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>
21 #include "transaction.h"
24 #include "print-tree.h"
28 /* magic values for the inode_only field in btrfs_log_inode:
30 * LOG_INODE_ALL means to log everything
31 * LOG_INODE_EXISTS means to log just enough to recreate the inode
34 #define LOG_INODE_ALL 0
35 #define LOG_INODE_EXISTS 1
38 * directory trouble cases
40 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
41 * log, we must force a full commit before doing an fsync of the directory
42 * where the unlink was done.
43 * ---> record transid of last unlink/rename per directory
47 * rename foo/some_dir foo2/some_dir
49 * fsync foo/some_dir/some_file
51 * The fsync above will unlink the original some_dir without recording
52 * it in its new location (foo2). After a crash, some_dir will be gone
53 * unless the fsync of some_file forces a full commit
55 * 2) we must log any new names for any file or dir that is in the fsync
56 * log. ---> check inode while renaming/linking.
58 * 2a) we must log any new names for any file or dir during rename
59 * when the directory they are being removed from was logged.
60 * ---> check inode and old parent dir during rename
62 * 2a is actually the more important variant. With the extra logging
63 * a crash might unlink the old name without recreating the new one
65 * 3) after a crash, we must go through any directories with a link count
66 * of zero and redo the rm -rf
73 * The directory f1 was fully removed from the FS, but fsync was never
74 * called on f1, only its parent dir. After a crash the rm -rf must
75 * be replayed. This must be able to recurse down the entire
76 * directory tree. The inode link count fixup code takes care of the
81 * stages for the tree walking. The first
82 * stage (0) is to only pin down the blocks we find
83 * the second stage (1) is to make sure that all the inodes
84 * we find in the log are created in the subvolume.
86 * The last stage is to deal with directories and links and extents
87 * and all the other fun semantics
89 #define LOG_WALK_PIN_ONLY 0
90 #define LOG_WALK_REPLAY_INODES 1
91 #define LOG_WALK_REPLAY_ALL 2
93 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
94 struct btrfs_root *root, struct inode *inode,
96 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root,
98 struct btrfs_path *path, u64 objectid);
99 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
100 struct btrfs_root *root,
101 struct btrfs_root *log,
102 struct btrfs_path *path,
103 u64 dirid, int del_all);
106 * tree logging is a special write ahead log used to make sure that
107 * fsyncs and O_SYNCs can happen without doing full tree commits.
109 * Full tree commits are expensive because they require commonly
110 * modified blocks to be recowed, creating many dirty pages in the
111 * extent tree an 4x-6x higher write load than ext3.
113 * Instead of doing a tree commit on every fsync, we use the
114 * key ranges and transaction ids to find items for a given file or directory
115 * that have changed in this transaction. Those items are copied into
116 * a special tree (one per subvolume root), that tree is written to disk
117 * and then the fsync is considered complete.
119 * After a crash, items are copied out of the log-tree back into the
120 * subvolume tree. Any file data extents found are recorded in the extent
121 * allocation tree, and the log-tree freed.
123 * The log tree is read three times, once to pin down all the extents it is
124 * using in ram and once, once to create all the inodes logged in the tree
125 * and once to do all the other items.
129 * start a sub transaction and setup the log tree
130 * this increments the log tree writer count to make the people
131 * syncing the tree wait for us to finish
133 static int start_log_trans(struct btrfs_trans_handle *trans,
134 struct btrfs_root *root)
138 mutex_lock(&root->log_mutex);
139 if (root->log_root) {
140 if (!root->log_start_pid) {
141 root->log_start_pid = current->pid;
142 root->log_multiple_pids = false;
143 } else if (root->log_start_pid != current->pid) {
144 root->log_multiple_pids = true;
148 atomic_inc(&root->log_writers);
149 mutex_unlock(&root->log_mutex);
152 root->log_multiple_pids = false;
153 root->log_start_pid = current->pid;
154 mutex_lock(&root->fs_info->tree_log_mutex);
155 if (!root->fs_info->log_root_tree) {
156 ret = btrfs_init_log_root_tree(trans, root->fs_info);
159 if (!root->log_root) {
160 ret = btrfs_add_log_tree(trans, root);
163 mutex_unlock(&root->fs_info->tree_log_mutex);
165 atomic_inc(&root->log_writers);
166 mutex_unlock(&root->log_mutex);
171 * returns 0 if there was a log transaction running and we were able
172 * to join, or returns -ENOENT if there were not transactions
175 static int join_running_log_trans(struct btrfs_root *root)
183 mutex_lock(&root->log_mutex);
184 if (root->log_root) {
186 atomic_inc(&root->log_writers);
188 mutex_unlock(&root->log_mutex);
193 * This either makes the current running log transaction wait
194 * until you call btrfs_end_log_trans() or it makes any future
195 * log transactions wait until you call btrfs_end_log_trans()
197 int btrfs_pin_log_trans(struct btrfs_root *root)
201 mutex_lock(&root->log_mutex);
202 atomic_inc(&root->log_writers);
203 mutex_unlock(&root->log_mutex);
208 * indicate we're done making changes to the log tree
209 * and wake up anyone waiting to do a sync
211 int btrfs_end_log_trans(struct btrfs_root *root)
213 if (atomic_dec_and_test(&root->log_writers)) {
215 if (waitqueue_active(&root->log_writer_wait))
216 wake_up(&root->log_writer_wait);
223 * the walk control struct is used to pass state down the chain when
224 * processing the log tree. The stage field tells us which part
225 * of the log tree processing we are currently doing. The others
226 * are state fields used for that specific part
228 struct walk_control {
229 /* should we free the extent on disk when done? This is used
230 * at transaction commit time while freeing a log tree
234 /* should we write out the extent buffer? This is used
235 * while flushing the log tree to disk during a sync
239 /* should we wait for the extent buffer io to finish? Also used
240 * while flushing the log tree to disk for a sync
244 /* pin only walk, we record which extents on disk belong to the
249 /* what stage of the replay code we're currently in */
252 /* the root we are currently replaying */
253 struct btrfs_root *replay_dest;
255 /* the trans handle for the current replay */
256 struct btrfs_trans_handle *trans;
258 /* the function that gets used to process blocks we find in the
259 * tree. Note the extent_buffer might not be up to date when it is
260 * passed in, and it must be checked or read if you need the data
263 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
264 struct walk_control *wc, u64 gen);
268 * process_func used to pin down extents, write them or wait on them
270 static int process_one_buffer(struct btrfs_root *log,
271 struct extent_buffer *eb,
272 struct walk_control *wc, u64 gen)
275 btrfs_pin_extent(log->fs_info->extent_root,
276 eb->start, eb->len, 0);
278 if (btrfs_buffer_uptodate(eb, gen)) {
280 btrfs_write_tree_block(eb);
282 btrfs_wait_tree_block_writeback(eb);
288 * Item overwrite used by replay and tree logging. eb, slot and key all refer
289 * to the src data we are copying out.
291 * root is the tree we are copying into, and path is a scratch
292 * path for use in this function (it should be released on entry and
293 * will be released on exit).
295 * If the key is already in the destination tree the existing item is
296 * overwritten. If the existing item isn't big enough, it is extended.
297 * If it is too large, it is truncated.
299 * If the key isn't in the destination yet, a new item is inserted.
301 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
302 struct btrfs_root *root,
303 struct btrfs_path *path,
304 struct extent_buffer *eb, int slot,
305 struct btrfs_key *key)
309 u64 saved_i_size = 0;
310 int save_old_i_size = 0;
311 unsigned long src_ptr;
312 unsigned long dst_ptr;
313 int overwrite_root = 0;
315 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
318 item_size = btrfs_item_size_nr(eb, slot);
319 src_ptr = btrfs_item_ptr_offset(eb, slot);
321 /* look for the key in the destination tree */
322 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
326 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
328 if (dst_size != item_size)
331 if (item_size == 0) {
332 btrfs_release_path(root, path);
335 dst_copy = kmalloc(item_size, GFP_NOFS);
336 src_copy = kmalloc(item_size, GFP_NOFS);
338 read_extent_buffer(eb, src_copy, src_ptr, item_size);
340 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
341 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
343 ret = memcmp(dst_copy, src_copy, item_size);
348 * they have the same contents, just return, this saves
349 * us from cowing blocks in the destination tree and doing
350 * extra writes that may not have been done by a previous
354 btrfs_release_path(root, path);
360 btrfs_release_path(root, path);
361 /* try to insert the key into the destination tree */
362 ret = btrfs_insert_empty_item(trans, root, path,
365 /* make sure any existing item is the correct size */
366 if (ret == -EEXIST) {
368 found_size = btrfs_item_size_nr(path->nodes[0],
370 if (found_size > item_size) {
371 btrfs_truncate_item(trans, root, path, item_size, 1);
372 } else if (found_size < item_size) {
373 ret = btrfs_extend_item(trans, root, path,
374 item_size - found_size);
380 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
383 /* don't overwrite an existing inode if the generation number
384 * was logged as zero. This is done when the tree logging code
385 * is just logging an inode to make sure it exists after recovery.
