1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/blkdev.h>
9 #include <linux/list_sort.h>
10 #include <linux/iversion.h>
15 #include "print-tree.h"
17 #include "compression.h"
19 #include "inode-map.h"
21 /* magic values for the inode_only field in btrfs_log_inode:
23 * LOG_INODE_ALL means to log everything
24 * LOG_INODE_EXISTS means to log just enough to recreate the inode
27 #define LOG_INODE_ALL 0
28 #define LOG_INODE_EXISTS 1
29 #define LOG_OTHER_INODE 2
32 * directory trouble cases
34 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
35 * log, we must force a full commit before doing an fsync of the directory
36 * where the unlink was done.
37 * ---> record transid of last unlink/rename per directory
41 * rename foo/some_dir foo2/some_dir
43 * fsync foo/some_dir/some_file
45 * The fsync above will unlink the original some_dir without recording
46 * it in its new location (foo2). After a crash, some_dir will be gone
47 * unless the fsync of some_file forces a full commit
49 * 2) we must log any new names for any file or dir that is in the fsync
50 * log. ---> check inode while renaming/linking.
52 * 2a) we must log any new names for any file or dir during rename
53 * when the directory they are being removed from was logged.
54 * ---> check inode and old parent dir during rename
56 * 2a is actually the more important variant. With the extra logging
57 * a crash might unlink the old name without recreating the new one
59 * 3) after a crash, we must go through any directories with a link count
60 * of zero and redo the rm -rf
67 * The directory f1 was fully removed from the FS, but fsync was never
68 * called on f1, only its parent dir. After a crash the rm -rf must
69 * be replayed. This must be able to recurse down the entire
70 * directory tree. The inode link count fixup code takes care of the
75 * stages for the tree walking. The first
76 * stage (0) is to only pin down the blocks we find
77 * the second stage (1) is to make sure that all the inodes
78 * we find in the log are created in the subvolume.
80 * The last stage is to deal with directories and links and extents
81 * and all the other fun semantics
83 #define LOG_WALK_PIN_ONLY 0
84 #define LOG_WALK_REPLAY_INODES 1
85 #define LOG_WALK_REPLAY_DIR_INDEX 2
86 #define LOG_WALK_REPLAY_ALL 3
88 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root, struct btrfs_inode *inode,
93 struct btrfs_log_ctx *ctx);
94 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root,
96 struct btrfs_path *path, u64 objectid);
97 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root,
99 struct btrfs_root *log,
100 struct btrfs_path *path,
101 u64 dirid, int del_all);
104 * tree logging is a special write ahead log used to make sure that
105 * fsyncs and O_SYNCs can happen without doing full tree commits.
107 * Full tree commits are expensive because they require commonly
108 * modified blocks to be recowed, creating many dirty pages in the
109 * extent tree an 4x-6x higher write load than ext3.
111 * Instead of doing a tree commit on every fsync, we use the
112 * key ranges and transaction ids to find items for a given file or directory
113 * that have changed in this transaction. Those items are copied into
114 * a special tree (one per subvolume root), that tree is written to disk
115 * and then the fsync is considered complete.
117 * After a crash, items are copied out of the log-tree back into the
118 * subvolume tree. Any file data extents found are recorded in the extent
119 * allocation tree, and the log-tree freed.
121 * The log tree is read three times, once to pin down all the extents it is
122 * using in ram and once, once to create all the inodes logged in the tree
123 * and once to do all the other items.
127 * start a sub transaction and setup the log tree
128 * this increments the log tree writer count to make the people
129 * syncing the tree wait for us to finish
131 static int start_log_trans(struct btrfs_trans_handle *trans,
132 struct btrfs_root *root,
133 struct btrfs_log_ctx *ctx)
135 struct btrfs_fs_info *fs_info = root->fs_info;
138 mutex_lock(&root->log_mutex);
140 if (root->log_root) {
141 if (btrfs_need_log_full_commit(fs_info, trans)) {
146 if (!root->log_start_pid) {
147 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
148 root->log_start_pid = current->pid;
149 } else if (root->log_start_pid != current->pid) {
150 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
153 mutex_lock(&fs_info->tree_log_mutex);
154 if (!fs_info->log_root_tree)
155 ret = btrfs_init_log_root_tree(trans, fs_info);
156 mutex_unlock(&fs_info->tree_log_mutex);
160 ret = btrfs_add_log_tree(trans, root);
164 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
165 root->log_start_pid = current->pid;
168 atomic_inc(&root->log_batch);
169 atomic_inc(&root->log_writers);
171 int index = root->log_transid % 2;
172 list_add_tail(&ctx->list, &root->log_ctxs[index]);
173 ctx->log_transid = root->log_transid;
177 mutex_unlock(&root->log_mutex);
182 * returns 0 if there was a log transaction running and we were able
183 * to join, or returns -ENOENT if there were not transactions
186 static int join_running_log_trans(struct btrfs_root *root)
194 mutex_lock(&root->log_mutex);
195 if (root->log_root) {
197 atomic_inc(&root->log_writers);
199 mutex_unlock(&root->log_mutex);
204 * This either makes the current running log transaction wait
205 * until you call btrfs_end_log_trans() or it makes any future
206 * log transactions wait until you call btrfs_end_log_trans()
208 int btrfs_pin_log_trans(struct btrfs_root *root)
212 mutex_lock(&root->log_mutex);
213 atomic_inc(&root->log_writers);
214 mutex_unlock(&root->log_mutex);
219 * indicate we're done making changes to the log tree
220 * and wake up anyone waiting to do a sync
222 void btrfs_end_log_trans(struct btrfs_root *root)
224 if (atomic_dec_and_test(&root->log_writers)) {
226 * Implicit memory barrier after atomic_dec_and_test
228 if (waitqueue_active(&root->log_writer_wait))
229 wake_up(&root->log_writer_wait);
235 * the walk control struct is used to pass state down the chain when
236 * processing the log tree. The stage field tells us which part
237 * of the log tree processing we are currently doing. The others
238 * are state fields used for that specific part
240 struct walk_control {
241 /* should we free the extent on disk when done? This is used
242 * at transaction commit time while freeing a log tree
246 /* should we write out the extent buffer? This is used
247 * while flushing the log tree to disk during a sync
251 /* should we wait for the extent buffer io to finish? Also used
252 * while flushing the log tree to disk for a sync
256 /* pin only walk, we record which extents on disk belong to the
261 /* what stage of the replay code we're currently in */
264 /* the root we are currently replaying */
265 struct btrfs_root *replay_dest;
267 /* the trans handle for the current replay */
268 struct btrfs_trans_handle *trans;
270 /* the function that gets used to process blocks we find in the
271 * tree. Note the extent_buffer might not be up to date when it is
272 * passed in, and it must be checked or read if you need the data
275 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
276 struct walk_control *wc, u64 gen, int level);
280 * process_func used to pin down extents, write them or wait on them
282 static int process_one_buffer(struct btrfs_root *log,
283 struct extent_buffer *eb,
284 struct walk_control *wc, u64 gen, int level)
286 struct btrfs_fs_info *fs_info = log->fs_info;
290 * If this fs is mixed then we need to be able to process the leaves to
291 * pin down any logged extents, so we have to read the block.
293 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
294 ret = btrfs_read_buffer(eb, gen, level, NULL);
300 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
303 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
304 if (wc->pin && btrfs_header_level(eb) == 0)
305 ret = btrfs_exclude_logged_extents(fs_info, eb);
307 btrfs_write_tree_block(eb);
309 btrfs_wait_tree_block_writeback(eb);
315 * Item overwrite used by replay and tree logging. eb, slot and key all refer
316 * to the src data we are copying out.
318 * root is the tree we are copying into, and path is a scratch
319 * path for use in this function (it should be released on entry and
320 * will be released on exit).
322 * If the key is already in the destination tree the existing item is
323 * overwritten. If the existing item isn't big enough, it is extended.
324 * If it is too large, it is truncated.
326 * If the key isn't in the destination yet, a new item is inserted.
328 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
329 struct btrfs_root *root,
330 struct btrfs_path *path,
331 struct extent_buffer *eb, int slot,
332 struct btrfs_key *key)
334 struct btrfs_fs_info *fs_info = root->fs_info;
337 u64 saved_i_size = 0;
338 int save_old_i_size = 0;
339 unsigned long src_ptr;
340 unsigned long dst_ptr;
341 int overwrite_root = 0;
342 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
344 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
347 item_size = btrfs_item_size_nr(eb, slot);
348 src_ptr = btrfs_item_ptr_offset(eb, slot);
350 /* look for the key in the destination tree */
351 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
358 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
360 if (dst_size != item_size)
363 if (item_size == 0) {
364 btrfs_release_path(path);
367 dst_copy = kmalloc(item_size, GFP_NOFS);
368 src_copy = kmalloc(item_size, GFP_NOFS);
369 if (!dst_copy || !src_copy) {
370 btrfs_release_path(path);
376 read_extent_buffer(eb, src_copy, src_ptr, item_size);
378 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
379 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
381 ret = memcmp(dst_copy, src_copy, item_size);
386 * they have the same contents, just return, this saves
387 * us from cowing blocks in the destination tree and doing
388 * extra writes that may not have been done by a previous
392 btrfs_release_path(path);
397 * We need to load the old nbytes into the inode so when we
398 * replay the extents we've logged we get the right nbytes.
401 struct btrfs_inode_item *item;
405 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
406 struct btrfs_inode_item);
407 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
408 item = btrfs_item_ptr(eb, slot,
409 struct btrfs_inode_item);
410 btrfs_set_inode_nbytes(eb, item, nbytes);
413 * If this is a directory we need to reset the i_size to
414 * 0 so that we can set it up properly when replaying
415 * the rest of the items in this log.
417 mode = btrfs_inode_mode(eb, item);
419 btrfs_set_inode_size(eb, item, 0);
421 } else if (inode_item) {
422 struct btrfs_inode_item *item;
426 * New inode, set nbytes to 0 so that the nbytes comes out
427 * properly when we replay the extents.
429 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
430 btrfs_set_inode_nbytes(eb, item, 0);
433 * If this is a directory we need to reset the i_size to 0 so
434 * that we can set it up properly when replaying the rest of
435 * the items in this log.
437 mode = btrfs_inode_mode(eb, item);
439 btrfs_set_inode_size(eb, item, 0);
442 btrfs_release_path(path);
443 /* try to insert the key into the destination tree */
444 path->skip_release_on_error = 1;
445 ret = btrfs_insert_empty_item(trans, root, path,
447 path->skip_release_on_error = 0;
449 /* make sure any existing item is the correct size */
450 if (ret == -EEXIST || ret == -EOVERFLOW) {
452 found_size = btrfs_item_size_nr(path->nodes[0],
454 if (found_size > item_size)
455 btrfs_truncate_item(fs_info, path, item_size, 1);
456 else if (found_size < item_size)
457 btrfs_extend_item(fs_info, path,
458 item_size - found_size);
462 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
465 /* don't overwrite an existing inode if the generation number
466 * was logged as zero. This is done when the tree logging code
467 * is just logging an inode to make sure it exists after recovery.
469 * Also, don't overwrite i_size on directories during replay.
470 * log replay inserts and removes directory items based on the
471 * state of the tree found in the subvolume, and i_size is modified
474 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
475 struct btrfs_inode_item *src_item;
476 struct btrfs_inode_item *dst_item;
478 src_item = (struct btrfs_inode_item *)src_ptr;
479 dst_item = (struct btrfs_inode_item *)dst_ptr;
481 if (btrfs_inode_generation(eb, src_item) == 0) {
482 struct extent_buffer *dst_eb = path->nodes[0];
483 const u64 ino_size = btrfs_inode_size(eb, src_item);
486 * For regular files an ino_size == 0 is used only when
487 * logging that an inode exists, as part of a directory
488 * fsync, and the inode wasn't fsynced before. In this
489 * case don't set the size of the inode in the fs/subvol
490 * tree, otherwise we would be throwing valid data away.
492 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
493 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
495 struct btrfs_map_token token;
497 btrfs_init_map_token(&token);
498 btrfs_set_token_inode_size(dst_eb, dst_item,
504 if (overwrite_root &&
505 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
506 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
508 saved_i_size = btrfs_inode_size(path->nodes[0],
513 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
516 if (save_old_i_size) {
517 struct btrfs_inode_item *dst_item;
518 dst_item = (struct btrfs_inode_item *)dst_ptr;
519 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
522 /* make sure the generation is filled in */
523 if (key->type == BTRFS_INODE_ITEM_KEY) {
524 struct btrfs_inode_item *dst_item;
525 dst_item = (struct btrfs_inode_item *)dst_ptr;
526 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
527 btrfs_set_inode_generation(path->nodes[0], dst_item,
532 btrfs_mark_buffer_dirty(path->nodes[0]);
533 btrfs_release_path(path);
538 * simple helper to read an inode off the disk from a given root
539 * This can only be called for subvolume roots and not for the log
541 static noinline struct inode *read_one_inode(struct btrfs_root *root,
544 struct btrfs_key key;
547 key.objectid = objectid;
548 key.type = BTRFS_INODE_ITEM_KEY;
550 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
553 } else if (is_bad_inode(inode)) {
560 /* replays a single extent in 'eb' at 'slot' with 'key' into the
561 * subvolume 'root'. path is released on entry and should be released
564 * extents in the log tree have not been allocated out of the extent
565 * tree yet. So, this completes the allocation, taking a reference
566 * as required if the extent already exists or creating a new extent
567 * if it isn't in the extent allocation tree yet.
569 * The extent is inserted into the file, dropping any existing extents
570 * from the file that overlap the new one.
572 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
573 struct btrfs_root *root,
574 struct btrfs_path *path,
575 struct extent_buffer *eb, int slot,
576 struct btrfs_key *key)
578 struct btrfs_fs_info *fs_info = root->fs_info;
581 u64 start = key->offset;
583 struct btrfs_file_extent_item *item;
584 struct inode *inode = NULL;
588 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
589 found_type = btrfs_file_extent_type(eb, item);
591 if (found_type == BTRFS_FILE_EXTENT_REG ||
592 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
593 nbytes = btrfs_file_extent_num_bytes(eb, item);
594 extent_end = start + nbytes;
597 * We don't add to the inodes nbytes if we are prealloc or a
600 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
602 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
603 size = btrfs_file_extent_inline_len(eb, slot, item);
604 nbytes = btrfs_file_extent_ram_bytes(eb, item);
605 extent_end = ALIGN(start + size,
606 fs_info->sectorsize);
612 inode = read_one_inode(root, key->objectid);
619 * first check to see if we already have this extent in the
620 * file. This must be done before the btrfs_drop_extents run
621 * so we don't try to drop this extent.
623 ret = btrfs_lookup_file_extent(trans, root, path,
624 btrfs_ino(BTRFS_I(inode)), start, 0);
627 (found_type == BTRFS_FILE_EXTENT_REG ||
628 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
629 struct btrfs_file_extent_item cmp1;
630 struct btrfs_file_extent_item cmp2;
631 struct btrfs_file_extent_item *existing;
632 struct extent_buffer *leaf;
634 leaf = path->nodes[0];
635 existing = btrfs_item_ptr(leaf, path->slots[0],
636 struct btrfs_file_extent_item);
638 read_extent_buffer(eb, &cmp1, (unsigned long)item,
640 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
644 * we already have a pointer to this exact extent,
645 * we don't have to do anything
647 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
648 btrfs_release_path(path);
652 btrfs_release_path(path);
654 /* drop any overlapping extents */
655 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
659 if (found_type == BTRFS_FILE_EXTENT_REG ||
660 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
662 unsigned long dest_offset;
663 struct btrfs_key ins;
665 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
666 btrfs_fs_incompat(fs_info, NO_HOLES))
669 ret = btrfs_insert_empty_item(trans, root, path, key,
673 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
675 copy_extent_buffer(path->nodes[0], eb, dest_offset,
676 (unsigned long)item, sizeof(*item));
678 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
679 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
680 ins.type = BTRFS_EXTENT_ITEM_KEY;
681 offset = key->offset - btrfs_file_extent_offset(eb, item);
684 * Manually record dirty extent, as here we did a shallow
685 * file extent item copy and skip normal backref update,
686 * but modifying extent tree all by ourselves.
