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 #include "inode-map.h"
33 /* magic values for the inode_only field in btrfs_log_inode:
35 * LOG_INODE_ALL means to log everything
36 * LOG_INODE_EXISTS means to log just enough to recreate the inode
39 #define LOG_INODE_ALL 0
40 #define LOG_INODE_EXISTS 1
41 #define LOG_OTHER_INODE 2
44 * directory trouble cases
46 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
47 * log, we must force a full commit before doing an fsync of the directory
48 * where the unlink was done.
49 * ---> record transid of last unlink/rename per directory
53 * rename foo/some_dir foo2/some_dir
55 * fsync foo/some_dir/some_file
57 * The fsync above will unlink the original some_dir without recording
58 * it in its new location (foo2). After a crash, some_dir will be gone
59 * unless the fsync of some_file forces a full commit
61 * 2) we must log any new names for any file or dir that is in the fsync
62 * log. ---> check inode while renaming/linking.
64 * 2a) we must log any new names for any file or dir during rename
65 * when the directory they are being removed from was logged.
66 * ---> check inode and old parent dir during rename
68 * 2a is actually the more important variant. With the extra logging
69 * a crash might unlink the old name without recreating the new one
71 * 3) after a crash, we must go through any directories with a link count
72 * of zero and redo the rm -rf
79 * The directory f1 was fully removed from the FS, but fsync was never
80 * called on f1, only its parent dir. After a crash the rm -rf must
81 * be replayed. This must be able to recurse down the entire
82 * directory tree. The inode link count fixup code takes care of the
87 * stages for the tree walking. The first
88 * stage (0) is to only pin down the blocks we find
89 * the second stage (1) is to make sure that all the inodes
90 * we find in the log are created in the subvolume.
92 * The last stage is to deal with directories and links and extents
93 * and all the other fun semantics
95 #define LOG_WALK_PIN_ONLY 0
96 #define LOG_WALK_REPLAY_INODES 1
97 #define LOG_WALK_REPLAY_DIR_INDEX 2
98 #define LOG_WALK_REPLAY_ALL 3
100 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root, struct inode *inode,
105 struct btrfs_log_ctx *ctx);
106 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
107 struct btrfs_root *root,
108 struct btrfs_path *path, u64 objectid);
109 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
110 struct btrfs_root *root,
111 struct btrfs_root *log,
112 struct btrfs_path *path,
113 u64 dirid, int del_all);
116 * tree logging is a special write ahead log used to make sure that
117 * fsyncs and O_SYNCs can happen without doing full tree commits.
119 * Full tree commits are expensive because they require commonly
120 * modified blocks to be recowed, creating many dirty pages in the
121 * extent tree an 4x-6x higher write load than ext3.
123 * Instead of doing a tree commit on every fsync, we use the
124 * key ranges and transaction ids to find items for a given file or directory
125 * that have changed in this transaction. Those items are copied into
126 * a special tree (one per subvolume root), that tree is written to disk
127 * and then the fsync is considered complete.
129 * After a crash, items are copied out of the log-tree back into the
130 * subvolume tree. Any file data extents found are recorded in the extent
131 * allocation tree, and the log-tree freed.
133 * The log tree is read three times, once to pin down all the extents it is
134 * using in ram and once, once to create all the inodes logged in the tree
135 * and once to do all the other items.
139 * start a sub transaction and setup the log tree
140 * this increments the log tree writer count to make the people
141 * syncing the tree wait for us to finish
143 static int start_log_trans(struct btrfs_trans_handle *trans,
144 struct btrfs_root *root,
145 struct btrfs_log_ctx *ctx)
149 mutex_lock(&root->log_mutex);
151 if (root->log_root) {
152 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
157 if (!root->log_start_pid) {
158 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
159 root->log_start_pid = current->pid;
160 } else if (root->log_start_pid != current->pid) {
161 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
164 mutex_lock(&root->fs_info->tree_log_mutex);
165 if (!root->fs_info->log_root_tree)
166 ret = btrfs_init_log_root_tree(trans, root->fs_info);
167 mutex_unlock(&root->fs_info->tree_log_mutex);
171 ret = btrfs_add_log_tree(trans, root);
175 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
176 root->log_start_pid = current->pid;
179 atomic_inc(&root->log_batch);
180 atomic_inc(&root->log_writers);
182 int index = root->log_transid % 2;
183 list_add_tail(&ctx->list, &root->log_ctxs[index]);
184 ctx->log_transid = root->log_transid;
188 mutex_unlock(&root->log_mutex);
193 * returns 0 if there was a log transaction running and we were able
194 * to join, or returns -ENOENT if there were not transactions
197 static int join_running_log_trans(struct btrfs_root *root)
205 mutex_lock(&root->log_mutex);
206 if (root->log_root) {
208 atomic_inc(&root->log_writers);
210 mutex_unlock(&root->log_mutex);
215 * This either makes the current running log transaction wait
216 * until you call btrfs_end_log_trans() or it makes any future
217 * log transactions wait until you call btrfs_end_log_trans()
219 int btrfs_pin_log_trans(struct btrfs_root *root)
223 mutex_lock(&root->log_mutex);
224 atomic_inc(&root->log_writers);
225 mutex_unlock(&root->log_mutex);
230 * indicate we're done making changes to the log tree
231 * and wake up anyone waiting to do a sync
233 void btrfs_end_log_trans(struct btrfs_root *root)
235 if (atomic_dec_and_test(&root->log_writers)) {
237 * Implicit memory barrier after atomic_dec_and_test
239 if (waitqueue_active(&root->log_writer_wait))
240 wake_up(&root->log_writer_wait);
246 * the walk control struct is used to pass state down the chain when
247 * processing the log tree. The stage field tells us which part
248 * of the log tree processing we are currently doing. The others
249 * are state fields used for that specific part
251 struct walk_control {
252 /* should we free the extent on disk when done? This is used
253 * at transaction commit time while freeing a log tree
257 /* should we write out the extent buffer? This is used
258 * while flushing the log tree to disk during a sync
262 /* should we wait for the extent buffer io to finish? Also used
263 * while flushing the log tree to disk for a sync
267 /* pin only walk, we record which extents on disk belong to the
272 /* what stage of the replay code we're currently in */
275 /* the root we are currently replaying */
276 struct btrfs_root *replay_dest;
278 /* the trans handle for the current replay */
279 struct btrfs_trans_handle *trans;
281 /* the function that gets used to process blocks we find in the
282 * tree. Note the extent_buffer might not be up to date when it is
283 * passed in, and it must be checked or read if you need the data
286 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
287 struct walk_control *wc, u64 gen);
291 * process_func used to pin down extents, write them or wait on them
293 static int process_one_buffer(struct btrfs_root *log,
294 struct extent_buffer *eb,
295 struct walk_control *wc, u64 gen)
300 * If this fs is mixed then we need to be able to process the leaves to
301 * pin down any logged extents, so we have to read the block.
303 if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) {
304 ret = btrfs_read_buffer(eb, gen);
310 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
313 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
314 if (wc->pin && btrfs_header_level(eb) == 0)
315 ret = btrfs_exclude_logged_extents(log, eb);
317 btrfs_write_tree_block(eb);
319 btrfs_wait_tree_block_writeback(eb);
325 * Item overwrite used by replay and tree logging. eb, slot and key all refer
326 * to the src data we are copying out.
328 * root is the tree we are copying into, and path is a scratch
329 * path for use in this function (it should be released on entry and
330 * will be released on exit).
332 * If the key is already in the destination tree the existing item is
333 * overwritten. If the existing item isn't big enough, it is extended.
334 * If it is too large, it is truncated.
336 * If the key isn't in the destination yet, a new item is inserted.
338 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
339 struct btrfs_root *root,
340 struct btrfs_path *path,
341 struct extent_buffer *eb, int slot,
342 struct btrfs_key *key)
346 u64 saved_i_size = 0;
347 int save_old_i_size = 0;
348 unsigned long src_ptr;
349 unsigned long dst_ptr;
350 int overwrite_root = 0;
351 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
353 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
356 item_size = btrfs_item_size_nr(eb, slot);
357 src_ptr = btrfs_item_ptr_offset(eb, slot);
359 /* look for the key in the destination tree */
360 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
367 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
369 if (dst_size != item_size)
372 if (item_size == 0) {
373 btrfs_release_path(path);
376 dst_copy = kmalloc(item_size, GFP_NOFS);
377 src_copy = kmalloc(item_size, GFP_NOFS);
378 if (!dst_copy || !src_copy) {
379 btrfs_release_path(path);
385 read_extent_buffer(eb, src_copy, src_ptr, item_size);
387 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
388 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
390 ret = memcmp(dst_copy, src_copy, item_size);
395 * they have the same contents, just return, this saves
396 * us from cowing blocks in the destination tree and doing
397 * extra writes that may not have been done by a previous
401 btrfs_release_path(path);
406 * We need to load the old nbytes into the inode so when we
407 * replay the extents we've logged we get the right nbytes.
410 struct btrfs_inode_item *item;
414 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
415 struct btrfs_inode_item);
416 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
417 item = btrfs_item_ptr(eb, slot,
418 struct btrfs_inode_item);
419 btrfs_set_inode_nbytes(eb, item, nbytes);
422 * If this is a directory we need to reset the i_size to
423 * 0 so that we can set it up properly when replaying
424 * the rest of the items in this log.
426 mode = btrfs_inode_mode(eb, item);
428 btrfs_set_inode_size(eb, item, 0);
430 } else if (inode_item) {
431 struct btrfs_inode_item *item;
435 * New inode, set nbytes to 0 so that the nbytes comes out
436 * properly when we replay the extents.
438 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
439 btrfs_set_inode_nbytes(eb, item, 0);
442 * If this is a directory we need to reset the i_size to 0 so
443 * that we can set it up properly when replaying the rest of
444 * the items in this log.
446 mode = btrfs_inode_mode(eb, item);
448 btrfs_set_inode_size(eb, item, 0);
451 btrfs_release_path(path);
452 /* try to insert the key into the destination tree */
453 path->skip_release_on_error = 1;
454 ret = btrfs_insert_empty_item(trans, root, path,
456 path->skip_release_on_error = 0;
458 /* make sure any existing item is the correct size */
459 if (ret == -EEXIST || ret == -EOVERFLOW) {
461 found_size = btrfs_item_size_nr(path->nodes[0],
463 if (found_size > item_size)
464 btrfs_truncate_item(root, path, item_size, 1);
465 else if (found_size < item_size)
466 btrfs_extend_item(root, path,
467 item_size - found_size);
471 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
474 /* don't overwrite an existing inode if the generation number
475 * was logged as zero. This is done when the tree logging code
476 * is just logging an inode to make sure it exists after recovery.
478 * Also, don't overwrite i_size on directories during replay.
479 * log replay inserts and removes directory items based on the
480 * state of the tree found in the subvolume, and i_size is modified
483 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
484 struct btrfs_inode_item *src_item;
485 struct btrfs_inode_item *dst_item;
487 src_item = (struct btrfs_inode_item *)src_ptr;
488 dst_item = (struct btrfs_inode_item *)dst_ptr;
490 if (btrfs_inode_generation(eb, src_item) == 0) {
491 struct extent_buffer *dst_eb = path->nodes[0];
492 const u64 ino_size = btrfs_inode_size(eb, src_item);
495 * For regular files an ino_size == 0 is used only when
496 * logging that an inode exists, as part of a directory
497 * fsync, and the inode wasn't fsynced before. In this
498 * case don't set the size of the inode in the fs/subvol
499 * tree, otherwise we would be throwing valid data away.
501 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
502 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
504 struct btrfs_map_token token;
506 btrfs_init_map_token(&token);
507 btrfs_set_token_inode_size(dst_eb, dst_item,
513 if (overwrite_root &&
514 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
515 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
517 saved_i_size = btrfs_inode_size(path->nodes[0],
522 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
525 if (save_old_i_size) {
526 struct btrfs_inode_item *dst_item;
527 dst_item = (struct btrfs_inode_item *)dst_ptr;
528 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
531 /* make sure the generation is filled in */
532 if (key->type == BTRFS_INODE_ITEM_KEY) {
533 struct btrfs_inode_item *dst_item;
534 dst_item = (struct btrfs_inode_item *)dst_ptr;
535 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
536 btrfs_set_inode_generation(path->nodes[0], dst_item,
541 btrfs_mark_buffer_dirty(path->nodes[0]);
542 btrfs_release_path(path);
547 * simple helper to read an inode off the disk from a given root
548 * This can only be called for subvolume roots and not for the log
550 static noinline struct inode *read_one_inode(struct btrfs_root *root,
553 struct btrfs_key key;
556 key.objectid = objectid;
557 key.type = BTRFS_INODE_ITEM_KEY;
559 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
562 } else if (is_bad_inode(inode)) {
569 /* replays a single extent in 'eb' at 'slot' with 'key' into the
570 * subvolume 'root'. path is released on entry and should be released
573 * extents in the log tree have not been allocated out of the extent
574 * tree yet. So, this completes the allocation, taking a reference
575 * as required if the extent already exists or creating a new extent
576 * if it isn't in the extent allocation tree yet.
578 * The extent is inserted into the file, dropping any existing extents
579 * from the file that overlap the new one.
581 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
582 struct btrfs_root *root,
583 struct btrfs_path *path,
584 struct extent_buffer *eb, int slot,
585 struct btrfs_key *key)
589 u64 start = key->offset;
591 struct btrfs_file_extent_item *item;
592 struct inode *inode = NULL;
596 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
597 found_type = btrfs_file_extent_type(eb, item);
599 if (found_type == BTRFS_FILE_EXTENT_REG ||
600 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
601 nbytes = btrfs_file_extent_num_bytes(eb, item);
602 extent_end = start + nbytes;
605 * We don't add to the inodes nbytes if we are prealloc or a
608 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
610 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
611 size = btrfs_file_extent_inline_len(eb, slot, item);
612 nbytes = btrfs_file_extent_ram_bytes(eb, item);
613 extent_end = ALIGN(start + size, root->sectorsize);
619 inode = read_one_inode(root, key->objectid);
626 * first check to see if we already have this extent in the
627 * file. This must be done before the btrfs_drop_extents run
628 * so we don't try to drop this extent.
630 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
634 (found_type == BTRFS_FILE_EXTENT_REG ||
635 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
636 struct btrfs_file_extent_item cmp1;
637 struct btrfs_file_extent_item cmp2;
638 struct btrfs_file_extent_item *existing;
639 struct extent_buffer *leaf;
641 leaf = path->nodes[0];
642 existing = btrfs_item_ptr(leaf, path->slots[0],
643 struct btrfs_file_extent_item);
645 read_extent_buffer(eb, &cmp1, (unsigned long)item,
647 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
651 * we already have a pointer to this exact extent,
652 * we don't have to do anything
654 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
655 btrfs_release_path(path);
659 btrfs_release_path(path);
661 /* drop any overlapping extents */
662 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
666 if (found_type == BTRFS_FILE_EXTENT_REG ||
667 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
669 unsigned long dest_offset;
670 struct btrfs_key ins;
672 ret = btrfs_insert_empty_item(trans, root, path, key,
676 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
678 copy_extent_buffer(path->nodes[0], eb, dest_offset,
679 (unsigned long)item, sizeof(*item));
681 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
682 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
683 ins.type = BTRFS_EXTENT_ITEM_KEY;
684 offset = key->offset - btrfs_file_extent_offset(eb, item);
687 * Manually record dirty extent, as here we did a shallow
688 * file extent item copy and skip normal backref update,
689 * but modifying extent tree all by ourselves.
