2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/blkdev.h>
22 #include <linux/list_sort.h>
26 #include "print-tree.h"
29 #include "compression.h"
32 /* magic values for the inode_only field in btrfs_log_inode:
34 * LOG_INODE_ALL means to log everything
35 * LOG_INODE_EXISTS means to log just enough to recreate the inode
38 #define LOG_INODE_ALL 0
39 #define LOG_INODE_EXISTS 1
40 #define LOG_OTHER_INODE 2
43 * directory trouble cases
45 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
46 * log, we must force a full commit before doing an fsync of the directory
47 * where the unlink was done.
48 * ---> record transid of last unlink/rename per directory
52 * rename foo/some_dir foo2/some_dir
54 * fsync foo/some_dir/some_file
56 * The fsync above will unlink the original some_dir without recording
57 * it in its new location (foo2). After a crash, some_dir will be gone
58 * unless the fsync of some_file forces a full commit
60 * 2) we must log any new names for any file or dir that is in the fsync
61 * log. ---> check inode while renaming/linking.
63 * 2a) we must log any new names for any file or dir during rename
64 * when the directory they are being removed from was logged.
65 * ---> check inode and old parent dir during rename
67 * 2a is actually the more important variant. With the extra logging
68 * a crash might unlink the old name without recreating the new one
70 * 3) after a crash, we must go through any directories with a link count
71 * of zero and redo the rm -rf
78 * The directory f1 was fully removed from the FS, but fsync was never
79 * called on f1, only its parent dir. After a crash the rm -rf must
80 * be replayed. This must be able to recurse down the entire
81 * directory tree. The inode link count fixup code takes care of the
86 * stages for the tree walking. The first
87 * stage (0) is to only pin down the blocks we find
88 * the second stage (1) is to make sure that all the inodes
89 * we find in the log are created in the subvolume.
91 * The last stage is to deal with directories and links and extents
92 * and all the other fun semantics
94 #define LOG_WALK_PIN_ONLY 0
95 #define LOG_WALK_REPLAY_INODES 1
96 #define LOG_WALK_REPLAY_DIR_INDEX 2
97 #define LOG_WALK_REPLAY_ALL 3
99 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
100 struct btrfs_root *root, struct btrfs_inode *inode,
104 struct btrfs_log_ctx *ctx);
105 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
107 struct btrfs_path *path, u64 objectid);
108 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
109 struct btrfs_root *root,
110 struct btrfs_root *log,
111 struct btrfs_path *path,
112 u64 dirid, int del_all);
115 * tree logging is a special write ahead log used to make sure that
116 * fsyncs and O_SYNCs can happen without doing full tree commits.
118 * Full tree commits are expensive because they require commonly
119 * modified blocks to be recowed, creating many dirty pages in the
120 * extent tree an 4x-6x higher write load than ext3.
122 * Instead of doing a tree commit on every fsync, we use the
123 * key ranges and transaction ids to find items for a given file or directory
124 * that have changed in this transaction. Those items are copied into
125 * a special tree (one per subvolume root), that tree is written to disk
126 * and then the fsync is considered complete.
128 * After a crash, items are copied out of the log-tree back into the
129 * subvolume tree. Any file data extents found are recorded in the extent
130 * allocation tree, and the log-tree freed.
132 * The log tree is read three times, once to pin down all the extents it is
133 * using in ram and once, once to create all the inodes logged in the tree
134 * and once to do all the other items.
138 * start a sub transaction and setup the log tree
139 * this increments the log tree writer count to make the people
140 * syncing the tree wait for us to finish
142 static int start_log_trans(struct btrfs_trans_handle *trans,
143 struct btrfs_root *root,
144 struct btrfs_log_ctx *ctx)
146 struct btrfs_fs_info *fs_info = root->fs_info;
149 mutex_lock(&root->log_mutex);
151 if (root->log_root) {
152 if (btrfs_need_log_full_commit(fs_info, trans)) {
157 if (!root->log_start_pid) {
158 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
159 root->log_start_pid = current->pid;
160 } else if (root->log_start_pid != current->pid) {
161 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
164 mutex_lock(&fs_info->tree_log_mutex);
165 if (!fs_info->log_root_tree)
166 ret = btrfs_init_log_root_tree(trans, fs_info);
167 mutex_unlock(&fs_info->tree_log_mutex);
171 ret = btrfs_add_log_tree(trans, root);
175 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
176 root->log_start_pid = current->pid;
179 atomic_inc(&root->log_batch);
180 atomic_inc(&root->log_writers);
182 int index = root->log_transid % 2;
183 list_add_tail(&ctx->list, &root->log_ctxs[index]);
184 ctx->log_transid = root->log_transid;
188 mutex_unlock(&root->log_mutex);
193 * returns 0 if there was a log transaction running and we were able
194 * to join, or returns -ENOENT if there were not transactions
197 static int join_running_log_trans(struct btrfs_root *root)
205 mutex_lock(&root->log_mutex);
206 if (root->log_root) {
208 atomic_inc(&root->log_writers);
210 mutex_unlock(&root->log_mutex);
215 * This either makes the current running log transaction wait
216 * until you call btrfs_end_log_trans() or it makes any future
217 * log transactions wait until you call btrfs_end_log_trans()
219 int btrfs_pin_log_trans(struct btrfs_root *root)
223 mutex_lock(&root->log_mutex);
224 atomic_inc(&root->log_writers);
225 mutex_unlock(&root->log_mutex);
230 * indicate we're done making changes to the log tree
231 * and wake up anyone waiting to do a sync
233 void btrfs_end_log_trans(struct btrfs_root *root)
235 if (atomic_dec_and_test(&root->log_writers)) {
237 * Implicit memory barrier after atomic_dec_and_test
239 if (waitqueue_active(&root->log_writer_wait))
240 wake_up(&root->log_writer_wait);
246 * the walk control struct is used to pass state down the chain when
247 * processing the log tree. The stage field tells us which part
248 * of the log tree processing we are currently doing. The others
249 * are state fields used for that specific part
251 struct walk_control {
252 /* should we free the extent on disk when done? This is used
253 * at transaction commit time while freeing a log tree
257 /* should we write out the extent buffer? This is used
258 * while flushing the log tree to disk during a sync
262 /* should we wait for the extent buffer io to finish? Also used
263 * while flushing the log tree to disk for a sync
267 /* pin only walk, we record which extents on disk belong to the
272 /* what stage of the replay code we're currently in */
275 /* the root we are currently replaying */
276 struct btrfs_root *replay_dest;
278 /* the trans handle for the current replay */
279 struct btrfs_trans_handle *trans;
281 /* the function that gets used to process blocks we find in the
282 * tree. Note the extent_buffer might not be up to date when it is
283 * passed in, and it must be checked or read if you need the data
286 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
287 struct walk_control *wc, u64 gen);
291 * process_func used to pin down extents, write them or wait on them
293 static int process_one_buffer(struct btrfs_root *log,
294 struct extent_buffer *eb,
295 struct walk_control *wc, u64 gen)
297 struct btrfs_fs_info *fs_info = log->fs_info;
301 * If this fs is mixed then we need to be able to process the leaves to
302 * pin down any logged extents, so we have to read the block.
304 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
305 ret = btrfs_read_buffer(eb, gen);
311 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
314 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
315 if (wc->pin && btrfs_header_level(eb) == 0)
316 ret = btrfs_exclude_logged_extents(fs_info, eb);
318 btrfs_write_tree_block(eb);
320 btrfs_wait_tree_block_writeback(eb);
326 * Item overwrite used by replay and tree logging. eb, slot and key all refer
327 * to the src data we are copying out.
329 * root is the tree we are copying into, and path is a scratch
330 * path for use in this function (it should be released on entry and
331 * will be released on exit).
333 * If the key is already in the destination tree the existing item is
334 * overwritten. If the existing item isn't big enough, it is extended.
335 * If it is too large, it is truncated.
337 * If the key isn't in the destination yet, a new item is inserted.
339 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
340 struct btrfs_root *root,
341 struct btrfs_path *path,
342 struct extent_buffer *eb, int slot,
343 struct btrfs_key *key)
345 struct btrfs_fs_info *fs_info = root->fs_info;
348 u64 saved_i_size = 0;
349 int save_old_i_size = 0;
350 unsigned long src_ptr;
351 unsigned long dst_ptr;
352 int overwrite_root = 0;
353 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
355 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
358 item_size = btrfs_item_size_nr(eb, slot);
359 src_ptr = btrfs_item_ptr_offset(eb, slot);
361 /* look for the key in the destination tree */
362 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
369 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
371 if (dst_size != item_size)
374 if (item_size == 0) {
375 btrfs_release_path(path);
378 dst_copy = kmalloc(item_size, GFP_NOFS);
379 src_copy = kmalloc(item_size, GFP_NOFS);
380 if (!dst_copy || !src_copy) {
381 btrfs_release_path(path);
387 read_extent_buffer(eb, src_copy, src_ptr, item_size);
389 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
390 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
392 ret = memcmp(dst_copy, src_copy, item_size);
397 * they have the same contents, just return, this saves
398 * us from cowing blocks in the destination tree and doing
399 * extra writes that may not have been done by a previous
403 btrfs_release_path(path);
408 * We need to load the old nbytes into the inode so when we
409 * replay the extents we've logged we get the right nbytes.
412 struct btrfs_inode_item *item;
416 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
417 struct btrfs_inode_item);
418 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
419 item = btrfs_item_ptr(eb, slot,
420 struct btrfs_inode_item);
421 btrfs_set_inode_nbytes(eb, item, nbytes);
424 * If this is a directory we need to reset the i_size to
425 * 0 so that we can set it up properly when replaying
426 * the rest of the items in this log.
428 mode = btrfs_inode_mode(eb, item);
430 btrfs_set_inode_size(eb, item, 0);
432 } else if (inode_item) {
433 struct btrfs_inode_item *item;
437 * New inode, set nbytes to 0 so that the nbytes comes out
438 * properly when we replay the extents.
440 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
441 btrfs_set_inode_nbytes(eb, item, 0);
444 * If this is a directory we need to reset the i_size to 0 so
445 * that we can set it up properly when replaying the rest of
446 * the items in this log.
448 mode = btrfs_inode_mode(eb, item);
450 btrfs_set_inode_size(eb, item, 0);
453 btrfs_release_path(path);
454 /* try to insert the key into the destination tree */
455 path->skip_release_on_error = 1;
456 ret = btrfs_insert_empty_item(trans, root, path,
458 path->skip_release_on_error = 0;
460 /* make sure any existing item is the correct size */
461 if (ret == -EEXIST || ret == -EOVERFLOW) {
463 found_size = btrfs_item_size_nr(path->nodes[0],
465 if (found_size > item_size)
466 btrfs_truncate_item(fs_info, path, item_size, 1);
467 else if (found_size < item_size)
468 btrfs_extend_item(fs_info, path,
469 item_size - found_size);
473 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
476 /* don't overwrite an existing inode if the generation number
477 * was logged as zero. This is done when the tree logging code
478 * is just logging an inode to make sure it exists after recovery.
480 * Also, don't overwrite i_size on directories during replay.
481 * log replay inserts and removes directory items based on the
482 * state of the tree found in the subvolume, and i_size is modified
485 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
486 struct btrfs_inode_item *src_item;
487 struct btrfs_inode_item *dst_item;
489 src_item = (struct btrfs_inode_item *)src_ptr;
490 dst_item = (struct btrfs_inode_item *)dst_ptr;
492 if (btrfs_inode_generation(eb, src_item) == 0) {
493 struct extent_buffer *dst_eb = path->nodes[0];
494 const u64 ino_size = btrfs_inode_size(eb, src_item);
497 * For regular files an ino_size == 0 is used only when
498 * logging that an inode exists, as part of a directory
499 * fsync, and the inode wasn't fsynced before. In this
500 * case don't set the size of the inode in the fs/subvol
501 * tree, otherwise we would be throwing valid data away.
503 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
504 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
506 struct btrfs_map_token token;
508 btrfs_init_map_token(&token);
509 btrfs_set_token_inode_size(dst_eb, dst_item,
515 if (overwrite_root &&
516 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
517 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
519 saved_i_size = btrfs_inode_size(path->nodes[0],
524 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
527 if (save_old_i_size) {
528 struct btrfs_inode_item *dst_item;
529 dst_item = (struct btrfs_inode_item *)dst_ptr;
530 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
533 /* make sure the generation is filled in */
534 if (key->type == BTRFS_INODE_ITEM_KEY) {
535 struct btrfs_inode_item *dst_item;
536 dst_item = (struct btrfs_inode_item *)dst_ptr;
537 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
538 btrfs_set_inode_generation(path->nodes[0], dst_item,
543 btrfs_mark_buffer_dirty(path->nodes[0]);
544 btrfs_release_path(path);
549 * simple helper to read an inode off the disk from a given root
550 * This can only be called for subvolume roots and not for the log
552 static noinline struct inode *read_one_inode(struct btrfs_root *root,
555 struct btrfs_key key;
558 key.objectid = objectid;
559 key.type = BTRFS_INODE_ITEM_KEY;
561 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
564 } else if (is_bad_inode(inode)) {
571 /* replays a single extent in 'eb' at 'slot' with 'key' into the
572 * subvolume 'root'. path is released on entry and should be released
575 * extents in the log tree have not been allocated out of the extent
576 * tree yet. So, this completes the allocation, taking a reference
577 * as required if the extent already exists or creating a new extent
578 * if it isn't in the extent allocation tree yet.
580 * The extent is inserted into the file, dropping any existing extents
581 * from the file that overlap the new one.
583 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
584 struct btrfs_root *root,
585 struct btrfs_path *path,
586 struct extent_buffer *eb, int slot,
587 struct btrfs_key *key)
589 struct btrfs_fs_info *fs_info = root->fs_info;
592 u64 start = key->offset;
594 struct btrfs_file_extent_item *item;
595 struct inode *inode = NULL;
599 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
600 found_type = btrfs_file_extent_type(eb, item);
602 if (found_type == BTRFS_FILE_EXTENT_REG ||
603 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
604 nbytes = btrfs_file_extent_num_bytes(eb, item);
605 extent_end = start + nbytes;
608 * We don't add to the inodes nbytes if we are prealloc or a
611 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
613 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
614 size = btrfs_file_extent_inline_len(eb, slot, item);
615 nbytes = btrfs_file_extent_ram_bytes(eb, item);
616 extent_end = ALIGN(start + size,
617 fs_info->sectorsize);
623 inode = read_one_inode(root, key->objectid);
630 * first check to see if we already have this extent in the
631 * file. This must be done before the btrfs_drop_extents run
632 * so we don't try to drop this extent.
634 ret = btrfs_lookup_file_extent(trans, root, path,
635 btrfs_ino(BTRFS_I(inode)), start, 0);
638 (found_type == BTRFS_FILE_EXTENT_REG ||
639 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
640 struct btrfs_file_extent_item cmp1;
641 struct btrfs_file_extent_item cmp2;
642 struct btrfs_file_extent_item *existing;
643 struct extent_buffer *leaf;
645 leaf = path->nodes[0];
646 existing = btrfs_item_ptr(leaf, path->slots[0],
647 struct btrfs_file_extent_item);
649 read_extent_buffer(eb, &cmp1, (unsigned long)item,
651 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
655 * we already have a pointer to this exact extent,
656 * we don't have to do anything
658 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
659 btrfs_release_path(path);
663 btrfs_release_path(path);
665 /* drop any overlapping extents */
666 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
670 if (found_type == BTRFS_FILE_EXTENT_REG ||
671 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
673 unsigned long dest_offset;
674 struct btrfs_key ins;
676 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
677 btrfs_fs_incompat(fs_info, NO_HOLES))
680 ret = btrfs_insert_empty_item(trans, root, path, key,
684 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
686 copy_extent_buffer(path->nodes[0], eb, dest_offset,
687 (unsigned long)item, sizeof(*item));
689 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
690 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
691 ins.type = BTRFS_EXTENT_ITEM_KEY;
692 offset = key->offset - btrfs_file_extent_offset(eb, item);
695 * Manually record dirty extent, as here we did a shallow
696 * file extent item copy and skip normal backref update,
697 * but modifying extent tree all by ourselves.
