2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/list_sort.h>
23 #include "transaction.h"
26 #include "print-tree.h"
32 /* magic values for the inode_only field in btrfs_log_inode:
34 * LOG_INODE_ALL means to log everything
35 * LOG_INODE_EXISTS means to log just enough to recreate the inode
38 #define LOG_INODE_ALL 0
39 #define LOG_INODE_EXISTS 1
42 * directory trouble cases
44 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
45 * log, we must force a full commit before doing an fsync of the directory
46 * where the unlink was done.
47 * ---> record transid of last unlink/rename per directory
51 * rename foo/some_dir foo2/some_dir
53 * fsync foo/some_dir/some_file
55 * The fsync above will unlink the original some_dir without recording
56 * it in its new location (foo2). After a crash, some_dir will be gone
57 * unless the fsync of some_file forces a full commit
59 * 2) we must log any new names for any file or dir that is in the fsync
60 * log. ---> check inode while renaming/linking.
62 * 2a) we must log any new names for any file or dir during rename
63 * when the directory they are being removed from was logged.
64 * ---> check inode and old parent dir during rename
66 * 2a is actually the more important variant. With the extra logging
67 * a crash might unlink the old name without recreating the new one
69 * 3) after a crash, we must go through any directories with a link count
70 * of zero and redo the rm -rf
77 * The directory f1 was fully removed from the FS, but fsync was never
78 * called on f1, only its parent dir. After a crash the rm -rf must
79 * be replayed. This must be able to recurse down the entire
80 * directory tree. The inode link count fixup code takes care of the
85 * stages for the tree walking. The first
86 * stage (0) is to only pin down the blocks we find
87 * the second stage (1) is to make sure that all the inodes
88 * we find in the log are created in the subvolume.
90 * The last stage is to deal with directories and links and extents
91 * and all the other fun semantics
93 #define LOG_WALK_PIN_ONLY 0
94 #define LOG_WALK_REPLAY_INODES 1
95 #define LOG_WALK_REPLAY_ALL 2
97 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode,
100 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root,
102 struct btrfs_path *path, u64 objectid);
103 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
104 struct btrfs_root *root,
105 struct btrfs_root *log,
106 struct btrfs_path *path,
107 u64 dirid, int del_all);
110 * tree logging is a special write ahead log used to make sure that
111 * fsyncs and O_SYNCs can happen without doing full tree commits.
113 * Full tree commits are expensive because they require commonly
114 * modified blocks to be recowed, creating many dirty pages in the
115 * extent tree an 4x-6x higher write load than ext3.
117 * Instead of doing a tree commit on every fsync, we use the
118 * key ranges and transaction ids to find items for a given file or directory
119 * that have changed in this transaction. Those items are copied into
120 * a special tree (one per subvolume root), that tree is written to disk
121 * and then the fsync is considered complete.
123 * After a crash, items are copied out of the log-tree back into the
124 * subvolume tree. Any file data extents found are recorded in the extent
125 * allocation tree, and the log-tree freed.
127 * The log tree is read three times, once to pin down all the extents it is
128 * using in ram and once, once to create all the inodes logged in the tree
129 * and once to do all the other items.
133 * start a sub transaction and setup the log tree
134 * this increments the log tree writer count to make the people
135 * syncing the tree wait for us to finish
137 static int start_log_trans(struct btrfs_trans_handle *trans,
138 struct btrfs_root *root)
143 mutex_lock(&root->log_mutex);
144 if (root->log_root) {
145 if (!root->log_start_pid) {
146 root->log_start_pid = current->pid;
147 root->log_multiple_pids = false;
148 } else if (root->log_start_pid != current->pid) {
149 root->log_multiple_pids = true;
152 atomic_inc(&root->log_batch);
153 atomic_inc(&root->log_writers);
154 mutex_unlock(&root->log_mutex);
157 root->log_multiple_pids = false;
158 root->log_start_pid = current->pid;
159 mutex_lock(&root->fs_info->tree_log_mutex);
160 if (!root->fs_info->log_root_tree) {
161 ret = btrfs_init_log_root_tree(trans, root->fs_info);
165 if (err == 0 && !root->log_root) {
166 ret = btrfs_add_log_tree(trans, root);
170 mutex_unlock(&root->fs_info->tree_log_mutex);
171 atomic_inc(&root->log_batch);
172 atomic_inc(&root->log_writers);
173 mutex_unlock(&root->log_mutex);
178 * returns 0 if there was a log transaction running and we were able
179 * to join, or returns -ENOENT if there were not transactions
182 static int join_running_log_trans(struct btrfs_root *root)
190 mutex_lock(&root->log_mutex);
191 if (root->log_root) {
193 atomic_inc(&root->log_writers);
195 mutex_unlock(&root->log_mutex);
200 * This either makes the current running log transaction wait
201 * until you call btrfs_end_log_trans() or it makes any future
202 * log transactions wait until you call btrfs_end_log_trans()
204 int btrfs_pin_log_trans(struct btrfs_root *root)
208 mutex_lock(&root->log_mutex);
209 atomic_inc(&root->log_writers);
210 mutex_unlock(&root->log_mutex);
215 * indicate we're done making changes to the log tree
216 * and wake up anyone waiting to do a sync
218 void btrfs_end_log_trans(struct btrfs_root *root)
220 if (atomic_dec_and_test(&root->log_writers)) {
222 if (waitqueue_active(&root->log_writer_wait))
223 wake_up(&root->log_writer_wait);
229 * the walk control struct is used to pass state down the chain when
230 * processing the log tree. The stage field tells us which part
231 * of the log tree processing we are currently doing. The others
232 * are state fields used for that specific part
234 struct walk_control {
235 /* should we free the extent on disk when done? This is used
236 * at transaction commit time while freeing a log tree
240 /* should we write out the extent buffer? This is used
241 * while flushing the log tree to disk during a sync
245 /* should we wait for the extent buffer io to finish? Also used
246 * while flushing the log tree to disk for a sync
250 /* pin only walk, we record which extents on disk belong to the
255 /* what stage of the replay code we're currently in */
258 /* the root we are currently replaying */
259 struct btrfs_root *replay_dest;
261 /* the trans handle for the current replay */
262 struct btrfs_trans_handle *trans;
264 /* the function that gets used to process blocks we find in the
265 * tree. Note the extent_buffer might not be up to date when it is
266 * passed in, and it must be checked or read if you need the data
269 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
270 struct walk_control *wc, u64 gen);
274 * process_func used to pin down extents, write them or wait on them
276 static int process_one_buffer(struct btrfs_root *log,
277 struct extent_buffer *eb,
278 struct walk_control *wc, u64 gen)
281 btrfs_pin_extent_for_log_replay(wc->trans,
282 log->fs_info->extent_root,
285 if (btrfs_buffer_uptodate(eb, gen, 0)) {
287 btrfs_write_tree_block(eb);
289 btrfs_wait_tree_block_writeback(eb);
295 * Item overwrite used by replay and tree logging. eb, slot and key all refer
296 * to the src data we are copying out.
298 * root is the tree we are copying into, and path is a scratch
299 * path for use in this function (it should be released on entry and
300 * will be released on exit).
302 * If the key is already in the destination tree the existing item is
303 * overwritten. If the existing item isn't big enough, it is extended.
304 * If it is too large, it is truncated.
306 * If the key isn't in the destination yet, a new item is inserted.
308 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
309 struct btrfs_root *root,
310 struct btrfs_path *path,
311 struct extent_buffer *eb, int slot,
312 struct btrfs_key *key)
316 u64 saved_i_size = 0;
317 int save_old_i_size = 0;
318 unsigned long src_ptr;
319 unsigned long dst_ptr;
320 int overwrite_root = 0;
322 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
325 item_size = btrfs_item_size_nr(eb, slot);
326 src_ptr = btrfs_item_ptr_offset(eb, slot);
328 /* look for the key in the destination tree */
329 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
333 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
335 if (dst_size != item_size)
338 if (item_size == 0) {
339 btrfs_release_path(path);
342 dst_copy = kmalloc(item_size, GFP_NOFS);
343 src_copy = kmalloc(item_size, GFP_NOFS);
344 if (!dst_copy || !src_copy) {
345 btrfs_release_path(path);
351 read_extent_buffer(eb, src_copy, src_ptr, item_size);
353 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
354 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
356 ret = memcmp(dst_copy, src_copy, item_size);
361 * they have the same contents, just return, this saves
362 * us from cowing blocks in the destination tree and doing
363 * extra writes that may not have been done by a previous
367 btrfs_release_path(path);
373 btrfs_release_path(path);
374 /* try to insert the key into the destination tree */
375 ret = btrfs_insert_empty_item(trans, root, path,
378 /* make sure any existing item is the correct size */
379 if (ret == -EEXIST) {
381 found_size = btrfs_item_size_nr(path->nodes[0],
383 if (found_size > item_size)
384 btrfs_truncate_item(trans, root, path, item_size, 1);
385 else if (found_size < item_size)
386 btrfs_extend_item(trans, root, path,
387 item_size - found_size);
391 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
394 /* don't overwrite an existing inode if the generation number
395 * was logged as zero. This is done when the tree logging code
396 * is just logging an inode to make sure it exists after recovery.
398 * Also, don't overwrite i_size on directories during replay.
399 * log replay inserts and removes directory items based on the
400 * state of the tree found in the subvolume, and i_size is modified
403 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
404 struct btrfs_inode_item *src_item;
405 struct btrfs_inode_item *dst_item;
407 src_item = (struct btrfs_inode_item *)src_ptr;
408 dst_item = (struct btrfs_inode_item *)dst_ptr;
410 if (btrfs_inode_generation(eb, src_item) == 0)
413 if (overwrite_root &&
414 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
415 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
417 saved_i_size = btrfs_inode_size(path->nodes[0],
422 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
425 if (save_old_i_size) {
426 struct btrfs_inode_item *dst_item;
427 dst_item = (struct btrfs_inode_item *)dst_ptr;
428 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
431 /* make sure the generation is filled in */
432 if (key->type == BTRFS_INODE_ITEM_KEY) {
433 struct btrfs_inode_item *dst_item;
434 dst_item = (struct btrfs_inode_item *)dst_ptr;
435 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
436 btrfs_set_inode_generation(path->nodes[0], dst_item,
441 btrfs_mark_buffer_dirty(path->nodes[0]);
442 btrfs_release_path(path);
447 * simple helper to read an inode off the disk from a given root
448 * This can only be called for subvolume roots and not for the log
450 static noinline struct inode *read_one_inode(struct btrfs_root *root,
453 struct btrfs_key key;
456 key.objectid = objectid;
457 key.type = BTRFS_INODE_ITEM_KEY;
459 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
462 } else if (is_bad_inode(inode)) {
469 /* replays a single extent in 'eb' at 'slot' with 'key' into the
470 * subvolume 'root'. path is released on entry and should be released
473 * extents in the log tree have not been allocated out of the extent
474 * tree yet. So, this completes the allocation, taking a reference
475 * as required if the extent already exists or creating a new extent
476 * if it isn't in the extent allocation tree yet.
