2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/blkdev.h>
22 #include <linux/list_sort.h>
24 #include "transaction.h"
27 #include "print-tree.h"
33 /* magic values for the inode_only field in btrfs_log_inode:
35 * LOG_INODE_ALL means to log everything
36 * LOG_INODE_EXISTS means to log just enough to recreate the inode
39 #define LOG_INODE_ALL 0
40 #define LOG_INODE_EXISTS 1
43 * directory trouble cases
45 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
46 * log, we must force a full commit before doing an fsync of the directory
47 * where the unlink was done.
48 * ---> record transid of last unlink/rename per directory
52 * rename foo/some_dir foo2/some_dir
54 * fsync foo/some_dir/some_file
56 * The fsync above will unlink the original some_dir without recording
57 * it in its new location (foo2). After a crash, some_dir will be gone
58 * unless the fsync of some_file forces a full commit
60 * 2) we must log any new names for any file or dir that is in the fsync
61 * log. ---> check inode while renaming/linking.
63 * 2a) we must log any new names for any file or dir during rename
64 * when the directory they are being removed from was logged.
65 * ---> check inode and old parent dir during rename
67 * 2a is actually the more important variant. With the extra logging
68 * a crash might unlink the old name without recreating the new one
70 * 3) after a crash, we must go through any directories with a link count
71 * of zero and redo the rm -rf
78 * The directory f1 was fully removed from the FS, but fsync was never
79 * called on f1, only its parent dir. After a crash the rm -rf must
80 * be replayed. This must be able to recurse down the entire
81 * directory tree. The inode link count fixup code takes care of the
86 * stages for the tree walking. The first
87 * stage (0) is to only pin down the blocks we find
88 * the second stage (1) is to make sure that all the inodes
89 * we find in the log are created in the subvolume.
91 * The last stage is to deal with directories and links and extents
92 * and all the other fun semantics
94 #define LOG_WALK_PIN_ONLY 0
95 #define LOG_WALK_REPLAY_INODES 1
96 #define LOG_WALK_REPLAY_DIR_INDEX 2
97 #define LOG_WALK_REPLAY_ALL 3
99 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
100 struct btrfs_root *root, struct inode *inode,
102 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
103 struct btrfs_root *root,
104 struct btrfs_path *path, u64 objectid);
105 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
107 struct btrfs_root *log,
108 struct btrfs_path *path,
109 u64 dirid, int del_all);
112 * tree logging is a special write ahead log used to make sure that
113 * fsyncs and O_SYNCs can happen without doing full tree commits.
115 * Full tree commits are expensive because they require commonly
116 * modified blocks to be recowed, creating many dirty pages in the
117 * extent tree an 4x-6x higher write load than ext3.
119 * Instead of doing a tree commit on every fsync, we use the
120 * key ranges and transaction ids to find items for a given file or directory
121 * that have changed in this transaction. Those items are copied into
122 * a special tree (one per subvolume root), that tree is written to disk
123 * and then the fsync is considered complete.
125 * After a crash, items are copied out of the log-tree back into the
126 * subvolume tree. Any file data extents found are recorded in the extent
127 * allocation tree, and the log-tree freed.
129 * The log tree is read three times, once to pin down all the extents it is
130 * using in ram and once, once to create all the inodes logged in the tree
131 * and once to do all the other items.
135 * start a sub transaction and setup the log tree
136 * this increments the log tree writer count to make the people
137 * syncing the tree wait for us to finish
139 static int start_log_trans(struct btrfs_trans_handle *trans,
140 struct btrfs_root *root)
145 mutex_lock(&root->log_mutex);
146 if (root->log_root) {
147 if (!root->log_start_pid) {
148 root->log_start_pid = current->pid;
149 root->log_multiple_pids = false;
150 } else if (root->log_start_pid != current->pid) {
151 root->log_multiple_pids = true;
154 atomic_inc(&root->log_batch);
155 atomic_inc(&root->log_writers);
156 mutex_unlock(&root->log_mutex);
159 root->log_multiple_pids = false;
160 root->log_start_pid = current->pid;
161 mutex_lock(&root->fs_info->tree_log_mutex);
162 if (!root->fs_info->log_root_tree) {
163 ret = btrfs_init_log_root_tree(trans, root->fs_info);
167 if (err == 0 && !root->log_root) {
168 ret = btrfs_add_log_tree(trans, root);
172 mutex_unlock(&root->fs_info->tree_log_mutex);
173 atomic_inc(&root->log_batch);
174 atomic_inc(&root->log_writers);
175 mutex_unlock(&root->log_mutex);
180 * returns 0 if there was a log transaction running and we were able
181 * to join, or returns -ENOENT if there were not transactions
184 static int join_running_log_trans(struct btrfs_root *root)
192 mutex_lock(&root->log_mutex);
193 if (root->log_root) {
195 atomic_inc(&root->log_writers);
197 mutex_unlock(&root->log_mutex);
202 * This either makes the current running log transaction wait
203 * until you call btrfs_end_log_trans() or it makes any future
204 * log transactions wait until you call btrfs_end_log_trans()
206 int btrfs_pin_log_trans(struct btrfs_root *root)
210 mutex_lock(&root->log_mutex);
211 atomic_inc(&root->log_writers);
212 mutex_unlock(&root->log_mutex);
217 * indicate we're done making changes to the log tree
218 * and wake up anyone waiting to do a sync
220 void btrfs_end_log_trans(struct btrfs_root *root)
222 if (atomic_dec_and_test(&root->log_writers)) {
224 if (waitqueue_active(&root->log_writer_wait))
225 wake_up(&root->log_writer_wait);
231 * the walk control struct is used to pass state down the chain when
232 * processing the log tree. The stage field tells us which part
233 * of the log tree processing we are currently doing. The others
234 * are state fields used for that specific part
236 struct walk_control {
237 /* should we free the extent on disk when done? This is used
238 * at transaction commit time while freeing a log tree
242 /* should we write out the extent buffer? This is used
243 * while flushing the log tree to disk during a sync
247 /* should we wait for the extent buffer io to finish? Also used
248 * while flushing the log tree to disk for a sync
252 /* pin only walk, we record which extents on disk belong to the
257 /* what stage of the replay code we're currently in */
260 /* the root we are currently replaying */
261 struct btrfs_root *replay_dest;
263 /* the trans handle for the current replay */
264 struct btrfs_trans_handle *trans;
266 /* the function that gets used to process blocks we find in the
267 * tree. Note the extent_buffer might not be up to date when it is
268 * passed in, and it must be checked or read if you need the data
271 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
272 struct walk_control *wc, u64 gen);
276 * process_func used to pin down extents, write them or wait on them
278 static int process_one_buffer(struct btrfs_root *log,
279 struct extent_buffer *eb,
280 struct walk_control *wc, u64 gen)
285 * If this fs is mixed then we need to be able to process the leaves to
286 * pin down any logged extents, so we have to read the block.
288 if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) {
289 ret = btrfs_read_buffer(eb, gen);
295 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
298 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
299 if (wc->pin && btrfs_header_level(eb) == 0)
300 ret = btrfs_exclude_logged_extents(log, eb);
302 btrfs_write_tree_block(eb);
304 btrfs_wait_tree_block_writeback(eb);
310 * Item overwrite used by replay and tree logging. eb, slot and key all refer
311 * to the src data we are copying out.
313 * root is the tree we are copying into, and path is a scratch
314 * path for use in this function (it should be released on entry and
315 * will be released on exit).
317 * If the key is already in the destination tree the existing item is
318 * overwritten. If the existing item isn't big enough, it is extended.
319 * If it is too large, it is truncated.
321 * If the key isn't in the destination yet, a new item is inserted.
323 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
324 struct btrfs_root *root,
325 struct btrfs_path *path,
326 struct extent_buffer *eb, int slot,
327 struct btrfs_key *key)
331 u64 saved_i_size = 0;
332 int save_old_i_size = 0;
333 unsigned long src_ptr;
334 unsigned long dst_ptr;
335 int overwrite_root = 0;
336 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
338 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
341 item_size = btrfs_item_size_nr(eb, slot);
342 src_ptr = btrfs_item_ptr_offset(eb, slot);
344 /* look for the key in the destination tree */
345 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
352 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
354 if (dst_size != item_size)
357 if (item_size == 0) {
358 btrfs_release_path(path);
361 dst_copy = kmalloc(item_size, GFP_NOFS);
362 src_copy = kmalloc(item_size, GFP_NOFS);
363 if (!dst_copy || !src_copy) {
364 btrfs_release_path(path);
370 read_extent_buffer(eb, src_copy, src_ptr, item_size);
372 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
373 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
375 ret = memcmp(dst_copy, src_copy, item_size);
380 * they have the same contents, just return, this saves
381 * us from cowing blocks in the destination tree and doing
382 * extra writes that may not have been done by a previous
386 btrfs_release_path(path);
391 * We need to load the old nbytes into the inode so when we
392 * replay the extents we've logged we get the right nbytes.
395 struct btrfs_inode_item *item;
399 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
400 struct btrfs_inode_item);
401 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
402 item = btrfs_item_ptr(eb, slot,
403 struct btrfs_inode_item);
404 btrfs_set_inode_nbytes(eb, item, nbytes);
407 * If this is a directory we need to reset the i_size to
408 * 0 so that we can set it up properly when replaying
409 * the rest of the items in this log.
411 mode = btrfs_inode_mode(eb, item);
413 btrfs_set_inode_size(eb, item, 0);
415 } else if (inode_item) {
416 struct btrfs_inode_item *item;
420 * New inode, set nbytes to 0 so that the nbytes comes out
421 * properly when we replay the extents.
423 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
424 btrfs_set_inode_nbytes(eb, item, 0);
427 * If this is a directory we need to reset the i_size to 0 so
428 * that we can set it up properly when replaying the rest of
429 * the items in this log.
431 mode = btrfs_inode_mode(eb, item);
433 btrfs_set_inode_size(eb, item, 0);
436 btrfs_release_path(path);
437 /* try to insert the key into the destination tree */
438 ret = btrfs_insert_empty_item(trans, root, path,
441 /* make sure any existing item is the correct size */
442 if (ret == -EEXIST) {
444 found_size = btrfs_item_size_nr(path->nodes[0],
446 if (found_size > item_size)
447 btrfs_truncate_item(root, path, item_size, 1);
448 else if (found_size < item_size)
449 btrfs_extend_item(root, path,
450 item_size - found_size);
454 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
457 /* don't overwrite an existing inode if the generation number
458 * was logged as zero. This is done when the tree logging code
459 * is just logging an inode to make sure it exists after recovery.
461 * Also, don't overwrite i_size on directories during replay.