698 * So need to manually record dirty extent for qgroup,
699 * as the owner of the file extent changed from log tree
700 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
702 ret = btrfs_qgroup_trace_extent(trans, fs_info,
703 btrfs_file_extent_disk_bytenr(eb, item),
704 btrfs_file_extent_disk_num_bytes(eb, item),
709 if (ins.objectid > 0) {
712 LIST_HEAD(ordered_sums);
714 * is this extent already allocated in the extent
715 * allocation tree? If so, just add a reference
717 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
720 ret = btrfs_inc_extent_ref(trans, fs_info,
721 ins.objectid, ins.offset,
722 0, root->root_key.objectid,
723 key->objectid, offset);
728 * insert the extent pointer in the extent
731 ret = btrfs_alloc_logged_file_extent(trans,
733 root->root_key.objectid,
734 key->objectid, offset, &ins);
738 btrfs_release_path(path);
740 if (btrfs_file_extent_compression(eb, item)) {
741 csum_start = ins.objectid;
742 csum_end = csum_start + ins.offset;
744 csum_start = ins.objectid +
745 btrfs_file_extent_offset(eb, item);
746 csum_end = csum_start +
747 btrfs_file_extent_num_bytes(eb, item);
750 ret = btrfs_lookup_csums_range(root->log_root,
751 csum_start, csum_end - 1,
756 * Now delete all existing cums in the csum root that
757 * cover our range. We do this because we can have an
758 * extent that is completely referenced by one file
759 * extent item and partially referenced by another
760 * file extent item (like after using the clone or
761 * extent_same ioctls). In this case if we end up doing
762 * the replay of the one that partially references the
763 * extent first, and we do not do the csum deletion
764 * below, we can get 2 csum items in the csum tree that
765 * overlap each other. For example, imagine our log has
766 * the two following file extent items:
768 * key (257 EXTENT_DATA 409600)
769 * extent data disk byte 12845056 nr 102400
770 * extent data offset 20480 nr 20480 ram 102400
772 * key (257 EXTENT_DATA 819200)
773 * extent data disk byte 12845056 nr 102400
774 * extent data offset 0 nr 102400 ram 102400
776 * Where the second one fully references the 100K extent
777 * that starts at disk byte 12845056, and the log tree
778 * has a single csum item that covers the entire range
781 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
783 * After the first file extent item is replayed, the
784 * csum tree gets the following csum item:
786 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
788 * Which covers the 20K sub-range starting at offset 20K
789 * of our extent. Now when we replay the second file
790 * extent item, if we do not delete existing csum items
791 * that cover any of its blocks, we end up getting two
792 * csum items in our csum tree that overlap each other:
794 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
795 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
797 * Which is a problem, because after this anyone trying
798 * to lookup up for the checksum of any block of our
799 * extent starting at an offset of 40K or higher, will
800 * end up looking at the second csum item only, which
801 * does not contain the checksum for any block starting
802 * at offset 40K or higher of our extent.
804 while (!list_empty(&ordered_sums)) {
805 struct btrfs_ordered_sum *sums;
806 sums = list_entry(ordered_sums.next,
807 struct btrfs_ordered_sum,
810 ret = btrfs_del_csums(trans, fs_info,
814 ret = btrfs_csum_file_blocks(trans,
815 fs_info->csum_root, sums);
816 list_del(&sums->list);
822 btrfs_release_path(path);
824 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
825 /* inline extents are easy, we just overwrite them */
826 ret = overwrite_item(trans, root, path, eb, slot, key);
831 inode_add_bytes(inode, nbytes);
833 ret = btrfs_update_inode(trans, root, inode);
841 * when cleaning up conflicts between the directory names in the
842 * subvolume, directory names in the log and directory names in the
843 * inode back references, we may have to unlink inodes from directories.
845 * This is a helper function to do the unlink of a specific directory
848 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
849 struct btrfs_root *root,
850 struct btrfs_path *path,
851 struct btrfs_inode *dir,
852 struct btrfs_dir_item *di)
854 struct btrfs_fs_info *fs_info = root->fs_info;
858 struct extent_buffer *leaf;
859 struct btrfs_key location;
862 leaf = path->nodes[0];
864 btrfs_dir_item_key_to_cpu(leaf, di, &location);
865 name_len = btrfs_dir_name_len(leaf, di);
866 name = kmalloc(name_len, GFP_NOFS);
870 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
871 btrfs_release_path(path);
873 inode = read_one_inode(root, location.objectid);
879 ret = link_to_fixup_dir(trans, root, path, location.objectid);
883 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
888 ret = btrfs_run_delayed_items(trans, fs_info);
896 * helper function to see if a given name and sequence number found
897 * in an inode back reference are already in a directory and correctly
898 * point to this inode
900 static noinline int inode_in_dir(struct btrfs_root *root,
901 struct btrfs_path *path,
902 u64 dirid, u64 objectid, u64 index,
903 const char *name, int name_len)
905 struct btrfs_dir_item *di;
906 struct btrfs_key location;
909 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
910 index, name, name_len, 0);
911 if (di && !IS_ERR(di)) {
912 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
913 if (location.objectid != objectid)
917 btrfs_release_path(path);
919 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
920 if (di && !IS_ERR(di)) {
921 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
922 if (location.objectid != objectid)
928 btrfs_release_path(path);
933 * helper function to check a log tree for a named back reference in
934 * an inode. This is used to decide if a back reference that is
935 * found in the subvolume conflicts with what we find in the log.
937 * inode backreferences may have multiple refs in a single item,
938 * during replay we process one reference at a time, and we don't
939 * want to delete valid links to a file from the subvolume if that
940 * link is also in the log.
942 static noinline int backref_in_log(struct btrfs_root *log,
943 struct btrfs_key *key,
945 const char *name, int namelen)
947 struct btrfs_path *path;
948 struct btrfs_inode_ref *ref;
950 unsigned long ptr_end;
951 unsigned long name_ptr;
957 path = btrfs_alloc_path();
961 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
965 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
967 if (key->type == BTRFS_INODE_EXTREF_KEY) {
968 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
969 name, namelen, NULL))
975 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
976 ptr_end = ptr + item_size;
977 while (ptr < ptr_end) {
978 ref = (struct btrfs_inode_ref *)ptr;
979 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
980 if (found_name_len == namelen) {
981 name_ptr = (unsigned long)(ref + 1);
982 ret = memcmp_extent_buffer(path->nodes[0], name,
989 ptr = (unsigned long)(ref + 1) + found_name_len;
992 btrfs_free_path(path);
996 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
997 struct btrfs_root *root,
998 struct btrfs_path *path,
999 struct btrfs_root *log_root,
1000 struct btrfs_inode *dir,
1001 struct btrfs_inode *inode,
1002 u64 inode_objectid, u64 parent_objectid,
1003 u64 ref_index, char *name, int namelen,
1006 struct btrfs_fs_info *fs_info = root->fs_info;
1009 int victim_name_len;
1010 struct extent_buffer *leaf;
1011 struct btrfs_dir_item *di;
1012 struct btrfs_key search_key;
1013 struct btrfs_inode_extref *extref;
1016 /* Search old style refs */
1017 search_key.objectid = inode_objectid;
1018 search_key.type = BTRFS_INODE_REF_KEY;
1019 search_key.offset = parent_objectid;
1020 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1022 struct btrfs_inode_ref *victim_ref;
1024 unsigned long ptr_end;
1026 leaf = path->nodes[0];
1028 /* are we trying to overwrite a back ref for the root directory
1029 * if so, just jump out, we're done
1031 if (search_key.objectid == search_key.offset)
1034 /* check all the names in this back reference to see
1035 * if they are in the log. if so, we allow them to stay
1036 * otherwise they must be unlinked as a conflict
1038 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1039 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1040 while (ptr < ptr_end) {
1041 victim_ref = (struct btrfs_inode_ref *)ptr;
1042 victim_name_len = btrfs_inode_ref_name_len(leaf,
1044 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1048 read_extent_buffer(leaf, victim_name,
1049 (unsigned long)(victim_ref + 1),
1052 if (!backref_in_log(log_root, &search_key,
1056 inc_nlink(&inode->vfs_inode);
1057 btrfs_release_path(path);
1059 ret = btrfs_unlink_inode(trans, root, dir, inode,
1060 victim_name, victim_name_len);
1064 ret = btrfs_run_delayed_items(trans, fs_info);
1072 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1076 * NOTE: we have searched root tree and checked the
1077 * corresponding ref, it does not need to check again.
1081 btrfs_release_path(path);
1083 /* Same search but for extended refs */
1084 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1085 inode_objectid, parent_objectid, 0,
1087 if (!IS_ERR_OR_NULL(extref)) {
1091 struct inode *victim_parent;
1093 leaf = path->nodes[0];
1095 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1096 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1098 while (cur_offset < item_size) {
1099 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1101 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1103 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1106 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1109 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1112 search_key.objectid = inode_objectid;
1113 search_key.type = BTRFS_INODE_EXTREF_KEY;
1114 search_key.offset = btrfs_extref_hash(parent_objectid,
1118 if (!backref_in_log(log_root, &search_key,
1119 parent_objectid, victim_name,
1122 victim_parent = read_one_inode(root,
1124 if (victim_parent) {
1125 inc_nlink(&inode->vfs_inode);
1126 btrfs_release_path(path);
1128 ret = btrfs_unlink_inode(trans, root,
1129 BTRFS_I(victim_parent),
1134 ret = btrfs_run_delayed_items(
1138 iput(victim_parent);
1147 cur_offset += victim_name_len + sizeof(*extref);
1151 btrfs_release_path(path);
1153 /* look for a conflicting sequence number */
1154 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1155 ref_index, name, namelen, 0);
1156 if (di && !IS_ERR(di)) {
1157 ret = drop_one_dir_item(trans, root, path, dir, di);
1161 btrfs_release_path(path);
1163 /* look for a conflicing name */
1164 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1166 if (di && !IS_ERR(di)) {
1167 ret = drop_one_dir_item(trans, root, path, dir, di);
1171 btrfs_release_path(path);
1176 static int extref_get_fields(struct extent_buffer *eb, int slot,
1177 unsigned long ref_ptr, u32 *namelen, char **name,
1178 u64 *index, u64 *parent_objectid)
1180 struct btrfs_inode_extref *extref;
1182 extref = (struct btrfs_inode_extref *)ref_ptr;
1184 *namelen = btrfs_inode_extref_name_len(eb, extref);
1185 if (!btrfs_is_name_len_valid(eb, slot, (unsigned long)&extref->name,
1189 *name = kmalloc(*namelen, GFP_NOFS);
1193 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1196 *index = btrfs_inode_extref_index(eb, extref);
1197 if (parent_objectid)
1198 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1203 static int ref_get_fields(struct extent_buffer *eb, int slot,
1204 unsigned long ref_ptr, u32 *namelen, char **name,
1207 struct btrfs_inode_ref *ref;
1209 ref = (struct btrfs_inode_ref *)ref_ptr;
1211 *namelen = btrfs_inode_ref_name_len(eb, ref);
1212 if (!btrfs_is_name_len_valid(eb, slot, (unsigned long)(ref + 1),
1216 *name = kmalloc(*namelen, GFP_NOFS);
1220 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1222 *index = btrfs_inode_ref_index(eb, ref);
1228 * replay one inode back reference item found in the log tree.
1229 * eb, slot and key refer to the buffer and key found in the log tree.
1230 * root is the destination we are replaying into, and path is for temp
1231 * use by this function. (it should be released on return).
1233 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1234 struct btrfs_root *root,
1235 struct btrfs_root *log,
1236 struct btrfs_path *path,
1237 struct extent_buffer *eb, int slot,
1238 struct btrfs_key *key)
1240 struct inode *dir = NULL;
1241 struct inode *inode = NULL;
1242 unsigned long ref_ptr;
1243 unsigned long ref_end;
1247 int search_done = 0;
1248 int log_ref_ver = 0;
1249 u64 parent_objectid;
1252 int ref_struct_size;
1254 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1255 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1257 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1258 struct btrfs_inode_extref *r;
1260 ref_struct_size = sizeof(struct btrfs_inode_extref);
1262 r = (struct btrfs_inode_extref *)ref_ptr;
1263 parent_objectid = btrfs_inode_extref_parent(eb, r);
1265 ref_struct_size = sizeof(struct btrfs_inode_ref);
1266 parent_objectid = key->offset;
1268 inode_objectid = key->objectid;
1271 * it is possible that we didn't log all the parent directories
1272 * for a given inode. If we don't find the dir, just don't
1273 * copy the back ref in. The link count fixup code will take
1276 dir = read_one_inode(root, parent_objectid);
1282 inode = read_one_inode(root, inode_objectid);
1288 while (ref_ptr < ref_end) {
1290 ret = extref_get_fields(eb, slot, ref_ptr, &namelen,
1291 &name, &ref_index, &parent_objectid);
1293 * parent object can change from one array
1297 dir = read_one_inode(root, parent_objectid);
1303 ret = ref_get_fields(eb, slot, ref_ptr, &namelen,
1309 /* if we already have a perfect match, we're done */
1310 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1311 btrfs_ino(BTRFS_I(inode)), ref_index,
1314 * look for a conflicting back reference in the
1315 * metadata. if we find one we have to unlink that name
1316 * of the file before we add our new link. Later on, we
1317 * overwrite any existing back reference, and we don't
1318 * want to create dangling pointers in the directory.
1322 ret = __add_inode_ref(trans, root, path, log,
1327 ref_index, name, namelen,
1336 /* insert our name */
1337 ret = btrfs_add_link(trans, BTRFS_I(dir),
1339 name, namelen, 0, ref_index);
1343 btrfs_update_inode(trans, root, inode);
1346 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1355 /* finally write the back reference in the inode */
1356 ret = overwrite_item(trans, root, path, eb, slot, key);
1358 btrfs_release_path(path);
1365 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1366 struct btrfs_root *root, u64 ino)
1370 ret = btrfs_insert_orphan_item(trans, root, ino);
1377 static int count_inode_extrefs(struct btrfs_root *root,
1378 struct btrfs_inode *inode, struct btrfs_path *path)
1382 unsigned int nlink = 0;
1385 u64 inode_objectid = btrfs_ino(inode);
1388 struct btrfs_inode_extref *extref;
1389 struct extent_buffer *leaf;
1392 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1397 leaf = path->nodes[0];
1398 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1399 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1402 while (cur_offset < item_size) {
1403 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1404 name_len = btrfs_inode_extref_name_len(leaf, extref);
1408 cur_offset += name_len + sizeof(*extref);
1412 btrfs_release_path(path);
1414 btrfs_release_path(path);
1416 if (ret < 0 && ret != -ENOENT)
1421 static int count_inode_refs(struct btrfs_root *root,
1422 struct btrfs_inode *inode, struct btrfs_path *path)
1425 struct btrfs_key key;
1426 unsigned int nlink = 0;
1428 unsigned long ptr_end;
1430 u64 ino = btrfs_ino(inode);
1433 key.type = BTRFS_INODE_REF_KEY;
1434 key.offset = (u64)-1;
1437 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1441 if (path->slots[0] == 0)
1446 btrfs_item_key_to_cpu(path->nodes[0], &key,
1448 if (key.objectid != ino ||
1449 key.type != BTRFS_INODE_REF_KEY)
1451 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1452 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1454 while (ptr < ptr_end) {
1455 struct btrfs_inode_ref *ref;
1457 ref = (struct btrfs_inode_ref *)ptr;
1458 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1460 ptr = (unsigned long)(ref + 1) + name_len;
1464 if (key.offset == 0)
1466 if (path->slots[0] > 0) {
1471 btrfs_release_path(path);
1473 btrfs_release_path(path);
1479 * There are a few corners where the link count of the file can't
1480 * be properly maintained during replay. So, instead of adding
1481 * lots of complexity to the log code, we just scan the backrefs
1482 * for any file that has been through replay.
1484 * The scan will update the link count on the inode to reflect the
1485 * number of back refs found. If it goes down to zero, the iput
1486 * will free the inode.