478 * The extent is inserted into the file, dropping any existing extents
479 * from the file that overlap the new one.
481 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
482 struct btrfs_root *root,
483 struct btrfs_path *path,
484 struct extent_buffer *eb, int slot,
485 struct btrfs_key *key)
488 u64 mask = root->sectorsize - 1;
490 u64 start = key->offset;
492 struct btrfs_file_extent_item *item;
493 struct inode *inode = NULL;
497 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
498 found_type = btrfs_file_extent_type(eb, item);
500 if (found_type == BTRFS_FILE_EXTENT_REG ||
501 found_type == BTRFS_FILE_EXTENT_PREALLOC)
502 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
503 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
504 size = btrfs_file_extent_inline_len(eb, item);
505 extent_end = (start + size + mask) & ~mask;
511 inode = read_one_inode(root, key->objectid);
518 * first check to see if we already have this extent in the
519 * file. This must be done before the btrfs_drop_extents run
520 * so we don't try to drop this extent.
522 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
526 (found_type == BTRFS_FILE_EXTENT_REG ||
527 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
528 struct btrfs_file_extent_item cmp1;
529 struct btrfs_file_extent_item cmp2;
530 struct btrfs_file_extent_item *existing;
531 struct extent_buffer *leaf;
533 leaf = path->nodes[0];
534 existing = btrfs_item_ptr(leaf, path->slots[0],
535 struct btrfs_file_extent_item);
537 read_extent_buffer(eb, &cmp1, (unsigned long)item,
539 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
543 * we already have a pointer to this exact extent,
544 * we don't have to do anything
546 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
547 btrfs_release_path(path);
551 btrfs_release_path(path);
553 saved_nbytes = inode_get_bytes(inode);
554 /* drop any overlapping extents */
555 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
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, 0);
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 ret = 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);
695 btrfs_run_delayed_items(trans, root);
700 * helper function to see if a given name and sequence number found
701 * in an inode back reference are already in a directory and correctly
702 * point to this inode
704 static noinline int inode_in_dir(struct btrfs_root *root,
705 struct btrfs_path *path,
706 u64 dirid, u64 objectid, u64 index,
707 const char *name, int name_len)
709 struct btrfs_dir_item *di;
710 struct btrfs_key location;
713 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
714 index, name, name_len, 0);
715 if (di && !IS_ERR(di)) {
716 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
717 if (location.objectid != objectid)
721 btrfs_release_path(path);
723 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
724 if (di && !IS_ERR(di)) {
725 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
726 if (location.objectid != objectid)
732 btrfs_release_path(path);
737 * helper function to check a log tree for a named back reference in
738 * an inode. This is used to decide if a back reference that is
739 * found in the subvolume conflicts with what we find in the log.
741 * inode backreferences may have multiple refs in a single item,
742 * during replay we process one reference at a time, and we don't
743 * want to delete valid links to a file from the subvolume if that
744 * link is also in the log.
746 static noinline int backref_in_log(struct btrfs_root *log,
747 struct btrfs_key *key,
749 char *name, int namelen)
751 struct btrfs_path *path;
752 struct btrfs_inode_ref *ref;
754 unsigned long ptr_end;
755 unsigned long name_ptr;
761 path = btrfs_alloc_path();
765 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
769 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
771 if (key->type == BTRFS_INODE_EXTREF_KEY) {
772 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
773 name, namelen, NULL))
779 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
780 ptr_end = ptr + item_size;
781 while (ptr < ptr_end) {
782 ref = (struct btrfs_inode_ref *)ptr;
783 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
784 if (found_name_len == namelen) {
785 name_ptr = (unsigned long)(ref + 1);
786 ret = memcmp_extent_buffer(path->nodes[0], name,
793 ptr = (unsigned long)(ref + 1) + found_name_len;
796 btrfs_free_path(path);
800 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
801 struct btrfs_root *root,
802 struct btrfs_path *path,
803 struct btrfs_root *log_root,
804 struct inode *dir, struct inode *inode,
805 struct extent_buffer *eb,
806 u64 inode_objectid, u64 parent_objectid,
807 u64 ref_index, char *name, int namelen,
813 struct extent_buffer *leaf;
814 struct btrfs_dir_item *di;
815 struct btrfs_key search_key;
816 struct btrfs_inode_extref *extref;
819 /* Search old style refs */
820 search_key.objectid = inode_objectid;
821 search_key.type = BTRFS_INODE_REF_KEY;
822 search_key.offset = parent_objectid;
823 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
825 struct btrfs_inode_ref *victim_ref;
827 unsigned long ptr_end;
829 leaf = path->nodes[0];
831 /* are we trying to overwrite a back ref for the root directory
832 * if so, just jump out, we're done
834 if (search_key.objectid == search_key.offset)
837 /* check all the names in this back reference to see
838 * if they are in the log. if so, we allow them to stay
839 * otherwise they must be unlinked as a conflict
841 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
842 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
843 while (ptr < ptr_end) {
844 victim_ref = (struct btrfs_inode_ref *)ptr;
845 victim_name_len = btrfs_inode_ref_name_len(leaf,
847 victim_name = kmalloc(victim_name_len, GFP_NOFS);
848 BUG_ON(!victim_name);
850 read_extent_buffer(leaf, victim_name,
851 (unsigned long)(victim_ref + 1),
854 if (!backref_in_log(log_root, &search_key,
858 btrfs_inc_nlink(inode);
859 btrfs_release_path(path);
861 ret = btrfs_unlink_inode(trans, root, dir,
865 btrfs_run_delayed_items(trans, root);
872 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
877 * NOTE: we have searched root tree and checked the
878 * coresponding ref, it does not need to check again.
882 btrfs_release_path(path);
884 /* Same search but for extended refs */
885 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
886 inode_objectid, parent_objectid, 0,
888 if (!IS_ERR_OR_NULL(extref)) {
892 struct inode *victim_parent;
894 leaf = path->nodes[0];
896 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
897 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
899 while (cur_offset < item_size) {
900 extref = (struct btrfs_inode_extref *)base + cur_offset;
902 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
904 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
907 victim_name = kmalloc(victim_name_len, GFP_NOFS);
908 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
911 search_key.objectid = inode_objectid;
912 search_key.type = BTRFS_INODE_EXTREF_KEY;
913 search_key.offset = btrfs_extref_hash(parent_objectid,
917 if (!backref_in_log(log_root, &search_key,
918 parent_objectid, victim_name,
921 victim_parent = read_one_inode(root,
924 btrfs_inc_nlink(inode);
925 btrfs_release_path(path);
927 ret = btrfs_unlink_inode(trans, root,
932 btrfs_run_delayed_items(trans, root);
943 cur_offset += victim_name_len + sizeof(*extref);
947 btrfs_release_path(path);
949 /* look for a conflicting sequence number */
950 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
951 ref_index, name, namelen, 0);
952 if (di && !IS_ERR(di)) {
953 ret = drop_one_dir_item(trans, root, path, dir, di);
956 btrfs_release_path(path);
958 /* look for a conflicing name */
959 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
961 if (di && !IS_ERR(di)) {
962 ret = drop_one_dir_item(trans, root, path, dir, di);
965 btrfs_release_path(path);
970 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
971 u32 *namelen, char **name, u64 *index,
972 u64 *parent_objectid)
974 struct btrfs_inode_extref *extref;
976 extref = (struct btrfs_inode_extref *)ref_ptr;
978 *namelen = btrfs_inode_extref_name_len(eb, extref);
979 *name = kmalloc(*namelen, GFP_NOFS);
983 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
986 *index = btrfs_inode_extref_index(eb, extref);
988 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
993 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
994 u32 *namelen, char **name, u64 *index)
996 struct btrfs_inode_ref *ref;
998 ref = (struct btrfs_inode_ref *)ref_ptr;
1000 *namelen = btrfs_inode_ref_name_len(eb, ref);
1001 *name = kmalloc(*namelen, GFP_NOFS);
1005 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1007 *index = btrfs_inode_ref_index(eb, ref);
1013 * replay one inode back reference item found in the log tree.
1014 * eb, slot and key refer to the buffer and key found in the log tree.
1015 * root is the destination we are replaying into, and path is for temp
1016 * use by this function. (it should be released on return).
1018 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1019 struct btrfs_root *root,
1020 struct btrfs_root *log,
1021 struct btrfs_path *path,
1022 struct extent_buffer *eb, int slot,
1023 struct btrfs_key *key)
1026 struct inode *inode;
1027 unsigned long ref_ptr;
1028 unsigned long ref_end;
1032 int search_done = 0;
1033 int log_ref_ver = 0;
1034 u64 parent_objectid;
1037 int ref_struct_size;
1039 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1040 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1042 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1043 struct btrfs_inode_extref *r;
1045 ref_struct_size = sizeof(struct btrfs_inode_extref);
1047 r = (struct btrfs_inode_extref *)ref_ptr;
1048 parent_objectid = btrfs_inode_extref_parent(eb, r);
1050 ref_struct_size = sizeof(struct btrfs_inode_ref);
1051 parent_objectid = key->offset;
1053 inode_objectid = key->objectid;
1056 * it is possible that we didn't log all the parent directories
1057 * for a given inode. If we don't find the dir, just don't
1058 * copy the back ref in. The link count fixup code will take
1061 dir = read_one_inode(root, parent_objectid);
1065 inode = read_one_inode(root, inode_objectid);
1071 while (ref_ptr < ref_end) {
1073 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1074 &ref_index, &parent_objectid);
1076 * parent object can change from one array
1080 dir = read_one_inode(root, parent_objectid);
1084 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1090 /* if we already have a perfect match, we're done */
1091 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
1092 ref_index, name, namelen)) {
1094 * look for a conflicting back reference in the
1095 * metadata. if we find one we have to unlink that name
1096 * of the file before we add our new link. Later on, we
1097 * overwrite any existing back reference, and we don't
1098 * want to create dangling pointers in the directory.