690 * So need to manually record dirty extent for qgroup,
691 * as the owner of the file extent changed from log tree
692 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
694 ret = btrfs_qgroup_insert_dirty_extent(trans, root->fs_info,
695 btrfs_file_extent_disk_bytenr(eb, item),
696 btrfs_file_extent_disk_num_bytes(eb, item),
701 if (ins.objectid > 0) {
704 LIST_HEAD(ordered_sums);
706 * is this extent already allocated in the extent
707 * allocation tree? If so, just add a reference
709 ret = btrfs_lookup_data_extent(root, ins.objectid,
712 ret = btrfs_inc_extent_ref(trans, root,
713 ins.objectid, ins.offset,
714 0, root->root_key.objectid,
715 key->objectid, offset);
720 * insert the extent pointer in the extent
723 ret = btrfs_alloc_logged_file_extent(trans,
724 root, root->root_key.objectid,
725 key->objectid, offset, &ins);
729 btrfs_release_path(path);
731 if (btrfs_file_extent_compression(eb, item)) {
732 csum_start = ins.objectid;
733 csum_end = csum_start + ins.offset;
735 csum_start = ins.objectid +
736 btrfs_file_extent_offset(eb, item);
737 csum_end = csum_start +
738 btrfs_file_extent_num_bytes(eb, item);
741 ret = btrfs_lookup_csums_range(root->log_root,
742 csum_start, csum_end - 1,
747 * Now delete all existing cums in the csum root that
748 * cover our range. We do this because we can have an
749 * extent that is completely referenced by one file
750 * extent item and partially referenced by another
751 * file extent item (like after using the clone or
752 * extent_same ioctls). In this case if we end up doing
753 * the replay of the one that partially references the
754 * extent first, and we do not do the csum deletion
755 * below, we can get 2 csum items in the csum tree that
756 * overlap each other. For example, imagine our log has
757 * the two following file extent items:
759 * key (257 EXTENT_DATA 409600)
760 * extent data disk byte 12845056 nr 102400
761 * extent data offset 20480 nr 20480 ram 102400
763 * key (257 EXTENT_DATA 819200)
764 * extent data disk byte 12845056 nr 102400
765 * extent data offset 0 nr 102400 ram 102400
767 * Where the second one fully references the 100K extent
768 * that starts at disk byte 12845056, and the log tree
769 * has a single csum item that covers the entire range
772 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
774 * After the first file extent item is replayed, the
775 * csum tree gets the following csum item:
777 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
779 * Which covers the 20K sub-range starting at offset 20K
780 * of our extent. Now when we replay the second file
781 * extent item, if we do not delete existing csum items
782 * that cover any of its blocks, we end up getting two
783 * csum items in our csum tree that overlap each other:
785 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
786 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
788 * Which is a problem, because after this anyone trying
789 * to lookup up for the checksum of any block of our
790 * extent starting at an offset of 40K or higher, will
791 * end up looking at the second csum item only, which
792 * does not contain the checksum for any block starting
793 * at offset 40K or higher of our extent.
795 while (!list_empty(&ordered_sums)) {
796 struct btrfs_ordered_sum *sums;
797 sums = list_entry(ordered_sums.next,
798 struct btrfs_ordered_sum,
801 ret = btrfs_del_csums(trans,
802 root->fs_info->csum_root,
806 ret = btrfs_csum_file_blocks(trans,
807 root->fs_info->csum_root,
809 list_del(&sums->list);
815 btrfs_release_path(path);
817 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
818 /* inline extents are easy, we just overwrite them */
819 ret = overwrite_item(trans, root, path, eb, slot, key);
824 inode_add_bytes(inode, nbytes);
825 ret = btrfs_update_inode(trans, root, inode);
833 * when cleaning up conflicts between the directory names in the
834 * subvolume, directory names in the log and directory names in the
835 * inode back references, we may have to unlink inodes from directories.
837 * This is a helper function to do the unlink of a specific directory
840 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
841 struct btrfs_root *root,
842 struct btrfs_path *path,
844 struct btrfs_dir_item *di)
849 struct extent_buffer *leaf;
850 struct btrfs_key location;
853 leaf = path->nodes[0];
855 btrfs_dir_item_key_to_cpu(leaf, di, &location);
856 name_len = btrfs_dir_name_len(leaf, di);
857 name = kmalloc(name_len, GFP_NOFS);
861 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
862 btrfs_release_path(path);
864 inode = read_one_inode(root, location.objectid);
870 ret = link_to_fixup_dir(trans, root, path, location.objectid);
874 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
878 ret = btrfs_run_delayed_items(trans, root);
886 * helper function to see if a given name and sequence number found
887 * in an inode back reference are already in a directory and correctly
888 * point to this inode
890 static noinline int inode_in_dir(struct btrfs_root *root,
891 struct btrfs_path *path,
892 u64 dirid, u64 objectid, u64 index,
893 const char *name, int name_len)
895 struct btrfs_dir_item *di;
896 struct btrfs_key location;
899 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
900 index, name, name_len, 0);
901 if (di && !IS_ERR(di)) {
902 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
903 if (location.objectid != objectid)
907 btrfs_release_path(path);
909 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
910 if (di && !IS_ERR(di)) {
911 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
912 if (location.objectid != objectid)
918 btrfs_release_path(path);
923 * helper function to check a log tree for a named back reference in
924 * an inode. This is used to decide if a back reference that is
925 * found in the subvolume conflicts with what we find in the log.
927 * inode backreferences may have multiple refs in a single item,
928 * during replay we process one reference at a time, and we don't
929 * want to delete valid links to a file from the subvolume if that
930 * link is also in the log.
932 static noinline int backref_in_log(struct btrfs_root *log,
933 struct btrfs_key *key,
935 const char *name, int namelen)
937 struct btrfs_path *path;
938 struct btrfs_inode_ref *ref;
940 unsigned long ptr_end;
941 unsigned long name_ptr;
947 path = btrfs_alloc_path();
951 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
955 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
957 if (key->type == BTRFS_INODE_EXTREF_KEY) {
958 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
959 name, namelen, NULL))
965 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
966 ptr_end = ptr + item_size;
967 while (ptr < ptr_end) {
968 ref = (struct btrfs_inode_ref *)ptr;
969 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
970 if (found_name_len == namelen) {
971 name_ptr = (unsigned long)(ref + 1);
972 ret = memcmp_extent_buffer(path->nodes[0], name,
979 ptr = (unsigned long)(ref + 1) + found_name_len;
982 btrfs_free_path(path);
986 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
987 struct btrfs_root *root,
988 struct btrfs_path *path,
989 struct btrfs_root *log_root,
990 struct inode *dir, struct inode *inode,
991 struct extent_buffer *eb,
992 u64 inode_objectid, u64 parent_objectid,
993 u64 ref_index, char *name, int namelen,
999 struct extent_buffer *leaf;
1000 struct btrfs_dir_item *di;
1001 struct btrfs_key search_key;
1002 struct btrfs_inode_extref *extref;
1005 /* Search old style refs */
1006 search_key.objectid = inode_objectid;
1007 search_key.type = BTRFS_INODE_REF_KEY;
1008 search_key.offset = parent_objectid;
1009 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1011 struct btrfs_inode_ref *victim_ref;
1013 unsigned long ptr_end;
1015 leaf = path->nodes[0];
1017 /* are we trying to overwrite a back ref for the root directory
1018 * if so, just jump out, we're done
1020 if (search_key.objectid == search_key.offset)
1023 /* check all the names in this back reference to see
1024 * if they are in the log. if so, we allow them to stay
1025 * otherwise they must be unlinked as a conflict
1027 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1028 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1029 while (ptr < ptr_end) {
1030 victim_ref = (struct btrfs_inode_ref *)ptr;
1031 victim_name_len = btrfs_inode_ref_name_len(leaf,
1033 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1037 read_extent_buffer(leaf, victim_name,
1038 (unsigned long)(victim_ref + 1),
1041 if (!backref_in_log(log_root, &search_key,
1046 btrfs_release_path(path);
1048 ret = btrfs_unlink_inode(trans, root, dir,
1054 ret = btrfs_run_delayed_items(trans, root);
1062 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1066 * NOTE: we have searched root tree and checked the
1067 * corresponding ref, it does not need to check again.
1071 btrfs_release_path(path);
1073 /* Same search but for extended refs */
1074 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1075 inode_objectid, parent_objectid, 0,
1077 if (!IS_ERR_OR_NULL(extref)) {
1081 struct inode *victim_parent;
1083 leaf = path->nodes[0];
1085 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1086 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1088 while (cur_offset < item_size) {
1089 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1091 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1093 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1096 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1099 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1102 search_key.objectid = inode_objectid;
1103 search_key.type = BTRFS_INODE_EXTREF_KEY;
1104 search_key.offset = btrfs_extref_hash(parent_objectid,
1108 if (!backref_in_log(log_root, &search_key,
1109 parent_objectid, victim_name,
1112 victim_parent = read_one_inode(root,
1114 if (victim_parent) {
1116 btrfs_release_path(path);
1118 ret = btrfs_unlink_inode(trans, root,
1124 ret = btrfs_run_delayed_items(
1127 iput(victim_parent);
1138 cur_offset += victim_name_len + sizeof(*extref);
1142 btrfs_release_path(path);
1144 /* look for a conflicting sequence number */
1145 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1146 ref_index, name, namelen, 0);
1147 if (di && !IS_ERR(di)) {
1148 ret = drop_one_dir_item(trans, root, path, dir, di);
1152 btrfs_release_path(path);
1154 /* look for a conflicing name */
1155 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1157 if (di && !IS_ERR(di)) {
1158 ret = drop_one_dir_item(trans, root, path, dir, di);
1162 btrfs_release_path(path);
1167 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1168 u32 *namelen, char **name, u64 *index,
1169 u64 *parent_objectid)
1171 struct btrfs_inode_extref *extref;
1173 extref = (struct btrfs_inode_extref *)ref_ptr;
1175 *namelen = btrfs_inode_extref_name_len(eb, extref);
1176 *name = kmalloc(*namelen, GFP_NOFS);
1180 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1183 *index = btrfs_inode_extref_index(eb, extref);
1184 if (parent_objectid)
1185 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1190 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1191 u32 *namelen, char **name, u64 *index)
1193 struct btrfs_inode_ref *ref;
1195 ref = (struct btrfs_inode_ref *)ref_ptr;
1197 *namelen = btrfs_inode_ref_name_len(eb, ref);
1198 *name = kmalloc(*namelen, GFP_NOFS);
1202 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1204 *index = btrfs_inode_ref_index(eb, ref);
1210 * replay one inode back reference item found in the log tree.
1211 * eb, slot and key refer to the buffer and key found in the log tree.
1212 * root is the destination we are replaying into, and path is for temp
1213 * use by this function. (it should be released on return).
1215 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1216 struct btrfs_root *root,
1217 struct btrfs_root *log,
1218 struct btrfs_path *path,
1219 struct extent_buffer *eb, int slot,
1220 struct btrfs_key *key)
1222 struct inode *dir = NULL;
1223 struct inode *inode = NULL;
1224 unsigned long ref_ptr;
1225 unsigned long ref_end;
1229 int search_done = 0;
1230 int log_ref_ver = 0;
1231 u64 parent_objectid;
1234 int ref_struct_size;
1236 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1237 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1239 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1240 struct btrfs_inode_extref *r;
1242 ref_struct_size = sizeof(struct btrfs_inode_extref);
1244 r = (struct btrfs_inode_extref *)ref_ptr;
1245 parent_objectid = btrfs_inode_extref_parent(eb, r);
1247 ref_struct_size = sizeof(struct btrfs_inode_ref);
1248 parent_objectid = key->offset;
1250 inode_objectid = key->objectid;
1253 * it is possible that we didn't log all the parent directories
1254 * for a given inode. If we don't find the dir, just don't
1255 * copy the back ref in. The link count fixup code will take
1258 dir = read_one_inode(root, parent_objectid);
1264 inode = read_one_inode(root, inode_objectid);
1270 while (ref_ptr < ref_end) {
1272 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1273 &ref_index, &parent_objectid);
1275 * parent object can change from one array
1279 dir = read_one_inode(root, parent_objectid);
1285 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1291 /* if we already have a perfect match, we're done */
1292 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
1293 ref_index, name, namelen)) {
1295 * look for a conflicting back reference in the
1296 * metadata. if we find one we have to unlink that name
1297 * of the file before we add our new link. Later on, we
1298 * overwrite any existing back reference, and we don't
1299 * want to create dangling pointers in the directory.
1303 ret = __add_inode_ref(trans, root, path, log,
1307 ref_index, name, namelen,
1316 /* insert our name */
1317 ret = btrfs_add_link(trans, dir, inode, name, namelen,
1322 btrfs_update_inode(trans, root, inode);
1325 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1334 /* finally write the back reference in the inode */
1335 ret = overwrite_item(trans, root, path, eb, slot, key);
1337 btrfs_release_path(path);
1344 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1345 struct btrfs_root *root, u64 ino)
1349 ret = btrfs_insert_orphan_item(trans, root, ino);
1356 static int count_inode_extrefs(struct btrfs_root *root,
1357 struct inode *inode, struct btrfs_path *path)
1361 unsigned int nlink = 0;
1364 u64 inode_objectid = btrfs_ino(inode);
1367 struct btrfs_inode_extref *extref;
1368 struct extent_buffer *leaf;
1371 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1376 leaf = path->nodes[0];
1377 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1378 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1381 while (cur_offset < item_size) {
1382 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1383 name_len = btrfs_inode_extref_name_len(leaf, extref);
1387 cur_offset += name_len + sizeof(*extref);
1391 btrfs_release_path(path);
1393 btrfs_release_path(path);
1395 if (ret < 0 && ret != -ENOENT)
1400 static int count_inode_refs(struct btrfs_root *root,
1401 struct inode *inode, struct btrfs_path *path)
1404 struct btrfs_key key;
1405 unsigned int nlink = 0;
1407 unsigned long ptr_end;
1409 u64 ino = btrfs_ino(inode);
1412 key.type = BTRFS_INODE_REF_KEY;
1413 key.offset = (u64)-1;
1416 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1420 if (path->slots[0] == 0)
1425 btrfs_item_key_to_cpu(path->nodes[0], &key,
1427 if (key.objectid != ino ||
1428 key.type != BTRFS_INODE_REF_KEY)
1430 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1431 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1433 while (ptr < ptr_end) {
1434 struct btrfs_inode_ref *ref;
1436 ref = (struct btrfs_inode_ref *)ptr;
1437 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1439 ptr = (unsigned long)(ref + 1) + name_len;
1443 if (key.offset == 0)
1445 if (path->slots[0] > 0) {
1450 btrfs_release_path(path);
1452 btrfs_release_path(path);
1458 * There are a few corners where the link count of the file can't
1459 * be properly maintained during replay. So, instead of adding
1460 * lots of complexity to the log code, we just scan the backrefs
1461 * for any file that has been through replay.
1463 * The scan will update the link count on the inode to reflect the
1464 * number of back refs found. If it goes down to zero, the iput
1465 * will free the inode.
1467 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1468 struct btrfs_root *root,
1469 struct inode *inode)
1471 struct btrfs_path *path;
1474 u64 ino = btrfs_ino(inode);
1476 path = btrfs_alloc_path();
1480 ret = count_inode_refs(root, inode, path);
1486 ret = count_inode_extrefs(root, inode, path);
1494 if (nlink != inode->i_nlink) {
1495 set_nlink(inode, nlink);
1496 btrfs_update_inode(trans, root, inode);
1498 BTRFS_I(inode)->index_cnt = (u64)-1;
1500 if (inode->i_nlink == 0) {
1501 if (S_ISDIR(inode->i_mode)) {
1502 ret = replay_dir_deletes(trans, root, NULL, path,
1507 ret = insert_orphan_item(trans, root, ino);
1511 btrfs_free_path(path);
1515 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1516 struct btrfs_root *root,
1517 struct btrfs_path *path)
1520 struct btrfs_key key;
1521 struct inode *inode;
1523 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1524 key.type = BTRFS_ORPHAN_ITEM_KEY;
1525 key.offset = (u64)-1;
1527 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1533 if (path->slots[0] == 0)
1538 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1539 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1540 key.type != BTRFS_ORPHAN_ITEM_KEY)
1543 ret = btrfs_del_item(trans, root, path);
1547 btrfs_release_path(path);
1548 inode = read_one_inode(root, key.offset);
1554 ret = fixup_inode_link_count(trans, root, inode);
1560 * fixup on a directory may create new entries,
1561 * make sure we always look for the highset possible
1564 key.offset = (u64)-1;
1566 btrfs_release_path(path);
1572 * record a given inode in the fixup dir so we can check its link
1573 * count when replay is done. The link count is incremented here
1574 * so the inode won't go away until we check it
1576 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1577 struct btrfs_root *root,
1578 struct btrfs_path *path,
1581 struct btrfs_key key;
1583 struct inode *inode;
1585 inode = read_one_inode(root, objectid);
1589 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1590 key.type = BTRFS_ORPHAN_ITEM_KEY;
1591 key.offset = objectid;
1593 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1595 btrfs_release_path(path);
1597 if (!inode->i_nlink)
1598 set_nlink(inode, 1);
1601 ret = btrfs_update_inode(trans, root, inode);
1602 } else if (ret == -EEXIST) {
1611 * when replaying the log for a directory, we only insert names
1612 * for inodes that actually exist. This means an fsync on a directory
1613 * does not implicitly fsync all the new files in it
1615 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1616 struct btrfs_root *root,
1617 u64 dirid, u64 index,
1618 char *name, int name_len,
1619 struct btrfs_key *location)
1621 struct inode *inode;
1625 inode = read_one_inode(root, location->objectid);
1629 dir = read_one_inode(root, dirid);
1635 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1637 /* FIXME, put inode into FIXUP list */
1645 * Return true if an inode reference exists in the log for the given name,
1646 * inode and parent inode.
