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>
22 #include "transaction.h"
25 #include "print-tree.h"
29 /* magic values for the inode_only field in btrfs_log_inode:
31 * LOG_INODE_ALL means to log everything
32 * LOG_INODE_EXISTS means to log just enough to recreate the inode
35 #define LOG_INODE_ALL 0
36 #define LOG_INODE_EXISTS 1
39 * directory trouble cases
41 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
42 * log, we must force a full commit before doing an fsync of the directory
43 * where the unlink was done.
44 * ---> record transid of last unlink/rename per directory
48 * rename foo/some_dir foo2/some_dir
50 * fsync foo/some_dir/some_file
52 * The fsync above will unlink the original some_dir without recording
53 * it in its new location (foo2). After a crash, some_dir will be gone
54 * unless the fsync of some_file forces a full commit
56 * 2) we must log any new names for any file or dir that is in the fsync
57 * log. ---> check inode while renaming/linking.
59 * 2a) we must log any new names for any file or dir during rename
60 * when the directory they are being removed from was logged.
61 * ---> check inode and old parent dir during rename
63 * 2a is actually the more important variant. With the extra logging
64 * a crash might unlink the old name without recreating the new one
66 * 3) after a crash, we must go through any directories with a link count
67 * of zero and redo the rm -rf
74 * The directory f1 was fully removed from the FS, but fsync was never
75 * called on f1, only its parent dir. After a crash the rm -rf must
76 * be replayed. This must be able to recurse down the entire
77 * directory tree. The inode link count fixup code takes care of the
82 * stages for the tree walking. The first
83 * stage (0) is to only pin down the blocks we find
84 * the second stage (1) is to make sure that all the inodes
85 * we find in the log are created in the subvolume.
87 * The last stage is to deal with directories and links and extents
88 * and all the other fun semantics
90 #define LOG_WALK_PIN_ONLY 0
91 #define LOG_WALK_REPLAY_INODES 1
92 #define LOG_WALK_REPLAY_ALL 2
94 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root, struct inode *inode,
97 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root,
99 struct btrfs_path *path, u64 objectid);
100 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root,
102 struct btrfs_root *log,
103 struct btrfs_path *path,
104 u64 dirid, int del_all);
107 * tree logging is a special write ahead log used to make sure that
108 * fsyncs and O_SYNCs can happen without doing full tree commits.
110 * Full tree commits are expensive because they require commonly
111 * modified blocks to be recowed, creating many dirty pages in the
112 * extent tree an 4x-6x higher write load than ext3.
114 * Instead of doing a tree commit on every fsync, we use the
115 * key ranges and transaction ids to find items for a given file or directory
116 * that have changed in this transaction. Those items are copied into
117 * a special tree (one per subvolume root), that tree is written to disk
118 * and then the fsync is considered complete.
120 * After a crash, items are copied out of the log-tree back into the
121 * subvolume tree. Any file data extents found are recorded in the extent
122 * allocation tree, and the log-tree freed.
124 * The log tree is read three times, once to pin down all the extents it is
125 * using in ram and once, once to create all the inodes logged in the tree
126 * and once to do all the other items.
130 * start a sub transaction and setup the log tree
131 * this increments the log tree writer count to make the people
132 * syncing the tree wait for us to finish
134 static int start_log_trans(struct btrfs_trans_handle *trans,
135 struct btrfs_root *root)
140 mutex_lock(&root->log_mutex);
141 if (root->log_root) {
142 if (!root->log_start_pid) {
143 root->log_start_pid = current->pid;
144 root->log_multiple_pids = false;
145 } else if (root->log_start_pid != current->pid) {
146 root->log_multiple_pids = true;
150 atomic_inc(&root->log_writers);
151 mutex_unlock(&root->log_mutex);
154 root->log_multiple_pids = false;
155 root->log_start_pid = current->pid;
156 mutex_lock(&root->fs_info->tree_log_mutex);
157 if (!root->fs_info->log_root_tree) {
158 ret = btrfs_init_log_root_tree(trans, root->fs_info);
162 if (err == 0 && !root->log_root) {
163 ret = btrfs_add_log_tree(trans, root);
167 mutex_unlock(&root->fs_info->tree_log_mutex);
169 atomic_inc(&root->log_writers);
170 mutex_unlock(&root->log_mutex);
175 * returns 0 if there was a log transaction running and we were able
176 * to join, or returns -ENOENT if there were not transactions
179 static int join_running_log_trans(struct btrfs_root *root)
187 mutex_lock(&root->log_mutex);
188 if (root->log_root) {
190 atomic_inc(&root->log_writers);
192 mutex_unlock(&root->log_mutex);
197 * This either makes the current running log transaction wait
198 * until you call btrfs_end_log_trans() or it makes any future
199 * log transactions wait until you call btrfs_end_log_trans()
201 int btrfs_pin_log_trans(struct btrfs_root *root)
205 mutex_lock(&root->log_mutex);
206 atomic_inc(&root->log_writers);
207 mutex_unlock(&root->log_mutex);
212 * indicate we're done making changes to the log tree
213 * and wake up anyone waiting to do a sync
215 int btrfs_end_log_trans(struct btrfs_root *root)
217 if (atomic_dec_and_test(&root->log_writers)) {
219 if (waitqueue_active(&root->log_writer_wait))
220 wake_up(&root->log_writer_wait);
227 * the walk control struct is used to pass state down the chain when
228 * processing the log tree. The stage field tells us which part
229 * of the log tree processing we are currently doing. The others
230 * are state fields used for that specific part
232 struct walk_control {
233 /* should we free the extent on disk when done? This is used
234 * at transaction commit time while freeing a log tree
238 /* should we write out the extent buffer? This is used
239 * while flushing the log tree to disk during a sync
243 /* should we wait for the extent buffer io to finish? Also used
244 * while flushing the log tree to disk for a sync
248 /* pin only walk, we record which extents on disk belong to the
253 /* what stage of the replay code we're currently in */
256 /* the root we are currently replaying */
257 struct btrfs_root *replay_dest;
259 /* the trans handle for the current replay */
260 struct btrfs_trans_handle *trans;
262 /* the function that gets used to process blocks we find in the
263 * tree. Note the extent_buffer might not be up to date when it is
264 * passed in, and it must be checked or read if you need the data
267 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
268 struct walk_control *wc, u64 gen);
272 * process_func used to pin down extents, write them or wait on them
274 static int process_one_buffer(struct btrfs_root *log,
275 struct extent_buffer *eb,
276 struct walk_control *wc, u64 gen)
279 btrfs_pin_extent_for_log_replay(wc->trans,
280 log->fs_info->extent_root,
283 if (btrfs_buffer_uptodate(eb, gen)) {
285 btrfs_write_tree_block(eb);
287 btrfs_wait_tree_block_writeback(eb);
293 * Item overwrite used by replay and tree logging. eb, slot and key all refer
294 * to the src data we are copying out.
296 * root is the tree we are copying into, and path is a scratch
297 * path for use in this function (it should be released on entry and
298 * will be released on exit).
300 * If the key is already in the destination tree the existing item is
301 * overwritten. If the existing item isn't big enough, it is extended.
302 * If it is too large, it is truncated.
304 * If the key isn't in the destination yet, a new item is inserted.
306 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
307 struct btrfs_root *root,
308 struct btrfs_path *path,
309 struct extent_buffer *eb, int slot,
310 struct btrfs_key *key)
314 u64 saved_i_size = 0;
315 int save_old_i_size = 0;
316 unsigned long src_ptr;
317 unsigned long dst_ptr;
318 int overwrite_root = 0;
320 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
323 item_size = btrfs_item_size_nr(eb, slot);
324 src_ptr = btrfs_item_ptr_offset(eb, slot);
326 /* look for the key in the destination tree */
327 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
331 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
333 if (dst_size != item_size)
336 if (item_size == 0) {
337 btrfs_release_path(path);
340 dst_copy = kmalloc(item_size, GFP_NOFS);
341 src_copy = kmalloc(item_size, GFP_NOFS);
342 if (!dst_copy || !src_copy) {
343 btrfs_release_path(path);
349 read_extent_buffer(eb, src_copy, src_ptr, item_size);
351 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
352 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
354 ret = memcmp(dst_copy, src_copy, item_size);
359 * they have the same contents, just return, this saves
360 * us from cowing blocks in the destination tree and doing
361 * extra writes that may not have been done by a previous
365 btrfs_release_path(path);
371 btrfs_release_path(path);
372 /* try to insert the key into the destination tree */
373 ret = btrfs_insert_empty_item(trans, root, path,
376 /* make sure any existing item is the correct size */
377 if (ret == -EEXIST) {
379 found_size = btrfs_item_size_nr(path->nodes[0],
381 if (found_size > item_size) {
382 btrfs_truncate_item(trans, root, path, item_size, 1);
383 } else if (found_size < item_size) {
384 ret = btrfs_extend_item(trans, root, path,
385 item_size - found_size);
390 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
393 /* don't overwrite an existing inode if the generation number
394 * was logged as zero. This is done when the tree logging code
395 * is just logging an inode to make sure it exists after recovery.