1102 ret = __add_inode_ref(trans, root, path, log,
1106 ref_index, name, namelen,
1113 /* insert our name */
1114 ret = btrfs_add_link(trans, dir, inode, name, namelen,
1118 btrfs_update_inode(trans, root, inode);
1121 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1129 /* finally write the back reference in the inode */
1130 ret = overwrite_item(trans, root, path, eb, slot, key);
1134 btrfs_release_path(path);
1140 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1141 struct btrfs_root *root, u64 offset)
1144 ret = btrfs_find_orphan_item(root, offset);
1146 ret = btrfs_insert_orphan_item(trans, root, offset);
1150 static int count_inode_extrefs(struct btrfs_root *root,
1151 struct inode *inode, struct btrfs_path *path)
1155 unsigned int nlink = 0;
1158 u64 inode_objectid = btrfs_ino(inode);
1161 struct btrfs_inode_extref *extref;
1162 struct extent_buffer *leaf;
1165 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1170 leaf = path->nodes[0];
1171 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1172 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1174 while (cur_offset < item_size) {
1175 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1176 name_len = btrfs_inode_extref_name_len(leaf, extref);
1180 cur_offset += name_len + sizeof(*extref);
1184 btrfs_release_path(path);
1186 btrfs_release_path(path);
1193 static int count_inode_refs(struct btrfs_root *root,
1194 struct inode *inode, struct btrfs_path *path)
1197 struct btrfs_key key;
1198 unsigned int nlink = 0;
1200 unsigned long ptr_end;
1202 u64 ino = btrfs_ino(inode);
1205 key.type = BTRFS_INODE_REF_KEY;
1206 key.offset = (u64)-1;
1209 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1213 if (path->slots[0] == 0)
1217 btrfs_item_key_to_cpu(path->nodes[0], &key,
1219 if (key.objectid != ino ||
1220 key.type != BTRFS_INODE_REF_KEY)
1222 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1223 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1225 while (ptr < ptr_end) {
1226 struct btrfs_inode_ref *ref;
1228 ref = (struct btrfs_inode_ref *)ptr;
1229 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1231 ptr = (unsigned long)(ref + 1) + name_len;
1235 if (key.offset == 0)
1238 btrfs_release_path(path);
1240 btrfs_release_path(path);
1246 * There are a few corners where the link count of the file can't
1247 * be properly maintained during replay. So, instead of adding
1248 * lots of complexity to the log code, we just scan the backrefs
1249 * for any file that has been through replay.
1251 * The scan will update the link count on the inode to reflect the
1252 * number of back refs found. If it goes down to zero, the iput
1253 * will free the inode.
1255 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1256 struct btrfs_root *root,
1257 struct inode *inode)
1259 struct btrfs_path *path;
1262 u64 ino = btrfs_ino(inode);
1264 path = btrfs_alloc_path();
1268 ret = count_inode_refs(root, inode, path);
1274 ret = count_inode_extrefs(root, inode, path);
1285 if (nlink != inode->i_nlink) {
1286 set_nlink(inode, nlink);
1287 btrfs_update_inode(trans, root, inode);
1289 BTRFS_I(inode)->index_cnt = (u64)-1;
1291 if (inode->i_nlink == 0) {
1292 if (S_ISDIR(inode->i_mode)) {
1293 ret = replay_dir_deletes(trans, root, NULL, path,
1297 ret = insert_orphan_item(trans, root, ino);
1302 btrfs_free_path(path);
1306 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1307 struct btrfs_root *root,
1308 struct btrfs_path *path)
1311 struct btrfs_key key;
1312 struct inode *inode;
1314 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1315 key.type = BTRFS_ORPHAN_ITEM_KEY;
1316 key.offset = (u64)-1;
1318 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1323 if (path->slots[0] == 0)
1328 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1329 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1330 key.type != BTRFS_ORPHAN_ITEM_KEY)
1333 ret = btrfs_del_item(trans, root, path);
1337 btrfs_release_path(path);
1338 inode = read_one_inode(root, key.offset);
1342 ret = fixup_inode_link_count(trans, root, inode);
1348 * fixup on a directory may create new entries,
1349 * make sure we always look for the highset possible
1352 key.offset = (u64)-1;
1356 btrfs_release_path(path);
1362 * record a given inode in the fixup dir so we can check its link
1363 * count when replay is done. The link count is incremented here
1364 * so the inode won't go away until we check it
1366 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1367 struct btrfs_root *root,
1368 struct btrfs_path *path,
1371 struct btrfs_key key;
1373 struct inode *inode;
1375 inode = read_one_inode(root, objectid);
1379 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1380 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1381 key.offset = objectid;
1383 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1385 btrfs_release_path(path);
1387 btrfs_inc_nlink(inode);
1388 ret = btrfs_update_inode(trans, root, inode);
1389 } else if (ret == -EEXIST) {
1400 * when replaying the log for a directory, we only insert names
1401 * for inodes that actually exist. This means an fsync on a directory
1402 * does not implicitly fsync all the new files in it
1404 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1405 struct btrfs_root *root,
1406 struct btrfs_path *path,
1407 u64 dirid, u64 index,
1408 char *name, int name_len, u8 type,
1409 struct btrfs_key *location)
1411 struct inode *inode;
1415 inode = read_one_inode(root, location->objectid);
1419 dir = read_one_inode(root, dirid);
1424 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1426 /* FIXME, put inode into FIXUP list */
1434 * take a single entry in a log directory item and replay it into
1437 * if a conflicting item exists in the subdirectory already,
1438 * the inode it points to is unlinked and put into the link count
1441 * If a name from the log points to a file or directory that does
1442 * not exist in the FS, it is skipped. fsyncs on directories
1443 * do not force down inodes inside that directory, just changes to the
1444 * names or unlinks in a directory.
1446 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1447 struct btrfs_root *root,
1448 struct btrfs_path *path,
1449 struct extent_buffer *eb,
1450 struct btrfs_dir_item *di,
1451 struct btrfs_key *key)
1455 struct btrfs_dir_item *dst_di;
1456 struct btrfs_key found_key;
1457 struct btrfs_key log_key;
1463 dir = read_one_inode(root, key->objectid);
1467 name_len = btrfs_dir_name_len(eb, di);
1468 name = kmalloc(name_len, GFP_NOFS);
1472 log_type = btrfs_dir_type(eb, di);
1473 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1476 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1477 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1482 btrfs_release_path(path);
1484 if (key->type == BTRFS_DIR_ITEM_KEY) {
1485 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1487 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1488 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1495 if (IS_ERR_OR_NULL(dst_di)) {
1496 /* we need a sequence number to insert, so we only
1497 * do inserts for the BTRFS_DIR_INDEX_KEY types
1499 if (key->type != BTRFS_DIR_INDEX_KEY)
1504 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1505 /* the existing item matches the logged item */
1506 if (found_key.objectid == log_key.objectid &&
1507 found_key.type == log_key.type &&
1508 found_key.offset == log_key.offset &&
1509 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1514 * don't drop the conflicting directory entry if the inode
1515 * for the new entry doesn't exist
1520 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1523 if (key->type == BTRFS_DIR_INDEX_KEY)
1526 btrfs_release_path(path);
1532 btrfs_release_path(path);
1533 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1534 name, name_len, log_type, &log_key);
1536 BUG_ON(ret && ret != -ENOENT);
1541 * find all the names in a directory item and reconcile them into
1542 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1543 * one name in a directory item, but the same code gets used for
1544 * both directory index types
1546 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1547 struct btrfs_root *root,
1548 struct btrfs_path *path,
1549 struct extent_buffer *eb, int slot,
1550 struct btrfs_key *key)
1553 u32 item_size = btrfs_item_size_nr(eb, slot);
1554 struct btrfs_dir_item *di;
1557 unsigned long ptr_end;
1559 ptr = btrfs_item_ptr_offset(eb, slot);
1560 ptr_end = ptr + item_size;
1561 while (ptr < ptr_end) {
1562 di = (struct btrfs_dir_item *)ptr;
1563 if (verify_dir_item(root, eb, di))
1565 name_len = btrfs_dir_name_len(eb, di);
1566 ret = replay_one_name(trans, root, path, eb, di, key);
1568 ptr = (unsigned long)(di + 1);
1575 * directory replay has two parts. There are the standard directory
1576 * items in the log copied from the subvolume, and range items
1577 * created in the log while the subvolume was logged.
1579 * The range items tell us which parts of the key space the log
1580 * is authoritative for. During replay, if a key in the subvolume
1581 * directory is in a logged range item, but not actually in the log
1582 * that means it was deleted from the directory before the fsync
1583 * and should be removed.
1585 static noinline int find_dir_range(struct btrfs_root *root,
1586 struct btrfs_path *path,
1587 u64 dirid, int key_type,
1588 u64 *start_ret, u64 *end_ret)
1590 struct btrfs_key key;
1592 struct btrfs_dir_log_item *item;
1596 if (*start_ret == (u64)-1)
1599 key.objectid = dirid;
1600 key.type = key_type;
1601 key.offset = *start_ret;
1603 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1607 if (path->slots[0] == 0)
1612 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1614 if (key.type != key_type || key.objectid != dirid) {
1618 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1619 struct btrfs_dir_log_item);
1620 found_end = btrfs_dir_log_end(path->nodes[0], item);
1622 if (*start_ret >= key.offset && *start_ret <= found_end) {
1624 *start_ret = key.offset;
1625 *end_ret = found_end;
1630 /* check the next slot in the tree to see if it is a valid item */
1631 nritems = btrfs_header_nritems(path->nodes[0]);
1632 if (path->slots[0] >= nritems) {
1633 ret = btrfs_next_leaf(root, path);
1640 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1642 if (key.type != key_type || key.objectid != dirid) {
1646 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1647 struct btrfs_dir_log_item);
1648 found_end = btrfs_dir_log_end(path->nodes[0], item);
1649 *start_ret = key.offset;
1650 *end_ret = found_end;
1653 btrfs_release_path(path);
1658 * this looks for a given directory item in the log. If the directory
1659 * item is not in the log, the item is removed and the inode it points
1662 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1663 struct btrfs_root *root,
1664 struct btrfs_root *log,
1665 struct btrfs_path *path,
1666 struct btrfs_path *log_path,
1668 struct btrfs_key *dir_key)
1671 struct extent_buffer *eb;
1674 struct btrfs_dir_item *di;
1675 struct btrfs_dir_item *log_di;
1678 unsigned long ptr_end;
1680 struct inode *inode;
1681 struct btrfs_key location;
1684 eb = path->nodes[0];
1685 slot = path->slots[0];
1686 item_size = btrfs_item_size_nr(eb, slot);
1687 ptr = btrfs_item_ptr_offset(eb, slot);
1688 ptr_end = ptr + item_size;
1689 while (ptr < ptr_end) {
1690 di = (struct btrfs_dir_item *)ptr;
1691 if (verify_dir_item(root, eb, di)) {
1696 name_len = btrfs_dir_name_len(eb, di);
1697 name = kmalloc(name_len, GFP_NOFS);
1702 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1705 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1706 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1709 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1710 log_di = btrfs_lookup_dir_index_item(trans, log,
1716 if (IS_ERR_OR_NULL(log_di)) {
1717 btrfs_dir_item_key_to_cpu(eb, di, &location);
1718 btrfs_release_path(path);
1719 btrfs_release_path(log_path);
1720 inode = read_one_inode(root, location.objectid);
1726 ret = link_to_fixup_dir(trans, root,
1727 path, location.objectid);
1729 btrfs_inc_nlink(inode);
1730 ret = btrfs_unlink_inode(trans, root, dir, inode,
1734 btrfs_run_delayed_items(trans, root);
1739 /* there might still be more names under this key
1740 * check and repeat if required
1742 ret = btrfs_search_slot(NULL, root, dir_key, path,
1749 btrfs_release_path(log_path);
1752 ptr = (unsigned long)(di + 1);
1757 btrfs_release_path(path);
1758 btrfs_release_path(log_path);
1763 * deletion replay happens before we copy any new directory items
1764 * out of the log or out of backreferences from inodes. It
1765 * scans the log to find ranges of keys that log is authoritative for,
1766 * and then scans the directory to find items in those ranges that are
1767 * not present in the log.