462 * log replay inserts and removes directory items based on the
463 * state of the tree found in the subvolume, and i_size is modified
466 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
467 struct btrfs_inode_item *src_item;
468 struct btrfs_inode_item *dst_item;
470 src_item = (struct btrfs_inode_item *)src_ptr;
471 dst_item = (struct btrfs_inode_item *)dst_ptr;
473 if (btrfs_inode_generation(eb, src_item) == 0)
476 if (overwrite_root &&
477 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
478 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
480 saved_i_size = btrfs_inode_size(path->nodes[0],
485 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
488 if (save_old_i_size) {
489 struct btrfs_inode_item *dst_item;
490 dst_item = (struct btrfs_inode_item *)dst_ptr;
491 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
494 /* make sure the generation is filled in */
495 if (key->type == BTRFS_INODE_ITEM_KEY) {
496 struct btrfs_inode_item *dst_item;
497 dst_item = (struct btrfs_inode_item *)dst_ptr;
498 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
499 btrfs_set_inode_generation(path->nodes[0], dst_item,
504 btrfs_mark_buffer_dirty(path->nodes[0]);
505 btrfs_release_path(path);
510 * simple helper to read an inode off the disk from a given root
511 * This can only be called for subvolume roots and not for the log
513 static noinline struct inode *read_one_inode(struct btrfs_root *root,
516 struct btrfs_key key;
519 key.objectid = objectid;
520 key.type = BTRFS_INODE_ITEM_KEY;
522 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
525 } else if (is_bad_inode(inode)) {
532 /* replays a single extent in 'eb' at 'slot' with 'key' into the
533 * subvolume 'root'. path is released on entry and should be released
536 * extents in the log tree have not been allocated out of the extent
537 * tree yet. So, this completes the allocation, taking a reference
538 * as required if the extent already exists or creating a new extent
539 * if it isn't in the extent allocation tree yet.
541 * The extent is inserted into the file, dropping any existing extents
542 * from the file that overlap the new one.
544 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
545 struct btrfs_root *root,
546 struct btrfs_path *path,
547 struct extent_buffer *eb, int slot,
548 struct btrfs_key *key)
552 u64 start = key->offset;
554 struct btrfs_file_extent_item *item;
555 struct inode *inode = NULL;
559 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
560 found_type = btrfs_file_extent_type(eb, item);
562 if (found_type == BTRFS_FILE_EXTENT_REG ||
563 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
564 nbytes = btrfs_file_extent_num_bytes(eb, item);
565 extent_end = start + nbytes;
568 * We don't add to the inodes nbytes if we are prealloc or a
571 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
573 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
574 size = btrfs_file_extent_inline_len(eb, item);
575 nbytes = btrfs_file_extent_ram_bytes(eb, item);
576 extent_end = ALIGN(start + size, root->sectorsize);
582 inode = read_one_inode(root, key->objectid);
589 * first check to see if we already have this extent in the
590 * file. This must be done before the btrfs_drop_extents run
591 * so we don't try to drop this extent.
593 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
597 (found_type == BTRFS_FILE_EXTENT_REG ||
598 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
599 struct btrfs_file_extent_item cmp1;
600 struct btrfs_file_extent_item cmp2;
601 struct btrfs_file_extent_item *existing;
602 struct extent_buffer *leaf;
604 leaf = path->nodes[0];
605 existing = btrfs_item_ptr(leaf, path->slots[0],
606 struct btrfs_file_extent_item);
608 read_extent_buffer(eb, &cmp1, (unsigned long)item,
610 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
614 * we already have a pointer to this exact extent,
615 * we don't have to do anything
617 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
618 btrfs_release_path(path);
622 btrfs_release_path(path);
624 /* drop any overlapping extents */
625 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
629 if (found_type == BTRFS_FILE_EXTENT_REG ||
630 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
632 unsigned long dest_offset;
633 struct btrfs_key ins;
635 ret = btrfs_insert_empty_item(trans, root, path, key,
639 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
641 copy_extent_buffer(path->nodes[0], eb, dest_offset,
642 (unsigned long)item, sizeof(*item));
644 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
645 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
646 ins.type = BTRFS_EXTENT_ITEM_KEY;
647 offset = key->offset - btrfs_file_extent_offset(eb, item);
649 if (ins.objectid > 0) {
652 LIST_HEAD(ordered_sums);
654 * is this extent already allocated in the extent
655 * allocation tree? If so, just add a reference
657 ret = btrfs_lookup_extent(root, ins.objectid,
660 ret = btrfs_inc_extent_ref(trans, root,
661 ins.objectid, ins.offset,
662 0, root->root_key.objectid,
663 key->objectid, offset, 0);
668 * insert the extent pointer in the extent
671 ret = btrfs_alloc_logged_file_extent(trans,
672 root, root->root_key.objectid,
673 key->objectid, offset, &ins);
677 btrfs_release_path(path);
679 if (btrfs_file_extent_compression(eb, item)) {
680 csum_start = ins.objectid;
681 csum_end = csum_start + ins.offset;
683 csum_start = ins.objectid +
684 btrfs_file_extent_offset(eb, item);
685 csum_end = csum_start +
686 btrfs_file_extent_num_bytes(eb, item);
689 ret = btrfs_lookup_csums_range(root->log_root,
690 csum_start, csum_end - 1,
694 while (!list_empty(&ordered_sums)) {
695 struct btrfs_ordered_sum *sums;
696 sums = list_entry(ordered_sums.next,
697 struct btrfs_ordered_sum,
700 ret = btrfs_csum_file_blocks(trans,
701 root->fs_info->csum_root,
703 list_del(&sums->list);
709 btrfs_release_path(path);
711 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
712 /* inline extents are easy, we just overwrite them */
713 ret = overwrite_item(trans, root, path, eb, slot, key);
718 inode_add_bytes(inode, nbytes);
719 ret = btrfs_update_inode(trans, root, inode);
727 * when cleaning up conflicts between the directory names in the
728 * subvolume, directory names in the log and directory names in the
729 * inode back references, we may have to unlink inodes from directories.
731 * This is a helper function to do the unlink of a specific directory
734 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
735 struct btrfs_root *root,
736 struct btrfs_path *path,
738 struct btrfs_dir_item *di)
743 struct extent_buffer *leaf;
744 struct btrfs_key location;
747 leaf = path->nodes[0];
749 btrfs_dir_item_key_to_cpu(leaf, di, &location);
750 name_len = btrfs_dir_name_len(leaf, di);
751 name = kmalloc(name_len, GFP_NOFS);
755 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
756 btrfs_release_path(path);
758 inode = read_one_inode(root, location.objectid);
764 ret = link_to_fixup_dir(trans, root, path, location.objectid);
768 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
772 ret = btrfs_run_delayed_items(trans, root);
780 * helper function to see if a given name and sequence number found
781 * in an inode back reference are already in a directory and correctly
782 * point to this inode
784 static noinline int inode_in_dir(struct btrfs_root *root,
785 struct btrfs_path *path,
786 u64 dirid, u64 objectid, u64 index,
787 const char *name, int name_len)
789 struct btrfs_dir_item *di;
790 struct btrfs_key location;
793 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
794 index, name, name_len, 0);
795 if (di && !IS_ERR(di)) {
796 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
797 if (location.objectid != objectid)
801 btrfs_release_path(path);
803 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
804 if (di && !IS_ERR(di)) {
805 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
806 if (location.objectid != objectid)
812 btrfs_release_path(path);
817 * helper function to check a log tree for a named back reference in
818 * an inode. This is used to decide if a back reference that is
819 * found in the subvolume conflicts with what we find in the log.
821 * inode backreferences may have multiple refs in a single item,
822 * during replay we process one reference at a time, and we don't
823 * want to delete valid links to a file from the subvolume if that
824 * link is also in the log.
826 static noinline int backref_in_log(struct btrfs_root *log,
827 struct btrfs_key *key,
829 char *name, int namelen)
831 struct btrfs_path *path;
832 struct btrfs_inode_ref *ref;
834 unsigned long ptr_end;
835 unsigned long name_ptr;
841 path = btrfs_alloc_path();
845 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
849 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
851 if (key->type == BTRFS_INODE_EXTREF_KEY) {
852 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
853 name, namelen, NULL))
859 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
860 ptr_end = ptr + item_size;
861 while (ptr < ptr_end) {
862 ref = (struct btrfs_inode_ref *)ptr;
863 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
864 if (found_name_len == namelen) {
865 name_ptr = (unsigned long)(ref + 1);
866 ret = memcmp_extent_buffer(path->nodes[0], name,
873 ptr = (unsigned long)(ref + 1) + found_name_len;
876 btrfs_free_path(path);
880 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
881 struct btrfs_root *root,
882 struct btrfs_path *path,
883 struct btrfs_root *log_root,
884 struct inode *dir, struct inode *inode,
885 struct extent_buffer *eb,
886 u64 inode_objectid, u64 parent_objectid,
887 u64 ref_index, char *name, int namelen,
893 struct extent_buffer *leaf;
894 struct btrfs_dir_item *di;
895 struct btrfs_key search_key;
896 struct btrfs_inode_extref *extref;
899 /* Search old style refs */
900 search_key.objectid = inode_objectid;
901 search_key.type = BTRFS_INODE_REF_KEY;
902 search_key.offset = parent_objectid;
903 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
905 struct btrfs_inode_ref *victim_ref;
907 unsigned long ptr_end;
909 leaf = path->nodes[0];
911 /* are we trying to overwrite a back ref for the root directory
912 * if so, just jump out, we're done
914 if (search_key.objectid == search_key.offset)
917 /* check all the names in this back reference to see
918 * if they are in the log. if so, we allow them to stay
919 * otherwise they must be unlinked as a conflict
921 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
922 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
923 while (ptr < ptr_end) {
924 victim_ref = (struct btrfs_inode_ref *)ptr;
925 victim_name_len = btrfs_inode_ref_name_len(leaf,
927 victim_name = kmalloc(victim_name_len, GFP_NOFS);
931 read_extent_buffer(leaf, victim_name,
932 (unsigned long)(victim_ref + 1),
935 if (!backref_in_log(log_root, &search_key,
939 btrfs_inc_nlink(inode);
940 btrfs_release_path(path);
942 ret = btrfs_unlink_inode(trans, root, dir,
948 ret = btrfs_run_delayed_items(trans, root);
956 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
960 * NOTE: we have searched root tree and checked the
961 * coresponding ref, it does not need to check again.
965 btrfs_release_path(path);
967 /* Same search but for extended refs */
968 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
969 inode_objectid, parent_objectid, 0,
971 if (!IS_ERR_OR_NULL(extref)) {
975 struct inode *victim_parent;
977 leaf = path->nodes[0];
979 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
980 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
982 while (cur_offset < item_size) {
983 extref = (struct btrfs_inode_extref *)base + cur_offset;
985 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
987 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
990 victim_name = kmalloc(victim_name_len, GFP_NOFS);
993 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
996 search_key.objectid = inode_objectid;
997 search_key.type = BTRFS_INODE_EXTREF_KEY;
998 search_key.offset = btrfs_extref_hash(parent_objectid,
1002 if (!backref_in_log(log_root, &search_key,
1003 parent_objectid, victim_name,
1006 victim_parent = read_one_inode(root,
1008 if (victim_parent) {
1009 btrfs_inc_nlink(inode);
1010 btrfs_release_path(path);
1012 ret = btrfs_unlink_inode(trans, root,
1018 ret = btrfs_run_delayed_items(
1021 iput(victim_parent);
1032 cur_offset += victim_name_len + sizeof(*extref);
1036 btrfs_release_path(path);
1038 /* look for a conflicting sequence number */
1039 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1040 ref_index, name, namelen, 0);
1041 if (di && !IS_ERR(di)) {
1042 ret = drop_one_dir_item(trans, root, path, dir, di);
1046 btrfs_release_path(path);
1048 /* look for a conflicing name */
1049 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1051 if (di && !IS_ERR(di)) {
1052 ret = drop_one_dir_item(trans, root, path, dir, di);
1056 btrfs_release_path(path);
1061 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1062 u32 *namelen, char **name, u64 *index,
1063 u64 *parent_objectid)
1065 struct btrfs_inode_extref *extref;
1067 extref = (struct btrfs_inode_extref *)ref_ptr;
1069 *namelen = btrfs_inode_extref_name_len(eb, extref);
1070 *name = kmalloc(*namelen, GFP_NOFS);
1074 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1077 *index = btrfs_inode_extref_index(eb, extref);
1078 if (parent_objectid)
1079 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1084 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1085 u32 *namelen, char **name, u64 *index)
1087 struct btrfs_inode_ref *ref;
1089 ref = (struct btrfs_inode_ref *)ref_ptr;
1091 *namelen = btrfs_inode_ref_name_len(eb, ref);
1092 *name = kmalloc(*namelen, GFP_NOFS);
1096 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1098 *index = btrfs_inode_ref_index(eb, ref);
1104 * replay one inode back reference item found in the log tree.