387 * Also, don't overwrite i_size on directories during replay.
388 * log replay inserts and removes directory items based on the
389 * state of the tree found in the subvolume, and i_size is modified
392 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
393 struct btrfs_inode_item *src_item;
394 struct btrfs_inode_item *dst_item;
396 src_item = (struct btrfs_inode_item *)src_ptr;
397 dst_item = (struct btrfs_inode_item *)dst_ptr;
399 if (btrfs_inode_generation(eb, src_item) == 0)
402 if (overwrite_root &&
403 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
404 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
406 saved_i_size = btrfs_inode_size(path->nodes[0],
411 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
414 if (save_old_i_size) {
415 struct btrfs_inode_item *dst_item;
416 dst_item = (struct btrfs_inode_item *)dst_ptr;
417 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
420 /* make sure the generation is filled in */
421 if (key->type == BTRFS_INODE_ITEM_KEY) {
422 struct btrfs_inode_item *dst_item;
423 dst_item = (struct btrfs_inode_item *)dst_ptr;
424 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
425 btrfs_set_inode_generation(path->nodes[0], dst_item,
430 btrfs_mark_buffer_dirty(path->nodes[0]);
431 btrfs_release_path(root, path);
436 * simple helper to read an inode off the disk from a given root
437 * This can only be called for subvolume roots and not for the log
439 static noinline struct inode *read_one_inode(struct btrfs_root *root,
442 struct btrfs_key key;
445 key.objectid = objectid;
446 key.type = BTRFS_INODE_ITEM_KEY;
448 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
451 } else if (is_bad_inode(inode)) {
458 /* replays a single extent in 'eb' at 'slot' with 'key' into the
459 * subvolume 'root'. path is released on entry and should be released
462 * extents in the log tree have not been allocated out of the extent
463 * tree yet. So, this completes the allocation, taking a reference
464 * as required if the extent already exists or creating a new extent
465 * if it isn't in the extent allocation tree yet.
467 * The extent is inserted into the file, dropping any existing extents
468 * from the file that overlap the new one.
470 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
471 struct btrfs_root *root,
472 struct btrfs_path *path,
473 struct extent_buffer *eb, int slot,
474 struct btrfs_key *key)
477 u64 mask = root->sectorsize - 1;
480 u64 start = key->offset;
482 struct btrfs_file_extent_item *item;
483 struct inode *inode = NULL;
487 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
488 found_type = btrfs_file_extent_type(eb, item);
490 if (found_type == BTRFS_FILE_EXTENT_REG ||
491 found_type == BTRFS_FILE_EXTENT_PREALLOC)
492 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
493 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
494 size = btrfs_file_extent_inline_len(eb, item);
495 extent_end = (start + size + mask) & ~mask;
501 inode = read_one_inode(root, key->objectid);
508 * first check to see if we already have this extent in the
509 * file. This must be done before the btrfs_drop_extents run
510 * so we don't try to drop this extent.
512 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
516 (found_type == BTRFS_FILE_EXTENT_REG ||
517 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
518 struct btrfs_file_extent_item cmp1;
519 struct btrfs_file_extent_item cmp2;
520 struct btrfs_file_extent_item *existing;
521 struct extent_buffer *leaf;
523 leaf = path->nodes[0];
524 existing = btrfs_item_ptr(leaf, path->slots[0],
525 struct btrfs_file_extent_item);
527 read_extent_buffer(eb, &cmp1, (unsigned long)item,
529 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
533 * we already have a pointer to this exact extent,
534 * we don't have to do anything
536 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
537 btrfs_release_path(root, path);
541 btrfs_release_path(root, path);
543 saved_nbytes = inode_get_bytes(inode);
544 /* drop any overlapping extents */
545 ret = btrfs_drop_extents(trans, inode, start, extent_end,
549 if (found_type == BTRFS_FILE_EXTENT_REG ||
550 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
552 unsigned long dest_offset;
553 struct btrfs_key ins;
555 ret = btrfs_insert_empty_item(trans, root, path, key,
558 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
560 copy_extent_buffer(path->nodes[0], eb, dest_offset,
561 (unsigned long)item, sizeof(*item));
563 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
564 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
565 ins.type = BTRFS_EXTENT_ITEM_KEY;
566 offset = key->offset - btrfs_file_extent_offset(eb, item);
568 if (ins.objectid > 0) {
571 LIST_HEAD(ordered_sums);
573 * is this extent already allocated in the extent
574 * allocation tree? If so, just add a reference
576 ret = btrfs_lookup_extent(root, ins.objectid,
579 ret = btrfs_inc_extent_ref(trans, root,
580 ins.objectid, ins.offset,
581 0, root->root_key.objectid,
582 key->objectid, offset);
585 * insert the extent pointer in the extent
588 ret = btrfs_alloc_logged_file_extent(trans,
589 root, root->root_key.objectid,
590 key->objectid, offset, &ins);
593 btrfs_release_path(root, path);
595 if (btrfs_file_extent_compression(eb, item)) {
596 csum_start = ins.objectid;
597 csum_end = csum_start + ins.offset;
599 csum_start = ins.objectid +
600 btrfs_file_extent_offset(eb, item);
601 csum_end = csum_start +
602 btrfs_file_extent_num_bytes(eb, item);
605 ret = btrfs_lookup_csums_range(root->log_root,
606 csum_start, csum_end - 1,
609 while (!list_empty(&ordered_sums)) {
610 struct btrfs_ordered_sum *sums;
611 sums = list_entry(ordered_sums.next,
612 struct btrfs_ordered_sum,
614 ret = btrfs_csum_file_blocks(trans,
615 root->fs_info->csum_root,
618 list_del(&sums->list);
622 btrfs_release_path(root, path);
624 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
625 /* inline extents are easy, we just overwrite them */
626 ret = overwrite_item(trans, root, path, eb, slot, key);
630 inode_set_bytes(inode, saved_nbytes);
631 btrfs_update_inode(trans, root, inode);
639 * when cleaning up conflicts between the directory names in the
640 * subvolume, directory names in the log and directory names in the
641 * inode back references, we may have to unlink inodes from directories.
643 * This is a helper function to do the unlink of a specific directory
646 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
647 struct btrfs_root *root,
648 struct btrfs_path *path,
650 struct btrfs_dir_item *di)
655 struct extent_buffer *leaf;
656 struct btrfs_key location;
659 leaf = path->nodes[0];
661 btrfs_dir_item_key_to_cpu(leaf, di, &location);
662 name_len = btrfs_dir_name_len(leaf, di);
663 name = kmalloc(name_len, GFP_NOFS);
664 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
665 btrfs_release_path(root, path);
667 inode = read_one_inode(root, location.objectid);
670 ret = link_to_fixup_dir(trans, root, path, location.objectid);
673 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
682 * helper function to see if a given name and sequence number found
683 * in an inode back reference are already in a directory and correctly
684 * point to this inode
686 static noinline int inode_in_dir(struct btrfs_root *root,
687 struct btrfs_path *path,
688 u64 dirid, u64 objectid, u64 index,
689 const char *name, int name_len)
691 struct btrfs_dir_item *di;
692 struct btrfs_key location;
695 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
696 index, name, name_len, 0);
697 if (di && !IS_ERR(di)) {
698 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
699 if (location.objectid != objectid)
703 btrfs_release_path(root, path);
705 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
706 if (di && !IS_ERR(di)) {
707 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
708 if (location.objectid != objectid)
714 btrfs_release_path(root, path);
719 * helper function to check a log tree for a named back reference in
720 * an inode. This is used to decide if a back reference that is
721 * found in the subvolume conflicts with what we find in the log.
723 * inode backreferences may have multiple refs in a single item,
724 * during replay we process one reference at a time, and we don't
725 * want to delete valid links to a file from the subvolume if that
726 * link is also in the log.
728 static noinline int backref_in_log(struct btrfs_root *log,
729 struct btrfs_key *key,
730 char *name, int namelen)
732 struct btrfs_path *path;
733 struct btrfs_inode_ref *ref;
735 unsigned long ptr_end;
736 unsigned long name_ptr;
742 path = btrfs_alloc_path();
743 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
747 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
748 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
749 ptr_end = ptr + item_size;
750 while (ptr < ptr_end) {
751 ref = (struct btrfs_inode_ref *)ptr;
752 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
753 if (found_name_len == namelen) {
754 name_ptr = (unsigned long)(ref + 1);
755 ret = memcmp_extent_buffer(path->nodes[0], name,
762 ptr = (unsigned long)(ref + 1) + found_name_len;
765 btrfs_free_path(path);
771 * replay one inode back reference item found in the log tree.
772 * eb, slot and key refer to the buffer and key found in the log tree.
773 * root is the destination we are replaying into, and path is for temp
774 * use by this function. (it should be released on return).