1488 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1489 struct btrfs_root *root,
1490 struct inode *inode)
1492 struct btrfs_path *path;
1495 u64 ino = btrfs_ino(BTRFS_I(inode));
1497 path = btrfs_alloc_path();
1501 ret = count_inode_refs(root, BTRFS_I(inode), path);
1507 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1515 if (nlink != inode->i_nlink) {
1516 set_nlink(inode, nlink);
1517 btrfs_update_inode(trans, root, inode);
1519 BTRFS_I(inode)->index_cnt = (u64)-1;
1521 if (inode->i_nlink == 0) {
1522 if (S_ISDIR(inode->i_mode)) {
1523 ret = replay_dir_deletes(trans, root, NULL, path,
1528 ret = insert_orphan_item(trans, root, ino);
1532 btrfs_free_path(path);
1536 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1537 struct btrfs_root *root,
1538 struct btrfs_path *path)
1541 struct btrfs_key key;
1542 struct inode *inode;
1544 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1545 key.type = BTRFS_ORPHAN_ITEM_KEY;
1546 key.offset = (u64)-1;
1548 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1553 if (path->slots[0] == 0)
1558 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1559 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1560 key.type != BTRFS_ORPHAN_ITEM_KEY)
1563 ret = btrfs_del_item(trans, root, path);
1567 btrfs_release_path(path);
1568 inode = read_one_inode(root, key.offset);
1572 ret = fixup_inode_link_count(trans, root, inode);
1578 * fixup on a directory may create new entries,
1579 * make sure we always look for the highset possible
1582 key.offset = (u64)-1;
1586 btrfs_release_path(path);
1592 * record a given inode in the fixup dir so we can check its link
1593 * count when replay is done. The link count is incremented here
1594 * so the inode won't go away until we check it
1596 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1597 struct btrfs_root *root,
1598 struct btrfs_path *path,
1601 struct btrfs_key key;
1603 struct inode *inode;
1605 inode = read_one_inode(root, objectid);
1609 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1610 key.type = BTRFS_ORPHAN_ITEM_KEY;
1611 key.offset = objectid;
1613 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1615 btrfs_release_path(path);
1617 if (!inode->i_nlink)
1618 set_nlink(inode, 1);
1621 ret = btrfs_update_inode(trans, root, inode);
1622 } else if (ret == -EEXIST) {
1625 BUG(); /* Logic Error */
1633 * when replaying the log for a directory, we only insert names
1634 * for inodes that actually exist. This means an fsync on a directory
1635 * does not implicitly fsync all the new files in it
1637 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1638 struct btrfs_root *root,
1639 u64 dirid, u64 index,
1640 char *name, int name_len,
1641 struct btrfs_key *location)
1643 struct inode *inode;
1647 inode = read_one_inode(root, location->objectid);
1651 dir = read_one_inode(root, dirid);
1657 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1658 name_len, 1, index);
1660 /* FIXME, put inode into FIXUP list */
1668 * Return true if an inode reference exists in the log for the given name,
1669 * inode and parent inode.
1671 static bool name_in_log_ref(struct btrfs_root *log_root,
1672 const char *name, const int name_len,
1673 const u64 dirid, const u64 ino)
1675 struct btrfs_key search_key;
1677 search_key.objectid = ino;
1678 search_key.type = BTRFS_INODE_REF_KEY;
1679 search_key.offset = dirid;
1680 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1683 search_key.type = BTRFS_INODE_EXTREF_KEY;
1684 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1685 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1692 * take a single entry in a log directory item and replay it into
1695 * if a conflicting item exists in the subdirectory already,
1696 * the inode it points to is unlinked and put into the link count
1699 * If a name from the log points to a file or directory that does
1700 * not exist in the FS, it is skipped. fsyncs on directories
1701 * do not force down inodes inside that directory, just changes to the
1702 * names or unlinks in a directory.
1704 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1705 * non-existing inode) and 1 if the name was replayed.
1707 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1708 struct btrfs_root *root,
1709 struct btrfs_path *path,
1710 struct extent_buffer *eb,
1711 struct btrfs_dir_item *di,
1712 struct btrfs_key *key)
1716 struct btrfs_dir_item *dst_di;
1717 struct btrfs_key found_key;
1718 struct btrfs_key log_key;
1723 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1724 bool name_added = false;
1726 dir = read_one_inode(root, key->objectid);
1730 name_len = btrfs_dir_name_len(eb, di);
1731 name = kmalloc(name_len, GFP_NOFS);
1737 log_type = btrfs_dir_type(eb, di);
1738 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1741 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1742 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1747 btrfs_release_path(path);
1749 if (key->type == BTRFS_DIR_ITEM_KEY) {
1750 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1752 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1753 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1762 if (IS_ERR_OR_NULL(dst_di)) {
1763 /* we need a sequence number to insert, so we only
1764 * do inserts for the BTRFS_DIR_INDEX_KEY types
1766 if (key->type != BTRFS_DIR_INDEX_KEY)
1771 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1772 /* the existing item matches the logged item */
1773 if (found_key.objectid == log_key.objectid &&
1774 found_key.type == log_key.type &&
1775 found_key.offset == log_key.offset &&
1776 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1777 update_size = false;
1782 * don't drop the conflicting directory entry if the inode
1783 * for the new entry doesn't exist
1788 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
1792 if (key->type == BTRFS_DIR_INDEX_KEY)
1795 btrfs_release_path(path);
1796 if (!ret && update_size) {
1797 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
1798 ret = btrfs_update_inode(trans, root, dir);
1802 if (!ret && name_added)
1807 if (name_in_log_ref(root->log_root, name, name_len,
1808 key->objectid, log_key.objectid)) {
1809 /* The dentry will be added later. */
1811 update_size = false;
1814 btrfs_release_path(path);
1815 ret = insert_one_name(trans, root, key->objectid, key->offset,
1816 name, name_len, &log_key);
1817 if (ret && ret != -ENOENT && ret != -EEXIST)
1821 update_size = false;
1827 * find all the names in a directory item and reconcile them into
1828 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1829 * one name in a directory item, but the same code gets used for
1830 * both directory index types
1832 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1833 struct btrfs_root *root,
1834 struct btrfs_path *path,
1835 struct extent_buffer *eb, int slot,
1836 struct btrfs_key *key)
1838 struct btrfs_fs_info *fs_info = root->fs_info;
1840 u32 item_size = btrfs_item_size_nr(eb, slot);
1841 struct btrfs_dir_item *di;
1844 unsigned long ptr_end;
1845 struct btrfs_path *fixup_path = NULL;
1847 ptr = btrfs_item_ptr_offset(eb, slot);
1848 ptr_end = ptr + item_size;
1849 while (ptr < ptr_end) {
1850 di = (struct btrfs_dir_item *)ptr;
1851 if (verify_dir_item(fs_info, eb, slot, di))
1853 name_len = btrfs_dir_name_len(eb, di);
1854 ret = replay_one_name(trans, root, path, eb, di, key);
1857 ptr = (unsigned long)(di + 1);
1861 * If this entry refers to a non-directory (directories can not
1862 * have a link count > 1) and it was added in the transaction
1863 * that was not committed, make sure we fixup the link count of
1864 * the inode it the entry points to. Otherwise something like
1865 * the following would result in a directory pointing to an
1866 * inode with a wrong link that does not account for this dir
1874 * ln testdir/bar testdir/bar_link
1875 * ln testdir/foo testdir/foo_link
1876 * xfs_io -c "fsync" testdir/bar
1880 * mount fs, log replay happens
1882 * File foo would remain with a link count of 1 when it has two
1883 * entries pointing to it in the directory testdir. This would
1884 * make it impossible to ever delete the parent directory has
1885 * it would result in stale dentries that can never be deleted.
1887 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1888 struct btrfs_key di_key;
1891 fixup_path = btrfs_alloc_path();
1898 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1899 ret = link_to_fixup_dir(trans, root, fixup_path,
1906 btrfs_free_path(fixup_path);
1911 * directory replay has two parts. There are the standard directory
1912 * items in the log copied from the subvolume, and range items
1913 * created in the log while the subvolume was logged.
1915 * The range items tell us which parts of the key space the log
1916 * is authoritative for. During replay, if a key in the subvolume
1917 * directory is in a logged range item, but not actually in the log
1918 * that means it was deleted from the directory before the fsync
1919 * and should be removed.
1921 static noinline int find_dir_range(struct btrfs_root *root,
1922 struct btrfs_path *path,
1923 u64 dirid, int key_type,
1924 u64 *start_ret, u64 *end_ret)
1926 struct btrfs_key key;
1928 struct btrfs_dir_log_item *item;
1932 if (*start_ret == (u64)-1)
1935 key.objectid = dirid;
1936 key.type = key_type;
1937 key.offset = *start_ret;
1939 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1943 if (path->slots[0] == 0)
1948 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1950 if (key.type != key_type || key.objectid != dirid) {
1954 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1955 struct btrfs_dir_log_item);
1956 found_end = btrfs_dir_log_end(path->nodes[0], item);
1958 if (*start_ret >= key.offset && *start_ret <= found_end) {
1960 *start_ret = key.offset;
1961 *end_ret = found_end;
1966 /* check the next slot in the tree to see if it is a valid item */
1967 nritems = btrfs_header_nritems(path->nodes[0]);
1969 if (path->slots[0] >= nritems) {
1970 ret = btrfs_next_leaf(root, path);
1975 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1977 if (key.type != key_type || key.objectid != dirid) {
1981 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1982 struct btrfs_dir_log_item);
1983 found_end = btrfs_dir_log_end(path->nodes[0], item);
1984 *start_ret = key.offset;
1985 *end_ret = found_end;
1988 btrfs_release_path(path);
1993 * this looks for a given directory item in the log. If the directory
1994 * item is not in the log, the item is removed and the inode it points
1997 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1998 struct btrfs_root *root,
1999 struct btrfs_root *log,
2000 struct btrfs_path *path,
2001 struct btrfs_path *log_path,
2003 struct btrfs_key *dir_key)
2005 struct btrfs_fs_info *fs_info = root->fs_info;
2007 struct extent_buffer *eb;
2010 struct btrfs_dir_item *di;
2011 struct btrfs_dir_item *log_di;
2014 unsigned long ptr_end;
2016 struct inode *inode;
2017 struct btrfs_key location;
2020 eb = path->nodes[0];
2021 slot = path->slots[0];
2022 item_size = btrfs_item_size_nr(eb, slot);
2023 ptr = btrfs_item_ptr_offset(eb, slot);
2024 ptr_end = ptr + item_size;
2025 while (ptr < ptr_end) {
2026 di = (struct btrfs_dir_item *)ptr;
2027 if (verify_dir_item(fs_info, eb, slot, di)) {
2032 name_len = btrfs_dir_name_len(eb, di);
2033 name = kmalloc(name_len, GFP_NOFS);
2038 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2041 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2042 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2045 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2046 log_di = btrfs_lookup_dir_index_item(trans, log,
2052 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
2053 btrfs_dir_item_key_to_cpu(eb, di, &location);
2054 btrfs_release_path(path);
2055 btrfs_release_path(log_path);
2056 inode = read_one_inode(root, location.objectid);
2062 ret = link_to_fixup_dir(trans, root,
2063 path, location.objectid);
2071 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2072 BTRFS_I(inode), name, name_len);
2074 ret = btrfs_run_delayed_items(trans, fs_info);
2080 /* there might still be more names under this key
2081 * check and repeat if required
2083 ret = btrfs_search_slot(NULL, root, dir_key, path,
2089 } else if (IS_ERR(log_di)) {
2091 return PTR_ERR(log_di);
2093 btrfs_release_path(log_path);
2096 ptr = (unsigned long)(di + 1);
2101 btrfs_release_path(path);
2102 btrfs_release_path(log_path);
2106 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2107 struct btrfs_root *root,
2108 struct btrfs_root *log,
2109 struct btrfs_path *path,
2112 struct btrfs_fs_info *fs_info = root->fs_info;
2113 struct btrfs_key search_key;
2114 struct btrfs_path *log_path;
2119 log_path = btrfs_alloc_path();
2123 search_key.objectid = ino;
2124 search_key.type = BTRFS_XATTR_ITEM_KEY;
2125 search_key.offset = 0;
2127 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2131 nritems = btrfs_header_nritems(path->nodes[0]);
2132 for (i = path->slots[0]; i < nritems; i++) {
2133 struct btrfs_key key;
2134 struct btrfs_dir_item *di;
2135 struct btrfs_dir_item *log_di;
2139 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2140 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2145 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2146 total_size = btrfs_item_size_nr(path->nodes[0], i);
2148 while (cur < total_size) {
2149 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2150 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2151 u32 this_len = sizeof(*di) + name_len + data_len;
2154 ret = verify_dir_item(fs_info, path->nodes[0], i, di);
2159 name = kmalloc(name_len, GFP_NOFS);
2164 read_extent_buffer(path->nodes[0], name,
2165 (unsigned long)(di + 1), name_len);
2167 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2169 btrfs_release_path(log_path);
2171 /* Doesn't exist in log tree, so delete it. */
2172 btrfs_release_path(path);
2173 di = btrfs_lookup_xattr(trans, root, path, ino,
2174 name, name_len, -1);
2181 ret = btrfs_delete_one_dir_name(trans, root,
2185 btrfs_release_path(path);
2190 if (IS_ERR(log_di)) {
2191 ret = PTR_ERR(log_di);
2195 di = (struct btrfs_dir_item *)((char *)di + this_len);
2198 ret = btrfs_next_leaf(root, path);
2204 btrfs_free_path(log_path);
2205 btrfs_release_path(path);
2211 * deletion replay happens before we copy any new directory items
2212 * out of the log or out of backreferences from inodes. It
2213 * scans the log to find ranges of keys that log is authoritative for,
2214 * and then scans the directory to find items in those ranges that are
2215 * not present in the log.
2217 * Anything we don't find in the log is unlinked and removed from the
2220 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2221 struct btrfs_root *root,
2222 struct btrfs_root *log,
2223 struct btrfs_path *path,
2224 u64 dirid, int del_all)
2228 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2230 struct btrfs_key dir_key;
2231 struct btrfs_key found_key;
2232 struct btrfs_path *log_path;
2235 dir_key.objectid = dirid;
2236 dir_key.type = BTRFS_DIR_ITEM_KEY;
2237 log_path = btrfs_alloc_path();
2241 dir = read_one_inode(root, dirid);
2242 /* it isn't an error if the inode isn't there, that can happen
2243 * because we replay the deletes before we copy in the inode item
2247 btrfs_free_path(log_path);
2255 range_end = (u64)-1;
2257 ret = find_dir_range(log, path, dirid, key_type,
2258 &range_start, &range_end);
2263 dir_key.offset = range_start;
2266 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2271 nritems = btrfs_header_nritems(path->nodes[0]);
2272 if (path->slots[0] >= nritems) {
2273 ret = btrfs_next_leaf(root, path);
2277 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2279 if (found_key.objectid != dirid ||
2280 found_key.type != dir_key.type)
2283 if (found_key.offset > range_end)
2286 ret = check_item_in_log(trans, root, log, path,
2291 if (found_key.offset == (u64)-1)
2293 dir_key.offset = found_key.offset + 1;
2295 btrfs_release_path(path);
2296 if (range_end == (u64)-1)
2298 range_start = range_end + 1;
2303 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2304 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2305 dir_key.type = BTRFS_DIR_INDEX_KEY;
2306 btrfs_release_path(path);
2310 btrfs_release_path(path);
2311 btrfs_free_path(log_path);
2317 * the process_func used to replay items from the log tree. This
2318 * gets called in two different stages. The first stage just looks
2319 * for inodes and makes sure they are all copied into the subvolume.