1105 * eb, slot and key refer to the buffer and key found in the log tree.
1106 * root is the destination we are replaying into, and path is for temp
1107 * use by this function. (it should be released on return).
1109 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1110 struct btrfs_root *root,
1111 struct btrfs_root *log,
1112 struct btrfs_path *path,
1113 struct extent_buffer *eb, int slot,
1114 struct btrfs_key *key)
1116 struct inode *dir = NULL;
1117 struct inode *inode = NULL;
1118 unsigned long ref_ptr;
1119 unsigned long ref_end;
1123 int search_done = 0;
1124 int log_ref_ver = 0;
1125 u64 parent_objectid;
1128 int ref_struct_size;
1130 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1131 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1133 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1134 struct btrfs_inode_extref *r;
1136 ref_struct_size = sizeof(struct btrfs_inode_extref);
1138 r = (struct btrfs_inode_extref *)ref_ptr;
1139 parent_objectid = btrfs_inode_extref_parent(eb, r);
1141 ref_struct_size = sizeof(struct btrfs_inode_ref);
1142 parent_objectid = key->offset;
1144 inode_objectid = key->objectid;
1147 * it is possible that we didn't log all the parent directories
1148 * for a given inode. If we don't find the dir, just don't
1149 * copy the back ref in. The link count fixup code will take
1152 dir = read_one_inode(root, parent_objectid);
1158 inode = read_one_inode(root, inode_objectid);
1164 while (ref_ptr < ref_end) {
1166 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1167 &ref_index, &parent_objectid);
1169 * parent object can change from one array
1173 dir = read_one_inode(root, parent_objectid);
1179 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1185 /* if we already have a perfect match, we're done */
1186 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
1187 ref_index, name, namelen)) {
1189 * look for a conflicting back reference in the
1190 * metadata. if we find one we have to unlink that name
1191 * of the file before we add our new link. Later on, we
1192 * overwrite any existing back reference, and we don't
1193 * want to create dangling pointers in the directory.
1197 ret = __add_inode_ref(trans, root, path, log,
1201 ref_index, name, namelen,
1210 /* insert our name */
1211 ret = btrfs_add_link(trans, dir, inode, name, namelen,
1216 btrfs_update_inode(trans, root, inode);
1219 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1228 /* finally write the back reference in the inode */
1229 ret = overwrite_item(trans, root, path, eb, slot, key);
1231 btrfs_release_path(path);
1238 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1239 struct btrfs_root *root, u64 offset)
1242 ret = btrfs_find_orphan_item(root, offset);
1244 ret = btrfs_insert_orphan_item(trans, root, offset);
1248 static int count_inode_extrefs(struct btrfs_root *root,
1249 struct inode *inode, struct btrfs_path *path)
1253 unsigned int nlink = 0;
1256 u64 inode_objectid = btrfs_ino(inode);
1259 struct btrfs_inode_extref *extref;
1260 struct extent_buffer *leaf;
1263 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1268 leaf = path->nodes[0];
1269 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1270 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1272 while (cur_offset < item_size) {
1273 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1274 name_len = btrfs_inode_extref_name_len(leaf, extref);
1278 cur_offset += name_len + sizeof(*extref);
1282 btrfs_release_path(path);
1284 btrfs_release_path(path);
1291 static int count_inode_refs(struct btrfs_root *root,
1292 struct inode *inode, struct btrfs_path *path)
1295 struct btrfs_key key;
1296 unsigned int nlink = 0;
1298 unsigned long ptr_end;
1300 u64 ino = btrfs_ino(inode);
1303 key.type = BTRFS_INODE_REF_KEY;
1304 key.offset = (u64)-1;
1307 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1311 if (path->slots[0] == 0)
1315 btrfs_item_key_to_cpu(path->nodes[0], &key,
1317 if (key.objectid != ino ||
1318 key.type != BTRFS_INODE_REF_KEY)
1320 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1321 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1323 while (ptr < ptr_end) {
1324 struct btrfs_inode_ref *ref;
1326 ref = (struct btrfs_inode_ref *)ptr;
1327 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1329 ptr = (unsigned long)(ref + 1) + name_len;
1333 if (key.offset == 0)
1336 btrfs_release_path(path);
1338 btrfs_release_path(path);
1344 * There are a few corners where the link count of the file can't
1345 * be properly maintained during replay. So, instead of adding
1346 * lots of complexity to the log code, we just scan the backrefs
1347 * for any file that has been through replay.
1349 * The scan will update the link count on the inode to reflect the
1350 * number of back refs found. If it goes down to zero, the iput
1351 * will free the inode.
1353 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1354 struct btrfs_root *root,
1355 struct inode *inode)
1357 struct btrfs_path *path;
1360 u64 ino = btrfs_ino(inode);
1362 path = btrfs_alloc_path();
1366 ret = count_inode_refs(root, inode, path);
1372 ret = count_inode_extrefs(root, inode, path);
1383 if (nlink != inode->i_nlink) {
1384 set_nlink(inode, nlink);
1385 btrfs_update_inode(trans, root, inode);
1387 BTRFS_I(inode)->index_cnt = (u64)-1;
1389 if (inode->i_nlink == 0) {
1390 if (S_ISDIR(inode->i_mode)) {
1391 ret = replay_dir_deletes(trans, root, NULL, path,
1396 ret = insert_orphan_item(trans, root, ino);
1400 btrfs_free_path(path);
1404 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1405 struct btrfs_root *root,
1406 struct btrfs_path *path)
1409 struct btrfs_key key;
1410 struct inode *inode;
1412 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1413 key.type = BTRFS_ORPHAN_ITEM_KEY;
1414 key.offset = (u64)-1;
1416 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1421 if (path->slots[0] == 0)
1426 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1427 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1428 key.type != BTRFS_ORPHAN_ITEM_KEY)
1431 ret = btrfs_del_item(trans, root, path);
1435 btrfs_release_path(path);
1436 inode = read_one_inode(root, key.offset);
1440 ret = fixup_inode_link_count(trans, root, inode);
1446 * fixup on a directory may create new entries,
1447 * make sure we always look for the highset possible
1450 key.offset = (u64)-1;
1454 btrfs_release_path(path);
1460 * record a given inode in the fixup dir so we can check its link
1461 * count when replay is done. The link count is incremented here
1462 * so the inode won't go away until we check it
1464 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1465 struct btrfs_root *root,
1466 struct btrfs_path *path,
1469 struct btrfs_key key;
1471 struct inode *inode;
1473 inode = read_one_inode(root, objectid);
1477 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1478 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1479 key.offset = objectid;
1481 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1483 btrfs_release_path(path);
1485 if (!inode->i_nlink)
1486 set_nlink(inode, 1);
1488 btrfs_inc_nlink(inode);
1489 ret = btrfs_update_inode(trans, root, inode);
1490 } else if (ret == -EEXIST) {
1493 BUG(); /* Logic Error */
1501 * when replaying the log for a directory, we only insert names
1502 * for inodes that actually exist. This means an fsync on a directory
1503 * does not implicitly fsync all the new files in it
1505 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1506 struct btrfs_root *root,
1507 struct btrfs_path *path,
1508 u64 dirid, u64 index,
1509 char *name, int name_len, u8 type,
1510 struct btrfs_key *location)
1512 struct inode *inode;
1516 inode = read_one_inode(root, location->objectid);
1520 dir = read_one_inode(root, dirid);
1526 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1528 /* FIXME, put inode into FIXUP list */
1536 * take a single entry in a log directory item and replay it into
1539 * if a conflicting item exists in the subdirectory already,
1540 * the inode it points to is unlinked and put into the link count
1543 * If a name from the log points to a file or directory that does
1544 * not exist in the FS, it is skipped. fsyncs on directories
1545 * do not force down inodes inside that directory, just changes to the
1546 * names or unlinks in a directory.
1548 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1549 struct btrfs_root *root,
1550 struct btrfs_path *path,
1551 struct extent_buffer *eb,
1552 struct btrfs_dir_item *di,
1553 struct btrfs_key *key)
1557 struct btrfs_dir_item *dst_di;
1558 struct btrfs_key found_key;
1559 struct btrfs_key log_key;
1564 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1566 dir = read_one_inode(root, key->objectid);
1570 name_len = btrfs_dir_name_len(eb, di);
1571 name = kmalloc(name_len, GFP_NOFS);
1577 log_type = btrfs_dir_type(eb, di);
1578 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1581 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1582 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1587 btrfs_release_path(path);
1589 if (key->type == BTRFS_DIR_ITEM_KEY) {
1590 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1592 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1593 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1602 if (IS_ERR_OR_NULL(dst_di)) {
1603 /* we need a sequence number to insert, so we only
1604 * do inserts for the BTRFS_DIR_INDEX_KEY types
1606 if (key->type != BTRFS_DIR_INDEX_KEY)
1611 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1612 /* the existing item matches the logged item */
1613 if (found_key.objectid == log_key.objectid &&
1614 found_key.type == log_key.type &&
1615 found_key.offset == log_key.offset &&
1616 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1621 * don't drop the conflicting directory entry if the inode
1622 * for the new entry doesn't exist
1627 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1631 if (key->type == BTRFS_DIR_INDEX_KEY)
1634 btrfs_release_path(path);
1635 if (!ret && update_size) {
1636 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1637 ret = btrfs_update_inode(trans, root, dir);
1644 btrfs_release_path(path);
1645 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1646 name, name_len, log_type, &log_key);
1647 if (ret && ret != -ENOENT)
1649 update_size = false;
1655 * find all the names in a directory item and reconcile them into
1656 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1657 * one name in a directory item, but the same code gets used for
1658 * both directory index types
1660 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1661 struct btrfs_root *root,
1662 struct btrfs_path *path,
1663 struct extent_buffer *eb, int slot,
1664 struct btrfs_key *key)
1667 u32 item_size = btrfs_item_size_nr(eb, slot);
1668 struct btrfs_dir_item *di;
1671 unsigned long ptr_end;
1673 ptr = btrfs_item_ptr_offset(eb, slot);
1674 ptr_end = ptr + item_size;
1675 while (ptr < ptr_end) {
1676 di = (struct btrfs_dir_item *)ptr;
1677 if (verify_dir_item(root, eb, di))
1679 name_len = btrfs_dir_name_len(eb, di);
1680 ret = replay_one_name(trans, root, path, eb, di, key);
1683 ptr = (unsigned long)(di + 1);
1690 * directory replay has two parts. There are the standard directory
1691 * items in the log copied from the subvolume, and range items
1692 * created in the log while the subvolume was logged.
