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(log->fs_info->extent_root,
280 eb->start, eb->len, 0);
282 if (btrfs_buffer_uptodate(eb, gen)) {
284 btrfs_write_tree_block(eb);
286 btrfs_wait_tree_block_writeback(eb);
292 * Item overwrite used by replay and tree logging. eb, slot and key all refer
293 * to the src data we are copying out.
295 * root is the tree we are copying into, and path is a scratch
296 * path for use in this function (it should be released on entry and
297 * will be released on exit).
299 * If the key is already in the destination tree the existing item is
300 * overwritten. If the existing item isn't big enough, it is extended.
301 * If it is too large, it is truncated.
303 * If the key isn't in the destination yet, a new item is inserted.
305 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
306 struct btrfs_root *root,
307 struct btrfs_path *path,
308 struct extent_buffer *eb, int slot,
309 struct btrfs_key *key)
313 u64 saved_i_size = 0;
314 int save_old_i_size = 0;
315 unsigned long src_ptr;
316 unsigned long dst_ptr;
317 int overwrite_root = 0;
319 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
322 item_size = btrfs_item_size_nr(eb, slot);
323 src_ptr = btrfs_item_ptr_offset(eb, slot);
325 /* look for the key in the destination tree */
326 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
330 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
332 if (dst_size != item_size)
335 if (item_size == 0) {
336 btrfs_release_path(path);
339 dst_copy = kmalloc(item_size, GFP_NOFS);
340 src_copy = kmalloc(item_size, GFP_NOFS);
341 if (!dst_copy || !src_copy) {
342 btrfs_release_path(path);
348 read_extent_buffer(eb, src_copy, src_ptr, item_size);
350 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
351 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
353 ret = memcmp(dst_copy, src_copy, item_size);
358 * they have the same contents, just return, this saves
359 * us from cowing blocks in the destination tree and doing
360 * extra writes that may not have been done by a previous
364 btrfs_release_path(path);
370 btrfs_release_path(path);
371 /* try to insert the key into the destination tree */
372 ret = btrfs_insert_empty_item(trans, root, path,
375 /* make sure any existing item is the correct size */
376 if (ret == -EEXIST) {
378 found_size = btrfs_item_size_nr(path->nodes[0],
380 if (found_size > item_size) {
381 btrfs_truncate_item(trans, root, path, item_size, 1);
382 } else if (found_size < item_size) {
383 ret = btrfs_extend_item(trans, root, path,
384 item_size - found_size);
389 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
392 /* don't overwrite an existing inode if the generation number
393 * was logged as zero. This is done when the tree logging code
394 * is just logging an inode to make sure it exists after recovery.
396 * Also, don't overwrite i_size on directories during replay.
397 * log replay inserts and removes directory items based on the
398 * state of the tree found in the subvolume, and i_size is modified
401 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
402 struct btrfs_inode_item *src_item;
403 struct btrfs_inode_item *dst_item;
405 src_item = (struct btrfs_inode_item *)src_ptr;
406 dst_item = (struct btrfs_inode_item *)dst_ptr;
408 if (btrfs_inode_generation(eb, src_item) == 0)
411 if (overwrite_root &&
412 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
413 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
415 saved_i_size = btrfs_inode_size(path->nodes[0],
420 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
423 if (save_old_i_size) {
424 struct btrfs_inode_item *dst_item;
425 dst_item = (struct btrfs_inode_item *)dst_ptr;
426 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
429 /* make sure the generation is filled in */
430 if (key->type == BTRFS_INODE_ITEM_KEY) {
431 struct btrfs_inode_item *dst_item;
432 dst_item = (struct btrfs_inode_item *)dst_ptr;
433 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
434 btrfs_set_inode_generation(path->nodes[0], dst_item,
439 btrfs_mark_buffer_dirty(path->nodes[0]);
440 btrfs_release_path(path);
445 * simple helper to read an inode off the disk from a given root
446 * This can only be called for subvolume roots and not for the log
448 static noinline struct inode *read_one_inode(struct btrfs_root *root,
451 struct btrfs_key key;
454 key.objectid = objectid;
455 key.type = BTRFS_INODE_ITEM_KEY;
457 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
460 } else if (is_bad_inode(inode)) {
467 /* replays a single extent in 'eb' at 'slot' with 'key' into the
468 * subvolume 'root'. path is released on entry and should be released
471 * extents in the log tree have not been allocated out of the extent
472 * tree yet. So, this completes the allocation, taking a reference
473 * as required if the extent already exists or creating a new extent
474 * if it isn't in the extent allocation tree yet.
476 * The extent is inserted into the file, dropping any existing extents
477 * from the file that overlap the new one.
479 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
480 struct btrfs_root *root,
481 struct btrfs_path *path,
482 struct extent_buffer *eb, int slot,
483 struct btrfs_key *key)
486 u64 mask = root->sectorsize - 1;
489 u64 start = key->offset;
491 struct btrfs_file_extent_item *item;
492 struct inode *inode = NULL;
496 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
497 found_type = btrfs_file_extent_type(eb, item);
499 if (found_type == BTRFS_FILE_EXTENT_REG ||
500 found_type == BTRFS_FILE_EXTENT_PREALLOC)
501 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
502 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
503 size = btrfs_file_extent_inline_len(eb, item);
504 extent_end = (start + size + mask) & ~mask;
510 inode = read_one_inode(root, key->objectid);
517 * first check to see if we already have this extent in the
518 * file. This must be done before the btrfs_drop_extents run
519 * so we don't try to drop this extent.
521 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
525 (found_type == BTRFS_FILE_EXTENT_REG ||
526 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
527 struct btrfs_file_extent_item cmp1;
528 struct btrfs_file_extent_item cmp2;
529 struct btrfs_file_extent_item *existing;
530 struct extent_buffer *leaf;
532 leaf = path->nodes[0];
533 existing = btrfs_item_ptr(leaf, path->slots[0],
534 struct btrfs_file_extent_item);
536 read_extent_buffer(eb, &cmp1, (unsigned long)item,
538 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
542 * we already have a pointer to this exact extent,
543 * we don't have to do anything
545 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
546 btrfs_release_path(path);
550 btrfs_release_path(path);
552 saved_nbytes = inode_get_bytes(inode);
553 /* drop any overlapping extents */
554 ret = btrfs_drop_extents(trans, inode, start, extent_end,
558 if (found_type == BTRFS_FILE_EXTENT_REG ||
559 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
561 unsigned long dest_offset;
562 struct btrfs_key ins;
564 ret = btrfs_insert_empty_item(trans, root, path, key,
567 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
569 copy_extent_buffer(path->nodes[0], eb, dest_offset,
570 (unsigned long)item, sizeof(*item));
572 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
573 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
574 ins.type = BTRFS_EXTENT_ITEM_KEY;
575 offset = key->offset - btrfs_file_extent_offset(eb, item);
577 if (ins.objectid > 0) {
580 LIST_HEAD(ordered_sums);
582 * is this extent already allocated in the extent
583 * allocation tree? If so, just add a reference
585 ret = btrfs_lookup_extent(root, ins.objectid,
588 ret = btrfs_inc_extent_ref(trans, root,
589 ins.objectid, ins.offset,
590 0, root->root_key.objectid,
591 key->objectid, offset);
595 * insert the extent pointer in the extent
598 ret = btrfs_alloc_logged_file_extent(trans,
599 root, root->root_key.objectid,
600 key->objectid, offset, &ins);
603 btrfs_release_path(path);
605 if (btrfs_file_extent_compression(eb, item)) {
606 csum_start = ins.objectid;
607 csum_end = csum_start + ins.offset;
609 csum_start = ins.objectid +
610 btrfs_file_extent_offset(eb, item);
611 csum_end = csum_start +
612 btrfs_file_extent_num_bytes(eb, item);
615 ret = btrfs_lookup_csums_range(root->log_root,
616 csum_start, csum_end - 1,
619 while (!list_empty(&ordered_sums)) {
620 struct btrfs_ordered_sum *sums;
621 sums = list_entry(ordered_sums.next,
622 struct btrfs_ordered_sum,
624 ret = btrfs_csum_file_blocks(trans,
625 root->fs_info->csum_root,
628 list_del(&sums->list);
632 btrfs_release_path(path);
634 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
635 /* inline extents are easy, we just overwrite them */
636 ret = overwrite_item(trans, root, path, eb, slot, key);
640 inode_set_bytes(inode, saved_nbytes);
641 btrfs_update_inode(trans, root, inode);
649 * when cleaning up conflicts between the directory names in the
650 * subvolume, directory names in the log and directory names in the
651 * inode back references, we may have to unlink inodes from directories.
