3244 struct btrfs_log_ctx *ctx,
3245 u64 min_offset, u64 *last_offset_ret)
3247 struct btrfs_key min_key;
3248 struct btrfs_root *log = root->log_root;
3249 struct extent_buffer *src;
3254 u64 first_offset = min_offset;
3255 u64 last_offset = (u64)-1;
3256 u64 ino = btrfs_ino(inode);
3258 log = root->log_root;
3260 min_key.objectid = ino;
3261 min_key.type = key_type;
3262 min_key.offset = min_offset;
3264 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3267 * we didn't find anything from this transaction, see if there
3268 * is anything at all
3270 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3271 min_key.objectid = ino;
3272 min_key.type = key_type;
3273 min_key.offset = (u64)-1;
3274 btrfs_release_path(path);
3275 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3277 btrfs_release_path(path);
3280 ret = btrfs_previous_item(root, path, ino, key_type);
3282 /* if ret == 0 there are items for this type,
3283 * create a range to tell us the last key of this type.
3284 * otherwise, there are no items in this directory after
3285 * *min_offset, and we create a range to indicate that.
3288 struct btrfs_key tmp;
3289 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3291 if (key_type == tmp.type)
3292 first_offset = max(min_offset, tmp.offset) + 1;
3297 /* go backward to find any previous key */
3298 ret = btrfs_previous_item(root, path, ino, key_type);
3300 struct btrfs_key tmp;
3301 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3302 if (key_type == tmp.type) {
3303 first_offset = tmp.offset;
3304 ret = overwrite_item(trans, log, dst_path,
3305 path->nodes[0], path->slots[0],
3313 btrfs_release_path(path);
3315 /* find the first key from this transaction again */
3316 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3317 if (WARN_ON(ret != 0))
3321 * we have a block from this transaction, log every item in it
3322 * from our directory
3325 struct btrfs_key tmp;
3326 src = path->nodes[0];
3327 nritems = btrfs_header_nritems(src);
3328 for (i = path->slots[0]; i < nritems; i++) {
3329 struct btrfs_dir_item *di;
3331 btrfs_item_key_to_cpu(src, &min_key, i);
3333 if (min_key.objectid != ino || min_key.type != key_type)
3335 ret = overwrite_item(trans, log, dst_path, src, i,
3343 * We must make sure that when we log a directory entry,
3344 * the corresponding inode, after log replay, has a
3345 * matching link count. For example:
3351 * xfs_io -c "fsync" mydir
3353 * <mount fs and log replay>
3355 * Would result in a fsync log that when replayed, our
3356 * file inode would have a link count of 1, but we get
3357 * two directory entries pointing to the same inode.
3358 * After removing one of the names, it would not be
3359 * possible to remove the other name, which resulted
3360 * always in stale file handle errors, and would not
3361 * be possible to rmdir the parent directory, since
3362 * its i_size could never decrement to the value
3363 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3365 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3366 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3368 (btrfs_dir_transid(src, di) == trans->transid ||
3369 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3370 tmp.type != BTRFS_ROOT_ITEM_KEY)
3371 ctx->log_new_dentries = true;
3373 path->slots[0] = nritems;
3376 * look ahead to the next item and see if it is also
3377 * from this directory and from this transaction
3379 ret = btrfs_next_leaf(root, path);
3381 last_offset = (u64)-1;
3384 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3385 if (tmp.objectid != ino || tmp.type != key_type) {
3386 last_offset = (u64)-1;
3389 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3390 ret = overwrite_item(trans, log, dst_path,
3391 path->nodes[0], path->slots[0],
3396 last_offset = tmp.offset;
3401 btrfs_release_path(path);
3402 btrfs_release_path(dst_path);
3405 *last_offset_ret = last_offset;
3407 * insert the log range keys to indicate where the log
3410 ret = insert_dir_log_key(trans, log, path, key_type,
3411 ino, first_offset, last_offset);
3419 * logging directories is very similar to logging inodes, We find all the items
3420 * from the current transaction and write them to the log.
3422 * The recovery code scans the directory in the subvolume, and if it finds a
3423 * key in the range logged that is not present in the log tree, then it means
3424 * that dir entry was unlinked during the transaction.
3426 * In order for that scan to work, we must include one key smaller than
3427 * the smallest logged by this transaction and one key larger than the largest
3428 * key logged by this transaction.
3430 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3431 struct btrfs_root *root, struct inode *inode,
3432 struct btrfs_path *path,
3433 struct btrfs_path *dst_path,
3434 struct btrfs_log_ctx *ctx)
3439 int key_type = BTRFS_DIR_ITEM_KEY;
3445 ret = log_dir_items(trans, root, inode, path,
3446 dst_path, key_type, ctx, min_key,
3450 if (max_key == (u64)-1)
3452 min_key = max_key + 1;
3455 if (key_type == BTRFS_DIR_ITEM_KEY) {
3456 key_type = BTRFS_DIR_INDEX_KEY;
3463 * a helper function to drop items from the log before we relog an
3464 * inode. max_key_type indicates the highest item type to remove.
3465 * This cannot be run for file data extents because it does not
3466 * free the extents they point to.
3468 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3469 struct btrfs_root *log,
3470 struct btrfs_path *path,
3471 u64 objectid, int max_key_type)
3474 struct btrfs_key key;
3475 struct btrfs_key found_key;
3478 key.objectid = objectid;
3479 key.type = max_key_type;
3480 key.offset = (u64)-1;
3483 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3484 BUG_ON(ret == 0); /* Logic error */
3488 if (path->slots[0] == 0)
3492 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3495 if (found_key.objectid != objectid)
3498 found_key.offset = 0;
3500 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3503 ret = btrfs_del_items(trans, log, path, start_slot,
3504 path->slots[0] - start_slot + 1);
3506 * If start slot isn't 0 then we don't need to re-search, we've
3507 * found the last guy with the objectid in this tree.