776 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
777 struct btrfs_root *root,
778 struct btrfs_root *log,
779 struct btrfs_path *path,
780 struct extent_buffer *eb, int slot,
781 struct btrfs_key *key)
785 struct btrfs_key location;
786 struct btrfs_inode_ref *ref;
787 struct btrfs_dir_item *di;
791 unsigned long ref_ptr;
792 unsigned long ref_end;
794 location.objectid = key->objectid;
795 location.type = BTRFS_INODE_ITEM_KEY;
799 * it is possible that we didn't log all the parent directories
800 * for a given inode. If we don't find the dir, just don't
801 * copy the back ref in. The link count fixup code will take
804 dir = read_one_inode(root, key->offset);
808 inode = read_one_inode(root, key->objectid);
811 ref_ptr = btrfs_item_ptr_offset(eb, slot);
812 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
815 ref = (struct btrfs_inode_ref *)ref_ptr;
817 namelen = btrfs_inode_ref_name_len(eb, ref);
818 name = kmalloc(namelen, GFP_NOFS);
821 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
823 /* if we already have a perfect match, we're done */
824 if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
825 btrfs_inode_ref_index(eb, ref),
831 * look for a conflicting back reference in the metadata.
832 * if we find one we have to unlink that name of the file
833 * before we add our new link. Later on, we overwrite any
834 * existing back reference, and we don't want to create
835 * dangling pointers in the directory.
838 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
842 struct btrfs_inode_ref *victim_ref;
844 unsigned long ptr_end;
845 struct extent_buffer *leaf = path->nodes[0];
847 /* are we trying to overwrite a back ref for the root directory
848 * if so, just jump out, we're done
850 if (key->objectid == key->offset)
853 /* check all the names in this back reference to see
854 * if they are in the log. if so, we allow them to stay
855 * otherwise they must be unlinked as a conflict
857 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
858 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
859 while (ptr < ptr_end) {
860 victim_ref = (struct btrfs_inode_ref *)ptr;
861 victim_name_len = btrfs_inode_ref_name_len(leaf,
863 victim_name = kmalloc(victim_name_len, GFP_NOFS);
864 BUG_ON(!victim_name);
866 read_extent_buffer(leaf, victim_name,
867 (unsigned long)(victim_ref + 1),
870 if (!backref_in_log(log, key, victim_name,
872 btrfs_inc_nlink(inode);
873 btrfs_release_path(root, path);
875 ret = btrfs_unlink_inode(trans, root, dir,
879 btrfs_release_path(root, path);
883 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
887 btrfs_release_path(root, path);
889 /* look for a conflicting sequence number */
890 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
891 btrfs_inode_ref_index(eb, ref),
893 if (di && !IS_ERR(di)) {
894 ret = drop_one_dir_item(trans, root, path, dir, di);
897 btrfs_release_path(root, path);
900 /* look for a conflicting name */
901 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
903 if (di && !IS_ERR(di)) {
904 ret = drop_one_dir_item(trans, root, path, dir, di);
907 btrfs_release_path(root, path);
909 /* insert our name */
910 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
911 btrfs_inode_ref_index(eb, ref));
914 btrfs_update_inode(trans, root, inode);
917 ref_ptr = (unsigned long)(ref + 1) + namelen;
919 if (ref_ptr < ref_end)
922 /* finally write the back reference in the inode */
923 ret = overwrite_item(trans, root, path, eb, slot, key);
927 btrfs_release_path(root, path);
933 static int insert_orphan_item(struct btrfs_trans_handle *trans,
934 struct btrfs_root *root, u64 offset)
937 ret = btrfs_find_orphan_item(root, offset);
939 ret = btrfs_insert_orphan_item(trans, root, offset);
945 * There are a few corners where the link count of the file can't
946 * be properly maintained during replay. So, instead of adding
947 * lots of complexity to the log code, we just scan the backrefs
948 * for any file that has been through replay.
950 * The scan will update the link count on the inode to reflect the
951 * number of back refs found. If it goes down to zero, the iput
952 * will free the inode.
954 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
955 struct btrfs_root *root,
958 struct btrfs_path *path;
960 struct btrfs_key key;
963 unsigned long ptr_end;
966 key.objectid = inode->i_ino;
967 key.type = BTRFS_INODE_REF_KEY;
968 key.offset = (u64)-1;
970 path = btrfs_alloc_path();
973 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
977 if (path->slots[0] == 0)
981 btrfs_item_key_to_cpu(path->nodes[0], &key,
983 if (key.objectid != inode->i_ino ||
984 key.type != BTRFS_INODE_REF_KEY)
986 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
987 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
989 while (ptr < ptr_end) {
990 struct btrfs_inode_ref *ref;
992 ref = (struct btrfs_inode_ref *)ptr;
993 name_len = btrfs_inode_ref_name_len(path->nodes[0],
995 ptr = (unsigned long)(ref + 1) + name_len;
1002 btrfs_release_path(root, path);
1004 btrfs_release_path(root, path);
1005 if (nlink != inode->i_nlink) {
1006 inode->i_nlink = nlink;
1007 btrfs_update_inode(trans, root, inode);
1009 BTRFS_I(inode)->index_cnt = (u64)-1;
1011 if (inode->i_nlink == 0) {
1012 if (S_ISDIR(inode->i_mode)) {
1013 ret = replay_dir_deletes(trans, root, NULL, path,
1017 ret = insert_orphan_item(trans, root, inode->i_ino);
1020 btrfs_free_path(path);
1025 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1026 struct btrfs_root *root,
1027 struct btrfs_path *path)
1030 struct btrfs_key key;
1031 struct inode *inode;
1033 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1034 key.type = BTRFS_ORPHAN_ITEM_KEY;
1035 key.offset = (u64)-1;
1037 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1042 if (path->slots[0] == 0)
1047 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1048 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1049 key.type != BTRFS_ORPHAN_ITEM_KEY)
1052 ret = btrfs_del_item(trans, root, path);
1055 btrfs_release_path(root, path);
1056 inode = read_one_inode(root, key.offset);
1059 ret = fixup_inode_link_count(trans, root, inode);
1065 * fixup on a directory may create new entries,
1066 * make sure we always look for the highset possible
1069 key.offset = (u64)-1;
1071 btrfs_release_path(root, path);
1077 * record a given inode in the fixup dir so we can check its link
1078 * count when replay is done. The link count is incremented here
1079 * so the inode won't go away until we check it
1081 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1082 struct btrfs_root *root,
1083 struct btrfs_path *path,
1086 struct btrfs_key key;
1088 struct inode *inode;
1090 inode = read_one_inode(root, objectid);
1093 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1094 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1095 key.offset = objectid;
1097 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1099 btrfs_release_path(root, path);
1101 btrfs_inc_nlink(inode);
1102 btrfs_update_inode(trans, root, inode);
1103 } else if (ret == -EEXIST) {
1114 * when replaying the log for a directory, we only insert names
1115 * for inodes that actually exist. This means an fsync on a directory
1116 * does not implicitly fsync all the new files in it
1118 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1119 struct btrfs_root *root,
1120 struct btrfs_path *path,
1121 u64 dirid, u64 index,
1122 char *name, int name_len, u8 type,
1123 struct btrfs_key *location)
1125 struct inode *inode;
1129 inode = read_one_inode(root, location->objectid);
1133 dir = read_one_inode(root, dirid);
1138 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1140 /* FIXME, put inode into FIXUP list */
1148 * take a single entry in a log directory item and replay it into
1151 * if a conflicting item exists in the subdirectory already,
1152 * the inode it points to is unlinked and put into the link count
1155 * If a name from the log points to a file or directory that does
1156 * not exist in the FS, it is skipped. fsyncs on directories
1157 * do not force down inodes inside that directory, just changes to the
1158 * names or unlinks in a directory.
1160 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1161 struct btrfs_root *root,
1162 struct btrfs_path *path,
1163 struct extent_buffer *eb,
1164 struct btrfs_dir_item *di,
1165 struct btrfs_key *key)
1169 struct btrfs_dir_item *dst_di;
1170 struct btrfs_key found_key;
1171 struct btrfs_key log_key;
1177 dir = read_one_inode(root, key->objectid);
1180 name_len = btrfs_dir_name_len(eb, di);
1181 name = kmalloc(name_len, GFP_NOFS);
1182 log_type = btrfs_dir_type(eb, di);
1183 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1186 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1187 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1192 btrfs_release_path(root, path);
1194 if (key->type == BTRFS_DIR_ITEM_KEY) {
1195 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1197 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1198 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1205 if (!dst_di || IS_ERR(dst_di)) {
1206 /* we need a sequence number to insert, so we only
1207 * do inserts for the BTRFS_DIR_INDEX_KEY types
1209 if (key->type != BTRFS_DIR_INDEX_KEY)
1214 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1215 /* the existing item matches the logged item */
1216 if (found_key.objectid == log_key.objectid &&
1217 found_key.type == log_key.type &&
1218 found_key.offset == log_key.offset &&
1219 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1224 * don't drop the conflicting directory entry if the inode
1225 * for the new entry doesn't exist
1230 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1233 if (key->type == BTRFS_DIR_INDEX_KEY)
1236 btrfs_release_path(root, path);
1242 btrfs_release_path(root, path);
1243 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1244 name, name_len, log_type, &log_key);
1246 BUG_ON(ret && ret != -ENOENT);
1251 * find all the names in a directory item and reconcile them into
1252 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1253 * one name in a directory item, but the same code gets used for
1254 * both directory index types
1256 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1257 struct btrfs_root *root,
1258 struct btrfs_path *path,
1259 struct extent_buffer *eb, int slot,
1260 struct btrfs_key *key)
1263 u32 item_size = btrfs_item_size_nr(eb, slot);
1264 struct btrfs_dir_item *di;
1267 unsigned long ptr_end;
1269 ptr = btrfs_item_ptr_offset(eb, slot);
1270 ptr_end = ptr + item_size;
1271 while (ptr < ptr_end) {
1272 di = (struct btrfs_dir_item *)ptr;
1273 name_len = btrfs_dir_name_len(eb, di);
1274 ret = replay_one_name(trans, root, path, eb, di, key);
1276 ptr = (unsigned long)(di + 1);
1283 * directory replay has two parts. There are the standard directory
1284 * items in the log copied from the subvolume, and range items
1285 * created in the log while the subvolume was logged.