397 * Also, don't overwrite i_size on directories during replay.
398 * log replay inserts and removes directory items based on the
399 * state of the tree found in the subvolume, and i_size is modified
402 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
403 struct btrfs_inode_item *src_item;
404 struct btrfs_inode_item *dst_item;
406 src_item = (struct btrfs_inode_item *)src_ptr;
407 dst_item = (struct btrfs_inode_item *)dst_ptr;
409 if (btrfs_inode_generation(eb, src_item) == 0)
412 if (overwrite_root &&
413 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
414 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
416 saved_i_size = btrfs_inode_size(path->nodes[0],
421 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
424 if (save_old_i_size) {
425 struct btrfs_inode_item *dst_item;
426 dst_item = (struct btrfs_inode_item *)dst_ptr;
427 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
430 /* make sure the generation is filled in */
431 if (key->type == BTRFS_INODE_ITEM_KEY) {
432 struct btrfs_inode_item *dst_item;
433 dst_item = (struct btrfs_inode_item *)dst_ptr;
434 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
435 btrfs_set_inode_generation(path->nodes[0], dst_item,
440 btrfs_mark_buffer_dirty(path->nodes[0]);
441 btrfs_release_path(path);
446 * simple helper to read an inode off the disk from a given root
447 * This can only be called for subvolume roots and not for the log
449 static noinline struct inode *read_one_inode(struct btrfs_root *root,
452 struct btrfs_key key;
455 key.objectid = objectid;
456 key.type = BTRFS_INODE_ITEM_KEY;
458 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
461 } else if (is_bad_inode(inode)) {
468 /* replays a single extent in 'eb' at 'slot' with 'key' into the
469 * subvolume 'root'. path is released on entry and should be released
472 * extents in the log tree have not been allocated out of the extent
473 * tree yet. So, this completes the allocation, taking a reference
474 * as required if the extent already exists or creating a new extent
475 * if it isn't in the extent allocation tree yet.
477 * The extent is inserted into the file, dropping any existing extents
478 * from the file that overlap the new one.
480 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
481 struct btrfs_root *root,
482 struct btrfs_path *path,
483 struct extent_buffer *eb, int slot,
484 struct btrfs_key *key)
487 u64 mask = root->sectorsize - 1;
490 u64 start = key->offset;
492 struct btrfs_file_extent_item *item;
493 struct inode *inode = NULL;
497 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
498 found_type = btrfs_file_extent_type(eb, item);
500 if (found_type == BTRFS_FILE_EXTENT_REG ||
501 found_type == BTRFS_FILE_EXTENT_PREALLOC)
502 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
503 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
504 size = btrfs_file_extent_inline_len(eb, item);
505 extent_end = (start + size + mask) & ~mask;
511 inode = read_one_inode(root, key->objectid);
518 * first check to see if we already have this extent in the
519 * file. This must be done before the btrfs_drop_extents run
520 * so we don't try to drop this extent.
522 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
526 (found_type == BTRFS_FILE_EXTENT_REG ||
527 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
528 struct btrfs_file_extent_item cmp1;
529 struct btrfs_file_extent_item cmp2;
530 struct btrfs_file_extent_item *existing;
531 struct extent_buffer *leaf;
533 leaf = path->nodes[0];
534 existing = btrfs_item_ptr(leaf, path->slots[0],
535 struct btrfs_file_extent_item);
537 read_extent_buffer(eb, &cmp1, (unsigned long)item,
539 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
543 * we already have a pointer to this exact extent,
544 * we don't have to do anything
546 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
547 btrfs_release_path(path);
551 btrfs_release_path(path);
553 saved_nbytes = inode_get_bytes(inode);
554 /* drop any overlapping extents */
555 ret = btrfs_drop_extents(trans, inode, start, extent_end,
559 if (found_type == BTRFS_FILE_EXTENT_REG ||
560 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
562 unsigned long dest_offset;
563 struct btrfs_key ins;
565 ret = btrfs_insert_empty_item(trans, root, path, key,
568 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
570 copy_extent_buffer(path->nodes[0], eb, dest_offset,
571 (unsigned long)item, sizeof(*item));
573 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
574 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
575 ins.type = BTRFS_EXTENT_ITEM_KEY;
576 offset = key->offset - btrfs_file_extent_offset(eb, item);
578 if (ins.objectid > 0) {
581 LIST_HEAD(ordered_sums);
583 * is this extent already allocated in the extent
584 * allocation tree? If so, just add a reference
586 ret = btrfs_lookup_extent(root, ins.objectid,
589 ret = btrfs_inc_extent_ref(trans, root,
590 ins.objectid, ins.offset,
591 0, root->root_key.objectid,
592 key->objectid, offset);
596 * insert the extent pointer in the extent
599 ret = btrfs_alloc_logged_file_extent(trans,
600 root, root->root_key.objectid,
601 key->objectid, offset, &ins);
604 btrfs_release_path(path);
606 if (btrfs_file_extent_compression(eb, item)) {
607 csum_start = ins.objectid;
608 csum_end = csum_start + ins.offset;
610 csum_start = ins.objectid +
611 btrfs_file_extent_offset(eb, item);
612 csum_end = csum_start +
613 btrfs_file_extent_num_bytes(eb, item);
616 ret = btrfs_lookup_csums_range(root->log_root,
617 csum_start, csum_end - 1,
620 while (!list_empty(&ordered_sums)) {
621 struct btrfs_ordered_sum *sums;
622 sums = list_entry(ordered_sums.next,
623 struct btrfs_ordered_sum,
625 ret = btrfs_csum_file_blocks(trans,
626 root->fs_info->csum_root,
629 list_del(&sums->list);
633 btrfs_release_path(path);
635 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
636 /* inline extents are easy, we just overwrite them */
637 ret = overwrite_item(trans, root, path, eb, slot, key);
641 inode_set_bytes(inode, saved_nbytes);
642 btrfs_update_inode(trans, root, inode);
650 * when cleaning up conflicts between the directory names in the
651 * subvolume, directory names in the log and directory names in the
652 * inode back references, we may have to unlink inodes from directories.
654 * This is a helper function to do the unlink of a specific directory
657 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
658 struct btrfs_root *root,
659 struct btrfs_path *path,
661 struct btrfs_dir_item *di)
666 struct extent_buffer *leaf;
667 struct btrfs_key location;
670 leaf = path->nodes[0];
672 btrfs_dir_item_key_to_cpu(leaf, di, &location);
673 name_len = btrfs_dir_name_len(leaf, di);
674 name = kmalloc(name_len, GFP_NOFS);
678 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
679 btrfs_release_path(path);
681 inode = read_one_inode(root, location.objectid);
687 ret = link_to_fixup_dir(trans, root, path, location.objectid);
690 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
696 btrfs_run_delayed_items(trans, root);
701 * helper function to see if a given name and sequence number found
702 * in an inode back reference are already in a directory and correctly
703 * point to this inode
705 static noinline int inode_in_dir(struct btrfs_root *root,
706 struct btrfs_path *path,
707 u64 dirid, u64 objectid, u64 index,
708 const char *name, int name_len)
710 struct btrfs_dir_item *di;
711 struct btrfs_key location;
714 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
715 index, name, name_len, 0);
716 if (di && !IS_ERR(di)) {
717 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
718 if (location.objectid != objectid)
722 btrfs_release_path(path);
724 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
725 if (di && !IS_ERR(di)) {
726 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
727 if (location.objectid != objectid)
733 btrfs_release_path(path);
738 * helper function to check a log tree for a named back reference in
739 * an inode. This is used to decide if a back reference that is
740 * found in the subvolume conflicts with what we find in the log.
742 * inode backreferences may have multiple refs in a single item,
743 * during replay we process one reference at a time, and we don't
744 * want to delete valid links to a file from the subvolume if that
745 * link is also in the log.
747 static noinline int backref_in_log(struct btrfs_root *log,
748 struct btrfs_key *key,
749 char *name, int namelen)
751 struct btrfs_path *path;
752 struct btrfs_inode_ref *ref;
754 unsigned long ptr_end;
755 unsigned long name_ptr;
761 path = btrfs_alloc_path();
765 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
769 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
770 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
771 ptr_end = ptr + item_size;
772 while (ptr < ptr_end) {
773 ref = (struct btrfs_inode_ref *)ptr;
774 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
775 if (found_name_len == namelen) {
776 name_ptr = (unsigned long)(ref + 1);
777 ret = memcmp_extent_buffer(path->nodes[0], name,
784 ptr = (unsigned long)(ref + 1) + found_name_len;
787 btrfs_free_path(path);
793 * replay one inode back reference item found in the log tree.
794 * eb, slot and key refer to the buffer and key found in the log tree.
795 * root is the destination we are replaying into, and path is for temp
796 * use by this function. (it should be released on return).