3509 if (ret || start_slot != 0)
3511 btrfs_release_path(path);
3513 btrfs_release_path(path);
3519 static void fill_inode_item(struct btrfs_trans_handle *trans,
3520 struct extent_buffer *leaf,
3521 struct btrfs_inode_item *item,
3522 struct inode *inode, int log_inode_only,
3525 struct btrfs_map_token token;
3527 btrfs_init_map_token(&token);
3529 if (log_inode_only) {
3530 /* set the generation to zero so the recover code
3531 * can tell the difference between an logging
3532 * just to say 'this inode exists' and a logging
3533 * to say 'update this inode with these values'
3535 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3536 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3538 btrfs_set_token_inode_generation(leaf, item,
3539 BTRFS_I(inode)->generation,
3541 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3544 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3545 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3546 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3547 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3549 btrfs_set_token_timespec_sec(leaf, &item->atime,
3550 inode->i_atime.tv_sec, &token);
3551 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3552 inode->i_atime.tv_nsec, &token);
3554 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3555 inode->i_mtime.tv_sec, &token);
3556 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3557 inode->i_mtime.tv_nsec, &token);
3559 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3560 inode->i_ctime.tv_sec, &token);
3561 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3562 inode->i_ctime.tv_nsec, &token);
3564 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3567 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3568 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3569 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3570 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3571 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3574 static int log_inode_item(struct btrfs_trans_handle *trans,
3575 struct btrfs_root *log, struct btrfs_path *path,
3576 struct inode *inode)
3578 struct btrfs_inode_item *inode_item;
3581 ret = btrfs_insert_empty_item(trans, log, path,
3582 &BTRFS_I(inode)->location,
3583 sizeof(*inode_item));
3584 if (ret && ret != -EEXIST)
3586 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3587 struct btrfs_inode_item);
3588 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0);
3589 btrfs_release_path(path);
3593 static noinline int copy_items(struct btrfs_trans_handle *trans,
3594 struct inode *inode,
3595 struct btrfs_path *dst_path,
3596 struct btrfs_path *src_path, u64 *last_extent,
3597 int start_slot, int nr, int inode_only,
3600 unsigned long src_offset;
3601 unsigned long dst_offset;
3602 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3603 struct btrfs_file_extent_item *extent;
3604 struct btrfs_inode_item *inode_item;
3605 struct extent_buffer *src = src_path->nodes[0];
3606 struct btrfs_key first_key, last_key, key;
3608 struct btrfs_key *ins_keys;
3612 struct list_head ordered_sums;
3613 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3614 bool has_extents = false;
3615 bool need_find_last_extent = true;
3618 INIT_LIST_HEAD(&ordered_sums);
3620 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3621 nr * sizeof(u32), GFP_NOFS);
3625 first_key.objectid = (u64)-1;
3627 ins_sizes = (u32 *)ins_data;
3628 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3630 for (i = 0; i < nr; i++) {
3631 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3632 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3634 ret = btrfs_insert_empty_items(trans, log, dst_path,
3635 ins_keys, ins_sizes, nr);
3641 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3642 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3643 dst_path->slots[0]);
3645 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3647 if ((i == (nr - 1)))
3648 last_key = ins_keys[i];
3650 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3651 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3653 struct btrfs_inode_item);
3654 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3655 inode, inode_only == LOG_INODE_EXISTS,
3658 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3659 src_offset, ins_sizes[i]);
3663 * We set need_find_last_extent here in case we know we were
3664 * processing other items and then walk into the first extent in
3665 * the inode. If we don't hit an extent then nothing changes,
3666 * we'll do the last search the next time around.
3668 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3670 if (first_key.objectid == (u64)-1)
3671 first_key = ins_keys[i];
3673 need_find_last_extent = false;
3676 /* take a reference on file data extents so that truncates
3677 * or deletes of this inode don't have to relog the inode
3680 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3683 extent = btrfs_item_ptr(src, start_slot + i,
3684 struct btrfs_file_extent_item);
3686 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3689 found_type = btrfs_file_extent_type(src, extent);
3690 if (found_type == BTRFS_FILE_EXTENT_REG) {
3692 ds = btrfs_file_extent_disk_bytenr(src,
3694 /* ds == 0 is a hole */
3698 dl = btrfs_file_extent_disk_num_bytes(src,
3700 cs = btrfs_file_extent_offset(src, extent);
3701 cl = btrfs_file_extent_num_bytes(src,
3703 if (btrfs_file_extent_compression(src,
3709 ret = btrfs_lookup_csums_range(
3710 log->fs_info->csum_root,
3711 ds + cs, ds + cs + cl - 1,
3714 btrfs_release_path(dst_path);
3722 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3723 btrfs_release_path(dst_path);
3727 * we have to do this after the loop above to avoid changing the
3728 * log tree while trying to change the log tree.
3731 while (!list_empty(&ordered_sums)) {
3732 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3733 struct btrfs_ordered_sum,
3736 ret = btrfs_csum_file_blocks(trans, log, sums);
3737 list_del(&sums->list);
3744 if (need_find_last_extent && *last_extent == first_key.offset) {
3746 * We don't have any leafs between our current one and the one
3747 * we processed before that can have file extent items for our
3748 * inode (and have a generation number smaller than our current
3751 need_find_last_extent = false;
3755 * Because we use btrfs_search_forward we could skip leaves that were
3756 * not modified and then assume *last_extent is valid when it really
3757 * isn't. So back up to the previous leaf and read the end of the last
3758 * extent before we go and fill in holes.
3760 if (need_find_last_extent) {
3763 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3768 if (src_path->slots[0])
3769 src_path->slots[0]--;
3770 src = src_path->nodes[0];
3771 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3772 if (key.objectid != btrfs_ino(inode) ||
3773 key.type != BTRFS_EXTENT_DATA_KEY)
3775 extent = btrfs_item_ptr(src, src_path->slots[0],
3776 struct btrfs_file_extent_item);
3777 if (btrfs_file_extent_type(src, extent) ==
3778 BTRFS_FILE_EXTENT_INLINE) {
3779 len = btrfs_file_extent_inline_len(src,
3782 *last_extent = ALIGN(key.offset + len,
3785 len = btrfs_file_extent_num_bytes(src, extent);
3786 *last_extent = key.offset + len;
3790 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3791 * things could have happened
3793 * 1) A merge could have happened, so we could currently be on a leaf
3794 * that holds what we were copying in the first place.
3795 * 2) A split could have happened, and now not all of the items we want
3796 * are on the same leaf.
3798 * So we need to adjust how we search for holes, we need to drop the
3799 * path and re-search for the first extent key we found, and then walk
3800 * forward until we hit the last one we copied.
3802 if (need_find_last_extent) {
3803 /* btrfs_prev_leaf could return 1 without releasing the path */
3804 btrfs_release_path(src_path);
3805 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3810 src = src_path->nodes[0];
3811 i = src_path->slots[0];
3817 * Ok so here we need to go through and fill in any holes we may have
3818 * to make sure that holes are punched for those areas in case they had
3819 * extents previously.
3825 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3826 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3830 src = src_path->nodes[0];
3834 btrfs_item_key_to_cpu(src, &key, i);
3835 if (!btrfs_comp_cpu_keys(&key, &last_key))
3837 if (key.objectid != btrfs_ino(inode) ||
3838 key.type != BTRFS_EXTENT_DATA_KEY) {
3842 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3843 if (btrfs_file_extent_type(src, extent) ==
3844 BTRFS_FILE_EXTENT_INLINE) {
3845 len = btrfs_file_extent_inline_len(src, i, extent);
3846 extent_end = ALIGN(key.offset + len, log->sectorsize);
3848 len = btrfs_file_extent_num_bytes(src, extent);
3849 extent_end = key.offset + len;
3853 if (*last_extent == key.offset) {
3854 *last_extent = extent_end;
3857 offset = *last_extent;
3858 len = key.offset - *last_extent;
3859 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3860 offset, 0, 0, len, 0, len, 0,
3864 *last_extent = extent_end;
3867 * Need to let the callers know we dropped the path so they should
3870 if (!ret && need_find_last_extent)
3875 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3877 struct extent_map *em1, *em2;
3879 em1 = list_entry(a, struct extent_map, list);
3880 em2 = list_entry(b, struct extent_map, list);
3882 if (em1->start < em2->start)
3884 else if (em1->start > em2->start)
3889 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3890 struct inode *inode,
3891 struct btrfs_root *root,
3892 const struct extent_map *em,
3893 const struct list_head *logged_list,
3894 bool *ordered_io_error)
3896 struct btrfs_ordered_extent *ordered;
3897 struct btrfs_root *log = root->log_root;
3898 u64 mod_start = em->mod_start;
3899 u64 mod_len = em->mod_len;
3900 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3903 LIST_HEAD(ordered_sums);
3906 *ordered_io_error = false;
3908 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3909 em->block_start == EXTENT_MAP_HOLE)
3913 * Wait far any ordered extent that covers our extent map. If it
3914 * finishes without an error, first check and see if our csums are on
3915 * our outstanding ordered extents.