1287 * The range items tell us which parts of the key space the log
1288 * is authoritative for. During replay, if a key in the subvolume
1289 * directory is in a logged range item, but not actually in the log
1290 * that means it was deleted from the directory before the fsync
1291 * and should be removed.
1293 static noinline int find_dir_range(struct btrfs_root *root,
1294 struct btrfs_path *path,
1295 u64 dirid, int key_type,
1296 u64 *start_ret, u64 *end_ret)
1298 struct btrfs_key key;
1300 struct btrfs_dir_log_item *item;
1304 if (*start_ret == (u64)-1)
1307 key.objectid = dirid;
1308 key.type = key_type;
1309 key.offset = *start_ret;
1311 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1315 if (path->slots[0] == 0)
1320 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1322 if (key.type != key_type || key.objectid != dirid) {
1326 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1327 struct btrfs_dir_log_item);
1328 found_end = btrfs_dir_log_end(path->nodes[0], item);
1330 if (*start_ret >= key.offset && *start_ret <= found_end) {
1332 *start_ret = key.offset;
1333 *end_ret = found_end;
1338 /* check the next slot in the tree to see if it is a valid item */
1339 nritems = btrfs_header_nritems(path->nodes[0]);
1340 if (path->slots[0] >= nritems) {
1341 ret = btrfs_next_leaf(root, path);
1348 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1350 if (key.type != key_type || key.objectid != dirid) {
1354 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1355 struct btrfs_dir_log_item);
1356 found_end = btrfs_dir_log_end(path->nodes[0], item);
1357 *start_ret = key.offset;
1358 *end_ret = found_end;
1361 btrfs_release_path(root, path);
1366 * this looks for a given directory item in the log. If the directory
1367 * item is not in the log, the item is removed and the inode it points
1370 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_root *log,
1373 struct btrfs_path *path,
1374 struct btrfs_path *log_path,
1376 struct btrfs_key *dir_key)
1379 struct extent_buffer *eb;
1382 struct btrfs_dir_item *di;
1383 struct btrfs_dir_item *log_di;
1386 unsigned long ptr_end;
1388 struct inode *inode;
1389 struct btrfs_key location;
1392 eb = path->nodes[0];
1393 slot = path->slots[0];
1394 item_size = btrfs_item_size_nr(eb, slot);
1395 ptr = btrfs_item_ptr_offset(eb, slot);
1396 ptr_end = ptr + item_size;
1397 while (ptr < ptr_end) {
1398 di = (struct btrfs_dir_item *)ptr;
1399 name_len = btrfs_dir_name_len(eb, di);
1400 name = kmalloc(name_len, GFP_NOFS);
1405 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1408 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1409 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1412 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1413 log_di = btrfs_lookup_dir_index_item(trans, log,
1419 if (!log_di || IS_ERR(log_di)) {
1420 btrfs_dir_item_key_to_cpu(eb, di, &location);
1421 btrfs_release_path(root, path);
1422 btrfs_release_path(log, log_path);
1423 inode = read_one_inode(root, location.objectid);
1426 ret = link_to_fixup_dir(trans, root,
1427 path, location.objectid);
1429 btrfs_inc_nlink(inode);
1430 ret = btrfs_unlink_inode(trans, root, dir, inode,
1436 /* there might still be more names under this key
1437 * check and repeat if required
1439 ret = btrfs_search_slot(NULL, root, dir_key, path,
1446 btrfs_release_path(log, log_path);
1449 ptr = (unsigned long)(di + 1);
1454 btrfs_release_path(root, path);
1455 btrfs_release_path(log, log_path);
1460 * deletion replay happens before we copy any new directory items
1461 * out of the log or out of backreferences from inodes. It
1462 * scans the log to find ranges of keys that log is authoritative for,
1463 * and then scans the directory to find items in those ranges that are
1464 * not present in the log.
1466 * Anything we don't find in the log is unlinked and removed from the
1469 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1470 struct btrfs_root *root,
1471 struct btrfs_root *log,
1472 struct btrfs_path *path,
1473 u64 dirid, int del_all)
1477 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1479 struct btrfs_key dir_key;
1480 struct btrfs_key found_key;
1481 struct btrfs_path *log_path;
1484 dir_key.objectid = dirid;
1485 dir_key.type = BTRFS_DIR_ITEM_KEY;
1486 log_path = btrfs_alloc_path();
1490 dir = read_one_inode(root, dirid);
1491 /* it isn't an error if the inode isn't there, that can happen
1492 * because we replay the deletes before we copy in the inode item
1496 btrfs_free_path(log_path);
1504 range_end = (u64)-1;
1506 ret = find_dir_range(log, path, dirid, key_type,
1507 &range_start, &range_end);
1512 dir_key.offset = range_start;
1515 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1520 nritems = btrfs_header_nritems(path->nodes[0]);
1521 if (path->slots[0] >= nritems) {
1522 ret = btrfs_next_leaf(root, path);
1526 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1528 if (found_key.objectid != dirid ||
1529 found_key.type != dir_key.type)
1532 if (found_key.offset > range_end)
1535 ret = check_item_in_log(trans, root, log, path,
1539 if (found_key.offset == (u64)-1)
1541 dir_key.offset = found_key.offset + 1;
1543 btrfs_release_path(root, path);
1544 if (range_end == (u64)-1)
1546 range_start = range_end + 1;
1551 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1552 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1553 dir_key.type = BTRFS_DIR_INDEX_KEY;
1554 btrfs_release_path(root, path);
1558 btrfs_release_path(root, path);
1559 btrfs_free_path(log_path);
1565 * the process_func used to replay items from the log tree. This
1566 * gets called in two different stages. The first stage just looks
1567 * for inodes and makes sure they are all copied into the subvolume.
1569 * The second stage copies all the other item types from the log into
1570 * the subvolume. The two stage approach is slower, but gets rid of
1571 * lots of complexity around inodes referencing other inodes that exist
1572 * only in the log (references come from either directory items or inode
1575 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1576 struct walk_control *wc, u64 gen)
1579 struct btrfs_path *path;
1580 struct btrfs_root *root = wc->replay_dest;
1581 struct btrfs_key key;
1587 btrfs_read_buffer(eb, gen);
1589 level = btrfs_header_level(eb);
1594 path = btrfs_alloc_path();
1597 nritems = btrfs_header_nritems(eb);
1598 for (i = 0; i < nritems; i++) {
1599 btrfs_item_key_to_cpu(eb, &key, i);
1600 item_size = btrfs_item_size_nr(eb, i);
1602 /* inode keys are done during the first stage */
1603 if (key.type == BTRFS_INODE_ITEM_KEY &&
1604 wc->stage == LOG_WALK_REPLAY_INODES) {
1605 struct btrfs_inode_item *inode_item;
1608 inode_item = btrfs_item_ptr(eb, i,
1609 struct btrfs_inode_item);
1610 mode = btrfs_inode_mode(eb, inode_item);
1611 if (S_ISDIR(mode)) {
1612 ret = replay_dir_deletes(wc->trans,
1613 root, log, path, key.objectid, 0);
1616 ret = overwrite_item(wc->trans, root, path,
1620 /* for regular files, make sure corresponding
1621 * orhpan item exist. extents past the new EOF
1622 * will be truncated later by orphan cleanup.