2321 * The second stage copies all the other item types from the log into
2322 * the subvolume. The two stage approach is slower, but gets rid of
2323 * lots of complexity around inodes referencing other inodes that exist
2324 * only in the log (references come from either directory items or inode
2327 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2328 struct walk_control *wc, u64 gen)
2331 struct btrfs_path *path;
2332 struct btrfs_root *root = wc->replay_dest;
2333 struct btrfs_key key;
2338 ret = btrfs_read_buffer(eb, gen);
2342 level = btrfs_header_level(eb);
2347 path = btrfs_alloc_path();
2351 nritems = btrfs_header_nritems(eb);
2352 for (i = 0; i < nritems; i++) {
2353 btrfs_item_key_to_cpu(eb, &key, i);
2355 /* inode keys are done during the first stage */
2356 if (key.type == BTRFS_INODE_ITEM_KEY &&
2357 wc->stage == LOG_WALK_REPLAY_INODES) {
2358 struct btrfs_inode_item *inode_item;
2361 inode_item = btrfs_item_ptr(eb, i,
2362 struct btrfs_inode_item);
2363 ret = replay_xattr_deletes(wc->trans, root, log,
2364 path, key.objectid);
2367 mode = btrfs_inode_mode(eb, inode_item);
2368 if (S_ISDIR(mode)) {
2369 ret = replay_dir_deletes(wc->trans,
2370 root, log, path, key.objectid, 0);
2374 ret = overwrite_item(wc->trans, root, path,
2379 /* for regular files, make sure corresponding
2380 * orphan item exist. extents past the new EOF
2381 * will be truncated later by orphan cleanup.
2383 if (S_ISREG(mode)) {
2384 ret = insert_orphan_item(wc->trans, root,
2390 ret = link_to_fixup_dir(wc->trans, root,
2391 path, key.objectid);
2396 if (key.type == BTRFS_DIR_INDEX_KEY &&
2397 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2398 ret = replay_one_dir_item(wc->trans, root, path,
2404 if (wc->stage < LOG_WALK_REPLAY_ALL)
2407 /* these keys are simply copied */
2408 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2409 ret = overwrite_item(wc->trans, root, path,
2413 } else if (key.type == BTRFS_INODE_REF_KEY ||
2414 key.type == BTRFS_INODE_EXTREF_KEY) {
2415 ret = add_inode_ref(wc->trans, root, log, path,
2417 if (ret && ret != -ENOENT)
2420 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2421 ret = replay_one_extent(wc->trans, root, path,
2425 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2426 ret = replay_one_dir_item(wc->trans, root, path,
2432 btrfs_free_path(path);
2436 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2437 struct btrfs_root *root,
2438 struct btrfs_path *path, int *level,
2439 struct walk_control *wc)
2441 struct btrfs_fs_info *fs_info = root->fs_info;
2445 struct extent_buffer *next;
2446 struct extent_buffer *cur;
2447 struct extent_buffer *parent;
2451 WARN_ON(*level < 0);
2452 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2454 while (*level > 0) {
2455 WARN_ON(*level < 0);
2456 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2457 cur = path->nodes[*level];
2459 WARN_ON(btrfs_header_level(cur) != *level);
2461 if (path->slots[*level] >=
2462 btrfs_header_nritems(cur))
2465 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2466 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2467 blocksize = fs_info->nodesize;
2469 parent = path->nodes[*level];
2470 root_owner = btrfs_header_owner(parent);
2472 next = btrfs_find_create_tree_block(fs_info, bytenr);
2474 return PTR_ERR(next);
2477 ret = wc->process_func(root, next, wc, ptr_gen);
2479 free_extent_buffer(next);
2483 path->slots[*level]++;
2485 ret = btrfs_read_buffer(next, ptr_gen);
2487 free_extent_buffer(next);
2492 btrfs_tree_lock(next);
2493 btrfs_set_lock_blocking(next);
2494 clean_tree_block(fs_info, next);
2495 btrfs_wait_tree_block_writeback(next);
2496 btrfs_tree_unlock(next);
2499 WARN_ON(root_owner !=
2500 BTRFS_TREE_LOG_OBJECTID);
2501 ret = btrfs_free_and_pin_reserved_extent(
2505 free_extent_buffer(next);
2509 free_extent_buffer(next);
2512 ret = btrfs_read_buffer(next, ptr_gen);
2514 free_extent_buffer(next);
2518 WARN_ON(*level <= 0);
2519 if (path->nodes[*level-1])
2520 free_extent_buffer(path->nodes[*level-1]);
2521 path->nodes[*level-1] = next;
2522 *level = btrfs_header_level(next);
2523 path->slots[*level] = 0;
2526 WARN_ON(*level < 0);
2527 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2529 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2535 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2536 struct btrfs_root *root,
2537 struct btrfs_path *path, int *level,
2538 struct walk_control *wc)
2540 struct btrfs_fs_info *fs_info = root->fs_info;
2546 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2547 slot = path->slots[i];
2548 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2551 WARN_ON(*level == 0);
2554 struct extent_buffer *parent;
2555 if (path->nodes[*level] == root->node)
2556 parent = path->nodes[*level];
2558 parent = path->nodes[*level + 1];
2560 root_owner = btrfs_header_owner(parent);
2561 ret = wc->process_func(root, path->nodes[*level], wc,
2562 btrfs_header_generation(path->nodes[*level]));
2567 struct extent_buffer *next;
2569 next = path->nodes[*level];
2572 btrfs_tree_lock(next);
2573 btrfs_set_lock_blocking(next);
2574 clean_tree_block(fs_info, next);
2575 btrfs_wait_tree_block_writeback(next);
2576 btrfs_tree_unlock(next);
2579 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2580 ret = btrfs_free_and_pin_reserved_extent(
2582 path->nodes[*level]->start,
2583 path->nodes[*level]->len);
2587 free_extent_buffer(path->nodes[*level]);
2588 path->nodes[*level] = NULL;
2596 * drop the reference count on the tree rooted at 'snap'. This traverses
2597 * the tree freeing any blocks that have a ref count of zero after being
2600 static int walk_log_tree(struct btrfs_trans_handle *trans,
2601 struct btrfs_root *log, struct walk_control *wc)
2603 struct btrfs_fs_info *fs_info = log->fs_info;
2607 struct btrfs_path *path;
2610 path = btrfs_alloc_path();
2614 level = btrfs_header_level(log->node);
2616 path->nodes[level] = log->node;
2617 extent_buffer_get(log->node);
2618 path->slots[level] = 0;
2621 wret = walk_down_log_tree(trans, log, path, &level, wc);
2629 wret = walk_up_log_tree(trans, log, path, &level, wc);
2638 /* was the root node processed? if not, catch it here */
2639 if (path->nodes[orig_level]) {
2640 ret = wc->process_func(log, path->nodes[orig_level], wc,
2641 btrfs_header_generation(path->nodes[orig_level]));
2645 struct extent_buffer *next;
2647 next = path->nodes[orig_level];
2650 btrfs_tree_lock(next);
2651 btrfs_set_lock_blocking(next);
2652 clean_tree_block(fs_info, next);
2653 btrfs_wait_tree_block_writeback(next);
2654 btrfs_tree_unlock(next);
2657 WARN_ON(log->root_key.objectid !=
2658 BTRFS_TREE_LOG_OBJECTID);
2659 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2660 next->start, next->len);
2667 btrfs_free_path(path);
2672 * helper function to update the item for a given subvolumes log root
2673 * in the tree of log roots
2675 static int update_log_root(struct btrfs_trans_handle *trans,
2676 struct btrfs_root *log)
2678 struct btrfs_fs_info *fs_info = log->fs_info;
2681 if (log->log_transid == 1) {
2682 /* insert root item on the first sync */
2683 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2684 &log->root_key, &log->root_item);
2686 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2687 &log->root_key, &log->root_item);
2692 static void wait_log_commit(struct btrfs_root *root, int transid)
2695 int index = transid % 2;
2698 * we only allow two pending log transactions at a time,
2699 * so we know that if ours is more than 2 older than the
2700 * current transaction, we're done
2703 prepare_to_wait(&root->log_commit_wait[index],
2704 &wait, TASK_UNINTERRUPTIBLE);
2705 mutex_unlock(&root->log_mutex);
2707 if (root->log_transid_committed < transid &&
2708 atomic_read(&root->log_commit[index]))
2711 finish_wait(&root->log_commit_wait[index], &wait);
2712 mutex_lock(&root->log_mutex);
2713 } while (root->log_transid_committed < transid &&
2714 atomic_read(&root->log_commit[index]));
2717 static void wait_for_writer(struct btrfs_root *root)
2721 while (atomic_read(&root->log_writers)) {
2722 prepare_to_wait(&root->log_writer_wait,
2723 &wait, TASK_UNINTERRUPTIBLE);
2724 mutex_unlock(&root->log_mutex);
2725 if (atomic_read(&root->log_writers))
2727 finish_wait(&root->log_writer_wait, &wait);
2728 mutex_lock(&root->log_mutex);
2732 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2733 struct btrfs_log_ctx *ctx)
2738 mutex_lock(&root->log_mutex);
2739 list_del_init(&ctx->list);
2740 mutex_unlock(&root->log_mutex);
2744 * Invoked in log mutex context, or be sure there is no other task which
2745 * can access the list.
2747 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2748 int index, int error)
2750 struct btrfs_log_ctx *ctx;
2751 struct btrfs_log_ctx *safe;
2753 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2754 list_del_init(&ctx->list);
2755 ctx->log_ret = error;
2758 INIT_LIST_HEAD(&root->log_ctxs[index]);
2762 * btrfs_sync_log does sends a given tree log down to the disk and
2763 * updates the super blocks to record it. When this call is done,
2764 * you know that any inodes previously logged are safely on disk only
2767 * Any other return value means you need to call btrfs_commit_transaction.
2768 * Some of the edge cases for fsyncing directories that have had unlinks
2769 * or renames done in the past mean that sometimes the only safe
2770 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2771 * that has happened.
2773 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2774 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2780 struct btrfs_fs_info *fs_info = root->fs_info;
2781 struct btrfs_root *log = root->log_root;
2782 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2783 int log_transid = 0;
2784 struct btrfs_log_ctx root_log_ctx;
2785 struct blk_plug plug;
2787 mutex_lock(&root->log_mutex);
2788 log_transid = ctx->log_transid;
2789 if (root->log_transid_committed >= log_transid) {
2790 mutex_unlock(&root->log_mutex);
2791 return ctx->log_ret;
2794 index1 = log_transid % 2;
2795 if (atomic_read(&root->log_commit[index1])) {
2796 wait_log_commit(root, log_transid);
2797 mutex_unlock(&root->log_mutex);
2798 return ctx->log_ret;
2800 ASSERT(log_transid == root->log_transid);
2801 atomic_set(&root->log_commit[index1], 1);
2803 /* wait for previous tree log sync to complete */
2804 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2805 wait_log_commit(root, log_transid - 1);
2808 int batch = atomic_read(&root->log_batch);
2809 /* when we're on an ssd, just kick the log commit out */
2810 if (!btrfs_test_opt(fs_info, SSD) &&
2811 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2812 mutex_unlock(&root->log_mutex);
2813 schedule_timeout_uninterruptible(1);
2814 mutex_lock(&root->log_mutex);
2816 wait_for_writer(root);
2817 if (batch == atomic_read(&root->log_batch))
2821 /* bail out if we need to do a full commit */
2822 if (btrfs_need_log_full_commit(fs_info, trans)) {
2824 btrfs_free_logged_extents(log, log_transid);
2825 mutex_unlock(&root->log_mutex);
2829 if (log_transid % 2 == 0)
2830 mark = EXTENT_DIRTY;
2834 /* we start IO on all the marked extents here, but we don't actually
2835 * wait for them until later.
2837 blk_start_plug(&plug);
2838 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2840 blk_finish_plug(&plug);
2841 btrfs_abort_transaction(trans, ret);
2842 btrfs_free_logged_extents(log, log_transid);
2843 btrfs_set_log_full_commit(fs_info, trans);
2844 mutex_unlock(&root->log_mutex);
2848 btrfs_set_root_node(&log->root_item, log->node);
2850 root->log_transid++;
2851 log->log_transid = root->log_transid;
2852 root->log_start_pid = 0;
2854 * IO has been started, blocks of the log tree have WRITTEN flag set
2855 * in their headers. new modifications of the log will be written to
2856 * new positions. so it's safe to allow log writers to go in.
2858 mutex_unlock(&root->log_mutex);
2860 btrfs_init_log_ctx(&root_log_ctx, NULL);
2862 mutex_lock(&log_root_tree->log_mutex);
2863 atomic_inc(&log_root_tree->log_batch);
2864 atomic_inc(&log_root_tree->log_writers);
2866 index2 = log_root_tree->log_transid % 2;
2867 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2868 root_log_ctx.log_transid = log_root_tree->log_transid;
2870 mutex_unlock(&log_root_tree->log_mutex);
2872 ret = update_log_root(trans, log);
2874 mutex_lock(&log_root_tree->log_mutex);
2875 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2877 * Implicit memory barrier after atomic_dec_and_test
2879 if (waitqueue_active(&log_root_tree->log_writer_wait))
2880 wake_up(&log_root_tree->log_writer_wait);
2884 if (!list_empty(&root_log_ctx.list))
2885 list_del_init(&root_log_ctx.list);
2887 blk_finish_plug(&plug);
2888 btrfs_set_log_full_commit(fs_info, trans);
2890 if (ret != -ENOSPC) {
2891 btrfs_abort_transaction(trans, ret);
2892 mutex_unlock(&log_root_tree->log_mutex);
2895 btrfs_wait_tree_log_extents(log, mark);
2896 btrfs_free_logged_extents(log, log_transid);
2897 mutex_unlock(&log_root_tree->log_mutex);
2902 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2903 blk_finish_plug(&plug);
2904 list_del_init(&root_log_ctx.list);
2905 mutex_unlock(&log_root_tree->log_mutex);
2906 ret = root_log_ctx.log_ret;
2910 index2 = root_log_ctx.log_transid % 2;
2911 if (atomic_read(&log_root_tree->log_commit[index2])) {
2912 blk_finish_plug(&plug);
2913 ret = btrfs_wait_tree_log_extents(log, mark);
2914 btrfs_wait_logged_extents(trans, log, log_transid);
2915 wait_log_commit(log_root_tree,
2916 root_log_ctx.log_transid);
2917 mutex_unlock(&log_root_tree->log_mutex);
2919 ret = root_log_ctx.log_ret;
2922 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
2923 atomic_set(&log_root_tree->log_commit[index2], 1);
2925 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2926 wait_log_commit(log_root_tree,
2927 root_log_ctx.log_transid - 1);
2930 wait_for_writer(log_root_tree);
2933 * now that we've moved on to the tree of log tree roots,
2934 * check the full commit flag again
2936 if (btrfs_need_log_full_commit(fs_info, trans)) {
2937 blk_finish_plug(&plug);
2938 btrfs_wait_tree_log_extents(log, mark);
2939 btrfs_free_logged_extents(log, log_transid);
2940 mutex_unlock(&log_root_tree->log_mutex);
2942 goto out_wake_log_root;
2945 ret = btrfs_write_marked_extents(fs_info,
2946 &log_root_tree->dirty_log_pages,
2947 EXTENT_DIRTY | EXTENT_NEW);
2948 blk_finish_plug(&plug);
2950 btrfs_set_log_full_commit(fs_info, trans);
2951 btrfs_abort_transaction(trans, ret);
2952 btrfs_free_logged_extents(log, log_transid);
2953 mutex_unlock(&log_root_tree->log_mutex);
2954 goto out_wake_log_root;
2956 ret = btrfs_wait_tree_log_extents(log, mark);
2958 ret = btrfs_wait_tree_log_extents(log_root_tree,
2959 EXTENT_NEW | EXTENT_DIRTY);
2961 btrfs_set_log_full_commit(fs_info, trans);
2962 btrfs_free_logged_extents(log, log_transid);
2963 mutex_unlock(&log_root_tree->log_mutex);
2964 goto out_wake_log_root;
2966 btrfs_wait_logged_extents(trans, log, log_transid);
2968 btrfs_set_super_log_root(fs_info->super_for_commit,
2969 log_root_tree->node->start);
2970 btrfs_set_super_log_root_level(fs_info->super_for_commit,
2971 btrfs_header_level(log_root_tree->node));
2973 log_root_tree->log_transid++;
2974 mutex_unlock(&log_root_tree->log_mutex);
2977 * nobody else is going to jump in and write the the ctree
2978 * super here because the log_commit atomic below is protecting
2979 * us. We must be called with a transaction handle pinning
2980 * the running transaction open, so a full commit can't hop
2981 * in and cause problems either.