3917 list_for_each_entry(ordered, logged_list, log_list) {
3918 struct btrfs_ordered_sum *sum;
3923 if (ordered->file_offset + ordered->len <= mod_start ||
3924 mod_start + mod_len <= ordered->file_offset)
3927 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3928 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3929 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3930 const u64 start = ordered->file_offset;
3931 const u64 end = ordered->file_offset + ordered->len - 1;
3933 WARN_ON(ordered->inode != inode);
3934 filemap_fdatawrite_range(inode->i_mapping, start, end);
3937 wait_event(ordered->wait,
3938 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3939 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3941 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
3943 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3944 * i_mapping flags, so that the next fsync won't get
3945 * an outdated io error too.
3947 btrfs_inode_check_errors(inode);
3948 *ordered_io_error = true;
3952 * We are going to copy all the csums on this ordered extent, so
3953 * go ahead and adjust mod_start and mod_len in case this
3954 * ordered extent has already been logged.
3956 if (ordered->file_offset > mod_start) {
3957 if (ordered->file_offset + ordered->len >=
3958 mod_start + mod_len)
3959 mod_len = ordered->file_offset - mod_start;
3961 * If we have this case
3963 * |--------- logged extent ---------|
3964 * |----- ordered extent ----|
3966 * Just don't mess with mod_start and mod_len, we'll
3967 * just end up logging more csums than we need and it
3971 if (ordered->file_offset + ordered->len <
3972 mod_start + mod_len) {
3973 mod_len = (mod_start + mod_len) -
3974 (ordered->file_offset + ordered->len);
3975 mod_start = ordered->file_offset +
3986 * To keep us from looping for the above case of an ordered
3987 * extent that falls inside of the logged extent.
3989 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3993 list_for_each_entry(sum, &ordered->list, list) {
3994 ret = btrfs_csum_file_blocks(trans, log, sum);
4000 if (*ordered_io_error || !mod_len || ret || skip_csum)
4003 if (em->compress_type) {
4005 csum_len = max(em->block_len, em->orig_block_len);
4007 csum_offset = mod_start - em->start;
4011 /* block start is already adjusted for the file extent offset. */
4012 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
4013 em->block_start + csum_offset,
4014 em->block_start + csum_offset +
4015 csum_len - 1, &ordered_sums, 0);
4019 while (!list_empty(&ordered_sums)) {
4020 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4021 struct btrfs_ordered_sum,
4024 ret = btrfs_csum_file_blocks(trans, log, sums);
4025 list_del(&sums->list);
4032 static int log_one_extent(struct btrfs_trans_handle *trans,
4033 struct inode *inode, struct btrfs_root *root,
4034 const struct extent_map *em,
4035 struct btrfs_path *path,
4036 const struct list_head *logged_list,
4037 struct btrfs_log_ctx *ctx)
4039 struct btrfs_root *log = root->log_root;
4040 struct btrfs_file_extent_item *fi;
4041 struct extent_buffer *leaf;
4042 struct btrfs_map_token token;
4043 struct btrfs_key key;
4044 u64 extent_offset = em->start - em->orig_start;
4047 int extent_inserted = 0;
4048 bool ordered_io_err = false;
4050 ret = wait_ordered_extents(trans, inode, root, em, logged_list,
4055 if (ordered_io_err) {
4060 btrfs_init_map_token(&token);
4062 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
4063 em->start + em->len, NULL, 0, 1,
4064 sizeof(*fi), &extent_inserted);
4068 if (!extent_inserted) {
4069 key.objectid = btrfs_ino(inode);
4070 key.type = BTRFS_EXTENT_DATA_KEY;
4071 key.offset = em->start;
4073 ret = btrfs_insert_empty_item(trans, log, path, &key,
4078 leaf = path->nodes[0];
4079 fi = btrfs_item_ptr(leaf, path->slots[0],
4080 struct btrfs_file_extent_item);
4082 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4084 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4085 btrfs_set_token_file_extent_type(leaf, fi,
4086 BTRFS_FILE_EXTENT_PREALLOC,
4089 btrfs_set_token_file_extent_type(leaf, fi,
4090 BTRFS_FILE_EXTENT_REG,
4093 block_len = max(em->block_len, em->orig_block_len);
4094 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4095 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4098 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4100 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4101 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4103 extent_offset, &token);
4104 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4107 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4108 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4112 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4113 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4114 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4115 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4117 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4118 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4119 btrfs_mark_buffer_dirty(leaf);
4121 btrfs_release_path(path);
4126 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4127 struct btrfs_root *root,
4128 struct inode *inode,
4129 struct btrfs_path *path,
4130 struct list_head *logged_list,
4131 struct btrfs_log_ctx *ctx,
4135 struct extent_map *em, *n;
4136 struct list_head extents;
4137 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
4142 INIT_LIST_HEAD(&extents);
4144 down_write(&BTRFS_I(inode)->dio_sem);
4145 write_lock(&tree->lock);
4146 test_gen = root->fs_info->last_trans_committed;
4148 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4149 list_del_init(&em->list);
4152 * Just an arbitrary number, this can be really CPU intensive
4153 * once we start getting a lot of extents, and really once we
4154 * have a bunch of extents we just want to commit since it will
4157 if (++num > 32768) {
4158 list_del_init(&tree->modified_extents);
4163 if (em->generation <= test_gen)
4165 /* Need a ref to keep it from getting evicted from cache */
4166 atomic_inc(&em->refs);
4167 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4168 list_add_tail(&em->list, &extents);
4172 list_sort(NULL, &extents, extent_cmp);
4173 btrfs_get_logged_extents(inode, logged_list, start, end);
4175 * Some ordered extents started by fsync might have completed
4176 * before we could collect them into the list logged_list, which
4177 * means they're gone, not in our logged_list nor in the inode's
4178 * ordered tree. We want the application/user space to know an
4179 * error happened while attempting to persist file data so that
4180 * it can take proper action. If such error happened, we leave
4181 * without writing to the log tree and the fsync must report the
4182 * file data write error and not commit the current transaction.