1624 if (S_ISREG(mode)) {
1625 ret = insert_orphan_item(wc->trans, root,
1630 ret = link_to_fixup_dir(wc->trans, root,
1631 path, key.objectid);
1634 if (wc->stage < LOG_WALK_REPLAY_ALL)
1637 /* these keys are simply copied */
1638 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1639 ret = overwrite_item(wc->trans, root, path,
1642 } else if (key.type == BTRFS_INODE_REF_KEY) {
1643 ret = add_inode_ref(wc->trans, root, log, path,
1645 BUG_ON(ret && ret != -ENOENT);
1646 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1647 ret = replay_one_extent(wc->trans, root, path,
1650 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1651 key.type == BTRFS_DIR_INDEX_KEY) {
1652 ret = replay_one_dir_item(wc->trans, root, path,
1657 btrfs_free_path(path);
1661 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
1662 struct btrfs_root *root,
1663 struct btrfs_path *path, int *level,
1664 struct walk_control *wc)
1670 struct extent_buffer *next;
1671 struct extent_buffer *cur;
1672 struct extent_buffer *parent;
1676 WARN_ON(*level < 0);
1677 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1679 while (*level > 0) {
1680 WARN_ON(*level < 0);
1681 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1682 cur = path->nodes[*level];
1684 if (btrfs_header_level(cur) != *level)
1687 if (path->slots[*level] >=
1688 btrfs_header_nritems(cur))
1691 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1692 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1693 blocksize = btrfs_level_size(root, *level - 1);
1695 parent = path->nodes[*level];
1696 root_owner = btrfs_header_owner(parent);
1697 root_gen = btrfs_header_generation(parent);
1699 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1701 wc->process_func(root, next, wc, ptr_gen);
1704 path->slots[*level]++;
1706 btrfs_read_buffer(next, ptr_gen);
1708 btrfs_tree_lock(next);
1709 clean_tree_block(trans, root, next);
1710 btrfs_set_lock_blocking(next);
1711 btrfs_wait_tree_block_writeback(next);
1712 btrfs_tree_unlock(next);
1714 WARN_ON(root_owner !=
1715 BTRFS_TREE_LOG_OBJECTID);
1716 ret = btrfs_free_reserved_extent(root,
1720 free_extent_buffer(next);
1723 btrfs_read_buffer(next, ptr_gen);
1725 WARN_ON(*level <= 0);
1726 if (path->nodes[*level-1])
1727 free_extent_buffer(path->nodes[*level-1]);
1728 path->nodes[*level-1] = next;
1729 *level = btrfs_header_level(next);
1730 path->slots[*level] = 0;
1733 WARN_ON(*level < 0);
1734 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1736 if (path->nodes[*level] == root->node)
1737 parent = path->nodes[*level];
1739 parent = path->nodes[*level + 1];
1741 bytenr = path->nodes[*level]->start;
1743 blocksize = btrfs_level_size(root, *level);
1744 root_owner = btrfs_header_owner(parent);
1745 root_gen = btrfs_header_generation(parent);
1747 wc->process_func(root, path->nodes[*level], wc,
1748 btrfs_header_generation(path->nodes[*level]));
1751 next = path->nodes[*level];
1752 btrfs_tree_lock(next);
1753 clean_tree_block(trans, root, next);
1754 btrfs_set_lock_blocking(next);
1755 btrfs_wait_tree_block_writeback(next);
1756 btrfs_tree_unlock(next);
1758 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1759 ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
1762 free_extent_buffer(path->nodes[*level]);
1763 path->nodes[*level] = NULL;
1770 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
1771 struct btrfs_root *root,
1772 struct btrfs_path *path, int *level,
1773 struct walk_control *wc)
1781 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1782 slot = path->slots[i];
1783 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1784 struct extent_buffer *node;
1785 node = path->nodes[i];
1788 WARN_ON(*level == 0);
1791 struct extent_buffer *parent;
1792 if (path->nodes[*level] == root->node)
1793 parent = path->nodes[*level];
1795 parent = path->nodes[*level + 1];
1797 root_owner = btrfs_header_owner(parent);
1798 root_gen = btrfs_header_generation(parent);
1799 wc->process_func(root, path->nodes[*level], wc,
1800 btrfs_header_generation(path->nodes[*level]));
1802 struct extent_buffer *next;
1804 next = path->nodes[*level];
1806 btrfs_tree_lock(next);
1807 clean_tree_block(trans, root, next);
1808 btrfs_set_lock_blocking(next);
1809 btrfs_wait_tree_block_writeback(next);
1810 btrfs_tree_unlock(next);
1812 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1813 ret = btrfs_free_reserved_extent(root,
1814 path->nodes[*level]->start,
1815 path->nodes[*level]->len);
1818 free_extent_buffer(path->nodes[*level]);
1819 path->nodes[*level] = NULL;
1827 * drop the reference count on the tree rooted at 'snap'. This traverses
1828 * the tree freeing any blocks that have a ref count of zero after being
1831 static int walk_log_tree(struct btrfs_trans_handle *trans,
1832 struct btrfs_root *log, struct walk_control *wc)
1837 struct btrfs_path *path;
1841 path = btrfs_alloc_path();
1844 level = btrfs_header_level(log->node);
1846 path->nodes[level] = log->node;
1847 extent_buffer_get(log->node);
1848 path->slots[level] = 0;
1851 wret = walk_down_log_tree(trans, log, path, &level, wc);
1857 wret = walk_up_log_tree(trans, log, path, &level, wc);
1864 /* was the root node processed? if not, catch it here */
1865 if (path->nodes[orig_level]) {
1866 wc->process_func(log, path->nodes[orig_level], wc,
1867 btrfs_header_generation(path->nodes[orig_level]));
1869 struct extent_buffer *next;
1871 next = path->nodes[orig_level];
1873 btrfs_tree_lock(next);
1874 clean_tree_block(trans, log, next);
1875 btrfs_set_lock_blocking(next);
1876 btrfs_wait_tree_block_writeback(next);
1877 btrfs_tree_unlock(next);
1879 WARN_ON(log->root_key.objectid !=
1880 BTRFS_TREE_LOG_OBJECTID);
1881 ret = btrfs_free_reserved_extent(log, next->start,
1887 for (i = 0; i <= orig_level; i++) {
1888 if (path->nodes[i]) {
1889 free_extent_buffer(path->nodes[i]);
1890 path->nodes[i] = NULL;
1893 btrfs_free_path(path);
1898 * helper function to update the item for a given subvolumes log root
1899 * in the tree of log roots
1901 static int update_log_root(struct btrfs_trans_handle *trans,
1902 struct btrfs_root *log)
1906 if (log->log_transid == 1) {
1907 /* insert root item on the first sync */
1908 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1909 &log->root_key, &log->root_item);
1911 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1912 &log->root_key, &log->root_item);
1917 static int wait_log_commit(struct btrfs_trans_handle *trans,
1918 struct btrfs_root *root, unsigned long transid)
1921 int index = transid % 2;
1924 * we only allow two pending log transactions at a time,
1925 * so we know that if ours is more than 2 older than the
1926 * current transaction, we're done
1929 prepare_to_wait(&root->log_commit_wait[index],
1930 &wait, TASK_UNINTERRUPTIBLE);
1931 mutex_unlock(&root->log_mutex);
1933 if (root->fs_info->last_trans_log_full_commit !=
1934 trans->transid && root->log_transid < transid + 2 &&
1935 atomic_read(&root->log_commit[index]))
1938 finish_wait(&root->log_commit_wait[index], &wait);
1939 mutex_lock(&root->log_mutex);
1940 } while (root->log_transid < transid + 2 &&
1941 atomic_read(&root->log_commit[index]));
1945 static int wait_for_writer(struct btrfs_trans_handle *trans,
1946 struct btrfs_root *root)
1949 while (atomic_read(&root->log_writers)) {
1950 prepare_to_wait(&root->log_writer_wait,
1951 &wait, TASK_UNINTERRUPTIBLE);
1952 mutex_unlock(&root->log_mutex);
1953 if (root->fs_info->last_trans_log_full_commit !=
1954 trans->transid && atomic_read(&root->log_writers))
1956 mutex_lock(&root->log_mutex);
1957 finish_wait(&root->log_writer_wait, &wait);
1963 * btrfs_sync_log does sends a given tree log down to the disk and
1964 * updates the super blocks to record it. When this call is done,
1965 * you know that any inodes previously logged are safely on disk only
1968 * Any other return value means you need to call btrfs_commit_transaction.
1969 * Some of the edge cases for fsyncing directories that have had unlinks
1970 * or renames done in the past mean that sometimes the only safe
1971 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1972 * that has happened.
1974 int btrfs_sync_log(struct btrfs_trans_handle *trans,
1975 struct btrfs_root *root)
1981 struct btrfs_root *log = root->log_root;
1982 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
1983 unsigned long log_transid = 0;
1985 mutex_lock(&root->log_mutex);
1986 index1 = root->log_transid % 2;
1987 if (atomic_read(&root->log_commit[index1])) {
1988 wait_log_commit(trans, root, root->log_transid);
1989 mutex_unlock(&root->log_mutex);
1992 atomic_set(&root->log_commit[index1], 1);
1994 /* wait for previous tree log sync to complete */
1995 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
1996 wait_log_commit(trans, root, root->log_transid - 1);
1999 unsigned long batch = root->log_batch;
2000 if (root->log_multiple_pids) {
2001 mutex_unlock(&root->log_mutex);
2002 schedule_timeout_uninterruptible(1);
2003 mutex_lock(&root->log_mutex);
2005 wait_for_writer(trans, root);
2006 if (batch == root->log_batch)
2010 /* bail out if we need to do a full commit */
2011 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2013 mutex_unlock(&root->log_mutex);
2017 log_transid = root->log_transid;
2018 if (log_transid % 2 == 0)
2019 mark = EXTENT_DIRTY;
2023 /* we start IO on all the marked extents here, but we don't actually
2024 * wait for them until later.