653 * This is a helper function to do the unlink of a specific directory
656 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
657 struct btrfs_root *root,
658 struct btrfs_path *path,
660 struct btrfs_dir_item *di)
665 struct extent_buffer *leaf;
666 struct btrfs_key location;
669 leaf = path->nodes[0];
671 btrfs_dir_item_key_to_cpu(leaf, di, &location);
672 name_len = btrfs_dir_name_len(leaf, di);
673 name = kmalloc(name_len, GFP_NOFS);
677 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
678 btrfs_release_path(path);
680 inode = read_one_inode(root, location.objectid);
686 ret = link_to_fixup_dir(trans, root, path, location.objectid);
689 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
698 * helper function to see if a given name and sequence number found
699 * in an inode back reference are already in a directory and correctly
700 * point to this inode
702 static noinline int inode_in_dir(struct btrfs_root *root,
703 struct btrfs_path *path,
704 u64 dirid, u64 objectid, u64 index,
705 const char *name, int name_len)
707 struct btrfs_dir_item *di;
708 struct btrfs_key location;
711 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
712 index, name, name_len, 0);
713 if (di && !IS_ERR(di)) {
714 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
715 if (location.objectid != objectid)
719 btrfs_release_path(path);
721 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
722 if (di && !IS_ERR(di)) {
723 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
724 if (location.objectid != objectid)
730 btrfs_release_path(path);
735 * helper function to check a log tree for a named back reference in
736 * an inode. This is used to decide if a back reference that is
737 * found in the subvolume conflicts with what we find in the log.
739 * inode backreferences may have multiple refs in a single item,
740 * during replay we process one reference at a time, and we don't
741 * want to delete valid links to a file from the subvolume if that
742 * link is also in the log.
744 static noinline int backref_in_log(struct btrfs_root *log,
745 struct btrfs_key *key,
746 char *name, int namelen)
748 struct btrfs_path *path;
749 struct btrfs_inode_ref *ref;
751 unsigned long ptr_end;
752 unsigned long name_ptr;
758 path = btrfs_alloc_path();
762 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
766 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
767 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
768 ptr_end = ptr + item_size;
769 while (ptr < ptr_end) {
770 ref = (struct btrfs_inode_ref *)ptr;
771 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
772 if (found_name_len == namelen) {
773 name_ptr = (unsigned long)(ref + 1);
774 ret = memcmp_extent_buffer(path->nodes[0], name,
781 ptr = (unsigned long)(ref + 1) + found_name_len;
784 btrfs_free_path(path);
790 * replay one inode back reference item found in the log tree.
791 * eb, slot and key refer to the buffer and key found in the log tree.
792 * root is the destination we are replaying into, and path is for temp
793 * use by this function. (it should be released on return).
795 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
796 struct btrfs_root *root,
797 struct btrfs_root *log,
798 struct btrfs_path *path,
799 struct extent_buffer *eb, int slot,
800 struct btrfs_key *key)
802 struct btrfs_inode_ref *ref;
803 struct btrfs_dir_item *di;
806 unsigned long ref_ptr;
807 unsigned long ref_end;
814 * it is possible that we didn't log all the parent directories
815 * for a given inode. If we don't find the dir, just don't
816 * copy the back ref in. The link count fixup code will take
819 dir = read_one_inode(root, key->offset);
823 inode = read_one_inode(root, key->objectid);
829 ref_ptr = btrfs_item_ptr_offset(eb, slot);
830 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
833 ref = (struct btrfs_inode_ref *)ref_ptr;
835 namelen = btrfs_inode_ref_name_len(eb, ref);
836 name = kmalloc(namelen, GFP_NOFS);
839 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
841 /* if we already have a perfect match, we're done */
842 if (inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
843 btrfs_inode_ref_index(eb, ref),
849 * look for a conflicting back reference in the metadata.
850 * if we find one we have to unlink that name of the file
851 * before we add our new link. Later on, we overwrite any
852 * existing back reference, and we don't want to create
853 * dangling pointers in the directory.
859 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
863 struct btrfs_inode_ref *victim_ref;
865 unsigned long ptr_end;
866 struct extent_buffer *leaf = path->nodes[0];
868 /* are we trying to overwrite a back ref for the root directory
869 * if so, just jump out, we're done
871 if (key->objectid == key->offset)
874 /* check all the names in this back reference to see
875 * if they are in the log. if so, we allow them to stay
876 * otherwise they must be unlinked as a conflict
878 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
879 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
880 while (ptr < ptr_end) {
881 victim_ref = (struct btrfs_inode_ref *)ptr;
882 victim_name_len = btrfs_inode_ref_name_len(leaf,
884 victim_name = kmalloc(victim_name_len, GFP_NOFS);
885 BUG_ON(!victim_name);
887 read_extent_buffer(leaf, victim_name,
888 (unsigned long)(victim_ref + 1),
891 if (!backref_in_log(log, key, victim_name,
893 btrfs_inc_nlink(inode);
894 btrfs_release_path(path);
896 ret = btrfs_unlink_inode(trans, root, dir,
901 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
906 * NOTE: we have searched root tree and checked the
907 * coresponding ref, it does not need to check again.
911 btrfs_release_path(path);
913 /* look for a conflicting sequence number */
914 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
915 btrfs_inode_ref_index(eb, ref),
917 if (di && !IS_ERR(di)) {
918 ret = drop_one_dir_item(trans, root, path, dir, di);
921 btrfs_release_path(path);
923 /* look for a conflicing name */
924 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
926 if (di && !IS_ERR(di)) {
927 ret = drop_one_dir_item(trans, root, path, dir, di);
930 btrfs_release_path(path);
933 /* insert our name */
934 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
935 btrfs_inode_ref_index(eb, ref));
938 btrfs_update_inode(trans, root, inode);
941 ref_ptr = (unsigned long)(ref + 1) + namelen;
943 if (ref_ptr < ref_end)
946 /* finally write the back reference in the inode */
947 ret = overwrite_item(trans, root, path, eb, slot, key);
951 btrfs_release_path(path);
957 static int insert_orphan_item(struct btrfs_trans_handle *trans,
958 struct btrfs_root *root, u64 offset)
961 ret = btrfs_find_orphan_item(root, offset);
963 ret = btrfs_insert_orphan_item(trans, root, offset);
969 * There are a few corners where the link count of the file can't
970 * be properly maintained during replay. So, instead of adding
971 * lots of complexity to the log code, we just scan the backrefs
972 * for any file that has been through replay.
974 * The scan will update the link count on the inode to reflect the
975 * number of back refs found. If it goes down to zero, the iput
976 * will free the inode.