687 * So need to manually record dirty extent for qgroup,
688 * as the owner of the file extent changed from log tree
689 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
691 ret = btrfs_qgroup_trace_extent(trans, fs_info,
692 btrfs_file_extent_disk_bytenr(eb, item),
693 btrfs_file_extent_disk_num_bytes(eb, item),
698 if (ins.objectid > 0) {
701 LIST_HEAD(ordered_sums);
703 * is this extent already allocated in the extent
704 * allocation tree? If so, just add a reference
706 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
709 ret = btrfs_inc_extent_ref(trans, root,
710 ins.objectid, ins.offset,
711 0, root->root_key.objectid,
712 key->objectid, offset);
717 * insert the extent pointer in the extent
720 ret = btrfs_alloc_logged_file_extent(trans,
722 root->root_key.objectid,
723 key->objectid, offset, &ins);
727 btrfs_release_path(path);
729 if (btrfs_file_extent_compression(eb, item)) {
730 csum_start = ins.objectid;
731 csum_end = csum_start + ins.offset;
733 csum_start = ins.objectid +
734 btrfs_file_extent_offset(eb, item);
735 csum_end = csum_start +
736 btrfs_file_extent_num_bytes(eb, item);
739 ret = btrfs_lookup_csums_range(root->log_root,
740 csum_start, csum_end - 1,
745 * Now delete all existing cums in the csum root that
746 * cover our range. We do this because we can have an
747 * extent that is completely referenced by one file
748 * extent item and partially referenced by another
749 * file extent item (like after using the clone or
750 * extent_same ioctls). In this case if we end up doing
751 * the replay of the one that partially references the
752 * extent first, and we do not do the csum deletion
753 * below, we can get 2 csum items in the csum tree that
754 * overlap each other. For example, imagine our log has
755 * the two following file extent items:
757 * key (257 EXTENT_DATA 409600)
758 * extent data disk byte 12845056 nr 102400
759 * extent data offset 20480 nr 20480 ram 102400
761 * key (257 EXTENT_DATA 819200)
762 * extent data disk byte 12845056 nr 102400
763 * extent data offset 0 nr 102400 ram 102400
765 * Where the second one fully references the 100K extent
766 * that starts at disk byte 12845056, and the log tree
767 * has a single csum item that covers the entire range
770 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
772 * After the first file extent item is replayed, the
773 * csum tree gets the following csum item:
775 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
777 * Which covers the 20K sub-range starting at offset 20K
778 * of our extent. Now when we replay the second file
779 * extent item, if we do not delete existing csum items
780 * that cover any of its blocks, we end up getting two
781 * csum items in our csum tree that overlap each other:
783 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
784 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
786 * Which is a problem, because after this anyone trying
787 * to lookup up for the checksum of any block of our
788 * extent starting at an offset of 40K or higher, will
789 * end up looking at the second csum item only, which
790 * does not contain the checksum for any block starting
791 * at offset 40K or higher of our extent.
793 while (!list_empty(&ordered_sums)) {
794 struct btrfs_ordered_sum *sums;
795 sums = list_entry(ordered_sums.next,
796 struct btrfs_ordered_sum,
799 ret = btrfs_del_csums(trans, fs_info,
803 ret = btrfs_csum_file_blocks(trans,
804 fs_info->csum_root, sums);
805 list_del(&sums->list);
811 btrfs_release_path(path);
813 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
814 /* inline extents are easy, we just overwrite them */
815 ret = overwrite_item(trans, root, path, eb, slot, key);
820 inode_add_bytes(inode, nbytes);
822 ret = btrfs_update_inode(trans, root, inode);
830 * when cleaning up conflicts between the directory names in the
831 * subvolume, directory names in the log and directory names in the
832 * inode back references, we may have to unlink inodes from directories.
834 * This is a helper function to do the unlink of a specific directory
837 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
838 struct btrfs_root *root,
839 struct btrfs_path *path,
840 struct btrfs_inode *dir,
841 struct btrfs_dir_item *di)
846 struct extent_buffer *leaf;
847 struct btrfs_key location;
850 leaf = path->nodes[0];
852 btrfs_dir_item_key_to_cpu(leaf, di, &location);
853 name_len = btrfs_dir_name_len(leaf, di);
854 name = kmalloc(name_len, GFP_NOFS);
858 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
859 btrfs_release_path(path);
861 inode = read_one_inode(root, location.objectid);
867 ret = link_to_fixup_dir(trans, root, path, location.objectid);
871 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
876 ret = btrfs_run_delayed_items(trans);
884 * helper function to see if a given name and sequence number found
885 * in an inode back reference are already in a directory and correctly
886 * point to this inode
888 static noinline int inode_in_dir(struct btrfs_root *root,
889 struct btrfs_path *path,
890 u64 dirid, u64 objectid, u64 index,
891 const char *name, int name_len)
893 struct btrfs_dir_item *di;
894 struct btrfs_key location;
897 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
898 index, name, name_len, 0);
899 if (di && !IS_ERR(di)) {
900 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
901 if (location.objectid != objectid)
905 btrfs_release_path(path);
907 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
908 if (di && !IS_ERR(di)) {
909 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
910 if (location.objectid != objectid)
916 btrfs_release_path(path);
921 * helper function to check a log tree for a named back reference in
922 * an inode. This is used to decide if a back reference that is
923 * found in the subvolume conflicts with what we find in the log.
925 * inode backreferences may have multiple refs in a single item,
926 * during replay we process one reference at a time, and we don't
927 * want to delete valid links to a file from the subvolume if that
928 * link is also in the log.
930 static noinline int backref_in_log(struct btrfs_root *log,
931 struct btrfs_key *key,
933 const char *name, int namelen)
935 struct btrfs_path *path;
936 struct btrfs_inode_ref *ref;
938 unsigned long ptr_end;
939 unsigned long name_ptr;
945 path = btrfs_alloc_path();
949 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
953 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
955 if (key->type == BTRFS_INODE_EXTREF_KEY) {
956 if (btrfs_find_name_in_ext_backref(path->nodes[0],
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 btrfs_inode *dir,
991 struct btrfs_inode *inode,
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,
1045 inc_nlink(&inode->vfs_inode);
1046 btrfs_release_path(path);
1048 ret = btrfs_unlink_inode(trans, root, dir, inode,
1049 victim_name, victim_name_len);
1053 ret = btrfs_run_delayed_items(trans);
1061 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1065 * NOTE: we have searched root tree and checked the
1066 * corresponding ref, it does not need to check again.
1070 btrfs_release_path(path);
1072 /* Same search but for extended refs */
1073 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1074 inode_objectid, parent_objectid, 0,
1076 if (!IS_ERR_OR_NULL(extref)) {
1080 struct inode *victim_parent;
1082 leaf = path->nodes[0];
1084 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1085 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1087 while (cur_offset < item_size) {
1088 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1090 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1092 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1095 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1098 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1101 search_key.objectid = inode_objectid;
1102 search_key.type = BTRFS_INODE_EXTREF_KEY;
1103 search_key.offset = btrfs_extref_hash(parent_objectid,
1107 if (!backref_in_log(log_root, &search_key,
1108 parent_objectid, victim_name,
1111 victim_parent = read_one_inode(root,
1113 if (victim_parent) {
1114 inc_nlink(&inode->vfs_inode);
1115 btrfs_release_path(path);
1117 ret = btrfs_unlink_inode(trans, root,
1118 BTRFS_I(victim_parent),
1123 ret = btrfs_run_delayed_items(
1126 iput(victim_parent);
1135 cur_offset += victim_name_len + sizeof(*extref);
1139 btrfs_release_path(path);
1141 /* look for a conflicting sequence number */
1142 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1143 ref_index, name, namelen, 0);
1144 if (di && !IS_ERR(di)) {
1145 ret = drop_one_dir_item(trans, root, path, dir, di);
1149 btrfs_release_path(path);
1151 /* look for a conflicing name */
1152 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1154 if (di && !IS_ERR(di)) {
1155 ret = drop_one_dir_item(trans, root, path, dir, di);
1159 btrfs_release_path(path);
1164 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1165 u32 *namelen, char **name, u64 *index,
1166 u64 *parent_objectid)
1168 struct btrfs_inode_extref *extref;
1170 extref = (struct btrfs_inode_extref *)ref_ptr;
1172 *namelen = btrfs_inode_extref_name_len(eb, extref);
1173 *name = kmalloc(*namelen, GFP_NOFS);
1177 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1181 *index = btrfs_inode_extref_index(eb, extref);
1182 if (parent_objectid)
1183 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1188 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1189 u32 *namelen, char **name, u64 *index)
1191 struct btrfs_inode_ref *ref;
1193 ref = (struct btrfs_inode_ref *)ref_ptr;
1195 *namelen = btrfs_inode_ref_name_len(eb, ref);
1196 *name = kmalloc(*namelen, GFP_NOFS);
1200 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1203 *index = btrfs_inode_ref_index(eb, ref);
1209 * Take an inode reference item from the log tree and iterate all names from the
1210 * inode reference item in the subvolume tree with the same key (if it exists).
1211 * For any name that is not in the inode reference item from the log tree, do a
1212 * proper unlink of that name (that is, remove its entry from the inode
1213 * reference item and both dir index keys).
1215 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1216 struct btrfs_root *root,
1217 struct btrfs_path *path,
1218 struct btrfs_inode *inode,
1219 struct extent_buffer *log_eb,
1221 struct btrfs_key *key)
1224 unsigned long ref_ptr;
1225 unsigned long ref_end;
1226 struct extent_buffer *eb;
1229 btrfs_release_path(path);
1230 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1238 eb = path->nodes[0];
1239 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1240 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1241 while (ref_ptr < ref_end) {
1246 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1247 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1250 parent_id = key->offset;
1251 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1257 if (key->type == BTRFS_INODE_EXTREF_KEY)
1258 ret = btrfs_find_name_in_ext_backref(log_eb, log_slot,
1262 ret = btrfs_find_name_in_backref(log_eb, log_slot, name,
1268 btrfs_release_path(path);
1269 dir = read_one_inode(root, parent_id);
1275 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1276 inode, name, namelen);
1286 if (key->type == BTRFS_INODE_EXTREF_KEY)
1287 ref_ptr += sizeof(struct btrfs_inode_extref);
1289 ref_ptr += sizeof(struct btrfs_inode_ref);
1293 btrfs_release_path(path);
1298 * replay one inode back reference item found in the log tree.
1299 * eb, slot and key refer to the buffer and key found in the log tree.
1300 * root is the destination we are replaying into, and path is for temp
1301 * use by this function. (it should be released on return).
1303 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1304 struct btrfs_root *root,
1305 struct btrfs_root *log,
1306 struct btrfs_path *path,
1307 struct extent_buffer *eb, int slot,
1308 struct btrfs_key *key)
1310 struct inode *dir = NULL;
1311 struct inode *inode = NULL;
1312 unsigned long ref_ptr;
1313 unsigned long ref_end;
1317 int search_done = 0;
1318 int log_ref_ver = 0;
1319 u64 parent_objectid;
1322 int ref_struct_size;
1324 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1325 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1327 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1328 struct btrfs_inode_extref *r;
1330 ref_struct_size = sizeof(struct btrfs_inode_extref);
1332 r = (struct btrfs_inode_extref *)ref_ptr;
1333 parent_objectid = btrfs_inode_extref_parent(eb, r);
1335 ref_struct_size = sizeof(struct btrfs_inode_ref);
1336 parent_objectid = key->offset;
1338 inode_objectid = key->objectid;
1341 * it is possible that we didn't log all the parent directories
1342 * for a given inode. If we don't find the dir, just don't
1343 * copy the back ref in. The link count fixup code will take
1346 dir = read_one_inode(root, parent_objectid);
1352 inode = read_one_inode(root, inode_objectid);
1358 while (ref_ptr < ref_end) {
1360 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1361 &ref_index, &parent_objectid);
1363 * parent object can change from one array
1367 dir = read_one_inode(root, parent_objectid);
1373 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1379 /* if we already have a perfect match, we're done */
1380 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1381 btrfs_ino(BTRFS_I(inode)), ref_index,
1384 * look for a conflicting back reference in the
1385 * metadata. if we find one we have to unlink that name
1386 * of the file before we add our new link. Later on, we
1387 * overwrite any existing back reference, and we don't
1388 * want to create dangling pointers in the directory.
1392 ret = __add_inode_ref(trans, root, path, log,
1397 ref_index, name, namelen,
1406 /* insert our name */
1407 ret = btrfs_add_link(trans, BTRFS_I(dir),
1409 name, namelen, 0, ref_index);
1413 btrfs_update_inode(trans, root, inode);
1416 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1426 * Before we overwrite the inode reference item in the subvolume tree
1427 * with the item from the log tree, we must unlink all names from the
1428 * parent directory that are in the subvolume's tree inode reference
1429 * item, otherwise we end up with an inconsistent subvolume tree where
1430 * dir index entries exist for a name but there is no inode reference
1431 * item with the same name.
1433 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1438 /* finally write the back reference in the inode */
1439 ret = overwrite_item(trans, root, path, eb, slot, key);
1441 btrfs_release_path(path);
1448 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1449 struct btrfs_root *root, u64 ino)
1453 ret = btrfs_insert_orphan_item(trans, root, ino);
1460 static int count_inode_extrefs(struct btrfs_root *root,
1461 struct btrfs_inode *inode, struct btrfs_path *path)
1465 unsigned int nlink = 0;
1468 u64 inode_objectid = btrfs_ino(inode);
1471 struct btrfs_inode_extref *extref;
1472 struct extent_buffer *leaf;
1475 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1480 leaf = path->nodes[0];
1481 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1482 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1485 while (cur_offset < item_size) {
1486 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1487 name_len = btrfs_inode_extref_name_len(leaf, extref);
1491 cur_offset += name_len + sizeof(*extref);
1495 btrfs_release_path(path);
1497 btrfs_release_path(path);
1499 if (ret < 0 && ret != -ENOENT)
1504 static int count_inode_refs(struct btrfs_root *root,
1505 struct btrfs_inode *inode, struct btrfs_path *path)
1508 struct btrfs_key key;
1509 unsigned int nlink = 0;
1511 unsigned long ptr_end;
1513 u64 ino = btrfs_ino(inode);
1516 key.type = BTRFS_INODE_REF_KEY;
1517 key.offset = (u64)-1;
1520 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1524 if (path->slots[0] == 0)
1529 btrfs_item_key_to_cpu(path->nodes[0], &key,
1531 if (key.objectid != ino ||
1532 key.type != BTRFS_INODE_REF_KEY)
1534 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1535 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1537 while (ptr < ptr_end) {
1538 struct btrfs_inode_ref *ref;
1540 ref = (struct btrfs_inode_ref *)ptr;
1541 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1543 ptr = (unsigned long)(ref + 1) + name_len;
1547 if (key.offset == 0)
1549 if (path->slots[0] > 0) {
1554 btrfs_release_path(path);
1556 btrfs_release_path(path);
1562 * There are a few corners where the link count of the file can't
1563 * be properly maintained during replay. So, instead of adding
1564 * lots of complexity to the log code, we just scan the backrefs
1565 * for any file that has been through replay.
1567 * The scan will update the link count on the inode to reflect the
1568 * number of back refs found. If it goes down to zero, the iput
1569 * will free the inode.