2026 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2029 btrfs_set_root_node(&log->root_item, log->node);
2031 root->log_batch = 0;
2032 root->log_transid++;
2033 log->log_transid = root->log_transid;
2034 root->log_start_pid = 0;
2037 * IO has been started, blocks of the log tree have WRITTEN flag set
2038 * in their headers. new modifications of the log will be written to
2039 * new positions. so it's safe to allow log writers to go in.
2041 mutex_unlock(&root->log_mutex);
2043 mutex_lock(&log_root_tree->log_mutex);
2044 log_root_tree->log_batch++;
2045 atomic_inc(&log_root_tree->log_writers);
2046 mutex_unlock(&log_root_tree->log_mutex);
2048 ret = update_log_root(trans, log);
2051 mutex_lock(&log_root_tree->log_mutex);
2052 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2054 if (waitqueue_active(&log_root_tree->log_writer_wait))
2055 wake_up(&log_root_tree->log_writer_wait);
2058 index2 = log_root_tree->log_transid % 2;
2059 if (atomic_read(&log_root_tree->log_commit[index2])) {
2060 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2061 wait_log_commit(trans, log_root_tree,
2062 log_root_tree->log_transid);
2063 mutex_unlock(&log_root_tree->log_mutex);
2066 atomic_set(&log_root_tree->log_commit[index2], 1);
2068 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2069 wait_log_commit(trans, log_root_tree,
2070 log_root_tree->log_transid - 1);
2073 wait_for_writer(trans, log_root_tree);
2076 * now that we've moved on to the tree of log tree roots,
2077 * check the full commit flag again
2079 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2080 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2081 mutex_unlock(&log_root_tree->log_mutex);
2083 goto out_wake_log_root;
2086 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2087 &log_root_tree->dirty_log_pages,
2088 EXTENT_DIRTY | EXTENT_NEW);
2090 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2092 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
2093 log_root_tree->node->start);
2094 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
2095 btrfs_header_level(log_root_tree->node));
2097 log_root_tree->log_batch = 0;
2098 log_root_tree->log_transid++;
2101 mutex_unlock(&log_root_tree->log_mutex);
2104 * nobody else is going to jump in and write the the ctree
2105 * super here because the log_commit atomic below is protecting
2106 * us. We must be called with a transaction handle pinning
2107 * the running transaction open, so a full commit can't hop
2108 * in and cause problems either.
2110 write_ctree_super(trans, root->fs_info->tree_root, 1);
2113 mutex_lock(&root->log_mutex);
2114 if (root->last_log_commit < log_transid)
2115 root->last_log_commit = log_transid;
2116 mutex_unlock(&root->log_mutex);
2119 atomic_set(&log_root_tree->log_commit[index2], 0);
2121 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2122 wake_up(&log_root_tree->log_commit_wait[index2]);
2124 atomic_set(&root->log_commit[index1], 0);
2126 if (waitqueue_active(&root->log_commit_wait[index1]))
2127 wake_up(&root->log_commit_wait[index1]);
2132 * free all the extents used by the tree log. This should be called
2133 * at commit time of the full transaction
2135 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2138 struct btrfs_root *log;
2142 struct walk_control wc = {
2144 .process_func = process_one_buffer
2147 if (!root->log_root || root->fs_info->log_root_recovering)
2150 log = root->log_root;
2151 ret = walk_log_tree(trans, log, &wc);
2155 ret = find_first_extent_bit(&log->dirty_log_pages,
2156 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
2160 clear_extent_bits(&log->dirty_log_pages, start, end,
2161 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2164 if (log->log_transid > 0) {
2165 ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
2169 root->log_root = NULL;
2170 free_extent_buffer(log->node);
2176 * If both a file and directory are logged, and unlinks or renames are
2177 * mixed in, we have a few interesting corners:
2179 * create file X in dir Y
2180 * link file X to X.link in dir Y
2182 * unlink file X but leave X.link
2185 * After a crash we would expect only X.link to exist. But file X
2186 * didn't get fsync'd again so the log has back refs for X and X.link.
2188 * We solve this by removing directory entries and inode backrefs from the
2189 * log when a file that was logged in the current transaction is
2190 * unlinked. Any later fsync will include the updated log entries, and
2191 * we'll be able to reconstruct the proper directory items from backrefs.
2193 * This optimizations allows us to avoid relogging the entire inode
2194 * or the entire directory.
2196 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2197 struct btrfs_root *root,
2198 const char *name, int name_len,
2199 struct inode *dir, u64 index)
2201 struct btrfs_root *log;
2202 struct btrfs_dir_item *di;
2203 struct btrfs_path *path;
2207 if (BTRFS_I(dir)->logged_trans < trans->transid)
2210 ret = join_running_log_trans(root);
2214 mutex_lock(&BTRFS_I(dir)->log_mutex);
2216 log = root->log_root;
2217 path = btrfs_alloc_path();
2218 di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2219 name, name_len, -1);
2220 if (di && !IS_ERR(di)) {
2221 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2222 bytes_del += name_len;
2225 btrfs_release_path(log, path);
2226 di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2227 index, name, name_len, -1);
2228 if (di && !IS_ERR(di)) {
2229 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2230 bytes_del += name_len;
2234 /* update the directory size in the log to reflect the names
2238 struct btrfs_key key;
2240 key.objectid = dir->i_ino;
2242 key.type = BTRFS_INODE_ITEM_KEY;
2243 btrfs_release_path(log, path);
2245 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2247 struct btrfs_inode_item *item;
2250 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2251 struct btrfs_inode_item);
2252 i_size = btrfs_inode_size(path->nodes[0], item);
2253 if (i_size > bytes_del)
2254 i_size -= bytes_del;
2257 btrfs_set_inode_size(path->nodes[0], item, i_size);
2258 btrfs_mark_buffer_dirty(path->nodes[0]);
2261 btrfs_release_path(log, path);
2264 btrfs_free_path(path);
2265 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2266 btrfs_end_log_trans(root);
2271 /* see comments for btrfs_del_dir_entries_in_log */
2272 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2273 struct btrfs_root *root,
2274 const char *name, int name_len,
2275 struct inode *inode, u64 dirid)
2277 struct btrfs_root *log;
2281 if (BTRFS_I(inode)->logged_trans < trans->transid)
2284 ret = join_running_log_trans(root);
2287 log = root->log_root;
2288 mutex_lock(&BTRFS_I(inode)->log_mutex);
2290 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2292 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2293 btrfs_end_log_trans(root);
2299 * creates a range item in the log for 'dirid'. first_offset and
2300 * last_offset tell us which parts of the key space the log should
2301 * be considered authoritative for.
2303 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2304 struct btrfs_root *log,
2305 struct btrfs_path *path,
2306 int key_type, u64 dirid,
2307 u64 first_offset, u64 last_offset)
2310 struct btrfs_key key;
2311 struct btrfs_dir_log_item *item;
2313 key.objectid = dirid;
2314 key.offset = first_offset;
2315 if (key_type == BTRFS_DIR_ITEM_KEY)
2316 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2318 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2319 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2322 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2323 struct btrfs_dir_log_item);
2324 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2325 btrfs_mark_buffer_dirty(path->nodes[0]);
2326 btrfs_release_path(log, path);
2331 * log all the items included in the current transaction for a given
2332 * directory. This also creates the range items in the log tree required
2333 * to replay anything deleted before the fsync
2335 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2336 struct btrfs_root *root, struct inode *inode,
2337 struct btrfs_path *path,
2338 struct btrfs_path *dst_path, int key_type,
2339 u64 min_offset, u64 *last_offset_ret)
2341 struct btrfs_key min_key;
2342 struct btrfs_key max_key;
2343 struct btrfs_root *log = root->log_root;
2344 struct extent_buffer *src;
2348 u64 first_offset = min_offset;
2349 u64 last_offset = (u64)-1;
2351 log = root->log_root;
2352 max_key.objectid = inode->i_ino;
2353 max_key.offset = (u64)-1;
2354 max_key.type = key_type;
2356 min_key.objectid = inode->i_ino;
2357 min_key.type = key_type;
2358 min_key.offset = min_offset;
2360 path->keep_locks = 1;
2362 ret = btrfs_search_forward(root, &min_key, &max_key,
2363 path, 0, trans->transid);
2366 * we didn't find anything from this transaction, see if there
2367 * is anything at all
2369 if (ret != 0 || min_key.objectid != inode->i_ino ||
2370 min_key.type != key_type) {
2371 min_key.objectid = inode->i_ino;
2372 min_key.type = key_type;
2373 min_key.offset = (u64)-1;
2374 btrfs_release_path(root, path);
2375 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2377 btrfs_release_path(root, path);
2380 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2382 /* if ret == 0 there are items for this type,
2383 * create a range to tell us the last key of this type.