978 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
979 struct btrfs_root *root,
982 struct btrfs_path *path;
984 struct btrfs_key key;
987 unsigned long ptr_end;
989 u64 ino = btrfs_ino(inode);
992 key.type = BTRFS_INODE_REF_KEY;
993 key.offset = (u64)-1;
995 path = btrfs_alloc_path();
1000 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1004 if (path->slots[0] == 0)
1008 btrfs_item_key_to_cpu(path->nodes[0], &key,
1010 if (key.objectid != ino ||
1011 key.type != BTRFS_INODE_REF_KEY)
1013 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1014 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1016 while (ptr < ptr_end) {
1017 struct btrfs_inode_ref *ref;
1019 ref = (struct btrfs_inode_ref *)ptr;
1020 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1022 ptr = (unsigned long)(ref + 1) + name_len;
1026 if (key.offset == 0)
1029 btrfs_release_path(path);
1031 btrfs_release_path(path);
1032 if (nlink != inode->i_nlink) {
1033 set_nlink(inode, nlink);
1034 btrfs_update_inode(trans, root, inode);
1036 BTRFS_I(inode)->index_cnt = (u64)-1;
1038 if (inode->i_nlink == 0) {
1039 if (S_ISDIR(inode->i_mode)) {
1040 ret = replay_dir_deletes(trans, root, NULL, path,
1044 ret = insert_orphan_item(trans, root, ino);
1047 btrfs_free_path(path);
1052 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1053 struct btrfs_root *root,
1054 struct btrfs_path *path)
1057 struct btrfs_key key;
1058 struct inode *inode;
1060 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1061 key.type = BTRFS_ORPHAN_ITEM_KEY;
1062 key.offset = (u64)-1;
1064 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1069 if (path->slots[0] == 0)
1074 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1075 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1076 key.type != BTRFS_ORPHAN_ITEM_KEY)
1079 ret = btrfs_del_item(trans, root, path);
1083 btrfs_release_path(path);
1084 inode = read_one_inode(root, key.offset);
1088 ret = fixup_inode_link_count(trans, root, inode);
1094 * fixup on a directory may create new entries,
1095 * make sure we always look for the highset possible
1098 key.offset = (u64)-1;
1102 btrfs_release_path(path);
1108 * record a given inode in the fixup dir so we can check its link
1109 * count when replay is done. The link count is incremented here
1110 * so the inode won't go away until we check it
1112 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1113 struct btrfs_root *root,
1114 struct btrfs_path *path,
1117 struct btrfs_key key;
1119 struct inode *inode;
1121 inode = read_one_inode(root, objectid);
1125 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1126 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1127 key.offset = objectid;
1129 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1131 btrfs_release_path(path);
1133 btrfs_inc_nlink(inode);
1134 btrfs_update_inode(trans, root, inode);
1135 } else if (ret == -EEXIST) {
1146 * when replaying the log for a directory, we only insert names
1147 * for inodes that actually exist. This means an fsync on a directory
1148 * does not implicitly fsync all the new files in it
1150 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1151 struct btrfs_root *root,
1152 struct btrfs_path *path,
1153 u64 dirid, u64 index,
1154 char *name, int name_len, u8 type,
1155 struct btrfs_key *location)
1157 struct inode *inode;
1161 inode = read_one_inode(root, location->objectid);
1165 dir = read_one_inode(root, dirid);
1170 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1172 /* FIXME, put inode into FIXUP list */
1180 * take a single entry in a log directory item and replay it into
1183 * if a conflicting item exists in the subdirectory already,
1184 * the inode it points to is unlinked and put into the link count
1187 * If a name from the log points to a file or directory that does
1188 * not exist in the FS, it is skipped. fsyncs on directories
1189 * do not force down inodes inside that directory, just changes to the
1190 * names or unlinks in a directory.
1192 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1193 struct btrfs_root *root,
1194 struct btrfs_path *path,
1195 struct extent_buffer *eb,
1196 struct btrfs_dir_item *di,
1197 struct btrfs_key *key)
1201 struct btrfs_dir_item *dst_di;
1202 struct btrfs_key found_key;
1203 struct btrfs_key log_key;
1209 dir = read_one_inode(root, key->objectid);
1213 name_len = btrfs_dir_name_len(eb, di);
1214 name = kmalloc(name_len, GFP_NOFS);
1218 log_type = btrfs_dir_type(eb, di);
1219 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1222 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1223 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1228 btrfs_release_path(path);
1230 if (key->type == BTRFS_DIR_ITEM_KEY) {
1231 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1233 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1234 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1241 if (IS_ERR_OR_NULL(dst_di)) {
1242 /* we need a sequence number to insert, so we only
1243 * do inserts for the BTRFS_DIR_INDEX_KEY types
1245 if (key->type != BTRFS_DIR_INDEX_KEY)
1250 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1251 /* the existing item matches the logged item */
1252 if (found_key.objectid == log_key.objectid &&
1253 found_key.type == log_key.type &&
1254 found_key.offset == log_key.offset &&
1255 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1260 * don't drop the conflicting directory entry if the inode
1261 * for the new entry doesn't exist
1266 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1269 if (key->type == BTRFS_DIR_INDEX_KEY)
1272 btrfs_release_path(path);
1278 btrfs_release_path(path);
1279 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1280 name, name_len, log_type, &log_key);
1282 BUG_ON(ret && ret != -ENOENT);
1287 * find all the names in a directory item and reconcile them into
1288 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1289 * one name in a directory item, but the same code gets used for
1290 * both directory index types
1292 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1293 struct btrfs_root *root,
1294 struct btrfs_path *path,
1295 struct extent_buffer *eb, int slot,
1296 struct btrfs_key *key)
1299 u32 item_size = btrfs_item_size_nr(eb, slot);
1300 struct btrfs_dir_item *di;
1303 unsigned long ptr_end;
1305 ptr = btrfs_item_ptr_offset(eb, slot);
1306 ptr_end = ptr + item_size;
1307 while (ptr < ptr_end) {
1308 di = (struct btrfs_dir_item *)ptr;
1309 if (verify_dir_item(root, eb, di))
1311 name_len = btrfs_dir_name_len(eb, di);
1312 ret = replay_one_name(trans, root, path, eb, di, key);
1314 ptr = (unsigned long)(di + 1);
1321 * directory replay has two parts. There are the standard directory
1322 * items in the log copied from the subvolume, and range items
1323 * created in the log while the subvolume was logged.
1325 * The range items tell us which parts of the key space the log
1326 * is authoritative for. During replay, if a key in the subvolume
1327 * directory is in a logged range item, but not actually in the log
1328 * that means it was deleted from the directory before the fsync
1329 * and should be removed.
1331 static noinline int find_dir_range(struct btrfs_root *root,
1332 struct btrfs_path *path,
1333 u64 dirid, int key_type,
1334 u64 *start_ret, u64 *end_ret)
1336 struct btrfs_key key;
1338 struct btrfs_dir_log_item *item;
1342 if (*start_ret == (u64)-1)
1345 key.objectid = dirid;
1346 key.type = key_type;
1347 key.offset = *start_ret;
1349 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1353 if (path->slots[0] == 0)
1358 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1360 if (key.type != key_type || key.objectid != dirid) {
1364 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1365 struct btrfs_dir_log_item);
1366 found_end = btrfs_dir_log_end(path->nodes[0], item);
1368 if (*start_ret >= key.offset && *start_ret <= found_end) {
1370 *start_ret = key.offset;
1371 *end_ret = found_end;
1376 /* check the next slot in the tree to see if it is a valid item */
1377 nritems = btrfs_header_nritems(path->nodes[0]);
1378 if (path->slots[0] >= nritems) {
1379 ret = btrfs_next_leaf(root, path);
1386 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1388 if (key.type != key_type || key.objectid != dirid) {
1392 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1393 struct btrfs_dir_log_item);
1394 found_end = btrfs_dir_log_end(path->nodes[0], item);
1395 *start_ret = key.offset;
1396 *end_ret = found_end;
1399 btrfs_release_path(path);
1404 * this looks for a given directory item in the log. If the directory
1405 * item is not in the log, the item is removed and the inode it points
1408 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1409 struct btrfs_root *root,
1410 struct btrfs_root *log,
1411 struct btrfs_path *path,
1412 struct btrfs_path *log_path,
1414 struct btrfs_key *dir_key)
1417 struct extent_buffer *eb;
1420 struct btrfs_dir_item *di;
1421 struct btrfs_dir_item *log_di;
1424 unsigned long ptr_end;
1426 struct inode *inode;
1427 struct btrfs_key location;
1430 eb = path->nodes[0];
1431 slot = path->slots[0];
1432 item_size = btrfs_item_size_nr(eb, slot);
1433 ptr = btrfs_item_ptr_offset(eb, slot);
1434 ptr_end = ptr + item_size;
1435 while (ptr < ptr_end) {
1436 di = (struct btrfs_dir_item *)ptr;
1437 if (verify_dir_item(root, eb, di)) {
1442 name_len = btrfs_dir_name_len(eb, di);
1443 name = kmalloc(name_len, GFP_NOFS);
1448 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1451 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1452 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1455 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1456 log_di = btrfs_lookup_dir_index_item(trans, log,
1462 if (IS_ERR_OR_NULL(log_di)) {
1463 btrfs_dir_item_key_to_cpu(eb, di, &location);
1464 btrfs_release_path(path);
1465 btrfs_release_path(log_path);
1466 inode = read_one_inode(root, location.objectid);
1472 ret = link_to_fixup_dir(trans, root,
1473 path, location.objectid);
1475 btrfs_inc_nlink(inode);
1476 ret = btrfs_unlink_inode(trans, root, dir, inode,
1482 /* there might still be more names under this key
1483 * check and repeat if required
1485 ret = btrfs_search_slot(NULL, root, dir_key, path,
1492 btrfs_release_path(log_path);
1495 ptr = (unsigned long)(di + 1);
1500 btrfs_release_path(path);
1501 btrfs_release_path(log_path);
1506 * deletion replay happens before we copy any new directory items
1507 * out of the log or out of backreferences from inodes. It
1508 * scans the log to find ranges of keys that log is authoritative for,
1509 * and then scans the directory to find items in those ranges that are
1510 * not present in the log.