1648 static bool name_in_log_ref(struct btrfs_root *log_root,
1649 const char *name, const int name_len,
1650 const u64 dirid, const u64 ino)
1652 struct btrfs_key search_key;
1654 search_key.objectid = ino;
1655 search_key.type = BTRFS_INODE_REF_KEY;
1656 search_key.offset = dirid;
1657 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1660 search_key.type = BTRFS_INODE_EXTREF_KEY;
1661 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1662 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1669 * take a single entry in a log directory item and replay it into
1672 * if a conflicting item exists in the subdirectory already,
1673 * the inode it points to is unlinked and put into the link count
1676 * If a name from the log points to a file or directory that does
1677 * not exist in the FS, it is skipped. fsyncs on directories
1678 * do not force down inodes inside that directory, just changes to the
1679 * names or unlinks in a directory.
1681 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1682 * non-existing inode) and 1 if the name was replayed.
1684 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1685 struct btrfs_root *root,
1686 struct btrfs_path *path,
1687 struct extent_buffer *eb,
1688 struct btrfs_dir_item *di,
1689 struct btrfs_key *key)
1693 struct btrfs_dir_item *dst_di;
1694 struct btrfs_key found_key;
1695 struct btrfs_key log_key;
1700 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1701 bool name_added = false;
1703 dir = read_one_inode(root, key->objectid);
1707 name_len = btrfs_dir_name_len(eb, di);
1708 name = kmalloc(name_len, GFP_NOFS);
1714 log_type = btrfs_dir_type(eb, di);
1715 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1718 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1719 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1724 btrfs_release_path(path);
1726 if (key->type == BTRFS_DIR_ITEM_KEY) {
1727 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1729 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1730 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1739 if (IS_ERR_OR_NULL(dst_di)) {
1740 /* we need a sequence number to insert, so we only
1741 * do inserts for the BTRFS_DIR_INDEX_KEY types
1743 if (key->type != BTRFS_DIR_INDEX_KEY)
1748 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1749 /* the existing item matches the logged item */
1750 if (found_key.objectid == log_key.objectid &&
1751 found_key.type == log_key.type &&
1752 found_key.offset == log_key.offset &&
1753 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1754 update_size = false;
1759 * don't drop the conflicting directory entry if the inode
1760 * for the new entry doesn't exist
1765 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1769 if (key->type == BTRFS_DIR_INDEX_KEY)
1772 btrfs_release_path(path);
1773 if (!ret && update_size) {
1774 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1775 ret = btrfs_update_inode(trans, root, dir);
1779 if (!ret && name_added)
1784 if (name_in_log_ref(root->log_root, name, name_len,
1785 key->objectid, log_key.objectid)) {
1786 /* The dentry will be added later. */
1788 update_size = false;
1791 btrfs_release_path(path);
1792 ret = insert_one_name(trans, root, key->objectid, key->offset,
1793 name, name_len, &log_key);
1794 if (ret && ret != -ENOENT && ret != -EEXIST)
1798 update_size = false;
1804 * find all the names in a directory item and reconcile them into
1805 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1806 * one name in a directory item, but the same code gets used for
1807 * both directory index types
1809 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1810 struct btrfs_root *root,
1811 struct btrfs_path *path,
1812 struct extent_buffer *eb, int slot,
1813 struct btrfs_key *key)
1816 u32 item_size = btrfs_item_size_nr(eb, slot);
1817 struct btrfs_dir_item *di;
1820 unsigned long ptr_end;
1821 struct btrfs_path *fixup_path = NULL;
1823 ptr = btrfs_item_ptr_offset(eb, slot);
1824 ptr_end = ptr + item_size;
1825 while (ptr < ptr_end) {
1826 di = (struct btrfs_dir_item *)ptr;
1827 if (verify_dir_item(root, eb, di))
1829 name_len = btrfs_dir_name_len(eb, di);
1830 ret = replay_one_name(trans, root, path, eb, di, key);
1833 ptr = (unsigned long)(di + 1);
1837 * If this entry refers to a non-directory (directories can not
1838 * have a link count > 1) and it was added in the transaction
1839 * that was not committed, make sure we fixup the link count of
1840 * the inode it the entry points to. Otherwise something like
1841 * the following would result in a directory pointing to an
1842 * inode with a wrong link that does not account for this dir
1850 * ln testdir/bar testdir/bar_link
1851 * ln testdir/foo testdir/foo_link
1852 * xfs_io -c "fsync" testdir/bar
1856 * mount fs, log replay happens
1858 * File foo would remain with a link count of 1 when it has two
1859 * entries pointing to it in the directory testdir. This would
1860 * make it impossible to ever delete the parent directory has
1861 * it would result in stale dentries that can never be deleted.
1863 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1864 struct btrfs_key di_key;
1867 fixup_path = btrfs_alloc_path();
1874 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1875 ret = link_to_fixup_dir(trans, root, fixup_path,
1882 btrfs_free_path(fixup_path);
1887 * directory replay has two parts. There are the standard directory
1888 * items in the log copied from the subvolume, and range items
1889 * created in the log while the subvolume was logged.
1891 * The range items tell us which parts of the key space the log
1892 * is authoritative for. During replay, if a key in the subvolume
1893 * directory is in a logged range item, but not actually in the log
1894 * that means it was deleted from the directory before the fsync
1895 * and should be removed.
1897 static noinline int find_dir_range(struct btrfs_root *root,
1898 struct btrfs_path *path,
1899 u64 dirid, int key_type,
1900 u64 *start_ret, u64 *end_ret)
1902 struct btrfs_key key;
1904 struct btrfs_dir_log_item *item;
1908 if (*start_ret == (u64)-1)
1911 key.objectid = dirid;
1912 key.type = key_type;
1913 key.offset = *start_ret;
1915 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1919 if (path->slots[0] == 0)
1924 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1926 if (key.type != key_type || key.objectid != dirid) {
1930 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1931 struct btrfs_dir_log_item);
1932 found_end = btrfs_dir_log_end(path->nodes[0], item);
1934 if (*start_ret >= key.offset && *start_ret <= found_end) {
1936 *start_ret = key.offset;
1937 *end_ret = found_end;
1942 /* check the next slot in the tree to see if it is a valid item */
1943 nritems = btrfs_header_nritems(path->nodes[0]);
1945 if (path->slots[0] >= nritems) {
1946 ret = btrfs_next_leaf(root, path);
1951 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1953 if (key.type != key_type || key.objectid != dirid) {
1957 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1958 struct btrfs_dir_log_item);
1959 found_end = btrfs_dir_log_end(path->nodes[0], item);
1960 *start_ret = key.offset;
1961 *end_ret = found_end;
1964 btrfs_release_path(path);
1969 * this looks for a given directory item in the log. If the directory
1970 * item is not in the log, the item is removed and the inode it points
1973 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1974 struct btrfs_root *root,
1975 struct btrfs_root *log,
1976 struct btrfs_path *path,
1977 struct btrfs_path *log_path,
1979 struct btrfs_key *dir_key)
1982 struct extent_buffer *eb;
1985 struct btrfs_dir_item *di;
1986 struct btrfs_dir_item *log_di;
1989 unsigned long ptr_end;
1991 struct inode *inode;
1992 struct btrfs_key location;
1995 eb = path->nodes[0];
1996 slot = path->slots[0];
1997 item_size = btrfs_item_size_nr(eb, slot);
1998 ptr = btrfs_item_ptr_offset(eb, slot);
1999 ptr_end = ptr + item_size;
2000 while (ptr < ptr_end) {
2001 di = (struct btrfs_dir_item *)ptr;
2002 if (verify_dir_item(root, eb, di)) {
2007 name_len = btrfs_dir_name_len(eb, di);
2008 name = kmalloc(name_len, GFP_NOFS);
2013 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2016 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2017 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2020 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2021 log_di = btrfs_lookup_dir_index_item(trans, log,
2027 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
2028 btrfs_dir_item_key_to_cpu(eb, di, &location);
2029 btrfs_release_path(path);
2030 btrfs_release_path(log_path);
2031 inode = read_one_inode(root, location.objectid);
2037 ret = link_to_fixup_dir(trans, root,
2038 path, location.objectid);
2046 ret = btrfs_unlink_inode(trans, root, dir, inode,
2049 ret = btrfs_run_delayed_items(trans, root);
2055 /* there might still be more names under this key
2056 * check and repeat if required
2058 ret = btrfs_search_slot(NULL, root, dir_key, path,
2064 } else if (IS_ERR(log_di)) {
2066 return PTR_ERR(log_di);
2068 btrfs_release_path(log_path);
2071 ptr = (unsigned long)(di + 1);
2076 btrfs_release_path(path);
2077 btrfs_release_path(log_path);
2081 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2082 struct btrfs_root *root,
2083 struct btrfs_root *log,
2084 struct btrfs_path *path,
2087 struct btrfs_key search_key;
2088 struct btrfs_path *log_path;
2093 log_path = btrfs_alloc_path();
2097 search_key.objectid = ino;
2098 search_key.type = BTRFS_XATTR_ITEM_KEY;
2099 search_key.offset = 0;
2101 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2105 nritems = btrfs_header_nritems(path->nodes[0]);
2106 for (i = path->slots[0]; i < nritems; i++) {
2107 struct btrfs_key key;
2108 struct btrfs_dir_item *di;
2109 struct btrfs_dir_item *log_di;
2113 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2114 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2119 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2120 total_size = btrfs_item_size_nr(path->nodes[0], i);
2122 while (cur < total_size) {
2123 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2124 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2125 u32 this_len = sizeof(*di) + name_len + data_len;
2128 name = kmalloc(name_len, GFP_NOFS);
2133 read_extent_buffer(path->nodes[0], name,
2134 (unsigned long)(di + 1), name_len);
2136 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2138 btrfs_release_path(log_path);
2140 /* Doesn't exist in log tree, so delete it. */
2141 btrfs_release_path(path);
2142 di = btrfs_lookup_xattr(trans, root, path, ino,
2143 name, name_len, -1);
2150 ret = btrfs_delete_one_dir_name(trans, root,
2154 btrfs_release_path(path);
2159 if (IS_ERR(log_di)) {
2160 ret = PTR_ERR(log_di);
2164 di = (struct btrfs_dir_item *)((char *)di + this_len);
2167 ret = btrfs_next_leaf(root, path);
2173 btrfs_free_path(log_path);
2174 btrfs_release_path(path);
2180 * deletion replay happens before we copy any new directory items
2181 * out of the log or out of backreferences from inodes. It
2182 * scans the log to find ranges of keys that log is authoritative for,
2183 * and then scans the directory to find items in those ranges that are
2184 * not present in the log.
2186 * Anything we don't find in the log is unlinked and removed from the
2189 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2190 struct btrfs_root *root,
2191 struct btrfs_root *log,
2192 struct btrfs_path *path,
2193 u64 dirid, int del_all)
2197 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2199 struct btrfs_key dir_key;
2200 struct btrfs_key found_key;
2201 struct btrfs_path *log_path;
2204 dir_key.objectid = dirid;
2205 dir_key.type = BTRFS_DIR_ITEM_KEY;
2206 log_path = btrfs_alloc_path();
2210 dir = read_one_inode(root, dirid);
2211 /* it isn't an error if the inode isn't there, that can happen
2212 * because we replay the deletes before we copy in the inode item
2216 btrfs_free_path(log_path);
2224 range_end = (u64)-1;
2226 ret = find_dir_range(log, path, dirid, key_type,
2227 &range_start, &range_end);
2232 dir_key.offset = range_start;
2235 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2240 nritems = btrfs_header_nritems(path->nodes[0]);
2241 if (path->slots[0] >= nritems) {
2242 ret = btrfs_next_leaf(root, path);
2248 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2250 if (found_key.objectid != dirid ||
2251 found_key.type != dir_key.type)
2254 if (found_key.offset > range_end)
2257 ret = check_item_in_log(trans, root, log, path,
2262 if (found_key.offset == (u64)-1)
2264 dir_key.offset = found_key.offset + 1;
2266 btrfs_release_path(path);
2267 if (range_end == (u64)-1)
2269 range_start = range_end + 1;
2274 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2275 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2276 dir_key.type = BTRFS_DIR_INDEX_KEY;
2277 btrfs_release_path(path);
2281 btrfs_release_path(path);
2282 btrfs_free_path(log_path);
2288 * the process_func used to replay items from the log tree. This
2289 * gets called in two different stages. The first stage just looks
2290 * for inodes and makes sure they are all copied into the subvolume.
2292 * The second stage copies all the other item types from the log into
2293 * the subvolume. The two stage approach is slower, but gets rid of
2294 * lots of complexity around inodes referencing other inodes that exist
2295 * only in the log (references come from either directory items or inode
2298 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2299 struct walk_control *wc, u64 gen)
2302 struct btrfs_path *path;
2303 struct btrfs_root *root = wc->replay_dest;
2304 struct btrfs_key key;
2309 ret = btrfs_read_buffer(eb, gen);
2313 level = btrfs_header_level(eb);
2318 path = btrfs_alloc_path();
2322 nritems = btrfs_header_nritems(eb);
2323 for (i = 0; i < nritems; i++) {
2324 btrfs_item_key_to_cpu(eb, &key, i);
2326 /* inode keys are done during the first stage */
2327 if (key.type == BTRFS_INODE_ITEM_KEY &&
2328 wc->stage == LOG_WALK_REPLAY_INODES) {
2329 struct btrfs_inode_item *inode_item;
2332 inode_item = btrfs_item_ptr(eb, i,
2333 struct btrfs_inode_item);
2334 ret = replay_xattr_deletes(wc->trans, root, log,
2335 path, key.objectid);
2338 mode = btrfs_inode_mode(eb, inode_item);
2339 if (S_ISDIR(mode)) {
2340 ret = replay_dir_deletes(wc->trans,
2341 root, log, path, key.objectid, 0);
2345 ret = overwrite_item(wc->trans, root, path,
2350 /* for regular files, make sure corresponding
2351 * orphan item exist. extents past the new EOF
2352 * will be truncated later by orphan cleanup.