1769 * Anything we don't find in the log is unlinked and removed from the
1772 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1773 struct btrfs_root *root,
1774 struct btrfs_root *log,
1775 struct btrfs_path *path,
1776 u64 dirid, int del_all)
1780 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1782 struct btrfs_key dir_key;
1783 struct btrfs_key found_key;
1784 struct btrfs_path *log_path;
1787 dir_key.objectid = dirid;
1788 dir_key.type = BTRFS_DIR_ITEM_KEY;
1789 log_path = btrfs_alloc_path();
1793 dir = read_one_inode(root, dirid);
1794 /* it isn't an error if the inode isn't there, that can happen
1795 * because we replay the deletes before we copy in the inode item
1799 btrfs_free_path(log_path);
1807 range_end = (u64)-1;
1809 ret = find_dir_range(log, path, dirid, key_type,
1810 &range_start, &range_end);
1815 dir_key.offset = range_start;
1818 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1823 nritems = btrfs_header_nritems(path->nodes[0]);
1824 if (path->slots[0] >= nritems) {
1825 ret = btrfs_next_leaf(root, path);
1829 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1831 if (found_key.objectid != dirid ||
1832 found_key.type != dir_key.type)
1835 if (found_key.offset > range_end)
1838 ret = check_item_in_log(trans, root, log, path,
1842 if (found_key.offset == (u64)-1)
1844 dir_key.offset = found_key.offset + 1;
1846 btrfs_release_path(path);
1847 if (range_end == (u64)-1)
1849 range_start = range_end + 1;
1854 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1855 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1856 dir_key.type = BTRFS_DIR_INDEX_KEY;
1857 btrfs_release_path(path);
1861 btrfs_release_path(path);
1862 btrfs_free_path(log_path);
1868 * the process_func used to replay items from the log tree. This
1869 * gets called in two different stages. The first stage just looks
1870 * for inodes and makes sure they are all copied into the subvolume.
1872 * The second stage copies all the other item types from the log into
1873 * the subvolume. The two stage approach is slower, but gets rid of
1874 * lots of complexity around inodes referencing other inodes that exist
1875 * only in the log (references come from either directory items or inode
1878 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1879 struct walk_control *wc, u64 gen)
1882 struct btrfs_path *path;
1883 struct btrfs_root *root = wc->replay_dest;
1884 struct btrfs_key key;
1889 ret = btrfs_read_buffer(eb, gen);
1893 level = btrfs_header_level(eb);
1898 path = btrfs_alloc_path();
1902 nritems = btrfs_header_nritems(eb);
1903 for (i = 0; i < nritems; i++) {
1904 btrfs_item_key_to_cpu(eb, &key, i);
1906 /* inode keys are done during the first stage */
1907 if (key.type == BTRFS_INODE_ITEM_KEY &&
1908 wc->stage == LOG_WALK_REPLAY_INODES) {
1909 struct btrfs_inode_item *inode_item;
1912 inode_item = btrfs_item_ptr(eb, i,
1913 struct btrfs_inode_item);
1914 mode = btrfs_inode_mode(eb, inode_item);
1915 if (S_ISDIR(mode)) {
1916 ret = replay_dir_deletes(wc->trans,
1917 root, log, path, key.objectid, 0);
1920 ret = overwrite_item(wc->trans, root, path,
1924 /* for regular files, make sure corresponding
1925 * orhpan item exist. extents past the new EOF
1926 * will be truncated later by orphan cleanup.
1928 if (S_ISREG(mode)) {
1929 ret = insert_orphan_item(wc->trans, root,
1934 ret = link_to_fixup_dir(wc->trans, root,
1935 path, key.objectid);
1938 if (wc->stage < LOG_WALK_REPLAY_ALL)
1941 /* these keys are simply copied */
1942 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1943 ret = overwrite_item(wc->trans, root, path,
1946 } else if (key.type == BTRFS_INODE_REF_KEY) {
1947 ret = add_inode_ref(wc->trans, root, log, path,
1949 BUG_ON(ret && ret != -ENOENT);
1950 } else if (key.type == BTRFS_INODE_EXTREF_KEY) {
1951 ret = add_inode_ref(wc->trans, root, log, path,
1953 BUG_ON(ret && ret != -ENOENT);
1954 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1955 ret = replay_one_extent(wc->trans, root, path,
1958 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1959 key.type == BTRFS_DIR_INDEX_KEY) {
1960 ret = replay_one_dir_item(wc->trans, root, path,
1965 btrfs_free_path(path);
1969 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
1970 struct btrfs_root *root,
1971 struct btrfs_path *path, int *level,
1972 struct walk_control *wc)
1977 struct extent_buffer *next;
1978 struct extent_buffer *cur;
1979 struct extent_buffer *parent;
1983 WARN_ON(*level < 0);
1984 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1986 while (*level > 0) {
1987 WARN_ON(*level < 0);
1988 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1989 cur = path->nodes[*level];
1991 if (btrfs_header_level(cur) != *level)
1994 if (path->slots[*level] >=
1995 btrfs_header_nritems(cur))
1998 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1999 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2000 blocksize = btrfs_level_size(root, *level - 1);
2002 parent = path->nodes[*level];
2003 root_owner = btrfs_header_owner(parent);
2005 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
2010 ret = wc->process_func(root, next, wc, ptr_gen);
2014 path->slots[*level]++;
2016 ret = btrfs_read_buffer(next, ptr_gen);
2018 free_extent_buffer(next);
2022 btrfs_tree_lock(next);
2023 btrfs_set_lock_blocking(next);
2024 clean_tree_block(trans, root, next);
2025 btrfs_wait_tree_block_writeback(next);
2026 btrfs_tree_unlock(next);
2028 WARN_ON(root_owner !=
2029 BTRFS_TREE_LOG_OBJECTID);
2030 ret = btrfs_free_and_pin_reserved_extent(root,
2032 BUG_ON(ret); /* -ENOMEM or logic errors */
2034 free_extent_buffer(next);
2037 ret = btrfs_read_buffer(next, ptr_gen);
2039 free_extent_buffer(next);
2043 WARN_ON(*level <= 0);
2044 if (path->nodes[*level-1])
2045 free_extent_buffer(path->nodes[*level-1]);
2046 path->nodes[*level-1] = next;
2047 *level = btrfs_header_level(next);
2048 path->slots[*level] = 0;
2051 WARN_ON(*level < 0);
2052 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2054 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2060 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2061 struct btrfs_root *root,
2062 struct btrfs_path *path, int *level,
2063 struct walk_control *wc)
2070 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2071 slot = path->slots[i];
2072 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2075 WARN_ON(*level == 0);
2078 struct extent_buffer *parent;
2079 if (path->nodes[*level] == root->node)
2080 parent = path->nodes[*level];
2082 parent = path->nodes[*level + 1];
2084 root_owner = btrfs_header_owner(parent);
2085 ret = wc->process_func(root, path->nodes[*level], wc,
2086 btrfs_header_generation(path->nodes[*level]));
2091 struct extent_buffer *next;
2093 next = path->nodes[*level];
2095 btrfs_tree_lock(next);
2096 btrfs_set_lock_blocking(next);
2097 clean_tree_block(trans, root, next);
2098 btrfs_wait_tree_block_writeback(next);
2099 btrfs_tree_unlock(next);
2101 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2102 ret = btrfs_free_and_pin_reserved_extent(root,
2103 path->nodes[*level]->start,
2104 path->nodes[*level]->len);
2107 free_extent_buffer(path->nodes[*level]);
2108 path->nodes[*level] = NULL;
2116 * drop the reference count on the tree rooted at 'snap'. This traverses
2117 * the tree freeing any blocks that have a ref count of zero after being
2120 static int walk_log_tree(struct btrfs_trans_handle *trans,
2121 struct btrfs_root *log, struct walk_control *wc)
2126 struct btrfs_path *path;
2130 path = btrfs_alloc_path();
2134 level = btrfs_header_level(log->node);
2136 path->nodes[level] = log->node;
2137 extent_buffer_get(log->node);
2138 path->slots[level] = 0;
2141 wret = walk_down_log_tree(trans, log, path, &level, wc);
2149 wret = walk_up_log_tree(trans, log, path, &level, wc);
2158 /* was the root node processed? if not, catch it here */
2159 if (path->nodes[orig_level]) {
2160 ret = wc->process_func(log, path->nodes[orig_level], wc,
2161 btrfs_header_generation(path->nodes[orig_level]));
2165 struct extent_buffer *next;
2167 next = path->nodes[orig_level];
2169 btrfs_tree_lock(next);
2170 btrfs_set_lock_blocking(next);
2171 clean_tree_block(trans, log, next);
2172 btrfs_wait_tree_block_writeback(next);
2173 btrfs_tree_unlock(next);
2175 WARN_ON(log->root_key.objectid !=
2176 BTRFS_TREE_LOG_OBJECTID);
2177 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2179 BUG_ON(ret); /* -ENOMEM or logic errors */
2184 for (i = 0; i <= orig_level; i++) {
2185 if (path->nodes[i]) {
2186 free_extent_buffer(path->nodes[i]);
2187 path->nodes[i] = NULL;
2190 btrfs_free_path(path);
2195 * helper function to update the item for a given subvolumes log root
2196 * in the tree of log roots
2198 static int update_log_root(struct btrfs_trans_handle *trans,
2199 struct btrfs_root *log)
2203 if (log->log_transid == 1) {
2204 /* insert root item on the first sync */
2205 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2206 &log->root_key, &log->root_item);
2208 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2209 &log->root_key, &log->root_item);
2214 static int wait_log_commit(struct btrfs_trans_handle *trans,
2215 struct btrfs_root *root, unsigned long transid)
2218 int index = transid % 2;
2221 * we only allow two pending log transactions at a time,
2222 * so we know that if ours is more than 2 older than the
2223 * current transaction, we're done
2226 prepare_to_wait(&root->log_commit_wait[index],
2227 &wait, TASK_UNINTERRUPTIBLE);
2228 mutex_unlock(&root->log_mutex);
2230 if (root->fs_info->last_trans_log_full_commit !=
2231 trans->transid && root->log_transid < transid + 2 &&
2232 atomic_read(&root->log_commit[index]))
2235 finish_wait(&root->log_commit_wait[index], &wait);
2236 mutex_lock(&root->log_mutex);
2237 } while (root->fs_info->last_trans_log_full_commit !=
2238 trans->transid && root->log_transid < transid + 2 &&
2239 atomic_read(&root->log_commit[index]));
2243 static void wait_for_writer(struct btrfs_trans_handle *trans,
2244 struct btrfs_root *root)
2247 while (root->fs_info->last_trans_log_full_commit !=
2248 trans->transid && atomic_read(&root->log_writers)) {
2249 prepare_to_wait(&root->log_writer_wait,
2250 &wait, TASK_UNINTERRUPTIBLE);
2251 mutex_unlock(&root->log_mutex);
2252 if (root->fs_info->last_trans_log_full_commit !=
2253 trans->transid && atomic_read(&root->log_writers))
2255 mutex_lock(&root->log_mutex);
2256 finish_wait(&root->log_writer_wait, &wait);
2261 * btrfs_sync_log does sends a given tree log down to the disk and
2262 * updates the super blocks to record it. When this call is done,
2263 * you know that any inodes previously logged are safely on disk only
2266 * Any other return value means you need to call btrfs_commit_transaction.