798 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
799 struct btrfs_root *root,
800 struct btrfs_root *log,
801 struct btrfs_path *path,
802 struct extent_buffer *eb, int slot,
803 struct btrfs_key *key)
805 struct btrfs_inode_ref *ref;
806 struct btrfs_dir_item *di;
809 unsigned long ref_ptr;
810 unsigned long ref_end;
817 * it is possible that we didn't log all the parent directories
818 * for a given inode. If we don't find the dir, just don't
819 * copy the back ref in. The link count fixup code will take
822 dir = read_one_inode(root, key->offset);
826 inode = read_one_inode(root, key->objectid);
832 ref_ptr = btrfs_item_ptr_offset(eb, slot);
833 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
836 ref = (struct btrfs_inode_ref *)ref_ptr;
838 namelen = btrfs_inode_ref_name_len(eb, ref);
839 name = kmalloc(namelen, GFP_NOFS);
842 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
844 /* if we already have a perfect match, we're done */
845 if (inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
846 btrfs_inode_ref_index(eb, ref),
852 * look for a conflicting back reference in the metadata.
853 * if we find one we have to unlink that name of the file
854 * before we add our new link. Later on, we overwrite any
855 * existing back reference, and we don't want to create
856 * dangling pointers in the directory.
862 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
866 struct btrfs_inode_ref *victim_ref;
868 unsigned long ptr_end;
869 struct extent_buffer *leaf = path->nodes[0];
871 /* are we trying to overwrite a back ref for the root directory
872 * if so, just jump out, we're done
874 if (key->objectid == key->offset)
877 /* check all the names in this back reference to see
878 * if they are in the log. if so, we allow them to stay
879 * otherwise they must be unlinked as a conflict
881 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
882 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
883 while (ptr < ptr_end) {
884 victim_ref = (struct btrfs_inode_ref *)ptr;
885 victim_name_len = btrfs_inode_ref_name_len(leaf,
887 victim_name = kmalloc(victim_name_len, GFP_NOFS);
888 BUG_ON(!victim_name);
890 read_extent_buffer(leaf, victim_name,
891 (unsigned long)(victim_ref + 1),
894 if (!backref_in_log(log, key, victim_name,
896 btrfs_inc_nlink(inode);
897 btrfs_release_path(path);
899 ret = btrfs_unlink_inode(trans, root, dir,
902 btrfs_run_delayed_items(trans, root);
905 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
910 * NOTE: we have searched root tree and checked the
911 * coresponding ref, it does not need to check again.
915 btrfs_release_path(path);
917 /* look for a conflicting sequence number */
918 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
919 btrfs_inode_ref_index(eb, ref),
921 if (di && !IS_ERR(di)) {
922 ret = drop_one_dir_item(trans, root, path, dir, di);
925 btrfs_release_path(path);
927 /* look for a conflicing name */
928 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
930 if (di && !IS_ERR(di)) {
931 ret = drop_one_dir_item(trans, root, path, dir, di);
934 btrfs_release_path(path);
937 /* insert our name */
938 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
939 btrfs_inode_ref_index(eb, ref));
942 btrfs_update_inode(trans, root, inode);
945 ref_ptr = (unsigned long)(ref + 1) + namelen;
947 if (ref_ptr < ref_end)
950 /* finally write the back reference in the inode */
951 ret = overwrite_item(trans, root, path, eb, slot, key);
955 btrfs_release_path(path);
961 static int insert_orphan_item(struct btrfs_trans_handle *trans,
962 struct btrfs_root *root, u64 offset)
965 ret = btrfs_find_orphan_item(root, offset);
967 ret = btrfs_insert_orphan_item(trans, root, offset);
973 * There are a few corners where the link count of the file can't
974 * be properly maintained during replay. So, instead of adding
975 * lots of complexity to the log code, we just scan the backrefs
976 * for any file that has been through replay.
978 * The scan will update the link count on the inode to reflect the
979 * number of back refs found. If it goes down to zero, the iput
980 * will free the inode.
982 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
983 struct btrfs_root *root,
986 struct btrfs_path *path;
988 struct btrfs_key key;
991 unsigned long ptr_end;
993 u64 ino = btrfs_ino(inode);
996 key.type = BTRFS_INODE_REF_KEY;
997 key.offset = (u64)-1;
999 path = btrfs_alloc_path();
1004 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1008 if (path->slots[0] == 0)
1012 btrfs_item_key_to_cpu(path->nodes[0], &key,
1014 if (key.objectid != ino ||
1015 key.type != BTRFS_INODE_REF_KEY)
1017 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1018 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1020 while (ptr < ptr_end) {
1021 struct btrfs_inode_ref *ref;
1023 ref = (struct btrfs_inode_ref *)ptr;
1024 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1026 ptr = (unsigned long)(ref + 1) + name_len;
1030 if (key.offset == 0)
1033 btrfs_release_path(path);
1035 btrfs_release_path(path);
1036 if (nlink != inode->i_nlink) {
1037 set_nlink(inode, nlink);
1038 btrfs_update_inode(trans, root, inode);
1040 BTRFS_I(inode)->index_cnt = (u64)-1;
1042 if (inode->i_nlink == 0) {
1043 if (S_ISDIR(inode->i_mode)) {
1044 ret = replay_dir_deletes(trans, root, NULL, path,
1048 ret = insert_orphan_item(trans, root, ino);
1051 btrfs_free_path(path);
1056 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1057 struct btrfs_root *root,
1058 struct btrfs_path *path)
1061 struct btrfs_key key;
1062 struct inode *inode;
1064 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1065 key.type = BTRFS_ORPHAN_ITEM_KEY;
1066 key.offset = (u64)-1;
1068 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1073 if (path->slots[0] == 0)
1078 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1079 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1080 key.type != BTRFS_ORPHAN_ITEM_KEY)
1083 ret = btrfs_del_item(trans, root, path);
1087 btrfs_release_path(path);
1088 inode = read_one_inode(root, key.offset);
1092 ret = fixup_inode_link_count(trans, root, inode);
1098 * fixup on a directory may create new entries,
1099 * make sure we always look for the highset possible
1102 key.offset = (u64)-1;
1106 btrfs_release_path(path);
1112 * record a given inode in the fixup dir so we can check its link
1113 * count when replay is done. The link count is incremented here
1114 * so the inode won't go away until we check it
1116 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1117 struct btrfs_root *root,
1118 struct btrfs_path *path,
1121 struct btrfs_key key;
1123 struct inode *inode;
1125 inode = read_one_inode(root, objectid);
1129 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1130 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1131 key.offset = objectid;
1133 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1135 btrfs_release_path(path);
1137 btrfs_inc_nlink(inode);
1138 btrfs_update_inode(trans, root, inode);
1139 } else if (ret == -EEXIST) {
1150 * when replaying the log for a directory, we only insert names
1151 * for inodes that actually exist. This means an fsync on a directory
1152 * does not implicitly fsync all the new files in it
1154 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1155 struct btrfs_root *root,
1156 struct btrfs_path *path,
1157 u64 dirid, u64 index,
1158 char *name, int name_len, u8 type,
1159 struct btrfs_key *location)
1161 struct inode *inode;
1165 inode = read_one_inode(root, location->objectid);
1169 dir = read_one_inode(root, dirid);
1174 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1176 /* FIXME, put inode into FIXUP list */
1184 * take a single entry in a log directory item and replay it into
1187 * if a conflicting item exists in the subdirectory already,
1188 * the inode it points to is unlinked and put into the link count
1191 * If a name from the log points to a file or directory that does
1192 * not exist in the FS, it is skipped. fsyncs on directories
1193 * do not force down inodes inside that directory, just changes to the
1194 * names or unlinks in a directory.
1196 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1197 struct btrfs_root *root,
1198 struct btrfs_path *path,
1199 struct extent_buffer *eb,
1200 struct btrfs_dir_item *di,
1201 struct btrfs_key *key)
1205 struct btrfs_dir_item *dst_di;
1206 struct btrfs_key found_key;
1207 struct btrfs_key log_key;
1213 dir = read_one_inode(root, key->objectid);
1217 name_len = btrfs_dir_name_len(eb, di);
1218 name = kmalloc(name_len, GFP_NOFS);
1222 log_type = btrfs_dir_type(eb, di);
1223 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1226 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1227 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1232 btrfs_release_path(path);
1234 if (key->type == BTRFS_DIR_ITEM_KEY) {
1235 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1237 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1238 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1245 if (IS_ERR_OR_NULL(dst_di)) {
1246 /* we need a sequence number to insert, so we only
1247 * do inserts for the BTRFS_DIR_INDEX_KEY types
1249 if (key->type != BTRFS_DIR_INDEX_KEY)
1254 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1255 /* the existing item matches the logged item */
1256 if (found_key.objectid == log_key.objectid &&
1257 found_key.type == log_key.type &&
1258 found_key.offset == log_key.offset &&
1259 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1264 * don't drop the conflicting directory entry if the inode
1265 * for the new entry doesn't exist
1270 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1273 if (key->type == BTRFS_DIR_INDEX_KEY)
1276 btrfs_release_path(path);
1282 btrfs_release_path(path);
1283 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1284 name, name_len, log_type, &log_key);
1286 BUG_ON(ret && ret != -ENOENT);
1291 * find all the names in a directory item and reconcile them into
1292 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1293 * one name in a directory item, but the same code gets used for
1294 * both directory index types
1296 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1297 struct btrfs_root *root,
1298 struct btrfs_path *path,
1299 struct extent_buffer *eb, int slot,
1300 struct btrfs_key *key)
1303 u32 item_size = btrfs_item_size_nr(eb, slot);
1304 struct btrfs_dir_item *di;
1307 unsigned long ptr_end;
1309 ptr = btrfs_item_ptr_offset(eb, slot);
1310 ptr_end = ptr + item_size;
1311 while (ptr < ptr_end) {
1312 di = (struct btrfs_dir_item *)ptr;
1313 if (verify_dir_item(root, eb, di))
1315 name_len = btrfs_dir_name_len(eb, di);
1316 ret = replay_one_name(trans, root, path, eb, di, key);
1318 ptr = (unsigned long)(di + 1);
1325 * directory replay has two parts. There are the standard directory
1326 * items in the log copied from the subvolume, and range items
1327 * created in the log while the subvolume was logged.