1694 * The range items tell us which parts of the key space the log
1695 * is authoritative for. During replay, if a key in the subvolume
1696 * directory is in a logged range item, but not actually in the log
1697 * that means it was deleted from the directory before the fsync
1698 * and should be removed.
1700 static noinline int find_dir_range(struct btrfs_root *root,
1701 struct btrfs_path *path,
1702 u64 dirid, int key_type,
1703 u64 *start_ret, u64 *end_ret)
1705 struct btrfs_key key;
1707 struct btrfs_dir_log_item *item;
1711 if (*start_ret == (u64)-1)
1714 key.objectid = dirid;
1715 key.type = key_type;
1716 key.offset = *start_ret;
1718 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1722 if (path->slots[0] == 0)
1727 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1729 if (key.type != key_type || key.objectid != dirid) {
1733 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1734 struct btrfs_dir_log_item);
1735 found_end = btrfs_dir_log_end(path->nodes[0], item);
1737 if (*start_ret >= key.offset && *start_ret <= found_end) {
1739 *start_ret = key.offset;
1740 *end_ret = found_end;
1745 /* check the next slot in the tree to see if it is a valid item */
1746 nritems = btrfs_header_nritems(path->nodes[0]);
1747 if (path->slots[0] >= nritems) {
1748 ret = btrfs_next_leaf(root, path);
1755 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1757 if (key.type != key_type || key.objectid != dirid) {
1761 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1762 struct btrfs_dir_log_item);
1763 found_end = btrfs_dir_log_end(path->nodes[0], item);
1764 *start_ret = key.offset;
1765 *end_ret = found_end;
1768 btrfs_release_path(path);
1773 * this looks for a given directory item in the log. If the directory
1774 * item is not in the log, the item is removed and the inode it points
1777 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1778 struct btrfs_root *root,
1779 struct btrfs_root *log,
1780 struct btrfs_path *path,
1781 struct btrfs_path *log_path,
1783 struct btrfs_key *dir_key)
1786 struct extent_buffer *eb;
1789 struct btrfs_dir_item *di;
1790 struct btrfs_dir_item *log_di;
1793 unsigned long ptr_end;
1795 struct inode *inode;
1796 struct btrfs_key location;
1799 eb = path->nodes[0];
1800 slot = path->slots[0];
1801 item_size = btrfs_item_size_nr(eb, slot);
1802 ptr = btrfs_item_ptr_offset(eb, slot);
1803 ptr_end = ptr + item_size;
1804 while (ptr < ptr_end) {
1805 di = (struct btrfs_dir_item *)ptr;
1806 if (verify_dir_item(root, eb, di)) {
1811 name_len = btrfs_dir_name_len(eb, di);
1812 name = kmalloc(name_len, GFP_NOFS);
1817 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1820 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1821 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1824 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1825 log_di = btrfs_lookup_dir_index_item(trans, log,
1831 if (IS_ERR_OR_NULL(log_di)) {
1832 btrfs_dir_item_key_to_cpu(eb, di, &location);
1833 btrfs_release_path(path);
1834 btrfs_release_path(log_path);
1835 inode = read_one_inode(root, location.objectid);
1841 ret = link_to_fixup_dir(trans, root,
1842 path, location.objectid);
1849 btrfs_inc_nlink(inode);
1850 ret = btrfs_unlink_inode(trans, root, dir, inode,
1853 ret = btrfs_run_delayed_items(trans, root);
1859 /* there might still be more names under this key
1860 * check and repeat if required
1862 ret = btrfs_search_slot(NULL, root, dir_key, path,
1869 btrfs_release_path(log_path);
1872 ptr = (unsigned long)(di + 1);
1877 btrfs_release_path(path);
1878 btrfs_release_path(log_path);
1883 * deletion replay happens before we copy any new directory items
1884 * out of the log or out of backreferences from inodes. It
1885 * scans the log to find ranges of keys that log is authoritative for,
1886 * and then scans the directory to find items in those ranges that are
1887 * not present in the log.
1889 * Anything we don't find in the log is unlinked and removed from the
1892 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1893 struct btrfs_root *root,
1894 struct btrfs_root *log,
1895 struct btrfs_path *path,
1896 u64 dirid, int del_all)
1900 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1902 struct btrfs_key dir_key;
1903 struct btrfs_key found_key;
1904 struct btrfs_path *log_path;
1907 dir_key.objectid = dirid;
1908 dir_key.type = BTRFS_DIR_ITEM_KEY;
1909 log_path = btrfs_alloc_path();
1913 dir = read_one_inode(root, dirid);
1914 /* it isn't an error if the inode isn't there, that can happen
1915 * because we replay the deletes before we copy in the inode item
1919 btrfs_free_path(log_path);
1927 range_end = (u64)-1;
1929 ret = find_dir_range(log, path, dirid, key_type,
1930 &range_start, &range_end);
1935 dir_key.offset = range_start;
1938 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1943 nritems = btrfs_header_nritems(path->nodes[0]);
1944 if (path->slots[0] >= nritems) {
1945 ret = btrfs_next_leaf(root, path);
1949 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1951 if (found_key.objectid != dirid ||
1952 found_key.type != dir_key.type)
1955 if (found_key.offset > range_end)
1958 ret = check_item_in_log(trans, root, log, path,
1963 if (found_key.offset == (u64)-1)
1965 dir_key.offset = found_key.offset + 1;
1967 btrfs_release_path(path);
1968 if (range_end == (u64)-1)
1970 range_start = range_end + 1;
1975 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1976 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1977 dir_key.type = BTRFS_DIR_INDEX_KEY;
1978 btrfs_release_path(path);
1982 btrfs_release_path(path);
1983 btrfs_free_path(log_path);
1989 * the process_func used to replay items from the log tree. This
1990 * gets called in two different stages. The first stage just looks
1991 * for inodes and makes sure they are all copied into the subvolume.
1993 * The second stage copies all the other item types from the log into
1994 * the subvolume. The two stage approach is slower, but gets rid of
1995 * lots of complexity around inodes referencing other inodes that exist
1996 * only in the log (references come from either directory items or inode
1999 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2000 struct walk_control *wc, u64 gen)
2003 struct btrfs_path *path;
2004 struct btrfs_root *root = wc->replay_dest;
2005 struct btrfs_key key;
2010 ret = btrfs_read_buffer(eb, gen);
2014 level = btrfs_header_level(eb);
2019 path = btrfs_alloc_path();
2023 nritems = btrfs_header_nritems(eb);
2024 for (i = 0; i < nritems; i++) {
2025 btrfs_item_key_to_cpu(eb, &key, i);
2027 /* inode keys are done during the first stage */
2028 if (key.type == BTRFS_INODE_ITEM_KEY &&
2029 wc->stage == LOG_WALK_REPLAY_INODES) {
2030 struct btrfs_inode_item *inode_item;
2033 inode_item = btrfs_item_ptr(eb, i,
2034 struct btrfs_inode_item);
2035 mode = btrfs_inode_mode(eb, inode_item);
2036 if (S_ISDIR(mode)) {
2037 ret = replay_dir_deletes(wc->trans,
2038 root, log, path, key.objectid, 0);
2042 ret = overwrite_item(wc->trans, root, path,
2047 /* for regular files, make sure corresponding
2048 * orhpan item exist. extents past the new EOF
2049 * will be truncated later by orphan cleanup.
2051 if (S_ISREG(mode)) {
2052 ret = insert_orphan_item(wc->trans, root,
2058 ret = link_to_fixup_dir(wc->trans, root,
2059 path, key.objectid);
2064 if (key.type == BTRFS_DIR_INDEX_KEY &&
2065 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2066 ret = replay_one_dir_item(wc->trans, root, path,
2072 if (wc->stage < LOG_WALK_REPLAY_ALL)
2075 /* these keys are simply copied */
2076 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2077 ret = overwrite_item(wc->trans, root, path,
2081 } else if (key.type == BTRFS_INODE_REF_KEY ||
2082 key.type == BTRFS_INODE_EXTREF_KEY) {
2083 ret = add_inode_ref(wc->trans, root, log, path,
2085 if (ret && ret != -ENOENT)
2088 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2089 ret = replay_one_extent(wc->trans, root, path,
2093 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2094 ret = replay_one_dir_item(wc->trans, root, path,
2100 btrfs_free_path(path);
2104 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2105 struct btrfs_root *root,
2106 struct btrfs_path *path, int *level,
2107 struct walk_control *wc)
2112 struct extent_buffer *next;
2113 struct extent_buffer *cur;
2114 struct extent_buffer *parent;
2118 WARN_ON(*level < 0);
2119 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2121 while (*level > 0) {
2122 WARN_ON(*level < 0);
2123 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2124 cur = path->nodes[*level];
2126 if (btrfs_header_level(cur) != *level)
2129 if (path->slots[*level] >=
2130 btrfs_header_nritems(cur))
2133 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2134 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2135 blocksize = btrfs_level_size(root, *level - 1);
2137 parent = path->nodes[*level];
2138 root_owner = btrfs_header_owner(parent);
2140 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
2145 ret = wc->process_func(root, next, wc, ptr_gen);
2147 free_extent_buffer(next);
2151 path->slots[*level]++;
2153 ret = btrfs_read_buffer(next, ptr_gen);
2155 free_extent_buffer(next);
2160 btrfs_tree_lock(next);
2161 btrfs_set_lock_blocking(next);
2162 clean_tree_block(trans, root, next);
2163 btrfs_wait_tree_block_writeback(next);
2164 btrfs_tree_unlock(next);
2167 WARN_ON(root_owner !=
2168 BTRFS_TREE_LOG_OBJECTID);
2169 ret = btrfs_free_and_pin_reserved_extent(root,
2172 free_extent_buffer(next);
2176 free_extent_buffer(next);
2179 ret = btrfs_read_buffer(next, ptr_gen);
2181 free_extent_buffer(next);
2185 WARN_ON(*level <= 0);
2186 if (path->nodes[*level-1])
2187 free_extent_buffer(path->nodes[*level-1]);
2188 path->nodes[*level-1] = next;
2189 *level = btrfs_header_level(next);
2190 path->slots[*level] = 0;
2193 WARN_ON(*level < 0);
2194 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2196 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2202 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2203 struct btrfs_root *root,
2204 struct btrfs_path *path, int *level,
2205 struct walk_control *wc)
2212 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2213 slot = path->slots[i];
2214 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2217 WARN_ON(*level == 0);
2220 struct extent_buffer *parent;
2221 if (path->nodes[*level] == root->node)
2222 parent = path->nodes[*level];
2224 parent = path->nodes[*level + 1];
2226 root_owner = btrfs_header_owner(parent);
2227 ret = wc->process_func(root, path->nodes[*level], wc,