2384 * otherwise, there are no items in this directory after
2385 * *min_offset, and we create a range to indicate that.
2388 struct btrfs_key tmp;
2389 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2391 if (key_type == tmp.type)
2392 first_offset = max(min_offset, tmp.offset) + 1;
2397 /* go backward to find any previous key */
2398 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2400 struct btrfs_key tmp;
2401 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2402 if (key_type == tmp.type) {
2403 first_offset = tmp.offset;
2404 ret = overwrite_item(trans, log, dst_path,
2405 path->nodes[0], path->slots[0],
2409 btrfs_release_path(root, path);
2411 /* find the first key from this transaction again */
2412 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2419 * we have a block from this transaction, log every item in it
2420 * from our directory
2423 struct btrfs_key tmp;
2424 src = path->nodes[0];
2425 nritems = btrfs_header_nritems(src);
2426 for (i = path->slots[0]; i < nritems; i++) {
2427 btrfs_item_key_to_cpu(src, &min_key, i);
2429 if (min_key.objectid != inode->i_ino ||
2430 min_key.type != key_type)
2432 ret = overwrite_item(trans, log, dst_path, src, i,
2436 path->slots[0] = nritems;
2439 * look ahead to the next item and see if it is also
2440 * from this directory and from this transaction
2442 ret = btrfs_next_leaf(root, path);
2444 last_offset = (u64)-1;
2447 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2448 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2449 last_offset = (u64)-1;
2452 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2453 ret = overwrite_item(trans, log, dst_path,
2454 path->nodes[0], path->slots[0],
2458 last_offset = tmp.offset;
2463 *last_offset_ret = last_offset;
2464 btrfs_release_path(root, path);
2465 btrfs_release_path(log, dst_path);
2467 /* insert the log range keys to indicate where the log is valid */
2468 ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
2469 first_offset, last_offset);
2475 * logging directories is very similar to logging inodes, We find all the items
2476 * from the current transaction and write them to the log.
2478 * The recovery code scans the directory in the subvolume, and if it finds a
2479 * key in the range logged that is not present in the log tree, then it means
2480 * that dir entry was unlinked during the transaction.
2482 * In order for that scan to work, we must include one key smaller than
2483 * the smallest logged by this transaction and one key larger than the largest
2484 * key logged by this transaction.
2486 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2487 struct btrfs_root *root, struct inode *inode,
2488 struct btrfs_path *path,
2489 struct btrfs_path *dst_path)
2494 int key_type = BTRFS_DIR_ITEM_KEY;
2500 ret = log_dir_items(trans, root, inode, path,
2501 dst_path, key_type, min_key,
2504 if (max_key == (u64)-1)
2506 min_key = max_key + 1;
2509 if (key_type == BTRFS_DIR_ITEM_KEY) {
2510 key_type = BTRFS_DIR_INDEX_KEY;
2517 * a helper function to drop items from the log before we relog an
2518 * inode. max_key_type indicates the highest item type to remove.
2519 * This cannot be run for file data extents because it does not
2520 * free the extents they point to.
2522 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2523 struct btrfs_root *log,
2524 struct btrfs_path *path,
2525 u64 objectid, int max_key_type)
2528 struct btrfs_key key;
2529 struct btrfs_key found_key;
2531 key.objectid = objectid;
2532 key.type = max_key_type;
2533 key.offset = (u64)-1;
2536 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2541 if (path->slots[0] == 0)
2545 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2548 if (found_key.objectid != objectid)
2551 ret = btrfs_del_item(trans, log, path);
2553 btrfs_release_path(log, path);
2555 btrfs_release_path(log, path);
2559 static noinline int copy_items(struct btrfs_trans_handle *trans,
2560 struct btrfs_root *log,
2561 struct btrfs_path *dst_path,
2562 struct extent_buffer *src,
2563 int start_slot, int nr, int inode_only)
2565 unsigned long src_offset;
2566 unsigned long dst_offset;
2567 struct btrfs_file_extent_item *extent;
2568 struct btrfs_inode_item *inode_item;
2570 struct btrfs_key *ins_keys;
2574 struct list_head ordered_sums;
2576 INIT_LIST_HEAD(&ordered_sums);
2578 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2579 nr * sizeof(u32), GFP_NOFS);
2580 ins_sizes = (u32 *)ins_data;
2581 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2583 for (i = 0; i < nr; i++) {
2584 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2585 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2587 ret = btrfs_insert_empty_items(trans, log, dst_path,
2588 ins_keys, ins_sizes, nr);
2591 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
2592 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2593 dst_path->slots[0]);
2595 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2597 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2598 src_offset, ins_sizes[i]);
2600 if (inode_only == LOG_INODE_EXISTS &&
2601 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2602 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2604 struct btrfs_inode_item);
2605 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2607 /* set the generation to zero so the recover code
2608 * can tell the difference between an logging
2609 * just to say 'this inode exists' and a logging
2610 * to say 'update this inode with these values'
2612 btrfs_set_inode_generation(dst_path->nodes[0],
2615 /* take a reference on file data extents so that truncates
2616 * or deletes of this inode don't have to relog the inode
2619 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2621 extent = btrfs_item_ptr(src, start_slot + i,
2622 struct btrfs_file_extent_item);
2624 found_type = btrfs_file_extent_type(src, extent);
2625 if (found_type == BTRFS_FILE_EXTENT_REG ||
2626 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
2628 ds = btrfs_file_extent_disk_bytenr(src,
2630 /* ds == 0 is a hole */
2634 dl = btrfs_file_extent_disk_num_bytes(src,
2636 cs = btrfs_file_extent_offset(src, extent);
2637 cl = btrfs_file_extent_num_bytes(src,
2639 if (btrfs_file_extent_compression(src,
2645 ret = btrfs_lookup_csums_range(
2646 log->fs_info->csum_root,
2647 ds + cs, ds + cs + cl - 1,
2654 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2655 btrfs_release_path(log, dst_path);
2659 * we have to do this after the loop above to avoid changing the
2660 * log tree while trying to change the log tree.
2662 while (!list_empty(&ordered_sums)) {
2663 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2664 struct btrfs_ordered_sum,
2666 ret = btrfs_csum_file_blocks(trans, log, sums);
2668 list_del(&sums->list);
2674 /* log a single inode in the tree log.
2675 * At least one parent directory for this inode must exist in the tree
2676 * or be logged already.
2678 * Any items from this inode changed by the current transaction are copied
2679 * to the log tree. An extra reference is taken on any extents in this
2680 * file, allowing us to avoid a whole pile of corner cases around logging
2681 * blocks that have been removed from the tree.
2683 * See LOG_INODE_ALL and related defines for a description of what inode_only
2686 * This handles both files and directories.
2688 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
2689 struct btrfs_root *root, struct inode *inode,
2692 struct btrfs_path *path;
2693 struct btrfs_path *dst_path;
2694 struct btrfs_key min_key;
2695 struct btrfs_key max_key;
2696 struct btrfs_root *log = root->log_root;
2697 struct extent_buffer *src = NULL;
2701 int ins_start_slot = 0;
2704 log = root->log_root;
2706 path = btrfs_alloc_path();
2707 dst_path = btrfs_alloc_path();
2709 min_key.objectid = inode->i_ino;
2710 min_key.type = BTRFS_INODE_ITEM_KEY;
2713 max_key.objectid = inode->i_ino;
2715 /* today the code can only do partial logging of directories */
2716 if (!S_ISDIR(inode->i_mode))
2717 inode_only = LOG_INODE_ALL;
2719 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2720 max_key.type = BTRFS_XATTR_ITEM_KEY;
2722 max_key.type = (u8)-1;
2723 max_key.offset = (u64)-1;
2725 mutex_lock(&BTRFS_I(inode)->log_mutex);
2728 * a brute force approach to making sure we get the most uptodate
2729 * copies of everything.