1329 * The range items tell us which parts of the key space the log
1330 * is authoritative for. During replay, if a key in the subvolume
1331 * directory is in a logged range item, but not actually in the log
1332 * that means it was deleted from the directory before the fsync
1333 * and should be removed.
1335 static noinline int find_dir_range(struct btrfs_root *root,
1336 struct btrfs_path *path,
1337 u64 dirid, int key_type,
1338 u64 *start_ret, u64 *end_ret)
1340 struct btrfs_key key;
1342 struct btrfs_dir_log_item *item;
1346 if (*start_ret == (u64)-1)
1349 key.objectid = dirid;
1350 key.type = key_type;
1351 key.offset = *start_ret;
1353 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1357 if (path->slots[0] == 0)
1362 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1364 if (key.type != key_type || key.objectid != dirid) {
1368 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1369 struct btrfs_dir_log_item);
1370 found_end = btrfs_dir_log_end(path->nodes[0], item);
1372 if (*start_ret >= key.offset && *start_ret <= found_end) {
1374 *start_ret = key.offset;
1375 *end_ret = found_end;
1380 /* check the next slot in the tree to see if it is a valid item */
1381 nritems = btrfs_header_nritems(path->nodes[0]);
1382 if (path->slots[0] >= nritems) {
1383 ret = btrfs_next_leaf(root, path);
1390 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1392 if (key.type != key_type || key.objectid != dirid) {
1396 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1397 struct btrfs_dir_log_item);
1398 found_end = btrfs_dir_log_end(path->nodes[0], item);
1399 *start_ret = key.offset;
1400 *end_ret = found_end;
1403 btrfs_release_path(path);
1408 * this looks for a given directory item in the log. If the directory
1409 * item is not in the log, the item is removed and the inode it points
1412 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1413 struct btrfs_root *root,
1414 struct btrfs_root *log,
1415 struct btrfs_path *path,
1416 struct btrfs_path *log_path,
1418 struct btrfs_key *dir_key)
1421 struct extent_buffer *eb;
1424 struct btrfs_dir_item *di;
1425 struct btrfs_dir_item *log_di;
1428 unsigned long ptr_end;
1430 struct inode *inode;
1431 struct btrfs_key location;
1434 eb = path->nodes[0];
1435 slot = path->slots[0];
1436 item_size = btrfs_item_size_nr(eb, slot);
1437 ptr = btrfs_item_ptr_offset(eb, slot);
1438 ptr_end = ptr + item_size;
1439 while (ptr < ptr_end) {
1440 di = (struct btrfs_dir_item *)ptr;
1441 if (verify_dir_item(root, eb, di)) {
1446 name_len = btrfs_dir_name_len(eb, di);
1447 name = kmalloc(name_len, GFP_NOFS);
1452 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1455 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1456 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1459 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1460 log_di = btrfs_lookup_dir_index_item(trans, log,
1466 if (IS_ERR_OR_NULL(log_di)) {
1467 btrfs_dir_item_key_to_cpu(eb, di, &location);
1468 btrfs_release_path(path);
1469 btrfs_release_path(log_path);
1470 inode = read_one_inode(root, location.objectid);
1476 ret = link_to_fixup_dir(trans, root,
1477 path, location.objectid);
1479 btrfs_inc_nlink(inode);
1480 ret = btrfs_unlink_inode(trans, root, dir, inode,
1484 btrfs_run_delayed_items(trans, root);
1489 /* there might still be more names under this key
1490 * check and repeat if required
1492 ret = btrfs_search_slot(NULL, root, dir_key, path,
1499 btrfs_release_path(log_path);
1502 ptr = (unsigned long)(di + 1);
1507 btrfs_release_path(path);
1508 btrfs_release_path(log_path);
1513 * deletion replay happens before we copy any new directory items
1514 * out of the log or out of backreferences from inodes. It
1515 * scans the log to find ranges of keys that log is authoritative for,
1516 * and then scans the directory to find items in those ranges that are
1517 * not present in the log.
1519 * Anything we don't find in the log is unlinked and removed from the
1522 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1523 struct btrfs_root *root,
1524 struct btrfs_root *log,
1525 struct btrfs_path *path,
1526 u64 dirid, int del_all)
1530 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1532 struct btrfs_key dir_key;
1533 struct btrfs_key found_key;
1534 struct btrfs_path *log_path;
1537 dir_key.objectid = dirid;
1538 dir_key.type = BTRFS_DIR_ITEM_KEY;
1539 log_path = btrfs_alloc_path();
1543 dir = read_one_inode(root, dirid);
1544 /* it isn't an error if the inode isn't there, that can happen
1545 * because we replay the deletes before we copy in the inode item
1549 btrfs_free_path(log_path);
1557 range_end = (u64)-1;
1559 ret = find_dir_range(log, path, dirid, key_type,
1560 &range_start, &range_end);
1565 dir_key.offset = range_start;
1568 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1573 nritems = btrfs_header_nritems(path->nodes[0]);
1574 if (path->slots[0] >= nritems) {
1575 ret = btrfs_next_leaf(root, path);
1579 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1581 if (found_key.objectid != dirid ||
1582 found_key.type != dir_key.type)
1585 if (found_key.offset > range_end)
1588 ret = check_item_in_log(trans, root, log, path,
1592 if (found_key.offset == (u64)-1)
1594 dir_key.offset = found_key.offset + 1;
1596 btrfs_release_path(path);
1597 if (range_end == (u64)-1)
1599 range_start = range_end + 1;
1604 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1605 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1606 dir_key.type = BTRFS_DIR_INDEX_KEY;
1607 btrfs_release_path(path);
1611 btrfs_release_path(path);
1612 btrfs_free_path(log_path);
1618 * the process_func used to replay items from the log tree. This
1619 * gets called in two different stages. The first stage just looks
1620 * for inodes and makes sure they are all copied into the subvolume.
1622 * The second stage copies all the other item types from the log into
1623 * the subvolume. The two stage approach is slower, but gets rid of
1624 * lots of complexity around inodes referencing other inodes that exist
1625 * only in the log (references come from either directory items or inode
1628 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1629 struct walk_control *wc, u64 gen)
1632 struct btrfs_path *path;
1633 struct btrfs_root *root = wc->replay_dest;
1634 struct btrfs_key key;
1639 btrfs_read_buffer(eb, gen);
1641 level = btrfs_header_level(eb);
1646 path = btrfs_alloc_path();
1650 nritems = btrfs_header_nritems(eb);
1651 for (i = 0; i < nritems; i++) {
1652 btrfs_item_key_to_cpu(eb, &key, i);
1654 /* inode keys are done during the first stage */
1655 if (key.type == BTRFS_INODE_ITEM_KEY &&
1656 wc->stage == LOG_WALK_REPLAY_INODES) {
1657 struct btrfs_inode_item *inode_item;
1660 inode_item = btrfs_item_ptr(eb, i,
1661 struct btrfs_inode_item);
1662 mode = btrfs_inode_mode(eb, inode_item);
1663 if (S_ISDIR(mode)) {
1664 ret = replay_dir_deletes(wc->trans,
1665 root, log, path, key.objectid, 0);
1668 ret = overwrite_item(wc->trans, root, path,
1672 /* for regular files, make sure corresponding
1673 * orhpan item exist. extents past the new EOF
1674 * will be truncated later by orphan cleanup.