4184 ret = btrfs_inode_check_errors(inode);
4188 while (!list_empty(&extents)) {
4189 em = list_entry(extents.next, struct extent_map, list);
4191 list_del_init(&em->list);
4194 * If we had an error we just need to delete everybody from our
4198 clear_em_logging(tree, em);
4199 free_extent_map(em);
4203 write_unlock(&tree->lock);
4205 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4207 write_lock(&tree->lock);
4208 clear_em_logging(tree, em);
4209 free_extent_map(em);
4211 WARN_ON(!list_empty(&extents));
4212 write_unlock(&tree->lock);
4213 up_write(&BTRFS_I(inode)->dio_sem);
4215 btrfs_release_path(path);
4219 static int logged_inode_size(struct btrfs_root *log, struct inode *inode,
4220 struct btrfs_path *path, u64 *size_ret)
4222 struct btrfs_key key;
4225 key.objectid = btrfs_ino(inode);
4226 key.type = BTRFS_INODE_ITEM_KEY;
4229 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4232 } else if (ret > 0) {
4235 struct btrfs_inode_item *item;
4237 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4238 struct btrfs_inode_item);
4239 *size_ret = btrfs_inode_size(path->nodes[0], item);
4242 btrfs_release_path(path);
4247 * At the moment we always log all xattrs. This is to figure out at log replay
4248 * time which xattrs must have their deletion replayed. If a xattr is missing
4249 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4250 * because if a xattr is deleted, the inode is fsynced and a power failure
4251 * happens, causing the log to be replayed the next time the fs is mounted,
4252 * we want the xattr to not exist anymore (same behaviour as other filesystems
4253 * with a journal, ext3/4, xfs, f2fs, etc).
4255 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4256 struct btrfs_root *root,
4257 struct inode *inode,
4258 struct btrfs_path *path,
4259 struct btrfs_path *dst_path)
4262 struct btrfs_key key;
4263 const u64 ino = btrfs_ino(inode);
4268 key.type = BTRFS_XATTR_ITEM_KEY;
4271 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4276 int slot = path->slots[0];
4277 struct extent_buffer *leaf = path->nodes[0];
4278 int nritems = btrfs_header_nritems(leaf);
4280 if (slot >= nritems) {
4282 u64 last_extent = 0;
4284 ret = copy_items(trans, inode, dst_path, path,
4285 &last_extent, start_slot,
4287 /* can't be 1, extent items aren't processed */
4293 ret = btrfs_next_leaf(root, path);
4301 btrfs_item_key_to_cpu(leaf, &key, slot);
4302 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4312 u64 last_extent = 0;
4314 ret = copy_items(trans, inode, dst_path, path,
4315 &last_extent, start_slot,
4317 /* can't be 1, extent items aren't processed */
4327 * If the no holes feature is enabled we need to make sure any hole between the
4328 * last extent and the i_size of our inode is explicitly marked in the log. This
4329 * is to make sure that doing something like:
4331 * 1) create file with 128Kb of data
4332 * 2) truncate file to 64Kb
4333 * 3) truncate file to 256Kb
4335 * 5) <crash/power failure>
4336 * 6) mount fs and trigger log replay
4338 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4339 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4340 * file correspond to a hole. The presence of explicit holes in a log tree is
4341 * what guarantees that log replay will remove/adjust file extent items in the
4344 * Here we do not need to care about holes between extents, that is already done
4345 * by copy_items(). We also only need to do this in the full sync path, where we
4346 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4347 * lookup the list of modified extent maps and if any represents a hole, we
4348 * insert a corresponding extent representing a hole in the log tree.
4350 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4351 struct btrfs_root *root,
4352 struct inode *inode,
4353 struct btrfs_path *path)
4356 struct btrfs_key key;
4359 struct extent_buffer *leaf;
4360 struct btrfs_root *log = root->log_root;
4361 const u64 ino = btrfs_ino(inode);
4362 const u64 i_size = i_size_read(inode);
4364 if (!btrfs_fs_incompat(root->fs_info, NO_HOLES))
4368 key.type = BTRFS_EXTENT_DATA_KEY;
4369 key.offset = (u64)-1;
4371 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4376 ASSERT(path->slots[0] > 0);
4378 leaf = path->nodes[0];
4379 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4381 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4382 /* inode does not have any extents */
4386 struct btrfs_file_extent_item *extent;
4390 * If there's an extent beyond i_size, an explicit hole was
4391 * already inserted by copy_items().
4393 if (key.offset >= i_size)
4396 extent = btrfs_item_ptr(leaf, path->slots[0],
4397 struct btrfs_file_extent_item);
4399 if (btrfs_file_extent_type(leaf, extent) ==
4400 BTRFS_FILE_EXTENT_INLINE) {
4401 len = btrfs_file_extent_inline_len(leaf,
4404 ASSERT(len == i_size);
4408 len = btrfs_file_extent_num_bytes(leaf, extent);
4409 /* Last extent goes beyond i_size, no need to log a hole. */
4410 if (key.offset + len > i_size)
4412 hole_start = key.offset + len;
4413 hole_size = i_size - hole_start;
4415 btrfs_release_path(path);
4417 /* Last extent ends at i_size. */
4421 hole_size = ALIGN(hole_size, root->sectorsize);
4422 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4423 hole_size, 0, hole_size, 0, 0, 0);
4428 * When we are logging a new inode X, check if it doesn't have a reference that
4429 * matches the reference from some other inode Y created in a past transaction
4430 * and that was renamed in the current transaction. If we don't do this, then at
4431 * log replay time we can lose inode Y (and all its files if it's a directory):
4434 * echo "hello world" > /mnt/x/foobar
4437 * mkdir /mnt/x # or touch /mnt/x
4438 * xfs_io -c fsync /mnt/x
4440 * mount fs, trigger log replay
4442 * After the log replay procedure, we would lose the first directory and all its
4443 * files (file foobar).
4444 * For the case where inode Y is not a directory we simply end up losing it:
4446 * echo "123" > /mnt/foo
4448 * mv /mnt/foo /mnt/bar
4449 * echo "abc" > /mnt/foo
4450 * xfs_io -c fsync /mnt/foo
4453 * We also need this for cases where a snapshot entry is replaced by some other
4454 * entry (file or directory) otherwise we end up with an unreplayable log due to
4455 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4456 * if it were a regular entry:
4459 * btrfs subvolume snapshot /mnt /mnt/x/snap
4460 * btrfs subvolume delete /mnt/x/snap
4463 * fsync /mnt/x or fsync some new file inside it
4466 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4467 * the same transaction.