1512 * Anything we don't find in the log is unlinked and removed from the
1515 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1516 struct btrfs_root *root,
1517 struct btrfs_root *log,
1518 struct btrfs_path *path,
1519 u64 dirid, int del_all)
1523 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1525 struct btrfs_key dir_key;
1526 struct btrfs_key found_key;
1527 struct btrfs_path *log_path;
1530 dir_key.objectid = dirid;
1531 dir_key.type = BTRFS_DIR_ITEM_KEY;
1532 log_path = btrfs_alloc_path();
1536 dir = read_one_inode(root, dirid);
1537 /* it isn't an error if the inode isn't there, that can happen
1538 * because we replay the deletes before we copy in the inode item
1542 btrfs_free_path(log_path);
1550 range_end = (u64)-1;
1552 ret = find_dir_range(log, path, dirid, key_type,
1553 &range_start, &range_end);
1558 dir_key.offset = range_start;
1561 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1566 nritems = btrfs_header_nritems(path->nodes[0]);
1567 if (path->slots[0] >= nritems) {
1568 ret = btrfs_next_leaf(root, path);
1572 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1574 if (found_key.objectid != dirid ||
1575 found_key.type != dir_key.type)
1578 if (found_key.offset > range_end)
1581 ret = check_item_in_log(trans, root, log, path,
1585 if (found_key.offset == (u64)-1)
1587 dir_key.offset = found_key.offset + 1;
1589 btrfs_release_path(path);
1590 if (range_end == (u64)-1)
1592 range_start = range_end + 1;
1597 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1598 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1599 dir_key.type = BTRFS_DIR_INDEX_KEY;
1600 btrfs_release_path(path);
1604 btrfs_release_path(path);
1605 btrfs_free_path(log_path);
1611 * the process_func used to replay items from the log tree. This
1612 * gets called in two different stages. The first stage just looks
1613 * for inodes and makes sure they are all copied into the subvolume.
1615 * The second stage copies all the other item types from the log into
1616 * the subvolume. The two stage approach is slower, but gets rid of
1617 * lots of complexity around inodes referencing other inodes that exist
1618 * only in the log (references come from either directory items or inode
1621 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1622 struct walk_control *wc, u64 gen)
1625 struct btrfs_path *path;
1626 struct btrfs_root *root = wc->replay_dest;
1627 struct btrfs_key key;
1632 btrfs_read_buffer(eb, gen);
1634 level = btrfs_header_level(eb);
1639 path = btrfs_alloc_path();
1643 nritems = btrfs_header_nritems(eb);
1644 for (i = 0; i < nritems; i++) {
1645 btrfs_item_key_to_cpu(eb, &key, i);
1647 /* inode keys are done during the first stage */
1648 if (key.type == BTRFS_INODE_ITEM_KEY &&
1649 wc->stage == LOG_WALK_REPLAY_INODES) {
1650 struct btrfs_inode_item *inode_item;
1653 inode_item = btrfs_item_ptr(eb, i,
1654 struct btrfs_inode_item);
1655 mode = btrfs_inode_mode(eb, inode_item);
1656 if (S_ISDIR(mode)) {
1657 ret = replay_dir_deletes(wc->trans,
1658 root, log, path, key.objectid, 0);
1661 ret = overwrite_item(wc->trans, root, path,
1665 /* for regular files, make sure corresponding
1666 * orhpan item exist. extents past the new EOF
1667 * will be truncated later by orphan cleanup.
1669 if (S_ISREG(mode)) {
1670 ret = insert_orphan_item(wc->trans, root,
1675 ret = link_to_fixup_dir(wc->trans, root,
1676 path, key.objectid);
1679 if (wc->stage < LOG_WALK_REPLAY_ALL)
1682 /* these keys are simply copied */
1683 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1684 ret = overwrite_item(wc->trans, root, path,
1687 } else if (key.type == BTRFS_INODE_REF_KEY) {
1688 ret = add_inode_ref(wc->trans, root, log, path,
1690 BUG_ON(ret && ret != -ENOENT);
1691 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1692 ret = replay_one_extent(wc->trans, root, path,
1695 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1696 key.type == BTRFS_DIR_INDEX_KEY) {
1697 ret = replay_one_dir_item(wc->trans, root, path,
1702 btrfs_free_path(path);
1706 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
1707 struct btrfs_root *root,
1708 struct btrfs_path *path, int *level,
1709 struct walk_control *wc)
1714 struct extent_buffer *next;
1715 struct extent_buffer *cur;
1716 struct extent_buffer *parent;
1720 WARN_ON(*level < 0);
1721 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1723 while (*level > 0) {
1724 WARN_ON(*level < 0);
1725 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1726 cur = path->nodes[*level];
1728 if (btrfs_header_level(cur) != *level)
1731 if (path->slots[*level] >=
1732 btrfs_header_nritems(cur))
1735 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1736 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1737 blocksize = btrfs_level_size(root, *level - 1);
1739 parent = path->nodes[*level];
1740 root_owner = btrfs_header_owner(parent);
1742 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1747 ret = wc->process_func(root, next, wc, ptr_gen);
1751 path->slots[*level]++;
1753 btrfs_read_buffer(next, ptr_gen);
1755 btrfs_tree_lock(next);
1756 btrfs_set_lock_blocking(next);
1757 clean_tree_block(trans, root, next);
1758 btrfs_wait_tree_block_writeback(next);
1759 btrfs_tree_unlock(next);
1761 WARN_ON(root_owner !=
1762 BTRFS_TREE_LOG_OBJECTID);
1763 ret = btrfs_free_reserved_extent(root,
1767 free_extent_buffer(next);
1770 btrfs_read_buffer(next, ptr_gen);
1772 WARN_ON(*level <= 0);
1773 if (path->nodes[*level-1])
1774 free_extent_buffer(path->nodes[*level-1]);
1775 path->nodes[*level-1] = next;
1776 *level = btrfs_header_level(next);
1777 path->slots[*level] = 0;
1780 WARN_ON(*level < 0);
1781 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1783 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1789 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
1790 struct btrfs_root *root,
1791 struct btrfs_path *path, int *level,
1792 struct walk_control *wc)
1799 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1800 slot = path->slots[i];
1801 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
1804 WARN_ON(*level == 0);
1807 struct extent_buffer *parent;
1808 if (path->nodes[*level] == root->node)
1809 parent = path->nodes[*level];
1811 parent = path->nodes[*level + 1];
1813 root_owner = btrfs_header_owner(parent);
1814 ret = wc->process_func(root, path->nodes[*level], wc,
1815 btrfs_header_generation(path->nodes[*level]));
1820 struct extent_buffer *next;
1822 next = path->nodes[*level];
1824 btrfs_tree_lock(next);
1825 btrfs_set_lock_blocking(next);
1826 clean_tree_block(trans, root, next);
1827 btrfs_wait_tree_block_writeback(next);
1828 btrfs_tree_unlock(next);
1830 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1831 ret = btrfs_free_reserved_extent(root,
1832 path->nodes[*level]->start,
1833 path->nodes[*level]->len);
1836 free_extent_buffer(path->nodes[*level]);
1837 path->nodes[*level] = NULL;
1845 * drop the reference count on the tree rooted at 'snap'. This traverses
1846 * the tree freeing any blocks that have a ref count of zero after being
1849 static int walk_log_tree(struct btrfs_trans_handle *trans,
1850 struct btrfs_root *log, struct walk_control *wc)
1855 struct btrfs_path *path;
1859 path = btrfs_alloc_path();