2267 * Some of the edge cases for fsyncing directories that have had unlinks
2268 * or renames done in the past mean that sometimes the only safe
2269 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2270 * that has happened.
2272 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2273 struct btrfs_root *root)
2279 struct btrfs_root *log = root->log_root;
2280 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2281 unsigned long log_transid = 0;
2283 mutex_lock(&root->log_mutex);
2284 index1 = root->log_transid % 2;
2285 if (atomic_read(&root->log_commit[index1])) {
2286 wait_log_commit(trans, root, root->log_transid);
2287 mutex_unlock(&root->log_mutex);
2290 atomic_set(&root->log_commit[index1], 1);
2292 /* wait for previous tree log sync to complete */
2293 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2294 wait_log_commit(trans, root, root->log_transid - 1);
2296 int batch = atomic_read(&root->log_batch);
2297 /* when we're on an ssd, just kick the log commit out */
2298 if (!btrfs_test_opt(root, SSD) && root->log_multiple_pids) {
2299 mutex_unlock(&root->log_mutex);
2300 schedule_timeout_uninterruptible(1);
2301 mutex_lock(&root->log_mutex);
2303 wait_for_writer(trans, root);
2304 if (batch == atomic_read(&root->log_batch))
2308 /* bail out if we need to do a full commit */
2309 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2311 mutex_unlock(&root->log_mutex);
2315 log_transid = root->log_transid;
2316 if (log_transid % 2 == 0)
2317 mark = EXTENT_DIRTY;
2321 /* we start IO on all the marked extents here, but we don't actually
2322 * wait for them until later.
2324 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2326 btrfs_abort_transaction(trans, root, ret);
2327 mutex_unlock(&root->log_mutex);
2331 btrfs_set_root_node(&log->root_item, log->node);
2333 root->log_transid++;
2334 log->log_transid = root->log_transid;
2335 root->log_start_pid = 0;
2338 * IO has been started, blocks of the log tree have WRITTEN flag set
2339 * in their headers. new modifications of the log will be written to
2340 * new positions. so it's safe to allow log writers to go in.
2342 mutex_unlock(&root->log_mutex);
2344 mutex_lock(&log_root_tree->log_mutex);
2345 atomic_inc(&log_root_tree->log_batch);
2346 atomic_inc(&log_root_tree->log_writers);
2347 mutex_unlock(&log_root_tree->log_mutex);
2349 ret = update_log_root(trans, log);
2351 mutex_lock(&log_root_tree->log_mutex);
2352 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2354 if (waitqueue_active(&log_root_tree->log_writer_wait))
2355 wake_up(&log_root_tree->log_writer_wait);
2359 if (ret != -ENOSPC) {
2360 btrfs_abort_transaction(trans, root, ret);
2361 mutex_unlock(&log_root_tree->log_mutex);
2364 root->fs_info->last_trans_log_full_commit = trans->transid;
2365 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2366 mutex_unlock(&log_root_tree->log_mutex);
2371 index2 = log_root_tree->log_transid % 2;
2372 if (atomic_read(&log_root_tree->log_commit[index2])) {
2373 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2374 wait_log_commit(trans, log_root_tree,
2375 log_root_tree->log_transid);
2376 mutex_unlock(&log_root_tree->log_mutex);
2380 atomic_set(&log_root_tree->log_commit[index2], 1);
2382 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2383 wait_log_commit(trans, log_root_tree,
2384 log_root_tree->log_transid - 1);
2387 wait_for_writer(trans, log_root_tree);
2390 * now that we've moved on to the tree of log tree roots,
2391 * check the full commit flag again
2393 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2394 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2395 mutex_unlock(&log_root_tree->log_mutex);
2397 goto out_wake_log_root;
2400 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2401 &log_root_tree->dirty_log_pages,
2402 EXTENT_DIRTY | EXTENT_NEW);
2404 btrfs_abort_transaction(trans, root, ret);
2405 mutex_unlock(&log_root_tree->log_mutex);
2406 goto out_wake_log_root;
2408 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2410 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2411 log_root_tree->node->start);
2412 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2413 btrfs_header_level(log_root_tree->node));
2415 log_root_tree->log_transid++;
2418 mutex_unlock(&log_root_tree->log_mutex);
2421 * nobody else is going to jump in and write the the ctree
2422 * super here because the log_commit atomic below is protecting
2423 * us. We must be called with a transaction handle pinning
2424 * the running transaction open, so a full commit can't hop
2425 * in and cause problems either.
2427 btrfs_scrub_pause_super(root);
2428 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2429 btrfs_scrub_continue_super(root);
2431 btrfs_abort_transaction(trans, root, ret);
2432 goto out_wake_log_root;
2435 mutex_lock(&root->log_mutex);
2436 if (root->last_log_commit < log_transid)
2437 root->last_log_commit = log_transid;
2438 mutex_unlock(&root->log_mutex);
2441 atomic_set(&log_root_tree->log_commit[index2], 0);
2443 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2444 wake_up(&log_root_tree->log_commit_wait[index2]);
2446 atomic_set(&root->log_commit[index1], 0);
2448 if (waitqueue_active(&root->log_commit_wait[index1]))
2449 wake_up(&root->log_commit_wait[index1]);
2453 static void free_log_tree(struct btrfs_trans_handle *trans,
2454 struct btrfs_root *log)
2459 struct walk_control wc = {
2461 .process_func = process_one_buffer
2464 ret = walk_log_tree(trans, log, &wc);
2468 ret = find_first_extent_bit(&log->dirty_log_pages,
2469 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2474 clear_extent_bits(&log->dirty_log_pages, start, end,
2475 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2478 free_extent_buffer(log->node);
2483 * free all the extents used by the tree log. This should be called
2484 * at commit time of the full transaction
2486 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2488 if (root->log_root) {
2489 free_log_tree(trans, root->log_root);
2490 root->log_root = NULL;
2495 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2496 struct btrfs_fs_info *fs_info)
2498 if (fs_info->log_root_tree) {
2499 free_log_tree(trans, fs_info->log_root_tree);
2500 fs_info->log_root_tree = NULL;
2506 * If both a file and directory are logged, and unlinks or renames are
2507 * mixed in, we have a few interesting corners:
2509 * create file X in dir Y
2510 * link file X to X.link in dir Y
2512 * unlink file X but leave X.link
2515 * After a crash we would expect only X.link to exist. But file X
2516 * didn't get fsync'd again so the log has back refs for X and X.link.
2518 * We solve this by removing directory entries and inode backrefs from the
2519 * log when a file that was logged in the current transaction is
2520 * unlinked. Any later fsync will include the updated log entries, and
2521 * we'll be able to reconstruct the proper directory items from backrefs.
2523 * This optimizations allows us to avoid relogging the entire inode
2524 * or the entire directory.
2526 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2527 struct btrfs_root *root,
2528 const char *name, int name_len,
2529 struct inode *dir, u64 index)
2531 struct btrfs_root *log;
2532 struct btrfs_dir_item *di;
2533 struct btrfs_path *path;
2537 u64 dir_ino = btrfs_ino(dir);
2539 if (BTRFS_I(dir)->logged_trans < trans->transid)
2542 ret = join_running_log_trans(root);
2546 mutex_lock(&BTRFS_I(dir)->log_mutex);
2548 log = root->log_root;
2549 path = btrfs_alloc_path();
2555 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
2556 name, name_len, -1);
2562 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2563 bytes_del += name_len;
2566 btrfs_release_path(path);
2567 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
2568 index, name, name_len, -1);
2574 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2575 bytes_del += name_len;
2579 /* update the directory size in the log to reflect the names
2583 struct btrfs_key key;
2585 key.objectid = dir_ino;
2587 key.type = BTRFS_INODE_ITEM_KEY;
2588 btrfs_release_path(path);
2590 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2596 struct btrfs_inode_item *item;
2599 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2600 struct btrfs_inode_item);
2601 i_size = btrfs_inode_size(path->nodes[0], item);
2602 if (i_size > bytes_del)
2603 i_size -= bytes_del;
2606 btrfs_set_inode_size(path->nodes[0], item, i_size);
2607 btrfs_mark_buffer_dirty(path->nodes[0]);
2610 btrfs_release_path(path);
2613 btrfs_free_path(path);
2615 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2616 if (ret == -ENOSPC) {
2617 root->fs_info->last_trans_log_full_commit = trans->transid;
2620 btrfs_abort_transaction(trans, root, ret);
2622 btrfs_end_log_trans(root);
2627 /* see comments for btrfs_del_dir_entries_in_log */
2628 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2629 struct btrfs_root *root,
2630 const char *name, int name_len,
2631 struct inode *inode, u64 dirid)
2633 struct btrfs_root *log;
2637 if (BTRFS_I(inode)->logged_trans < trans->transid)
2640 ret = join_running_log_trans(root);
2643 log = root->log_root;
2644 mutex_lock(&BTRFS_I(inode)->log_mutex);
2646 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
2648 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2649 if (ret == -ENOSPC) {
2650 root->fs_info->last_trans_log_full_commit = trans->transid;
2652 } else if (ret < 0 && ret != -ENOENT)
2653 btrfs_abort_transaction(trans, root, ret);
2654 btrfs_end_log_trans(root);
2660 * creates a range item in the log for 'dirid'. first_offset and
2661 * last_offset tell us which parts of the key space the log should
2662 * be considered authoritative for.