4469 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4471 const struct btrfs_key *key,
4472 struct inode *inode)
4475 struct btrfs_path *search_path;
4478 u32 item_size = btrfs_item_size_nr(eb, slot);
4480 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4482 search_path = btrfs_alloc_path();
4485 search_path->search_commit_root = 1;
4486 search_path->skip_locking = 1;
4488 while (cur_offset < item_size) {
4492 unsigned long name_ptr;
4493 struct btrfs_dir_item *di;
4495 if (key->type == BTRFS_INODE_REF_KEY) {
4496 struct btrfs_inode_ref *iref;
4498 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4499 parent = key->offset;
4500 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4501 name_ptr = (unsigned long)(iref + 1);
4502 this_len = sizeof(*iref) + this_name_len;
4504 struct btrfs_inode_extref *extref;
4506 extref = (struct btrfs_inode_extref *)(ptr +
4508 parent = btrfs_inode_extref_parent(eb, extref);
4509 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4510 name_ptr = (unsigned long)&extref->name;
4511 this_len = sizeof(*extref) + this_name_len;
4514 if (this_name_len > name_len) {
4517 new_name = krealloc(name, this_name_len, GFP_NOFS);
4522 name_len = this_name_len;
4526 read_extent_buffer(eb, name, name_ptr, this_name_len);
4527 di = btrfs_lookup_dir_item(NULL, BTRFS_I(inode)->root,
4528 search_path, parent,
4529 name, this_name_len, 0);
4530 if (di && !IS_ERR(di)) {
4533 } else if (IS_ERR(di)) {
4537 btrfs_release_path(search_path);
4539 cur_offset += this_len;
4543 btrfs_free_path(search_path);
4548 /* log a single inode in the tree log.
4549 * At least one parent directory for this inode must exist in the tree
4550 * or be logged already.
4552 * Any items from this inode changed by the current transaction are copied
4553 * to the log tree. An extra reference is taken on any extents in this
4554 * file, allowing us to avoid a whole pile of corner cases around logging
4555 * blocks that have been removed from the tree.
4557 * See LOG_INODE_ALL and related defines for a description of what inode_only
4560 * This handles both files and directories.
4562 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4563 struct btrfs_root *root, struct inode *inode,
4567 struct btrfs_log_ctx *ctx)
4569 struct btrfs_path *path;
4570 struct btrfs_path *dst_path;
4571 struct btrfs_key min_key;
4572 struct btrfs_key max_key;
4573 struct btrfs_root *log = root->log_root;
4574 struct extent_buffer *src = NULL;
4575 LIST_HEAD(logged_list);
4576 u64 last_extent = 0;
4580 int ins_start_slot = 0;
4582 bool fast_search = false;
4583 u64 ino = btrfs_ino(inode);
4584 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4585 u64 logged_isize = 0;
4586 bool need_log_inode_item = true;
4588 path = btrfs_alloc_path();
4591 dst_path = btrfs_alloc_path();
4593 btrfs_free_path(path);
4597 min_key.objectid = ino;
4598 min_key.type = BTRFS_INODE_ITEM_KEY;
4601 max_key.objectid = ino;
4604 /* today the code can only do partial logging of directories */
4605 if (S_ISDIR(inode->i_mode) ||
4606 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4607 &BTRFS_I(inode)->runtime_flags) &&
4608 inode_only == LOG_INODE_EXISTS))
4609 max_key.type = BTRFS_XATTR_ITEM_KEY;
4611 max_key.type = (u8)-1;
4612 max_key.offset = (u64)-1;
4615 * Only run delayed items if we are a dir or a new file.
4616 * Otherwise commit the delayed inode only, which is needed in
4617 * order for the log replay code to mark inodes for link count
4618 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4620 if (S_ISDIR(inode->i_mode) ||
4621 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed)
4622 ret = btrfs_commit_inode_delayed_items(trans, inode);
4624 ret = btrfs_commit_inode_delayed_inode(inode);
4627 btrfs_free_path(path);
4628 btrfs_free_path(dst_path);
4632 mutex_lock(&BTRFS_I(inode)->log_mutex);
4635 * a brute force approach to making sure we get the most uptodate
4636 * copies of everything.
4638 if (S_ISDIR(inode->i_mode)) {
4639 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4641 if (inode_only == LOG_INODE_EXISTS)
4642 max_key_type = BTRFS_XATTR_ITEM_KEY;
4643 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
4645 if (inode_only == LOG_INODE_EXISTS) {
4647 * Make sure the new inode item we write to the log has
4648 * the same isize as the current one (if it exists).
4649 * This is necessary to prevent data loss after log
4650 * replay, and also to prevent doing a wrong expanding
4651 * truncate - for e.g. create file, write 4K into offset
4652 * 0, fsync, write 4K into offset 4096, add hard link,
4653 * fsync some other file (to sync log), power fail - if
4654 * we use the inode's current i_size, after log replay
4655 * we get a 8Kb file, with the last 4Kb extent as a hole
4656 * (zeroes), as if an expanding truncate happened,
4657 * instead of getting a file of 4Kb only.
4659 err = logged_inode_size(log, inode, path,
4664 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4665 &BTRFS_I(inode)->runtime_flags)) {
4666 if (inode_only == LOG_INODE_EXISTS) {
4667 max_key.type = BTRFS_XATTR_ITEM_KEY;
4668 ret = drop_objectid_items(trans, log, path, ino,
4671 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4672 &BTRFS_I(inode)->runtime_flags);
4673 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4674 &BTRFS_I(inode)->runtime_flags);
4676 ret = btrfs_truncate_inode_items(trans,
4682 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4683 &BTRFS_I(inode)->runtime_flags) ||
4684 inode_only == LOG_INODE_EXISTS) {
4685 if (inode_only == LOG_INODE_ALL)
4687 max_key.type = BTRFS_XATTR_ITEM_KEY;
4688 ret = drop_objectid_items(trans, log, path, ino,
4691 if (inode_only == LOG_INODE_ALL)
4704 ret = btrfs_search_forward(root, &min_key,
4705 path, trans->transid);
4713 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4714 if (min_key.objectid != ino)
4716 if (min_key.type > max_key.type)
4719 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4720 need_log_inode_item = false;
4722 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4723 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4724 BTRFS_I(inode)->generation == trans->transid) {
4725 ret = btrfs_check_ref_name_override(path->nodes[0],
4731 } else if (ret > 0) {
4733 btrfs_set_log_full_commit(root->fs_info, trans);
4738 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4739 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4742 ret = copy_items(trans, inode, dst_path, path,
4743 &last_extent, ins_start_slot,
4744 ins_nr, inode_only, logged_isize);
4751 btrfs_release_path(path);
4757 src = path->nodes[0];
4758 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4761 } else if (!ins_nr) {
4762 ins_start_slot = path->slots[0];
4767 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4768 ins_start_slot, ins_nr, inode_only,
4776 btrfs_release_path(path);
4780 ins_start_slot = path->slots[0];
4783 nritems = btrfs_header_nritems(path->nodes[0]);
4785 if (path->slots[0] < nritems) {
4786 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4791 ret = copy_items(trans, inode, dst_path, path,
4792 &last_extent, ins_start_slot,
4793 ins_nr, inode_only, logged_isize);
4801 btrfs_release_path(path);
4803 if (min_key.offset < (u64)-1) {
4805 } else if (min_key.type < max_key.type) {
4813 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4814 ins_start_slot, ins_nr, inode_only,
4824 btrfs_release_path(path);
4825 btrfs_release_path(dst_path);
4826 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
4829 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4830 btrfs_release_path(path);
4831 btrfs_release_path(dst_path);
4832 err = btrfs_log_trailing_hole(trans, root, inode, path);
4837 btrfs_release_path(path);
4838 btrfs_release_path(dst_path);
4839 if (need_log_inode_item) {
4840 err = log_inode_item(trans, log, dst_path, inode);
4845 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
4846 &logged_list, ctx, start, end);
4851 } else if (inode_only == LOG_INODE_ALL) {
4852 struct extent_map *em, *n;
4854 write_lock(&em_tree->lock);
4856 * We can't just remove every em if we're called for a ranged
4857 * fsync - that is, one that doesn't cover the whole possible
4858 * file range (0 to LLONG_MAX). This is because we can have
4859 * em's that fall outside the range we're logging and therefore
4860 * their ordered operations haven't completed yet
4861 * (btrfs_finish_ordered_io() not invoked yet). This means we
4862 * didn't get their respective file extent item in the fs/subvol
4863 * tree yet, and need to let the next fast fsync (one which
4864 * consults the list of modified extent maps) find the em so
4865 * that it logs a matching file extent item and waits for the
4866 * respective ordered operation to complete (if it's still
4869 * Removing every em outside the range we're logging would make
4870 * the next fast fsync not log their matching file extent items,
4871 * therefore making us lose data after a log replay.