1863 level = btrfs_header_level(log->node);
1865 path->nodes[level] = log->node;
1866 extent_buffer_get(log->node);
1867 path->slots[level] = 0;
1870 wret = walk_down_log_tree(trans, log, path, &level, wc);
1876 wret = walk_up_log_tree(trans, log, path, &level, wc);
1883 /* was the root node processed? if not, catch it here */
1884 if (path->nodes[orig_level]) {
1885 wc->process_func(log, path->nodes[orig_level], wc,
1886 btrfs_header_generation(path->nodes[orig_level]));
1888 struct extent_buffer *next;
1890 next = path->nodes[orig_level];
1892 btrfs_tree_lock(next);
1893 btrfs_set_lock_blocking(next);
1894 clean_tree_block(trans, log, next);
1895 btrfs_wait_tree_block_writeback(next);
1896 btrfs_tree_unlock(next);
1898 WARN_ON(log->root_key.objectid !=
1899 BTRFS_TREE_LOG_OBJECTID);
1900 ret = btrfs_free_reserved_extent(log, next->start,
1906 for (i = 0; i <= orig_level; i++) {
1907 if (path->nodes[i]) {
1908 free_extent_buffer(path->nodes[i]);
1909 path->nodes[i] = NULL;
1912 btrfs_free_path(path);
1917 * helper function to update the item for a given subvolumes log root
1918 * in the tree of log roots
1920 static int update_log_root(struct btrfs_trans_handle *trans,
1921 struct btrfs_root *log)
1925 if (log->log_transid == 1) {
1926 /* insert root item on the first sync */
1927 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1928 &log->root_key, &log->root_item);
1930 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1931 &log->root_key, &log->root_item);
1936 static int wait_log_commit(struct btrfs_trans_handle *trans,
1937 struct btrfs_root *root, unsigned long transid)
1940 int index = transid % 2;
1943 * we only allow two pending log transactions at a time,
1944 * so we know that if ours is more than 2 older than the
1945 * current transaction, we're done
1948 prepare_to_wait(&root->log_commit_wait[index],
1949 &wait, TASK_UNINTERRUPTIBLE);
1950 mutex_unlock(&root->log_mutex);
1952 if (root->fs_info->last_trans_log_full_commit !=
1953 trans->transid && root->log_transid < transid + 2 &&
1954 atomic_read(&root->log_commit[index]))
1957 finish_wait(&root->log_commit_wait[index], &wait);
1958 mutex_lock(&root->log_mutex);
1959 } while (root->log_transid < transid + 2 &&
1960 atomic_read(&root->log_commit[index]));
1964 static int wait_for_writer(struct btrfs_trans_handle *trans,
1965 struct btrfs_root *root)
1968 while (atomic_read(&root->log_writers)) {
1969 prepare_to_wait(&root->log_writer_wait,
1970 &wait, TASK_UNINTERRUPTIBLE);
1971 mutex_unlock(&root->log_mutex);
1972 if (root->fs_info->last_trans_log_full_commit !=
1973 trans->transid && atomic_read(&root->log_writers))
1975 mutex_lock(&root->log_mutex);
1976 finish_wait(&root->log_writer_wait, &wait);
1982 * btrfs_sync_log does sends a given tree log down to the disk and
1983 * updates the super blocks to record it. When this call is done,
1984 * you know that any inodes previously logged are safely on disk only
1987 * Any other return value means you need to call btrfs_commit_transaction.
1988 * Some of the edge cases for fsyncing directories that have had unlinks
1989 * or renames done in the past mean that sometimes the only safe
1990 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1991 * that has happened.
1993 int btrfs_sync_log(struct btrfs_trans_handle *trans,
1994 struct btrfs_root *root)
2000 struct btrfs_root *log = root->log_root;
2001 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2002 unsigned long log_transid = 0;
2004 mutex_lock(&root->log_mutex);
2005 index1 = root->log_transid % 2;
2006 if (atomic_read(&root->log_commit[index1])) {
2007 wait_log_commit(trans, root, root->log_transid);
2008 mutex_unlock(&root->log_mutex);
2011 atomic_set(&root->log_commit[index1], 1);
2013 /* wait for previous tree log sync to complete */
2014 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2015 wait_log_commit(trans, root, root->log_transid - 1);
2018 unsigned long batch = root->log_batch;
2019 if (root->log_multiple_pids) {
2020 mutex_unlock(&root->log_mutex);
2021 schedule_timeout_uninterruptible(1);
2022 mutex_lock(&root->log_mutex);
2024 wait_for_writer(trans, root);
2025 if (batch == root->log_batch)
2029 /* bail out if we need to do a full commit */
2030 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2032 mutex_unlock(&root->log_mutex);
2036 log_transid = root->log_transid;
2037 if (log_transid % 2 == 0)
2038 mark = EXTENT_DIRTY;
2042 /* we start IO on all the marked extents here, but we don't actually
2043 * wait for them until later.
2045 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2048 btrfs_set_root_node(&log->root_item, log->node);
2050 root->log_batch = 0;
2051 root->log_transid++;
2052 log->log_transid = root->log_transid;
2053 root->log_start_pid = 0;
2056 * IO has been started, blocks of the log tree have WRITTEN flag set
2057 * in their headers. new modifications of the log will be written to
2058 * new positions. so it's safe to allow log writers to go in.
2060 mutex_unlock(&root->log_mutex);
2062 mutex_lock(&log_root_tree->log_mutex);
2063 log_root_tree->log_batch++;
2064 atomic_inc(&log_root_tree->log_writers);
2065 mutex_unlock(&log_root_tree->log_mutex);
2067 ret = update_log_root(trans, log);
2069 mutex_lock(&log_root_tree->log_mutex);
2070 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2072 if (waitqueue_active(&log_root_tree->log_writer_wait))
2073 wake_up(&log_root_tree->log_writer_wait);
2077 BUG_ON(ret != -ENOSPC);
2078 root->fs_info->last_trans_log_full_commit = trans->transid;
2079 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2080 mutex_unlock(&log_root_tree->log_mutex);
2085 index2 = log_root_tree->log_transid % 2;
2086 if (atomic_read(&log_root_tree->log_commit[index2])) {
2087 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2088 wait_log_commit(trans, log_root_tree,
2089 log_root_tree->log_transid);
2090 mutex_unlock(&log_root_tree->log_mutex);
2094 atomic_set(&log_root_tree->log_commit[index2], 1);
2096 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2097 wait_log_commit(trans, log_root_tree,
2098 log_root_tree->log_transid - 1);
2101 wait_for_writer(trans, log_root_tree);
2104 * now that we've moved on to the tree of log tree roots,
2105 * check the full commit flag again
2107 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2108 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2109 mutex_unlock(&log_root_tree->log_mutex);
2111 goto out_wake_log_root;
2114 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2115 &log_root_tree->dirty_log_pages,
2116 EXTENT_DIRTY | EXTENT_NEW);
2118 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2120 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
2121 log_root_tree->node->start);
2122 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
2123 btrfs_header_level(log_root_tree->node));
2125 log_root_tree->log_batch = 0;
2126 log_root_tree->log_transid++;
2129 mutex_unlock(&log_root_tree->log_mutex);
2132 * nobody else is going to jump in and write the the ctree
2133 * super here because the log_commit atomic below is protecting
2134 * us. We must be called with a transaction handle pinning
2135 * the running transaction open, so a full commit can't hop
2136 * in and cause problems either.
