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>
26 #include "print-tree.h"
30 /* magic values for the inode_only field in btrfs_log_inode:
32 * LOG_INODE_ALL means to log everything
33 * LOG_INODE_EXISTS means to log just enough to recreate the inode
36 #define LOG_INODE_ALL 0
37 #define LOG_INODE_EXISTS 1
40 * directory trouble cases
42 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
43 * log, we must force a full commit before doing an fsync of the directory
44 * where the unlink was done.
45 * ---> record transid of last unlink/rename per directory
49 * rename foo/some_dir foo2/some_dir
51 * fsync foo/some_dir/some_file
53 * The fsync above will unlink the original some_dir without recording
54 * it in its new location (foo2). After a crash, some_dir will be gone
55 * unless the fsync of some_file forces a full commit
57 * 2) we must log any new names for any file or dir that is in the fsync
58 * log. ---> check inode while renaming/linking.
60 * 2a) we must log any new names for any file or dir during rename
61 * when the directory they are being removed from was logged.
62 * ---> check inode and old parent dir during rename
64 * 2a is actually the more important variant. With the extra logging
65 * a crash might unlink the old name without recreating the new one
67 * 3) after a crash, we must go through any directories with a link count
68 * of zero and redo the rm -rf
75 * The directory f1 was fully removed from the FS, but fsync was never
76 * called on f1, only its parent dir. After a crash the rm -rf must
77 * be replayed. This must be able to recurse down the entire
78 * directory tree. The inode link count fixup code takes care of the
83 * stages for the tree walking. The first
84 * stage (0) is to only pin down the blocks we find
85 * the second stage (1) is to make sure that all the inodes
86 * we find in the log are created in the subvolume.
88 * The last stage is to deal with directories and links and extents
89 * and all the other fun semantics
91 #define LOG_WALK_PIN_ONLY 0
92 #define LOG_WALK_REPLAY_INODES 1
93 #define LOG_WALK_REPLAY_DIR_INDEX 2
94 #define LOG_WALK_REPLAY_ALL 3
96 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root, struct inode *inode,
99 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
100 struct btrfs_root *root,
101 struct btrfs_path *path, u64 objectid);
102 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
103 struct btrfs_root *root,
104 struct btrfs_root *log,
105 struct btrfs_path *path,
106 u64 dirid, int del_all);
109 * tree logging is a special write ahead log used to make sure that
110 * fsyncs and O_SYNCs can happen without doing full tree commits.
112 * Full tree commits are expensive because they require commonly
113 * modified blocks to be recowed, creating many dirty pages in the
114 * extent tree an 4x-6x higher write load than ext3.
116 * Instead of doing a tree commit on every fsync, we use the
117 * key ranges and transaction ids to find items for a given file or directory
118 * that have changed in this transaction. Those items are copied into
119 * a special tree (one per subvolume root), that tree is written to disk
120 * and then the fsync is considered complete.
122 * After a crash, items are copied out of the log-tree back into the
123 * subvolume tree. Any file data extents found are recorded in the extent
124 * allocation tree, and the log-tree freed.
126 * The log tree is read three times, once to pin down all the extents it is
127 * using in ram and once, once to create all the inodes logged in the tree
128 * and once to do all the other items.