4873 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4875 const u64 mod_end = em->mod_start + em->mod_len - 1;
4877 if (em->mod_start >= start && mod_end <= end)
4878 list_del_init(&em->list);
4880 write_unlock(&em_tree->lock);
4883 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
4884 ret = log_directory_changes(trans, root, inode, path, dst_path,
4892 spin_lock(&BTRFS_I(inode)->lock);
4893 BTRFS_I(inode)->logged_trans = trans->transid;
4894 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
4895 spin_unlock(&BTRFS_I(inode)->lock);
4898 btrfs_put_logged_extents(&logged_list);
4900 btrfs_submit_logged_extents(&logged_list, log);
4901 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4903 btrfs_free_path(path);
4904 btrfs_free_path(dst_path);
4909 * Check if we must fallback to a transaction commit when logging an inode.
4910 * This must be called after logging the inode and is used only in the context
4911 * when fsyncing an inode requires the need to log some other inode - in which
4912 * case we can't lock the i_mutex of each other inode we need to log as that
4913 * can lead to deadlocks with concurrent fsync against other inodes (as we can
4914 * log inodes up or down in the hierarchy) or rename operations for example. So
4915 * we take the log_mutex of the inode after we have logged it and then check for
4916 * its last_unlink_trans value - this is safe because any task setting
4917 * last_unlink_trans must take the log_mutex and it must do this before it does
4918 * the actual unlink operation, so if we do this check before a concurrent task
4919 * sets last_unlink_trans it means we've logged a consistent version/state of
4920 * all the inode items, otherwise we are not sure and must do a transaction
4921 * commit (the concurrent task might have only updated last_unlink_trans before
4922 * we logged the inode or it might have also done the unlink).
4924 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
4925 struct inode *inode)
4927 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4930 mutex_lock(&BTRFS_I(inode)->log_mutex);
4931 if (BTRFS_I(inode)->last_unlink_trans > fs_info->last_trans_committed) {
4933 * Make sure any commits to the log are forced to be full
4936 btrfs_set_log_full_commit(fs_info, trans);
4939 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4945 * follow the dentry parent pointers up the chain and see if any
4946 * of the directories in it require a full commit before they can
4947 * be logged. Returns zero if nothing special needs to be done or 1 if
4948 * a full commit is required.
4950 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
4951 struct inode *inode,
4952 struct dentry *parent,
4953 struct super_block *sb,
4957 struct dentry *old_parent = NULL;
4958 struct inode *orig_inode = inode;
4961 * for regular files, if its inode is already on disk, we don't
4962 * have to worry about the parents at all. This is because
4963 * we can use the last_unlink_trans field to record renames
4964 * and other fun in this file.
4966 if (S_ISREG(inode->i_mode) &&
4967 BTRFS_I(inode)->generation <= last_committed &&
4968 BTRFS_I(inode)->last_unlink_trans <= last_committed)
4971 if (!S_ISDIR(inode->i_mode)) {
4972 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
4974 inode = d_inode(parent);
4979 * If we are logging a directory then we start with our inode,
4980 * not our parent's inode, so we need to skip setting the
4981 * logged_trans so that further down in the log code we don't
4982 * think this inode has already been logged.
4984 if (inode != orig_inode)
4985 BTRFS_I(inode)->logged_trans = trans->transid;
4988 if (btrfs_must_commit_transaction(trans, inode)) {
4993 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
4996 if (IS_ROOT(parent))
4999 parent = dget_parent(parent);
5001 old_parent = parent;
5002 inode = d_inode(parent);
5010 struct btrfs_dir_list {
5012 struct list_head list;
5016 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5017 * details about the why it is needed.
5018 * This is a recursive operation - if an existing dentry corresponds to a
5019 * directory, that directory's new entries are logged too (same behaviour as
5020 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5021 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5022 * complains about the following circular lock dependency / possible deadlock:
5026 * lock(&type->i_mutex_dir_key#3/2);
5027 * lock(sb_internal#2);
5028 * lock(&type->i_mutex_dir_key#3/2);
5029 * lock(&sb->s_type->i_mutex_key#14);
5031 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5032 * sb_start_intwrite() in btrfs_start_transaction().
5033 * Not locking i_mutex of the inodes is still safe because:
5035 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5036 * that while logging the inode new references (names) are added or removed
5037 * from the inode, leaving the logged inode item with a link count that does
5038 * not match the number of logged inode reference items. This is fine because
5039 * at log replay time we compute the real number of links and correct the
5040 * link count in the inode item (see replay_one_buffer() and
5041 * link_to_fixup_dir());
5043 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5044 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5045 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5046 * has a size that doesn't match the sum of the lengths of all the logged
5047 * names. This does not result in a problem because if a dir_item key is
5048 * logged but its matching dir_index key is not logged, at log replay time we
5049 * don't use it to replay the respective name (see replay_one_name()). On the
5050 * other hand if only the dir_index key ends up being logged, the respective
5051 * name is added to the fs/subvol tree with both the dir_item and dir_index
5052 * keys created (see replay_one_name()).
5053 * The directory's inode item with a wrong i_size is not a problem as well,
5054 * since we don't use it at log replay time to set the i_size in the inode
5055 * item of the fs/subvol tree (see overwrite_item()).