2983 ret = write_all_supers(fs_info, 1);
2985 btrfs_set_log_full_commit(fs_info, trans);
2986 btrfs_abort_transaction(trans, ret);
2987 goto out_wake_log_root;
2990 mutex_lock(&root->log_mutex);
2991 if (root->last_log_commit < log_transid)
2992 root->last_log_commit = log_transid;
2993 mutex_unlock(&root->log_mutex);
2996 mutex_lock(&log_root_tree->log_mutex);
2997 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2999 log_root_tree->log_transid_committed++;
3000 atomic_set(&log_root_tree->log_commit[index2], 0);
3001 mutex_unlock(&log_root_tree->log_mutex);
3004 * The barrier before waitqueue_active is implied by mutex_unlock
3006 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
3007 wake_up(&log_root_tree->log_commit_wait[index2]);
3009 mutex_lock(&root->log_mutex);
3010 btrfs_remove_all_log_ctxs(root, index1, ret);
3011 root->log_transid_committed++;
3012 atomic_set(&root->log_commit[index1], 0);
3013 mutex_unlock(&root->log_mutex);
3016 * The barrier before waitqueue_active is implied by mutex_unlock
3018 if (waitqueue_active(&root->log_commit_wait[index1]))
3019 wake_up(&root->log_commit_wait[index1]);
3023 static void free_log_tree(struct btrfs_trans_handle *trans,
3024 struct btrfs_root *log)
3029 struct walk_control wc = {
3031 .process_func = process_one_buffer
3034 ret = walk_log_tree(trans, log, &wc);
3035 /* I don't think this can happen but just in case */
3037 btrfs_abort_transaction(trans, ret);
3040 ret = find_first_extent_bit(&log->dirty_log_pages,
3041 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
3046 clear_extent_bits(&log->dirty_log_pages, start, end,
3047 EXTENT_DIRTY | EXTENT_NEW);
3051 * We may have short-circuited the log tree with the full commit logic
3052 * and left ordered extents on our list, so clear these out to keep us
3053 * from leaking inodes and memory.
3055 btrfs_free_logged_extents(log, 0);
3056 btrfs_free_logged_extents(log, 1);
3058 free_extent_buffer(log->node);
3063 * free all the extents used by the tree log. This should be called
3064 * at commit time of the full transaction
3066 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3068 if (root->log_root) {
3069 free_log_tree(trans, root->log_root);
3070 root->log_root = NULL;
3075 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3076 struct btrfs_fs_info *fs_info)
3078 if (fs_info->log_root_tree) {
3079 free_log_tree(trans, fs_info->log_root_tree);
3080 fs_info->log_root_tree = NULL;
3086 * If both a file and directory are logged, and unlinks or renames are
3087 * mixed in, we have a few interesting corners:
3089 * create file X in dir Y
3090 * link file X to X.link in dir Y
3092 * unlink file X but leave X.link
3095 * After a crash we would expect only X.link to exist. But file X
3096 * didn't get fsync'd again so the log has back refs for X and X.link.
3098 * We solve this by removing directory entries and inode backrefs from the
3099 * log when a file that was logged in the current transaction is
3100 * unlinked. Any later fsync will include the updated log entries, and
3101 * we'll be able to reconstruct the proper directory items from backrefs.
3103 * This optimizations allows us to avoid relogging the entire inode
3104 * or the entire directory.
3106 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3107 struct btrfs_root *root,
3108 const char *name, int name_len,
3109 struct btrfs_inode *dir, u64 index)
3111 struct btrfs_root *log;
3112 struct btrfs_dir_item *di;
3113 struct btrfs_path *path;
3117 u64 dir_ino = btrfs_ino(dir);
3119 if (dir->logged_trans < trans->transid)
3122 ret = join_running_log_trans(root);
3126 mutex_lock(&dir->log_mutex);
3128 log = root->log_root;
3129 path = btrfs_alloc_path();
3135 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3136 name, name_len, -1);
3142 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3143 bytes_del += name_len;
3149 btrfs_release_path(path);
3150 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3151 index, name, name_len, -1);
3157 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3158 bytes_del += name_len;
3165 /* update the directory size in the log to reflect the names
3169 struct btrfs_key key;
3171 key.objectid = dir_ino;
3173 key.type = BTRFS_INODE_ITEM_KEY;
3174 btrfs_release_path(path);
3176 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3182 struct btrfs_inode_item *item;
3185 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3186 struct btrfs_inode_item);
3187 i_size = btrfs_inode_size(path->nodes[0], item);
3188 if (i_size > bytes_del)
3189 i_size -= bytes_del;
3192 btrfs_set_inode_size(path->nodes[0], item, i_size);
3193 btrfs_mark_buffer_dirty(path->nodes[0]);
3196 btrfs_release_path(path);
3199 btrfs_free_path(path);
3201 mutex_unlock(&dir->log_mutex);
3202 if (ret == -ENOSPC) {
3203 btrfs_set_log_full_commit(root->fs_info, trans);
3206 btrfs_abort_transaction(trans, ret);
3208 btrfs_end_log_trans(root);
3213 /* see comments for btrfs_del_dir_entries_in_log */
3214 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3215 struct btrfs_root *root,
3216 const char *name, int name_len,
3217 struct btrfs_inode *inode, u64 dirid)
3219 struct btrfs_fs_info *fs_info = root->fs_info;
3220 struct btrfs_root *log;
3224 if (inode->logged_trans < trans->transid)
3227 ret = join_running_log_trans(root);
3230 log = root->log_root;
3231 mutex_lock(&inode->log_mutex);
3233 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3235 mutex_unlock(&inode->log_mutex);
3236 if (ret == -ENOSPC) {
3237 btrfs_set_log_full_commit(fs_info, trans);
3239 } else if (ret < 0 && ret != -ENOENT)
3240 btrfs_abort_transaction(trans, ret);
3241 btrfs_end_log_trans(root);
3247 * creates a range item in the log for 'dirid'. first_offset and
3248 * last_offset tell us which parts of the key space the log should
3249 * be considered authoritative for.
3251 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3252 struct btrfs_root *log,
3253 struct btrfs_path *path,
3254 int key_type, u64 dirid,
3255 u64 first_offset, u64 last_offset)
3258 struct btrfs_key key;
3259 struct btrfs_dir_log_item *item;
3261 key.objectid = dirid;
3262 key.offset = first_offset;
3263 if (key_type == BTRFS_DIR_ITEM_KEY)
3264 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3266 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3267 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3271 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3272 struct btrfs_dir_log_item);
3273 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3274 btrfs_mark_buffer_dirty(path->nodes[0]);
3275 btrfs_release_path(path);
3280 * log all the items included in the current transaction for a given
3281 * directory. This also creates the range items in the log tree required
3282 * to replay anything deleted before the fsync
3284 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3285 struct btrfs_root *root, struct btrfs_inode *inode,
3286 struct btrfs_path *path,
3287 struct btrfs_path *dst_path, int key_type,
3288 struct btrfs_log_ctx *ctx,
3289 u64 min_offset, u64 *last_offset_ret)
3291 struct btrfs_key min_key;
3292 struct btrfs_root *log = root->log_root;
3293 struct extent_buffer *src;
3298 u64 first_offset = min_offset;
3299 u64 last_offset = (u64)-1;
3300 u64 ino = btrfs_ino(inode);
3302 log = root->log_root;
3304 min_key.objectid = ino;
3305 min_key.type = key_type;
3306 min_key.offset = min_offset;
3308 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3311 * we didn't find anything from this transaction, see if there
3312 * is anything at all
3314 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3315 min_key.objectid = ino;
3316 min_key.type = key_type;
3317 min_key.offset = (u64)-1;
3318 btrfs_release_path(path);
3319 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3321 btrfs_release_path(path);
3324 ret = btrfs_previous_item(root, path, ino, key_type);
3326 /* if ret == 0 there are items for this type,
3327 * create a range to tell us the last key of this type.
3328 * otherwise, there are no items in this directory after
3329 * *min_offset, and we create a range to indicate that.
3332 struct btrfs_key tmp;
3333 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3335 if (key_type == tmp.type)
3336 first_offset = max(min_offset, tmp.offset) + 1;
3341 /* go backward to find any previous key */
3342 ret = btrfs_previous_item(root, path, ino, key_type);
3344 struct btrfs_key tmp;
3345 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3346 if (key_type == tmp.type) {
3347 first_offset = tmp.offset;
3348 ret = overwrite_item(trans, log, dst_path,
3349 path->nodes[0], path->slots[0],
3357 btrfs_release_path(path);
3359 /* find the first key from this transaction again */
3360 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3361 if (WARN_ON(ret != 0))
3365 * we have a block from this transaction, log every item in it
3366 * from our directory
3369 struct btrfs_key tmp;
3370 src = path->nodes[0];
3371 nritems = btrfs_header_nritems(src);
3372 for (i = path->slots[0]; i < nritems; i++) {
3373 struct btrfs_dir_item *di;
3375 btrfs_item_key_to_cpu(src, &min_key, i);
3377 if (min_key.objectid != ino || min_key.type != key_type)
3379 ret = overwrite_item(trans, log, dst_path, src, i,
3387 * We must make sure that when we log a directory entry,
3388 * the corresponding inode, after log replay, has a
3389 * matching link count. For example:
3395 * xfs_io -c "fsync" mydir
3397 * <mount fs and log replay>
3399 * Would result in a fsync log that when replayed, our
3400 * file inode would have a link count of 1, but we get
3401 * two directory entries pointing to the same inode.
3402 * After removing one of the names, it would not be
3403 * possible to remove the other name, which resulted
3404 * always in stale file handle errors, and would not
3405 * be possible to rmdir the parent directory, since
3406 * its i_size could never decrement to the value
3407 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3409 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3410 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3412 (btrfs_dir_transid(src, di) == trans->transid ||
3413 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3414 tmp.type != BTRFS_ROOT_ITEM_KEY)
3415 ctx->log_new_dentries = true;
3417 path->slots[0] = nritems;
3420 * look ahead to the next item and see if it is also
3421 * from this directory and from this transaction
3423 ret = btrfs_next_leaf(root, path);
3425 last_offset = (u64)-1;
3428 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3429 if (tmp.objectid != ino || tmp.type != key_type) {
3430 last_offset = (u64)-1;
3433 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3434 ret = overwrite_item(trans, log, dst_path,
3435 path->nodes[0], path->slots[0],
3440 last_offset = tmp.offset;
3445 btrfs_release_path(path);
3446 btrfs_release_path(dst_path);
3449 *last_offset_ret = last_offset;
3451 * insert the log range keys to indicate where the log
3454 ret = insert_dir_log_key(trans, log, path, key_type,
3455 ino, first_offset, last_offset);
3463 * logging directories is very similar to logging inodes, We find all the items
3464 * from the current transaction and write them to the log.
3466 * The recovery code scans the directory in the subvolume, and if it finds a
3467 * key in the range logged that is not present in the log tree, then it means
3468 * that dir entry was unlinked during the transaction.
3470 * In order for that scan to work, we must include one key smaller than
3471 * the smallest logged by this transaction and one key larger than the largest
3472 * key logged by this transaction.
3474 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3475 struct btrfs_root *root, struct btrfs_inode *inode,
3476 struct btrfs_path *path,
3477 struct btrfs_path *dst_path,
3478 struct btrfs_log_ctx *ctx)
3483 int key_type = BTRFS_DIR_ITEM_KEY;
3489 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3490 ctx, min_key, &max_key);
3493 if (max_key == (u64)-1)
3495 min_key = max_key + 1;
3498 if (key_type == BTRFS_DIR_ITEM_KEY) {
3499 key_type = BTRFS_DIR_INDEX_KEY;
3506 * a helper function to drop items from the log before we relog an
3507 * inode. max_key_type indicates the highest item type to remove.
3508 * This cannot be run for file data extents because it does not
3509 * free the extents they point to.
3511 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3512 struct btrfs_root *log,
3513 struct btrfs_path *path,
3514 u64 objectid, int max_key_type)
3517 struct btrfs_key key;
3518 struct btrfs_key found_key;
3521 key.objectid = objectid;
3522 key.type = max_key_type;
3523 key.offset = (u64)-1;
3526 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3527 BUG_ON(ret == 0); /* Logic error */
3531 if (path->slots[0] == 0)
3535 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3538 if (found_key.objectid != objectid)
3541 found_key.offset = 0;
3543 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3546 ret = btrfs_del_items(trans, log, path, start_slot,
3547 path->slots[0] - start_slot + 1);
3549 * If start slot isn't 0 then we don't need to re-search, we've
3550 * found the last guy with the objectid in this tree.
3552 if (ret || start_slot != 0)
3554 btrfs_release_path(path);
3556 btrfs_release_path(path);
3562 static void fill_inode_item(struct btrfs_trans_handle *trans,
3563 struct extent_buffer *leaf,
3564 struct btrfs_inode_item *item,
3565 struct inode *inode, int log_inode_only,
3568 struct btrfs_map_token token;
3570 btrfs_init_map_token(&token);
3572 if (log_inode_only) {
3573 /* set the generation to zero so the recover code
3574 * can tell the difference between an logging
3575 * just to say 'this inode exists' and a logging
3576 * to say 'update this inode with these values'
3578 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3579 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3581 btrfs_set_token_inode_generation(leaf, item,
3582 BTRFS_I(inode)->generation,
3584 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3587 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3588 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3589 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3590 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3592 btrfs_set_token_timespec_sec(leaf, &item->atime,
3593 inode->i_atime.tv_sec, &token);
3594 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3595 inode->i_atime.tv_nsec, &token);
3597 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3598 inode->i_mtime.tv_sec, &token);
3599 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3600 inode->i_mtime.tv_nsec, &token);
3602 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3603 inode->i_ctime.tv_sec, &token);
3604 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3605 inode->i_ctime.tv_nsec, &token);
3607 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3610 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3611 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3612 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3613 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3614 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3617 static int log_inode_item(struct btrfs_trans_handle *trans,
3618 struct btrfs_root *log, struct btrfs_path *path,
3619 struct btrfs_inode *inode)
3621 struct btrfs_inode_item *inode_item;
3624 ret = btrfs_insert_empty_item(trans, log, path,
3625 &inode->location, sizeof(*inode_item));
3626 if (ret && ret != -EEXIST)
3628 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3629 struct btrfs_inode_item);
3630 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3632 btrfs_release_path(path);
3636 static noinline int copy_items(struct btrfs_trans_handle *trans,
3637 struct btrfs_inode *inode,
3638 struct btrfs_path *dst_path,
3639 struct btrfs_path *src_path, u64 *last_extent,
3640 int start_slot, int nr, int inode_only,
3643 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3644 unsigned long src_offset;
3645 unsigned long dst_offset;
3646 struct btrfs_root *log = inode->root->log_root;
3647 struct btrfs_file_extent_item *extent;
3648 struct btrfs_inode_item *inode_item;
3649 struct extent_buffer *src = src_path->nodes[0];
3650 struct btrfs_key first_key, last_key, key;
3652 struct btrfs_key *ins_keys;
3656 struct list_head ordered_sums;
3657 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3658 bool has_extents = false;
3659 bool need_find_last_extent = true;
3662 INIT_LIST_HEAD(&ordered_sums);
3664 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3665 nr * sizeof(u32), GFP_NOFS);
3669 first_key.objectid = (u64)-1;
3671 ins_sizes = (u32 *)ins_data;
3672 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3674 for (i = 0; i < nr; i++) {
3675 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3676 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3678 ret = btrfs_insert_empty_items(trans, log, dst_path,
3679 ins_keys, ins_sizes, nr);
3685 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3686 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3687 dst_path->slots[0]);
3689 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3692 last_key = ins_keys[i];
3694 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3695 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3697 struct btrfs_inode_item);
3698 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3700 inode_only == LOG_INODE_EXISTS,
3703 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3704 src_offset, ins_sizes[i]);
3708 * We set need_find_last_extent here in case we know we were
3709 * processing other items and then walk into the first extent in
3710 * the inode. If we don't hit an extent then nothing changes,
3711 * we'll do the last search the next time around.