132 * start a sub transaction and setup the log tree
133 * this increments the log tree writer count to make the people
134 * syncing the tree wait for us to finish
136 static int start_log_trans(struct btrfs_trans_handle *trans,
137 struct btrfs_root *root,
138 struct btrfs_log_ctx *ctx)
143 mutex_lock(&root->log_mutex);
144 if (root->log_root) {
145 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
149 if (!root->log_start_pid) {
150 root->log_start_pid = current->pid;
151 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
152 } else if (root->log_start_pid != current->pid) {
153 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
156 atomic_inc(&root->log_batch);
157 atomic_inc(&root->log_writers);
159 index = root->log_transid % 2;
160 list_add_tail(&ctx->list, &root->log_ctxs[index]);
161 ctx->log_transid = root->log_transid;
163 mutex_unlock(&root->log_mutex);
168 mutex_lock(&root->fs_info->tree_log_mutex);
169 if (!root->fs_info->log_root_tree)
170 ret = btrfs_init_log_root_tree(trans, root->fs_info);
171 mutex_unlock(&root->fs_info->tree_log_mutex);
175 if (!root->log_root) {
176 ret = btrfs_add_log_tree(trans, root);
180 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
181 root->log_start_pid = current->pid;
182 atomic_inc(&root->log_batch);
183 atomic_inc(&root->log_writers);
185 index = root->log_transid % 2;
186 list_add_tail(&ctx->list, &root->log_ctxs[index]);
187 ctx->log_transid = root->log_transid;
190 mutex_unlock(&root->log_mutex);
195 * returns 0 if there was a log transaction running and we were able
196 * to join, or returns -ENOENT if there were not transactions
199 static int join_running_log_trans(struct btrfs_root *root)
207 mutex_lock(&root->log_mutex);
208 if (root->log_root) {
210 atomic_inc(&root->log_writers);
212 mutex_unlock(&root->log_mutex);
217 * This either makes the current running log transaction wait
218 * until you call btrfs_end_log_trans() or it makes any future
219 * log transactions wait until you call btrfs_end_log_trans()
221 int btrfs_pin_log_trans(struct btrfs_root *root)
225 mutex_lock(&root->log_mutex);
226 atomic_inc(&root->log_writers);
227 mutex_unlock(&root->log_mutex);
232 * indicate we're done making changes to the log tree
233 * and wake up anyone waiting to do a sync
235 void btrfs_end_log_trans(struct btrfs_root *root)
237 if (atomic_dec_and_test(&root->log_writers)) {
239 if (waitqueue_active(&root->log_writer_wait))
240 wake_up(&root->log_writer_wait);
246 * the walk control struct is used to pass state down the chain when
247 * processing the log tree. The stage field tells us which part
248 * of the log tree processing we are currently doing. The others
249 * are state fields used for that specific part
251 struct walk_control {
252 /* should we free the extent on disk when done? This is used
253 * at transaction commit time while freeing a log tree
257 /* should we write out the extent buffer? This is used
258 * while flushing the log tree to disk during a sync
262 /* should we wait for the extent buffer io to finish? Also used
263 * while flushing the log tree to disk for a sync
267 /* pin only walk, we record which extents on disk belong to the
272 /* what stage of the replay code we're currently in */
275 /* the root we are currently replaying */
276 struct btrfs_root *replay_dest;
278 /* the trans handle for the current replay */
279 struct btrfs_trans_handle *trans;
281 /* the function that gets used to process blocks we find in the
282 * tree. Note the extent_buffer might not be up to date when it is
283 * passed in, and it must be checked or read if you need the data
286 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
287 struct walk_control *wc, u64 gen);
291 * process_func used to pin down extents, write them or wait on them
293 static int process_one_buffer(struct btrfs_root *log,
294 struct extent_buffer *eb,
295 struct walk_control *wc, u64 gen)
300 * If this fs is mixed then we need to be able to process the leaves to
301 * pin down any logged extents, so we have to read the block.
303 if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) {
304 ret = btrfs_read_buffer(eb, gen);
310 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
313 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
314 if (wc->pin && btrfs_header_level(eb) == 0)
315 ret = btrfs_exclude_logged_extents(log, eb);
317 btrfs_write_tree_block(eb);
319 btrfs_wait_tree_block_writeback(eb);
325 * Item overwrite used by replay and tree logging. eb, slot and key all refer
326 * to the src data we are copying out.
328 * root is the tree we are copying into, and path is a scratch
329 * path for use in this function (it should be released on entry and
330 * will be released on exit).
332 * If the key is already in the destination tree the existing item is
333 * overwritten. If the existing item isn't big enough, it is extended.
334 * If it is too large, it is truncated.
336 * If the key isn't in the destination yet, a new item is inserted.