5057 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5058 struct btrfs_root *root,
5059 struct inode *start_inode,
5060 struct btrfs_log_ctx *ctx)
5062 struct btrfs_root *log = root->log_root;
5063 struct btrfs_path *path;
5064 LIST_HEAD(dir_list);
5065 struct btrfs_dir_list *dir_elem;
5068 path = btrfs_alloc_path();
5072 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5074 btrfs_free_path(path);
5077 dir_elem->ino = btrfs_ino(start_inode);
5078 list_add_tail(&dir_elem->list, &dir_list);
5080 while (!list_empty(&dir_list)) {
5081 struct extent_buffer *leaf;
5082 struct btrfs_key min_key;
5086 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5089 goto next_dir_inode;
5091 min_key.objectid = dir_elem->ino;
5092 min_key.type = BTRFS_DIR_ITEM_KEY;
5095 btrfs_release_path(path);
5096 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5098 goto next_dir_inode;
5099 } else if (ret > 0) {
5101 goto next_dir_inode;
5105 leaf = path->nodes[0];
5106 nritems = btrfs_header_nritems(leaf);
5107 for (i = path->slots[0]; i < nritems; i++) {
5108 struct btrfs_dir_item *di;
5109 struct btrfs_key di_key;
5110 struct inode *di_inode;
5111 struct btrfs_dir_list *new_dir_elem;
5112 int log_mode = LOG_INODE_EXISTS;
5115 btrfs_item_key_to_cpu(leaf, &min_key, i);
5116 if (min_key.objectid != dir_elem->ino ||
5117 min_key.type != BTRFS_DIR_ITEM_KEY)
5118 goto next_dir_inode;
5120 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5121 type = btrfs_dir_type(leaf, di);
5122 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5123 type != BTRFS_FT_DIR)
5125 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5126 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5129 di_inode = btrfs_iget(root->fs_info->sb, &di_key,
5131 if (IS_ERR(di_inode)) {
5132 ret = PTR_ERR(di_inode);
5133 goto next_dir_inode;
5136 if (btrfs_inode_in_log(di_inode, trans->transid)) {
5141 ctx->log_new_dentries = false;
5142 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5143 log_mode = LOG_INODE_ALL;
5144 btrfs_release_path(path);
5145 ret = btrfs_log_inode(trans, root, di_inode,
5146 log_mode, 0, LLONG_MAX, ctx);
5148 btrfs_must_commit_transaction(trans, di_inode))
5152 goto next_dir_inode;
5153 if (ctx->log_new_dentries) {
5154 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5156 if (!new_dir_elem) {
5158 goto next_dir_inode;
5160 new_dir_elem->ino = di_key.objectid;
5161 list_add_tail(&new_dir_elem->list, &dir_list);
5166 ret = btrfs_next_leaf(log, path);
5168 goto next_dir_inode;
5169 } else if (ret > 0) {
5171 goto next_dir_inode;
5175 if (min_key.offset < (u64)-1) {
5180 list_del(&dir_elem->list);
5184 btrfs_free_path(path);
5188 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5189 struct inode *inode,
5190 struct btrfs_log_ctx *ctx)
5193 struct btrfs_path *path;
5194 struct btrfs_key key;
5195 struct btrfs_root *root = BTRFS_I(inode)->root;
5196 const u64 ino = btrfs_ino(inode);
5198 path = btrfs_alloc_path();
5201 path->skip_locking = 1;
5202 path->search_commit_root = 1;
5205 key.type = BTRFS_INODE_REF_KEY;
5207 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5212 struct extent_buffer *leaf = path->nodes[0];
5213 int slot = path->slots[0];
5218 if (slot >= btrfs_header_nritems(leaf)) {
5219 ret = btrfs_next_leaf(root, path);
5227 btrfs_item_key_to_cpu(leaf, &key, slot);
5228 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5229 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5232 item_size = btrfs_item_size_nr(leaf, slot);
5233 ptr = btrfs_item_ptr_offset(leaf, slot);
5234 while (cur_offset < item_size) {
5235 struct btrfs_key inode_key;
5236 struct inode *dir_inode;
5238 inode_key.type = BTRFS_INODE_ITEM_KEY;
5239 inode_key.offset = 0;
5241 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5242 struct btrfs_inode_extref *extref;
5244 extref = (struct btrfs_inode_extref *)
5246 inode_key.objectid = btrfs_inode_extref_parent(
5248 cur_offset += sizeof(*extref);
5249 cur_offset += btrfs_inode_extref_name_len(leaf,
5252 inode_key.objectid = key.offset;
5253 cur_offset = item_size;
5256 dir_inode = btrfs_iget(root->fs_info->sb, &inode_key,
5258 /* If parent inode was deleted, skip it. */
5259 if (IS_ERR(dir_inode))
5263 ctx->log_new_dentries = false;
5264 ret = btrfs_log_inode(trans, root, dir_inode,
5265 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5267 btrfs_must_commit_transaction(trans, dir_inode))
5269 if (!ret && ctx && ctx->log_new_dentries)
5270 ret = log_new_dir_dentries(trans, root,
5280 btrfs_free_path(path);
5285 * helper function around btrfs_log_inode to make sure newly created
5286 * parent directories also end up in the log. A minimal inode and backref
5287 * only logging is done of any parent directories that are older than
5288 * the last committed transaction
5290 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5291 struct btrfs_root *root, struct inode *inode,
5292 struct dentry *parent,
5296 struct btrfs_log_ctx *ctx)
5298 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
5299 struct super_block *sb;
5300 struct dentry *old_parent = NULL;
5302 u64 last_committed = root->fs_info->last_trans_committed;
5303 bool log_dentries = false;
5304 struct inode *orig_inode = inode;
5308 if (btrfs_test_opt(root->fs_info, NOTREELOG)) {
5314 * The prev transaction commit doesn't complete, we need do
5315 * full commit by ourselves.
5317 if (root->fs_info->last_trans_log_full_commit >
5318 root->fs_info->last_trans_committed) {
5323 if (root != BTRFS_I(inode)->root ||
5324 btrfs_root_refs(&root->root_item) == 0) {
5329 ret = check_parent_dirs_for_sync(trans, inode, parent,
5330 sb, last_committed);
5334 if (btrfs_inode_in_log(inode, trans->transid)) {
5335 ret = BTRFS_NO_LOG_SYNC;
5339 ret = start_log_trans(trans, root, ctx);
5343 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5348 * for regular files, if its inode is already on disk, we don't
5349 * have to worry about the parents at all. This is because
5350 * we can use the last_unlink_trans field to record renames
5351 * and other fun in this file.
5353 if (S_ISREG(inode->i_mode) &&
5354 BTRFS_I(inode)->generation <= last_committed &&
5355 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
5360 if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries)
5361 log_dentries = true;
5364 * On unlink we must make sure all our current and old parent directory
5365 * inodes are fully logged. This is to prevent leaving dangling
5366 * directory index entries in directories that were our parents but are
5367 * not anymore. Not doing this results in old parent directory being
5368 * impossible to delete after log replay (rmdir will always fail with
5369 * error -ENOTEMPTY).
5375 * ln testdir/foo testdir/bar
5377 * unlink testdir/bar
5378 * xfs_io -c fsync testdir/foo
5380 * mount fs, triggers log replay
5382 * If we don't log the parent directory (testdir), after log replay the
5383 * directory still has an entry pointing to the file inode using the bar
5384 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5385 * the file inode has a link count of 1.
5391 * ln foo testdir/foo2
5392 * ln foo testdir/foo3
5394 * unlink testdir/foo3
5395 * xfs_io -c fsync foo
5397 * mount fs, triggers log replay
5399 * Similar as the first example, after log replay the parent directory
5400 * testdir still has an entry pointing to the inode file with name foo3
5401 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5402 * and has a link count of 2.