1676 if (S_ISREG(mode)) {
1677 ret = insert_orphan_item(wc->trans, root,
1682 ret = link_to_fixup_dir(wc->trans, root,
1683 path, key.objectid);
1686 if (wc->stage < LOG_WALK_REPLAY_ALL)
1689 /* these keys are simply copied */
1690 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1691 ret = overwrite_item(wc->trans, root, path,
1694 } else if (key.type == BTRFS_INODE_REF_KEY) {
1695 ret = add_inode_ref(wc->trans, root, log, path,
1697 BUG_ON(ret && ret != -ENOENT);
1698 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1699 ret = replay_one_extent(wc->trans, root, path,
1702 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1703 key.type == BTRFS_DIR_INDEX_KEY) {
1704 ret = replay_one_dir_item(wc->trans, root, path,
1709 btrfs_free_path(path);
1713 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
1714 struct btrfs_root *root,
1715 struct btrfs_path *path, int *level,
1716 struct walk_control *wc)
1721 struct extent_buffer *next;
1722 struct extent_buffer *cur;
1723 struct extent_buffer *parent;
1727 WARN_ON(*level < 0);
1728 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1730 while (*level > 0) {
1731 WARN_ON(*level < 0);
1732 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1733 cur = path->nodes[*level];
1735 if (btrfs_header_level(cur) != *level)
1738 if (path->slots[*level] >=
1739 btrfs_header_nritems(cur))
1742 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1743 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1744 blocksize = btrfs_level_size(root, *level - 1);
1746 parent = path->nodes[*level];
1747 root_owner = btrfs_header_owner(parent);
1749 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1754 ret = wc->process_func(root, next, wc, ptr_gen);
1758 path->slots[*level]++;
1760 btrfs_read_buffer(next, ptr_gen);
1762 btrfs_tree_lock(next);
1763 btrfs_set_lock_blocking(next);
1764 clean_tree_block(trans, root, next);
1765 btrfs_wait_tree_block_writeback(next);
1766 btrfs_tree_unlock(next);
1768 WARN_ON(root_owner !=
1769 BTRFS_TREE_LOG_OBJECTID);
1770 ret = btrfs_free_and_pin_reserved_extent(root,
1774 free_extent_buffer(next);
1777 btrfs_read_buffer(next, ptr_gen);
1779 WARN_ON(*level <= 0);
1780 if (path->nodes[*level-1])
1781 free_extent_buffer(path->nodes[*level-1]);
1782 path->nodes[*level-1] = next;
1783 *level = btrfs_header_level(next);
1784 path->slots[*level] = 0;
1787 WARN_ON(*level < 0);
1788 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1790 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1796 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
1797 struct btrfs_root *root,
1798 struct btrfs_path *path, int *level,
1799 struct walk_control *wc)
1806 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1807 slot = path->slots[i];
1808 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
1811 WARN_ON(*level == 0);
1814 struct extent_buffer *parent;
1815 if (path->nodes[*level] == root->node)
1816 parent = path->nodes[*level];
1818 parent = path->nodes[*level + 1];
1820 root_owner = btrfs_header_owner(parent);
1821 ret = wc->process_func(root, path->nodes[*level], wc,
1822 btrfs_header_generation(path->nodes[*level]));
1827 struct extent_buffer *next;
1829 next = path->nodes[*level];
1831 btrfs_tree_lock(next);
1832 btrfs_set_lock_blocking(next);
1833 clean_tree_block(trans, root, next);
1834 btrfs_wait_tree_block_writeback(next);
1835 btrfs_tree_unlock(next);
1837 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1838 ret = btrfs_free_and_pin_reserved_extent(root,
1839 path->nodes[*level]->start,
1840 path->nodes[*level]->len);
1843 free_extent_buffer(path->nodes[*level]);
1844 path->nodes[*level] = NULL;
1852 * drop the reference count on the tree rooted at 'snap'. This traverses
1853 * the tree freeing any blocks that have a ref count of zero after being
1856 static int walk_log_tree(struct btrfs_trans_handle *trans,
1857 struct btrfs_root *log, struct walk_control *wc)
1862 struct btrfs_path *path;
1866 path = btrfs_alloc_path();
1870 level = btrfs_header_level(log->node);
1872 path->nodes[level] = log->node;
1873 extent_buffer_get(log->node);
1874 path->slots[level] = 0;
1877 wret = walk_down_log_tree(trans, log, path, &level, wc);
1883 wret = walk_up_log_tree(trans, log, path, &level, wc);
1890 /* was the root node processed? if not, catch it here */
1891 if (path->nodes[orig_level]) {
1892 wc->process_func(log, path->nodes[orig_level], wc,
1893 btrfs_header_generation(path->nodes[orig_level]));
1895 struct extent_buffer *next;
1897 next = path->nodes[orig_level];
1899 btrfs_tree_lock(next);
1900 btrfs_set_lock_blocking(next);
1901 clean_tree_block(trans, log, next);
1902 btrfs_wait_tree_block_writeback(next);
1903 btrfs_tree_unlock(next);
1905 WARN_ON(log->root_key.objectid !=
1906 BTRFS_TREE_LOG_OBJECTID);
1907 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
1913 for (i = 0; i <= orig_level; i++) {
1914 if (path->nodes[i]) {
1915 free_extent_buffer(path->nodes[i]);
1916 path->nodes[i] = NULL;
1919 btrfs_free_path(path);
1924 * helper function to update the item for a given subvolumes log root
1925 * in the tree of log roots
1927 static int update_log_root(struct btrfs_trans_handle *trans,
1928 struct btrfs_root *log)
1932 if (log->log_transid == 1) {
1933 /* insert root item on the first sync */
1934 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1935 &log->root_key, &log->root_item);
1937 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1938 &log->root_key, &log->root_item);
1943 static int wait_log_commit(struct btrfs_trans_handle *trans,
1944 struct btrfs_root *root, unsigned long transid)
1947 int index = transid % 2;
1950 * we only allow two pending log transactions at a time,
1951 * so we know that if ours is more than 2 older than the
1952 * current transaction, we're done
1955 prepare_to_wait(&root->log_commit_wait[index],
1956 &wait, TASK_UNINTERRUPTIBLE);
1957 mutex_unlock(&root->log_mutex);
1959 if (root->fs_info->last_trans_log_full_commit !=
1960 trans->transid && root->log_transid < transid + 2 &&
1961 atomic_read(&root->log_commit[index]))
1964 finish_wait(&root->log_commit_wait[index], &wait);
1965 mutex_lock(&root->log_mutex);
1966 } while (root->log_transid < transid + 2 &&
1967 atomic_read(&root->log_commit[index]));
1971 static int wait_for_writer(struct btrfs_trans_handle *trans,
1972 struct btrfs_root *root)
1975 while (atomic_read(&root->log_writers)) {
1976 prepare_to_wait(&root->log_writer_wait,
1977 &wait, TASK_UNINTERRUPTIBLE);
1978 mutex_unlock(&root->log_mutex);
1979 if (root->fs_info->last_trans_log_full_commit !=
1980 trans->transid && atomic_read(&root->log_writers))
1982 mutex_lock(&root->log_mutex);
1983 finish_wait(&root->log_writer_wait, &wait);
1989 * btrfs_sync_log does sends a given tree log down to the disk and
1990 * updates the super blocks to record it. When this call is done,
1991 * you know that any inodes previously logged are safely on disk only
1994 * Any other return value means you need to call btrfs_commit_transaction.
1995 * Some of the edge cases for fsyncing directories that have had unlinks
1996 * or renames done in the past mean that sometimes the only safe
1997 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1998 * that has happened.
2000 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2001 struct btrfs_root *root)
2007 struct btrfs_root *log = root->log_root;
2008 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2009 unsigned long log_transid = 0;
2011 mutex_lock(&root->log_mutex);
2012 index1 = root->log_transid % 2;
2013 if (atomic_read(&root->log_commit[index1])) {
2014 wait_log_commit(trans, root, root->log_transid);
2015 mutex_unlock(&root->log_mutex);
2018 atomic_set(&root->log_commit[index1], 1);
2020 /* wait for previous tree log sync to complete */
2021 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2022 wait_log_commit(trans, root, root->log_transid - 1);
2024 unsigned long batch = root->log_batch;
2025 /* when we're on an ssd, just kick the log commit out */
2026 if (!btrfs_test_opt(root, SSD) && root->log_multiple_pids) {
2027 mutex_unlock(&root->log_mutex);
2028 schedule_timeout_uninterruptible(1);
2029 mutex_lock(&root->log_mutex);
2031 wait_for_writer(trans, root);
2032 if (batch == root->log_batch)
2036 /* bail out if we need to do a full commit */
2037 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2039 mutex_unlock(&root->log_mutex);
2043 log_transid = root->log_transid;
2044 if (log_transid % 2 == 0)
2045 mark = EXTENT_DIRTY;
2049 /* we start IO on all the marked extents here, but we don't actually
2050 * wait for them until later.
2052 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2055 btrfs_set_root_node(&log->root_item, log->node);
2057 root->log_batch = 0;
2058 root->log_transid++;
2059 log->log_transid = root->log_transid;
2060 root->log_start_pid = 0;
2063 * IO has been started, blocks of the log tree have WRITTEN flag set
2064 * in their headers. new modifications of the log will be written to
2065 * new positions. so it's safe to allow log writers to go in.