2228 btrfs_header_generation(path->nodes[*level]));
2233 struct extent_buffer *next;
2235 next = path->nodes[*level];
2238 btrfs_tree_lock(next);
2239 btrfs_set_lock_blocking(next);
2240 clean_tree_block(trans, root, next);
2241 btrfs_wait_tree_block_writeback(next);
2242 btrfs_tree_unlock(next);
2245 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2246 ret = btrfs_free_and_pin_reserved_extent(root,
2247 path->nodes[*level]->start,
2248 path->nodes[*level]->len);
2252 free_extent_buffer(path->nodes[*level]);
2253 path->nodes[*level] = NULL;
2261 * drop the reference count on the tree rooted at 'snap'. This traverses
2262 * the tree freeing any blocks that have a ref count of zero after being
2265 static int walk_log_tree(struct btrfs_trans_handle *trans,
2266 struct btrfs_root *log, struct walk_control *wc)
2271 struct btrfs_path *path;
2274 path = btrfs_alloc_path();
2278 level = btrfs_header_level(log->node);
2280 path->nodes[level] = log->node;
2281 extent_buffer_get(log->node);
2282 path->slots[level] = 0;
2285 wret = walk_down_log_tree(trans, log, path, &level, wc);
2293 wret = walk_up_log_tree(trans, log, path, &level, wc);
2302 /* was the root node processed? if not, catch it here */
2303 if (path->nodes[orig_level]) {
2304 ret = wc->process_func(log, path->nodes[orig_level], wc,
2305 btrfs_header_generation(path->nodes[orig_level]));
2309 struct extent_buffer *next;
2311 next = path->nodes[orig_level];
2314 btrfs_tree_lock(next);
2315 btrfs_set_lock_blocking(next);
2316 clean_tree_block(trans, log, next);
2317 btrfs_wait_tree_block_writeback(next);
2318 btrfs_tree_unlock(next);
2321 WARN_ON(log->root_key.objectid !=
2322 BTRFS_TREE_LOG_OBJECTID);
2323 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2331 btrfs_free_path(path);
2336 * helper function to update the item for a given subvolumes log root
2337 * in the tree of log roots
2339 static int update_log_root(struct btrfs_trans_handle *trans,
2340 struct btrfs_root *log)
2344 if (log->log_transid == 1) {
2345 /* insert root item on the first sync */
2346 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2347 &log->root_key, &log->root_item);
2349 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2350 &log->root_key, &log->root_item);
2355 static int wait_log_commit(struct btrfs_trans_handle *trans,
2356 struct btrfs_root *root, unsigned long transid)
2359 int index = transid % 2;
2362 * we only allow two pending log transactions at a time,
2363 * so we know that if ours is more than 2 older than the
2364 * current transaction, we're done
2367 prepare_to_wait(&root->log_commit_wait[index],
2368 &wait, TASK_UNINTERRUPTIBLE);
2369 mutex_unlock(&root->log_mutex);
2371 if (root->fs_info->last_trans_log_full_commit !=
2372 trans->transid && root->log_transid < transid + 2 &&
2373 atomic_read(&root->log_commit[index]))
2376 finish_wait(&root->log_commit_wait[index], &wait);
2377 mutex_lock(&root->log_mutex);
2378 } while (root->fs_info->last_trans_log_full_commit !=
2379 trans->transid && root->log_transid < transid + 2 &&
2380 atomic_read(&root->log_commit[index]));
2384 static void wait_for_writer(struct btrfs_trans_handle *trans,
2385 struct btrfs_root *root)
2388 while (root->fs_info->last_trans_log_full_commit !=
2389 trans->transid && atomic_read(&root->log_writers)) {
2390 prepare_to_wait(&root->log_writer_wait,
2391 &wait, TASK_UNINTERRUPTIBLE);
2392 mutex_unlock(&root->log_mutex);
2393 if (root->fs_info->last_trans_log_full_commit !=
2394 trans->transid && atomic_read(&root->log_writers))
2396 mutex_lock(&root->log_mutex);
2397 finish_wait(&root->log_writer_wait, &wait);
2402 * btrfs_sync_log does sends a given tree log down to the disk and
2403 * updates the super blocks to record it. When this call is done,
2404 * you know that any inodes previously logged are safely on disk only
2407 * Any other return value means you need to call btrfs_commit_transaction.
2408 * Some of the edge cases for fsyncing directories that have had unlinks
2409 * or renames done in the past mean that sometimes the only safe
2410 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2411 * that has happened.
2413 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2414 struct btrfs_root *root)
2420 struct btrfs_root *log = root->log_root;
2421 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2422 unsigned long log_transid = 0;
2423 struct blk_plug plug;
2425 mutex_lock(&root->log_mutex);
2426 log_transid = root->log_transid;
2427 index1 = root->log_transid % 2;
2428 if (atomic_read(&root->log_commit[index1])) {
2429 wait_log_commit(trans, root, root->log_transid);
2430 mutex_unlock(&root->log_mutex);
2433 atomic_set(&root->log_commit[index1], 1);
2435 /* wait for previous tree log sync to complete */
2436 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2437 wait_log_commit(trans, root, root->log_transid - 1);
2439 int batch = atomic_read(&root->log_batch);
2440 /* when we're on an ssd, just kick the log commit out */
2441 if (!btrfs_test_opt(root, SSD) && root->log_multiple_pids) {
2442 mutex_unlock(&root->log_mutex);
2443 schedule_timeout_uninterruptible(1);
2444 mutex_lock(&root->log_mutex);
2446 wait_for_writer(trans, root);
2447 if (batch == atomic_read(&root->log_batch))
2451 /* bail out if we need to do a full commit */
2452 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2454 btrfs_free_logged_extents(log, log_transid);
2455 mutex_unlock(&root->log_mutex);
2459 if (log_transid % 2 == 0)
2460 mark = EXTENT_DIRTY;
2464 /* we start IO on all the marked extents here, but we don't actually
2465 * wait for them until later.
2467 blk_start_plug(&plug);
2468 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2470 blk_finish_plug(&plug);
2471 btrfs_abort_transaction(trans, root, ret);
2472 btrfs_free_logged_extents(log, log_transid);
2473 mutex_unlock(&root->log_mutex);
2477 btrfs_set_root_node(&log->root_item, log->node);
2479 root->log_transid++;
2480 log->log_transid = root->log_transid;
2481 root->log_start_pid = 0;
2484 * IO has been started, blocks of the log tree have WRITTEN flag set
2485 * in their headers. new modifications of the log will be written to
2486 * new positions. so it's safe to allow log writers to go in.
2488 mutex_unlock(&root->log_mutex);
2490 mutex_lock(&log_root_tree->log_mutex);
2491 atomic_inc(&log_root_tree->log_batch);
2492 atomic_inc(&log_root_tree->log_writers);
2493 mutex_unlock(&log_root_tree->log_mutex);
2495 ret = update_log_root(trans, log);
2497 mutex_lock(&log_root_tree->log_mutex);
2498 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2500 if (waitqueue_active(&log_root_tree->log_writer_wait))
2501 wake_up(&log_root_tree->log_writer_wait);
2505 blk_finish_plug(&plug);
2506 if (ret != -ENOSPC) {
2507 btrfs_abort_transaction(trans, root, ret);
2508 mutex_unlock(&log_root_tree->log_mutex);
2511 root->fs_info->last_trans_log_full_commit = trans->transid;
2512 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2513 btrfs_free_logged_extents(log, log_transid);
2514 mutex_unlock(&log_root_tree->log_mutex);
2519 index2 = log_root_tree->log_transid % 2;
2520 if (atomic_read(&log_root_tree->log_commit[index2])) {
2521 blk_finish_plug(&plug);
2522 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2523 wait_log_commit(trans, log_root_tree,
2524 log_root_tree->log_transid);
2525 btrfs_free_logged_extents(log, log_transid);
2526 mutex_unlock(&log_root_tree->log_mutex);
2530 atomic_set(&log_root_tree->log_commit[index2], 1);
2532 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2533 wait_log_commit(trans, log_root_tree,
2534 log_root_tree->log_transid - 1);
2537 wait_for_writer(trans, log_root_tree);
2540 * now that we've moved on to the tree of log tree roots,
2541 * check the full commit flag again
2543 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2544 blk_finish_plug(&plug);
2545 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2546 btrfs_free_logged_extents(log, log_transid);
2547 mutex_unlock(&log_root_tree->log_mutex);
2549 goto out_wake_log_root;
2552 ret = btrfs_write_marked_extents(log_root_tree,
2553 &log_root_tree->dirty_log_pages,
2554 EXTENT_DIRTY | EXTENT_NEW);
2555 blk_finish_plug(&plug);
2557 btrfs_abort_transaction(trans, root, ret);
2558 btrfs_free_logged_extents(log, log_transid);
2559 mutex_unlock(&log_root_tree->log_mutex);
2560 goto out_wake_log_root;
2562 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2563 btrfs_wait_marked_extents(log_root_tree,
2564 &log_root_tree->dirty_log_pages,
2565 EXTENT_NEW | EXTENT_DIRTY);
2566 btrfs_wait_logged_extents(log, log_transid);
2568 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2569 log_root_tree->node->start);
2570 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2571 btrfs_header_level(log_root_tree->node));
2573 log_root_tree->log_transid++;
2576 mutex_unlock(&log_root_tree->log_mutex);
2579 * nobody else is going to jump in and write the the ctree
2580 * super here because the log_commit atomic below is protecting
2581 * us. We must be called with a transaction handle pinning
2582 * the running transaction open, so a full commit can't hop
2583 * in and cause problems either.
2585 btrfs_scrub_pause_super(root);
2586 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2587 btrfs_scrub_continue_super(root);
2589 btrfs_abort_transaction(trans, root, ret);
2590 goto out_wake_log_root;
2593 mutex_lock(&root->log_mutex);
2594 if (root->last_log_commit < log_transid)
2595 root->last_log_commit = log_transid;
2596 mutex_unlock(&root->log_mutex);
2599 atomic_set(&log_root_tree->log_commit[index2], 0);
2601 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2602 wake_up(&log_root_tree->log_commit_wait[index2]);
2604 atomic_set(&root->log_commit[index1], 0);
2606 if (waitqueue_active(&root->log_commit_wait[index1]))
2607 wake_up(&root->log_commit_wait[index1]);
2611 static void free_log_tree(struct btrfs_trans_handle *trans,
2612 struct btrfs_root *log)
2617 struct walk_control wc = {
2619 .process_func = process_one_buffer
2622 ret = walk_log_tree(trans, log, &wc);
2623 /* I don't think this can happen but just in case */
2625 btrfs_abort_transaction(trans, log, ret);
2628 ret = find_first_extent_bit(&log->dirty_log_pages,
2629 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2634 clear_extent_bits(&log->dirty_log_pages, start, end,
2635 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2639 * We may have short-circuited the log tree with the full commit logic
2640 * and left ordered extents on our list, so clear these out to keep us
2641 * from leaking inodes and memory.