2354 if (S_ISREG(mode)) {
2355 ret = insert_orphan_item(wc->trans, root,
2361 ret = link_to_fixup_dir(wc->trans, root,
2362 path, key.objectid);
2367 if (key.type == BTRFS_DIR_INDEX_KEY &&
2368 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2369 ret = replay_one_dir_item(wc->trans, root, path,
2375 if (wc->stage < LOG_WALK_REPLAY_ALL)
2378 /* these keys are simply copied */
2379 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2380 ret = overwrite_item(wc->trans, root, path,
2384 } else if (key.type == BTRFS_INODE_REF_KEY ||
2385 key.type == BTRFS_INODE_EXTREF_KEY) {
2386 ret = add_inode_ref(wc->trans, root, log, path,
2388 if (ret && ret != -ENOENT)
2391 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2392 ret = replay_one_extent(wc->trans, root, path,
2396 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2397 ret = replay_one_dir_item(wc->trans, root, path,
2403 btrfs_free_path(path);
2407 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2408 struct btrfs_root *root,
2409 struct btrfs_path *path, int *level,
2410 struct walk_control *wc)
2415 struct extent_buffer *next;
2416 struct extent_buffer *cur;
2417 struct extent_buffer *parent;
2421 WARN_ON(*level < 0);
2422 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2424 while (*level > 0) {
2425 WARN_ON(*level < 0);
2426 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2427 cur = path->nodes[*level];
2429 WARN_ON(btrfs_header_level(cur) != *level);
2431 if (path->slots[*level] >=
2432 btrfs_header_nritems(cur))
2435 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2436 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2437 blocksize = root->nodesize;
2439 parent = path->nodes[*level];
2440 root_owner = btrfs_header_owner(parent);
2442 next = btrfs_find_create_tree_block(root, bytenr);
2444 return PTR_ERR(next);
2447 ret = wc->process_func(root, next, wc, ptr_gen);
2449 free_extent_buffer(next);
2453 path->slots[*level]++;
2455 ret = btrfs_read_buffer(next, ptr_gen);
2457 free_extent_buffer(next);
2462 btrfs_tree_lock(next);
2463 btrfs_set_lock_blocking(next);
2464 clean_tree_block(trans, root->fs_info,
2466 btrfs_wait_tree_block_writeback(next);
2467 btrfs_tree_unlock(next);
2469 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2470 clear_extent_buffer_dirty(next);
2473 WARN_ON(root_owner !=
2474 BTRFS_TREE_LOG_OBJECTID);
2475 ret = btrfs_free_and_pin_reserved_extent(root,
2478 free_extent_buffer(next);
2482 free_extent_buffer(next);
2485 ret = btrfs_read_buffer(next, ptr_gen);
2487 free_extent_buffer(next);
2491 WARN_ON(*level <= 0);
2492 if (path->nodes[*level-1])
2493 free_extent_buffer(path->nodes[*level-1]);
2494 path->nodes[*level-1] = next;
2495 *level = btrfs_header_level(next);
2496 path->slots[*level] = 0;
2499 WARN_ON(*level < 0);
2500 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2502 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2508 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2509 struct btrfs_root *root,
2510 struct btrfs_path *path, int *level,
2511 struct walk_control *wc)
2518 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2519 slot = path->slots[i];
2520 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2523 WARN_ON(*level == 0);
2526 struct extent_buffer *parent;
2527 if (path->nodes[*level] == root->node)
2528 parent = path->nodes[*level];
2530 parent = path->nodes[*level + 1];
2532 root_owner = btrfs_header_owner(parent);
2533 ret = wc->process_func(root, path->nodes[*level], wc,
2534 btrfs_header_generation(path->nodes[*level]));
2539 struct extent_buffer *next;
2541 next = path->nodes[*level];
2544 btrfs_tree_lock(next);
2545 btrfs_set_lock_blocking(next);
2546 clean_tree_block(trans, root->fs_info,
2548 btrfs_wait_tree_block_writeback(next);
2549 btrfs_tree_unlock(next);
2551 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2552 clear_extent_buffer_dirty(next);
2555 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2556 ret = btrfs_free_and_pin_reserved_extent(root,
2557 path->nodes[*level]->start,
2558 path->nodes[*level]->len);
2562 free_extent_buffer(path->nodes[*level]);
2563 path->nodes[*level] = NULL;
2571 * drop the reference count on the tree rooted at 'snap'. This traverses
2572 * the tree freeing any blocks that have a ref count of zero after being
2575 static int walk_log_tree(struct btrfs_trans_handle *trans,
2576 struct btrfs_root *log, struct walk_control *wc)
2581 struct btrfs_path *path;
2584 path = btrfs_alloc_path();
2588 level = btrfs_header_level(log->node);
2590 path->nodes[level] = log->node;
2591 extent_buffer_get(log->node);
2592 path->slots[level] = 0;
2595 wret = walk_down_log_tree(trans, log, path, &level, wc);
2603 wret = walk_up_log_tree(trans, log, path, &level, wc);
2612 /* was the root node processed? if not, catch it here */
2613 if (path->nodes[orig_level]) {
2614 ret = wc->process_func(log, path->nodes[orig_level], wc,
2615 btrfs_header_generation(path->nodes[orig_level]));
2619 struct extent_buffer *next;
2621 next = path->nodes[orig_level];
2624 btrfs_tree_lock(next);
2625 btrfs_set_lock_blocking(next);
2626 clean_tree_block(trans, log->fs_info, next);
2627 btrfs_wait_tree_block_writeback(next);
2628 btrfs_tree_unlock(next);
2630 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2631 clear_extent_buffer_dirty(next);
2634 WARN_ON(log->root_key.objectid !=
2635 BTRFS_TREE_LOG_OBJECTID);
2636 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2644 btrfs_free_path(path);
2649 * helper function to update the item for a given subvolumes log root
2650 * in the tree of log roots
2652 static int update_log_root(struct btrfs_trans_handle *trans,
2653 struct btrfs_root *log)
2657 if (log->log_transid == 1) {
2658 /* insert root item on the first sync */
2659 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2660 &log->root_key, &log->root_item);
2662 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2663 &log->root_key, &log->root_item);
2668 static void wait_log_commit(struct btrfs_root *root, int transid)
2671 int index = transid % 2;
2674 * we only allow two pending log transactions at a time,
2675 * so we know that if ours is more than 2 older than the
2676 * current transaction, we're done
2679 prepare_to_wait(&root->log_commit_wait[index],
2680 &wait, TASK_UNINTERRUPTIBLE);
2681 mutex_unlock(&root->log_mutex);
2683 if (root->log_transid_committed < transid &&
2684 atomic_read(&root->log_commit[index]))
2687 finish_wait(&root->log_commit_wait[index], &wait);
2688 mutex_lock(&root->log_mutex);
2689 } while (root->log_transid_committed < transid &&
2690 atomic_read(&root->log_commit[index]));
2693 static void wait_for_writer(struct btrfs_root *root)
2697 while (atomic_read(&root->log_writers)) {
2698 prepare_to_wait(&root->log_writer_wait,
2699 &wait, TASK_UNINTERRUPTIBLE);
2700 mutex_unlock(&root->log_mutex);
2701 if (atomic_read(&root->log_writers))
2703 finish_wait(&root->log_writer_wait, &wait);
2704 mutex_lock(&root->log_mutex);
2708 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2709 struct btrfs_log_ctx *ctx)
2714 mutex_lock(&root->log_mutex);
2715 list_del_init(&ctx->list);
2716 mutex_unlock(&root->log_mutex);
2720 * Invoked in log mutex context, or be sure there is no other task which
2721 * can access the list.
2723 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2724 int index, int error)
2726 struct btrfs_log_ctx *ctx;
2727 struct btrfs_log_ctx *safe;
2729 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2730 list_del_init(&ctx->list);
2731 ctx->log_ret = error;
2734 INIT_LIST_HEAD(&root->log_ctxs[index]);
2738 * btrfs_sync_log does sends a given tree log down to the disk and
2739 * updates the super blocks to record it. When this call is done,
2740 * you know that any inodes previously logged are safely on disk only
2743 * Any other return value means you need to call btrfs_commit_transaction.
2744 * Some of the edge cases for fsyncing directories that have had unlinks
2745 * or renames done in the past mean that sometimes the only safe
2746 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2747 * that has happened.
2749 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2750 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2756 struct btrfs_root *log = root->log_root;
2757 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2758 int log_transid = 0;
2759 struct btrfs_log_ctx root_log_ctx;
2760 struct blk_plug plug;
2762 mutex_lock(&root->log_mutex);
2763 log_transid = ctx->log_transid;
2764 if (root->log_transid_committed >= log_transid) {
2765 mutex_unlock(&root->log_mutex);
2766 return ctx->log_ret;
2769 index1 = log_transid % 2;
2770 if (atomic_read(&root->log_commit[index1])) {
2771 wait_log_commit(root, log_transid);
2772 mutex_unlock(&root->log_mutex);
2773 return ctx->log_ret;
2775 ASSERT(log_transid == root->log_transid);
2776 atomic_set(&root->log_commit[index1], 1);
2778 /* wait for previous tree log sync to complete */
2779 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2780 wait_log_commit(root, log_transid - 1);
2783 int batch = atomic_read(&root->log_batch);
2784 /* when we're on an ssd, just kick the log commit out */
2785 if (!btrfs_test_opt(root->fs_info, SSD) &&
2786 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2787 mutex_unlock(&root->log_mutex);
2788 schedule_timeout_uninterruptible(1);
2789 mutex_lock(&root->log_mutex);
2791 wait_for_writer(root);
2792 if (batch == atomic_read(&root->log_batch))
2796 /* bail out if we need to do a full commit */
2797 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2799 btrfs_free_logged_extents(log, log_transid);
2800 mutex_unlock(&root->log_mutex);
2804 if (log_transid % 2 == 0)
2805 mark = EXTENT_DIRTY;
2809 /* we start IO on all the marked extents here, but we don't actually
2810 * wait for them until later.
2812 blk_start_plug(&plug);
2813 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2815 blk_finish_plug(&plug);
2816 btrfs_abort_transaction(trans, ret);
2817 btrfs_free_logged_extents(log, log_transid);
2818 btrfs_set_log_full_commit(root->fs_info, trans);
2819 mutex_unlock(&root->log_mutex);
2823 btrfs_set_root_node(&log->root_item, log->node);
2825 root->log_transid++;
2826 log->log_transid = root->log_transid;
2827 root->log_start_pid = 0;
2829 * Update or create log root item under the root's log_mutex to prevent
2830 * races with concurrent log syncs that can lead to failure to update
2831 * log root item because it was not created yet.
2833 ret = update_log_root(trans, log);
2835 * IO has been started, blocks of the log tree have WRITTEN flag set
2836 * in their headers. new modifications of the log will be written to
2837 * new positions. so it's safe to allow log writers to go in.
2839 mutex_unlock(&root->log_mutex);
2841 btrfs_init_log_ctx(&root_log_ctx, NULL);
2843 mutex_lock(&log_root_tree->log_mutex);
2844 atomic_inc(&log_root_tree->log_batch);
2845 atomic_inc(&log_root_tree->log_writers);
2847 index2 = log_root_tree->log_transid % 2;
2848 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2849 root_log_ctx.log_transid = log_root_tree->log_transid;
2851 mutex_unlock(&log_root_tree->log_mutex);
2853 mutex_lock(&log_root_tree->log_mutex);
2854 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2856 * Implicit memory barrier after atomic_dec_and_test
2858 if (waitqueue_active(&log_root_tree->log_writer_wait))
2859 wake_up(&log_root_tree->log_writer_wait);
2863 if (!list_empty(&root_log_ctx.list))
2864 list_del_init(&root_log_ctx.list);
2866 blk_finish_plug(&plug);
2867 btrfs_set_log_full_commit(root->fs_info, trans);
2869 if (ret != -ENOSPC) {
2870 btrfs_abort_transaction(trans, ret);
2871 mutex_unlock(&log_root_tree->log_mutex);
2874 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2875 btrfs_free_logged_extents(log, log_transid);
2876 mutex_unlock(&log_root_tree->log_mutex);
2881 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2882 blk_finish_plug(&plug);
2883 list_del_init(&root_log_ctx.list);
2884 mutex_unlock(&log_root_tree->log_mutex);
2885 ret = root_log_ctx.log_ret;
2889 index2 = root_log_ctx.log_transid % 2;
2890 if (atomic_read(&log_root_tree->log_commit[index2])) {
2891 blk_finish_plug(&plug);
2892 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages,
2894 btrfs_wait_logged_extents(trans, log, log_transid);
2895 wait_log_commit(log_root_tree,
2896 root_log_ctx.log_transid);
2897 mutex_unlock(&log_root_tree->log_mutex);
2899 ret = root_log_ctx.log_ret;
2902 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
2903 atomic_set(&log_root_tree->log_commit[index2], 1);
2905 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2906 wait_log_commit(log_root_tree,
2907 root_log_ctx.log_transid - 1);
2910 wait_for_writer(log_root_tree);
2913 * now that we've moved on to the tree of log tree roots,
2914 * check the full commit flag again
2916 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2917 blk_finish_plug(&plug);
2918 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2919 btrfs_free_logged_extents(log, log_transid);
2920 mutex_unlock(&log_root_tree->log_mutex);
2922 goto out_wake_log_root;
2925 ret = btrfs_write_marked_extents(log_root_tree,
2926 &log_root_tree->dirty_log_pages,
2927 EXTENT_DIRTY | EXTENT_NEW);
2928 blk_finish_plug(&plug);
2930 btrfs_set_log_full_commit(root->fs_info, trans);
2931 btrfs_abort_transaction(trans, ret);
2932 btrfs_free_logged_extents(log, log_transid);
2933 mutex_unlock(&log_root_tree->log_mutex);
2934 goto out_wake_log_root;
2936 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2938 ret = btrfs_wait_marked_extents(log_root_tree,
2939 &log_root_tree->dirty_log_pages,
2940 EXTENT_NEW | EXTENT_DIRTY);
2942 btrfs_set_log_full_commit(root->fs_info, trans);
2943 btrfs_free_logged_extents(log, log_transid);
2944 mutex_unlock(&log_root_tree->log_mutex);
2945 goto out_wake_log_root;
2947 btrfs_wait_logged_extents(trans, log, log_transid);
2949 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2950 log_root_tree->node->start);
2951 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2952 btrfs_header_level(log_root_tree->node));
2954 log_root_tree->log_transid++;
2955 mutex_unlock(&log_root_tree->log_mutex);
2958 * nobody else is going to jump in and write the the ctree
2959 * super here because the log_commit atomic below is protecting
2960 * us. We must be called with a transaction handle pinning
2961 * the running transaction open, so a full commit can't hop
2962 * in and cause problems either.
2964 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2966 btrfs_set_log_full_commit(root->fs_info, trans);
2967 btrfs_abort_transaction(trans, ret);
2968 goto out_wake_log_root;
2971 mutex_lock(&root->log_mutex);
2972 if (root->last_log_commit < log_transid)
2973 root->last_log_commit = log_transid;
2974 mutex_unlock(&root->log_mutex);
2977 mutex_lock(&log_root_tree->log_mutex);
2978 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2980 log_root_tree->log_transid_committed++;
2981 atomic_set(&log_root_tree->log_commit[index2], 0);
2982 mutex_unlock(&log_root_tree->log_mutex);
2985 * The barrier before waitqueue_active is needed so all the updates
2986 * above are seen by the woken threads. It might not be necessary, but
2987 * proving that seems to be hard.
2990 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2991 wake_up(&log_root_tree->log_commit_wait[index2]);
2993 mutex_lock(&root->log_mutex);
2994 btrfs_remove_all_log_ctxs(root, index1, ret);
2995 root->log_transid_committed++;
2996 atomic_set(&root->log_commit[index1], 0);
2997 mutex_unlock(&root->log_mutex);
3000 * The barrier before waitqueue_active is needed so all the updates
3001 * above are seen by the woken threads. It might not be necessary, but
3002 * proving that seems to be hard.
3005 if (waitqueue_active(&root->log_commit_wait[index1]))
3006 wake_up(&root->log_commit_wait[index1]);
3010 static void free_log_tree(struct btrfs_trans_handle *trans,
3011 struct btrfs_root *log)
3016 struct walk_control wc = {
3018 .process_func = process_one_buffer
3021 ret = walk_log_tree(trans, log, &wc);
3024 btrfs_abort_transaction(trans, ret);
3026 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3030 ret = find_first_extent_bit(&log->dirty_log_pages,
3032 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT,
3037 clear_extent_bits(&log->dirty_log_pages, start, end,
3038 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3042 * We may have short-circuited the log tree with the full commit logic
3043 * and left ordered extents on our list, so clear these out to keep us
3044 * from leaking inodes and memory.