2067 mutex_unlock(&root->log_mutex);
2069 mutex_lock(&log_root_tree->log_mutex);
2070 log_root_tree->log_batch++;
2071 atomic_inc(&log_root_tree->log_writers);
2072 mutex_unlock(&log_root_tree->log_mutex);
2074 ret = update_log_root(trans, log);
2076 mutex_lock(&log_root_tree->log_mutex);
2077 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2079 if (waitqueue_active(&log_root_tree->log_writer_wait))
2080 wake_up(&log_root_tree->log_writer_wait);
2084 BUG_ON(ret != -ENOSPC);
2085 root->fs_info->last_trans_log_full_commit = trans->transid;
2086 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2087 mutex_unlock(&log_root_tree->log_mutex);
2092 index2 = log_root_tree->log_transid % 2;
2093 if (atomic_read(&log_root_tree->log_commit[index2])) {
2094 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2095 wait_log_commit(trans, log_root_tree,
2096 log_root_tree->log_transid);
2097 mutex_unlock(&log_root_tree->log_mutex);
2101 atomic_set(&log_root_tree->log_commit[index2], 1);
2103 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2104 wait_log_commit(trans, log_root_tree,
2105 log_root_tree->log_transid - 1);
2108 wait_for_writer(trans, log_root_tree);
2111 * now that we've moved on to the tree of log tree roots,
2112 * check the full commit flag again
2114 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2115 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2116 mutex_unlock(&log_root_tree->log_mutex);
2118 goto out_wake_log_root;
2121 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2122 &log_root_tree->dirty_log_pages,
2123 EXTENT_DIRTY | EXTENT_NEW);
2125 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2127 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2128 log_root_tree->node->start);
2129 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2130 btrfs_header_level(log_root_tree->node));
2132 log_root_tree->log_batch = 0;
2133 log_root_tree->log_transid++;
2136 mutex_unlock(&log_root_tree->log_mutex);
2139 * nobody else is going to jump in and write the the ctree
2140 * super here because the log_commit atomic below is protecting
2141 * us. We must be called with a transaction handle pinning
2142 * the running transaction open, so a full commit can't hop
2143 * in and cause problems either.
2145 btrfs_scrub_pause_super(root);
2146 write_ctree_super(trans, root->fs_info->tree_root, 1);
2147 btrfs_scrub_continue_super(root);
2150 mutex_lock(&root->log_mutex);
2151 if (root->last_log_commit < log_transid)
2152 root->last_log_commit = log_transid;
2153 mutex_unlock(&root->log_mutex);
2156 atomic_set(&log_root_tree->log_commit[index2], 0);
2158 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2159 wake_up(&log_root_tree->log_commit_wait[index2]);
2161 atomic_set(&root->log_commit[index1], 0);
2163 if (waitqueue_active(&root->log_commit_wait[index1]))
2164 wake_up(&root->log_commit_wait[index1]);
2168 static void free_log_tree(struct btrfs_trans_handle *trans,
2169 struct btrfs_root *log)
2174 struct walk_control wc = {
2176 .process_func = process_one_buffer
2179 ret = walk_log_tree(trans, log, &wc);
2183 ret = find_first_extent_bit(&log->dirty_log_pages,
2184 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
2188 clear_extent_bits(&log->dirty_log_pages, start, end,
2189 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2192 free_extent_buffer(log->node);
2197 * free all the extents used by the tree log. This should be called
2198 * at commit time of the full transaction
2200 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2202 if (root->log_root) {
2203 free_log_tree(trans, root->log_root);
2204 root->log_root = NULL;
2209 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2210 struct btrfs_fs_info *fs_info)
2212 if (fs_info->log_root_tree) {
2213 free_log_tree(trans, fs_info->log_root_tree);
2214 fs_info->log_root_tree = NULL;
2220 * If both a file and directory are logged, and unlinks or renames are
2221 * mixed in, we have a few interesting corners:
2223 * create file X in dir Y
2224 * link file X to X.link in dir Y
2226 * unlink file X but leave X.link
2229 * After a crash we would expect only X.link to exist. But file X
2230 * didn't get fsync'd again so the log has back refs for X and X.link.
2232 * We solve this by removing directory entries and inode backrefs from the
2233 * log when a file that was logged in the current transaction is
2234 * unlinked. Any later fsync will include the updated log entries, and
2235 * we'll be able to reconstruct the proper directory items from backrefs.
2237 * This optimizations allows us to avoid relogging the entire inode
2238 * or the entire directory.
2240 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2241 struct btrfs_root *root,
2242 const char *name, int name_len,
2243 struct inode *dir, u64 index)
2245 struct btrfs_root *log;
2246 struct btrfs_dir_item *di;
2247 struct btrfs_path *path;
2251 u64 dir_ino = btrfs_ino(dir);
2253 if (BTRFS_I(dir)->logged_trans < trans->transid)
2256 ret = join_running_log_trans(root);
2260 mutex_lock(&BTRFS_I(dir)->log_mutex);
2262 log = root->log_root;
2263 path = btrfs_alloc_path();
2269 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
2270 name, name_len, -1);
2276 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2277 bytes_del += name_len;
2280 btrfs_release_path(path);
2281 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
2282 index, name, name_len, -1);
2288 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2289 bytes_del += name_len;
2293 /* update the directory size in the log to reflect the names
2297 struct btrfs_key key;
2299 key.objectid = dir_ino;
2301 key.type = BTRFS_INODE_ITEM_KEY;
2302 btrfs_release_path(path);
2304 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2310 struct btrfs_inode_item *item;
2313 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2314 struct btrfs_inode_item);
2315 i_size = btrfs_inode_size(path->nodes[0], item);
2316 if (i_size > bytes_del)
2317 i_size -= bytes_del;
2320 btrfs_set_inode_size(path->nodes[0], item, i_size);
2321 btrfs_mark_buffer_dirty(path->nodes[0]);
2324 btrfs_release_path(path);
2327 btrfs_free_path(path);
2329 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2330 if (ret == -ENOSPC) {
2331 root->fs_info->last_trans_log_full_commit = trans->transid;
2334 btrfs_end_log_trans(root);
2339 /* see comments for btrfs_del_dir_entries_in_log */
2340 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2341 struct btrfs_root *root,
2342 const char *name, int name_len,
2343 struct inode *inode, u64 dirid)
2345 struct btrfs_root *log;
2349 if (BTRFS_I(inode)->logged_trans < trans->transid)
2352 ret = join_running_log_trans(root);
2355 log = root->log_root;
2356 mutex_lock(&BTRFS_I(inode)->log_mutex);
2358 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
2360 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2361 if (ret == -ENOSPC) {
2362 root->fs_info->last_trans_log_full_commit = trans->transid;
2365 btrfs_end_log_trans(root);
2371 * creates a range item in the log for 'dirid'. first_offset and
2372 * last_offset tell us which parts of the key space the log should
2373 * be considered authoritative for.
2375 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2376 struct btrfs_root *log,
2377 struct btrfs_path *path,
2378 int key_type, u64 dirid,
2379 u64 first_offset, u64 last_offset)
2382 struct btrfs_key key;
2383 struct btrfs_dir_log_item *item;
2385 key.objectid = dirid;
2386 key.offset = first_offset;
2387 if (key_type == BTRFS_DIR_ITEM_KEY)
2388 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2390 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2391 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2395 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2396 struct btrfs_dir_log_item);
2397 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2398 btrfs_mark_buffer_dirty(path->nodes[0]);
2399 btrfs_release_path(path);
2404 * log all the items included in the current transaction for a given
2405 * directory. This also creates the range items in the log tree required
2406 * to replay anything deleted before the fsync
2408 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2409 struct btrfs_root *root, struct inode *inode,
2410 struct btrfs_path *path,
2411 struct btrfs_path *dst_path, int key_type,
2412 u64 min_offset, u64 *last_offset_ret)
2414 struct btrfs_key min_key;
2415 struct btrfs_key max_key;
2416 struct btrfs_root *log = root->log_root;
2417 struct extent_buffer *src;
2422 u64 first_offset = min_offset;
2423 u64 last_offset = (u64)-1;
2424 u64 ino = btrfs_ino(inode);
2426 log = root->log_root;
2427 max_key.objectid = ino;
2428 max_key.offset = (u64)-1;
2429 max_key.type = key_type;
2431 min_key.objectid = ino;
2432 min_key.type = key_type;
2433 min_key.offset = min_offset;
2435 path->keep_locks = 1;
2437 ret = btrfs_search_forward(root, &min_key, &max_key,
2438 path, 0, trans->transid);
2441 * we didn't find anything from this transaction, see if there
2442 * is anything at all
2444 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2445 min_key.objectid = ino;
2446 min_key.type = key_type;
2447 min_key.offset = (u64)-1;
2448 btrfs_release_path(path);
2449 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2451 btrfs_release_path(path);
2454 ret = btrfs_previous_item(root, path, ino, key_type);
2456 /* if ret == 0 there are items for this type,
2457 * create a range to tell us the last key of this type.
2458 * otherwise, there are no items in this directory after
2459 * *min_offset, and we create a range to indicate that.