2138 btrfs_scrub_pause_super(root);
2139 write_ctree_super(trans, root->fs_info->tree_root, 1);
2140 btrfs_scrub_continue_super(root);
2143 mutex_lock(&root->log_mutex);
2144 if (root->last_log_commit < log_transid)
2145 root->last_log_commit = log_transid;
2146 mutex_unlock(&root->log_mutex);
2149 atomic_set(&log_root_tree->log_commit[index2], 0);
2151 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2152 wake_up(&log_root_tree->log_commit_wait[index2]);
2154 atomic_set(&root->log_commit[index1], 0);
2156 if (waitqueue_active(&root->log_commit_wait[index1]))
2157 wake_up(&root->log_commit_wait[index1]);
2161 static void free_log_tree(struct btrfs_trans_handle *trans,
2162 struct btrfs_root *log)
2167 struct walk_control wc = {
2169 .process_func = process_one_buffer
2172 ret = walk_log_tree(trans, log, &wc);
2176 ret = find_first_extent_bit(&log->dirty_log_pages,
2177 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
2181 clear_extent_bits(&log->dirty_log_pages, start, end,
2182 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2185 free_extent_buffer(log->node);
2190 * free all the extents used by the tree log. This should be called
2191 * at commit time of the full transaction
2193 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2195 if (root->log_root) {
2196 free_log_tree(trans, root->log_root);
2197 root->log_root = NULL;
2202 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2203 struct btrfs_fs_info *fs_info)
2205 if (fs_info->log_root_tree) {
2206 free_log_tree(trans, fs_info->log_root_tree);
2207 fs_info->log_root_tree = NULL;
2213 * If both a file and directory are logged, and unlinks or renames are
2214 * mixed in, we have a few interesting corners:
2216 * create file X in dir Y
2217 * link file X to X.link in dir Y
2219 * unlink file X but leave X.link
2222 * After a crash we would expect only X.link to exist. But file X
2223 * didn't get fsync'd again so the log has back refs for X and X.link.
2225 * We solve this by removing directory entries and inode backrefs from the
2226 * log when a file that was logged in the current transaction is
2227 * unlinked. Any later fsync will include the updated log entries, and
2228 * we'll be able to reconstruct the proper directory items from backrefs.
2230 * This optimizations allows us to avoid relogging the entire inode
2231 * or the entire directory.
2233 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2234 struct btrfs_root *root,
2235 const char *name, int name_len,
2236 struct inode *dir, u64 index)
2238 struct btrfs_root *log;
2239 struct btrfs_dir_item *di;
2240 struct btrfs_path *path;
2244 u64 dir_ino = btrfs_ino(dir);
2246 if (BTRFS_I(dir)->logged_trans < trans->transid)
2249 ret = join_running_log_trans(root);
2253 mutex_lock(&BTRFS_I(dir)->log_mutex);
2255 log = root->log_root;
2256 path = btrfs_alloc_path();
2262 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
2263 name, name_len, -1);
2269 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2270 bytes_del += name_len;
2273 btrfs_release_path(path);
2274 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
2275 index, name, name_len, -1);
2281 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2282 bytes_del += name_len;
2286 /* update the directory size in the log to reflect the names
2290 struct btrfs_key key;
2292 key.objectid = dir_ino;
2294 key.type = BTRFS_INODE_ITEM_KEY;
2295 btrfs_release_path(path);
2297 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2303 struct btrfs_inode_item *item;
2306 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2307 struct btrfs_inode_item);
2308 i_size = btrfs_inode_size(path->nodes[0], item);
2309 if (i_size > bytes_del)
2310 i_size -= bytes_del;
2313 btrfs_set_inode_size(path->nodes[0], item, i_size);
2314 btrfs_mark_buffer_dirty(path->nodes[0]);
2317 btrfs_release_path(path);
2320 btrfs_free_path(path);
2322 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2323 if (ret == -ENOSPC) {
2324 root->fs_info->last_trans_log_full_commit = trans->transid;
2327 btrfs_end_log_trans(root);
2332 /* see comments for btrfs_del_dir_entries_in_log */
2333 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2334 struct btrfs_root *root,
2335 const char *name, int name_len,
2336 struct inode *inode, u64 dirid)
2338 struct btrfs_root *log;
2342 if (BTRFS_I(inode)->logged_trans < trans->transid)
2345 ret = join_running_log_trans(root);
2348 log = root->log_root;
2349 mutex_lock(&BTRFS_I(inode)->log_mutex);
2351 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
2353 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2354 if (ret == -ENOSPC) {
2355 root->fs_info->last_trans_log_full_commit = trans->transid;
2358 btrfs_end_log_trans(root);
2364 * creates a range item in the log for 'dirid'. first_offset and
2365 * last_offset tell us which parts of the key space the log should
2366 * be considered authoritative for.
2368 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2369 struct btrfs_root *log,
2370 struct btrfs_path *path,
2371 int key_type, u64 dirid,
2372 u64 first_offset, u64 last_offset)
2375 struct btrfs_key key;
2376 struct btrfs_dir_log_item *item;
2378 key.objectid = dirid;
2379 key.offset = first_offset;
2380 if (key_type == BTRFS_DIR_ITEM_KEY)
2381 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2383 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2384 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2388 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2389 struct btrfs_dir_log_item);
2390 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2391 btrfs_mark_buffer_dirty(path->nodes[0]);
2392 btrfs_release_path(path);
2397 * log all the items included in the current transaction for a given
2398 * directory. This also creates the range items in the log tree required
2399 * to replay anything deleted before the fsync
2401 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2402 struct btrfs_root *root, struct inode *inode,
2403 struct btrfs_path *path,
2404 struct btrfs_path *dst_path, int key_type,
2405 u64 min_offset, u64 *last_offset_ret)
2407 struct btrfs_key min_key;
2408 struct btrfs_key max_key;
2409 struct btrfs_root *log = root->log_root;
2410 struct extent_buffer *src;
2415 u64 first_offset = min_offset;
2416 u64 last_offset = (u64)-1;
2417 u64 ino = btrfs_ino(inode);
2419 log = root->log_root;
2420 max_key.objectid = ino;
2421 max_key.offset = (u64)-1;
2422 max_key.type = key_type;
2424 min_key.objectid = ino;
2425 min_key.type = key_type;
2426 min_key.offset = min_offset;
2428 path->keep_locks = 1;
2430 ret = btrfs_search_forward(root, &min_key, &max_key,
2431 path, 0, trans->transid);
2434 * we didn't find anything from this transaction, see if there
2435 * is anything at all
2437 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2438 min_key.objectid = ino;
2439 min_key.type = key_type;
2440 min_key.offset = (u64)-1;
2441 btrfs_release_path(path);
2442 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2444 btrfs_release_path(path);
2447 ret = btrfs_previous_item(root, path, ino, key_type);
2449 /* if ret == 0 there are items for this type,
2450 * create a range to tell us the last key of this type.
2451 * otherwise, there are no items in this directory after
2452 * *min_offset, and we create a range to indicate that.
2455 struct btrfs_key tmp;
2456 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2458 if (key_type == tmp.type)
2459 first_offset = max(min_offset, tmp.offset) + 1;
2464 /* go backward to find any previous key */
2465 ret = btrfs_previous_item(root, path, ino, key_type);
2467 struct btrfs_key tmp;
2468 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2469 if (key_type == tmp.type) {
2470 first_offset = tmp.offset;
2471 ret = overwrite_item(trans, log, dst_path,
2472 path->nodes[0], path->slots[0],
2480 btrfs_release_path(path);
2482 /* find the first key from this transaction again */
2483 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2490 * we have a block from this transaction, log every item in it
2491 * from our directory
2494 struct btrfs_key tmp;
2495 src = path->nodes[0];
2496 nritems = btrfs_header_nritems(src);
2497 for (i = path->slots[0]; i < nritems; i++) {
2498 btrfs_item_key_to_cpu(src, &min_key, i);
2500 if (min_key.objectid != ino || min_key.type != key_type)
2502 ret = overwrite_item(trans, log, dst_path, src, i,
2509 path->slots[0] = nritems;
2512 * look ahead to the next item and see if it is also
2513 * from this directory and from this transaction
2515 ret = btrfs_next_leaf(root, path);
2517 last_offset = (u64)-1;
2520 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2521 if (tmp.objectid != ino || tmp.type != key_type) {
2522 last_offset = (u64)-1;
2525 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2526 ret = overwrite_item(trans, log, dst_path,
2527 path->nodes[0], path->slots[0],
2532 last_offset = tmp.offset;
2537 btrfs_release_path(path);
2538 btrfs_release_path(dst_path);
2541 *last_offset_ret = last_offset;
2543 * insert the log range keys to indicate where the log
2546 ret = insert_dir_log_key(trans, log, path, key_type,
2547 ino, first_offset, last_offset);
2555 * logging directories is very similar to logging inodes, We find all the items
2556 * from the current transaction and write them to the log.