338 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
339 struct btrfs_root *root,
340 struct btrfs_path *path,
341 struct extent_buffer *eb, int slot,
342 struct btrfs_key *key)
346 u64 saved_i_size = 0;
347 int save_old_i_size = 0;
348 unsigned long src_ptr;
349 unsigned long dst_ptr;
350 int overwrite_root = 0;
351 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
353 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
356 item_size = btrfs_item_size_nr(eb, slot);
357 src_ptr = btrfs_item_ptr_offset(eb, slot);
359 /* look for the key in the destination tree */
360 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
367 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
369 if (dst_size != item_size)
372 if (item_size == 0) {
373 btrfs_release_path(path);
376 dst_copy = kmalloc(item_size, GFP_NOFS);
377 src_copy = kmalloc(item_size, GFP_NOFS);
378 if (!dst_copy || !src_copy) {
379 btrfs_release_path(path);
385 read_extent_buffer(eb, src_copy, src_ptr, item_size);
387 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
388 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
390 ret = memcmp(dst_copy, src_copy, item_size);
395 * they have the same contents, just return, this saves
396 * us from cowing blocks in the destination tree and doing
397 * extra writes that may not have been done by a previous
401 btrfs_release_path(path);
406 * We need to load the old nbytes into the inode so when we
407 * replay the extents we've logged we get the right nbytes.
410 struct btrfs_inode_item *item;
414 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
415 struct btrfs_inode_item);
416 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
417 item = btrfs_item_ptr(eb, slot,
418 struct btrfs_inode_item);
419 btrfs_set_inode_nbytes(eb, item, nbytes);
422 * If this is a directory we need to reset the i_size to
423 * 0 so that we can set it up properly when replaying
424 * the rest of the items in this log.
426 mode = btrfs_inode_mode(eb, item);
428 btrfs_set_inode_size(eb, item, 0);
430 } else if (inode_item) {
431 struct btrfs_inode_item *item;
435 * New inode, set nbytes to 0 so that the nbytes comes out
436 * properly when we replay the extents.
438 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
439 btrfs_set_inode_nbytes(eb, item, 0);
442 * If this is a directory we need to reset the i_size to 0 so
443 * that we can set it up properly when replaying the rest of
444 * the items in this log.
446 mode = btrfs_inode_mode(eb, item);
448 btrfs_set_inode_size(eb, item, 0);
451 btrfs_release_path(path);
452 /* try to insert the key into the destination tree */
453 ret = btrfs_insert_empty_item(trans, root, path,
456 /* make sure any existing item is the correct size */
457 if (ret == -EEXIST) {
459 found_size = btrfs_item_size_nr(path->nodes[0],
461 if (found_size > item_size)
462 btrfs_truncate_item(root, path, item_size, 1);
463 else if (found_size < item_size)
464 btrfs_extend_item(root, path,
465 item_size - found_size);
469 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
472 /* don't overwrite an existing inode if the generation number
473 * was logged as zero. This is done when the tree logging code
474 * is just logging an inode to make sure it exists after recovery.
476 * Also, don't overwrite i_size on directories during replay.
477 * log replay inserts and removes directory items based on the
478 * state of the tree found in the subvolume, and i_size is modified
481 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
482 struct btrfs_inode_item *src_item;
483 struct btrfs_inode_item *dst_item;
485 src_item = (struct btrfs_inode_item *)src_ptr;
486 dst_item = (struct btrfs_inode_item *)dst_ptr;
488 if (btrfs_inode_generation(eb, src_item) == 0)
491 if (overwrite_root &&
492 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
493 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
495 saved_i_size = btrfs_inode_size(path->nodes[0],
500 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
503 if (save_old_i_size) {
504 struct btrfs_inode_item *dst_item;
505 dst_item = (struct btrfs_inode_item *)dst_ptr;
506 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
509 /* make sure the generation is filled in */
510 if (key->type == BTRFS_INODE_ITEM_KEY) {
511 struct btrfs_inode_item *dst_item;
512 dst_item = (struct btrfs_inode_item *)dst_ptr;
513 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
514 btrfs_set_inode_generation(path->nodes[0], dst_item,
519 btrfs_mark_buffer_dirty(path->nodes[0]);
520 btrfs_release_path(path);
525 * simple helper to read an inode off the disk from a given root
526 * This can only be called for subvolume roots and not for the log
528 static noinline struct inode *read_one_inode(struct btrfs_root *root,
531 struct btrfs_key key;
534 key.objectid = objectid;
535 key.type = BTRFS_INODE_ITEM_KEY;
537 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
540 } else if (is_bad_inode(inode)) {
547 /* replays a single extent in 'eb' at 'slot' with 'key' into the
548 * subvolume 'root'. path is released on entry and should be released
551 * extents in the log tree have not been allocated out of the extent
552 * tree yet. So, this completes the allocation, taking a reference
553 * as required if the extent already exists or creating a new extent
554 * if it isn't in the extent allocation tree yet.