2462 struct btrfs_key tmp;
2463 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2465 if (key_type == tmp.type)
2466 first_offset = max(min_offset, tmp.offset) + 1;
2471 /* go backward to find any previous key */
2472 ret = btrfs_previous_item(root, path, ino, key_type);
2474 struct btrfs_key tmp;
2475 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2476 if (key_type == tmp.type) {
2477 first_offset = tmp.offset;
2478 ret = overwrite_item(trans, log, dst_path,
2479 path->nodes[0], path->slots[0],
2487 btrfs_release_path(path);
2489 /* find the first key from this transaction again */
2490 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2497 * we have a block from this transaction, log every item in it
2498 * from our directory
2501 struct btrfs_key tmp;
2502 src = path->nodes[0];
2503 nritems = btrfs_header_nritems(src);
2504 for (i = path->slots[0]; i < nritems; i++) {
2505 btrfs_item_key_to_cpu(src, &min_key, i);
2507 if (min_key.objectid != ino || min_key.type != key_type)
2509 ret = overwrite_item(trans, log, dst_path, src, i,
2516 path->slots[0] = nritems;
2519 * look ahead to the next item and see if it is also
2520 * from this directory and from this transaction
2522 ret = btrfs_next_leaf(root, path);
2524 last_offset = (u64)-1;
2527 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2528 if (tmp.objectid != ino || tmp.type != key_type) {
2529 last_offset = (u64)-1;
2532 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2533 ret = overwrite_item(trans, log, dst_path,
2534 path->nodes[0], path->slots[0],
2539 last_offset = tmp.offset;
2544 btrfs_release_path(path);
2545 btrfs_release_path(dst_path);
2548 *last_offset_ret = last_offset;
2550 * insert the log range keys to indicate where the log
2553 ret = insert_dir_log_key(trans, log, path, key_type,
2554 ino, first_offset, last_offset);
2562 * logging directories is very similar to logging inodes, We find all the items
2563 * from the current transaction and write them to the log.
2565 * The recovery code scans the directory in the subvolume, and if it finds a
2566 * key in the range logged that is not present in the log tree, then it means
2567 * that dir entry was unlinked during the transaction.
2569 * In order for that scan to work, we must include one key smaller than
2570 * the smallest logged by this transaction and one key larger than the largest
2571 * key logged by this transaction.
2573 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2574 struct btrfs_root *root, struct inode *inode,
2575 struct btrfs_path *path,
2576 struct btrfs_path *dst_path)
2581 int key_type = BTRFS_DIR_ITEM_KEY;
2587 ret = log_dir_items(trans, root, inode, path,
2588 dst_path, key_type, min_key,
2592 if (max_key == (u64)-1)
2594 min_key = max_key + 1;
2597 if (key_type == BTRFS_DIR_ITEM_KEY) {
2598 key_type = BTRFS_DIR_INDEX_KEY;
2605 * a helper function to drop items from the log before we relog an
2606 * inode. max_key_type indicates the highest item type to remove.
2607 * This cannot be run for file data extents because it does not
2608 * free the extents they point to.
2610 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2611 struct btrfs_root *log,
2612 struct btrfs_path *path,
2613 u64 objectid, int max_key_type)
2616 struct btrfs_key key;
2617 struct btrfs_key found_key;
2619 key.objectid = objectid;
2620 key.type = max_key_type;
2621 key.offset = (u64)-1;
2624 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2629 if (path->slots[0] == 0)
2633 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2636 if (found_key.objectid != objectid)
2639 ret = btrfs_del_item(trans, log, path);
2642 btrfs_release_path(path);
2644 btrfs_release_path(path);
2648 static noinline int copy_items(struct btrfs_trans_handle *trans,
2649 struct btrfs_root *log,
2650 struct btrfs_path *dst_path,
2651 struct extent_buffer *src,
2652 int start_slot, int nr, int inode_only)
2654 unsigned long src_offset;
2655 unsigned long dst_offset;
2656 struct btrfs_file_extent_item *extent;
2657 struct btrfs_inode_item *inode_item;
2659 struct btrfs_key *ins_keys;
2663 struct list_head ordered_sums;
2665 INIT_LIST_HEAD(&ordered_sums);
2667 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2668 nr * sizeof(u32), GFP_NOFS);
2672 ins_sizes = (u32 *)ins_data;
2673 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2675 for (i = 0; i < nr; i++) {
2676 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2677 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2679 ret = btrfs_insert_empty_items(trans, log, dst_path,
2680 ins_keys, ins_sizes, nr);
2686 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
2687 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2688 dst_path->slots[0]);
2690 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2692 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2693 src_offset, ins_sizes[i]);
2695 if (inode_only == LOG_INODE_EXISTS &&
2696 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2697 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2699 struct btrfs_inode_item);
2700 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2702 /* set the generation to zero so the recover code
2703 * can tell the difference between an logging
2704 * just to say 'this inode exists' and a logging
2705 * to say 'update this inode with these values'
2707 btrfs_set_inode_generation(dst_path->nodes[0],
2710 /* take a reference on file data extents so that truncates
2711 * or deletes of this inode don't have to relog the inode
2714 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2716 extent = btrfs_item_ptr(src, start_slot + i,
2717 struct btrfs_file_extent_item);
2719 if (btrfs_file_extent_generation(src, extent) < trans->transid)
2722 found_type = btrfs_file_extent_type(src, extent);
2723 if (found_type == BTRFS_FILE_EXTENT_REG ||
2724 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
2726 ds = btrfs_file_extent_disk_bytenr(src,
2728 /* ds == 0 is a hole */
2732 dl = btrfs_file_extent_disk_num_bytes(src,
2734 cs = btrfs_file_extent_offset(src, extent);
2735 cl = btrfs_file_extent_num_bytes(src,
2737 if (btrfs_file_extent_compression(src,
2743 ret = btrfs_lookup_csums_range(
2744 log->fs_info->csum_root,
2745 ds + cs, ds + cs + cl - 1,
2752 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2753 btrfs_release_path(dst_path);
2757 * we have to do this after the loop above to avoid changing the
2758 * log tree while trying to change the log tree.
2761 while (!list_empty(&ordered_sums)) {
2762 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2763 struct btrfs_ordered_sum,
2766 ret = btrfs_csum_file_blocks(trans, log, sums);
2767 list_del(&sums->list);
2773 /* log a single inode in the tree log.
2774 * At least one parent directory for this inode must exist in the tree
2775 * or be logged already.
2777 * Any items from this inode changed by the current transaction are copied
2778 * to the log tree. An extra reference is taken on any extents in this
2779 * file, allowing us to avoid a whole pile of corner cases around logging
2780 * blocks that have been removed from the tree.
2782 * See LOG_INODE_ALL and related defines for a description of what inode_only
2785 * This handles both files and directories.
2787 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
2788 struct btrfs_root *root, struct inode *inode,
2791 struct btrfs_path *path;
2792 struct btrfs_path *dst_path;
2793 struct btrfs_key min_key;
2794 struct btrfs_key max_key;
2795 struct btrfs_root *log = root->log_root;
2796 struct extent_buffer *src = NULL;
2800 int ins_start_slot = 0;
2802 u64 ino = btrfs_ino(inode);
2804 log = root->log_root;
2806 path = btrfs_alloc_path();
2809 dst_path = btrfs_alloc_path();
2811 btrfs_free_path(path);
2815 min_key.objectid = ino;
2816 min_key.type = BTRFS_INODE_ITEM_KEY;
2819 max_key.objectid = ino;
2821 /* today the code can only do partial logging of directories */
2822 if (!S_ISDIR(inode->i_mode))
2823 inode_only = LOG_INODE_ALL;
2825 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2826 max_key.type = BTRFS_XATTR_ITEM_KEY;
2828 max_key.type = (u8)-1;
2829 max_key.offset = (u64)-1;
2831 ret = btrfs_commit_inode_delayed_items(trans, inode);
2833 btrfs_free_path(path);
2834 btrfs_free_path(dst_path);
2838 mutex_lock(&BTRFS_I(inode)->log_mutex);
2841 * a brute force approach to making sure we get the most uptodate
2842 * copies of everything.