5404 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
5405 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5411 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5414 inode = d_inode(parent);
5415 if (root != BTRFS_I(inode)->root)
5418 if (BTRFS_I(inode)->generation > last_committed) {
5419 ret = btrfs_log_inode(trans, root, inode,
5425 if (IS_ROOT(parent))
5428 parent = dget_parent(parent);
5430 old_parent = parent;
5433 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5439 btrfs_set_log_full_commit(root->fs_info, trans);
5444 btrfs_remove_log_ctx(root, ctx);
5445 btrfs_end_log_trans(root);
5451 * it is not safe to log dentry if the chunk root has added new
5452 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5453 * If this returns 1, you must commit the transaction to safely get your
5456 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
5457 struct btrfs_root *root, struct dentry *dentry,
5460 struct btrfs_log_ctx *ctx)
5462 struct dentry *parent = dget_parent(dentry);
5465 ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent,
5466 start, end, 0, ctx);
5473 * should be called during mount to recover any replay any log trees
5476 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5479 struct btrfs_path *path;
5480 struct btrfs_trans_handle *trans;
5481 struct btrfs_key key;
5482 struct btrfs_key found_key;
5483 struct btrfs_key tmp_key;
5484 struct btrfs_root *log;
5485 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5486 struct walk_control wc = {
5487 .process_func = process_one_buffer,
5491 path = btrfs_alloc_path();
5495 fs_info->log_root_recovering = 1;
5497 trans = btrfs_start_transaction(fs_info->tree_root, 0);
5498 if (IS_ERR(trans)) {
5499 ret = PTR_ERR(trans);
5506 ret = walk_log_tree(trans, log_root_tree, &wc);
5508 btrfs_handle_fs_error(fs_info, ret, "Failed to pin buffers while "
5509 "recovering log root tree.");
5514 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5515 key.offset = (u64)-1;
5516 key.type = BTRFS_ROOT_ITEM_KEY;
5519 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
5522 btrfs_handle_fs_error(fs_info, ret,
5523 "Couldn't find tree log root.");
5527 if (path->slots[0] == 0)
5531 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5533 btrfs_release_path(path);
5534 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5537 log = btrfs_read_fs_root(log_root_tree, &found_key);
5540 btrfs_handle_fs_error(fs_info, ret,
5541 "Couldn't read tree log root.");
5545 tmp_key.objectid = found_key.offset;
5546 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5547 tmp_key.offset = (u64)-1;
5549 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
5550 if (IS_ERR(wc.replay_dest)) {
5551 ret = PTR_ERR(wc.replay_dest);
5552 free_extent_buffer(log->node);
5553 free_extent_buffer(log->commit_root);
5555 btrfs_handle_fs_error(fs_info, ret, "Couldn't read target root "
5556 "for tree log recovery.");
5560 wc.replay_dest->log_root = log;
5561 btrfs_record_root_in_trans(trans, wc.replay_dest);
5562 ret = walk_log_tree(trans, log, &wc);
5564 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5565 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5569 key.offset = found_key.offset - 1;
5570 wc.replay_dest->log_root = NULL;
5571 free_extent_buffer(log->node);
5572 free_extent_buffer(log->commit_root);
5578 if (found_key.offset == 0)
5581 btrfs_release_path(path);
5583 /* step one is to pin it all, step two is to replay just inodes */
5586 wc.process_func = replay_one_buffer;
5587 wc.stage = LOG_WALK_REPLAY_INODES;
5590 /* step three is to replay everything */
5591 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5596 btrfs_free_path(path);
5598 /* step 4: commit the transaction, which also unpins the blocks */
5599 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
5603 free_extent_buffer(log_root_tree->node);
5604 log_root_tree->log_root = NULL;
5605 fs_info->log_root_recovering = 0;
5606 kfree(log_root_tree);
5611 btrfs_end_transaction(wc.trans, fs_info->tree_root);
5612 btrfs_free_path(path);
5617 * there are some corner cases where we want to force a full
5618 * commit instead of allowing a directory to be logged.
5620 * They revolve around files there were unlinked from the directory, and
5621 * this function updates the parent directory so that a full commit is
5622 * properly done if it is fsync'd later after the unlinks are done.
5624 * Must be called before the unlink operations (updates to the subvolume tree,
5625 * inodes, etc) are done.
5627 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5628 struct inode *dir, struct inode *inode,
5632 * when we're logging a file, if it hasn't been renamed
5633 * or unlinked, and its inode is fully committed on disk,
5634 * we don't have to worry about walking up the directory chain
5635 * to log its parents.
5637 * So, we use the last_unlink_trans field to put this transid
5638 * into the file. When the file is logged we check it and
5639 * don't log the parents if the file is fully on disk.
5641 mutex_lock(&BTRFS_I(inode)->log_mutex);
5642 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5643 mutex_unlock(&BTRFS_I(inode)->log_mutex);
5646 * if this directory was already logged any new
5647 * names for this file/dir will get recorded
5650 if (BTRFS_I(dir)->logged_trans == trans->transid)
5654 * if the inode we're about to unlink was logged,
5655 * the log will be properly updated for any new names
5657 if (BTRFS_I(inode)->logged_trans == trans->transid)
5661 * when renaming files across directories, if the directory
5662 * there we're unlinking from gets fsync'd later on, there's
5663 * no way to find the destination directory later and fsync it
5664 * properly. So, we have to be conservative and force commits
5665 * so the new name gets discovered.
5670 /* we can safely do the unlink without any special recording */
5674 mutex_lock(&BTRFS_I(dir)->log_mutex);
5675 BTRFS_I(dir)->last_unlink_trans = trans->transid;
5676 mutex_unlock(&BTRFS_I(dir)->log_mutex);
5680 * Make sure that if someone attempts to fsync the parent directory of a deleted
5681 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5682 * that after replaying the log tree of the parent directory's root we will not
5683 * see the snapshot anymore and at log replay time we will not see any log tree
5684 * corresponding to the deleted snapshot's root, which could lead to replaying
5685 * it after replaying the log tree of the parent directory (which would replay
5686 * the snapshot delete operation).
5688 * Must be called before the actual snapshot destroy operation (updates to the
5689 * parent root and tree of tree roots trees, etc) are done.
5691 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
5694 mutex_lock(&BTRFS_I(dir)->log_mutex);
5695 BTRFS_I(dir)->last_unlink_trans = trans->transid;
5696 mutex_unlock(&BTRFS_I(dir)->log_mutex);
5700 * Call this after adding a new name for a file and it will properly
5701 * update the log to reflect the new name.
5703 * It will return zero if all goes well, and it will return 1 if a
5704 * full transaction commit is required.
5706 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5707 struct inode *inode, struct inode *old_dir,
5708 struct dentry *parent)
5710 struct btrfs_root * root = BTRFS_I(inode)->root;
5713 * this will force the logging code to walk the dentry chain
5716 if (S_ISREG(inode->i_mode))
5717 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5720 * if this inode hasn't been logged and directory we're renaming it
5721 * from hasn't been logged, we don't need to log it
5723 if (BTRFS_I(inode)->logged_trans <=
5724 root->fs_info->last_trans_committed &&
5725 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
5726 root->fs_info->last_trans_committed))
5729 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
5730 LLONG_MAX, 1, NULL);