3046 btrfs_free_logged_extents(log, 0);
3047 btrfs_free_logged_extents(log, 1);
3049 free_extent_buffer(log->node);
3054 * free all the extents used by the tree log. This should be called
3055 * at commit time of the full transaction
3057 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3059 if (root->log_root) {
3060 free_log_tree(trans, root->log_root);
3061 root->log_root = NULL;
3066 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3067 struct btrfs_fs_info *fs_info)
3069 if (fs_info->log_root_tree) {
3070 free_log_tree(trans, fs_info->log_root_tree);
3071 fs_info->log_root_tree = NULL;
3077 * If both a file and directory are logged, and unlinks or renames are
3078 * mixed in, we have a few interesting corners:
3080 * create file X in dir Y
3081 * link file X to X.link in dir Y
3083 * unlink file X but leave X.link
3086 * After a crash we would expect only X.link to exist. But file X
3087 * didn't get fsync'd again so the log has back refs for X and X.link.
3089 * We solve this by removing directory entries and inode backrefs from the
3090 * log when a file that was logged in the current transaction is
3091 * unlinked. Any later fsync will include the updated log entries, and
3092 * we'll be able to reconstruct the proper directory items from backrefs.
3094 * This optimizations allows us to avoid relogging the entire inode
3095 * or the entire directory.
3097 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3098 struct btrfs_root *root,
3099 const char *name, int name_len,
3100 struct inode *dir, u64 index)
3102 struct btrfs_root *log;
3103 struct btrfs_dir_item *di;
3104 struct btrfs_path *path;
3108 u64 dir_ino = btrfs_ino(dir);
3110 if (BTRFS_I(dir)->logged_trans < trans->transid)
3113 ret = join_running_log_trans(root);
3117 mutex_lock(&BTRFS_I(dir)->log_mutex);
3119 log = root->log_root;
3120 path = btrfs_alloc_path();
3126 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3127 name, name_len, -1);
3133 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3134 bytes_del += name_len;
3140 btrfs_release_path(path);
3141 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3142 index, name, name_len, -1);
3148 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3149 bytes_del += name_len;
3156 /* update the directory size in the log to reflect the names
3160 struct btrfs_key key;
3162 key.objectid = dir_ino;
3164 key.type = BTRFS_INODE_ITEM_KEY;
3165 btrfs_release_path(path);
3167 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3173 struct btrfs_inode_item *item;
3176 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3177 struct btrfs_inode_item);
3178 i_size = btrfs_inode_size(path->nodes[0], item);
3179 if (i_size > bytes_del)
3180 i_size -= bytes_del;
3183 btrfs_set_inode_size(path->nodes[0], item, i_size);
3184 btrfs_mark_buffer_dirty(path->nodes[0]);
3187 btrfs_release_path(path);
3190 btrfs_free_path(path);
3192 mutex_unlock(&BTRFS_I(dir)->log_mutex);
3193 if (err == -ENOSPC) {
3194 btrfs_set_log_full_commit(root->fs_info, trans);
3196 } else if (err < 0 && err != -ENOENT) {
3197 /* ENOENT can be returned if the entry hasn't been fsynced yet */
3198 btrfs_abort_transaction(trans, err);
3201 btrfs_end_log_trans(root);
3206 /* see comments for btrfs_del_dir_entries_in_log */
3207 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3208 struct btrfs_root *root,
3209 const char *name, int name_len,
3210 struct inode *inode, u64 dirid)
3212 struct btrfs_root *log;
3216 if (BTRFS_I(inode)->logged_trans < trans->transid)
3219 ret = join_running_log_trans(root);
3222 log = root->log_root;
3223 mutex_lock(&BTRFS_I(inode)->log_mutex);
3225 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3227 mutex_unlock(&BTRFS_I(inode)->log_mutex);
3228 if (ret == -ENOSPC) {
3229 btrfs_set_log_full_commit(root->fs_info, trans);
3231 } else if (ret < 0 && ret != -ENOENT)
3232 btrfs_abort_transaction(trans, ret);
3233 btrfs_end_log_trans(root);
3239 * creates a range item in the log for 'dirid'. first_offset and
3240 * last_offset tell us which parts of the key space the log should
3241 * be considered authoritative for.
3243 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3244 struct btrfs_root *log,
3245 struct btrfs_path *path,
3246 int key_type, u64 dirid,
3247 u64 first_offset, u64 last_offset)
3250 struct btrfs_key key;
3251 struct btrfs_dir_log_item *item;
3253 key.objectid = dirid;
3254 key.offset = first_offset;
3255 if (key_type == BTRFS_DIR_ITEM_KEY)
3256 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3258 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3259 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3263 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3264 struct btrfs_dir_log_item);
3265 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3266 btrfs_mark_buffer_dirty(path->nodes[0]);
3267 btrfs_release_path(path);
3272 * log all the items included in the current transaction for a given
3273 * directory. This also creates the range items in the log tree required
3274 * to replay anything deleted before the fsync
3276 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3277 struct btrfs_root *root, struct inode *inode,
3278 struct btrfs_path *path,
3279 struct btrfs_path *dst_path, int key_type,
3280 struct btrfs_log_ctx *ctx,
3281 u64 min_offset, u64 *last_offset_ret)
3283 struct btrfs_key min_key;
3284 struct btrfs_root *log = root->log_root;
3285 struct extent_buffer *src;
3290 u64 first_offset = min_offset;
3291 u64 last_offset = (u64)-1;
3292 u64 ino = btrfs_ino(inode);
3294 log = root->log_root;
3296 min_key.objectid = ino;
3297 min_key.type = key_type;
3298 min_key.offset = min_offset;
3300 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3303 * we didn't find anything from this transaction, see if there
3304 * is anything at all
3306 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3307 min_key.objectid = ino;
3308 min_key.type = key_type;
3309 min_key.offset = (u64)-1;
3310 btrfs_release_path(path);
3311 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3313 btrfs_release_path(path);
3316 ret = btrfs_previous_item(root, path, ino, key_type);
3318 /* if ret == 0 there are items for this type,
3319 * create a range to tell us the last key of this type.
3320 * otherwise, there are no items in this directory after
3321 * *min_offset, and we create a range to indicate that.
3324 struct btrfs_key tmp;
3325 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3327 if (key_type == tmp.type)
3328 first_offset = max(min_offset, tmp.offset) + 1;
3333 /* go backward to find any previous key */
3334 ret = btrfs_previous_item(root, path, ino, key_type);
3336 struct btrfs_key tmp;
3337 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3338 if (key_type == tmp.type) {
3339 first_offset = tmp.offset;
3340 ret = overwrite_item(trans, log, dst_path,
3341 path->nodes[0], path->slots[0],
3349 btrfs_release_path(path);
3352 * Find the first key from this transaction again. See the note for
3353 * log_new_dir_dentries, if we're logging a directory recursively we
3354 * won't be holding its i_mutex, which means we can modify the directory
3355 * while we're logging it. If we remove an entry between our first
3356 * search and this search we'll not find the key again and can just
3360 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3365 * we have a block from this transaction, log every item in it
3366 * from our directory
3369 struct btrfs_key tmp;
3370 src = path->nodes[0];
3371 nritems = btrfs_header_nritems(src);
3372 for (i = path->slots[0]; i < nritems; i++) {
3373 struct btrfs_dir_item *di;
3375 btrfs_item_key_to_cpu(src, &min_key, i);
3377 if (min_key.objectid != ino || min_key.type != key_type)
3380 if (need_resched()) {
3381 btrfs_release_path(path);
3386 ret = overwrite_item(trans, log, dst_path, src, i,
3394 * We must make sure that when we log a directory entry,
3395 * the corresponding inode, after log replay, has a
3396 * matching link count. For example:
3402 * xfs_io -c "fsync" mydir
3404 * <mount fs and log replay>
3406 * Would result in a fsync log that when replayed, our
3407 * file inode would have a link count of 1, but we get
3408 * two directory entries pointing to the same inode.
3409 * After removing one of the names, it would not be
3410 * possible to remove the other name, which resulted
3411 * always in stale file handle errors, and would not
3412 * be possible to rmdir the parent directory, since
3413 * its i_size could never decrement to the value
3414 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3416 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3417 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3419 (btrfs_dir_transid(src, di) == trans->transid ||
3420 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3421 tmp.type != BTRFS_ROOT_ITEM_KEY)
3422 ctx->log_new_dentries = true;
3424 path->slots[0] = nritems;
3427 * look ahead to the next item and see if it is also
3428 * from this directory and from this transaction
3430 ret = btrfs_next_leaf(root, path);
3433 last_offset = (u64)-1;
3438 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3439 if (tmp.objectid != ino || tmp.type != key_type) {
3440 last_offset = (u64)-1;
3443 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3444 ret = overwrite_item(trans, log, dst_path,
3445 path->nodes[0], path->slots[0],
3450 last_offset = tmp.offset;
3455 btrfs_release_path(path);
3456 btrfs_release_path(dst_path);
3459 *last_offset_ret = last_offset;
3461 * insert the log range keys to indicate where the log
3464 ret = insert_dir_log_key(trans, log, path, key_type,
3465 ino, first_offset, last_offset);
3473 * logging directories is very similar to logging inodes, We find all the items
3474 * from the current transaction and write them to the log.
3476 * The recovery code scans the directory in the subvolume, and if it finds a
3477 * key in the range logged that is not present in the log tree, then it means
3478 * that dir entry was unlinked during the transaction.
3480 * In order for that scan to work, we must include one key smaller than
3481 * the smallest logged by this transaction and one key larger than the largest
3482 * key logged by this transaction.
3484 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3485 struct btrfs_root *root, struct inode *inode,
3486 struct btrfs_path *path,
3487 struct btrfs_path *dst_path,
3488 struct btrfs_log_ctx *ctx)
3493 int key_type = BTRFS_DIR_ITEM_KEY;
3499 ret = log_dir_items(trans, root, inode, path,
3500 dst_path, key_type, ctx, min_key,
3504 if (max_key == (u64)-1)
3506 min_key = max_key + 1;
3509 if (key_type == BTRFS_DIR_ITEM_KEY) {
3510 key_type = BTRFS_DIR_INDEX_KEY;
3517 * a helper function to drop items from the log before we relog an
3518 * inode. max_key_type indicates the highest item type to remove.
3519 * This cannot be run for file data extents because it does not
3520 * free the extents they point to.
3522 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3523 struct btrfs_root *log,
3524 struct btrfs_path *path,
3525 u64 objectid, int max_key_type)
3528 struct btrfs_key key;
3529 struct btrfs_key found_key;
3532 key.objectid = objectid;
3533 key.type = max_key_type;
3534 key.offset = (u64)-1;
3537 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3538 BUG_ON(ret == 0); /* Logic error */
3542 if (path->slots[0] == 0)
3546 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3549 if (found_key.objectid != objectid)
3552 found_key.offset = 0;
3554 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3557 ret = btrfs_del_items(trans, log, path, start_slot,
3558 path->slots[0] - start_slot + 1);
3560 * If start slot isn't 0 then we don't need to re-search, we've
3561 * found the last guy with the objectid in this tree.
3563 if (ret || start_slot != 0)
3565 btrfs_release_path(path);
3567 btrfs_release_path(path);
3573 static void fill_inode_item(struct btrfs_trans_handle *trans,
3574 struct extent_buffer *leaf,
3575 struct btrfs_inode_item *item,
3576 struct inode *inode, int log_inode_only,
3579 struct btrfs_map_token token;
3581 btrfs_init_map_token(&token);
3583 if (log_inode_only) {
3584 /* set the generation to zero so the recover code
3585 * can tell the difference between an logging
3586 * just to say 'this inode exists' and a logging
3587 * to say 'update this inode with these values'
3589 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3590 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3592 btrfs_set_token_inode_generation(leaf, item,
3593 BTRFS_I(inode)->generation,
3595 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3598 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3599 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3600 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3601 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3603 btrfs_set_token_timespec_sec(leaf, &item->atime,
3604 inode->i_atime.tv_sec, &token);
3605 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3606 inode->i_atime.tv_nsec, &token);
3608 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3609 inode->i_mtime.tv_sec, &token);
3610 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3611 inode->i_mtime.tv_nsec, &token);
3613 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3614 inode->i_ctime.tv_sec, &token);
3615 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3616 inode->i_ctime.tv_nsec, &token);
3618 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3621 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3622 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3623 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3624 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3625 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3628 static int log_inode_item(struct btrfs_trans_handle *trans,
3629 struct btrfs_root *log, struct btrfs_path *path,
3630 struct inode *inode)
3632 struct btrfs_inode_item *inode_item;
3635 ret = btrfs_insert_empty_item(trans, log, path,
3636 &BTRFS_I(inode)->location,
3637 sizeof(*inode_item));
3638 if (ret && ret != -EEXIST)
3640 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3641 struct btrfs_inode_item);
3642 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0);
3643 btrfs_release_path(path);
3647 static noinline int copy_items(struct btrfs_trans_handle *trans,
3648 struct inode *inode,
3649 struct btrfs_path *dst_path,
3650 struct btrfs_path *src_path, u64 *last_extent,
3651 int start_slot, int nr, int inode_only,
3654 unsigned long src_offset;
3655 unsigned long dst_offset;
3656 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3657 struct btrfs_file_extent_item *extent;
3658 struct btrfs_inode_item *inode_item;
3659 struct extent_buffer *src = src_path->nodes[0];
3660 struct btrfs_key first_key, last_key, key;
3662 struct btrfs_key *ins_keys;
3666 struct list_head ordered_sums;
3667 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3668 bool has_extents = false;
3669 bool need_find_last_extent = true;
3672 INIT_LIST_HEAD(&ordered_sums);
3674 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3675 nr * sizeof(u32), GFP_NOFS);
3679 first_key.objectid = (u64)-1;
3681 ins_sizes = (u32 *)ins_data;
3682 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3684 for (i = 0; i < nr; i++) {
3685 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3686 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3688 ret = btrfs_insert_empty_items(trans, log, dst_path,
3689 ins_keys, ins_sizes, nr);
3695 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3696 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3697 dst_path->slots[0]);
3699 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3702 last_key = ins_keys[i];
3704 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3705 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3707 struct btrfs_inode_item);
3708 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3709 inode, inode_only == LOG_INODE_EXISTS,
3712 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3713 src_offset, ins_sizes[i]);
3717 * We set need_find_last_extent here in case we know we were
3718 * processing other items and then walk into the first extent in
3719 * the inode. If we don't hit an extent then nothing changes,
3720 * we'll do the last search the next time around.
3722 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3724 if (first_key.objectid == (u64)-1)
3725 first_key = ins_keys[i];
3727 need_find_last_extent = false;
3730 /* take a reference on file data extents so that truncates
3731 * or deletes of this inode don't have to relog the inode
3734 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3737 extent = btrfs_item_ptr(src, start_slot + i,
3738 struct btrfs_file_extent_item);
3740 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3743 found_type = btrfs_file_extent_type(src, extent);
3744 if (found_type == BTRFS_FILE_EXTENT_REG) {
3746 ds = btrfs_file_extent_disk_bytenr(src,
3748 /* ds == 0 is a hole */
3752 dl = btrfs_file_extent_disk_num_bytes(src,
3754 cs = btrfs_file_extent_offset(src, extent);
3755 cl = btrfs_file_extent_num_bytes(src,
3757 if (btrfs_file_extent_compression(src,
3763 ret = btrfs_lookup_csums_range(
3764 log->fs_info->csum_root,
3765 ds + cs, ds + cs + cl - 1,
3773 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3774 btrfs_release_path(dst_path);
3778 * we have to do this after the loop above to avoid changing the
3779 * log tree while trying to change the log tree.
3781 while (!list_empty(&ordered_sums)) {
3782 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3783 struct btrfs_ordered_sum,
3786 ret = btrfs_csum_file_blocks(trans, log, sums);
3787 list_del(&sums->list);
3794 if (need_find_last_extent && *last_extent == first_key.offset) {
3796 * We don't have any leafs between our current one and the one
3797 * we processed before that can have file extent items for our
3798 * inode (and have a generation number smaller than our current
3801 need_find_last_extent = false;
3805 * Because we use btrfs_search_forward we could skip leaves that were
3806 * not modified and then assume *last_extent is valid when it really
3807 * isn't. So back up to the previous leaf and read the end of the last
3808 * extent before we go and fill in holes.