2558 * The recovery code scans the directory in the subvolume, and if it finds a
2559 * key in the range logged that is not present in the log tree, then it means
2560 * that dir entry was unlinked during the transaction.
2562 * In order for that scan to work, we must include one key smaller than
2563 * the smallest logged by this transaction and one key larger than the largest
2564 * key logged by this transaction.
2566 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2567 struct btrfs_root *root, struct inode *inode,
2568 struct btrfs_path *path,
2569 struct btrfs_path *dst_path)
2574 int key_type = BTRFS_DIR_ITEM_KEY;
2580 ret = log_dir_items(trans, root, inode, path,
2581 dst_path, key_type, min_key,
2585 if (max_key == (u64)-1)
2587 min_key = max_key + 1;
2590 if (key_type == BTRFS_DIR_ITEM_KEY) {
2591 key_type = BTRFS_DIR_INDEX_KEY;
2598 * a helper function to drop items from the log before we relog an
2599 * inode. max_key_type indicates the highest item type to remove.
2600 * This cannot be run for file data extents because it does not
2601 * free the extents they point to.
2603 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2604 struct btrfs_root *log,
2605 struct btrfs_path *path,
2606 u64 objectid, int max_key_type)
2609 struct btrfs_key key;
2610 struct btrfs_key found_key;
2612 key.objectid = objectid;
2613 key.type = max_key_type;
2614 key.offset = (u64)-1;
2617 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2622 if (path->slots[0] == 0)
2626 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2629 if (found_key.objectid != objectid)
2632 ret = btrfs_del_item(trans, log, path);
2635 btrfs_release_path(path);
2637 btrfs_release_path(path);
2641 static noinline int copy_items(struct btrfs_trans_handle *trans,
2642 struct btrfs_root *log,
2643 struct btrfs_path *dst_path,
2644 struct extent_buffer *src,
2645 int start_slot, int nr, int inode_only)
2647 unsigned long src_offset;
2648 unsigned long dst_offset;
2649 struct btrfs_file_extent_item *extent;
2650 struct btrfs_inode_item *inode_item;
2652 struct btrfs_key *ins_keys;
2656 struct list_head ordered_sums;
2658 INIT_LIST_HEAD(&ordered_sums);
2660 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2661 nr * sizeof(u32), GFP_NOFS);
2665 ins_sizes = (u32 *)ins_data;
2666 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2668 for (i = 0; i < nr; i++) {
2669 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2670 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2672 ret = btrfs_insert_empty_items(trans, log, dst_path,
2673 ins_keys, ins_sizes, nr);
2679 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
2680 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2681 dst_path->slots[0]);
2683 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2685 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2686 src_offset, ins_sizes[i]);
2688 if (inode_only == LOG_INODE_EXISTS &&
2689 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2690 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2692 struct btrfs_inode_item);
2693 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2695 /* set the generation to zero so the recover code
2696 * can tell the difference between an logging
2697 * just to say 'this inode exists' and a logging
2698 * to say 'update this inode with these values'
2700 btrfs_set_inode_generation(dst_path->nodes[0],
2703 /* take a reference on file data extents so that truncates
2704 * or deletes of this inode don't have to relog the inode
2707 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2709 extent = btrfs_item_ptr(src, start_slot + i,
2710 struct btrfs_file_extent_item);
2712 if (btrfs_file_extent_generation(src, extent) < trans->transid)
2715 found_type = btrfs_file_extent_type(src, extent);
2716 if (found_type == BTRFS_FILE_EXTENT_REG ||
2717 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
2719 ds = btrfs_file_extent_disk_bytenr(src,
2721 /* ds == 0 is a hole */
2725 dl = btrfs_file_extent_disk_num_bytes(src,
2727 cs = btrfs_file_extent_offset(src, extent);
2728 cl = btrfs_file_extent_num_bytes(src,
2730 if (btrfs_file_extent_compression(src,
2736 ret = btrfs_lookup_csums_range(
2737 log->fs_info->csum_root,
2738 ds + cs, ds + cs + cl - 1,
2745 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2746 btrfs_release_path(dst_path);
2750 * we have to do this after the loop above to avoid changing the
2751 * log tree while trying to change the log tree.
2754 while (!list_empty(&ordered_sums)) {
2755 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2756 struct btrfs_ordered_sum,
2759 ret = btrfs_csum_file_blocks(trans, log, sums);
2760 list_del(&sums->list);
2766 /* log a single inode in the tree log.
2767 * At least one parent directory for this inode must exist in the tree
2768 * or be logged already.
2770 * Any items from this inode changed by the current transaction are copied
2771 * to the log tree. An extra reference is taken on any extents in this
2772 * file, allowing us to avoid a whole pile of corner cases around logging
2773 * blocks that have been removed from the tree.
2775 * See LOG_INODE_ALL and related defines for a description of what inode_only
2778 * This handles both files and directories.
2780 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
2781 struct btrfs_root *root, struct inode *inode,
2784 struct btrfs_path *path;
2785 struct btrfs_path *dst_path;
2786 struct btrfs_key min_key;
2787 struct btrfs_key max_key;
2788 struct btrfs_root *log = root->log_root;
2789 struct extent_buffer *src = NULL;
2793 int ins_start_slot = 0;
2795 u64 ino = btrfs_ino(inode);
2797 log = root->log_root;
2799 path = btrfs_alloc_path();
2802 dst_path = btrfs_alloc_path();
2804 btrfs_free_path(path);
2808 min_key.objectid = ino;
2809 min_key.type = BTRFS_INODE_ITEM_KEY;
2812 max_key.objectid = ino;
2814 /* today the code can only do partial logging of directories */
2815 if (!S_ISDIR(inode->i_mode))
2816 inode_only = LOG_INODE_ALL;
2818 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2819 max_key.type = BTRFS_XATTR_ITEM_KEY;
2821 max_key.type = (u8)-1;
2822 max_key.offset = (u64)-1;
2824 ret = btrfs_commit_inode_delayed_items(trans, inode);
2826 btrfs_free_path(path);
2827 btrfs_free_path(dst_path);
2831 mutex_lock(&BTRFS_I(inode)->log_mutex);
2834 * a brute force approach to making sure we get the most uptodate
2835 * copies of everything.
2837 if (S_ISDIR(inode->i_mode)) {
2838 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2840 if (inode_only == LOG_INODE_EXISTS)
2841 max_key_type = BTRFS_XATTR_ITEM_KEY;
2842 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
2844 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2850 path->keep_locks = 1;
2854 ret = btrfs_search_forward(root, &min_key, &max_key,
2855 path, 0, trans->transid);
2859 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2860 if (min_key.objectid != ino)
2862 if (min_key.type > max_key.type)
2865 src = path->nodes[0];
2866 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2869 } else if (!ins_nr) {
2870 ins_start_slot = path->slots[0];
2875 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2876 ins_nr, inode_only);
2882 ins_start_slot = path->slots[0];
2885 nritems = btrfs_header_nritems(path->nodes[0]);
2887 if (path->slots[0] < nritems) {
2888 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2893 ret = copy_items(trans, log, dst_path, src,
2895 ins_nr, inode_only);
2902 btrfs_release_path(path);
2904 if (min_key.offset < (u64)-1)
2906 else if (min_key.type < (u8)-1)
2908 else if (min_key.objectid < (u64)-1)
2914 ret = copy_items(trans, log, dst_path, src,
2916 ins_nr, inode_only);
2924 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2925 btrfs_release_path(path);
2926 btrfs_release_path(dst_path);
2927 ret = log_directory_changes(trans, root, inode, path, dst_path);
2933 BTRFS_I(inode)->logged_trans = trans->transid;
2935 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2937 btrfs_free_path(path);
2938 btrfs_free_path(dst_path);
2943 * follow the dentry parent pointers up the chain and see if any
2944 * of the directories in it require a full commit before they can
2945 * be logged. Returns zero if nothing special needs to be done or 1 if
2946 * a full commit is required.