1571 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1572 struct btrfs_root *root,
1573 struct inode *inode)
1575 struct btrfs_path *path;
1578 u64 ino = btrfs_ino(BTRFS_I(inode));
1580 path = btrfs_alloc_path();
1584 ret = count_inode_refs(root, BTRFS_I(inode), path);
1590 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1598 if (nlink != inode->i_nlink) {
1599 set_nlink(inode, nlink);
1600 btrfs_update_inode(trans, root, inode);
1602 BTRFS_I(inode)->index_cnt = (u64)-1;
1604 if (inode->i_nlink == 0) {
1605 if (S_ISDIR(inode->i_mode)) {
1606 ret = replay_dir_deletes(trans, root, NULL, path,
1611 ret = insert_orphan_item(trans, root, ino);
1615 btrfs_free_path(path);
1619 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1620 struct btrfs_root *root,
1621 struct btrfs_path *path)
1624 struct btrfs_key key;
1625 struct inode *inode;
1627 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1628 key.type = BTRFS_ORPHAN_ITEM_KEY;
1629 key.offset = (u64)-1;
1631 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1636 if (path->slots[0] == 0)
1641 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1642 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1643 key.type != BTRFS_ORPHAN_ITEM_KEY)
1646 ret = btrfs_del_item(trans, root, path);
1650 btrfs_release_path(path);
1651 inode = read_one_inode(root, key.offset);
1655 ret = fixup_inode_link_count(trans, root, inode);
1661 * fixup on a directory may create new entries,
1662 * make sure we always look for the highset possible
1665 key.offset = (u64)-1;
1669 btrfs_release_path(path);
1675 * record a given inode in the fixup dir so we can check its link
1676 * count when replay is done. The link count is incremented here
1677 * so the inode won't go away until we check it
1679 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1680 struct btrfs_root *root,
1681 struct btrfs_path *path,
1684 struct btrfs_key key;
1686 struct inode *inode;
1688 inode = read_one_inode(root, objectid);
1692 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1693 key.type = BTRFS_ORPHAN_ITEM_KEY;
1694 key.offset = objectid;
1696 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1698 btrfs_release_path(path);
1700 if (!inode->i_nlink)
1701 set_nlink(inode, 1);
1704 ret = btrfs_update_inode(trans, root, inode);
1705 } else if (ret == -EEXIST) {
1708 BUG(); /* Logic Error */
1716 * when replaying the log for a directory, we only insert names
1717 * for inodes that actually exist. This means an fsync on a directory
1718 * does not implicitly fsync all the new files in it
1720 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1721 struct btrfs_root *root,
1722 u64 dirid, u64 index,
1723 char *name, int name_len,
1724 struct btrfs_key *location)
1726 struct inode *inode;
1730 inode = read_one_inode(root, location->objectid);
1734 dir = read_one_inode(root, dirid);
1740 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1741 name_len, 1, index);
1743 /* FIXME, put inode into FIXUP list */
1751 * Return true if an inode reference exists in the log for the given name,
1752 * inode and parent inode.
1754 static bool name_in_log_ref(struct btrfs_root *log_root,
1755 const char *name, const int name_len,
1756 const u64 dirid, const u64 ino)
1758 struct btrfs_key search_key;
1760 search_key.objectid = ino;
1761 search_key.type = BTRFS_INODE_REF_KEY;
1762 search_key.offset = dirid;
1763 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1766 search_key.type = BTRFS_INODE_EXTREF_KEY;
1767 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1768 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1775 * take a single entry in a log directory item and replay it into
1778 * if a conflicting item exists in the subdirectory already,
1779 * the inode it points to is unlinked and put into the link count
1782 * If a name from the log points to a file or directory that does
1783 * not exist in the FS, it is skipped. fsyncs on directories
1784 * do not force down inodes inside that directory, just changes to the
1785 * names or unlinks in a directory.
1787 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1788 * non-existing inode) and 1 if the name was replayed.
1790 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1791 struct btrfs_root *root,
1792 struct btrfs_path *path,
1793 struct extent_buffer *eb,
1794 struct btrfs_dir_item *di,
1795 struct btrfs_key *key)
1799 struct btrfs_dir_item *dst_di;
1800 struct btrfs_key found_key;
1801 struct btrfs_key log_key;
1806 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1807 bool name_added = false;
1809 dir = read_one_inode(root, key->objectid);
1813 name_len = btrfs_dir_name_len(eb, di);
1814 name = kmalloc(name_len, GFP_NOFS);
1820 log_type = btrfs_dir_type(eb, di);
1821 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1824 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1825 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1830 btrfs_release_path(path);
1832 if (key->type == BTRFS_DIR_ITEM_KEY) {
1833 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1835 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1836 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1845 if (IS_ERR_OR_NULL(dst_di)) {
1846 /* we need a sequence number to insert, so we only
1847 * do inserts for the BTRFS_DIR_INDEX_KEY types
1849 if (key->type != BTRFS_DIR_INDEX_KEY)
1854 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1855 /* the existing item matches the logged item */
1856 if (found_key.objectid == log_key.objectid &&
1857 found_key.type == log_key.type &&
1858 found_key.offset == log_key.offset &&
1859 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1860 update_size = false;
1865 * don't drop the conflicting directory entry if the inode
1866 * for the new entry doesn't exist
1871 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
1875 if (key->type == BTRFS_DIR_INDEX_KEY)
1878 btrfs_release_path(path);
1879 if (!ret && update_size) {
1880 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
1881 ret = btrfs_update_inode(trans, root, dir);
1885 if (!ret && name_added)
1890 if (name_in_log_ref(root->log_root, name, name_len,
1891 key->objectid, log_key.objectid)) {
1892 /* The dentry will be added later. */
1894 update_size = false;
1897 btrfs_release_path(path);
1898 ret = insert_one_name(trans, root, key->objectid, key->offset,
1899 name, name_len, &log_key);
1900 if (ret && ret != -ENOENT && ret != -EEXIST)
1904 update_size = false;
1910 * find all the names in a directory item and reconcile them into
1911 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1912 * one name in a directory item, but the same code gets used for
1913 * both directory index types
1915 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1916 struct btrfs_root *root,
1917 struct btrfs_path *path,
1918 struct extent_buffer *eb, int slot,
1919 struct btrfs_key *key)
1922 u32 item_size = btrfs_item_size_nr(eb, slot);
1923 struct btrfs_dir_item *di;
1926 unsigned long ptr_end;
1927 struct btrfs_path *fixup_path = NULL;
1929 ptr = btrfs_item_ptr_offset(eb, slot);
1930 ptr_end = ptr + item_size;
1931 while (ptr < ptr_end) {
1932 di = (struct btrfs_dir_item *)ptr;
1933 name_len = btrfs_dir_name_len(eb, di);
1934 ret = replay_one_name(trans, root, path, eb, di, key);
1937 ptr = (unsigned long)(di + 1);
1941 * If this entry refers to a non-directory (directories can not
1942 * have a link count > 1) and it was added in the transaction
1943 * that was not committed, make sure we fixup the link count of
1944 * the inode it the entry points to. Otherwise something like
1945 * the following would result in a directory pointing to an
1946 * inode with a wrong link that does not account for this dir
1954 * ln testdir/bar testdir/bar_link
1955 * ln testdir/foo testdir/foo_link
1956 * xfs_io -c "fsync" testdir/bar
1960 * mount fs, log replay happens
1962 * File foo would remain with a link count of 1 when it has two
1963 * entries pointing to it in the directory testdir. This would
1964 * make it impossible to ever delete the parent directory has
1965 * it would result in stale dentries that can never be deleted.
1967 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1968 struct btrfs_key di_key;
1971 fixup_path = btrfs_alloc_path();
1978 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1979 ret = link_to_fixup_dir(trans, root, fixup_path,
1986 btrfs_free_path(fixup_path);
1991 * directory replay has two parts. There are the standard directory
1992 * items in the log copied from the subvolume, and range items
1993 * created in the log while the subvolume was logged.
1995 * The range items tell us which parts of the key space the log
1996 * is authoritative for. During replay, if a key in the subvolume
1997 * directory is in a logged range item, but not actually in the log
1998 * that means it was deleted from the directory before the fsync
1999 * and should be removed.
2001 static noinline int find_dir_range(struct btrfs_root *root,
2002 struct btrfs_path *path,
2003 u64 dirid, int key_type,
2004 u64 *start_ret, u64 *end_ret)
2006 struct btrfs_key key;
2008 struct btrfs_dir_log_item *item;
2012 if (*start_ret == (u64)-1)
2015 key.objectid = dirid;
2016 key.type = key_type;
2017 key.offset = *start_ret;
2019 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2023 if (path->slots[0] == 0)
2028 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2030 if (key.type != key_type || key.objectid != dirid) {
2034 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2035 struct btrfs_dir_log_item);
2036 found_end = btrfs_dir_log_end(path->nodes[0], item);
2038 if (*start_ret >= key.offset && *start_ret <= found_end) {
2040 *start_ret = key.offset;
2041 *end_ret = found_end;
2046 /* check the next slot in the tree to see if it is a valid item */
2047 nritems = btrfs_header_nritems(path->nodes[0]);
2049 if (path->slots[0] >= nritems) {
2050 ret = btrfs_next_leaf(root, path);
2055 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2057 if (key.type != key_type || key.objectid != dirid) {
2061 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2062 struct btrfs_dir_log_item);
2063 found_end = btrfs_dir_log_end(path->nodes[0], item);
2064 *start_ret = key.offset;
2065 *end_ret = found_end;
2068 btrfs_release_path(path);
2073 * this looks for a given directory item in the log. If the directory
2074 * item is not in the log, the item is removed and the inode it points
2077 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2078 struct btrfs_root *root,
2079 struct btrfs_root *log,
2080 struct btrfs_path *path,
2081 struct btrfs_path *log_path,
2083 struct btrfs_key *dir_key)
2086 struct extent_buffer *eb;
2089 struct btrfs_dir_item *di;
2090 struct btrfs_dir_item *log_di;
2093 unsigned long ptr_end;
2095 struct inode *inode;
2096 struct btrfs_key location;
2099 eb = path->nodes[0];
2100 slot = path->slots[0];
2101 item_size = btrfs_item_size_nr(eb, slot);
2102 ptr = btrfs_item_ptr_offset(eb, slot);
2103 ptr_end = ptr + item_size;
2104 while (ptr < ptr_end) {
2105 di = (struct btrfs_dir_item *)ptr;
2106 name_len = btrfs_dir_name_len(eb, di);
2107 name = kmalloc(name_len, GFP_NOFS);
2112 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2115 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2116 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2119 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2120 log_di = btrfs_lookup_dir_index_item(trans, log,
2126 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
2127 btrfs_dir_item_key_to_cpu(eb, di, &location);
2128 btrfs_release_path(path);
2129 btrfs_release_path(log_path);
2130 inode = read_one_inode(root, location.objectid);
2136 ret = link_to_fixup_dir(trans, root,
2137 path, location.objectid);
2145 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2146 BTRFS_I(inode), name, name_len);
2148 ret = btrfs_run_delayed_items(trans);
2154 /* there might still be more names under this key
2155 * check and repeat if required
2157 ret = btrfs_search_slot(NULL, root, dir_key, path,
2163 } else if (IS_ERR(log_di)) {
2165 return PTR_ERR(log_di);
2167 btrfs_release_path(log_path);
2170 ptr = (unsigned long)(di + 1);
2175 btrfs_release_path(path);
2176 btrfs_release_path(log_path);
2180 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2181 struct btrfs_root *root,
2182 struct btrfs_root *log,
2183 struct btrfs_path *path,
2186 struct btrfs_key search_key;
2187 struct btrfs_path *log_path;
2192 log_path = btrfs_alloc_path();
2196 search_key.objectid = ino;
2197 search_key.type = BTRFS_XATTR_ITEM_KEY;
2198 search_key.offset = 0;
2200 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2204 nritems = btrfs_header_nritems(path->nodes[0]);
2205 for (i = path->slots[0]; i < nritems; i++) {
2206 struct btrfs_key key;
2207 struct btrfs_dir_item *di;
2208 struct btrfs_dir_item *log_di;
2212 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2213 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2218 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2219 total_size = btrfs_item_size_nr(path->nodes[0], i);
2221 while (cur < total_size) {
2222 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2223 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2224 u32 this_len = sizeof(*di) + name_len + data_len;
2227 name = kmalloc(name_len, GFP_NOFS);
2232 read_extent_buffer(path->nodes[0], name,
2233 (unsigned long)(di + 1), name_len);
2235 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2237 btrfs_release_path(log_path);
2239 /* Doesn't exist in log tree, so delete it. */
2240 btrfs_release_path(path);
2241 di = btrfs_lookup_xattr(trans, root, path, ino,
2242 name, name_len, -1);
2249 ret = btrfs_delete_one_dir_name(trans, root,
2253 btrfs_release_path(path);
2258 if (IS_ERR(log_di)) {
2259 ret = PTR_ERR(log_di);
2263 di = (struct btrfs_dir_item *)((char *)di + this_len);
2266 ret = btrfs_next_leaf(root, path);
2272 btrfs_free_path(log_path);
2273 btrfs_release_path(path);
2279 * deletion replay happens before we copy any new directory items
2280 * out of the log or out of backreferences from inodes. It
2281 * scans the log to find ranges of keys that log is authoritative for,
2282 * and then scans the directory to find items in those ranges that are
2283 * not present in the log.
2285 * Anything we don't find in the log is unlinked and removed from the
2288 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2289 struct btrfs_root *root,
2290 struct btrfs_root *log,
2291 struct btrfs_path *path,
2292 u64 dirid, int del_all)
2296 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2298 struct btrfs_key dir_key;
2299 struct btrfs_key found_key;
2300 struct btrfs_path *log_path;
2303 dir_key.objectid = dirid;
2304 dir_key.type = BTRFS_DIR_ITEM_KEY;
2305 log_path = btrfs_alloc_path();
2309 dir = read_one_inode(root, dirid);
2310 /* it isn't an error if the inode isn't there, that can happen
2311 * because we replay the deletes before we copy in the inode item
2315 btrfs_free_path(log_path);
2323 range_end = (u64)-1;
2325 ret = find_dir_range(log, path, dirid, key_type,
2326 &range_start, &range_end);
2331 dir_key.offset = range_start;
2334 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2339 nritems = btrfs_header_nritems(path->nodes[0]);
2340 if (path->slots[0] >= nritems) {
2341 ret = btrfs_next_leaf(root, path);
2347 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2349 if (found_key.objectid != dirid ||
2350 found_key.type != dir_key.type)
2353 if (found_key.offset > range_end)
2356 ret = check_item_in_log(trans, root, log, path,
2361 if (found_key.offset == (u64)-1)
2363 dir_key.offset = found_key.offset + 1;
2365 btrfs_release_path(path);
2366 if (range_end == (u64)-1)
2368 range_start = range_end + 1;
2373 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2374 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2375 dir_key.type = BTRFS_DIR_INDEX_KEY;
2376 btrfs_release_path(path);
2380 btrfs_release_path(path);
2381 btrfs_free_path(log_path);
2387 * the process_func used to replay items from the log tree. This
2388 * gets called in two different stages. The first stage just looks
2389 * for inodes and makes sure they are all copied into the subvolume.
2391 * The second stage copies all the other item types from the log into
2392 * the subvolume. The two stage approach is slower, but gets rid of
2393 * lots of complexity around inodes referencing other inodes that exist
2394 * only in the log (references come from either directory items or inode
2397 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2398 struct walk_control *wc, u64 gen, int level)
2401 struct btrfs_path *path;
2402 struct btrfs_root *root = wc->replay_dest;
2403 struct btrfs_key key;
2407 ret = btrfs_read_buffer(eb, gen, level, NULL);
2411 level = btrfs_header_level(eb);
2416 path = btrfs_alloc_path();
2420 nritems = btrfs_header_nritems(eb);
2421 for (i = 0; i < nritems; i++) {
2422 btrfs_item_key_to_cpu(eb, &key, i);
2424 /* inode keys are done during the first stage */
2425 if (key.type == BTRFS_INODE_ITEM_KEY &&
2426 wc->stage == LOG_WALK_REPLAY_INODES) {
2427 struct btrfs_inode_item *inode_item;
2430 inode_item = btrfs_item_ptr(eb, i,
2431 struct btrfs_inode_item);
2432 ret = replay_xattr_deletes(wc->trans, root, log,
2433 path, key.objectid);
2436 mode = btrfs_inode_mode(eb, inode_item);
2437 if (S_ISDIR(mode)) {
2438 ret = replay_dir_deletes(wc->trans,
2439 root, log, path, key.objectid, 0);
2443 ret = overwrite_item(wc->trans, root, path,
2449 * Before replaying extents, truncate the inode to its
2450 * size. We need to do it now and not after log replay
2451 * because before an fsync we can have prealloc extents
2452 * added beyond the inode's i_size. If we did it after,
2453 * through orphan cleanup for example, we would drop
2454 * those prealloc extents just after replaying them.
2456 if (S_ISREG(mode)) {
2457 struct inode *inode;
2460 inode = read_one_inode(root, key.objectid);
2465 from = ALIGN(i_size_read(inode),
2466 root->fs_info->sectorsize);
2467 ret = btrfs_drop_extents(wc->trans, root, inode,
2470 * If the nlink count is zero here, the iput
2471 * will free the inode. We bump it to make
2472 * sure it doesn't get freed until the link
2473 * count fixup is done.