3713 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3715 if (first_key.objectid == (u64)-1)
3716 first_key = ins_keys[i];
3718 need_find_last_extent = false;
3721 /* take a reference on file data extents so that truncates
3722 * or deletes of this inode don't have to relog the inode
3725 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3728 extent = btrfs_item_ptr(src, start_slot + i,
3729 struct btrfs_file_extent_item);
3731 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3734 found_type = btrfs_file_extent_type(src, extent);
3735 if (found_type == BTRFS_FILE_EXTENT_REG) {
3737 ds = btrfs_file_extent_disk_bytenr(src,
3739 /* ds == 0 is a hole */
3743 dl = btrfs_file_extent_disk_num_bytes(src,
3745 cs = btrfs_file_extent_offset(src, extent);
3746 cl = btrfs_file_extent_num_bytes(src,
3748 if (btrfs_file_extent_compression(src,
3754 ret = btrfs_lookup_csums_range(
3756 ds + cs, ds + cs + cl - 1,
3759 btrfs_release_path(dst_path);
3767 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3768 btrfs_release_path(dst_path);
3772 * we have to do this after the loop above to avoid changing the
3773 * log tree while trying to change the log tree.
3776 while (!list_empty(&ordered_sums)) {
3777 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3778 struct btrfs_ordered_sum,
3781 ret = btrfs_csum_file_blocks(trans, log, sums);
3782 list_del(&sums->list);
3789 if (need_find_last_extent && *last_extent == first_key.offset) {
3791 * We don't have any leafs between our current one and the one
3792 * we processed before that can have file extent items for our
3793 * inode (and have a generation number smaller than our current
3796 need_find_last_extent = false;
3800 * Because we use btrfs_search_forward we could skip leaves that were
3801 * not modified and then assume *last_extent is valid when it really
3802 * isn't. So back up to the previous leaf and read the end of the last
3803 * extent before we go and fill in holes.
3805 if (need_find_last_extent) {
3808 ret = btrfs_prev_leaf(inode->root, src_path);
3813 if (src_path->slots[0])
3814 src_path->slots[0]--;
3815 src = src_path->nodes[0];
3816 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3817 if (key.objectid != btrfs_ino(inode) ||
3818 key.type != BTRFS_EXTENT_DATA_KEY)
3820 extent = btrfs_item_ptr(src, src_path->slots[0],
3821 struct btrfs_file_extent_item);
3822 if (btrfs_file_extent_type(src, extent) ==
3823 BTRFS_FILE_EXTENT_INLINE) {
3824 len = btrfs_file_extent_inline_len(src,
3827 *last_extent = ALIGN(key.offset + len,
3828 fs_info->sectorsize);
3830 len = btrfs_file_extent_num_bytes(src, extent);
3831 *last_extent = key.offset + len;
3835 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3836 * things could have happened
3838 * 1) A merge could have happened, so we could currently be on a leaf
3839 * that holds what we were copying in the first place.
3840 * 2) A split could have happened, and now not all of the items we want
3841 * are on the same leaf.
3843 * So we need to adjust how we search for holes, we need to drop the
3844 * path and re-search for the first extent key we found, and then walk
3845 * forward until we hit the last one we copied.
3847 if (need_find_last_extent) {
3848 /* btrfs_prev_leaf could return 1 without releasing the path */
3849 btrfs_release_path(src_path);
3850 ret = btrfs_search_slot(NULL, inode->root, &first_key,
3855 src = src_path->nodes[0];
3856 i = src_path->slots[0];
3862 * Ok so here we need to go through and fill in any holes we may have
3863 * to make sure that holes are punched for those areas in case they had
3864 * extents previously.
3870 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3871 ret = btrfs_next_leaf(inode->root, src_path);
3875 src = src_path->nodes[0];
3879 btrfs_item_key_to_cpu(src, &key, i);
3880 if (!btrfs_comp_cpu_keys(&key, &last_key))
3882 if (key.objectid != btrfs_ino(inode) ||
3883 key.type != BTRFS_EXTENT_DATA_KEY) {
3887 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3888 if (btrfs_file_extent_type(src, extent) ==
3889 BTRFS_FILE_EXTENT_INLINE) {
3890 len = btrfs_file_extent_inline_len(src, i, extent);
3891 extent_end = ALIGN(key.offset + len,
3892 fs_info->sectorsize);
3894 len = btrfs_file_extent_num_bytes(src, extent);
3895 extent_end = key.offset + len;
3899 if (*last_extent == key.offset) {
3900 *last_extent = extent_end;
3903 offset = *last_extent;
3904 len = key.offset - *last_extent;
3905 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3906 offset, 0, 0, len, 0, len, 0, 0, 0);
3909 *last_extent = extent_end;
3912 * Need to let the callers know we dropped the path so they should
3915 if (!ret && need_find_last_extent)
3920 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3922 struct extent_map *em1, *em2;
3924 em1 = list_entry(a, struct extent_map, list);
3925 em2 = list_entry(b, struct extent_map, list);
3927 if (em1->start < em2->start)
3929 else if (em1->start > em2->start)
3934 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3935 struct inode *inode,
3936 struct btrfs_root *root,
3937 const struct extent_map *em,
3938 const struct list_head *logged_list,
3939 bool *ordered_io_error)
3941 struct btrfs_fs_info *fs_info = root->fs_info;
3942 struct btrfs_ordered_extent *ordered;
3943 struct btrfs_root *log = root->log_root;
3944 u64 mod_start = em->mod_start;
3945 u64 mod_len = em->mod_len;
3946 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3949 LIST_HEAD(ordered_sums);
3952 *ordered_io_error = false;
3954 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3955 em->block_start == EXTENT_MAP_HOLE)
3959 * Wait far any ordered extent that covers our extent map. If it
3960 * finishes without an error, first check and see if our csums are on
3961 * our outstanding ordered extents.
3963 list_for_each_entry(ordered, logged_list, log_list) {
3964 struct btrfs_ordered_sum *sum;
3969 if (ordered->file_offset + ordered->len <= mod_start ||
3970 mod_start + mod_len <= ordered->file_offset)
3973 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3974 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3975 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3976 const u64 start = ordered->file_offset;
3977 const u64 end = ordered->file_offset + ordered->len - 1;
3979 WARN_ON(ordered->inode != inode);
3980 filemap_fdatawrite_range(inode->i_mapping, start, end);
3983 wait_event(ordered->wait,
3984 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3985 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3987 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
3989 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3990 * i_mapping flags, so that the next fsync won't get
3991 * an outdated io error too.
3993 filemap_check_errors(inode->i_mapping);
3994 *ordered_io_error = true;
3998 * We are going to copy all the csums on this ordered extent, so
3999 * go ahead and adjust mod_start and mod_len in case this
4000 * ordered extent has already been logged.
4002 if (ordered->file_offset > mod_start) {
4003 if (ordered->file_offset + ordered->len >=
4004 mod_start + mod_len)
4005 mod_len = ordered->file_offset - mod_start;
4007 * If we have this case
4009 * |--------- logged extent ---------|
4010 * |----- ordered extent ----|
4012 * Just don't mess with mod_start and mod_len, we'll
4013 * just end up logging more csums than we need and it
4017 if (ordered->file_offset + ordered->len <
4018 mod_start + mod_len) {
4019 mod_len = (mod_start + mod_len) -
4020 (ordered->file_offset + ordered->len);
4021 mod_start = ordered->file_offset +
4032 * To keep us from looping for the above case of an ordered
4033 * extent that falls inside of the logged extent.
4035 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
4039 list_for_each_entry(sum, &ordered->list, list) {
4040 ret = btrfs_csum_file_blocks(trans, log, sum);
4046 if (*ordered_io_error || !mod_len || ret || skip_csum)
4049 if (em->compress_type) {
4051 csum_len = max(em->block_len, em->orig_block_len);
4053 csum_offset = mod_start - em->start;
4057 /* block start is already adjusted for the file extent offset. */
4058 ret = btrfs_lookup_csums_range(fs_info->csum_root,
4059 em->block_start + csum_offset,
4060 em->block_start + csum_offset +
4061 csum_len - 1, &ordered_sums, 0);
4065 while (!list_empty(&ordered_sums)) {
4066 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4067 struct btrfs_ordered_sum,
4070 ret = btrfs_csum_file_blocks(trans, log, sums);
4071 list_del(&sums->list);
4078 static int log_one_extent(struct btrfs_trans_handle *trans,
4079 struct btrfs_inode *inode, struct btrfs_root *root,
4080 const struct extent_map *em,
4081 struct btrfs_path *path,
4082 const struct list_head *logged_list,
4083 struct btrfs_log_ctx *ctx)
4085 struct btrfs_root *log = root->log_root;
4086 struct btrfs_file_extent_item *fi;
4087 struct extent_buffer *leaf;
4088 struct btrfs_map_token token;
4089 struct btrfs_key key;
4090 u64 extent_offset = em->start - em->orig_start;
4093 int extent_inserted = 0;
4094 bool ordered_io_err = false;
4096 ret = wait_ordered_extents(trans, &inode->vfs_inode, root, em,
4097 logged_list, &ordered_io_err);
4101 if (ordered_io_err) {
4106 btrfs_init_map_token(&token);
4108 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4109 em->start + em->len, NULL, 0, 1,
4110 sizeof(*fi), &extent_inserted);
4114 if (!extent_inserted) {
4115 key.objectid = btrfs_ino(inode);
4116 key.type = BTRFS_EXTENT_DATA_KEY;
4117 key.offset = em->start;
4119 ret = btrfs_insert_empty_item(trans, log, path, &key,
4124 leaf = path->nodes[0];
4125 fi = btrfs_item_ptr(leaf, path->slots[0],
4126 struct btrfs_file_extent_item);
4128 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4130 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4131 btrfs_set_token_file_extent_type(leaf, fi,
4132 BTRFS_FILE_EXTENT_PREALLOC,
4135 btrfs_set_token_file_extent_type(leaf, fi,
4136 BTRFS_FILE_EXTENT_REG,
4139 block_len = max(em->block_len, em->orig_block_len);
4140 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4141 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4144 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4146 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4147 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4149 extent_offset, &token);
4150 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4153 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4154 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4158 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4159 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4160 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4161 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4163 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4164 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4165 btrfs_mark_buffer_dirty(leaf);
4167 btrfs_release_path(path);
4172 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4173 struct btrfs_root *root,
4174 struct btrfs_inode *inode,
4175 struct btrfs_path *path,
4176 struct list_head *logged_list,
4177 struct btrfs_log_ctx *ctx,
4181 struct extent_map *em, *n;
4182 struct list_head extents;
4183 struct extent_map_tree *tree = &inode->extent_tree;
4184 u64 logged_start, logged_end;
4189 INIT_LIST_HEAD(&extents);
4191 down_write(&inode->dio_sem);
4192 write_lock(&tree->lock);
4193 test_gen = root->fs_info->last_trans_committed;
4194 logged_start = start;
4197 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4198 list_del_init(&em->list);
4200 * Just an arbitrary number, this can be really CPU intensive
4201 * once we start getting a lot of extents, and really once we
4202 * have a bunch of extents we just want to commit since it will
4205 if (++num > 32768) {
4206 list_del_init(&tree->modified_extents);
4211 if (em->generation <= test_gen)
4214 if (em->start < logged_start)
4215 logged_start = em->start;
4216 if ((em->start + em->len - 1) > logged_end)
4217 logged_end = em->start + em->len - 1;
4219 /* Need a ref to keep it from getting evicted from cache */
4220 refcount_inc(&em->refs);
4221 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4222 list_add_tail(&em->list, &extents);
4226 list_sort(NULL, &extents, extent_cmp);
4227 btrfs_get_logged_extents(inode, logged_list, logged_start, logged_end);
4229 * Some ordered extents started by fsync might have completed
4230 * before we could collect them into the list logged_list, which
4231 * means they're gone, not in our logged_list nor in the inode's
4232 * ordered tree. We want the application/user space to know an
4233 * error happened while attempting to persist file data so that
4234 * it can take proper action. If such error happened, we leave
4235 * without writing to the log tree and the fsync must report the
4236 * file data write error and not commit the current transaction.
4238 ret = filemap_check_errors(inode->vfs_inode.i_mapping);
4242 while (!list_empty(&extents)) {
4243 em = list_entry(extents.next, struct extent_map, list);
4245 list_del_init(&em->list);
4248 * If we had an error we just need to delete everybody from our
4252 clear_em_logging(tree, em);
4253 free_extent_map(em);
4257 write_unlock(&tree->lock);
4259 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4261 write_lock(&tree->lock);
4262 clear_em_logging(tree, em);
4263 free_extent_map(em);
4265 WARN_ON(!list_empty(&extents));
4266 write_unlock(&tree->lock);
4267 up_write(&inode->dio_sem);
4269 btrfs_release_path(path);
4273 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4274 struct btrfs_path *path, u64 *size_ret)
4276 struct btrfs_key key;
4279 key.objectid = btrfs_ino(inode);
4280 key.type = BTRFS_INODE_ITEM_KEY;
4283 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4286 } else if (ret > 0) {
4289 struct btrfs_inode_item *item;
4291 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4292 struct btrfs_inode_item);
4293 *size_ret = btrfs_inode_size(path->nodes[0], item);
4296 btrfs_release_path(path);
4301 * At the moment we always log all xattrs. This is to figure out at log replay
4302 * time which xattrs must have their deletion replayed. If a xattr is missing
4303 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4304 * because if a xattr is deleted, the inode is fsynced and a power failure
4305 * happens, causing the log to be replayed the next time the fs is mounted,
4306 * we want the xattr to not exist anymore (same behaviour as other filesystems
4307 * with a journal, ext3/4, xfs, f2fs, etc).
4309 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4310 struct btrfs_root *root,
4311 struct btrfs_inode *inode,
4312 struct btrfs_path *path,
4313 struct btrfs_path *dst_path)
4316 struct btrfs_key key;
4317 const u64 ino = btrfs_ino(inode);
4322 key.type = BTRFS_XATTR_ITEM_KEY;
4325 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4330 int slot = path->slots[0];
4331 struct extent_buffer *leaf = path->nodes[0];
4332 int nritems = btrfs_header_nritems(leaf);
4334 if (slot >= nritems) {
4336 u64 last_extent = 0;
4338 ret = copy_items(trans, inode, dst_path, path,
4339 &last_extent, start_slot,
4341 /* can't be 1, extent items aren't processed */
4347 ret = btrfs_next_leaf(root, path);
4355 btrfs_item_key_to_cpu(leaf, &key, slot);
4356 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4366 u64 last_extent = 0;
4368 ret = copy_items(trans, inode, dst_path, path,
4369 &last_extent, start_slot,
4371 /* can't be 1, extent items aren't processed */
4381 * If the no holes feature is enabled we need to make sure any hole between the
4382 * last extent and the i_size of our inode is explicitly marked in the log. This
4383 * is to make sure that doing something like:
4385 * 1) create file with 128Kb of data
4386 * 2) truncate file to 64Kb
4387 * 3) truncate file to 256Kb
4389 * 5) <crash/power failure>
4390 * 6) mount fs and trigger log replay
4392 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4393 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4394 * file correspond to a hole. The presence of explicit holes in a log tree is
4395 * what guarantees that log replay will remove/adjust file extent items in the
4398 * Here we do not need to care about holes between extents, that is already done
4399 * by copy_items(). We also only need to do this in the full sync path, where we
4400 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4401 * lookup the list of modified extent maps and if any represents a hole, we
4402 * insert a corresponding extent representing a hole in the log tree.
4404 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4405 struct btrfs_root *root,
4406 struct btrfs_inode *inode,
4407 struct btrfs_path *path)
4409 struct btrfs_fs_info *fs_info = root->fs_info;
4411 struct btrfs_key key;
4414 struct extent_buffer *leaf;
4415 struct btrfs_root *log = root->log_root;
4416 const u64 ino = btrfs_ino(inode);
4417 const u64 i_size = i_size_read(&inode->vfs_inode);
4419 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
4423 key.type = BTRFS_EXTENT_DATA_KEY;
4424 key.offset = (u64)-1;
4426 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4431 ASSERT(path->slots[0] > 0);
4433 leaf = path->nodes[0];
4434 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4436 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4437 /* inode does not have any extents */
4441 struct btrfs_file_extent_item *extent;
4445 * If there's an extent beyond i_size, an explicit hole was
4446 * already inserted by copy_items().