2643 btrfs_free_logged_extents(log, 0);
2644 btrfs_free_logged_extents(log, 1);
2646 free_extent_buffer(log->node);
2651 * free all the extents used by the tree log. This should be called
2652 * at commit time of the full transaction
2654 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2656 if (root->log_root) {
2657 free_log_tree(trans, root->log_root);
2658 root->log_root = NULL;
2663 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2664 struct btrfs_fs_info *fs_info)
2666 if (fs_info->log_root_tree) {
2667 free_log_tree(trans, fs_info->log_root_tree);
2668 fs_info->log_root_tree = NULL;
2674 * If both a file and directory are logged, and unlinks or renames are
2675 * mixed in, we have a few interesting corners:
2677 * create file X in dir Y
2678 * link file X to X.link in dir Y
2680 * unlink file X but leave X.link
2683 * After a crash we would expect only X.link to exist. But file X
2684 * didn't get fsync'd again so the log has back refs for X and X.link.
2686 * We solve this by removing directory entries and inode backrefs from the
2687 * log when a file that was logged in the current transaction is
2688 * unlinked. Any later fsync will include the updated log entries, and
2689 * we'll be able to reconstruct the proper directory items from backrefs.
2691 * This optimizations allows us to avoid relogging the entire inode
2692 * or the entire directory.
2694 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2695 struct btrfs_root *root,
2696 const char *name, int name_len,
2697 struct inode *dir, u64 index)
2699 struct btrfs_root *log;
2700 struct btrfs_dir_item *di;
2701 struct btrfs_path *path;
2705 u64 dir_ino = btrfs_ino(dir);
2707 if (BTRFS_I(dir)->logged_trans < trans->transid)
2710 ret = join_running_log_trans(root);
2714 mutex_lock(&BTRFS_I(dir)->log_mutex);
2716 log = root->log_root;
2717 path = btrfs_alloc_path();
2723 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
2724 name, name_len, -1);
2730 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2731 bytes_del += name_len;
2737 btrfs_release_path(path);
2738 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
2739 index, name, name_len, -1);
2745 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2746 bytes_del += name_len;
2753 /* update the directory size in the log to reflect the names
2757 struct btrfs_key key;
2759 key.objectid = dir_ino;
2761 key.type = BTRFS_INODE_ITEM_KEY;
2762 btrfs_release_path(path);
2764 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2770 struct btrfs_inode_item *item;
2773 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2774 struct btrfs_inode_item);
2775 i_size = btrfs_inode_size(path->nodes[0], item);
2776 if (i_size > bytes_del)
2777 i_size -= bytes_del;
2780 btrfs_set_inode_size(path->nodes[0], item, i_size);
2781 btrfs_mark_buffer_dirty(path->nodes[0]);
2784 btrfs_release_path(path);
2787 btrfs_free_path(path);
2789 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2790 if (ret == -ENOSPC) {
2791 root->fs_info->last_trans_log_full_commit = trans->transid;
2794 btrfs_abort_transaction(trans, root, ret);
2796 btrfs_end_log_trans(root);
2801 /* see comments for btrfs_del_dir_entries_in_log */
2802 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2803 struct btrfs_root *root,
2804 const char *name, int name_len,
2805 struct inode *inode, u64 dirid)
2807 struct btrfs_root *log;
2811 if (BTRFS_I(inode)->logged_trans < trans->transid)
2814 ret = join_running_log_trans(root);
2817 log = root->log_root;
2818 mutex_lock(&BTRFS_I(inode)->log_mutex);
2820 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
2822 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2823 if (ret == -ENOSPC) {
2824 root->fs_info->last_trans_log_full_commit = trans->transid;
2826 } else if (ret < 0 && ret != -ENOENT)
2827 btrfs_abort_transaction(trans, root, ret);
2828 btrfs_end_log_trans(root);
2834 * creates a range item in the log for 'dirid'. first_offset and
2835 * last_offset tell us which parts of the key space the log should
2836 * be considered authoritative for.
2838 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2839 struct btrfs_root *log,
2840 struct btrfs_path *path,
2841 int key_type, u64 dirid,
2842 u64 first_offset, u64 last_offset)
2845 struct btrfs_key key;
2846 struct btrfs_dir_log_item *item;
2848 key.objectid = dirid;
2849 key.offset = first_offset;
2850 if (key_type == BTRFS_DIR_ITEM_KEY)
2851 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2853 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2854 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2858 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2859 struct btrfs_dir_log_item);
2860 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2861 btrfs_mark_buffer_dirty(path->nodes[0]);
2862 btrfs_release_path(path);
2867 * log all the items included in the current transaction for a given
2868 * directory. This also creates the range items in the log tree required
2869 * to replay anything deleted before the fsync
2871 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2872 struct btrfs_root *root, struct inode *inode,
2873 struct btrfs_path *path,
2874 struct btrfs_path *dst_path, int key_type,
2875 u64 min_offset, u64 *last_offset_ret)
2877 struct btrfs_key min_key;
2878 struct btrfs_root *log = root->log_root;
2879 struct extent_buffer *src;
2884 u64 first_offset = min_offset;
2885 u64 last_offset = (u64)-1;
2886 u64 ino = btrfs_ino(inode);
2888 log = root->log_root;
2890 min_key.objectid = ino;
2891 min_key.type = key_type;
2892 min_key.offset = min_offset;
2894 path->keep_locks = 1;
2896 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
2899 * we didn't find anything from this transaction, see if there
2900 * is anything at all
2902 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2903 min_key.objectid = ino;
2904 min_key.type = key_type;
2905 min_key.offset = (u64)-1;
2906 btrfs_release_path(path);
2907 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2909 btrfs_release_path(path);
2912 ret = btrfs_previous_item(root, path, ino, key_type);
2914 /* if ret == 0 there are items for this type,
2915 * create a range to tell us the last key of this type.
2916 * otherwise, there are no items in this directory after
2917 * *min_offset, and we create a range to indicate that.
2920 struct btrfs_key tmp;
2921 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2923 if (key_type == tmp.type)
2924 first_offset = max(min_offset, tmp.offset) + 1;
2929 /* go backward to find any previous key */
2930 ret = btrfs_previous_item(root, path, ino, key_type);
2932 struct btrfs_key tmp;
2933 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2934 if (key_type == tmp.type) {
2935 first_offset = tmp.offset;
2936 ret = overwrite_item(trans, log, dst_path,
2937 path->nodes[0], path->slots[0],
2945 btrfs_release_path(path);
2947 /* find the first key from this transaction again */
2948 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2955 * we have a block from this transaction, log every item in it
2956 * from our directory
2959 struct btrfs_key tmp;
2960 src = path->nodes[0];
2961 nritems = btrfs_header_nritems(src);
2962 for (i = path->slots[0]; i < nritems; i++) {
2963 btrfs_item_key_to_cpu(src, &min_key, i);
2965 if (min_key.objectid != ino || min_key.type != key_type)
2967 ret = overwrite_item(trans, log, dst_path, src, i,
2974 path->slots[0] = nritems;
2977 * look ahead to the next item and see if it is also
2978 * from this directory and from this transaction
2980 ret = btrfs_next_leaf(root, path);
2982 last_offset = (u64)-1;
2985 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2986 if (tmp.objectid != ino || tmp.type != key_type) {
2987 last_offset = (u64)-1;
2990 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2991 ret = overwrite_item(trans, log, dst_path,
2992 path->nodes[0], path->slots[0],
2997 last_offset = tmp.offset;
3002 btrfs_release_path(path);
3003 btrfs_release_path(dst_path);
3006 *last_offset_ret = last_offset;
3008 * insert the log range keys to indicate where the log
3011 ret = insert_dir_log_key(trans, log, path, key_type,
3012 ino, first_offset, last_offset);
3020 * logging directories is very similar to logging inodes, We find all the items
3021 * from the current transaction and write them to the log.
3023 * The recovery code scans the directory in the subvolume, and if it finds a
3024 * key in the range logged that is not present in the log tree, then it means
3025 * that dir entry was unlinked during the transaction.
3027 * In order for that scan to work, we must include one key smaller than
3028 * the smallest logged by this transaction and one key larger than the largest
3029 * key logged by this transaction.
3031 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3032 struct btrfs_root *root, struct inode *inode,
3033 struct btrfs_path *path,
3034 struct btrfs_path *dst_path)
3039 int key_type = BTRFS_DIR_ITEM_KEY;
3045 ret = log_dir_items(trans, root, inode, path,
3046 dst_path, key_type, min_key,
3050 if (max_key == (u64)-1)
3052 min_key = max_key + 1;
3055 if (key_type == BTRFS_DIR_ITEM_KEY) {
3056 key_type = BTRFS_DIR_INDEX_KEY;
3063 * a helper function to drop items from the log before we relog an
3064 * inode. max_key_type indicates the highest item type to remove.
3065 * This cannot be run for file data extents because it does not
3066 * free the extents they point to.
3068 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3069 struct btrfs_root *log,
3070 struct btrfs_path *path,
3071 u64 objectid, int max_key_type)
3074 struct btrfs_key key;
3075 struct btrfs_key found_key;
3078 key.objectid = objectid;
3079 key.type = max_key_type;
3080 key.offset = (u64)-1;
3083 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3084 BUG_ON(ret == 0); /* Logic error */
3088 if (path->slots[0] == 0)
3092 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3095 if (found_key.objectid != objectid)
3098 found_key.offset = 0;
3100 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3103 ret = btrfs_del_items(trans, log, path, start_slot,
3104 path->slots[0] - start_slot + 1);
3106 * If start slot isn't 0 then we don't need to re-search, we've
3107 * found the last guy with the objectid in this tree.
3109 if (ret || start_slot != 0)
3111 btrfs_release_path(path);
3113 btrfs_release_path(path);
3119 static void fill_inode_item(struct btrfs_trans_handle *trans,
3120 struct extent_buffer *leaf,
3121 struct btrfs_inode_item *item,
3122 struct inode *inode, int log_inode_only)
3124 struct btrfs_map_token token;
3126 btrfs_init_map_token(&token);
3128 if (log_inode_only) {
3129 /* set the generation to zero so the recover code
3130 * can tell the difference between an logging
3131 * just to say 'this inode exists' and a logging
3132 * to say 'update this inode with these values'
3134 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3135 btrfs_set_token_inode_size(leaf, item, 0, &token);
3137 btrfs_set_token_inode_generation(leaf, item,
3138 BTRFS_I(inode)->generation,
3140 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3143 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3144 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3145 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3146 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3148 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3149 inode->i_atime.tv_sec, &token);
3150 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3151 inode->i_atime.tv_nsec, &token);
3153 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3154 inode->i_mtime.tv_sec, &token);
3155 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3156 inode->i_mtime.tv_nsec, &token);
3158 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3159 inode->i_ctime.tv_sec, &token);
3160 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3161 inode->i_ctime.tv_nsec, &token);
3163 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3166 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3167 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3168 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3169 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3170 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3173 static int log_inode_item(struct btrfs_trans_handle *trans,
3174 struct btrfs_root *log, struct btrfs_path *path,
3175 struct inode *inode)
3177 struct btrfs_inode_item *inode_item;
3180 ret = btrfs_insert_empty_item(trans, log, path,
3181 &BTRFS_I(inode)->location,
3182 sizeof(*inode_item));
3183 if (ret && ret != -EEXIST)
3185 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3186 struct btrfs_inode_item);
3187 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0);
3188 btrfs_release_path(path);
3192 static noinline int copy_items(struct btrfs_trans_handle *trans,
3193 struct inode *inode,
3194 struct btrfs_path *dst_path,
3195 struct extent_buffer *src,
3196 int start_slot, int nr, int inode_only)
3198 unsigned long src_offset;
3199 unsigned long dst_offset;
3200 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3201 struct btrfs_file_extent_item *extent;
3202 struct btrfs_inode_item *inode_item;
3204 struct btrfs_key *ins_keys;
3208 struct list_head ordered_sums;
3209 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3211 INIT_LIST_HEAD(&ordered_sums);
3213 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3214 nr * sizeof(u32), GFP_NOFS);
3218 ins_sizes = (u32 *)ins_data;
3219 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3221 for (i = 0; i < nr; i++) {
3222 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3223 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3225 ret = btrfs_insert_empty_items(trans, log, dst_path,
3226 ins_keys, ins_sizes, nr);
3232 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3233 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3234 dst_path->slots[0]);
3236 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3238 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3239 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3241 struct btrfs_inode_item);
3242 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3243 inode, inode_only == LOG_INODE_EXISTS);
3245 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3246 src_offset, ins_sizes[i]);
3249 /* take a reference on file data extents so that truncates
3250 * or deletes of this inode don't have to relog the inode
3253 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY &&
3256 extent = btrfs_item_ptr(src, start_slot + i,
3257 struct btrfs_file_extent_item);
3259 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3262 found_type = btrfs_file_extent_type(src, extent);
3263 if (found_type == BTRFS_FILE_EXTENT_REG) {
3265 ds = btrfs_file_extent_disk_bytenr(src,
3267 /* ds == 0 is a hole */
3271 dl = btrfs_file_extent_disk_num_bytes(src,
3273 cs = btrfs_file_extent_offset(src, extent);
3274 cl = btrfs_file_extent_num_bytes(src,
3276 if (btrfs_file_extent_compression(src,
3282 ret = btrfs_lookup_csums_range(
3283 log->fs_info->csum_root,
3284 ds + cs, ds + cs + cl - 1,
3287 btrfs_release_path(dst_path);
3295 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3296 btrfs_release_path(dst_path);
3300 * we have to do this after the loop above to avoid changing the
3301 * log tree while trying to change the log tree.