3810 if (need_find_last_extent) {
3813 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3818 if (src_path->slots[0])
3819 src_path->slots[0]--;
3820 src = src_path->nodes[0];
3821 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3822 if (key.objectid != btrfs_ino(inode) ||
3823 key.type != BTRFS_EXTENT_DATA_KEY)
3825 extent = btrfs_item_ptr(src, src_path->slots[0],
3826 struct btrfs_file_extent_item);
3827 if (btrfs_file_extent_type(src, extent) ==
3828 BTRFS_FILE_EXTENT_INLINE) {
3829 len = btrfs_file_extent_inline_len(src,
3832 *last_extent = ALIGN(key.offset + len,
3835 len = btrfs_file_extent_num_bytes(src, extent);
3836 *last_extent = key.offset + len;
3840 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3841 * things could have happened
3843 * 1) A merge could have happened, so we could currently be on a leaf
3844 * that holds what we were copying in the first place.
3845 * 2) A split could have happened, and now not all of the items we want
3846 * are on the same leaf.
3848 * So we need to adjust how we search for holes, we need to drop the
3849 * path and re-search for the first extent key we found, and then walk
3850 * forward until we hit the last one we copied.
3852 if (need_find_last_extent) {
3853 /* btrfs_prev_leaf could return 1 without releasing the path */
3854 btrfs_release_path(src_path);
3855 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3860 src = src_path->nodes[0];
3861 i = src_path->slots[0];
3867 * Ok so here we need to go through and fill in any holes we may have
3868 * to make sure that holes are punched for those areas in case they had
3869 * extents previously.
3875 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3876 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3880 src = src_path->nodes[0];
3882 need_find_last_extent = true;
3885 btrfs_item_key_to_cpu(src, &key, i);
3886 if (!btrfs_comp_cpu_keys(&key, &last_key))
3888 if (key.objectid != btrfs_ino(inode) ||
3889 key.type != BTRFS_EXTENT_DATA_KEY) {
3893 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3894 if (btrfs_file_extent_type(src, extent) ==
3895 BTRFS_FILE_EXTENT_INLINE) {
3896 len = btrfs_file_extent_inline_len(src, i, extent);
3897 extent_end = ALIGN(key.offset + len, log->sectorsize);
3899 len = btrfs_file_extent_num_bytes(src, extent);
3900 extent_end = key.offset + len;
3904 if (*last_extent == key.offset) {
3905 *last_extent = extent_end;
3908 offset = *last_extent;
3909 len = key.offset - *last_extent;
3910 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3911 offset, 0, 0, len, 0, len, 0,
3915 *last_extent = extent_end;
3918 * Need to let the callers know we dropped the path so they should
3921 if (!ret && need_find_last_extent)
3926 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3928 struct extent_map *em1, *em2;
3930 em1 = list_entry(a, struct extent_map, list);
3931 em2 = list_entry(b, struct extent_map, list);
3933 if (em1->start < em2->start)
3935 else if (em1->start > em2->start)
3940 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3941 struct inode *inode,
3942 struct btrfs_root *root,
3943 const struct extent_map *em,
3944 const struct list_head *logged_list,
3945 bool *ordered_io_error)
3947 struct btrfs_ordered_extent *ordered;
3948 struct btrfs_root *log = root->log_root;
3949 u64 mod_start = em->mod_start;
3950 u64 mod_len = em->mod_len;
3951 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3954 LIST_HEAD(ordered_sums);
3957 *ordered_io_error = false;
3959 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3960 em->block_start == EXTENT_MAP_HOLE)
3964 * Wait far any ordered extent that covers our extent map. If it
3965 * finishes without an error, first check and see if our csums are on
3966 * our outstanding ordered extents.
3968 list_for_each_entry(ordered, logged_list, log_list) {
3969 struct btrfs_ordered_sum *sum;
3974 if (ordered->file_offset + ordered->len <= mod_start ||
3975 mod_start + mod_len <= ordered->file_offset)
3978 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3979 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3980 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3981 const u64 start = ordered->file_offset;
3982 const u64 end = ordered->file_offset + ordered->len - 1;
3984 WARN_ON(ordered->inode != inode);
3985 filemap_fdatawrite_range(inode->i_mapping, start, end);
3988 wait_event(ordered->wait,
3989 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3990 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3992 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
3994 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3995 * i_mapping flags, so that the next fsync won't get
3996 * an outdated io error too.
3998 filemap_check_errors(inode->i_mapping);
3999 *ordered_io_error = true;
4003 * We are going to copy all the csums on this ordered extent, so
4004 * go ahead and adjust mod_start and mod_len in case this
4005 * ordered extent has already been logged.
4007 if (ordered->file_offset > mod_start) {
4008 if (ordered->file_offset + ordered->len >=
4009 mod_start + mod_len)
4010 mod_len = ordered->file_offset - mod_start;
4012 * If we have this case
4014 * |--------- logged extent ---------|
4015 * |----- ordered extent ----|
4017 * Just don't mess with mod_start and mod_len, we'll
4018 * just end up logging more csums than we need and it
4022 if (ordered->file_offset + ordered->len <
4023 mod_start + mod_len) {
4024 mod_len = (mod_start + mod_len) -
4025 (ordered->file_offset + ordered->len);
4026 mod_start = ordered->file_offset +
4037 * To keep us from looping for the above case of an ordered
4038 * extent that falls inside of the logged extent.
4040 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
4044 list_for_each_entry(sum, &ordered->list, list) {
4045 ret = btrfs_csum_file_blocks(trans, log, sum);
4051 if (*ordered_io_error || !mod_len || ret || skip_csum)
4054 if (em->compress_type) {
4056 csum_len = max(em->block_len, em->orig_block_len);
4058 csum_offset = mod_start - em->start;
4062 /* block start is already adjusted for the file extent offset. */
4063 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
4064 em->block_start + csum_offset,
4065 em->block_start + csum_offset +
4066 csum_len - 1, &ordered_sums, 0);
4070 while (!list_empty(&ordered_sums)) {
4071 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4072 struct btrfs_ordered_sum,
4075 ret = btrfs_csum_file_blocks(trans, log, sums);
4076 list_del(&sums->list);
4083 static int log_one_extent(struct btrfs_trans_handle *trans,
4084 struct inode *inode, struct btrfs_root *root,
4085 const struct extent_map *em,
4086 struct btrfs_path *path,
4087 const struct list_head *logged_list,
4088 struct btrfs_log_ctx *ctx)
4090 struct btrfs_root *log = root->log_root;
4091 struct btrfs_file_extent_item *fi;
4092 struct extent_buffer *leaf;
4093 struct btrfs_map_token token;
4094 struct btrfs_key key;
4095 u64 extent_offset = em->start - em->orig_start;
4098 int extent_inserted = 0;
4099 bool ordered_io_err = false;
4101 ret = wait_ordered_extents(trans, inode, root, em, logged_list,
4106 if (ordered_io_err) {
4111 btrfs_init_map_token(&token);
4113 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
4114 em->start + em->len, NULL, 0, 1,
4115 sizeof(*fi), &extent_inserted);
4119 if (!extent_inserted) {
4120 key.objectid = btrfs_ino(inode);
4121 key.type = BTRFS_EXTENT_DATA_KEY;
4122 key.offset = em->start;
4124 ret = btrfs_insert_empty_item(trans, log, path, &key,
4129 leaf = path->nodes[0];
4130 fi = btrfs_item_ptr(leaf, path->slots[0],
4131 struct btrfs_file_extent_item);
4133 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4135 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4136 btrfs_set_token_file_extent_type(leaf, fi,
4137 BTRFS_FILE_EXTENT_PREALLOC,
4140 btrfs_set_token_file_extent_type(leaf, fi,
4141 BTRFS_FILE_EXTENT_REG,
4144 block_len = max(em->block_len, em->orig_block_len);
4145 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4146 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4149 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4151 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4152 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4154 extent_offset, &token);
4155 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4158 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4159 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4163 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4164 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4165 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4166 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4168 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4169 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4170 btrfs_mark_buffer_dirty(leaf);
4172 btrfs_release_path(path);
4177 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4178 struct btrfs_root *root,
4179 struct inode *inode,
4180 struct btrfs_path *path,
4181 struct list_head *logged_list,
4182 struct btrfs_log_ctx *ctx,
4186 struct extent_map *em, *n;
4187 struct list_head extents;
4188 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
4193 INIT_LIST_HEAD(&extents);
4195 down_write(&BTRFS_I(inode)->dio_sem);
4196 write_lock(&tree->lock);
4197 test_gen = root->fs_info->last_trans_committed;
4199 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4200 list_del_init(&em->list);
4203 * Just an arbitrary number, this can be really CPU intensive
4204 * once we start getting a lot of extents, and really once we
4205 * have a bunch of extents we just want to commit since it will
4208 if (++num > 32768) {
4209 list_del_init(&tree->modified_extents);
4214 if (em->generation <= test_gen)
4216 /* Need a ref to keep it from getting evicted from cache */
4217 atomic_inc(&em->refs);
4218 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4219 list_add_tail(&em->list, &extents);
4223 list_sort(NULL, &extents, extent_cmp);
4224 btrfs_get_logged_extents(inode, logged_list, start, end);
4226 * Some ordered extents started by fsync might have completed
4227 * before we could collect them into the list logged_list, which
4228 * means they're gone, not in our logged_list nor in the inode's
4229 * ordered tree. We want the application/user space to know an
4230 * error happened while attempting to persist file data so that
4231 * it can take proper action. If such error happened, we leave
4232 * without writing to the log tree and the fsync must report the
4233 * file data write error and not commit the current transaction.
4235 ret = filemap_check_errors(inode->i_mapping);
4239 while (!list_empty(&extents)) {
4240 em = list_entry(extents.next, struct extent_map, list);
4242 list_del_init(&em->list);
4245 * If we had an error we just need to delete everybody from our
4249 clear_em_logging(tree, em);
4250 free_extent_map(em);
4254 write_unlock(&tree->lock);
4256 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4258 write_lock(&tree->lock);
4259 clear_em_logging(tree, em);
4260 free_extent_map(em);
4262 WARN_ON(!list_empty(&extents));
4263 write_unlock(&tree->lock);
4264 up_write(&BTRFS_I(inode)->dio_sem);
4266 btrfs_release_path(path);
4270 static int logged_inode_size(struct btrfs_root *log, struct inode *inode,
4271 struct btrfs_path *path, u64 *size_ret)
4273 struct btrfs_key key;
4276 key.objectid = btrfs_ino(inode);
4277 key.type = BTRFS_INODE_ITEM_KEY;
4280 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4283 } else if (ret > 0) {
4286 struct btrfs_inode_item *item;
4288 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4289 struct btrfs_inode_item);
4290 *size_ret = btrfs_inode_size(path->nodes[0], item);
4293 btrfs_release_path(path);
4298 * At the moment we always log all xattrs. This is to figure out at log replay
4299 * time which xattrs must have their deletion replayed. If a xattr is missing
4300 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4301 * because if a xattr is deleted, the inode is fsynced and a power failure
4302 * happens, causing the log to be replayed the next time the fs is mounted,
4303 * we want the xattr to not exist anymore (same behaviour as other filesystems
4304 * with a journal, ext3/4, xfs, f2fs, etc).
4306 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4307 struct btrfs_root *root,
4308 struct inode *inode,
4309 struct btrfs_path *path,
4310 struct btrfs_path *dst_path)
4313 struct btrfs_key key;
4314 const u64 ino = btrfs_ino(inode);
4319 key.type = BTRFS_XATTR_ITEM_KEY;
4322 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4327 int slot = path->slots[0];
4328 struct extent_buffer *leaf = path->nodes[0];
4329 int nritems = btrfs_header_nritems(leaf);
4331 if (slot >= nritems) {
4333 u64 last_extent = 0;
4335 ret = copy_items(trans, inode, dst_path, path,
4336 &last_extent, start_slot,
4338 /* can't be 1, extent items aren't processed */
4344 ret = btrfs_next_leaf(root, path);
4352 btrfs_item_key_to_cpu(leaf, &key, slot);
4353 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4363 u64 last_extent = 0;
4365 ret = copy_items(trans, inode, dst_path, path,
4366 &last_extent, start_slot,
4368 /* can't be 1, extent items aren't processed */
4378 * If the no holes feature is enabled we need to make sure any hole between the
4379 * last extent and the i_size of our inode is explicitly marked in the log. This
4380 * is to make sure that doing something like:
4382 * 1) create file with 128Kb of data
4383 * 2) truncate file to 64Kb
4384 * 3) truncate file to 256Kb
4386 * 5) <crash/power failure>
4387 * 6) mount fs and trigger log replay
4389 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4390 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4391 * file correspond to a hole. The presence of explicit holes in a log tree is
4392 * what guarantees that log replay will remove/adjust file extent items in the
4395 * Here we do not need to care about holes between extents, that is already done
4396 * by copy_items(). We also only need to do this in the full sync path, where we
4397 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4398 * lookup the list of modified extent maps and if any represents a hole, we
4399 * insert a corresponding extent representing a hole in the log tree.
4401 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4402 struct btrfs_root *root,
4403 struct inode *inode,
4404 struct btrfs_path *path)
4407 struct btrfs_key key;
4410 struct extent_buffer *leaf;
4411 struct btrfs_root *log = root->log_root;
4412 const u64 ino = btrfs_ino(inode);
4413 const u64 i_size = i_size_read(inode);
4415 if (!btrfs_fs_incompat(root->fs_info, NO_HOLES))
4419 key.type = BTRFS_EXTENT_DATA_KEY;
4420 key.offset = (u64)-1;
4422 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4427 ASSERT(path->slots[0] > 0);
4429 leaf = path->nodes[0];
4430 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4432 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4433 /* inode does not have any extents */
4437 struct btrfs_file_extent_item *extent;
4441 * If there's an extent beyond i_size, an explicit hole was
4442 * already inserted by copy_items().
4444 if (key.offset >= i_size)
4447 extent = btrfs_item_ptr(leaf, path->slots[0],
4448 struct btrfs_file_extent_item);
4450 if (btrfs_file_extent_type(leaf, extent) ==
4451 BTRFS_FILE_EXTENT_INLINE)
4454 len = btrfs_file_extent_num_bytes(leaf, extent);
4455 /* Last extent goes beyond i_size, no need to log a hole. */
4456 if (key.offset + len > i_size)
4458 hole_start = key.offset + len;
4459 hole_size = i_size - hole_start;
4461 btrfs_release_path(path);
4463 /* Last extent ends at i_size. */
4467 hole_size = ALIGN(hole_size, root->sectorsize);
4468 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4469 hole_size, 0, hole_size, 0, 0, 0);
4474 * When we are logging a new inode X, check if it doesn't have a reference that
4475 * matches the reference from some other inode Y created in a past transaction
4476 * and that was renamed in the current transaction. If we don't do this, then at
4477 * log replay time we can lose inode Y (and all its files if it's a directory):
4480 * echo "hello world" > /mnt/x/foobar
4483 * mkdir /mnt/x # or touch /mnt/x
4484 * xfs_io -c fsync /mnt/x
4486 * mount fs, trigger log replay
4488 * After the log replay procedure, we would lose the first directory and all its
4489 * files (file foobar).
4490 * For the case where inode Y is not a directory we simply end up losing it:
4492 * echo "123" > /mnt/foo
4494 * mv /mnt/foo /mnt/bar
4495 * echo "abc" > /mnt/foo
4496 * xfs_io -c fsync /mnt/foo
4499 * We also need this for cases where a snapshot entry is replaced by some other
4500 * entry (file or directory) otherwise we end up with an unreplayable log due to
4501 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4502 * if it were a regular entry:
4505 * btrfs subvolume snapshot /mnt /mnt/x/snap
4506 * btrfs subvolume delete /mnt/x/snap
4509 * fsync /mnt/x or fsync some new file inside it
4512 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4513 * the same transaction.