2844 if (S_ISDIR(inode->i_mode)) {
2845 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2847 if (inode_only == LOG_INODE_EXISTS)
2848 max_key_type = BTRFS_XATTR_ITEM_KEY;
2849 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
2851 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2857 path->keep_locks = 1;
2861 ret = btrfs_search_forward(root, &min_key, &max_key,
2862 path, 0, trans->transid);
2866 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2867 if (min_key.objectid != ino)
2869 if (min_key.type > max_key.type)
2872 src = path->nodes[0];
2873 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2876 } else if (!ins_nr) {
2877 ins_start_slot = path->slots[0];
2882 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2883 ins_nr, inode_only);
2889 ins_start_slot = path->slots[0];
2892 nritems = btrfs_header_nritems(path->nodes[0]);
2894 if (path->slots[0] < nritems) {
2895 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2900 ret = copy_items(trans, log, dst_path, src,
2902 ins_nr, inode_only);
2909 btrfs_release_path(path);
2911 if (min_key.offset < (u64)-1)
2913 else if (min_key.type < (u8)-1)
2915 else if (min_key.objectid < (u64)-1)
2921 ret = copy_items(trans, log, dst_path, src,
2923 ins_nr, inode_only);
2931 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2932 btrfs_release_path(path);
2933 btrfs_release_path(dst_path);
2934 ret = log_directory_changes(trans, root, inode, path, dst_path);
2940 BTRFS_I(inode)->logged_trans = trans->transid;
2942 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2944 btrfs_free_path(path);
2945 btrfs_free_path(dst_path);
2950 * follow the dentry parent pointers up the chain and see if any
2951 * of the directories in it require a full commit before they can
2952 * be logged. Returns zero if nothing special needs to be done or 1 if
2953 * a full commit is required.
2955 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2956 struct inode *inode,
2957 struct dentry *parent,
2958 struct super_block *sb,
2962 struct btrfs_root *root;
2963 struct dentry *old_parent = NULL;
2966 * for regular files, if its inode is already on disk, we don't
2967 * have to worry about the parents at all. This is because
2968 * we can use the last_unlink_trans field to record renames
2969 * and other fun in this file.
2971 if (S_ISREG(inode->i_mode) &&
2972 BTRFS_I(inode)->generation <= last_committed &&
2973 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2976 if (!S_ISDIR(inode->i_mode)) {
2977 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2979 inode = parent->d_inode;
2983 BTRFS_I(inode)->logged_trans = trans->transid;
2986 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2987 root = BTRFS_I(inode)->root;
2990 * make sure any commits to the log are forced
2991 * to be full commits
2993 root->fs_info->last_trans_log_full_commit =
2999 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3002 if (IS_ROOT(parent))
3005 parent = dget_parent(parent);
3007 old_parent = parent;
3008 inode = parent->d_inode;
3016 static int inode_in_log(struct btrfs_trans_handle *trans,
3017 struct inode *inode)
3019 struct btrfs_root *root = BTRFS_I(inode)->root;
3022 mutex_lock(&root->log_mutex);
3023 if (BTRFS_I(inode)->logged_trans == trans->transid &&
3024 BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
3026 mutex_unlock(&root->log_mutex);
3032 * helper function around btrfs_log_inode to make sure newly created
3033 * parent directories also end up in the log. A minimal inode and backref
3034 * only logging is done of any parent directories that are older than
3035 * the last committed transaction
3037 int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
3038 struct btrfs_root *root, struct inode *inode,
3039 struct dentry *parent, int exists_only)
3041 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
3042 struct super_block *sb;
3043 struct dentry *old_parent = NULL;
3045 u64 last_committed = root->fs_info->last_trans_committed;
3049 if (btrfs_test_opt(root, NOTREELOG)) {
3054 if (root->fs_info->last_trans_log_full_commit >
3055 root->fs_info->last_trans_committed) {
3060 if (root != BTRFS_I(inode)->root ||
3061 btrfs_root_refs(&root->root_item) == 0) {
3066 ret = check_parent_dirs_for_sync(trans, inode, parent,
3067 sb, last_committed);
3071 if (inode_in_log(trans, inode)) {
3072 ret = BTRFS_NO_LOG_SYNC;
3076 ret = start_log_trans(trans, root);
3080 ret = btrfs_log_inode(trans, root, inode, inode_only);
3085 * for regular files, if its inode is already on disk, we don't
3086 * have to worry about the parents at all. This is because
3087 * we can use the last_unlink_trans field to record renames
3088 * and other fun in this file.
3090 if (S_ISREG(inode->i_mode) &&
3091 BTRFS_I(inode)->generation <= last_committed &&
3092 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3097 inode_only = LOG_INODE_EXISTS;
3099 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3102 inode = parent->d_inode;
3103 if (root != BTRFS_I(inode)->root)
3106 if (BTRFS_I(inode)->generation >
3107 root->fs_info->last_trans_committed) {
3108 ret = btrfs_log_inode(trans, root, inode, inode_only);
3112 if (IS_ROOT(parent))
3115 parent = dget_parent(parent);
3117 old_parent = parent;
3123 BUG_ON(ret != -ENOSPC);
3124 root->fs_info->last_trans_log_full_commit = trans->transid;
3127 btrfs_end_log_trans(root);
3133 * it is not safe to log dentry if the chunk root has added new
3134 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3135 * If this returns 1, you must commit the transaction to safely get your
3138 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3139 struct btrfs_root *root, struct dentry *dentry)
3141 struct dentry *parent = dget_parent(dentry);
3144 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
3151 * should be called during mount to recover any replay any log trees
3154 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3157 struct btrfs_path *path;
3158 struct btrfs_trans_handle *trans;
3159 struct btrfs_key key;
3160 struct btrfs_key found_key;
3161 struct btrfs_key tmp_key;
3162 struct btrfs_root *log;
3163 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3164 struct walk_control wc = {
3165 .process_func = process_one_buffer,
3169 path = btrfs_alloc_path();
3173 fs_info->log_root_recovering = 1;
3175 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3176 BUG_ON(IS_ERR(trans));
3181 ret = walk_log_tree(trans, log_root_tree, &wc);
3185 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3186 key.offset = (u64)-1;
3187 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3190 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3194 if (path->slots[0] == 0)
3198 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3200 btrfs_release_path(path);
3201 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3204 log = btrfs_read_fs_root_no_radix(log_root_tree,
3206 BUG_ON(IS_ERR(log));
3208 tmp_key.objectid = found_key.offset;
3209 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3210 tmp_key.offset = (u64)-1;
3212 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
3213 BUG_ON(IS_ERR_OR_NULL(wc.replay_dest));
3215 wc.replay_dest->log_root = log;
3216 btrfs_record_root_in_trans(trans, wc.replay_dest);
3217 ret = walk_log_tree(trans, log, &wc);
3220 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3221 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3226 key.offset = found_key.offset - 1;
3227 wc.replay_dest->log_root = NULL;
3228 free_extent_buffer(log->node);
3229 free_extent_buffer(log->commit_root);
3232 if (found_key.offset == 0)
3235 btrfs_release_path(path);
3237 /* step one is to pin it all, step two is to replay just inodes */
3240 wc.process_func = replay_one_buffer;
3241 wc.stage = LOG_WALK_REPLAY_INODES;
3244 /* step three is to replay everything */
3245 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3250 btrfs_free_path(path);
3252 free_extent_buffer(log_root_tree->node);
3253 log_root_tree->log_root = NULL;
3254 fs_info->log_root_recovering = 0;
3256 /* step 4: commit the transaction, which also unpins the blocks */
3257 btrfs_commit_transaction(trans, fs_info->tree_root);
3259 kfree(log_root_tree);
3264 * there are some corner cases where we want to force a full
3265 * commit instead of allowing a directory to be logged.
3267 * They revolve around files there were unlinked from the directory, and
3268 * this function updates the parent directory so that a full commit is
3269 * properly done if it is fsync'd later after the unlinks are done.
3271 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3272 struct inode *dir, struct inode *inode,
3276 * when we're logging a file, if it hasn't been renamed
3277 * or unlinked, and its inode is fully committed on disk,
3278 * we don't have to worry about walking up the directory chain
3279 * to log its parents.
3281 * So, we use the last_unlink_trans field to put this transid
3282 * into the file. When the file is logged we check it and
3283 * don't log the parents if the file is fully on disk.
3285 if (S_ISREG(inode->i_mode))
3286 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3289 * if this directory was already logged any new
3290 * names for this file/dir will get recorded
3293 if (BTRFS_I(dir)->logged_trans == trans->transid)
3297 * if the inode we're about to unlink was logged,
3298 * the log will be properly updated for any new names
3300 if (BTRFS_I(inode)->logged_trans == trans->transid)
3304 * when renaming files across directories, if the directory
3305 * there we're unlinking from gets fsync'd later on, there's
3306 * no way to find the destination directory later and fsync it
3307 * properly. So, we have to be conservative and force commits
3308 * so the new name gets discovered.
3313 /* we can safely do the unlink without any special recording */
3317 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3321 * Call this after adding a new name for a file and it will properly
3322 * update the log to reflect the new name.
3324 * It will return zero if all goes well, and it will return 1 if a
3325 * full transaction commit is required.
3327 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3328 struct inode *inode, struct inode *old_dir,
3329 struct dentry *parent)
3331 struct btrfs_root * root = BTRFS_I(inode)->root;
3334 * this will force the logging code to walk the dentry chain
3337 if (S_ISREG(inode->i_mode))
3338 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3341 * if this inode hasn't been logged and directory we're renaming it
3342 * from hasn't been logged, we don't need to log it
3344 if (BTRFS_I(inode)->logged_trans <=
3345 root->fs_info->last_trans_committed &&
3346 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3347 root->fs_info->last_trans_committed))
3350 return btrfs_log_inode_parent(trans, root, inode, parent, 1);