4448 if (key.offset >= i_size)
4451 extent = btrfs_item_ptr(leaf, path->slots[0],
4452 struct btrfs_file_extent_item);
4454 if (btrfs_file_extent_type(leaf, extent) ==
4455 BTRFS_FILE_EXTENT_INLINE) {
4456 len = btrfs_file_extent_inline_len(leaf,
4459 ASSERT(len == i_size ||
4460 (len == fs_info->sectorsize &&
4461 btrfs_file_extent_compression(leaf, extent) !=
4462 BTRFS_COMPRESS_NONE));
4466 len = btrfs_file_extent_num_bytes(leaf, extent);
4467 /* Last extent goes beyond i_size, no need to log a hole. */
4468 if (key.offset + len > i_size)
4470 hole_start = key.offset + len;
4471 hole_size = i_size - hole_start;
4473 btrfs_release_path(path);
4475 /* Last extent ends at i_size. */
4479 hole_size = ALIGN(hole_size, fs_info->sectorsize);
4480 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4481 hole_size, 0, hole_size, 0, 0, 0);
4486 * When we are logging a new inode X, check if it doesn't have a reference that
4487 * matches the reference from some other inode Y created in a past transaction
4488 * and that was renamed in the current transaction. If we don't do this, then at
4489 * log replay time we can lose inode Y (and all its files if it's a directory):
4492 * echo "hello world" > /mnt/x/foobar
4495 * mkdir /mnt/x # or touch /mnt/x
4496 * xfs_io -c fsync /mnt/x
4498 * mount fs, trigger log replay
4500 * After the log replay procedure, we would lose the first directory and all its
4501 * files (file foobar).
4502 * For the case where inode Y is not a directory we simply end up losing it:
4504 * echo "123" > /mnt/foo
4506 * mv /mnt/foo /mnt/bar
4507 * echo "abc" > /mnt/foo
4508 * xfs_io -c fsync /mnt/foo
4511 * We also need this for cases where a snapshot entry is replaced by some other
4512 * entry (file or directory) otherwise we end up with an unreplayable log due to
4513 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4514 * if it were a regular entry:
4517 * btrfs subvolume snapshot /mnt /mnt/x/snap
4518 * btrfs subvolume delete /mnt/x/snap
4521 * fsync /mnt/x or fsync some new file inside it
4524 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4525 * the same transaction.
4527 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4529 const struct btrfs_key *key,
4530 struct btrfs_inode *inode,
4534 struct btrfs_path *search_path;
4537 u32 item_size = btrfs_item_size_nr(eb, slot);
4539 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4541 search_path = btrfs_alloc_path();
4544 search_path->search_commit_root = 1;
4545 search_path->skip_locking = 1;
4547 while (cur_offset < item_size) {
4551 unsigned long name_ptr;
4552 struct btrfs_dir_item *di;
4554 if (key->type == BTRFS_INODE_REF_KEY) {
4555 struct btrfs_inode_ref *iref;
4557 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4558 parent = key->offset;
4559 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4560 name_ptr = (unsigned long)(iref + 1);
4561 this_len = sizeof(*iref) + this_name_len;
4563 struct btrfs_inode_extref *extref;
4565 extref = (struct btrfs_inode_extref *)(ptr +
4567 parent = btrfs_inode_extref_parent(eb, extref);
4568 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4569 name_ptr = (unsigned long)&extref->name;
4570 this_len = sizeof(*extref) + this_name_len;
4573 ret = btrfs_is_name_len_valid(eb, slot, name_ptr,
4579 if (this_name_len > name_len) {
4582 new_name = krealloc(name, this_name_len, GFP_NOFS);
4587 name_len = this_name_len;
4591 read_extent_buffer(eb, name, name_ptr, this_name_len);
4592 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4593 parent, name, this_name_len, 0);
4594 if (di && !IS_ERR(di)) {
4595 struct btrfs_key di_key;
4597 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4599 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4601 *other_ino = di_key.objectid;
4606 } else if (IS_ERR(di)) {
4610 btrfs_release_path(search_path);
4612 cur_offset += this_len;
4616 btrfs_free_path(search_path);
4621 /* log a single inode in the tree log.
4622 * At least one parent directory for this inode must exist in the tree
4623 * or be logged already.
4625 * Any items from this inode changed by the current transaction are copied
4626 * to the log tree. An extra reference is taken on any extents in this
4627 * file, allowing us to avoid a whole pile of corner cases around logging
4628 * blocks that have been removed from the tree.
4630 * See LOG_INODE_ALL and related defines for a description of what inode_only
4633 * This handles both files and directories.
4635 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4636 struct btrfs_root *root, struct btrfs_inode *inode,
4640 struct btrfs_log_ctx *ctx)
4642 struct btrfs_fs_info *fs_info = root->fs_info;
4643 struct btrfs_path *path;
4644 struct btrfs_path *dst_path;
4645 struct btrfs_key min_key;
4646 struct btrfs_key max_key;
4647 struct btrfs_root *log = root->log_root;
4648 struct extent_buffer *src = NULL;
4649 LIST_HEAD(logged_list);
4650 u64 last_extent = 0;
4654 int ins_start_slot = 0;
4656 bool fast_search = false;
4657 u64 ino = btrfs_ino(inode);
4658 struct extent_map_tree *em_tree = &inode->extent_tree;
4659 u64 logged_isize = 0;
4660 bool need_log_inode_item = true;
4662 path = btrfs_alloc_path();
4665 dst_path = btrfs_alloc_path();
4667 btrfs_free_path(path);
4671 min_key.objectid = ino;
4672 min_key.type = BTRFS_INODE_ITEM_KEY;
4675 max_key.objectid = ino;
4678 /* today the code can only do partial logging of directories */
4679 if (S_ISDIR(inode->vfs_inode.i_mode) ||
4680 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4681 &inode->runtime_flags) &&
4682 inode_only >= LOG_INODE_EXISTS))
4683 max_key.type = BTRFS_XATTR_ITEM_KEY;
4685 max_key.type = (u8)-1;
4686 max_key.offset = (u64)-1;
4689 * Only run delayed items if we are a dir or a new file.
4690 * Otherwise commit the delayed inode only, which is needed in
4691 * order for the log replay code to mark inodes for link count
4692 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4694 if (S_ISDIR(inode->vfs_inode.i_mode) ||
4695 inode->generation > fs_info->last_trans_committed)
4696 ret = btrfs_commit_inode_delayed_items(trans, inode);
4698 ret = btrfs_commit_inode_delayed_inode(inode);
4701 btrfs_free_path(path);
4702 btrfs_free_path(dst_path);
4706 if (inode_only == LOG_OTHER_INODE) {
4707 inode_only = LOG_INODE_EXISTS;
4708 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
4710 mutex_lock(&inode->log_mutex);
4714 * a brute force approach to making sure we get the most uptodate
4715 * copies of everything.
4717 if (S_ISDIR(inode->vfs_inode.i_mode)) {
4718 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4720 if (inode_only == LOG_INODE_EXISTS)
4721 max_key_type = BTRFS_XATTR_ITEM_KEY;
4722 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
4724 if (inode_only == LOG_INODE_EXISTS) {
4726 * Make sure the new inode item we write to the log has
4727 * the same isize as the current one (if it exists).
4728 * This is necessary to prevent data loss after log
4729 * replay, and also to prevent doing a wrong expanding
4730 * truncate - for e.g. create file, write 4K into offset
4731 * 0, fsync, write 4K into offset 4096, add hard link,
4732 * fsync some other file (to sync log), power fail - if
4733 * we use the inode's current i_size, after log replay
4734 * we get a 8Kb file, with the last 4Kb extent as a hole
4735 * (zeroes), as if an expanding truncate happened,
4736 * instead of getting a file of 4Kb only.
4738 err = logged_inode_size(log, inode, path, &logged_isize);
4742 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4743 &inode->runtime_flags)) {
4744 if (inode_only == LOG_INODE_EXISTS) {
4745 max_key.type = BTRFS_XATTR_ITEM_KEY;
4746 ret = drop_objectid_items(trans, log, path, ino,
4749 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4750 &inode->runtime_flags);
4751 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4752 &inode->runtime_flags);
4754 ret = btrfs_truncate_inode_items(trans,
4755 log, &inode->vfs_inode, 0, 0);
4760 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4761 &inode->runtime_flags) ||
4762 inode_only == LOG_INODE_EXISTS) {
4763 if (inode_only == LOG_INODE_ALL)
4765 max_key.type = BTRFS_XATTR_ITEM_KEY;
4766 ret = drop_objectid_items(trans, log, path, ino,
4769 if (inode_only == LOG_INODE_ALL)
4782 ret = btrfs_search_forward(root, &min_key,
4783 path, trans->transid);
4791 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4792 if (min_key.objectid != ino)
4794 if (min_key.type > max_key.type)
4797 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4798 need_log_inode_item = false;
4800 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4801 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4802 inode->generation == trans->transid) {
4805 ret = btrfs_check_ref_name_override(path->nodes[0],
4806 path->slots[0], &min_key, inode,
4811 } else if (ret > 0 && ctx &&
4812 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
4813 struct btrfs_key inode_key;
4814 struct inode *other_inode;
4820 ins_start_slot = path->slots[0];
4822 ret = copy_items(trans, inode, dst_path, path,
4823 &last_extent, ins_start_slot,
4831 btrfs_release_path(path);
4832 inode_key.objectid = other_ino;
4833 inode_key.type = BTRFS_INODE_ITEM_KEY;
4834 inode_key.offset = 0;
4835 other_inode = btrfs_iget(fs_info->sb,
4839 * If the other inode that had a conflicting dir
4840 * entry was deleted in the current transaction,
4841 * we don't need to do more work nor fallback to
4842 * a transaction commit.
4844 if (IS_ERR(other_inode) &&
4845 PTR_ERR(other_inode) == -ENOENT) {
4847 } else if (IS_ERR(other_inode)) {
4848 err = PTR_ERR(other_inode);
4852 * We are safe logging the other inode without
4853 * acquiring its i_mutex as long as we log with
4854 * the LOG_INODE_EXISTS mode. We're safe against
4855 * concurrent renames of the other inode as well
4856 * because during a rename we pin the log and
4857 * update the log with the new name before we
4860 err = btrfs_log_inode(trans, root,
4861 BTRFS_I(other_inode),
4862 LOG_OTHER_INODE, 0, LLONG_MAX,
4872 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4873 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4876 ret = copy_items(trans, inode, dst_path, path,
4877 &last_extent, ins_start_slot,
4878 ins_nr, inode_only, logged_isize);
4885 btrfs_release_path(path);
4891 src = path->nodes[0];
4892 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4895 } else if (!ins_nr) {
4896 ins_start_slot = path->slots[0];
4901 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4902 ins_start_slot, ins_nr, inode_only,
4910 btrfs_release_path(path);
4914 ins_start_slot = path->slots[0];
4917 nritems = btrfs_header_nritems(path->nodes[0]);
4919 if (path->slots[0] < nritems) {
4920 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4925 ret = copy_items(trans, inode, dst_path, path,
4926 &last_extent, ins_start_slot,
4927 ins_nr, inode_only, logged_isize);
4935 btrfs_release_path(path);
4937 if (min_key.offset < (u64)-1) {
4939 } else if (min_key.type < max_key.type) {
4947 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4948 ins_start_slot, ins_nr, inode_only,
4958 btrfs_release_path(path);
4959 btrfs_release_path(dst_path);
4960 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
4963 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4964 btrfs_release_path(path);
4965 btrfs_release_path(dst_path);
4966 err = btrfs_log_trailing_hole(trans, root, inode, path);
4971 btrfs_release_path(path);
4972 btrfs_release_path(dst_path);
4973 if (need_log_inode_item) {
4974 err = log_inode_item(trans, log, dst_path, inode);
4979 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
4980 &logged_list, ctx, start, end);
4985 } else if (inode_only == LOG_INODE_ALL) {
4986 struct extent_map *em, *n;
4988 write_lock(&em_tree->lock);
4990 * We can't just remove every em if we're called for a ranged
4991 * fsync - that is, one that doesn't cover the whole possible
4992 * file range (0 to LLONG_MAX). This is because we can have
4993 * em's that fall outside the range we're logging and therefore
4994 * their ordered operations haven't completed yet
4995 * (btrfs_finish_ordered_io() not invoked yet). This means we
4996 * didn't get their respective file extent item in the fs/subvol
4997 * tree yet, and need to let the next fast fsync (one which
4998 * consults the list of modified extent maps) find the em so
4999 * that it logs a matching file extent item and waits for the
5000 * respective ordered operation to complete (if it's still
5003 * Removing every em outside the range we're logging would make
5004 * the next fast fsync not log their matching file extent items,
5005 * therefore making us lose data after a log replay.
5007 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5009 const u64 mod_end = em->mod_start + em->mod_len - 1;
5011 if (em->mod_start >= start && mod_end <= end)
5012 list_del_init(&em->list);
5014 write_unlock(&em_tree->lock);
5017 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5018 ret = log_directory_changes(trans, root, inode, path, dst_path,
5026 spin_lock(&inode->lock);
5027 inode->logged_trans = trans->transid;
5028 inode->last_log_commit = inode->last_sub_trans;
5029 spin_unlock(&inode->lock);
5032 btrfs_put_logged_extents(&logged_list);
5034 btrfs_submit_logged_extents(&logged_list, log);
5035 mutex_unlock(&inode->log_mutex);
5037 btrfs_free_path(path);
5038 btrfs_free_path(dst_path);
5043 * Check if we must fallback to a transaction commit when logging an inode.
5044 * This must be called after logging the inode and is used only in the context
5045 * when fsyncing an inode requires the need to log some other inode - in which
5046 * case we can't lock the i_mutex of each other inode we need to log as that
5047 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5048 * log inodes up or down in the hierarchy) or rename operations for example. So
5049 * we take the log_mutex of the inode after we have logged it and then check for
5050 * its last_unlink_trans value - this is safe because any task setting
5051 * last_unlink_trans must take the log_mutex and it must do this before it does
5052 * the actual unlink operation, so if we do this check before a concurrent task
5053 * sets last_unlink_trans it means we've logged a consistent version/state of
5054 * all the inode items, otherwise we are not sure and must do a transaction
5055 * commit (the concurrent task might have only updated last_unlink_trans before
5056 * we logged the inode or it might have also done the unlink).
5058 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5059 struct btrfs_inode *inode)
5061 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5064 mutex_lock(&inode->log_mutex);
5065 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5067 * Make sure any commits to the log are forced to be full
5070 btrfs_set_log_full_commit(fs_info, trans);
5073 mutex_unlock(&inode->log_mutex);
5079 * follow the dentry parent pointers up the chain and see if any
5080 * of the directories in it require a full commit before they can
5081 * be logged. Returns zero if nothing special needs to be done or 1 if
5082 * a full commit is required.
5084 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5085 struct btrfs_inode *inode,
5086 struct dentry *parent,
5087 struct super_block *sb,
5091 struct dentry *old_parent = NULL;
5092 struct btrfs_inode *orig_inode = inode;
5095 * for regular files, if its inode is already on disk, we don't
5096 * have to worry about the parents at all. This is because
5097 * we can use the last_unlink_trans field to record renames
5098 * and other fun in this file.
5100 if (S_ISREG(inode->vfs_inode.i_mode) &&
5101 inode->generation <= last_committed &&
5102 inode->last_unlink_trans <= last_committed)
5105 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5106 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5108 inode = BTRFS_I(d_inode(parent));
5113 * If we are logging a directory then we start with our inode,
5114 * not our parent's inode, so we need to skip setting the
5115 * logged_trans so that further down in the log code we don't
5116 * think this inode has already been logged.
5118 if (inode != orig_inode)
5119 inode->logged_trans = trans->transid;
5122 if (btrfs_must_commit_transaction(trans, inode)) {
5127 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5130 if (IS_ROOT(parent)) {
5131 inode = BTRFS_I(d_inode(parent));
5132 if (btrfs_must_commit_transaction(trans, inode))
5137 parent = dget_parent(parent);
5139 old_parent = parent;
5140 inode = BTRFS_I(d_inode(parent));
5148 struct btrfs_dir_list {
5150 struct list_head list;
5154 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5155 * details about the why it is needed.