2476 if (inode->i_nlink == 0)
2478 /* Update link count and nbytes. */
2479 ret = btrfs_update_inode(wc->trans,
2487 ret = link_to_fixup_dir(wc->trans, root,
2488 path, key.objectid);
2493 if (key.type == BTRFS_DIR_INDEX_KEY &&
2494 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2495 ret = replay_one_dir_item(wc->trans, root, path,
2501 if (wc->stage < LOG_WALK_REPLAY_ALL)
2504 /* these keys are simply copied */
2505 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2506 ret = overwrite_item(wc->trans, root, path,
2510 } else if (key.type == BTRFS_INODE_REF_KEY ||
2511 key.type == BTRFS_INODE_EXTREF_KEY) {
2512 ret = add_inode_ref(wc->trans, root, log, path,
2514 if (ret && ret != -ENOENT)
2517 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2518 ret = replay_one_extent(wc->trans, root, path,
2522 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2523 ret = replay_one_dir_item(wc->trans, root, path,
2529 btrfs_free_path(path);
2533 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2534 struct btrfs_root *root,
2535 struct btrfs_path *path, int *level,
2536 struct walk_control *wc)
2538 struct btrfs_fs_info *fs_info = root->fs_info;
2542 struct extent_buffer *next;
2543 struct extent_buffer *cur;
2544 struct extent_buffer *parent;
2548 WARN_ON(*level < 0);
2549 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2551 while (*level > 0) {
2552 struct btrfs_key first_key;
2554 WARN_ON(*level < 0);
2555 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2556 cur = path->nodes[*level];
2558 WARN_ON(btrfs_header_level(cur) != *level);
2560 if (path->slots[*level] >=
2561 btrfs_header_nritems(cur))
2564 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2565 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2566 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2567 blocksize = fs_info->nodesize;
2569 parent = path->nodes[*level];
2570 root_owner = btrfs_header_owner(parent);
2572 next = btrfs_find_create_tree_block(fs_info, bytenr);
2574 return PTR_ERR(next);
2577 ret = wc->process_func(root, next, wc, ptr_gen,
2580 free_extent_buffer(next);
2584 path->slots[*level]++;
2586 ret = btrfs_read_buffer(next, ptr_gen,
2587 *level - 1, &first_key);
2589 free_extent_buffer(next);
2594 btrfs_tree_lock(next);
2595 btrfs_set_lock_blocking(next);
2596 clean_tree_block(fs_info, next);
2597 btrfs_wait_tree_block_writeback(next);
2598 btrfs_tree_unlock(next);
2600 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2601 clear_extent_buffer_dirty(next);
2604 WARN_ON(root_owner !=
2605 BTRFS_TREE_LOG_OBJECTID);
2606 ret = btrfs_free_and_pin_reserved_extent(
2610 free_extent_buffer(next);
2614 free_extent_buffer(next);
2617 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2619 free_extent_buffer(next);
2623 WARN_ON(*level <= 0);
2624 if (path->nodes[*level-1])
2625 free_extent_buffer(path->nodes[*level-1]);
2626 path->nodes[*level-1] = next;
2627 *level = btrfs_header_level(next);
2628 path->slots[*level] = 0;
2631 WARN_ON(*level < 0);
2632 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2634 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2640 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2641 struct btrfs_root *root,
2642 struct btrfs_path *path, int *level,
2643 struct walk_control *wc)
2645 struct btrfs_fs_info *fs_info = root->fs_info;
2651 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2652 slot = path->slots[i];
2653 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2656 WARN_ON(*level == 0);
2659 struct extent_buffer *parent;
2660 if (path->nodes[*level] == root->node)
2661 parent = path->nodes[*level];
2663 parent = path->nodes[*level + 1];
2665 root_owner = btrfs_header_owner(parent);
2666 ret = wc->process_func(root, path->nodes[*level], wc,
2667 btrfs_header_generation(path->nodes[*level]),
2673 struct extent_buffer *next;
2675 next = path->nodes[*level];
2678 btrfs_tree_lock(next);
2679 btrfs_set_lock_blocking(next);
2680 clean_tree_block(fs_info, next);
2681 btrfs_wait_tree_block_writeback(next);
2682 btrfs_tree_unlock(next);
2684 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2685 clear_extent_buffer_dirty(next);
2688 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2689 ret = btrfs_free_and_pin_reserved_extent(
2691 path->nodes[*level]->start,
2692 path->nodes[*level]->len);
2696 free_extent_buffer(path->nodes[*level]);
2697 path->nodes[*level] = NULL;
2705 * drop the reference count on the tree rooted at 'snap'. This traverses
2706 * the tree freeing any blocks that have a ref count of zero after being
2709 static int walk_log_tree(struct btrfs_trans_handle *trans,
2710 struct btrfs_root *log, struct walk_control *wc)
2712 struct btrfs_fs_info *fs_info = log->fs_info;
2716 struct btrfs_path *path;
2719 path = btrfs_alloc_path();
2723 level = btrfs_header_level(log->node);
2725 path->nodes[level] = log->node;
2726 extent_buffer_get(log->node);
2727 path->slots[level] = 0;
2730 wret = walk_down_log_tree(trans, log, path, &level, wc);
2738 wret = walk_up_log_tree(trans, log, path, &level, wc);
2747 /* was the root node processed? if not, catch it here */
2748 if (path->nodes[orig_level]) {
2749 ret = wc->process_func(log, path->nodes[orig_level], wc,
2750 btrfs_header_generation(path->nodes[orig_level]),
2755 struct extent_buffer *next;
2757 next = path->nodes[orig_level];
2760 btrfs_tree_lock(next);
2761 btrfs_set_lock_blocking(next);
2762 clean_tree_block(fs_info, next);
2763 btrfs_wait_tree_block_writeback(next);
2764 btrfs_tree_unlock(next);
2766 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2767 clear_extent_buffer_dirty(next);
2770 WARN_ON(log->root_key.objectid !=
2771 BTRFS_TREE_LOG_OBJECTID);
2772 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2773 next->start, next->len);
2780 btrfs_free_path(path);
2785 * helper function to update the item for a given subvolumes log root
2786 * in the tree of log roots
2788 static int update_log_root(struct btrfs_trans_handle *trans,
2789 struct btrfs_root *log)
2791 struct btrfs_fs_info *fs_info = log->fs_info;
2794 if (log->log_transid == 1) {
2795 /* insert root item on the first sync */
2796 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2797 &log->root_key, &log->root_item);
2799 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2800 &log->root_key, &log->root_item);
2805 static void wait_log_commit(struct btrfs_root *root, int transid)
2808 int index = transid % 2;
2811 * we only allow two pending log transactions at a time,
2812 * so we know that if ours is more than 2 older than the
2813 * current transaction, we're done
2816 prepare_to_wait(&root->log_commit_wait[index],
2817 &wait, TASK_UNINTERRUPTIBLE);
2819 if (!(root->log_transid_committed < transid &&
2820 atomic_read(&root->log_commit[index])))
2823 mutex_unlock(&root->log_mutex);
2825 mutex_lock(&root->log_mutex);
2827 finish_wait(&root->log_commit_wait[index], &wait);
2830 static void wait_for_writer(struct btrfs_root *root)
2835 prepare_to_wait(&root->log_writer_wait, &wait,
2836 TASK_UNINTERRUPTIBLE);
2837 if (!atomic_read(&root->log_writers))
2840 mutex_unlock(&root->log_mutex);
2842 mutex_lock(&root->log_mutex);
2844 finish_wait(&root->log_writer_wait, &wait);
2847 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2848 struct btrfs_log_ctx *ctx)
2853 mutex_lock(&root->log_mutex);
2854 list_del_init(&ctx->list);
2855 mutex_unlock(&root->log_mutex);
2859 * Invoked in log mutex context, or be sure there is no other task which
2860 * can access the list.
2862 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2863 int index, int error)
2865 struct btrfs_log_ctx *ctx;
2866 struct btrfs_log_ctx *safe;
2868 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2869 list_del_init(&ctx->list);
2870 ctx->log_ret = error;
2873 INIT_LIST_HEAD(&root->log_ctxs[index]);
2877 * btrfs_sync_log does sends a given tree log down to the disk and
2878 * updates the super blocks to record it. When this call is done,
2879 * you know that any inodes previously logged are safely on disk only
2882 * Any other return value means you need to call btrfs_commit_transaction.
2883 * Some of the edge cases for fsyncing directories that have had unlinks
2884 * or renames done in the past mean that sometimes the only safe
2885 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2886 * that has happened.
2888 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2889 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2895 struct btrfs_fs_info *fs_info = root->fs_info;
2896 struct btrfs_root *log = root->log_root;
2897 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2898 int log_transid = 0;
2899 struct btrfs_log_ctx root_log_ctx;
2900 struct blk_plug plug;
2902 mutex_lock(&root->log_mutex);
2903 log_transid = ctx->log_transid;
2904 if (root->log_transid_committed >= log_transid) {
2905 mutex_unlock(&root->log_mutex);
2906 return ctx->log_ret;
2909 index1 = log_transid % 2;
2910 if (atomic_read(&root->log_commit[index1])) {
2911 wait_log_commit(root, log_transid);
2912 mutex_unlock(&root->log_mutex);
2913 return ctx->log_ret;
2915 ASSERT(log_transid == root->log_transid);
2916 atomic_set(&root->log_commit[index1], 1);
2918 /* wait for previous tree log sync to complete */
2919 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2920 wait_log_commit(root, log_transid - 1);
2923 int batch = atomic_read(&root->log_batch);
2924 /* when we're on an ssd, just kick the log commit out */
2925 if (!btrfs_test_opt(fs_info, SSD) &&
2926 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2927 mutex_unlock(&root->log_mutex);
2928 schedule_timeout_uninterruptible(1);
2929 mutex_lock(&root->log_mutex);
2931 wait_for_writer(root);
2932 if (batch == atomic_read(&root->log_batch))
2936 /* bail out if we need to do a full commit */
2937 if (btrfs_need_log_full_commit(fs_info, trans)) {
2939 btrfs_free_logged_extents(log, log_transid);
2940 mutex_unlock(&root->log_mutex);
2944 if (log_transid % 2 == 0)
2945 mark = EXTENT_DIRTY;
2949 /* we start IO on all the marked extents here, but we don't actually
2950 * wait for them until later.
2952 blk_start_plug(&plug);
2953 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2955 blk_finish_plug(&plug);
2956 btrfs_abort_transaction(trans, ret);
2957 btrfs_free_logged_extents(log, log_transid);
2958 btrfs_set_log_full_commit(fs_info, trans);
2959 mutex_unlock(&root->log_mutex);
2963 btrfs_set_root_node(&log->root_item, log->node);
2965 root->log_transid++;
2966 log->log_transid = root->log_transid;
2967 root->log_start_pid = 0;
2969 * IO has been started, blocks of the log tree have WRITTEN flag set
2970 * in their headers. new modifications of the log will be written to
2971 * new positions. so it's safe to allow log writers to go in.
2973 mutex_unlock(&root->log_mutex);
2975 btrfs_init_log_ctx(&root_log_ctx, NULL);
2977 mutex_lock(&log_root_tree->log_mutex);
2978 atomic_inc(&log_root_tree->log_batch);
2979 atomic_inc(&log_root_tree->log_writers);
2981 index2 = log_root_tree->log_transid % 2;
2982 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2983 root_log_ctx.log_transid = log_root_tree->log_transid;
2985 mutex_unlock(&log_root_tree->log_mutex);
2987 ret = update_log_root(trans, log);
2989 mutex_lock(&log_root_tree->log_mutex);
2990 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2992 * Implicit memory barrier after atomic_dec_and_test
2994 if (waitqueue_active(&log_root_tree->log_writer_wait))
2995 wake_up(&log_root_tree->log_writer_wait);
2999 if (!list_empty(&root_log_ctx.list))
3000 list_del_init(&root_log_ctx.list);
3002 blk_finish_plug(&plug);
3003 btrfs_set_log_full_commit(fs_info, trans);
3005 if (ret != -ENOSPC) {
3006 btrfs_abort_transaction(trans, ret);
3007 mutex_unlock(&log_root_tree->log_mutex);
3010 btrfs_wait_tree_log_extents(log, mark);
3011 btrfs_free_logged_extents(log, log_transid);
3012 mutex_unlock(&log_root_tree->log_mutex);
3017 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3018 blk_finish_plug(&plug);
3019 list_del_init(&root_log_ctx.list);
3020 mutex_unlock(&log_root_tree->log_mutex);
3021 ret = root_log_ctx.log_ret;
3025 index2 = root_log_ctx.log_transid % 2;
3026 if (atomic_read(&log_root_tree->log_commit[index2])) {
3027 blk_finish_plug(&plug);
3028 ret = btrfs_wait_tree_log_extents(log, mark);
3029 btrfs_wait_logged_extents(trans, log, log_transid);
3030 wait_log_commit(log_root_tree,
3031 root_log_ctx.log_transid);
3032 mutex_unlock(&log_root_tree->log_mutex);
3034 ret = root_log_ctx.log_ret;
3037 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3038 atomic_set(&log_root_tree->log_commit[index2], 1);
3040 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3041 wait_log_commit(log_root_tree,
3042 root_log_ctx.log_transid - 1);
3045 wait_for_writer(log_root_tree);
3048 * now that we've moved on to the tree of log tree roots,
3049 * check the full commit flag again
3051 if (btrfs_need_log_full_commit(fs_info, trans)) {
3052 blk_finish_plug(&plug);
3053 btrfs_wait_tree_log_extents(log, mark);
3054 btrfs_free_logged_extents(log, log_transid);
3055 mutex_unlock(&log_root_tree->log_mutex);
3057 goto out_wake_log_root;
3060 ret = btrfs_write_marked_extents(fs_info,
3061 &log_root_tree->dirty_log_pages,
3062 EXTENT_DIRTY | EXTENT_NEW);
3063 blk_finish_plug(&plug);
3065 btrfs_set_log_full_commit(fs_info, trans);
3066 btrfs_abort_transaction(trans, ret);
3067 btrfs_free_logged_extents(log, log_transid);
3068 mutex_unlock(&log_root_tree->log_mutex);
3069 goto out_wake_log_root;
3071 ret = btrfs_wait_tree_log_extents(log, mark);
3073 ret = btrfs_wait_tree_log_extents(log_root_tree,
3074 EXTENT_NEW | EXTENT_DIRTY);
3076 btrfs_set_log_full_commit(fs_info, trans);
3077 btrfs_free_logged_extents(log, log_transid);
3078 mutex_unlock(&log_root_tree->log_mutex);
3079 goto out_wake_log_root;
3081 btrfs_wait_logged_extents(trans, log, log_transid);
3083 btrfs_set_super_log_root(fs_info->super_for_commit,
3084 log_root_tree->node->start);
3085 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3086 btrfs_header_level(log_root_tree->node));
3088 log_root_tree->log_transid++;
3089 mutex_unlock(&log_root_tree->log_mutex);
3092 * nobody else is going to jump in and write the the ctree
3093 * super here because the log_commit atomic below is protecting
3094 * us. We must be called with a transaction handle pinning
3095 * the running transaction open, so a full commit can't hop
3096 * in and cause problems either.
3098 ret = write_all_supers(fs_info, 1);
3100 btrfs_set_log_full_commit(fs_info, trans);
3101 btrfs_abort_transaction(trans, ret);
3102 goto out_wake_log_root;
3105 mutex_lock(&root->log_mutex);
3106 if (root->last_log_commit < log_transid)
3107 root->last_log_commit = log_transid;
3108 mutex_unlock(&root->log_mutex);
3111 mutex_lock(&log_root_tree->log_mutex);
3112 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3114 log_root_tree->log_transid_committed++;
3115 atomic_set(&log_root_tree->log_commit[index2], 0);
3116 mutex_unlock(&log_root_tree->log_mutex);
3119 * The barrier before waitqueue_active is implied by mutex_unlock
3121 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
3122 wake_up(&log_root_tree->log_commit_wait[index2]);
3124 mutex_lock(&root->log_mutex);
3125 btrfs_remove_all_log_ctxs(root, index1, ret);
3126 root->log_transid_committed++;
3127 atomic_set(&root->log_commit[index1], 0);
3128 mutex_unlock(&root->log_mutex);
3131 * The barrier before waitqueue_active is implied by mutex_unlock
3133 if (waitqueue_active(&root->log_commit_wait[index1]))
3134 wake_up(&root->log_commit_wait[index1]);
3138 static void free_log_tree(struct btrfs_trans_handle *trans,
3139 struct btrfs_root *log)
3144 struct walk_control wc = {
3146 .process_func = process_one_buffer
3149 ret = walk_log_tree(trans, log, &wc);
3150 /* I don't think this can happen but just in case */
3152 btrfs_abort_transaction(trans, ret);
3155 ret = find_first_extent_bit(&log->dirty_log_pages,
3157 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT,
3162 clear_extent_bits(&log->dirty_log_pages, start, end,
3163 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3167 * We may have short-circuited the log tree with the full commit logic
3168 * and left ordered extents on our list, so clear these out to keep us
3169 * from leaking inodes and memory.