2664 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2665 struct btrfs_root *log,
2666 struct btrfs_path *path,
2667 int key_type, u64 dirid,
2668 u64 first_offset, u64 last_offset)
2671 struct btrfs_key key;
2672 struct btrfs_dir_log_item *item;
2674 key.objectid = dirid;
2675 key.offset = first_offset;
2676 if (key_type == BTRFS_DIR_ITEM_KEY)
2677 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2679 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2680 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2684 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2685 struct btrfs_dir_log_item);
2686 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2687 btrfs_mark_buffer_dirty(path->nodes[0]);
2688 btrfs_release_path(path);
2693 * log all the items included in the current transaction for a given
2694 * directory. This also creates the range items in the log tree required
2695 * to replay anything deleted before the fsync
2697 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2698 struct btrfs_root *root, struct inode *inode,
2699 struct btrfs_path *path,
2700 struct btrfs_path *dst_path, int key_type,
2701 u64 min_offset, u64 *last_offset_ret)
2703 struct btrfs_key min_key;
2704 struct btrfs_key max_key;
2705 struct btrfs_root *log = root->log_root;
2706 struct extent_buffer *src;
2711 u64 first_offset = min_offset;
2712 u64 last_offset = (u64)-1;
2713 u64 ino = btrfs_ino(inode);
2715 log = root->log_root;
2716 max_key.objectid = ino;
2717 max_key.offset = (u64)-1;
2718 max_key.type = key_type;
2720 min_key.objectid = ino;
2721 min_key.type = key_type;
2722 min_key.offset = min_offset;
2724 path->keep_locks = 1;
2726 ret = btrfs_search_forward(root, &min_key, &max_key,
2727 path, 0, trans->transid);
2730 * we didn't find anything from this transaction, see if there
2731 * is anything at all
2733 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2734 min_key.objectid = ino;
2735 min_key.type = key_type;
2736 min_key.offset = (u64)-1;
2737 btrfs_release_path(path);
2738 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2740 btrfs_release_path(path);
2743 ret = btrfs_previous_item(root, path, ino, key_type);
2745 /* if ret == 0 there are items for this type,
2746 * create a range to tell us the last key of this type.
2747 * otherwise, there are no items in this directory after
2748 * *min_offset, and we create a range to indicate that.
2751 struct btrfs_key tmp;
2752 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2754 if (key_type == tmp.type)
2755 first_offset = max(min_offset, tmp.offset) + 1;
2760 /* go backward to find any previous key */
2761 ret = btrfs_previous_item(root, path, ino, key_type);
2763 struct btrfs_key tmp;
2764 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2765 if (key_type == tmp.type) {
2766 first_offset = tmp.offset;
2767 ret = overwrite_item(trans, log, dst_path,
2768 path->nodes[0], path->slots[0],
2776 btrfs_release_path(path);
2778 /* find the first key from this transaction again */
2779 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2786 * we have a block from this transaction, log every item in it
2787 * from our directory
2790 struct btrfs_key tmp;
2791 src = path->nodes[0];
2792 nritems = btrfs_header_nritems(src);
2793 for (i = path->slots[0]; i < nritems; i++) {
2794 btrfs_item_key_to_cpu(src, &min_key, i);
2796 if (min_key.objectid != ino || min_key.type != key_type)
2798 ret = overwrite_item(trans, log, dst_path, src, i,
2805 path->slots[0] = nritems;
2808 * look ahead to the next item and see if it is also
2809 * from this directory and from this transaction
2811 ret = btrfs_next_leaf(root, path);
2813 last_offset = (u64)-1;
2816 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2817 if (tmp.objectid != ino || tmp.type != key_type) {
2818 last_offset = (u64)-1;
2821 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2822 ret = overwrite_item(trans, log, dst_path,
2823 path->nodes[0], path->slots[0],
2828 last_offset = tmp.offset;
2833 btrfs_release_path(path);
2834 btrfs_release_path(dst_path);
2837 *last_offset_ret = last_offset;
2839 * insert the log range keys to indicate where the log
2842 ret = insert_dir_log_key(trans, log, path, key_type,
2843 ino, first_offset, last_offset);
2851 * logging directories is very similar to logging inodes, We find all the items
2852 * from the current transaction and write them to the log.
2854 * The recovery code scans the directory in the subvolume, and if it finds a
2855 * key in the range logged that is not present in the log tree, then it means
2856 * that dir entry was unlinked during the transaction.
2858 * In order for that scan to work, we must include one key smaller than
2859 * the smallest logged by this transaction and one key larger than the largest
2860 * key logged by this transaction.
2862 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2863 struct btrfs_root *root, struct inode *inode,
2864 struct btrfs_path *path,
2865 struct btrfs_path *dst_path)
2870 int key_type = BTRFS_DIR_ITEM_KEY;
2876 ret = log_dir_items(trans, root, inode, path,
2877 dst_path, key_type, min_key,
2881 if (max_key == (u64)-1)
2883 min_key = max_key + 1;
2886 if (key_type == BTRFS_DIR_ITEM_KEY) {
2887 key_type = BTRFS_DIR_INDEX_KEY;
2894 * a helper function to drop items from the log before we relog an
2895 * inode. max_key_type indicates the highest item type to remove.
2896 * This cannot be run for file data extents because it does not
2897 * free the extents they point to.
2899 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2900 struct btrfs_root *log,
2901 struct btrfs_path *path,
2902 u64 objectid, int max_key_type)
2905 struct btrfs_key key;
2906 struct btrfs_key found_key;
2909 key.objectid = objectid;
2910 key.type = max_key_type;
2911 key.offset = (u64)-1;
2914 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2919 if (path->slots[0] == 0)
2923 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2926 if (found_key.objectid != objectid)
2929 found_key.offset = 0;
2931 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
2934 ret = btrfs_del_items(trans, log, path, start_slot,
2935 path->slots[0] - start_slot + 1);
2937 * If start slot isn't 0 then we don't need to re-search, we've
2938 * found the last guy with the objectid in this tree.
2940 if (ret || start_slot != 0)
2942 btrfs_release_path(path);
2944 btrfs_release_path(path);
2950 static void fill_inode_item(struct btrfs_trans_handle *trans,
2951 struct extent_buffer *leaf,
2952 struct btrfs_inode_item *item,
2953 struct inode *inode, int log_inode_only)
2955 struct btrfs_map_token token;
2957 btrfs_init_map_token(&token);
2959 if (log_inode_only) {
2960 /* set the generation to zero so the recover code
2961 * can tell the difference between an logging
2962 * just to say 'this inode exists' and a logging
2963 * to say 'update this inode with these values'
2965 btrfs_set_token_inode_generation(leaf, item, 0, &token);
2966 btrfs_set_token_inode_size(leaf, item, 0, &token);
2968 btrfs_set_token_inode_generation(leaf, item,
2969 BTRFS_I(inode)->generation,
2971 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
2974 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
2975 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
2976 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
2977 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
2979 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
2980 inode->i_atime.tv_sec, &token);
2981 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
2982 inode->i_atime.tv_nsec, &token);
2984 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
2985 inode->i_mtime.tv_sec, &token);
2986 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
2987 inode->i_mtime.tv_nsec, &token);
2989 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
2990 inode->i_ctime.tv_sec, &token);
2991 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
2992 inode->i_ctime.tv_nsec, &token);
2994 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
2997 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
2998 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
2999 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3000 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3001 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3004 static int log_inode_item(struct btrfs_trans_handle *trans,
3005 struct btrfs_root *log, struct btrfs_path *path,
3006 struct inode *inode)
3008 struct btrfs_inode_item *inode_item;
3009 struct btrfs_key key;
3012 memcpy(&key, &BTRFS_I(inode)->location, sizeof(key));
3013 ret = btrfs_insert_empty_item(trans, log, path, &key,
3014 sizeof(*inode_item));
3015 if (ret && ret != -EEXIST)
3017 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3018 struct btrfs_inode_item);
3019 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0);
3020 btrfs_release_path(path);
3024 static noinline int copy_items(struct btrfs_trans_handle *trans,
3025 struct inode *inode,
3026 struct btrfs_path *dst_path,
3027 struct extent_buffer *src,
3028 int start_slot, int nr, int inode_only)
3030 unsigned long src_offset;
3031 unsigned long dst_offset;
3032 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3033 struct btrfs_file_extent_item *extent;
3034 struct btrfs_inode_item *inode_item;
3036 struct btrfs_key *ins_keys;
3040 struct list_head ordered_sums;
3041 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3043 INIT_LIST_HEAD(&ordered_sums);
3045 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3046 nr * sizeof(u32), GFP_NOFS);
3050 ins_sizes = (u32 *)ins_data;
3051 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3053 for (i = 0; i < nr; i++) {
3054 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3055 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3057 ret = btrfs_insert_empty_items(trans, log, dst_path,
3058 ins_keys, ins_sizes, nr);
3064 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3065 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3066 dst_path->slots[0]);
3068 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3070 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3071 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3073 struct btrfs_inode_item);
3074 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3075 inode, inode_only == LOG_INODE_EXISTS);
3077 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3078 src_offset, ins_sizes[i]);
3081 /* take a reference on file data extents so that truncates
3082 * or deletes of this inode don't have to relog the inode
3085 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY &&
3088 extent = btrfs_item_ptr(src, start_slot + i,
3089 struct btrfs_file_extent_item);
3091 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3094 found_type = btrfs_file_extent_type(src, extent);
3095 if (found_type == BTRFS_FILE_EXTENT_REG) {
3097 ds = btrfs_file_extent_disk_bytenr(src,
3099 /* ds == 0 is a hole */
3103 dl = btrfs_file_extent_disk_num_bytes(src,
3105 cs = btrfs_file_extent_offset(src, extent);
3106 cl = btrfs_file_extent_num_bytes(src,
3108 if (btrfs_file_extent_compression(src,
3114 ret = btrfs_lookup_csums_range(
3115 log->fs_info->csum_root,
3116 ds + cs, ds + cs + cl - 1,
3123 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3124 btrfs_release_path(dst_path);
3128 * we have to do this after the loop above to avoid changing the
3129 * log tree while trying to change the log tree.