3304 while (!list_empty(&ordered_sums)) {
3305 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3306 struct btrfs_ordered_sum,
3309 ret = btrfs_csum_file_blocks(trans, log, sums);
3310 list_del(&sums->list);
3316 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3318 struct extent_map *em1, *em2;
3320 em1 = list_entry(a, struct extent_map, list);
3321 em2 = list_entry(b, struct extent_map, list);
3323 if (em1->start < em2->start)
3325 else if (em1->start > em2->start)
3330 static int log_one_extent(struct btrfs_trans_handle *trans,
3331 struct inode *inode, struct btrfs_root *root,
3332 struct extent_map *em, struct btrfs_path *path)
3334 struct btrfs_root *log = root->log_root;
3335 struct btrfs_file_extent_item *fi;
3336 struct extent_buffer *leaf;
3337 struct btrfs_ordered_extent *ordered;
3338 struct list_head ordered_sums;
3339 struct btrfs_map_token token;
3340 struct btrfs_key key;
3341 u64 mod_start = em->mod_start;
3342 u64 mod_len = em->mod_len;
3345 u64 extent_offset = em->start - em->orig_start;
3348 int index = log->log_transid % 2;
3349 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3351 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
3352 em->start + em->len, NULL, 0);
3356 INIT_LIST_HEAD(&ordered_sums);
3357 btrfs_init_map_token(&token);
3358 key.objectid = btrfs_ino(inode);
3359 key.type = BTRFS_EXTENT_DATA_KEY;
3360 key.offset = em->start;
3362 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*fi));
3365 leaf = path->nodes[0];
3366 fi = btrfs_item_ptr(leaf, path->slots[0],
3367 struct btrfs_file_extent_item);
3369 btrfs_set_token_file_extent_generation(leaf, fi, em->generation,
3371 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3373 btrfs_set_token_file_extent_type(leaf, fi,
3374 BTRFS_FILE_EXTENT_PREALLOC,
3377 btrfs_set_token_file_extent_type(leaf, fi,
3378 BTRFS_FILE_EXTENT_REG,
3380 if (em->block_start == EXTENT_MAP_HOLE)
3384 block_len = max(em->block_len, em->orig_block_len);
3385 if (em->compress_type != BTRFS_COMPRESS_NONE) {
3386 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3389 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3391 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
3392 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3394 extent_offset, &token);
3395 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3398 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
3399 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
3403 btrfs_set_token_file_extent_offset(leaf, fi,
3404 em->start - em->orig_start,
3406 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
3407 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
3408 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
3410 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
3411 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
3412 btrfs_mark_buffer_dirty(leaf);
3414 btrfs_release_path(path);
3422 if (em->compress_type) {
3424 csum_len = block_len;
3428 * First check and see if our csums are on our outstanding ordered
3432 spin_lock_irq(&log->log_extents_lock[index]);
3433 list_for_each_entry(ordered, &log->logged_list[index], log_list) {
3434 struct btrfs_ordered_sum *sum;
3439 if (ordered->inode != inode)
3442 if (ordered->file_offset + ordered->len <= mod_start ||
3443 mod_start + mod_len <= ordered->file_offset)
3447 * We are going to copy all the csums on this ordered extent, so
3448 * go ahead and adjust mod_start and mod_len in case this
3449 * ordered extent has already been logged.
3451 if (ordered->file_offset > mod_start) {
3452 if (ordered->file_offset + ordered->len >=
3453 mod_start + mod_len)
3454 mod_len = ordered->file_offset - mod_start;
3456 * If we have this case
3458 * |--------- logged extent ---------|
3459 * |----- ordered extent ----|
3461 * Just don't mess with mod_start and mod_len, we'll
3462 * just end up logging more csums than we need and it
3466 if (ordered->file_offset + ordered->len <
3467 mod_start + mod_len) {
3468 mod_len = (mod_start + mod_len) -
3469 (ordered->file_offset + ordered->len);
3470 mod_start = ordered->file_offset +
3478 * To keep us from looping for the above case of an ordered
3479 * extent that falls inside of the logged extent.
3481 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3484 atomic_inc(&ordered->refs);
3485 spin_unlock_irq(&log->log_extents_lock[index]);
3487 * we've dropped the lock, we must either break or
3488 * start over after this.
3491 wait_event(ordered->wait, ordered->csum_bytes_left == 0);
3493 list_for_each_entry(sum, &ordered->list, list) {
3494 ret = btrfs_csum_file_blocks(trans, log, sum);
3496 btrfs_put_ordered_extent(ordered);
3500 btrfs_put_ordered_extent(ordered);
3504 spin_unlock_irq(&log->log_extents_lock[index]);
3507 if (!mod_len || ret)
3510 csum_offset = mod_start - em->start;
3513 /* block start is already adjusted for the file extent offset. */
3514 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
3515 em->block_start + csum_offset,
3516 em->block_start + csum_offset +
3517 csum_len - 1, &ordered_sums, 0);
3521 while (!list_empty(&ordered_sums)) {
3522 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3523 struct btrfs_ordered_sum,
3526 ret = btrfs_csum_file_blocks(trans, log, sums);
3527 list_del(&sums->list);
3534 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
3535 struct btrfs_root *root,
3536 struct inode *inode,
3537 struct btrfs_path *path)
3539 struct extent_map *em, *n;
3540 struct list_head extents;
3541 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3546 INIT_LIST_HEAD(&extents);
3548 write_lock(&tree->lock);
3549 test_gen = root->fs_info->last_trans_committed;
3551 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
3552 list_del_init(&em->list);
3555 * Just an arbitrary number, this can be really CPU intensive
3556 * once we start getting a lot of extents, and really once we
3557 * have a bunch of extents we just want to commit since it will
3560 if (++num > 32768) {
3561 list_del_init(&tree->modified_extents);
3566 if (em->generation <= test_gen)
3568 /* Need a ref to keep it from getting evicted from cache */
3569 atomic_inc(&em->refs);
3570 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
3571 list_add_tail(&em->list, &extents);
3575 list_sort(NULL, &extents, extent_cmp);
3578 while (!list_empty(&extents)) {
3579 em = list_entry(extents.next, struct extent_map, list);
3581 list_del_init(&em->list);
3584 * If we had an error we just need to delete everybody from our
3588 clear_em_logging(tree, em);
3589 free_extent_map(em);
3593 write_unlock(&tree->lock);
3595 ret = log_one_extent(trans, inode, root, em, path);
3596 write_lock(&tree->lock);
3597 clear_em_logging(tree, em);
3598 free_extent_map(em);
3600 WARN_ON(!list_empty(&extents));
3601 write_unlock(&tree->lock);
3603 btrfs_release_path(path);
3607 /* log a single inode in the tree log.
3608 * At least one parent directory for this inode must exist in the tree
3609 * or be logged already.
3611 * Any items from this inode changed by the current transaction are copied
3612 * to the log tree. An extra reference is taken on any extents in this
3613 * file, allowing us to avoid a whole pile of corner cases around logging
3614 * blocks that have been removed from the tree.
3616 * See LOG_INODE_ALL and related defines for a description of what inode_only
3619 * This handles both files and directories.
3621 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
3622 struct btrfs_root *root, struct inode *inode,
3625 struct btrfs_path *path;
3626 struct btrfs_path *dst_path;
3627 struct btrfs_key min_key;
3628 struct btrfs_key max_key;
3629 struct btrfs_root *log = root->log_root;
3630 struct extent_buffer *src = NULL;
3634 int ins_start_slot = 0;
3636 bool fast_search = false;
3637 u64 ino = btrfs_ino(inode);
3639 path = btrfs_alloc_path();
3642 dst_path = btrfs_alloc_path();
3644 btrfs_free_path(path);
3648 min_key.objectid = ino;
3649 min_key.type = BTRFS_INODE_ITEM_KEY;
3652 max_key.objectid = ino;
3655 /* today the code can only do partial logging of directories */
3656 if (S_ISDIR(inode->i_mode) ||
3657 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3658 &BTRFS_I(inode)->runtime_flags) &&
3659 inode_only == LOG_INODE_EXISTS))
3660 max_key.type = BTRFS_XATTR_ITEM_KEY;
3662 max_key.type = (u8)-1;
3663 max_key.offset = (u64)-1;
3665 /* Only run delayed items if we are a dir or a new file */
3666 if (S_ISDIR(inode->i_mode) ||
3667 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) {
3668 ret = btrfs_commit_inode_delayed_items(trans, inode);
3670 btrfs_free_path(path);
3671 btrfs_free_path(dst_path);
3676 mutex_lock(&BTRFS_I(inode)->log_mutex);
3678 btrfs_get_logged_extents(log, inode);
3681 * a brute force approach to making sure we get the most uptodate
3682 * copies of everything.