3171 btrfs_free_logged_extents(log, 0);
3172 btrfs_free_logged_extents(log, 1);
3174 free_extent_buffer(log->node);
3179 * free all the extents used by the tree log. This should be called
3180 * at commit time of the full transaction
3182 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3184 if (root->log_root) {
3185 free_log_tree(trans, root->log_root);
3186 root->log_root = NULL;
3191 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3192 struct btrfs_fs_info *fs_info)
3194 if (fs_info->log_root_tree) {
3195 free_log_tree(trans, fs_info->log_root_tree);
3196 fs_info->log_root_tree = NULL;
3202 * If both a file and directory are logged, and unlinks or renames are
3203 * mixed in, we have a few interesting corners:
3205 * create file X in dir Y
3206 * link file X to X.link in dir Y
3208 * unlink file X but leave X.link
3211 * After a crash we would expect only X.link to exist. But file X
3212 * didn't get fsync'd again so the log has back refs for X and X.link.
3214 * We solve this by removing directory entries and inode backrefs from the
3215 * log when a file that was logged in the current transaction is
3216 * unlinked. Any later fsync will include the updated log entries, and
3217 * we'll be able to reconstruct the proper directory items from backrefs.
3219 * This optimizations allows us to avoid relogging the entire inode
3220 * or the entire directory.
3222 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3223 struct btrfs_root *root,
3224 const char *name, int name_len,
3225 struct btrfs_inode *dir, u64 index)
3227 struct btrfs_root *log;
3228 struct btrfs_dir_item *di;
3229 struct btrfs_path *path;
3233 u64 dir_ino = btrfs_ino(dir);
3235 if (dir->logged_trans < trans->transid)
3238 ret = join_running_log_trans(root);
3242 mutex_lock(&dir->log_mutex);
3244 log = root->log_root;
3245 path = btrfs_alloc_path();
3251 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3252 name, name_len, -1);
3258 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3259 bytes_del += name_len;
3265 btrfs_release_path(path);
3266 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3267 index, name, name_len, -1);
3273 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3274 bytes_del += name_len;
3281 /* update the directory size in the log to reflect the names
3285 struct btrfs_key key;
3287 key.objectid = dir_ino;
3289 key.type = BTRFS_INODE_ITEM_KEY;
3290 btrfs_release_path(path);
3292 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3298 struct btrfs_inode_item *item;
3301 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3302 struct btrfs_inode_item);
3303 i_size = btrfs_inode_size(path->nodes[0], item);
3304 if (i_size > bytes_del)
3305 i_size -= bytes_del;
3308 btrfs_set_inode_size(path->nodes[0], item, i_size);
3309 btrfs_mark_buffer_dirty(path->nodes[0]);
3312 btrfs_release_path(path);
3315 btrfs_free_path(path);
3317 mutex_unlock(&dir->log_mutex);
3318 if (ret == -ENOSPC) {
3319 btrfs_set_log_full_commit(root->fs_info, trans);
3322 btrfs_abort_transaction(trans, ret);
3324 btrfs_end_log_trans(root);
3329 /* see comments for btrfs_del_dir_entries_in_log */
3330 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3331 struct btrfs_root *root,
3332 const char *name, int name_len,
3333 struct btrfs_inode *inode, u64 dirid)
3335 struct btrfs_fs_info *fs_info = root->fs_info;
3336 struct btrfs_root *log;
3340 if (inode->logged_trans < trans->transid)
3343 ret = join_running_log_trans(root);
3346 log = root->log_root;
3347 mutex_lock(&inode->log_mutex);
3349 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3351 mutex_unlock(&inode->log_mutex);
3352 if (ret == -ENOSPC) {
3353 btrfs_set_log_full_commit(fs_info, trans);
3355 } else if (ret < 0 && ret != -ENOENT)
3356 btrfs_abort_transaction(trans, ret);
3357 btrfs_end_log_trans(root);
3363 * creates a range item in the log for 'dirid'. first_offset and
3364 * last_offset tell us which parts of the key space the log should
3365 * be considered authoritative for.
3367 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3368 struct btrfs_root *log,
3369 struct btrfs_path *path,
3370 int key_type, u64 dirid,
3371 u64 first_offset, u64 last_offset)
3374 struct btrfs_key key;
3375 struct btrfs_dir_log_item *item;
3377 key.objectid = dirid;
3378 key.offset = first_offset;
3379 if (key_type == BTRFS_DIR_ITEM_KEY)
3380 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3382 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3383 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3387 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3388 struct btrfs_dir_log_item);
3389 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3390 btrfs_mark_buffer_dirty(path->nodes[0]);
3391 btrfs_release_path(path);
3396 * log all the items included in the current transaction for a given
3397 * directory. This also creates the range items in the log tree required
3398 * to replay anything deleted before the fsync
3400 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3401 struct btrfs_root *root, struct btrfs_inode *inode,
3402 struct btrfs_path *path,
3403 struct btrfs_path *dst_path, int key_type,
3404 struct btrfs_log_ctx *ctx,
3405 u64 min_offset, u64 *last_offset_ret)
3407 struct btrfs_key min_key;
3408 struct btrfs_root *log = root->log_root;
3409 struct extent_buffer *src;
3414 u64 first_offset = min_offset;
3415 u64 last_offset = (u64)-1;
3416 u64 ino = btrfs_ino(inode);
3418 log = root->log_root;
3420 min_key.objectid = ino;
3421 min_key.type = key_type;
3422 min_key.offset = min_offset;
3424 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3427 * we didn't find anything from this transaction, see if there
3428 * is anything at all
3430 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3431 min_key.objectid = ino;
3432 min_key.type = key_type;
3433 min_key.offset = (u64)-1;
3434 btrfs_release_path(path);
3435 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3437 btrfs_release_path(path);
3440 ret = btrfs_previous_item(root, path, ino, key_type);
3442 /* if ret == 0 there are items for this type,
3443 * create a range to tell us the last key of this type.
3444 * otherwise, there are no items in this directory after
3445 * *min_offset, and we create a range to indicate that.
3448 struct btrfs_key tmp;
3449 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3451 if (key_type == tmp.type)
3452 first_offset = max(min_offset, tmp.offset) + 1;
3457 /* go backward to find any previous key */
3458 ret = btrfs_previous_item(root, path, ino, key_type);
3460 struct btrfs_key tmp;
3461 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3462 if (key_type == tmp.type) {
3463 first_offset = tmp.offset;
3464 ret = overwrite_item(trans, log, dst_path,
3465 path->nodes[0], path->slots[0],
3473 btrfs_release_path(path);
3475 /* find the first key from this transaction again */
3476 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3477 if (WARN_ON(ret != 0))
3481 * we have a block from this transaction, log every item in it
3482 * from our directory
3485 struct btrfs_key tmp;
3486 src = path->nodes[0];
3487 nritems = btrfs_header_nritems(src);
3488 for (i = path->slots[0]; i < nritems; i++) {
3489 struct btrfs_dir_item *di;
3491 btrfs_item_key_to_cpu(src, &min_key, i);
3493 if (min_key.objectid != ino || min_key.type != key_type)
3495 ret = overwrite_item(trans, log, dst_path, src, i,
3503 * We must make sure that when we log a directory entry,
3504 * the corresponding inode, after log replay, has a
3505 * matching link count. For example:
3511 * xfs_io -c "fsync" mydir
3513 * <mount fs and log replay>
3515 * Would result in a fsync log that when replayed, our
3516 * file inode would have a link count of 1, but we get
3517 * two directory entries pointing to the same inode.
3518 * After removing one of the names, it would not be
3519 * possible to remove the other name, which resulted
3520 * always in stale file handle errors, and would not
3521 * be possible to rmdir the parent directory, since
3522 * its i_size could never decrement to the value
3523 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3525 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3526 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3528 (btrfs_dir_transid(src, di) == trans->transid ||
3529 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3530 tmp.type != BTRFS_ROOT_ITEM_KEY)
3531 ctx->log_new_dentries = true;
3533 path->slots[0] = nritems;
3536 * look ahead to the next item and see if it is also
3537 * from this directory and from this transaction
3539 ret = btrfs_next_leaf(root, path);
3542 last_offset = (u64)-1;
3547 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3548 if (tmp.objectid != ino || tmp.type != key_type) {
3549 last_offset = (u64)-1;
3552 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3553 ret = overwrite_item(trans, log, dst_path,
3554 path->nodes[0], path->slots[0],
3559 last_offset = tmp.offset;
3564 btrfs_release_path(path);
3565 btrfs_release_path(dst_path);
3568 *last_offset_ret = last_offset;
3570 * insert the log range keys to indicate where the log
3573 ret = insert_dir_log_key(trans, log, path, key_type,
3574 ino, first_offset, last_offset);
3582 * logging directories is very similar to logging inodes, We find all the items
3583 * from the current transaction and write them to the log.
3585 * The recovery code scans the directory in the subvolume, and if it finds a
3586 * key in the range logged that is not present in the log tree, then it means
3587 * that dir entry was unlinked during the transaction.
3589 * In order for that scan to work, we must include one key smaller than
3590 * the smallest logged by this transaction and one key larger than the largest
3591 * key logged by this transaction.
3593 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3594 struct btrfs_root *root, struct btrfs_inode *inode,
3595 struct btrfs_path *path,
3596 struct btrfs_path *dst_path,
3597 struct btrfs_log_ctx *ctx)
3602 int key_type = BTRFS_DIR_ITEM_KEY;
3608 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3609 ctx, min_key, &max_key);
3612 if (max_key == (u64)-1)
3614 min_key = max_key + 1;
3617 if (key_type == BTRFS_DIR_ITEM_KEY) {
3618 key_type = BTRFS_DIR_INDEX_KEY;
3625 * a helper function to drop items from the log before we relog an
3626 * inode. max_key_type indicates the highest item type to remove.
3627 * This cannot be run for file data extents because it does not
3628 * free the extents they point to.
3630 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3631 struct btrfs_root *log,
3632 struct btrfs_path *path,
3633 u64 objectid, int max_key_type)
3636 struct btrfs_key key;
3637 struct btrfs_key found_key;
3640 key.objectid = objectid;
3641 key.type = max_key_type;
3642 key.offset = (u64)-1;
3645 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3646 BUG_ON(ret == 0); /* Logic error */
3650 if (path->slots[0] == 0)
3654 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3657 if (found_key.objectid != objectid)
3660 found_key.offset = 0;
3662 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3665 ret = btrfs_del_items(trans, log, path, start_slot,
3666 path->slots[0] - start_slot + 1);
3668 * If start slot isn't 0 then we don't need to re-search, we've
3669 * found the last guy with the objectid in this tree.
3671 if (ret || start_slot != 0)
3673 btrfs_release_path(path);
3675 btrfs_release_path(path);
3681 static void fill_inode_item(struct btrfs_trans_handle *trans,
3682 struct extent_buffer *leaf,
3683 struct btrfs_inode_item *item,
3684 struct inode *inode, int log_inode_only,
3687 struct btrfs_map_token token;
3689 btrfs_init_map_token(&token);
3691 if (log_inode_only) {
3692 /* set the generation to zero so the recover code
3693 * can tell the difference between an logging
3694 * just to say 'this inode exists' and a logging
3695 * to say 'update this inode with these values'
3697 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3698 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3700 btrfs_set_token_inode_generation(leaf, item,
3701 BTRFS_I(inode)->generation,
3703 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3706 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3707 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3708 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3709 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3711 btrfs_set_token_timespec_sec(leaf, &item->atime,
3712 inode->i_atime.tv_sec, &token);
3713 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3714 inode->i_atime.tv_nsec, &token);
3716 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3717 inode->i_mtime.tv_sec, &token);
3718 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3719 inode->i_mtime.tv_nsec, &token);
3721 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3722 inode->i_ctime.tv_sec, &token);
3723 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3724 inode->i_ctime.tv_nsec, &token);
3726 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3729 btrfs_set_token_inode_sequence(leaf, item,
3730 inode_peek_iversion(inode), &token);
3731 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3732 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3733 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3734 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3737 static int log_inode_item(struct btrfs_trans_handle *trans,
3738 struct btrfs_root *log, struct btrfs_path *path,
3739 struct btrfs_inode *inode)
3741 struct btrfs_inode_item *inode_item;
3744 ret = btrfs_insert_empty_item(trans, log, path,
3745 &inode->location, sizeof(*inode_item));
3746 if (ret && ret != -EEXIST)
3748 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3749 struct btrfs_inode_item);
3750 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3752 btrfs_release_path(path);
3756 static noinline int copy_items(struct btrfs_trans_handle *trans,
3757 struct btrfs_inode *inode,
3758 struct btrfs_path *dst_path,
3759 struct btrfs_path *src_path, u64 *last_extent,
3760 int start_slot, int nr, int inode_only,
3763 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3764 unsigned long src_offset;
3765 unsigned long dst_offset;
3766 struct btrfs_root *log = inode->root->log_root;
3767 struct btrfs_file_extent_item *extent;
3768 struct btrfs_inode_item *inode_item;
3769 struct extent_buffer *src = src_path->nodes[0];
3770 struct btrfs_key first_key, last_key, key;
3772 struct btrfs_key *ins_keys;
3776 struct list_head ordered_sums;
3777 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3778 bool has_extents = false;
3779 bool need_find_last_extent = true;
3782 INIT_LIST_HEAD(&ordered_sums);
3784 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3785 nr * sizeof(u32), GFP_NOFS);
3789 first_key.objectid = (u64)-1;
3791 ins_sizes = (u32 *)ins_data;
3792 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3794 for (i = 0; i < nr; i++) {
3795 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3796 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3798 ret = btrfs_insert_empty_items(trans, log, dst_path,
3799 ins_keys, ins_sizes, nr);
3805 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3806 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3807 dst_path->slots[0]);
3809 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3812 last_key = ins_keys[i];
3814 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3815 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3817 struct btrfs_inode_item);
3818 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3820 inode_only == LOG_INODE_EXISTS,
3823 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3824 src_offset, ins_sizes[i]);
3828 * We set need_find_last_extent here in case we know we were
3829 * processing other items and then walk into the first extent in
3830 * the inode. If we don't hit an extent then nothing changes,
3831 * we'll do the last search the next time around.
3833 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3835 if (first_key.objectid == (u64)-1)
3836 first_key = ins_keys[i];
3838 need_find_last_extent = false;
3841 /* take a reference on file data extents so that truncates
3842 * or deletes of this inode don't have to relog the inode
3845 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3848 extent = btrfs_item_ptr(src, start_slot + i,
3849 struct btrfs_file_extent_item);
3851 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3854 found_type = btrfs_file_extent_type(src, extent);
3855 if (found_type == BTRFS_FILE_EXTENT_REG) {
3857 ds = btrfs_file_extent_disk_bytenr(src,
3859 /* ds == 0 is a hole */
3863 dl = btrfs_file_extent_disk_num_bytes(src,
3865 cs = btrfs_file_extent_offset(src, extent);
3866 cl = btrfs_file_extent_num_bytes(src,
3868 if (btrfs_file_extent_compression(src,
3874 ret = btrfs_lookup_csums_range(
3876 ds + cs, ds + cs + cl - 1,
3879 btrfs_release_path(dst_path);
3887 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3888 btrfs_release_path(dst_path);
3892 * we have to do this after the loop above to avoid changing the
3893 * log tree while trying to change the log tree.