2948 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2949 struct inode *inode,
2950 struct dentry *parent,
2951 struct super_block *sb,
2955 struct btrfs_root *root;
2956 struct dentry *old_parent = NULL;
2959 * for regular files, if its inode is already on disk, we don't
2960 * have to worry about the parents at all. This is because
2961 * we can use the last_unlink_trans field to record renames
2962 * and other fun in this file.
2964 if (S_ISREG(inode->i_mode) &&
2965 BTRFS_I(inode)->generation <= last_committed &&
2966 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2969 if (!S_ISDIR(inode->i_mode)) {
2970 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2972 inode = parent->d_inode;
2976 BTRFS_I(inode)->logged_trans = trans->transid;
2979 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2980 root = BTRFS_I(inode)->root;
2983 * make sure any commits to the log are forced
2984 * to be full commits
2986 root->fs_info->last_trans_log_full_commit =
2992 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2995 if (IS_ROOT(parent))
2998 parent = dget_parent(parent);
3000 old_parent = parent;
3001 inode = parent->d_inode;
3009 static int inode_in_log(struct btrfs_trans_handle *trans,
3010 struct inode *inode)
3012 struct btrfs_root *root = BTRFS_I(inode)->root;
3015 mutex_lock(&root->log_mutex);
3016 if (BTRFS_I(inode)->logged_trans == trans->transid &&
3017 BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
3019 mutex_unlock(&root->log_mutex);
3025 * helper function around btrfs_log_inode to make sure newly created
3026 * parent directories also end up in the log. A minimal inode and backref
3027 * only logging is done of any parent directories that are older than
3028 * the last committed transaction
3030 int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
3031 struct btrfs_root *root, struct inode *inode,
3032 struct dentry *parent, int exists_only)
3034 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
3035 struct super_block *sb;
3036 struct dentry *old_parent = NULL;
3038 u64 last_committed = root->fs_info->last_trans_committed;
3042 if (btrfs_test_opt(root, NOTREELOG)) {
3047 if (root->fs_info->last_trans_log_full_commit >
3048 root->fs_info->last_trans_committed) {
3053 if (root != BTRFS_I(inode)->root ||
3054 btrfs_root_refs(&root->root_item) == 0) {
3059 ret = check_parent_dirs_for_sync(trans, inode, parent,
3060 sb, last_committed);
3064 if (inode_in_log(trans, inode)) {
3065 ret = BTRFS_NO_LOG_SYNC;
3069 ret = start_log_trans(trans, root);
3073 ret = btrfs_log_inode(trans, root, inode, inode_only);
3078 * for regular files, if its inode is already on disk, we don't
3079 * have to worry about the parents at all. This is because
3080 * we can use the last_unlink_trans field to record renames
3081 * and other fun in this file.
3083 if (S_ISREG(inode->i_mode) &&
3084 BTRFS_I(inode)->generation <= last_committed &&
3085 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3090 inode_only = LOG_INODE_EXISTS;
3092 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3095 inode = parent->d_inode;
3096 if (root != BTRFS_I(inode)->root)
3099 if (BTRFS_I(inode)->generation >
3100 root->fs_info->last_trans_committed) {
3101 ret = btrfs_log_inode(trans, root, inode, inode_only);
3105 if (IS_ROOT(parent))
3108 parent = dget_parent(parent);
3110 old_parent = parent;
3116 BUG_ON(ret != -ENOSPC);
3117 root->fs_info->last_trans_log_full_commit = trans->transid;
3120 btrfs_end_log_trans(root);
3126 * it is not safe to log dentry if the chunk root has added new
3127 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3128 * If this returns 1, you must commit the transaction to safely get your
3131 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3132 struct btrfs_root *root, struct dentry *dentry)
3134 struct dentry *parent = dget_parent(dentry);
3137 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
3144 * should be called during mount to recover any replay any log trees
3147 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3150 struct btrfs_path *path;
3151 struct btrfs_trans_handle *trans;
3152 struct btrfs_key key;
3153 struct btrfs_key found_key;
3154 struct btrfs_key tmp_key;
3155 struct btrfs_root *log;
3156 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3157 struct walk_control wc = {
3158 .process_func = process_one_buffer,
3162 path = btrfs_alloc_path();
3166 fs_info->log_root_recovering = 1;
3168 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3169 BUG_ON(IS_ERR(trans));
3174 ret = walk_log_tree(trans, log_root_tree, &wc);
3178 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3179 key.offset = (u64)-1;
3180 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3183 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3187 if (path->slots[0] == 0)
3191 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3193 btrfs_release_path(path);
3194 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3197 log = btrfs_read_fs_root_no_radix(log_root_tree,
3199 BUG_ON(IS_ERR(log));
3201 tmp_key.objectid = found_key.offset;
3202 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3203 tmp_key.offset = (u64)-1;
3205 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
3206 BUG_ON(IS_ERR_OR_NULL(wc.replay_dest));
3208 wc.replay_dest->log_root = log;
3209 btrfs_record_root_in_trans(trans, wc.replay_dest);
3210 ret = walk_log_tree(trans, log, &wc);
3213 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3214 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3219 key.offset = found_key.offset - 1;
3220 wc.replay_dest->log_root = NULL;
3221 free_extent_buffer(log->node);
3222 free_extent_buffer(log->commit_root);
3225 if (found_key.offset == 0)
3228 btrfs_release_path(path);
3230 /* step one is to pin it all, step two is to replay just inodes */
3233 wc.process_func = replay_one_buffer;
3234 wc.stage = LOG_WALK_REPLAY_INODES;
3237 /* step three is to replay everything */
3238 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3243 btrfs_free_path(path);
3245 free_extent_buffer(log_root_tree->node);
3246 log_root_tree->log_root = NULL;
3247 fs_info->log_root_recovering = 0;
3249 /* step 4: commit the transaction, which also unpins the blocks */
3250 btrfs_commit_transaction(trans, fs_info->tree_root);
3252 kfree(log_root_tree);
3257 * there are some corner cases where we want to force a full
3258 * commit instead of allowing a directory to be logged.
3260 * They revolve around files there were unlinked from the directory, and
3261 * this function updates the parent directory so that a full commit is
3262 * properly done if it is fsync'd later after the unlinks are done.
3264 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3265 struct inode *dir, struct inode *inode,
3269 * when we're logging a file, if it hasn't been renamed
3270 * or unlinked, and its inode is fully committed on disk,
3271 * we don't have to worry about walking up the directory chain
3272 * to log its parents.
3274 * So, we use the last_unlink_trans field to put this transid
3275 * into the file. When the file is logged we check it and
3276 * don't log the parents if the file is fully on disk.
3278 if (S_ISREG(inode->i_mode))
3279 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3282 * if this directory was already logged any new
3283 * names for this file/dir will get recorded
3286 if (BTRFS_I(dir)->logged_trans == trans->transid)
3290 * if the inode we're about to unlink was logged,
3291 * the log will be properly updated for any new names
3293 if (BTRFS_I(inode)->logged_trans == trans->transid)
3297 * when renaming files across directories, if the directory
3298 * there we're unlinking from gets fsync'd later on, there's
3299 * no way to find the destination directory later and fsync it
3300 * properly. So, we have to be conservative and force commits
3301 * so the new name gets discovered.
3306 /* we can safely do the unlink without any special recording */
3310 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3314 * Call this after adding a new name for a file and it will properly
3315 * update the log to reflect the new name.
3317 * It will return zero if all goes well, and it will return 1 if a
3318 * full transaction commit is required.
3320 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3321 struct inode *inode, struct inode *old_dir,
3322 struct dentry *parent)
3324 struct btrfs_root * root = BTRFS_I(inode)->root;
3327 * this will force the logging code to walk the dentry chain
3330 if (S_ISREG(inode->i_mode))
3331 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3334 * if this inode hasn't been logged and directory we're renaming it
3335 * from hasn't been logged, we don't need to log it
3337 if (BTRFS_I(inode)->logged_trans <=
3338 root->fs_info->last_trans_committed &&
3339 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3340 root->fs_info->last_trans_committed))
3343 return btrfs_log_inode_parent(trans, root, inode, parent, 1);