3684 if (S_ISDIR(inode->i_mode)) {
3685 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
3687 if (inode_only == LOG_INODE_EXISTS)
3688 max_key_type = BTRFS_XATTR_ITEM_KEY;
3689 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
3691 if (test_and_clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3692 &BTRFS_I(inode)->runtime_flags)) {
3693 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3694 &BTRFS_I(inode)->runtime_flags);
3695 ret = btrfs_truncate_inode_items(trans, log,
3697 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3698 &BTRFS_I(inode)->runtime_flags)) {
3699 if (inode_only == LOG_INODE_ALL)
3701 max_key.type = BTRFS_XATTR_ITEM_KEY;
3702 ret = drop_objectid_items(trans, log, path, ino,
3705 if (inode_only == LOG_INODE_ALL)
3707 ret = log_inode_item(trans, log, dst_path, inode);
3720 path->keep_locks = 1;
3724 ret = btrfs_search_forward(root, &min_key,
3725 path, trans->transid);
3729 /* note, ins_nr might be > 0 here, cleanup outside the loop */
3730 if (min_key.objectid != ino)
3732 if (min_key.type > max_key.type)
3735 src = path->nodes[0];
3736 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
3739 } else if (!ins_nr) {
3740 ins_start_slot = path->slots[0];
3745 ret = copy_items(trans, inode, dst_path, src, ins_start_slot,
3746 ins_nr, inode_only);
3752 ins_start_slot = path->slots[0];
3755 nritems = btrfs_header_nritems(path->nodes[0]);
3757 if (path->slots[0] < nritems) {
3758 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
3763 ret = copy_items(trans, inode, dst_path, src,
3765 ins_nr, inode_only);
3772 btrfs_release_path(path);
3774 if (min_key.offset < (u64)-1) {
3776 } else if (min_key.type < max_key.type) {
3784 ret = copy_items(trans, inode, dst_path, src, ins_start_slot,
3785 ins_nr, inode_only);
3794 btrfs_release_path(path);
3795 btrfs_release_path(dst_path);
3797 ret = btrfs_log_changed_extents(trans, root, inode, dst_path);
3803 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3804 struct extent_map *em, *n;
3806 write_lock(&tree->lock);
3807 list_for_each_entry_safe(em, n, &tree->modified_extents, list)
3808 list_del_init(&em->list);
3809 write_unlock(&tree->lock);
3812 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
3813 ret = log_directory_changes(trans, root, inode, path, dst_path);
3819 BTRFS_I(inode)->logged_trans = trans->transid;
3820 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
3823 btrfs_free_logged_extents(log, log->log_transid);
3824 mutex_unlock(&BTRFS_I(inode)->log_mutex);
3826 btrfs_free_path(path);
3827 btrfs_free_path(dst_path);
3832 * follow the dentry parent pointers up the chain and see if any
3833 * of the directories in it require a full commit before they can
3834 * be logged. Returns zero if nothing special needs to be done or 1 if
3835 * a full commit is required.
3837 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
3838 struct inode *inode,
3839 struct dentry *parent,
3840 struct super_block *sb,
3844 struct btrfs_root *root;
3845 struct dentry *old_parent = NULL;
3846 struct inode *orig_inode = inode;
3849 * for regular files, if its inode is already on disk, we don't
3850 * have to worry about the parents at all. This is because
3851 * we can use the last_unlink_trans field to record renames
3852 * and other fun in this file.
3854 if (S_ISREG(inode->i_mode) &&
3855 BTRFS_I(inode)->generation <= last_committed &&
3856 BTRFS_I(inode)->last_unlink_trans <= last_committed)
3859 if (!S_ISDIR(inode->i_mode)) {
3860 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3862 inode = parent->d_inode;
3867 * If we are logging a directory then we start with our inode,
3868 * not our parents inode, so we need to skipp setting the
3869 * logged_trans so that further down in the log code we don't
3870 * think this inode has already been logged.
3872 if (inode != orig_inode)
3873 BTRFS_I(inode)->logged_trans = trans->transid;
3876 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
3877 root = BTRFS_I(inode)->root;
3880 * make sure any commits to the log are forced
3881 * to be full commits
3883 root->fs_info->last_trans_log_full_commit =
3889 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3892 if (IS_ROOT(parent))
3895 parent = dget_parent(parent);
3897 old_parent = parent;
3898 inode = parent->d_inode;
3907 * helper function around btrfs_log_inode to make sure newly created
3908 * parent directories also end up in the log. A minimal inode and backref
3909 * only logging is done of any parent directories that are older than
3910 * the last committed transaction
3912 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
3913 struct btrfs_root *root, struct inode *inode,
3914 struct dentry *parent, int exists_only)
3916 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
3917 struct super_block *sb;
3918 struct dentry *old_parent = NULL;
3920 u64 last_committed = root->fs_info->last_trans_committed;
3924 if (btrfs_test_opt(root, NOTREELOG)) {
3929 if (root->fs_info->last_trans_log_full_commit >
3930 root->fs_info->last_trans_committed) {
3935 if (root != BTRFS_I(inode)->root ||
3936 btrfs_root_refs(&root->root_item) == 0) {
3941 ret = check_parent_dirs_for_sync(trans, inode, parent,
3942 sb, last_committed);
3946 if (btrfs_inode_in_log(inode, trans->transid)) {
3947 ret = BTRFS_NO_LOG_SYNC;
3951 ret = start_log_trans(trans, root);
3955 ret = btrfs_log_inode(trans, root, inode, inode_only);
3960 * for regular files, if its inode is already on disk, we don't
3961 * have to worry about the parents at all. This is because
3962 * we can use the last_unlink_trans field to record renames
3963 * and other fun in this file.
3965 if (S_ISREG(inode->i_mode) &&
3966 BTRFS_I(inode)->generation <= last_committed &&
3967 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3972 inode_only = LOG_INODE_EXISTS;
3974 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3977 inode = parent->d_inode;
3978 if (root != BTRFS_I(inode)->root)
3981 if (BTRFS_I(inode)->generation >
3982 root->fs_info->last_trans_committed) {
3983 ret = btrfs_log_inode(trans, root, inode, inode_only);
3987 if (IS_ROOT(parent))
3990 parent = dget_parent(parent);
3992 old_parent = parent;
3998 root->fs_info->last_trans_log_full_commit = trans->transid;
4001 btrfs_end_log_trans(root);
4007 * it is not safe to log dentry if the chunk root has added new
4008 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
4009 * If this returns 1, you must commit the transaction to safely get your
4012 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
4013 struct btrfs_root *root, struct dentry *dentry)
4015 struct dentry *parent = dget_parent(dentry);
4018 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
4025 * should be called during mount to recover any replay any log trees
4028 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
4031 struct btrfs_path *path;
4032 struct btrfs_trans_handle *trans;
4033 struct btrfs_key key;
4034 struct btrfs_key found_key;
4035 struct btrfs_key tmp_key;
4036 struct btrfs_root *log;
4037 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
4038 struct walk_control wc = {
4039 .process_func = process_one_buffer,
4043 path = btrfs_alloc_path();
4047 fs_info->log_root_recovering = 1;
4049 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4050 if (IS_ERR(trans)) {
4051 ret = PTR_ERR(trans);
4058 ret = walk_log_tree(trans, log_root_tree, &wc);
4060 btrfs_error(fs_info, ret, "Failed to pin buffers while "
4061 "recovering log root tree.");
4066 key.objectid = BTRFS_TREE_LOG_OBJECTID;
4067 key.offset = (u64)-1;
4068 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
4071 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
4074 btrfs_error(fs_info, ret,
4075 "Couldn't find tree log root.");
4079 if (path->slots[0] == 0)
4083 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4085 btrfs_release_path(path);
4086 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4089 log = btrfs_read_fs_root(log_root_tree, &found_key);
4092 btrfs_error(fs_info, ret,
4093 "Couldn't read tree log root.");
4097 tmp_key.objectid = found_key.offset;
4098 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
4099 tmp_key.offset = (u64)-1;
4101 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
4102 if (IS_ERR(wc.replay_dest)) {
4103 ret = PTR_ERR(wc.replay_dest);
4104 free_extent_buffer(log->node);
4105 free_extent_buffer(log->commit_root);
4107 btrfs_error(fs_info, ret, "Couldn't read target root "
4108 "for tree log recovery.");
4112 wc.replay_dest->log_root = log;
4113 btrfs_record_root_in_trans(trans, wc.replay_dest);
4114 ret = walk_log_tree(trans, log, &wc);
4116 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
4117 ret = fixup_inode_link_counts(trans, wc.replay_dest,
4121 key.offset = found_key.offset - 1;
4122 wc.replay_dest->log_root = NULL;
4123 free_extent_buffer(log->node);
4124 free_extent_buffer(log->commit_root);
4130 if (found_key.offset == 0)
4133 btrfs_release_path(path);
4135 /* step one is to pin it all, step two is to replay just inodes */
4138 wc.process_func = replay_one_buffer;
4139 wc.stage = LOG_WALK_REPLAY_INODES;
4142 /* step three is to replay everything */
4143 if (wc.stage < LOG_WALK_REPLAY_ALL) {
4148 btrfs_free_path(path);
4150 /* step 4: commit the transaction, which also unpins the blocks */
4151 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
4155 free_extent_buffer(log_root_tree->node);
4156 log_root_tree->log_root = NULL;
4157 fs_info->log_root_recovering = 0;
4158 kfree(log_root_tree);
4163 btrfs_end_transaction(wc.trans, fs_info->tree_root);
4164 btrfs_free_path(path);
4169 * there are some corner cases where we want to force a full
4170 * commit instead of allowing a directory to be logged.
4172 * They revolve around files there were unlinked from the directory, and
4173 * this function updates the parent directory so that a full commit is
4174 * properly done if it is fsync'd later after the unlinks are done.
4176 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
4177 struct inode *dir, struct inode *inode,
4181 * when we're logging a file, if it hasn't been renamed
4182 * or unlinked, and its inode is fully committed on disk,
4183 * we don't have to worry about walking up the directory chain
4184 * to log its parents.
4186 * So, we use the last_unlink_trans field to put this transid
4187 * into the file. When the file is logged we check it and
4188 * don't log the parents if the file is fully on disk.
4190 if (S_ISREG(inode->i_mode))
4191 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4194 * if this directory was already logged any new
4195 * names for this file/dir will get recorded
4198 if (BTRFS_I(dir)->logged_trans == trans->transid)
4202 * if the inode we're about to unlink was logged,
4203 * the log will be properly updated for any new names
4205 if (BTRFS_I(inode)->logged_trans == trans->transid)
4209 * when renaming files across directories, if the directory
4210 * there we're unlinking from gets fsync'd later on, there's
4211 * no way to find the destination directory later and fsync it
4212 * properly. So, we have to be conservative and force commits
4213 * so the new name gets discovered.
4218 /* we can safely do the unlink without any special recording */
4222 BTRFS_I(dir)->last_unlink_trans = trans->transid;
4226 * Call this after adding a new name for a file and it will properly
4227 * update the log to reflect the new name.
4229 * It will return zero if all goes well, and it will return 1 if a
4230 * full transaction commit is required.
4232 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
4233 struct inode *inode, struct inode *old_dir,
4234 struct dentry *parent)
4236 struct btrfs_root * root = BTRFS_I(inode)->root;
4239 * this will force the logging code to walk the dentry chain
4242 if (S_ISREG(inode->i_mode))
4243 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4246 * if this inode hasn't been logged and directory we're renaming it
4247 * from hasn't been logged, we don't need to log it
4249 if (BTRFS_I(inode)->logged_trans <=
4250 root->fs_info->last_trans_committed &&
4251 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
4252 root->fs_info->last_trans_committed))
4255 return btrfs_log_inode_parent(trans, root, inode, parent, 1);