4515 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4517 const struct btrfs_key *key,
4518 struct inode *inode,
4522 struct btrfs_path *search_path;
4525 u32 item_size = btrfs_item_size_nr(eb, slot);
4527 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4529 search_path = btrfs_alloc_path();
4532 search_path->search_commit_root = 1;
4533 search_path->skip_locking = 1;
4535 while (cur_offset < item_size) {
4539 unsigned long name_ptr;
4540 struct btrfs_dir_item *di;
4542 if (key->type == BTRFS_INODE_REF_KEY) {
4543 struct btrfs_inode_ref *iref;
4545 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4546 parent = key->offset;
4547 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4548 name_ptr = (unsigned long)(iref + 1);
4549 this_len = sizeof(*iref) + this_name_len;
4551 struct btrfs_inode_extref *extref;
4553 extref = (struct btrfs_inode_extref *)(ptr +
4555 parent = btrfs_inode_extref_parent(eb, extref);
4556 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4557 name_ptr = (unsigned long)&extref->name;
4558 this_len = sizeof(*extref) + this_name_len;
4561 if (this_name_len > name_len) {
4564 new_name = krealloc(name, this_name_len, GFP_NOFS);
4569 name_len = this_name_len;
4573 read_extent_buffer(eb, name, name_ptr, this_name_len);
4574 di = btrfs_lookup_dir_item(NULL, BTRFS_I(inode)->root,
4575 search_path, parent,
4576 name, this_name_len, 0);
4577 if (di && !IS_ERR(di)) {
4578 struct btrfs_key di_key;
4580 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4582 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4584 *other_ino = di_key.objectid;
4589 } else if (IS_ERR(di)) {
4593 btrfs_release_path(search_path);
4595 cur_offset += this_len;
4599 btrfs_free_path(search_path);
4604 /* log a single inode in the tree log.
4605 * At least one parent directory for this inode must exist in the tree
4606 * or be logged already.
4608 * Any items from this inode changed by the current transaction are copied
4609 * to the log tree. An extra reference is taken on any extents in this
4610 * file, allowing us to avoid a whole pile of corner cases around logging
4611 * blocks that have been removed from the tree.
4613 * See LOG_INODE_ALL and related defines for a description of what inode_only
4616 * This handles both files and directories.
4618 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4619 struct btrfs_root *root, struct inode *inode,
4623 struct btrfs_log_ctx *ctx)
4625 struct btrfs_path *path;
4626 struct btrfs_path *dst_path;
4627 struct btrfs_key min_key;
4628 struct btrfs_key max_key;
4629 struct btrfs_root *log = root->log_root;
4630 struct extent_buffer *src = NULL;
4631 LIST_HEAD(logged_list);
4632 u64 last_extent = 0;
4636 int ins_start_slot = 0;
4638 bool fast_search = false;
4639 u64 ino = btrfs_ino(inode);
4640 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4641 u64 logged_isize = 0;
4642 bool need_log_inode_item = true;
4643 bool xattrs_logged = false;
4645 path = btrfs_alloc_path();
4648 dst_path = btrfs_alloc_path();
4650 btrfs_free_path(path);
4654 min_key.objectid = ino;
4655 min_key.type = BTRFS_INODE_ITEM_KEY;
4658 max_key.objectid = ino;
4661 /* today the code can only do partial logging of directories */
4662 if (S_ISDIR(inode->i_mode) ||
4663 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4664 &BTRFS_I(inode)->runtime_flags) &&
4665 inode_only >= LOG_INODE_EXISTS))
4666 max_key.type = BTRFS_XATTR_ITEM_KEY;
4668 max_key.type = (u8)-1;
4669 max_key.offset = (u64)-1;
4672 * Only run delayed items if we are a dir or a new file.
4673 * Otherwise commit the delayed inode only, which is needed in
4674 * order for the log replay code to mark inodes for link count
4675 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4677 if (S_ISDIR(inode->i_mode) ||
4678 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed)
4679 ret = btrfs_commit_inode_delayed_items(trans, inode);
4681 ret = btrfs_commit_inode_delayed_inode(inode);
4684 btrfs_free_path(path);
4685 btrfs_free_path(dst_path);
4689 if (inode_only == LOG_OTHER_INODE) {
4690 inode_only = LOG_INODE_EXISTS;
4691 mutex_lock_nested(&BTRFS_I(inode)->log_mutex,
4692 SINGLE_DEPTH_NESTING);
4694 mutex_lock(&BTRFS_I(inode)->log_mutex);
4698 * a brute force approach to making sure we get the most uptodate
4699 * copies of everything.
4701 if (S_ISDIR(inode->i_mode)) {
4702 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4704 if (inode_only == LOG_INODE_EXISTS)
4705 max_key_type = BTRFS_XATTR_ITEM_KEY;
4706 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
4708 if (inode_only == LOG_INODE_EXISTS) {
4710 * Make sure the new inode item we write to the log has
4711 * the same isize as the current one (if it exists).
4712 * This is necessary to prevent data loss after log
4713 * replay, and also to prevent doing a wrong expanding
4714 * truncate - for e.g. create file, write 4K into offset
4715 * 0, fsync, write 4K into offset 4096, add hard link,
4716 * fsync some other file (to sync log), power fail - if
4717 * we use the inode's current i_size, after log replay
4718 * we get a 8Kb file, with the last 4Kb extent as a hole
4719 * (zeroes), as if an expanding truncate happened,
4720 * instead of getting a file of 4Kb only.
4722 err = logged_inode_size(log, inode, path,
4727 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4728 &BTRFS_I(inode)->runtime_flags)) {
4729 if (inode_only == LOG_INODE_EXISTS) {
4730 max_key.type = BTRFS_XATTR_ITEM_KEY;
4731 ret = drop_objectid_items(trans, log, path, ino,
4734 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4735 &BTRFS_I(inode)->runtime_flags);
4736 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4737 &BTRFS_I(inode)->runtime_flags);
4739 ret = btrfs_truncate_inode_items(trans,
4745 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4746 &BTRFS_I(inode)->runtime_flags) ||
4747 inode_only == LOG_INODE_EXISTS) {
4748 if (inode_only == LOG_INODE_ALL)
4750 max_key.type = BTRFS_XATTR_ITEM_KEY;
4751 ret = drop_objectid_items(trans, log, path, ino,
4754 if (inode_only == LOG_INODE_ALL)
4767 ret = btrfs_search_forward(root, &min_key,
4768 path, trans->transid);
4776 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4777 if (min_key.objectid != ino)
4779 if (min_key.type > max_key.type)
4782 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4783 need_log_inode_item = false;
4785 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4786 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4787 BTRFS_I(inode)->generation == trans->transid) {
4790 ret = btrfs_check_ref_name_override(path->nodes[0],
4797 } else if (ret > 0 && ctx &&
4798 other_ino != btrfs_ino(ctx->inode)) {
4799 struct btrfs_key inode_key;
4800 struct inode *other_inode;
4806 ins_start_slot = path->slots[0];
4808 ret = copy_items(trans, inode, dst_path, path,
4809 &last_extent, ins_start_slot,
4817 btrfs_release_path(path);
4818 inode_key.objectid = other_ino;
4819 inode_key.type = BTRFS_INODE_ITEM_KEY;
4820 inode_key.offset = 0;
4821 other_inode = btrfs_iget(root->fs_info->sb,
4825 * If the other inode that had a conflicting dir
4826 * entry was deleted in the current transaction,
4827 * we don't need to do more work nor fallback to
4828 * a transaction commit.
4830 if (IS_ERR(other_inode) &&
4831 PTR_ERR(other_inode) == -ENOENT) {
4833 } else if (IS_ERR(other_inode)) {
4834 err = PTR_ERR(other_inode);
4838 * We are safe logging the other inode without
4839 * acquiring its i_mutex as long as we log with
4840 * the LOG_INODE_EXISTS mode. We're safe against
4841 * concurrent renames of the other inode as well
4842 * because during a rename we pin the log and
4843 * update the log with the new name before we
4846 err = btrfs_log_inode(trans, root, other_inode,
4849 btrfs_add_delayed_iput(other_inode);
4857 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4858 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4861 ret = copy_items(trans, inode, dst_path, path,
4862 &last_extent, ins_start_slot,
4863 ins_nr, inode_only, logged_isize);
4870 btrfs_release_path(path);
4876 src = path->nodes[0];
4877 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4880 } else if (!ins_nr) {
4881 ins_start_slot = path->slots[0];
4886 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4887 ins_start_slot, ins_nr, inode_only,
4895 btrfs_release_path(path);
4899 ins_start_slot = path->slots[0];
4902 nritems = btrfs_header_nritems(path->nodes[0]);
4904 if (path->slots[0] < nritems) {
4905 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4910 ret = copy_items(trans, inode, dst_path, path,
4911 &last_extent, ins_start_slot,
4912 ins_nr, inode_only, logged_isize);
4920 btrfs_release_path(path);
4922 if (min_key.offset < (u64)-1) {
4924 } else if (min_key.type < max_key.type) {
4932 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4933 ins_start_slot, ins_nr, inode_only,
4943 btrfs_release_path(path);
4944 btrfs_release_path(dst_path);
4945 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
4948 xattrs_logged = true;
4949 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4950 btrfs_release_path(path);
4951 btrfs_release_path(dst_path);
4952 err = btrfs_log_trailing_hole(trans, root, inode, path);
4957 btrfs_release_path(path);
4958 btrfs_release_path(dst_path);
4959 if (need_log_inode_item) {
4960 err = log_inode_item(trans, log, dst_path, inode);
4961 if (!err && !xattrs_logged) {
4962 err = btrfs_log_all_xattrs(trans, root, inode, path,
4964 btrfs_release_path(path);
4970 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
4971 &logged_list, ctx, start, end);
4976 } else if (inode_only == LOG_INODE_ALL) {
4977 struct extent_map *em, *n;
4979 write_lock(&em_tree->lock);
4981 * We can't just remove every em if we're called for a ranged
4982 * fsync - that is, one that doesn't cover the whole possible
4983 * file range (0 to LLONG_MAX). This is because we can have
4984 * em's that fall outside the range we're logging and therefore
4985 * their ordered operations haven't completed yet
4986 * (btrfs_finish_ordered_io() not invoked yet). This means we
4987 * didn't get their respective file extent item in the fs/subvol
4988 * tree yet, and need to let the next fast fsync (one which
4989 * consults the list of modified extent maps) find the em so
4990 * that it logs a matching file extent item and waits for the
4991 * respective ordered operation to complete (if it's still
4994 * Removing every em outside the range we're logging would make
4995 * the next fast fsync not log their matching file extent items,
4996 * therefore making us lose data after a log replay.
4998 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5000 const u64 mod_end = em->mod_start + em->mod_len - 1;
5002 if (em->mod_start >= start && mod_end <= end)
5003 list_del_init(&em->list);
5005 write_unlock(&em_tree->lock);
5008 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
5009 ret = log_directory_changes(trans, root, inode, path, dst_path,
5017 spin_lock(&BTRFS_I(inode)->lock);
5018 BTRFS_I(inode)->logged_trans = trans->transid;
5019 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
5020 spin_unlock(&BTRFS_I(inode)->lock);
5023 btrfs_put_logged_extents(&logged_list);
5025 btrfs_submit_logged_extents(&logged_list, log);
5026 mutex_unlock(&BTRFS_I(inode)->log_mutex);
5028 btrfs_free_path(path);
5029 btrfs_free_path(dst_path);
5034 * Check if we must fallback to a transaction commit when logging an inode.
5035 * This must be called after logging the inode and is used only in the context
5036 * when fsyncing an inode requires the need to log some other inode - in which
5037 * case we can't lock the i_mutex of each other inode we need to log as that
5038 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5039 * log inodes up or down in the hierarchy) or rename operations for example. So
5040 * we take the log_mutex of the inode after we have logged it and then check for
5041 * its last_unlink_trans value - this is safe because any task setting
5042 * last_unlink_trans must take the log_mutex and it must do this before it does
5043 * the actual unlink operation, so if we do this check before a concurrent task
5044 * sets last_unlink_trans it means we've logged a consistent version/state of
5045 * all the inode items, otherwise we are not sure and must do a transaction
5046 * commit (the concurrent task might have only updated last_unlink_trans before
5047 * we logged the inode or it might have also done the unlink).
5049 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5050 struct inode *inode)
5052 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
5055 mutex_lock(&BTRFS_I(inode)->log_mutex);
5056 if (BTRFS_I(inode)->last_unlink_trans > fs_info->last_trans_committed) {
5058 * Make sure any commits to the log are forced to be full
5061 btrfs_set_log_full_commit(fs_info, trans);
5064 mutex_unlock(&BTRFS_I(inode)->log_mutex);
5070 * follow the dentry parent pointers up the chain and see if any
5071 * of the directories in it require a full commit before they can
5072 * be logged. Returns zero if nothing special needs to be done or 1 if
5073 * a full commit is required.
5075 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5076 struct inode *inode,
5077 struct dentry *parent,
5078 struct super_block *sb,
5082 struct dentry *old_parent = NULL;
5083 struct inode *orig_inode = inode;
5086 * for regular files, if its inode is already on disk, we don't
5087 * have to worry about the parents at all. This is because
5088 * we can use the last_unlink_trans field to record renames
5089 * and other fun in this file.
5091 if (S_ISREG(inode->i_mode) &&
5092 BTRFS_I(inode)->generation <= last_committed &&
5093 BTRFS_I(inode)->last_unlink_trans <= last_committed)
5096 if (!S_ISDIR(inode->i_mode)) {
5097 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5099 inode = d_inode(parent);
5104 * If we are logging a directory then we start with our inode,
5105 * not our parent's inode, so we need to skip setting the
5106 * logged_trans so that further down in the log code we don't
5107 * think this inode has already been logged.
5109 if (inode != orig_inode)
5110 BTRFS_I(inode)->logged_trans = trans->transid;
5113 if (btrfs_must_commit_transaction(trans, inode)) {
5118 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5121 if (IS_ROOT(parent)) {
5122 inode = d_inode(parent);
5123 if (btrfs_must_commit_transaction(trans, inode))
5128 parent = dget_parent(parent);
5130 old_parent = parent;
5131 inode = d_inode(parent);
5139 struct btrfs_dir_list {
5141 struct list_head list;
5145 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5146 * details about the why it is needed.
5147 * This is a recursive operation - if an existing dentry corresponds to a
5148 * directory, that directory's new entries are logged too (same behaviour as
5149 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5150 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5151 * complains about the following circular lock dependency / possible deadlock:
5155 * lock(&type->i_mutex_dir_key#3/2);
5156 * lock(sb_internal#2);
5157 * lock(&type->i_mutex_dir_key#3/2);
5158 * lock(&sb->s_type->i_mutex_key#14);
5160 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5161 * sb_start_intwrite() in btrfs_start_transaction().
5162 * Not locking i_mutex of the inodes is still safe because:
5164 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5165 * that while logging the inode new references (names) are added or removed
5166 * from the inode, leaving the logged inode item with a link count that does
5167 * not match the number of logged inode reference items. This is fine because
5168 * at log replay time we compute the real number of links and correct the
5169 * link count in the inode item (see replay_one_buffer() and
5170 * link_to_fixup_dir());
5172 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5173 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5174 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5175 * has a size that doesn't match the sum of the lengths of all the logged
5176 * names. This does not result in a problem because if a dir_item key is
5177 * logged but its matching dir_index key is not logged, at log replay time we
5178 * don't use it to replay the respective name (see replay_one_name()). On the
5179 * other hand if only the dir_index key ends up being logged, the respective
5180 * name is added to the fs/subvol tree with both the dir_item and dir_index
5181 * keys created (see replay_one_name()).
5182 * The directory's inode item with a wrong i_size is not a problem as well,
5183 * since we don't use it at log replay time to set the i_size in the inode
5184 * item of the fs/subvol tree (see overwrite_item()).