2731 if (S_ISDIR(inode->i_mode)) {
2732 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2734 if (inode_only == LOG_INODE_EXISTS)
2735 max_key_type = BTRFS_XATTR_ITEM_KEY;
2736 ret = drop_objectid_items(trans, log, path,
2737 inode->i_ino, max_key_type);
2739 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2742 path->keep_locks = 1;
2746 ret = btrfs_search_forward(root, &min_key, &max_key,
2747 path, 0, trans->transid);
2751 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2752 if (min_key.objectid != inode->i_ino)
2754 if (min_key.type > max_key.type)
2757 src = path->nodes[0];
2758 size = btrfs_item_size_nr(src, path->slots[0]);
2759 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2762 } else if (!ins_nr) {
2763 ins_start_slot = path->slots[0];
2768 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2769 ins_nr, inode_only);
2772 ins_start_slot = path->slots[0];
2775 nritems = btrfs_header_nritems(path->nodes[0]);
2777 if (path->slots[0] < nritems) {
2778 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2783 ret = copy_items(trans, log, dst_path, src,
2785 ins_nr, inode_only);
2789 btrfs_release_path(root, path);
2791 if (min_key.offset < (u64)-1)
2793 else if (min_key.type < (u8)-1)
2795 else if (min_key.objectid < (u64)-1)
2801 ret = copy_items(trans, log, dst_path, src,
2803 ins_nr, inode_only);
2808 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2809 btrfs_release_path(root, path);
2810 btrfs_release_path(log, dst_path);
2811 ret = log_directory_changes(trans, root, inode, path, dst_path);
2814 BTRFS_I(inode)->logged_trans = trans->transid;
2815 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2817 btrfs_free_path(path);
2818 btrfs_free_path(dst_path);
2823 * follow the dentry parent pointers up the chain and see if any
2824 * of the directories in it require a full commit before they can
2825 * be logged. Returns zero if nothing special needs to be done or 1 if
2826 * a full commit is required.
2828 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2829 struct inode *inode,
2830 struct dentry *parent,
2831 struct super_block *sb,
2835 struct btrfs_root *root;
2838 * for regular files, if its inode is already on disk, we don't
2839 * have to worry about the parents at all. This is because
2840 * we can use the last_unlink_trans field to record renames
2841 * and other fun in this file.
2843 if (S_ISREG(inode->i_mode) &&
2844 BTRFS_I(inode)->generation <= last_committed &&
2845 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2848 if (!S_ISDIR(inode->i_mode)) {
2849 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2851 inode = parent->d_inode;
2855 BTRFS_I(inode)->logged_trans = trans->transid;
2858 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2859 root = BTRFS_I(inode)->root;
2862 * make sure any commits to the log are forced
2863 * to be full commits
2865 root->fs_info->last_trans_log_full_commit =
2871 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2874 if (IS_ROOT(parent))
2877 parent = parent->d_parent;
2878 inode = parent->d_inode;
2885 static int inode_in_log(struct btrfs_trans_handle *trans,
2886 struct inode *inode)
2888 struct btrfs_root *root = BTRFS_I(inode)->root;
2891 mutex_lock(&root->log_mutex);
2892 if (BTRFS_I(inode)->logged_trans == trans->transid &&
2893 BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
2895 mutex_unlock(&root->log_mutex);
2901 * helper function around btrfs_log_inode to make sure newly created
2902 * parent directories also end up in the log. A minimal inode and backref
2903 * only logging is done of any parent directories that are older than
2904 * the last committed transaction
2906 int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2907 struct btrfs_root *root, struct inode *inode,
2908 struct dentry *parent, int exists_only)
2910 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
2911 struct super_block *sb;
2913 u64 last_committed = root->fs_info->last_trans_committed;
2917 if (btrfs_test_opt(root, NOTREELOG)) {
2922 if (root->fs_info->last_trans_log_full_commit >
2923 root->fs_info->last_trans_committed) {
2928 if (root != BTRFS_I(inode)->root ||
2929 btrfs_root_refs(&root->root_item) == 0) {
2934 ret = check_parent_dirs_for_sync(trans, inode, parent,
2935 sb, last_committed);
2939 if (inode_in_log(trans, inode)) {
2940 ret = BTRFS_NO_LOG_SYNC;
2944 start_log_trans(trans, root);
2946 ret = btrfs_log_inode(trans, root, inode, inode_only);
2950 * for regular files, if its inode is already on disk, we don't
2951 * have to worry about the parents at all. This is because
2952 * we can use the last_unlink_trans field to record renames
2953 * and other fun in this file.
2955 if (S_ISREG(inode->i_mode) &&
2956 BTRFS_I(inode)->generation <= last_committed &&
2957 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2960 inode_only = LOG_INODE_EXISTS;
2962 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2965 inode = parent->d_inode;
2966 if (root != BTRFS_I(inode)->root)
2969 if (BTRFS_I(inode)->generation >
2970 root->fs_info->last_trans_committed) {
2971 ret = btrfs_log_inode(trans, root, inode, inode_only);
2974 if (IS_ROOT(parent))
2977 parent = parent->d_parent;
2981 btrfs_end_log_trans(root);
2987 * it is not safe to log dentry if the chunk root has added new
2988 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
2989 * If this returns 1, you must commit the transaction to safely get your
2992 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
2993 struct btrfs_root *root, struct dentry *dentry)
2995 return btrfs_log_inode_parent(trans, root, dentry->d_inode,
2996 dentry->d_parent, 0);
3000 * should be called during mount to recover any replay any log trees
3003 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3006 struct btrfs_path *path;
3007 struct btrfs_trans_handle *trans;
3008 struct btrfs_key key;
3009 struct btrfs_key found_key;
3010 struct btrfs_key tmp_key;
3011 struct btrfs_root *log;
3012 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3013 struct walk_control wc = {
3014 .process_func = process_one_buffer,
3018 fs_info->log_root_recovering = 1;
3019 path = btrfs_alloc_path();
3022 trans = btrfs_start_transaction(fs_info->tree_root, 1);
3027 walk_log_tree(trans, log_root_tree, &wc);
3030 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3031 key.offset = (u64)-1;
3032 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3035 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3039 if (path->slots[0] == 0)
3043 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3045 btrfs_release_path(log_root_tree, path);
3046 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3049 log = btrfs_read_fs_root_no_radix(log_root_tree,
3054 tmp_key.objectid = found_key.offset;
3055 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3056 tmp_key.offset = (u64)-1;
3058 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
3059 BUG_ON(!wc.replay_dest);
3061 wc.replay_dest->log_root = log;
3062 btrfs_record_root_in_trans(trans, wc.replay_dest);
3063 ret = walk_log_tree(trans, log, &wc);
3066 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3067 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3072 key.offset = found_key.offset - 1;
3073 wc.replay_dest->log_root = NULL;
3074 free_extent_buffer(log->node);
3075 free_extent_buffer(log->commit_root);
3078 if (found_key.offset == 0)
3081 btrfs_release_path(log_root_tree, path);
3083 /* step one is to pin it all, step two is to replay just inodes */
3086 wc.process_func = replay_one_buffer;
3087 wc.stage = LOG_WALK_REPLAY_INODES;
3090 /* step three is to replay everything */
3091 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3096 btrfs_free_path(path);
3098 free_extent_buffer(log_root_tree->node);
3099 log_root_tree->log_root = NULL;
3100 fs_info->log_root_recovering = 0;
3102 /* step 4: commit the transaction, which also unpins the blocks */
3103 btrfs_commit_transaction(trans, fs_info->tree_root);
3105 kfree(log_root_tree);
3110 * there are some corner cases where we want to force a full
3111 * commit instead of allowing a directory to be logged.
3113 * They revolve around files there were unlinked from the directory, and
3114 * this function updates the parent directory so that a full commit is
3115 * properly done if it is fsync'd later after the unlinks are done.
3117 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3118 struct inode *dir, struct inode *inode,
3122 * when we're logging a file, if it hasn't been renamed
3123 * or unlinked, and its inode is fully committed on disk,
3124 * we don't have to worry about walking up the directory chain
3125 * to log its parents.
3127 * So, we use the last_unlink_trans field to put this transid
3128 * into the file. When the file is logged we check it and
3129 * don't log the parents if the file is fully on disk.
3131 if (S_ISREG(inode->i_mode))
3132 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3135 * if this directory was already logged any new
3136 * names for this file/dir will get recorded
3139 if (BTRFS_I(dir)->logged_trans == trans->transid)
3143 * if the inode we're about to unlink was logged,
3144 * the log will be properly updated for any new names
3146 if (BTRFS_I(inode)->logged_trans == trans->transid)
3150 * when renaming files across directories, if the directory
3151 * there we're unlinking from gets fsync'd later on, there's
3152 * no way to find the destination directory later and fsync it
3153 * properly. So, we have to be conservative and force commits
3154 * so the new name gets discovered.
3159 /* we can safely do the unlink without any special recording */
3163 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3167 * Call this after adding a new name for a file and it will properly
3168 * update the log to reflect the new name.
3170 * It will return zero if all goes well, and it will return 1 if a
3171 * full transaction commit is required.
3173 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3174 struct inode *inode, struct inode *old_dir,
3175 struct dentry *parent)
3177 struct btrfs_root * root = BTRFS_I(inode)->root;
3180 * this will force the logging code to walk the dentry chain
3183 if (S_ISREG(inode->i_mode))
3184 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3187 * if this inode hasn't been logged and directory we're renaming it
3188 * from hasn't been logged, we don't need to log it
3190 if (BTRFS_I(inode)->logged_trans <=
3191 root->fs_info->last_trans_committed &&
3192 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3193 root->fs_info->last_trans_committed))
3196 return btrfs_log_inode_parent(trans, root, inode, parent, 1);