5156 * This is a recursive operation - if an existing dentry corresponds to a
5157 * directory, that directory's new entries are logged too (same behaviour as
5158 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5159 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5160 * complains about the following circular lock dependency / possible deadlock:
5164 * lock(&type->i_mutex_dir_key#3/2);
5165 * lock(sb_internal#2);
5166 * lock(&type->i_mutex_dir_key#3/2);
5167 * lock(&sb->s_type->i_mutex_key#14);
5169 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5170 * sb_start_intwrite() in btrfs_start_transaction().
5171 * Not locking i_mutex of the inodes is still safe because:
5173 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5174 * that while logging the inode new references (names) are added or removed
5175 * from the inode, leaving the logged inode item with a link count that does
5176 * not match the number of logged inode reference items. This is fine because
5177 * at log replay time we compute the real number of links and correct the
5178 * link count in the inode item (see replay_one_buffer() and
5179 * link_to_fixup_dir());
5181 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5182 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5183 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5184 * has a size that doesn't match the sum of the lengths of all the logged
5185 * names. This does not result in a problem because if a dir_item key is
5186 * logged but its matching dir_index key is not logged, at log replay time we
5187 * don't use it to replay the respective name (see replay_one_name()). On the
5188 * other hand if only the dir_index key ends up being logged, the respective
5189 * name is added to the fs/subvol tree with both the dir_item and dir_index
5190 * keys created (see replay_one_name()).
5191 * The directory's inode item with a wrong i_size is not a problem as well,
5192 * since we don't use it at log replay time to set the i_size in the inode
5193 * item of the fs/subvol tree (see overwrite_item()).
5195 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5196 struct btrfs_root *root,
5197 struct btrfs_inode *start_inode,
5198 struct btrfs_log_ctx *ctx)
5200 struct btrfs_fs_info *fs_info = root->fs_info;
5201 struct btrfs_root *log = root->log_root;
5202 struct btrfs_path *path;
5203 LIST_HEAD(dir_list);
5204 struct btrfs_dir_list *dir_elem;
5207 path = btrfs_alloc_path();
5211 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5213 btrfs_free_path(path);
5216 dir_elem->ino = btrfs_ino(start_inode);
5217 list_add_tail(&dir_elem->list, &dir_list);
5219 while (!list_empty(&dir_list)) {
5220 struct extent_buffer *leaf;
5221 struct btrfs_key min_key;
5225 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5228 goto next_dir_inode;
5230 min_key.objectid = dir_elem->ino;
5231 min_key.type = BTRFS_DIR_ITEM_KEY;
5234 btrfs_release_path(path);
5235 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5237 goto next_dir_inode;
5238 } else if (ret > 0) {
5240 goto next_dir_inode;
5244 leaf = path->nodes[0];
5245 nritems = btrfs_header_nritems(leaf);
5246 for (i = path->slots[0]; i < nritems; i++) {
5247 struct btrfs_dir_item *di;
5248 struct btrfs_key di_key;
5249 struct inode *di_inode;
5250 struct btrfs_dir_list *new_dir_elem;
5251 int log_mode = LOG_INODE_EXISTS;
5254 btrfs_item_key_to_cpu(leaf, &min_key, i);
5255 if (min_key.objectid != dir_elem->ino ||
5256 min_key.type != BTRFS_DIR_ITEM_KEY)
5257 goto next_dir_inode;
5259 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5260 type = btrfs_dir_type(leaf, di);
5261 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5262 type != BTRFS_FT_DIR)
5264 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5265 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5268 btrfs_release_path(path);
5269 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
5270 if (IS_ERR(di_inode)) {
5271 ret = PTR_ERR(di_inode);
5272 goto next_dir_inode;
5275 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5280 ctx->log_new_dentries = false;
5281 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5282 log_mode = LOG_INODE_ALL;
5283 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5284 log_mode, 0, LLONG_MAX, ctx);
5286 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5290 goto next_dir_inode;
5291 if (ctx->log_new_dentries) {
5292 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5294 if (!new_dir_elem) {
5296 goto next_dir_inode;
5298 new_dir_elem->ino = di_key.objectid;
5299 list_add_tail(&new_dir_elem->list, &dir_list);
5304 ret = btrfs_next_leaf(log, path);
5306 goto next_dir_inode;
5307 } else if (ret > 0) {
5309 goto next_dir_inode;
5313 if (min_key.offset < (u64)-1) {
5318 list_del(&dir_elem->list);
5322 btrfs_free_path(path);
5326 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5327 struct btrfs_inode *inode,
5328 struct btrfs_log_ctx *ctx)
5330 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5332 struct btrfs_path *path;
5333 struct btrfs_key key;
5334 struct btrfs_root *root = inode->root;
5335 const u64 ino = btrfs_ino(inode);
5337 path = btrfs_alloc_path();
5340 path->skip_locking = 1;
5341 path->search_commit_root = 1;
5344 key.type = BTRFS_INODE_REF_KEY;
5346 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5351 struct extent_buffer *leaf = path->nodes[0];
5352 int slot = path->slots[0];
5357 if (slot >= btrfs_header_nritems(leaf)) {
5358 ret = btrfs_next_leaf(root, path);
5366 btrfs_item_key_to_cpu(leaf, &key, slot);
5367 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5368 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5371 item_size = btrfs_item_size_nr(leaf, slot);
5372 ptr = btrfs_item_ptr_offset(leaf, slot);
5373 while (cur_offset < item_size) {
5374 struct btrfs_key inode_key;
5375 struct inode *dir_inode;
5377 inode_key.type = BTRFS_INODE_ITEM_KEY;
5378 inode_key.offset = 0;
5380 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5381 struct btrfs_inode_extref *extref;
5383 extref = (struct btrfs_inode_extref *)
5385 inode_key.objectid = btrfs_inode_extref_parent(
5387 cur_offset += sizeof(*extref);
5388 cur_offset += btrfs_inode_extref_name_len(leaf,
5391 inode_key.objectid = key.offset;
5392 cur_offset = item_size;
5395 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
5397 /* If parent inode was deleted, skip it. */
5398 if (IS_ERR(dir_inode))
5402 ctx->log_new_dentries = false;
5403 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5404 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5406 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5408 if (!ret && ctx && ctx->log_new_dentries)
5409 ret = log_new_dir_dentries(trans, root,
5410 BTRFS_I(dir_inode), ctx);
5419 btrfs_free_path(path);
5424 * helper function around btrfs_log_inode to make sure newly created
5425 * parent directories also end up in the log. A minimal inode and backref
5426 * only logging is done of any parent directories that are older than
5427 * the last committed transaction
5429 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5430 struct btrfs_root *root,
5431 struct btrfs_inode *inode,
5432 struct dentry *parent,
5436 struct btrfs_log_ctx *ctx)
5438 struct btrfs_fs_info *fs_info = root->fs_info;
5439 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
5440 struct super_block *sb;
5441 struct dentry *old_parent = NULL;
5443 u64 last_committed = fs_info->last_trans_committed;
5444 bool log_dentries = false;
5445 struct btrfs_inode *orig_inode = inode;
5447 sb = inode->vfs_inode.i_sb;
5449 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5455 * The prev transaction commit doesn't complete, we need do
5456 * full commit by ourselves.
5458 if (fs_info->last_trans_log_full_commit >
5459 fs_info->last_trans_committed) {
5464 if (root != inode->root || btrfs_root_refs(&root->root_item) == 0) {
5469 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
5474 if (btrfs_inode_in_log(inode, trans->transid)) {
5475 ret = BTRFS_NO_LOG_SYNC;
5479 ret = start_log_trans(trans, root, ctx);
5483 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5488 * for regular files, if its inode is already on disk, we don't
5489 * have to worry about the parents at all. This is because
5490 * we can use the last_unlink_trans field to record renames
5491 * and other fun in this file.
5493 if (S_ISREG(inode->vfs_inode.i_mode) &&
5494 inode->generation <= last_committed &&
5495 inode->last_unlink_trans <= last_committed) {
5500 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
5501 log_dentries = true;
5504 * On unlink we must make sure all our current and old parent directory
5505 * inodes are fully logged. This is to prevent leaving dangling
5506 * directory index entries in directories that were our parents but are
5507 * not anymore. Not doing this results in old parent directory being
5508 * impossible to delete after log replay (rmdir will always fail with
5509 * error -ENOTEMPTY).
5515 * ln testdir/foo testdir/bar
5517 * unlink testdir/bar
5518 * xfs_io -c fsync testdir/foo
5520 * mount fs, triggers log replay
5522 * If we don't log the parent directory (testdir), after log replay the
5523 * directory still has an entry pointing to the file inode using the bar
5524 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5525 * the file inode has a link count of 1.
5531 * ln foo testdir/foo2
5532 * ln foo testdir/foo3
5534 * unlink testdir/foo3
5535 * xfs_io -c fsync foo
5537 * mount fs, triggers log replay
5539 * Similar as the first example, after log replay the parent directory
5540 * testdir still has an entry pointing to the inode file with name foo3
5541 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5542 * and has a link count of 2.
5544 if (inode->last_unlink_trans > last_committed) {
5545 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5551 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5554 inode = BTRFS_I(d_inode(parent));
5555 if (root != inode->root)
5558 if (inode->generation > last_committed) {
5559 ret = btrfs_log_inode(trans, root, inode,
5560 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5564 if (IS_ROOT(parent))
5567 parent = dget_parent(parent);
5569 old_parent = parent;
5572 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5578 btrfs_set_log_full_commit(fs_info, trans);
5583 btrfs_remove_log_ctx(root, ctx);
5584 btrfs_end_log_trans(root);
5590 * it is not safe to log dentry if the chunk root has added new
5591 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5592 * If this returns 1, you must commit the transaction to safely get your
5595 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
5596 struct btrfs_root *root, struct dentry *dentry,
5599 struct btrfs_log_ctx *ctx)
5601 struct dentry *parent = dget_parent(dentry);
5604 ret = btrfs_log_inode_parent(trans, root, BTRFS_I(d_inode(dentry)),
5605 parent, start, end, 0, ctx);
5612 * should be called during mount to recover any replay any log trees
5615 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5618 struct btrfs_path *path;
5619 struct btrfs_trans_handle *trans;
5620 struct btrfs_key key;
5621 struct btrfs_key found_key;
5622 struct btrfs_key tmp_key;
5623 struct btrfs_root *log;
5624 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5625 struct walk_control wc = {
5626 .process_func = process_one_buffer,
5630 path = btrfs_alloc_path();
5634 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5636 trans = btrfs_start_transaction(fs_info->tree_root, 0);
5637 if (IS_ERR(trans)) {
5638 ret = PTR_ERR(trans);
5645 ret = walk_log_tree(trans, log_root_tree, &wc);
5647 btrfs_handle_fs_error(fs_info, ret,
5648 "Failed to pin buffers while recovering log root tree.");
5653 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5654 key.offset = (u64)-1;
5655 key.type = BTRFS_ROOT_ITEM_KEY;
5658 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
5661 btrfs_handle_fs_error(fs_info, ret,
5662 "Couldn't find tree log root.");
5666 if (path->slots[0] == 0)
5670 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5672 btrfs_release_path(path);
5673 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5676 log = btrfs_read_fs_root(log_root_tree, &found_key);
5679 btrfs_handle_fs_error(fs_info, ret,
5680 "Couldn't read tree log root.");
5684 tmp_key.objectid = found_key.offset;
5685 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5686 tmp_key.offset = (u64)-1;
5688 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
5689 if (IS_ERR(wc.replay_dest)) {
5690 ret = PTR_ERR(wc.replay_dest);
5691 free_extent_buffer(log->node);
5692 free_extent_buffer(log->commit_root);
5694 btrfs_handle_fs_error(fs_info, ret,
5695 "Couldn't read target root for tree log recovery.");
5699 wc.replay_dest->log_root = log;
5700 btrfs_record_root_in_trans(trans, wc.replay_dest);
5701 ret = walk_log_tree(trans, log, &wc);
5703 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5704 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5708 key.offset = found_key.offset - 1;
5709 wc.replay_dest->log_root = NULL;
5710 free_extent_buffer(log->node);
5711 free_extent_buffer(log->commit_root);
5717 if (found_key.offset == 0)
5720 btrfs_release_path(path);
5722 /* step one is to pin it all, step two is to replay just inodes */
5725 wc.process_func = replay_one_buffer;
5726 wc.stage = LOG_WALK_REPLAY_INODES;
5729 /* step three is to replay everything */
5730 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5735 btrfs_free_path(path);
5737 /* step 4: commit the transaction, which also unpins the blocks */
5738 ret = btrfs_commit_transaction(trans);
5742 free_extent_buffer(log_root_tree->node);
5743 log_root_tree->log_root = NULL;
5744 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5745 kfree(log_root_tree);
5750 btrfs_end_transaction(wc.trans);
5751 btrfs_free_path(path);
5756 * there are some corner cases where we want to force a full
5757 * commit instead of allowing a directory to be logged.
5759 * They revolve around files there were unlinked from the directory, and
5760 * this function updates the parent directory so that a full commit is
5761 * properly done if it is fsync'd later after the unlinks are done.
5763 * Must be called before the unlink operations (updates to the subvolume tree,
5764 * inodes, etc) are done.
5766 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5767 struct btrfs_inode *dir, struct btrfs_inode *inode,
5771 * when we're logging a file, if it hasn't been renamed
5772 * or unlinked, and its inode is fully committed on disk,
5773 * we don't have to worry about walking up the directory chain
5774 * to log its parents.
5776 * So, we use the last_unlink_trans field to put this transid
5777 * into the file. When the file is logged we check it and
5778 * don't log the parents if the file is fully on disk.
5780 mutex_lock(&inode->log_mutex);
5781 inode->last_unlink_trans = trans->transid;
5782 mutex_unlock(&inode->log_mutex);
5785 * if this directory was already logged any new
5786 * names for this file/dir will get recorded
5789 if (dir->logged_trans == trans->transid)
5793 * if the inode we're about to unlink was logged,
5794 * the log will be properly updated for any new names
5796 if (inode->logged_trans == trans->transid)
5800 * when renaming files across directories, if the directory
5801 * there we're unlinking from gets fsync'd later on, there's
5802 * no way to find the destination directory later and fsync it
5803 * properly. So, we have to be conservative and force commits
5804 * so the new name gets discovered.
5809 /* we can safely do the unlink without any special recording */
5813 mutex_lock(&dir->log_mutex);
5814 dir->last_unlink_trans = trans->transid;
5815 mutex_unlock(&dir->log_mutex);
5819 * Make sure that if someone attempts to fsync the parent directory of a deleted
5820 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5821 * that after replaying the log tree of the parent directory's root we will not
5822 * see the snapshot anymore and at log replay time we will not see any log tree
5823 * corresponding to the deleted snapshot's root, which could lead to replaying
5824 * it after replaying the log tree of the parent directory (which would replay
5825 * the snapshot delete operation).
5827 * Must be called before the actual snapshot destroy operation (updates to the
5828 * parent root and tree of tree roots trees, etc) are done.
5830 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
5831 struct btrfs_inode *dir)
5833 mutex_lock(&dir->log_mutex);
5834 dir->last_unlink_trans = trans->transid;
5835 mutex_unlock(&dir->log_mutex);
5839 * Call this after adding a new name for a file and it will properly
5840 * update the log to reflect the new name.
5842 * It will return zero if all goes well, and it will return 1 if a
5843 * full transaction commit is required.
5845 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5846 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
5847 struct dentry *parent)
5849 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5850 struct btrfs_root *root = inode->root;
5853 * this will force the logging code to walk the dentry chain
5856 if (S_ISREG(inode->vfs_inode.i_mode))
5857 inode->last_unlink_trans = trans->transid;
5860 * if this inode hasn't been logged and directory we're renaming it
5861 * from hasn't been logged, we don't need to log it
5863 if (inode->logged_trans <= fs_info->last_trans_committed &&
5864 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
5867 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
5868 LLONG_MAX, 1, NULL);