5186 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5187 struct btrfs_root *root,
5188 struct inode *start_inode,
5189 struct btrfs_log_ctx *ctx)
5191 struct btrfs_root *log = root->log_root;
5192 struct btrfs_path *path;
5193 LIST_HEAD(dir_list);
5194 struct btrfs_dir_list *dir_elem;
5197 path = btrfs_alloc_path();
5201 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5203 btrfs_free_path(path);
5206 dir_elem->ino = btrfs_ino(start_inode);
5207 list_add_tail(&dir_elem->list, &dir_list);
5209 while (!list_empty(&dir_list)) {
5210 struct extent_buffer *leaf;
5211 struct btrfs_key min_key;
5215 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5218 goto next_dir_inode;
5220 min_key.objectid = dir_elem->ino;
5221 min_key.type = BTRFS_DIR_ITEM_KEY;
5224 btrfs_release_path(path);
5225 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5227 goto next_dir_inode;
5228 } else if (ret > 0) {
5230 goto next_dir_inode;
5234 leaf = path->nodes[0];
5235 nritems = btrfs_header_nritems(leaf);
5236 for (i = path->slots[0]; i < nritems; i++) {
5237 struct btrfs_dir_item *di;
5238 struct btrfs_key di_key;
5239 struct inode *di_inode;
5240 struct btrfs_dir_list *new_dir_elem;
5241 int log_mode = LOG_INODE_EXISTS;
5244 btrfs_item_key_to_cpu(leaf, &min_key, i);
5245 if (min_key.objectid != dir_elem->ino ||
5246 min_key.type != BTRFS_DIR_ITEM_KEY)
5247 goto next_dir_inode;
5249 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5250 type = btrfs_dir_type(leaf, di);
5251 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5252 type != BTRFS_FT_DIR)
5254 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5255 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5258 btrfs_release_path(path);
5259 di_inode = btrfs_iget(root->fs_info->sb, &di_key,
5261 if (IS_ERR(di_inode)) {
5262 ret = PTR_ERR(di_inode);
5263 goto next_dir_inode;
5266 if (btrfs_inode_in_log(di_inode, trans->transid)) {
5267 btrfs_add_delayed_iput(di_inode);
5271 ctx->log_new_dentries = false;
5272 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5273 log_mode = LOG_INODE_ALL;
5274 ret = btrfs_log_inode(trans, root, di_inode,
5275 log_mode, 0, LLONG_MAX, ctx);
5277 btrfs_must_commit_transaction(trans, di_inode))
5279 btrfs_add_delayed_iput(di_inode);
5281 goto next_dir_inode;
5282 if (ctx->log_new_dentries) {
5283 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5285 if (!new_dir_elem) {
5287 goto next_dir_inode;
5289 new_dir_elem->ino = di_key.objectid;
5290 list_add_tail(&new_dir_elem->list, &dir_list);
5295 ret = btrfs_next_leaf(log, path);
5297 goto next_dir_inode;
5298 } else if (ret > 0) {
5300 goto next_dir_inode;
5304 if (min_key.offset < (u64)-1) {
5309 list_del(&dir_elem->list);
5313 btrfs_free_path(path);
5317 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5318 struct inode *inode,
5319 struct btrfs_log_ctx *ctx)
5322 struct btrfs_path *path;
5323 struct btrfs_key key;
5324 struct btrfs_root *root = BTRFS_I(inode)->root;
5325 const u64 ino = btrfs_ino(inode);
5327 path = btrfs_alloc_path();
5330 path->skip_locking = 1;
5331 path->search_commit_root = 1;
5334 key.type = BTRFS_INODE_REF_KEY;
5336 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5341 struct extent_buffer *leaf = path->nodes[0];
5342 int slot = path->slots[0];
5347 if (slot >= btrfs_header_nritems(leaf)) {
5348 ret = btrfs_next_leaf(root, path);
5356 btrfs_item_key_to_cpu(leaf, &key, slot);
5357 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5358 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5361 item_size = btrfs_item_size_nr(leaf, slot);
5362 ptr = btrfs_item_ptr_offset(leaf, slot);
5363 while (cur_offset < item_size) {
5364 struct btrfs_key inode_key;
5365 struct inode *dir_inode;
5367 inode_key.type = BTRFS_INODE_ITEM_KEY;
5368 inode_key.offset = 0;
5370 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5371 struct btrfs_inode_extref *extref;
5373 extref = (struct btrfs_inode_extref *)
5375 inode_key.objectid = btrfs_inode_extref_parent(
5377 cur_offset += sizeof(*extref);
5378 cur_offset += btrfs_inode_extref_name_len(leaf,
5381 inode_key.objectid = key.offset;
5382 cur_offset = item_size;
5385 dir_inode = btrfs_iget(root->fs_info->sb, &inode_key,
5388 * If the parent inode was deleted, return an error to
5389 * fallback to a transaction commit. This is to prevent
5390 * getting an inode that was moved from one parent A to
5391 * a parent B, got its former parent A deleted and then
5392 * it got fsync'ed, from existing at both parents after
5393 * a log replay (and the old parent still existing).
5400 * mv /mnt/B/bar /mnt/A/bar
5401 * mv -T /mnt/A /mnt/B
5405 * If we ignore the old parent B which got deleted,
5406 * after a log replay we would have file bar linked
5407 * at both parents and the old parent B would still
5410 if (IS_ERR(dir_inode)) {
5411 ret = PTR_ERR(dir_inode);
5416 ctx->log_new_dentries = false;
5417 ret = btrfs_log_inode(trans, root, dir_inode,
5418 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5420 btrfs_must_commit_transaction(trans, dir_inode))
5422 if (!ret && ctx && ctx->log_new_dentries)
5423 ret = log_new_dir_dentries(trans, root,
5425 btrfs_add_delayed_iput(dir_inode);
5433 btrfs_free_path(path);
5438 * helper function around btrfs_log_inode to make sure newly created
5439 * parent directories also end up in the log. A minimal inode and backref
5440 * only logging is done of any parent directories that are older than
5441 * the last committed transaction
5443 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5444 struct btrfs_root *root, struct inode *inode,
5445 struct dentry *parent,
5449 struct btrfs_log_ctx *ctx)
5451 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
5452 struct super_block *sb;
5453 struct dentry *old_parent = NULL;
5455 u64 last_committed = root->fs_info->last_trans_committed;
5456 bool log_dentries = false;
5457 struct inode *orig_inode = inode;
5461 if (btrfs_test_opt(root->fs_info, NOTREELOG)) {
5467 * The prev transaction commit doesn't complete, we need do
5468 * full commit by ourselves.
5470 if (root->fs_info->last_trans_log_full_commit >
5471 root->fs_info->last_trans_committed) {
5476 if (root != BTRFS_I(inode)->root ||
5477 btrfs_root_refs(&root->root_item) == 0) {
5482 ret = check_parent_dirs_for_sync(trans, inode, parent,
5483 sb, last_committed);
5487 if (btrfs_inode_in_log(inode, trans->transid)) {
5488 ret = BTRFS_NO_LOG_SYNC;
5492 ret = start_log_trans(trans, root, ctx);
5496 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5501 * for regular files, if its inode is already on disk, we don't
5502 * have to worry about the parents at all. This is because
5503 * we can use the last_unlink_trans field to record renames
5504 * and other fun in this file.
5506 if (S_ISREG(inode->i_mode) &&
5507 BTRFS_I(inode)->generation <= last_committed &&
5508 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
5513 if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries)
5514 log_dentries = true;
5517 * On unlink we must make sure all our current and old parent directory
5518 * inodes are fully logged. This is to prevent leaving dangling
5519 * directory index entries in directories that were our parents but are
5520 * not anymore. Not doing this results in old parent directory being
5521 * impossible to delete after log replay (rmdir will always fail with
5522 * error -ENOTEMPTY).
5528 * ln testdir/foo testdir/bar
5530 * unlink testdir/bar
5531 * xfs_io -c fsync testdir/foo
5533 * mount fs, triggers log replay
5535 * If we don't log the parent directory (testdir), after log replay the
5536 * directory still has an entry pointing to the file inode using the bar
5537 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5538 * the file inode has a link count of 1.
5544 * ln foo testdir/foo2
5545 * ln foo testdir/foo3
5547 * unlink testdir/foo3
5548 * xfs_io -c fsync foo
5550 * mount fs, triggers log replay
5552 * Similar as the first example, after log replay the parent directory
5553 * testdir still has an entry pointing to the inode file with name foo3
5554 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5555 * and has a link count of 2.
5557 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
5558 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5564 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5567 inode = d_inode(parent);
5568 if (root != BTRFS_I(inode)->root)
5571 if (BTRFS_I(inode)->generation > last_committed) {
5572 ret = btrfs_log_inode(trans, root, inode,
5578 if (IS_ROOT(parent))
5581 parent = dget_parent(parent);
5583 old_parent = parent;
5586 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5592 btrfs_set_log_full_commit(root->fs_info, trans);
5597 btrfs_remove_log_ctx(root, ctx);
5598 btrfs_end_log_trans(root);
5604 * it is not safe to log dentry if the chunk root has added new
5605 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5606 * If this returns 1, you must commit the transaction to safely get your
5609 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
5610 struct btrfs_root *root, struct dentry *dentry,
5613 struct btrfs_log_ctx *ctx)
5615 struct dentry *parent = dget_parent(dentry);
5618 ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent,
5619 start, end, 0, ctx);
5626 * should be called during mount to recover any replay any log trees
5629 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5632 struct btrfs_path *path;
5633 struct btrfs_trans_handle *trans;
5634 struct btrfs_key key;
5635 struct btrfs_key found_key;
5636 struct btrfs_key tmp_key;
5637 struct btrfs_root *log;
5638 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5639 struct walk_control wc = {
5640 .process_func = process_one_buffer,
5644 path = btrfs_alloc_path();
5648 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5650 trans = btrfs_start_transaction(fs_info->tree_root, 0);
5651 if (IS_ERR(trans)) {
5652 ret = PTR_ERR(trans);
5659 ret = walk_log_tree(trans, log_root_tree, &wc);
5661 btrfs_handle_fs_error(fs_info, ret,
5662 "Failed to pin buffers while recovering log root tree.");
5667 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5668 key.offset = (u64)-1;
5669 key.type = BTRFS_ROOT_ITEM_KEY;
5672 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
5675 btrfs_handle_fs_error(fs_info, ret,
5676 "Couldn't find tree log root.");
5680 if (path->slots[0] == 0)
5684 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5686 btrfs_release_path(path);
5687 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5690 log = btrfs_read_fs_root(log_root_tree, &found_key);
5693 btrfs_handle_fs_error(fs_info, ret,
5694 "Couldn't read tree log root.");
5698 tmp_key.objectid = found_key.offset;
5699 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5700 tmp_key.offset = (u64)-1;
5702 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
5703 if (IS_ERR(wc.replay_dest)) {
5704 ret = PTR_ERR(wc.replay_dest);
5707 * We didn't find the subvol, likely because it was
5708 * deleted. This is ok, simply skip this log and go to
5711 * We need to exclude the root because we can't have
5712 * other log replays overwriting this log as we'll read
5713 * it back in a few more times. This will keep our
5714 * block from being modified, and we'll just bail for
5715 * each subsequent pass.
5718 ret = btrfs_pin_extent_for_log_replay(fs_info->extent_root,
5721 free_extent_buffer(log->node);
5722 free_extent_buffer(log->commit_root);
5727 btrfs_handle_fs_error(fs_info, ret,
5728 "Couldn't read target root for tree log recovery.");
5732 wc.replay_dest->log_root = log;
5733 btrfs_record_root_in_trans(trans, wc.replay_dest);
5734 ret = walk_log_tree(trans, log, &wc);
5736 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5737 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5741 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5742 struct btrfs_root *root = wc.replay_dest;
5744 btrfs_release_path(path);
5747 * We have just replayed everything, and the highest
5748 * objectid of fs roots probably has changed in case
5749 * some inode_item's got replayed.
5751 * root->objectid_mutex is not acquired as log replay
5752 * could only happen during mount.
5754 ret = btrfs_find_highest_objectid(root,
5755 &root->highest_objectid);
5758 wc.replay_dest->log_root = NULL;
5759 free_extent_buffer(log->node);
5760 free_extent_buffer(log->commit_root);
5766 if (found_key.offset == 0)
5768 key.offset = found_key.offset - 1;
5770 btrfs_release_path(path);
5772 /* step one is to pin it all, step two is to replay just inodes */
5775 wc.process_func = replay_one_buffer;
5776 wc.stage = LOG_WALK_REPLAY_INODES;
5779 /* step three is to replay everything */
5780 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5785 btrfs_free_path(path);
5787 /* step 4: commit the transaction, which also unpins the blocks */
5788 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
5792 free_extent_buffer(log_root_tree->node);
5793 log_root_tree->log_root = NULL;
5794 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5795 kfree(log_root_tree);
5800 btrfs_end_transaction(wc.trans, fs_info->tree_root);
5801 btrfs_free_path(path);
5806 * there are some corner cases where we want to force a full
5807 * commit instead of allowing a directory to be logged.
5809 * They revolve around files there were unlinked from the directory, and
5810 * this function updates the parent directory so that a full commit is
5811 * properly done if it is fsync'd later after the unlinks are done.
5813 * Must be called before the unlink operations (updates to the subvolume tree,
5814 * inodes, etc) are done.
5816 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5817 struct inode *dir, struct inode *inode,
5821 * when we're logging a file, if it hasn't been renamed
5822 * or unlinked, and its inode is fully committed on disk,
5823 * we don't have to worry about walking up the directory chain
5824 * to log its parents.
5826 * So, we use the last_unlink_trans field to put this transid
5827 * into the file. When the file is logged we check it and
5828 * don't log the parents if the file is fully on disk.
5830 mutex_lock(&BTRFS_I(inode)->log_mutex);
5831 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5832 mutex_unlock(&BTRFS_I(inode)->log_mutex);
5835 * if this directory was already logged any new
5836 * names for this file/dir will get recorded
5839 if (BTRFS_I(dir)->logged_trans == trans->transid)
5843 * if the inode we're about to unlink was logged,
5844 * the log will be properly updated for any new names
5846 if (BTRFS_I(inode)->logged_trans == trans->transid)
5850 * when renaming files across directories, if the directory
5851 * there we're unlinking from gets fsync'd later on, there's
5852 * no way to find the destination directory later and fsync it
5853 * properly. So, we have to be conservative and force commits
5854 * so the new name gets discovered.
5859 /* we can safely do the unlink without any special recording */
5863 mutex_lock(&BTRFS_I(dir)->log_mutex);
5864 BTRFS_I(dir)->last_unlink_trans = trans->transid;
5865 mutex_unlock(&BTRFS_I(dir)->log_mutex);
5869 * Make sure that if someone attempts to fsync the parent directory of a deleted
5870 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5871 * that after replaying the log tree of the parent directory's root we will not
5872 * see the snapshot anymore and at log replay time we will not see any log tree
5873 * corresponding to the deleted snapshot's root, which could lead to replaying
5874 * it after replaying the log tree of the parent directory (which would replay
5875 * the snapshot delete operation).
5877 * Must be called before the actual snapshot destroy operation (updates to the
5878 * parent root and tree of tree roots trees, etc) are done.
5880 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
5883 mutex_lock(&BTRFS_I(dir)->log_mutex);
5884 BTRFS_I(dir)->last_unlink_trans = trans->transid;
5885 mutex_unlock(&BTRFS_I(dir)->log_mutex);
5889 * Call this after adding a new name for a file and it will properly
5890 * update the log to reflect the new name.
5892 * It will return zero if all goes well, and it will return 1 if a
5893 * full transaction commit is required.
5895 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5896 struct inode *inode, struct inode *old_dir,
5897 struct dentry *parent)
5899 struct btrfs_root * root = BTRFS_I(inode)->root;
5902 * this will force the logging code to walk the dentry chain
5905 if (S_ISREG(inode->i_mode))
5906 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5909 * if this inode hasn't been logged and directory we're renaming it
5910 * from hasn't been logged, we don't need to log it
5912 if (BTRFS_I(inode)->logged_trans <=
5913 root->fs_info->last_trans_committed &&
5914 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
5915 root->fs_info->last_trans_committed))
5918 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
5919 LLONG_MAX, 1, NULL);