3896 while (!list_empty(&ordered_sums)) {
3897 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3898 struct btrfs_ordered_sum,
3901 ret = btrfs_csum_file_blocks(trans, log, sums);
3902 list_del(&sums->list);
3909 if (need_find_last_extent && *last_extent == first_key.offset) {
3911 * We don't have any leafs between our current one and the one
3912 * we processed before that can have file extent items for our
3913 * inode (and have a generation number smaller than our current
3916 need_find_last_extent = false;
3920 * Because we use btrfs_search_forward we could skip leaves that were
3921 * not modified and then assume *last_extent is valid when it really
3922 * isn't. So back up to the previous leaf and read the end of the last
3923 * extent before we go and fill in holes.
3925 if (need_find_last_extent) {
3928 ret = btrfs_prev_leaf(inode->root, src_path);
3933 if (src_path->slots[0])
3934 src_path->slots[0]--;
3935 src = src_path->nodes[0];
3936 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3937 if (key.objectid != btrfs_ino(inode) ||
3938 key.type != BTRFS_EXTENT_DATA_KEY)
3940 extent = btrfs_item_ptr(src, src_path->slots[0],
3941 struct btrfs_file_extent_item);
3942 if (btrfs_file_extent_type(src, extent) ==
3943 BTRFS_FILE_EXTENT_INLINE) {
3944 len = btrfs_file_extent_inline_len(src,
3947 *last_extent = ALIGN(key.offset + len,
3948 fs_info->sectorsize);
3950 len = btrfs_file_extent_num_bytes(src, extent);
3951 *last_extent = key.offset + len;
3955 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3956 * things could have happened
3958 * 1) A merge could have happened, so we could currently be on a leaf
3959 * that holds what we were copying in the first place.
3960 * 2) A split could have happened, and now not all of the items we want
3961 * are on the same leaf.
3963 * So we need to adjust how we search for holes, we need to drop the
3964 * path and re-search for the first extent key we found, and then walk
3965 * forward until we hit the last one we copied.
3967 if (need_find_last_extent) {
3968 /* btrfs_prev_leaf could return 1 without releasing the path */
3969 btrfs_release_path(src_path);
3970 ret = btrfs_search_slot(NULL, inode->root, &first_key,
3975 src = src_path->nodes[0];
3976 i = src_path->slots[0];
3982 * Ok so here we need to go through and fill in any holes we may have
3983 * to make sure that holes are punched for those areas in case they had
3984 * extents previously.
3990 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3991 ret = btrfs_next_leaf(inode->root, src_path);
3995 src = src_path->nodes[0];
3997 need_find_last_extent = true;
4000 btrfs_item_key_to_cpu(src, &key, i);
4001 if (!btrfs_comp_cpu_keys(&key, &last_key))
4003 if (key.objectid != btrfs_ino(inode) ||
4004 key.type != BTRFS_EXTENT_DATA_KEY) {
4008 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
4009 if (btrfs_file_extent_type(src, extent) ==
4010 BTRFS_FILE_EXTENT_INLINE) {
4011 len = btrfs_file_extent_inline_len(src, i, extent);
4012 extent_end = ALIGN(key.offset + len,
4013 fs_info->sectorsize);
4015 len = btrfs_file_extent_num_bytes(src, extent);
4016 extent_end = key.offset + len;
4020 if (*last_extent == key.offset) {
4021 *last_extent = extent_end;
4024 offset = *last_extent;
4025 len = key.offset - *last_extent;
4026 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
4027 offset, 0, 0, len, 0, len, 0, 0, 0);
4030 *last_extent = extent_end;
4034 * Check if there is a hole between the last extent found in our leaf
4035 * and the first extent in the next leaf. If there is one, we need to
4036 * log an explicit hole so that at replay time we can punch the hole.
4039 key.objectid == btrfs_ino(inode) &&
4040 key.type == BTRFS_EXTENT_DATA_KEY &&
4041 i == btrfs_header_nritems(src_path->nodes[0])) {
4042 ret = btrfs_next_leaf(inode->root, src_path);
4043 need_find_last_extent = true;
4046 } else if (ret == 0) {
4047 btrfs_item_key_to_cpu(src_path->nodes[0], &key,
4048 src_path->slots[0]);
4049 if (key.objectid == btrfs_ino(inode) &&
4050 key.type == BTRFS_EXTENT_DATA_KEY &&
4051 *last_extent < key.offset) {
4052 const u64 len = key.offset - *last_extent;
4054 ret = btrfs_insert_file_extent(trans, log,
4063 * Need to let the callers know we dropped the path so they should
4066 if (!ret && need_find_last_extent)
4071 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4073 struct extent_map *em1, *em2;
4075 em1 = list_entry(a, struct extent_map, list);
4076 em2 = list_entry(b, struct extent_map, list);
4078 if (em1->start < em2->start)
4080 else if (em1->start > em2->start)
4085 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
4086 struct inode *inode,
4087 struct btrfs_root *root,
4088 const struct extent_map *em,
4089 const struct list_head *logged_list,
4090 bool *ordered_io_error)
4092 struct btrfs_fs_info *fs_info = root->fs_info;
4093 struct btrfs_ordered_extent *ordered;
4094 struct btrfs_root *log = root->log_root;
4095 u64 mod_start = em->mod_start;
4096 u64 mod_len = em->mod_len;
4097 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
4100 LIST_HEAD(ordered_sums);
4103 *ordered_io_error = false;
4105 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4106 em->block_start == EXTENT_MAP_HOLE)
4110 * Wait far any ordered extent that covers our extent map. If it
4111 * finishes without an error, first check and see if our csums are on
4112 * our outstanding ordered extents.
4114 list_for_each_entry(ordered, logged_list, log_list) {
4115 struct btrfs_ordered_sum *sum;
4120 if (ordered->file_offset + ordered->len <= mod_start ||
4121 mod_start + mod_len <= ordered->file_offset)
4124 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
4125 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
4126 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
4127 const u64 start = ordered->file_offset;
4128 const u64 end = ordered->file_offset + ordered->len - 1;
4130 WARN_ON(ordered->inode != inode);
4131 filemap_fdatawrite_range(inode->i_mapping, start, end);
4134 wait_event(ordered->wait,
4135 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
4136 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
4138 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
4140 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
4141 * i_mapping flags, so that the next fsync won't get
4142 * an outdated io error too.
4144 filemap_check_errors(inode->i_mapping);
4145 *ordered_io_error = true;
4149 * We are going to copy all the csums on this ordered extent, so
4150 * go ahead and adjust mod_start and mod_len in case this
4151 * ordered extent has already been logged.
4153 if (ordered->file_offset > mod_start) {
4154 if (ordered->file_offset + ordered->len >=
4155 mod_start + mod_len)
4156 mod_len = ordered->file_offset - mod_start;
4158 * If we have this case
4160 * |--------- logged extent ---------|
4161 * |----- ordered extent ----|
4163 * Just don't mess with mod_start and mod_len, we'll
4164 * just end up logging more csums than we need and it
4168 if (ordered->file_offset + ordered->len <
4169 mod_start + mod_len) {
4170 mod_len = (mod_start + mod_len) -
4171 (ordered->file_offset + ordered->len);
4172 mod_start = ordered->file_offset +
4183 * To keep us from looping for the above case of an ordered
4184 * extent that falls inside of the logged extent.
4186 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
4190 list_for_each_entry(sum, &ordered->list, list) {
4191 ret = btrfs_csum_file_blocks(trans, log, sum);
4197 if (*ordered_io_error || !mod_len || ret || skip_csum)
4200 if (em->compress_type) {
4202 csum_len = max(em->block_len, em->orig_block_len);
4204 csum_offset = mod_start - em->start;
4208 /* block start is already adjusted for the file extent offset. */
4209 ret = btrfs_lookup_csums_range(fs_info->csum_root,
4210 em->block_start + csum_offset,
4211 em->block_start + csum_offset +
4212 csum_len - 1, &ordered_sums, 0);
4216 while (!list_empty(&ordered_sums)) {
4217 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4218 struct btrfs_ordered_sum,
4221 ret = btrfs_csum_file_blocks(trans, log, sums);
4222 list_del(&sums->list);
4229 static int log_one_extent(struct btrfs_trans_handle *trans,
4230 struct btrfs_inode *inode, struct btrfs_root *root,
4231 const struct extent_map *em,
4232 struct btrfs_path *path,
4233 const struct list_head *logged_list,
4234 struct btrfs_log_ctx *ctx)
4236 struct btrfs_root *log = root->log_root;
4237 struct btrfs_file_extent_item *fi;
4238 struct extent_buffer *leaf;
4239 struct btrfs_map_token token;
4240 struct btrfs_key key;
4241 u64 extent_offset = em->start - em->orig_start;
4244 int extent_inserted = 0;
4245 bool ordered_io_err = false;
4247 ret = wait_ordered_extents(trans, &inode->vfs_inode, root, em,
4248 logged_list, &ordered_io_err);
4252 if (ordered_io_err) {
4257 btrfs_init_map_token(&token);
4259 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4260 em->start + em->len, NULL, 0, 1,
4261 sizeof(*fi), &extent_inserted);
4265 if (!extent_inserted) {
4266 key.objectid = btrfs_ino(inode);
4267 key.type = BTRFS_EXTENT_DATA_KEY;
4268 key.offset = em->start;
4270 ret = btrfs_insert_empty_item(trans, log, path, &key,
4275 leaf = path->nodes[0];
4276 fi = btrfs_item_ptr(leaf, path->slots[0],
4277 struct btrfs_file_extent_item);
4279 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4281 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4282 btrfs_set_token_file_extent_type(leaf, fi,
4283 BTRFS_FILE_EXTENT_PREALLOC,
4286 btrfs_set_token_file_extent_type(leaf, fi,
4287 BTRFS_FILE_EXTENT_REG,
4290 block_len = max(em->block_len, em->orig_block_len);
4291 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4292 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4295 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4297 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4298 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4300 extent_offset, &token);
4301 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4304 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4305 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4309 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4310 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4311 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4312 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4314 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4315 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4316 btrfs_mark_buffer_dirty(leaf);
4318 btrfs_release_path(path);
4323 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4324 struct btrfs_root *root,
4325 struct btrfs_inode *inode,
4326 struct btrfs_path *path,
4327 struct list_head *logged_list,
4328 struct btrfs_log_ctx *ctx,
4332 struct extent_map *em, *n;
4333 struct list_head extents;
4334 struct extent_map_tree *tree = &inode->extent_tree;
4335 u64 logged_start, logged_end;
4340 INIT_LIST_HEAD(&extents);
4342 down_write(&inode->dio_sem);
4343 write_lock(&tree->lock);
4344 test_gen = root->fs_info->last_trans_committed;
4345 logged_start = start;
4348 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4349 list_del_init(&em->list);
4351 * Just an arbitrary number, this can be really CPU intensive
4352 * once we start getting a lot of extents, and really once we
4353 * have a bunch of extents we just want to commit since it will
4356 if (++num > 32768) {
4357 list_del_init(&tree->modified_extents);
4362 if (em->generation <= test_gen)
4365 if (em->start < logged_start)
4366 logged_start = em->start;
4367 if ((em->start + em->len - 1) > logged_end)
4368 logged_end = em->start + em->len - 1;
4370 /* Need a ref to keep it from getting evicted from cache */
4371 refcount_inc(&em->refs);
4372 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4373 list_add_tail(&em->list, &extents);
4378 * Add all prealloc extents beyond the inode's i_size to make sure we
4379 * don't lose them after doing a fast fsync and replaying the log.
4381 if (inode->flags & BTRFS_INODE_PREALLOC) {
4382 struct rb_node *node;
4384 for (node = rb_last(&tree->map); node; node = rb_prev(node)) {
4385 em = rb_entry(node, struct extent_map, rb_node);
4386 if (em->start < i_size_read(&inode->vfs_inode))
4388 if (!list_empty(&em->list))
4390 /* Same as above loop. */
4391 if (++num > 32768) {
4392 list_del_init(&tree->modified_extents);
4396 refcount_inc(&em->refs);
4397 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4398 list_add_tail(&em->list, &extents);
4402 list_sort(NULL, &extents, extent_cmp);
4403 btrfs_get_logged_extents(inode, logged_list, logged_start, logged_end);
4405 * Some ordered extents started by fsync might have completed
4406 * before we could collect them into the list logged_list, which
4407 * means they're gone, not in our logged_list nor in the inode's
4408 * ordered tree. We want the application/user space to know an
4409 * error happened while attempting to persist file data so that
4410 * it can take proper action. If such error happened, we leave
4411 * without writing to the log tree and the fsync must report the
4412 * file data write error and not commit the current transaction.
4414 ret = filemap_check_errors(inode->vfs_inode.i_mapping);
4418 while (!list_empty(&extents)) {
4419 em = list_entry(extents.next, struct extent_map, list);
4421 list_del_init(&em->list);
4424 * If we had an error we just need to delete everybody from our
4428 clear_em_logging(tree, em);
4429 free_extent_map(em);
4433 write_unlock(&tree->lock);
4435 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4437 write_lock(&tree->lock);
4438 clear_em_logging(tree, em);
4439 free_extent_map(em);
4441 WARN_ON(!list_empty(&extents));
4442 write_unlock(&tree->lock);
4443 up_write(&inode->dio_sem);
4445 btrfs_release_path(path);
4449 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4450 struct btrfs_path *path, u64 *size_ret)
4452 struct btrfs_key key;
4455 key.objectid = btrfs_ino(inode);
4456 key.type = BTRFS_INODE_ITEM_KEY;
4459 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4462 } else if (ret > 0) {
4465 struct btrfs_inode_item *item;
4467 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4468 struct btrfs_inode_item);
4469 *size_ret = btrfs_inode_size(path->nodes[0], item);
4472 btrfs_release_path(path);
4477 * At the moment we always log all xattrs. This is to figure out at log replay
4478 * time which xattrs must have their deletion replayed. If a xattr is missing
4479 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4480 * because if a xattr is deleted, the inode is fsynced and a power failure
4481 * happens, causing the log to be replayed the next time the fs is mounted,
4482 * we want the xattr to not exist anymore (same behaviour as other filesystems
4483 * with a journal, ext3/4, xfs, f2fs, etc).
4485 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4486 struct btrfs_root *root,
4487 struct btrfs_inode *inode,
4488 struct btrfs_path *path,
4489 struct btrfs_path *dst_path)
4492 struct btrfs_key key;
4493 const u64 ino = btrfs_ino(inode);
4498 key.type = BTRFS_XATTR_ITEM_KEY;
4501 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4506 int slot = path->slots[0];
4507 struct extent_buffer *leaf = path->nodes[0];
4508 int nritems = btrfs_header_nritems(leaf);
4510 if (slot >= nritems) {
4512 u64 last_extent = 0;
4514 ret = copy_items(trans, inode, dst_path, path,
4515 &last_extent, start_slot,
4517 /* can't be 1, extent items aren't processed */
4523 ret = btrfs_next_leaf(root, path);
4531 btrfs_item_key_to_cpu(leaf, &key, slot);
4532 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4542 u64 last_extent = 0;
4544 ret = copy_items(trans, inode, dst_path, path,
4545 &last_extent, start_slot,
4547 /* can't be 1, extent items aren't processed */
4557 * If the no holes feature is enabled we need to make sure any hole between the
4558 * last extent and the i_size of our inode is explicitly marked in the log. This
4559 * is to make sure that doing something like:
4561 * 1) create file with 128Kb of data
4562 * 2) truncate file to 64Kb
4563 * 3) truncate file to 256Kb
4565 * 5) <crash/power failure>
4566 * 6) mount fs and trigger log replay
4568 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4569 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4570 * file correspond to a hole. The presence of explicit holes in a log tree is
4571 * what guarantees that log replay will remove/adjust file extent items in the
4574 * Here we do not need to care about holes between extents, that is already done
4575 * by copy_items(). We also only need to do this in the full sync path, where we
4576 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4577 * lookup the list of modified extent maps and if any represents a hole, we
4578 * insert a corresponding extent representing a hole in the log tree.
4580 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4581 struct btrfs_root *root,
4582 struct btrfs_inode *inode,
4583 struct btrfs_path *path)
4585 struct btrfs_fs_info *fs_info = root->fs_info;
4587 struct btrfs_key key;
4590 struct extent_buffer *leaf;
4591 struct btrfs_root *log = root->log_root;
4592 const u64 ino = btrfs_ino(inode);
4593 const u64 i_size = i_size_read(&inode->vfs_inode);
4595 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
4599 key.type = BTRFS_EXTENT_DATA_KEY;
4600 key.offset = (u64)-1;
4602 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4607 ASSERT(path->slots[0] > 0);
4609 leaf = path->nodes[0];
4610 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4612 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4613 /* inode does not have any extents */
4617 struct btrfs_file_extent_item *extent;
4621 * If there's an extent beyond i_size, an explicit hole was
4622 * already inserted by copy_items().