3132 while (!list_empty(&ordered_sums)) {
3133 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3134 struct btrfs_ordered_sum,
3137 ret = btrfs_csum_file_blocks(trans, log, sums);
3138 list_del(&sums->list);
3144 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3146 struct extent_map *em1, *em2;
3148 em1 = list_entry(a, struct extent_map, list);
3149 em2 = list_entry(b, struct extent_map, list);
3151 if (em1->start < em2->start)
3153 else if (em1->start > em2->start)
3158 static int drop_adjacent_extents(struct btrfs_trans_handle *trans,
3159 struct btrfs_root *root, struct inode *inode,
3160 struct extent_map *em,
3161 struct btrfs_path *path)
3163 struct btrfs_file_extent_item *fi;
3164 struct extent_buffer *leaf;
3165 struct btrfs_key key, new_key;
3166 struct btrfs_map_token token;
3168 u64 extent_offset = 0;
3175 btrfs_init_map_token(&token);
3176 leaf = path->nodes[0];
3178 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3180 ret = btrfs_del_items(trans, root, path,
3187 ret = btrfs_next_leaf_write(trans, root, path, 1);
3192 leaf = path->nodes[0];
3195 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3196 if (key.objectid != btrfs_ino(inode) ||
3197 key.type != BTRFS_EXTENT_DATA_KEY ||
3198 key.offset >= em->start + em->len)
3201 fi = btrfs_item_ptr(leaf, path->slots[0],
3202 struct btrfs_file_extent_item);
3203 extent_type = btrfs_token_file_extent_type(leaf, fi, &token);
3204 if (extent_type == BTRFS_FILE_EXTENT_REG ||
3205 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
3206 extent_offset = btrfs_token_file_extent_offset(leaf,
3208 extent_end = key.offset +
3209 btrfs_token_file_extent_num_bytes(leaf, fi,
3211 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3212 extent_end = key.offset +
3213 btrfs_file_extent_inline_len(leaf, fi);
3218 if (extent_end <= em->len + em->start) {
3220 del_slot = path->slots[0];
3227 * Ok so we'll ignore previous items if we log a new extent,
3228 * which can lead to overlapping extents, so if we have an
3229 * existing extent we want to adjust we _have_ to check the next
3230 * guy to make sure we even need this extent anymore, this keeps
3231 * us from panicing in set_item_key_safe.
3233 if (path->slots[0] < btrfs_header_nritems(leaf) - 1) {
3234 struct btrfs_key tmp_key;
3236 btrfs_item_key_to_cpu(leaf, &tmp_key,
3237 path->slots[0] + 1);
3238 if (tmp_key.objectid == btrfs_ino(inode) &&
3239 tmp_key.type == BTRFS_EXTENT_DATA_KEY &&
3240 tmp_key.offset <= em->start + em->len) {
3242 del_slot = path->slots[0];
3248 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
3249 memcpy(&new_key, &key, sizeof(new_key));
3250 new_key.offset = em->start + em->len;
3251 btrfs_set_item_key_safe(trans, root, path, &new_key);
3252 extent_offset += em->start + em->len - key.offset;
3253 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset,
3255 btrfs_set_token_file_extent_num_bytes(leaf, fi, extent_end -
3256 (em->start + em->len),
3258 btrfs_mark_buffer_dirty(leaf);
3262 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
3267 static int log_one_extent(struct btrfs_trans_handle *trans,
3268 struct inode *inode, struct btrfs_root *root,
3269 struct extent_map *em, struct btrfs_path *path)
3271 struct btrfs_root *log = root->log_root;
3272 struct btrfs_file_extent_item *fi;
3273 struct extent_buffer *leaf;
3274 struct list_head ordered_sums;
3275 struct btrfs_map_token token;
3276 struct btrfs_key key;
3277 u64 csum_offset = em->mod_start - em->start;
3278 u64 csum_len = em->mod_len;
3279 u64 extent_offset = em->start - em->orig_start;
3282 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3284 INIT_LIST_HEAD(&ordered_sums);
3285 btrfs_init_map_token(&token);
3286 key.objectid = btrfs_ino(inode);
3287 key.type = BTRFS_EXTENT_DATA_KEY;
3288 key.offset = em->start;
3289 path->really_keep_locks = 1;
3291 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*fi));
3292 if (ret && ret != -EEXIST) {
3293 path->really_keep_locks = 0;
3296 leaf = path->nodes[0];
3297 fi = btrfs_item_ptr(leaf, path->slots[0],
3298 struct btrfs_file_extent_item);
3299 btrfs_set_token_file_extent_generation(leaf, fi, em->generation,
3301 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3303 btrfs_set_token_file_extent_type(leaf, fi,
3304 BTRFS_FILE_EXTENT_PREALLOC,
3307 btrfs_set_token_file_extent_type(leaf, fi,
3308 BTRFS_FILE_EXTENT_REG,
3310 if (em->block_start == 0)
3314 block_len = max(em->block_len, em->orig_block_len);
3315 if (em->compress_type != BTRFS_COMPRESS_NONE) {
3316 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3319 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3321 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
3322 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3324 extent_offset, &token);
3325 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3328 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
3329 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
3333 btrfs_set_token_file_extent_offset(leaf, fi,
3334 em->start - em->orig_start,
3336 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
3337 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->len, &token);
3338 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
3340 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
3341 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
3342 btrfs_mark_buffer_dirty(leaf);
3345 * Have to check the extent to the right of us to make sure it doesn't
3346 * fall in our current range. We're ok if the previous extent is in our
3347 * range since the recovery stuff will run us in key order and thus just
3348 * drop the part we overwrote.
3350 ret = drop_adjacent_extents(trans, log, inode, em, path);
3351 btrfs_release_path(path);
3352 path->really_keep_locks = 0;
3360 /* block start is already adjusted for the file extent offset. */
3361 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
3362 em->block_start + csum_offset,
3363 em->block_start + csum_offset +
3364 csum_len - 1, &ordered_sums, 0);
3368 while (!list_empty(&ordered_sums)) {
3369 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3370 struct btrfs_ordered_sum,
3373 ret = btrfs_csum_file_blocks(trans, log, sums);
3374 list_del(&sums->list);
3381 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
3382 struct btrfs_root *root,
3383 struct inode *inode,
3384 struct btrfs_path *path)
3386 struct extent_map *em, *n;
3387 struct list_head extents;
3388 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3392 INIT_LIST_HEAD(&extents);
3394 write_lock(&tree->lock);
3395 test_gen = root->fs_info->last_trans_committed;
3397 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
3398 list_del_init(&em->list);
3399 if (em->generation <= test_gen)
3401 /* Need a ref to keep it from getting evicted from cache */
3402 atomic_inc(&em->refs);
3403 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
3404 list_add_tail(&em->list, &extents);
3407 list_sort(NULL, &extents, extent_cmp);
3409 while (!list_empty(&extents)) {
3410 em = list_entry(extents.next, struct extent_map, list);
3412 list_del_init(&em->list);
3413 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
3416 * If we had an error we just need to delete everybody from our
3420 free_extent_map(em);
3424 write_unlock(&tree->lock);
3426 ret = log_one_extent(trans, inode, root, em, path);
3427 free_extent_map(em);
3428 write_lock(&tree->lock);
3430 WARN_ON(!list_empty(&extents));
3431 write_unlock(&tree->lock);
3433 btrfs_release_path(path);
3437 /* log a single inode in the tree log.
3438 * At least one parent directory for this inode must exist in the tree
3439 * or be logged already.
3441 * Any items from this inode changed by the current transaction are copied
3442 * to the log tree. An extra reference is taken on any extents in this
3443 * file, allowing us to avoid a whole pile of corner cases around logging
3444 * blocks that have been removed from the tree.
3446 * See LOG_INODE_ALL and related defines for a description of what inode_only
3449 * This handles both files and directories.
3451 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
3452 struct btrfs_root *root, struct inode *inode,
3455 struct btrfs_path *path;
3456 struct btrfs_path *dst_path;
3457 struct btrfs_key min_key;
3458 struct btrfs_key max_key;
3459 struct btrfs_root *log = root->log_root;
3460 struct extent_buffer *src = NULL;
3464 int ins_start_slot = 0;
3466 bool fast_search = false;
3467 u64 ino = btrfs_ino(inode);
3469 log = root->log_root;
3471 path = btrfs_alloc_path();
3474 dst_path = btrfs_alloc_path();
3476 btrfs_free_path(path);
3480 min_key.objectid = ino;
3481 min_key.type = BTRFS_INODE_ITEM_KEY;
3484 max_key.objectid = ino;
3487 /* today the code can only do partial logging of directories */
3488 if (S_ISDIR(inode->i_mode) ||
3489 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3490 &BTRFS_I(inode)->runtime_flags) &&
3491 inode_only == LOG_INODE_EXISTS))
3492 max_key.type = BTRFS_XATTR_ITEM_KEY;
3494 max_key.type = (u8)-1;
3495 max_key.offset = (u64)-1;
3497 /* Only run delayed items if we are a dir or a new file */
3498 if (S_ISDIR(inode->i_mode) ||
3499 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) {
3500 ret = btrfs_commit_inode_delayed_items(trans, inode);
3502 btrfs_free_path(path);
3503 btrfs_free_path(dst_path);
3508 mutex_lock(&BTRFS_I(inode)->log_mutex);
3511 * a brute force approach to making sure we get the most uptodate
3512 * copies of everything.