4624 if (key.offset >= i_size)
4627 extent = btrfs_item_ptr(leaf, path->slots[0],
4628 struct btrfs_file_extent_item);
4630 if (btrfs_file_extent_type(leaf, extent) ==
4631 BTRFS_FILE_EXTENT_INLINE) {
4632 len = btrfs_file_extent_inline_len(leaf,
4635 ASSERT(len == i_size ||
4636 (len == fs_info->sectorsize &&
4637 btrfs_file_extent_compression(leaf, extent) !=
4638 BTRFS_COMPRESS_NONE));
4642 len = btrfs_file_extent_num_bytes(leaf, extent);
4643 /* Last extent goes beyond i_size, no need to log a hole. */
4644 if (key.offset + len > i_size)
4646 hole_start = key.offset + len;
4647 hole_size = i_size - hole_start;
4649 btrfs_release_path(path);
4651 /* Last extent ends at i_size. */
4655 hole_size = ALIGN(hole_size, fs_info->sectorsize);
4656 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4657 hole_size, 0, hole_size, 0, 0, 0);
4662 * When we are logging a new inode X, check if it doesn't have a reference that
4663 * matches the reference from some other inode Y created in a past transaction
4664 * and that was renamed in the current transaction. If we don't do this, then at
4665 * log replay time we can lose inode Y (and all its files if it's a directory):
4668 * echo "hello world" > /mnt/x/foobar
4671 * mkdir /mnt/x # or touch /mnt/x
4672 * xfs_io -c fsync /mnt/x
4674 * mount fs, trigger log replay
4676 * After the log replay procedure, we would lose the first directory and all its
4677 * files (file foobar).
4678 * For the case where inode Y is not a directory we simply end up losing it:
4680 * echo "123" > /mnt/foo
4682 * mv /mnt/foo /mnt/bar
4683 * echo "abc" > /mnt/foo
4684 * xfs_io -c fsync /mnt/foo
4687 * We also need this for cases where a snapshot entry is replaced by some other
4688 * entry (file or directory) otherwise we end up with an unreplayable log due to
4689 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4690 * if it were a regular entry:
4693 * btrfs subvolume snapshot /mnt /mnt/x/snap
4694 * btrfs subvolume delete /mnt/x/snap
4697 * fsync /mnt/x or fsync some new file inside it
4700 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4701 * the same transaction.
4703 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4705 const struct btrfs_key *key,
4706 struct btrfs_inode *inode,
4710 struct btrfs_path *search_path;
4713 u32 item_size = btrfs_item_size_nr(eb, slot);
4715 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4717 search_path = btrfs_alloc_path();
4720 search_path->search_commit_root = 1;
4721 search_path->skip_locking = 1;
4723 while (cur_offset < item_size) {
4727 unsigned long name_ptr;
4728 struct btrfs_dir_item *di;
4730 if (key->type == BTRFS_INODE_REF_KEY) {
4731 struct btrfs_inode_ref *iref;
4733 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4734 parent = key->offset;
4735 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4736 name_ptr = (unsigned long)(iref + 1);
4737 this_len = sizeof(*iref) + this_name_len;
4739 struct btrfs_inode_extref *extref;
4741 extref = (struct btrfs_inode_extref *)(ptr +
4743 parent = btrfs_inode_extref_parent(eb, extref);
4744 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4745 name_ptr = (unsigned long)&extref->name;
4746 this_len = sizeof(*extref) + this_name_len;
4749 if (this_name_len > name_len) {
4752 new_name = krealloc(name, this_name_len, GFP_NOFS);
4757 name_len = this_name_len;
4761 read_extent_buffer(eb, name, name_ptr, this_name_len);
4762 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4763 parent, name, this_name_len, 0);
4764 if (di && !IS_ERR(di)) {
4765 struct btrfs_key di_key;
4767 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4769 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4771 *other_ino = di_key.objectid;
4776 } else if (IS_ERR(di)) {
4780 btrfs_release_path(search_path);
4782 cur_offset += this_len;
4786 btrfs_free_path(search_path);
4791 /* log a single inode in the tree log.
4792 * At least one parent directory for this inode must exist in the tree
4793 * or be logged already.
4795 * Any items from this inode changed by the current transaction are copied
4796 * to the log tree. An extra reference is taken on any extents in this
4797 * file, allowing us to avoid a whole pile of corner cases around logging
4798 * blocks that have been removed from the tree.
4800 * See LOG_INODE_ALL and related defines for a description of what inode_only
4803 * This handles both files and directories.
4805 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4806 struct btrfs_root *root, struct btrfs_inode *inode,
4810 struct btrfs_log_ctx *ctx)
4812 struct btrfs_fs_info *fs_info = root->fs_info;
4813 struct btrfs_path *path;
4814 struct btrfs_path *dst_path;
4815 struct btrfs_key min_key;
4816 struct btrfs_key max_key;
4817 struct btrfs_root *log = root->log_root;
4818 LIST_HEAD(logged_list);
4819 u64 last_extent = 0;
4823 int ins_start_slot = 0;
4825 bool fast_search = false;
4826 u64 ino = btrfs_ino(inode);
4827 struct extent_map_tree *em_tree = &inode->extent_tree;
4828 u64 logged_isize = 0;
4829 bool need_log_inode_item = true;
4831 path = btrfs_alloc_path();
4834 dst_path = btrfs_alloc_path();
4836 btrfs_free_path(path);
4840 min_key.objectid = ino;
4841 min_key.type = BTRFS_INODE_ITEM_KEY;
4844 max_key.objectid = ino;
4847 /* today the code can only do partial logging of directories */
4848 if (S_ISDIR(inode->vfs_inode.i_mode) ||
4849 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4850 &inode->runtime_flags) &&
4851 inode_only >= LOG_INODE_EXISTS))
4852 max_key.type = BTRFS_XATTR_ITEM_KEY;
4854 max_key.type = (u8)-1;
4855 max_key.offset = (u64)-1;
4858 * Only run delayed items if we are a dir or a new file.
4859 * Otherwise commit the delayed inode only, which is needed in
4860 * order for the log replay code to mark inodes for link count
4861 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4863 if (S_ISDIR(inode->vfs_inode.i_mode) ||
4864 inode->generation > fs_info->last_trans_committed)
4865 ret = btrfs_commit_inode_delayed_items(trans, inode);
4867 ret = btrfs_commit_inode_delayed_inode(inode);
4870 btrfs_free_path(path);
4871 btrfs_free_path(dst_path);
4875 if (inode_only == LOG_OTHER_INODE) {
4876 inode_only = LOG_INODE_EXISTS;
4877 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
4879 mutex_lock(&inode->log_mutex);
4883 * a brute force approach to making sure we get the most uptodate
4884 * copies of everything.
4886 if (S_ISDIR(inode->vfs_inode.i_mode)) {
4887 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4889 if (inode_only == LOG_INODE_EXISTS)
4890 max_key_type = BTRFS_XATTR_ITEM_KEY;
4891 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
4893 if (inode_only == LOG_INODE_EXISTS) {
4895 * Make sure the new inode item we write to the log has
4896 * the same isize as the current one (if it exists).
4897 * This is necessary to prevent data loss after log
4898 * replay, and also to prevent doing a wrong expanding
4899 * truncate - for e.g. create file, write 4K into offset
4900 * 0, fsync, write 4K into offset 4096, add hard link,
4901 * fsync some other file (to sync log), power fail - if
4902 * we use the inode's current i_size, after log replay
4903 * we get a 8Kb file, with the last 4Kb extent as a hole
4904 * (zeroes), as if an expanding truncate happened,
4905 * instead of getting a file of 4Kb only.
4907 err = logged_inode_size(log, inode, path, &logged_isize);
4911 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4912 &inode->runtime_flags)) {
4913 if (inode_only == LOG_INODE_EXISTS) {
4914 max_key.type = BTRFS_XATTR_ITEM_KEY;
4915 ret = drop_objectid_items(trans, log, path, ino,
4918 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4919 &inode->runtime_flags);
4920 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4921 &inode->runtime_flags);
4923 ret = btrfs_truncate_inode_items(trans,
4924 log, &inode->vfs_inode, 0, 0);
4929 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4930 &inode->runtime_flags) ||
4931 inode_only == LOG_INODE_EXISTS) {
4932 if (inode_only == LOG_INODE_ALL)
4934 max_key.type = BTRFS_XATTR_ITEM_KEY;
4935 ret = drop_objectid_items(trans, log, path, ino,
4938 if (inode_only == LOG_INODE_ALL)
4951 ret = btrfs_search_forward(root, &min_key,
4952 path, trans->transid);
4960 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4961 if (min_key.objectid != ino)
4963 if (min_key.type > max_key.type)
4966 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4967 need_log_inode_item = false;
4969 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4970 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4971 inode->generation == trans->transid) {
4974 ret = btrfs_check_ref_name_override(path->nodes[0],
4975 path->slots[0], &min_key, inode,
4980 } else if (ret > 0 && ctx &&
4981 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
4982 struct btrfs_key inode_key;
4983 struct inode *other_inode;
4989 ins_start_slot = path->slots[0];
4991 ret = copy_items(trans, inode, dst_path, path,
4992 &last_extent, ins_start_slot,
5000 btrfs_release_path(path);
5001 inode_key.objectid = other_ino;
5002 inode_key.type = BTRFS_INODE_ITEM_KEY;
5003 inode_key.offset = 0;
5004 other_inode = btrfs_iget(fs_info->sb,
5008 * If the other inode that had a conflicting dir
5009 * entry was deleted in the current transaction,
5010 * we don't need to do more work nor fallback to
5011 * a transaction commit.
5013 if (IS_ERR(other_inode) &&
5014 PTR_ERR(other_inode) == -ENOENT) {
5016 } else if (IS_ERR(other_inode)) {
5017 err = PTR_ERR(other_inode);
5021 * We are safe logging the other inode without
5022 * acquiring its i_mutex as long as we log with
5023 * the LOG_INODE_EXISTS mode. We're safe against
5024 * concurrent renames of the other inode as well
5025 * because during a rename we pin the log and
5026 * update the log with the new name before we
5029 err = btrfs_log_inode(trans, root,
5030 BTRFS_I(other_inode),
5031 LOG_OTHER_INODE, 0, LLONG_MAX,
5041 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5042 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
5045 ret = copy_items(trans, inode, dst_path, path,
5046 &last_extent, ins_start_slot,
5047 ins_nr, inode_only, logged_isize);
5054 btrfs_release_path(path);
5060 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5063 } else if (!ins_nr) {
5064 ins_start_slot = path->slots[0];
5069 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5070 ins_start_slot, ins_nr, inode_only,
5078 btrfs_release_path(path);
5082 ins_start_slot = path->slots[0];
5085 nritems = btrfs_header_nritems(path->nodes[0]);
5087 if (path->slots[0] < nritems) {
5088 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
5093 ret = copy_items(trans, inode, dst_path, path,
5094 &last_extent, ins_start_slot,
5095 ins_nr, inode_only, logged_isize);
5103 btrfs_release_path(path);
5105 if (min_key.offset < (u64)-1) {
5107 } else if (min_key.type < max_key.type) {
5115 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5116 ins_start_slot, ins_nr, inode_only,
5126 btrfs_release_path(path);
5127 btrfs_release_path(dst_path);
5128 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5131 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5132 btrfs_release_path(path);
5133 btrfs_release_path(dst_path);
5134 err = btrfs_log_trailing_hole(trans, root, inode, path);
5139 btrfs_release_path(path);
5140 btrfs_release_path(dst_path);
5141 if (need_log_inode_item) {
5142 err = log_inode_item(trans, log, dst_path, inode);
5147 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5148 &logged_list, ctx, start, end);
5153 } else if (inode_only == LOG_INODE_ALL) {
5154 struct extent_map *em, *n;
5156 write_lock(&em_tree->lock);
5158 * We can't just remove every em if we're called for a ranged
5159 * fsync - that is, one that doesn't cover the whole possible
5160 * file range (0 to LLONG_MAX). This is because we can have
5161 * em's that fall outside the range we're logging and therefore
5162 * their ordered operations haven't completed yet
5163 * (btrfs_finish_ordered_io() not invoked yet). This means we
5164 * didn't get their respective file extent item in the fs/subvol
5165 * tree yet, and need to let the next fast fsync (one which
5166 * consults the list of modified extent maps) find the em so
5167 * that it logs a matching file extent item and waits for the
5168 * respective ordered operation to complete (if it's still
5171 * Removing every em outside the range we're logging would make
5172 * the next fast fsync not log their matching file extent items,
5173 * therefore making us lose data after a log replay.
5175 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5177 const u64 mod_end = em->mod_start + em->mod_len - 1;
5179 if (em->mod_start >= start && mod_end <= end)
5180 list_del_init(&em->list);
5182 write_unlock(&em_tree->lock);
5185 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5186 ret = log_directory_changes(trans, root, inode, path, dst_path,
5194 spin_lock(&inode->lock);
5195 inode->logged_trans = trans->transid;
5196 inode->last_log_commit = inode->last_sub_trans;
5197 spin_unlock(&inode->lock);
5200 btrfs_put_logged_extents(&logged_list);
5202 btrfs_submit_logged_extents(&logged_list, log);
5203 mutex_unlock(&inode->log_mutex);
5205 btrfs_free_path(path);
5206 btrfs_free_path(dst_path);
5211 * Check if we must fallback to a transaction commit when logging an inode.
5212 * This must be called after logging the inode and is used only in the context
5213 * when fsyncing an inode requires the need to log some other inode - in which
5214 * case we can't lock the i_mutex of each other inode we need to log as that
5215 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5216 * log inodes up or down in the hierarchy) or rename operations for example. So
5217 * we take the log_mutex of the inode after we have logged it and then check for
5218 * its last_unlink_trans value - this is safe because any task setting
5219 * last_unlink_trans must take the log_mutex and it must do this before it does
5220 * the actual unlink operation, so if we do this check before a concurrent task
5221 * sets last_unlink_trans it means we've logged a consistent version/state of
5222 * all the inode items, otherwise we are not sure and must do a transaction
5223 * commit (the concurrent task might have only updated last_unlink_trans before
5224 * we logged the inode or it might have also done the unlink).
5226 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5227 struct btrfs_inode *inode)
5229 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5232 mutex_lock(&inode->log_mutex);
5233 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5235 * Make sure any commits to the log are forced to be full
5238 btrfs_set_log_full_commit(fs_info, trans);
5241 mutex_unlock(&inode->log_mutex);
5247 * follow the dentry parent pointers up the chain and see if any
5248 * of the directories in it require a full commit before they can
5249 * be logged. Returns zero if nothing special needs to be done or 1 if
5250 * a full commit is required.
5252 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5253 struct btrfs_inode *inode,
5254 struct dentry *parent,
5255 struct super_block *sb,
5259 struct dentry *old_parent = NULL;
5260 struct btrfs_inode *orig_inode = inode;
5263 * for regular files, if its inode is already on disk, we don't
5264 * have to worry about the parents at all. This is because
5265 * we can use the last_unlink_trans field to record renames
5266 * and other fun in this file.
5268 if (S_ISREG(inode->vfs_inode.i_mode) &&
5269 inode->generation <= last_committed &&
5270 inode->last_unlink_trans <= last_committed)
5273 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5274 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5276 inode = BTRFS_I(d_inode(parent));
5281 * If we are logging a directory then we start with our inode,
5282 * not our parent's inode, so we need to skip setting the
5283 * logged_trans so that further down in the log code we don't
5284 * think this inode has already been logged.
5286 if (inode != orig_inode)
5287 inode->logged_trans = trans->transid;
5290 if (btrfs_must_commit_transaction(trans, inode)) {
5295 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5298 if (IS_ROOT(parent)) {
5299 inode = BTRFS_I(d_inode(parent));
5300 if (btrfs_must_commit_transaction(trans, inode))
5305 parent = dget_parent(parent);
5307 old_parent = parent;
5308 inode = BTRFS_I(d_inode(parent));
5316 struct btrfs_dir_list {
5318 struct list_head list;
5322 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5323 * details about the why it is needed.