3514 if (S_ISDIR(inode->i_mode)) {
3515 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
3517 if (inode_only == LOG_INODE_EXISTS)
3518 max_key_type = BTRFS_XATTR_ITEM_KEY;
3519 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
3521 if (test_and_clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3522 &BTRFS_I(inode)->runtime_flags)) {
3523 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3524 &BTRFS_I(inode)->runtime_flags);
3525 ret = btrfs_truncate_inode_items(trans, log,
3527 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3528 &BTRFS_I(inode)->runtime_flags)) {
3529 if (inode_only == LOG_INODE_ALL)
3531 max_key.type = BTRFS_XATTR_ITEM_KEY;
3532 ret = drop_objectid_items(trans, log, path, ino,
3535 if (inode_only == LOG_INODE_ALL)
3537 ret = log_inode_item(trans, log, dst_path, inode);
3550 path->keep_locks = 1;
3554 ret = btrfs_search_forward(root, &min_key, &max_key,
3555 path, 0, trans->transid);
3559 /* note, ins_nr might be > 0 here, cleanup outside the loop */
3560 if (min_key.objectid != ino)
3562 if (min_key.type > max_key.type)
3565 src = path->nodes[0];
3566 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
3569 } else if (!ins_nr) {
3570 ins_start_slot = path->slots[0];
3575 ret = copy_items(trans, inode, dst_path, src, ins_start_slot,
3576 ins_nr, inode_only);
3582 ins_start_slot = path->slots[0];
3585 nritems = btrfs_header_nritems(path->nodes[0]);
3587 if (path->slots[0] < nritems) {
3588 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
3593 ret = copy_items(trans, inode, dst_path, src,
3595 ins_nr, inode_only);
3602 btrfs_release_path(path);
3604 if (min_key.offset < (u64)-1)
3606 else if (min_key.type < (u8)-1)
3608 else if (min_key.objectid < (u64)-1)
3614 ret = copy_items(trans, inode, dst_path, src, ins_start_slot,
3615 ins_nr, inode_only);
3625 btrfs_release_path(dst_path);
3626 ret = btrfs_log_changed_extents(trans, root, inode, dst_path);
3632 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3633 struct extent_map *em, *n;
3635 write_lock(&tree->lock);
3636 list_for_each_entry_safe(em, n, &tree->modified_extents, list)
3637 list_del_init(&em->list);
3638 write_unlock(&tree->lock);
3641 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
3642 btrfs_release_path(path);
3643 btrfs_release_path(dst_path);
3644 ret = log_directory_changes(trans, root, inode, path, dst_path);
3650 BTRFS_I(inode)->logged_trans = trans->transid;
3651 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
3653 mutex_unlock(&BTRFS_I(inode)->log_mutex);
3655 btrfs_free_path(path);
3656 btrfs_free_path(dst_path);
3661 * follow the dentry parent pointers up the chain and see if any
3662 * of the directories in it require a full commit before they can
3663 * be logged. Returns zero if nothing special needs to be done or 1 if
3664 * a full commit is required.
3666 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
3667 struct inode *inode,
3668 struct dentry *parent,
3669 struct super_block *sb,
3673 struct btrfs_root *root;
3674 struct dentry *old_parent = NULL;
3677 * for regular files, if its inode is already on disk, we don't
3678 * have to worry about the parents at all. This is because
3679 * we can use the last_unlink_trans field to record renames
3680 * and other fun in this file.
3682 if (S_ISREG(inode->i_mode) &&
3683 BTRFS_I(inode)->generation <= last_committed &&
3684 BTRFS_I(inode)->last_unlink_trans <= last_committed)
3687 if (!S_ISDIR(inode->i_mode)) {
3688 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3690 inode = parent->d_inode;
3694 BTRFS_I(inode)->logged_trans = trans->transid;
3697 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
3698 root = BTRFS_I(inode)->root;
3701 * make sure any commits to the log are forced
3702 * to be full commits
3704 root->fs_info->last_trans_log_full_commit =
3710 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3713 if (IS_ROOT(parent))
3716 parent = dget_parent(parent);
3718 old_parent = parent;
3719 inode = parent->d_inode;
3728 * helper function around btrfs_log_inode to make sure newly created
3729 * parent directories also end up in the log. A minimal inode and backref
3730 * only logging is done of any parent directories that are older than
3731 * the last committed transaction
3733 int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
3734 struct btrfs_root *root, struct inode *inode,
3735 struct dentry *parent, int exists_only)
3737 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
3738 struct super_block *sb;
3739 struct dentry *old_parent = NULL;
3741 u64 last_committed = root->fs_info->last_trans_committed;
3745 if (btrfs_test_opt(root, NOTREELOG)) {
3750 if (root->fs_info->last_trans_log_full_commit >
3751 root->fs_info->last_trans_committed) {
3756 if (root != BTRFS_I(inode)->root ||
3757 btrfs_root_refs(&root->root_item) == 0) {
3762 ret = check_parent_dirs_for_sync(trans, inode, parent,
3763 sb, last_committed);
3767 if (btrfs_inode_in_log(inode, trans->transid)) {
3768 ret = BTRFS_NO_LOG_SYNC;
3772 ret = start_log_trans(trans, root);
3776 ret = btrfs_log_inode(trans, root, inode, inode_only);
3781 * for regular files, if its inode is already on disk, we don't
3782 * have to worry about the parents at all. This is because
3783 * we can use the last_unlink_trans field to record renames
3784 * and other fun in this file.
3786 if (S_ISREG(inode->i_mode) &&
3787 BTRFS_I(inode)->generation <= last_committed &&
3788 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3793 inode_only = LOG_INODE_EXISTS;
3795 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3798 inode = parent->d_inode;
3799 if (root != BTRFS_I(inode)->root)
3802 if (BTRFS_I(inode)->generation >
3803 root->fs_info->last_trans_committed) {
3804 ret = btrfs_log_inode(trans, root, inode, inode_only);
3808 if (IS_ROOT(parent))
3811 parent = dget_parent(parent);
3813 old_parent = parent;
3819 WARN_ON(ret != -ENOSPC);
3820 root->fs_info->last_trans_log_full_commit = trans->transid;
3823 btrfs_end_log_trans(root);
3829 * it is not safe to log dentry if the chunk root has added new
3830 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3831 * If this returns 1, you must commit the transaction to safely get your
3834 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3835 struct btrfs_root *root, struct dentry *dentry)
3837 struct dentry *parent = dget_parent(dentry);
3840 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
3847 * should be called during mount to recover any replay any log trees
3850 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3853 struct btrfs_path *path;
3854 struct btrfs_trans_handle *trans;
3855 struct btrfs_key key;
3856 struct btrfs_key found_key;
3857 struct btrfs_key tmp_key;
3858 struct btrfs_root *log;
3859 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3860 struct walk_control wc = {
3861 .process_func = process_one_buffer,
3865 path = btrfs_alloc_path();
3869 fs_info->log_root_recovering = 1;
3871 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3872 if (IS_ERR(trans)) {
3873 ret = PTR_ERR(trans);
3880 ret = walk_log_tree(trans, log_root_tree, &wc);
3882 btrfs_error(fs_info, ret, "Failed to pin buffers while "
3883 "recovering log root tree.");
3888 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3889 key.offset = (u64)-1;
3890 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3893 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3896 btrfs_error(fs_info, ret,
3897 "Couldn't find tree log root.");
3901 if (path->slots[0] == 0)
3905 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3907 btrfs_release_path(path);
3908 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3911 log = btrfs_read_fs_root_no_radix(log_root_tree,
3915 btrfs_error(fs_info, ret,
3916 "Couldn't read tree log root.");
3920 tmp_key.objectid = found_key.offset;
3921 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3922 tmp_key.offset = (u64)-1;
3924 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
3925 if (IS_ERR(wc.replay_dest)) {
3926 ret = PTR_ERR(wc.replay_dest);
3927 btrfs_error(fs_info, ret, "Couldn't read target root "
3928 "for tree log recovery.");
3932 wc.replay_dest->log_root = log;
3933 btrfs_record_root_in_trans(trans, wc.replay_dest);
3934 ret = walk_log_tree(trans, log, &wc);
3937 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3938 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3943 key.offset = found_key.offset - 1;
3944 wc.replay_dest->log_root = NULL;
3945 free_extent_buffer(log->node);
3946 free_extent_buffer(log->commit_root);
3949 if (found_key.offset == 0)
3952 btrfs_release_path(path);
3954 /* step one is to pin it all, step two is to replay just inodes */
3957 wc.process_func = replay_one_buffer;
3958 wc.stage = LOG_WALK_REPLAY_INODES;
3961 /* step three is to replay everything */
3962 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3967 btrfs_free_path(path);
3969 free_extent_buffer(log_root_tree->node);
3970 log_root_tree->log_root = NULL;
3971 fs_info->log_root_recovering = 0;
3973 /* step 4: commit the transaction, which also unpins the blocks */
3974 btrfs_commit_transaction(trans, fs_info->tree_root);
3976 kfree(log_root_tree);
3980 btrfs_free_path(path);
3985 * there are some corner cases where we want to force a full
3986 * commit instead of allowing a directory to be logged.
3988 * They revolve around files there were unlinked from the directory, and
3989 * this function updates the parent directory so that a full commit is
3990 * properly done if it is fsync'd later after the unlinks are done.
3992 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3993 struct inode *dir, struct inode *inode,
3997 * when we're logging a file, if it hasn't been renamed
3998 * or unlinked, and its inode is fully committed on disk,
3999 * we don't have to worry about walking up the directory chain
4000 * to log its parents.
4002 * So, we use the last_unlink_trans field to put this transid
4003 * into the file. When the file is logged we check it and
4004 * don't log the parents if the file is fully on disk.
4006 if (S_ISREG(inode->i_mode))
4007 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4010 * if this directory was already logged any new
4011 * names for this file/dir will get recorded
4014 if (BTRFS_I(dir)->logged_trans == trans->transid)
4018 * if the inode we're about to unlink was logged,
4019 * the log will be properly updated for any new names
4021 if (BTRFS_I(inode)->logged_trans == trans->transid)
4025 * when renaming files across directories, if the directory
4026 * there we're unlinking from gets fsync'd later on, there's
4027 * no way to find the destination directory later and fsync it
4028 * properly. So, we have to be conservative and force commits
4029 * so the new name gets discovered.
4034 /* we can safely do the unlink without any special recording */
4038 BTRFS_I(dir)->last_unlink_trans = trans->transid;
4042 * Call this after adding a new name for a file and it will properly
4043 * update the log to reflect the new name.
4045 * It will return zero if all goes well, and it will return 1 if a
4046 * full transaction commit is required.
4048 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
4049 struct inode *inode, struct inode *old_dir,
4050 struct dentry *parent)
4052 struct btrfs_root * root = BTRFS_I(inode)->root;
4055 * this will force the logging code to walk the dentry chain
4058 if (S_ISREG(inode->i_mode))
4059 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4062 * if this inode hasn't been logged and directory we're renaming it
4063 * from hasn't been logged, we don't need to log it
4065 if (BTRFS_I(inode)->logged_trans <=
4066 root->fs_info->last_trans_committed &&
4067 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
4068 root->fs_info->last_trans_committed))
4071 return btrfs_log_inode_parent(trans, root, inode, parent, 1);