5324 * This is a recursive operation - if an existing dentry corresponds to a
5325 * directory, that directory's new entries are logged too (same behaviour as
5326 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5327 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5328 * complains about the following circular lock dependency / possible deadlock:
5332 * lock(&type->i_mutex_dir_key#3/2);
5333 * lock(sb_internal#2);
5334 * lock(&type->i_mutex_dir_key#3/2);
5335 * lock(&sb->s_type->i_mutex_key#14);
5337 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5338 * sb_start_intwrite() in btrfs_start_transaction().
5339 * Not locking i_mutex of the inodes is still safe because:
5341 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5342 * that while logging the inode new references (names) are added or removed
5343 * from the inode, leaving the logged inode item with a link count that does
5344 * not match the number of logged inode reference items. This is fine because
5345 * at log replay time we compute the real number of links and correct the
5346 * link count in the inode item (see replay_one_buffer() and
5347 * link_to_fixup_dir());
5349 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5350 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5351 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5352 * has a size that doesn't match the sum of the lengths of all the logged
5353 * names. This does not result in a problem because if a dir_item key is
5354 * logged but its matching dir_index key is not logged, at log replay time we
5355 * don't use it to replay the respective name (see replay_one_name()). On the
5356 * other hand if only the dir_index key ends up being logged, the respective
5357 * name is added to the fs/subvol tree with both the dir_item and dir_index
5358 * keys created (see replay_one_name()).
5359 * The directory's inode item with a wrong i_size is not a problem as well,
5360 * since we don't use it at log replay time to set the i_size in the inode
5361 * item of the fs/subvol tree (see overwrite_item()).
5363 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5364 struct btrfs_root *root,
5365 struct btrfs_inode *start_inode,
5366 struct btrfs_log_ctx *ctx)
5368 struct btrfs_fs_info *fs_info = root->fs_info;
5369 struct btrfs_root *log = root->log_root;
5370 struct btrfs_path *path;
5371 LIST_HEAD(dir_list);
5372 struct btrfs_dir_list *dir_elem;
5375 path = btrfs_alloc_path();
5379 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5381 btrfs_free_path(path);
5384 dir_elem->ino = btrfs_ino(start_inode);
5385 list_add_tail(&dir_elem->list, &dir_list);
5387 while (!list_empty(&dir_list)) {
5388 struct extent_buffer *leaf;
5389 struct btrfs_key min_key;
5393 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5396 goto next_dir_inode;
5398 min_key.objectid = dir_elem->ino;
5399 min_key.type = BTRFS_DIR_ITEM_KEY;
5402 btrfs_release_path(path);
5403 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5405 goto next_dir_inode;
5406 } else if (ret > 0) {
5408 goto next_dir_inode;
5412 leaf = path->nodes[0];
5413 nritems = btrfs_header_nritems(leaf);
5414 for (i = path->slots[0]; i < nritems; i++) {
5415 struct btrfs_dir_item *di;
5416 struct btrfs_key di_key;
5417 struct inode *di_inode;
5418 struct btrfs_dir_list *new_dir_elem;
5419 int log_mode = LOG_INODE_EXISTS;
5422 btrfs_item_key_to_cpu(leaf, &min_key, i);
5423 if (min_key.objectid != dir_elem->ino ||
5424 min_key.type != BTRFS_DIR_ITEM_KEY)
5425 goto next_dir_inode;
5427 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5428 type = btrfs_dir_type(leaf, di);
5429 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5430 type != BTRFS_FT_DIR)
5432 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5433 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5436 btrfs_release_path(path);
5437 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
5438 if (IS_ERR(di_inode)) {
5439 ret = PTR_ERR(di_inode);
5440 goto next_dir_inode;
5443 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5448 ctx->log_new_dentries = false;
5449 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5450 log_mode = LOG_INODE_ALL;
5451 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5452 log_mode, 0, LLONG_MAX, ctx);
5454 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5458 goto next_dir_inode;
5459 if (ctx->log_new_dentries) {
5460 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5462 if (!new_dir_elem) {
5464 goto next_dir_inode;
5466 new_dir_elem->ino = di_key.objectid;
5467 list_add_tail(&new_dir_elem->list, &dir_list);
5472 ret = btrfs_next_leaf(log, path);
5474 goto next_dir_inode;
5475 } else if (ret > 0) {
5477 goto next_dir_inode;
5481 if (min_key.offset < (u64)-1) {
5486 list_del(&dir_elem->list);
5490 btrfs_free_path(path);
5494 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5495 struct btrfs_inode *inode,
5496 struct btrfs_log_ctx *ctx)
5498 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5500 struct btrfs_path *path;
5501 struct btrfs_key key;
5502 struct btrfs_root *root = inode->root;
5503 const u64 ino = btrfs_ino(inode);
5505 path = btrfs_alloc_path();
5508 path->skip_locking = 1;
5509 path->search_commit_root = 1;
5512 key.type = BTRFS_INODE_REF_KEY;
5514 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5519 struct extent_buffer *leaf = path->nodes[0];
5520 int slot = path->slots[0];
5525 if (slot >= btrfs_header_nritems(leaf)) {
5526 ret = btrfs_next_leaf(root, path);
5534 btrfs_item_key_to_cpu(leaf, &key, slot);
5535 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5536 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5539 item_size = btrfs_item_size_nr(leaf, slot);
5540 ptr = btrfs_item_ptr_offset(leaf, slot);
5541 while (cur_offset < item_size) {
5542 struct btrfs_key inode_key;
5543 struct inode *dir_inode;
5545 inode_key.type = BTRFS_INODE_ITEM_KEY;
5546 inode_key.offset = 0;
5548 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5549 struct btrfs_inode_extref *extref;
5551 extref = (struct btrfs_inode_extref *)
5553 inode_key.objectid = btrfs_inode_extref_parent(
5555 cur_offset += sizeof(*extref);
5556 cur_offset += btrfs_inode_extref_name_len(leaf,
5559 inode_key.objectid = key.offset;
5560 cur_offset = item_size;
5563 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
5565 /* If parent inode was deleted, skip it. */
5566 if (IS_ERR(dir_inode))
5570 ctx->log_new_dentries = false;
5571 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5572 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5574 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5576 if (!ret && ctx && ctx->log_new_dentries)
5577 ret = log_new_dir_dentries(trans, root,
5578 BTRFS_I(dir_inode), ctx);
5587 btrfs_free_path(path);
5592 * helper function around btrfs_log_inode to make sure newly created
5593 * parent directories also end up in the log. A minimal inode and backref
5594 * only logging is done of any parent directories that are older than
5595 * the last committed transaction
5597 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5598 struct btrfs_inode *inode,
5599 struct dentry *parent,
5603 struct btrfs_log_ctx *ctx)
5605 struct btrfs_root *root = inode->root;
5606 struct btrfs_fs_info *fs_info = root->fs_info;
5607 struct super_block *sb;
5608 struct dentry *old_parent = NULL;
5610 u64 last_committed = fs_info->last_trans_committed;
5611 bool log_dentries = false;
5612 struct btrfs_inode *orig_inode = inode;
5614 sb = inode->vfs_inode.i_sb;
5616 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5622 * The prev transaction commit doesn't complete, we need do
5623 * full commit by ourselves.
5625 if (fs_info->last_trans_log_full_commit >
5626 fs_info->last_trans_committed) {
5631 if (btrfs_root_refs(&root->root_item) == 0) {
5636 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
5641 if (btrfs_inode_in_log(inode, trans->transid)) {
5642 ret = BTRFS_NO_LOG_SYNC;
5646 ret = start_log_trans(trans, root, ctx);
5650 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5655 * for regular files, if its inode is already on disk, we don't
5656 * have to worry about the parents at all. This is because
5657 * we can use the last_unlink_trans field to record renames
5658 * and other fun in this file.
5660 if (S_ISREG(inode->vfs_inode.i_mode) &&
5661 inode->generation <= last_committed &&
5662 inode->last_unlink_trans <= last_committed) {
5667 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
5668 log_dentries = true;
5671 * On unlink we must make sure all our current and old parent directory
5672 * inodes are fully logged. This is to prevent leaving dangling
5673 * directory index entries in directories that were our parents but are
5674 * not anymore. Not doing this results in old parent directory being
5675 * impossible to delete after log replay (rmdir will always fail with
5676 * error -ENOTEMPTY).
5682 * ln testdir/foo testdir/bar
5684 * unlink testdir/bar
5685 * xfs_io -c fsync testdir/foo
5687 * mount fs, triggers log replay
5689 * If we don't log the parent directory (testdir), after log replay the
5690 * directory still has an entry pointing to the file inode using the bar
5691 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5692 * the file inode has a link count of 1.
5698 * ln foo testdir/foo2
5699 * ln foo testdir/foo3
5701 * unlink testdir/foo3
5702 * xfs_io -c fsync foo
5704 * mount fs, triggers log replay
5706 * Similar as the first example, after log replay the parent directory
5707 * testdir still has an entry pointing to the inode file with name foo3
5708 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5709 * and has a link count of 2.
5711 if (inode->last_unlink_trans > last_committed) {
5712 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5718 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5721 inode = BTRFS_I(d_inode(parent));
5722 if (root != inode->root)
5725 if (inode->generation > last_committed) {
5726 ret = btrfs_log_inode(trans, root, inode,
5727 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5731 if (IS_ROOT(parent))
5734 parent = dget_parent(parent);
5736 old_parent = parent;
5739 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5745 btrfs_set_log_full_commit(fs_info, trans);
5750 btrfs_remove_log_ctx(root, ctx);
5751 btrfs_end_log_trans(root);
5757 * it is not safe to log dentry if the chunk root has added new
5758 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5759 * If this returns 1, you must commit the transaction to safely get your
5762 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
5763 struct dentry *dentry,
5766 struct btrfs_log_ctx *ctx)
5768 struct dentry *parent = dget_parent(dentry);
5771 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
5772 start, end, LOG_INODE_ALL, ctx);
5779 * should be called during mount to recover any replay any log trees
5782 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5785 struct btrfs_path *path;
5786 struct btrfs_trans_handle *trans;
5787 struct btrfs_key key;
5788 struct btrfs_key found_key;
5789 struct btrfs_key tmp_key;
5790 struct btrfs_root *log;
5791 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5792 struct walk_control wc = {
5793 .process_func = process_one_buffer,
5797 path = btrfs_alloc_path();
5801 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5803 trans = btrfs_start_transaction(fs_info->tree_root, 0);
5804 if (IS_ERR(trans)) {
5805 ret = PTR_ERR(trans);
5812 ret = walk_log_tree(trans, log_root_tree, &wc);
5814 btrfs_handle_fs_error(fs_info, ret,
5815 "Failed to pin buffers while recovering log root tree.");
5820 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5821 key.offset = (u64)-1;
5822 key.type = BTRFS_ROOT_ITEM_KEY;
5825 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
5828 btrfs_handle_fs_error(fs_info, ret,
5829 "Couldn't find tree log root.");
5833 if (path->slots[0] == 0)
5837 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5839 btrfs_release_path(path);
5840 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5843 log = btrfs_read_fs_root(log_root_tree, &found_key);
5846 btrfs_handle_fs_error(fs_info, ret,
5847 "Couldn't read tree log root.");
5851 tmp_key.objectid = found_key.offset;
5852 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5853 tmp_key.offset = (u64)-1;
5855 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
5856 if (IS_ERR(wc.replay_dest)) {
5857 ret = PTR_ERR(wc.replay_dest);
5858 free_extent_buffer(log->node);
5859 free_extent_buffer(log->commit_root);
5861 btrfs_handle_fs_error(fs_info, ret,
5862 "Couldn't read target root for tree log recovery.");
5866 wc.replay_dest->log_root = log;
5867 btrfs_record_root_in_trans(trans, wc.replay_dest);
5868 ret = walk_log_tree(trans, log, &wc);
5870 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5871 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5875 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5876 struct btrfs_root *root = wc.replay_dest;
5878 btrfs_release_path(path);
5881 * We have just replayed everything, and the highest
5882 * objectid of fs roots probably has changed in case
5883 * some inode_item's got replayed.
5885 * root->objectid_mutex is not acquired as log replay
5886 * could only happen during mount.
5888 ret = btrfs_find_highest_objectid(root,
5889 &root->highest_objectid);
5892 key.offset = found_key.offset - 1;
5893 wc.replay_dest->log_root = NULL;
5894 free_extent_buffer(log->node);
5895 free_extent_buffer(log->commit_root);
5901 if (found_key.offset == 0)
5904 btrfs_release_path(path);
5906 /* step one is to pin it all, step two is to replay just inodes */
5909 wc.process_func = replay_one_buffer;
5910 wc.stage = LOG_WALK_REPLAY_INODES;
5913 /* step three is to replay everything */
5914 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5919 btrfs_free_path(path);
5921 /* step 4: commit the transaction, which also unpins the blocks */
5922 ret = btrfs_commit_transaction(trans);
5926 free_extent_buffer(log_root_tree->node);
5927 log_root_tree->log_root = NULL;
5928 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5929 kfree(log_root_tree);
5934 btrfs_end_transaction(wc.trans);
5935 btrfs_free_path(path);
5940 * there are some corner cases where we want to force a full
5941 * commit instead of allowing a directory to be logged.
5943 * They revolve around files there were unlinked from the directory, and
5944 * this function updates the parent directory so that a full commit is
5945 * properly done if it is fsync'd later after the unlinks are done.
5947 * Must be called before the unlink operations (updates to the subvolume tree,
5948 * inodes, etc) are done.
5950 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5951 struct btrfs_inode *dir, struct btrfs_inode *inode,
5955 * when we're logging a file, if it hasn't been renamed
5956 * or unlinked, and its inode is fully committed on disk,
5957 * we don't have to worry about walking up the directory chain
5958 * to log its parents.
5960 * So, we use the last_unlink_trans field to put this transid
5961 * into the file. When the file is logged we check it and
5962 * don't log the parents if the file is fully on disk.
5964 mutex_lock(&inode->log_mutex);
5965 inode->last_unlink_trans = trans->transid;
5966 mutex_unlock(&inode->log_mutex);
5969 * if this directory was already logged any new
5970 * names for this file/dir will get recorded
5973 if (dir->logged_trans == trans->transid)
5977 * if the inode we're about to unlink was logged,
5978 * the log will be properly updated for any new names
5980 if (inode->logged_trans == trans->transid)
5984 * when renaming files across directories, if the directory
5985 * there we're unlinking from gets fsync'd later on, there's
5986 * no way to find the destination directory later and fsync it
5987 * properly. So, we have to be conservative and force commits
5988 * so the new name gets discovered.
5993 /* we can safely do the unlink without any special recording */
5997 mutex_lock(&dir->log_mutex);
5998 dir->last_unlink_trans = trans->transid;
5999 mutex_unlock(&dir->log_mutex);
6003 * Make sure that if someone attempts to fsync the parent directory of a deleted
6004 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6005 * that after replaying the log tree of the parent directory's root we will not
6006 * see the snapshot anymore and at log replay time we will not see any log tree
6007 * corresponding to the deleted snapshot's root, which could lead to replaying
6008 * it after replaying the log tree of the parent directory (which would replay
6009 * the snapshot delete operation).
6011 * Must be called before the actual snapshot destroy operation (updates to the
6012 * parent root and tree of tree roots trees, etc) are done.
6014 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6015 struct btrfs_inode *dir)
6017 mutex_lock(&dir->log_mutex);
6018 dir->last_unlink_trans = trans->transid;
6019 mutex_unlock(&dir->log_mutex);
6023 * Call this after adding a new name for a file and it will properly
6024 * update the log to reflect the new name.
6026 * It will return zero if all goes well, and it will return 1 if a
6027 * full transaction commit is required.
6029 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
6030 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6031 struct dentry *parent)
6033 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6036 * this will force the logging code to walk the dentry chain
6039 if (!S_ISDIR(inode->vfs_inode.i_mode))
6040 inode->last_unlink_trans = trans->transid;
6043 * if this inode hasn't been logged and directory we're renaming it
6044 * from hasn't been logged, we don't need to log it
6046 if (inode->logged_trans <= fs_info->last_trans_committed &&
6047 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
6050 return btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6051 LOG_INODE_EXISTS, NULL);