556 * The extent is inserted into the file, dropping any existing extents
557 * from the file that overlap the new one.
559 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
560 struct btrfs_root *root,
561 struct btrfs_path *path,
562 struct extent_buffer *eb, int slot,
563 struct btrfs_key *key)
567 u64 start = key->offset;
569 struct btrfs_file_extent_item *item;
570 struct inode *inode = NULL;
574 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
575 found_type = btrfs_file_extent_type(eb, item);
577 if (found_type == BTRFS_FILE_EXTENT_REG ||
578 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
579 nbytes = btrfs_file_extent_num_bytes(eb, item);
580 extent_end = start + nbytes;
583 * We don't add to the inodes nbytes if we are prealloc or a
586 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
588 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
589 size = btrfs_file_extent_inline_len(eb, slot, item);
590 nbytes = btrfs_file_extent_ram_bytes(eb, item);
591 extent_end = ALIGN(start + size, root->sectorsize);
597 inode = read_one_inode(root, key->objectid);
604 * first check to see if we already have this extent in the
605 * file. This must be done before the btrfs_drop_extents run
606 * so we don't try to drop this extent.
608 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
612 (found_type == BTRFS_FILE_EXTENT_REG ||
613 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
614 struct btrfs_file_extent_item cmp1;
615 struct btrfs_file_extent_item cmp2;
616 struct btrfs_file_extent_item *existing;
617 struct extent_buffer *leaf;
619 leaf = path->nodes[0];
620 existing = btrfs_item_ptr(leaf, path->slots[0],
621 struct btrfs_file_extent_item);
623 read_extent_buffer(eb, &cmp1, (unsigned long)item,
625 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
629 * we already have a pointer to this exact extent,
630 * we don't have to do anything
632 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
633 btrfs_release_path(path);
637 btrfs_release_path(path);
639 /* drop any overlapping extents */
640 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
644 if (found_type == BTRFS_FILE_EXTENT_REG ||
645 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
647 unsigned long dest_offset;
648 struct btrfs_key ins;
650 ret = btrfs_insert_empty_item(trans, root, path, key,
654 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
656 copy_extent_buffer(path->nodes[0], eb, dest_offset,
657 (unsigned long)item, sizeof(*item));
659 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
660 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
661 ins.type = BTRFS_EXTENT_ITEM_KEY;
662 offset = key->offset - btrfs_file_extent_offset(eb, item);
664 if (ins.objectid > 0) {
667 LIST_HEAD(ordered_sums);
669 * is this extent already allocated in the extent
670 * allocation tree? If so, just add a reference
672 ret = btrfs_lookup_extent(root, ins.objectid,
675 ret = btrfs_inc_extent_ref(trans, root,
676 ins.objectid, ins.offset,
677 0, root->root_key.objectid,
678 key->objectid, offset, 0);
683 * insert the extent pointer in the extent
686 ret = btrfs_alloc_logged_file_extent(trans,
687 root, root->root_key.objectid,
688 key->objectid, offset, &ins);
692 btrfs_release_path(path);
694 if (btrfs_file_extent_compression(eb, item)) {
695 csum_start = ins.objectid;
696 csum_end = csum_start + ins.offset;
698 csum_start = ins.objectid +
699 btrfs_file_extent_offset(eb, item);
700 csum_end = csum_start +
701 btrfs_file_extent_num_bytes(eb, item);
704 ret = btrfs_lookup_csums_range(root->log_root,
705 csum_start, csum_end - 1,
709 while (!list_empty(&ordered_sums)) {
710 struct btrfs_ordered_sum *sums;
711 sums = list_entry(ordered_sums.next,
712 struct btrfs_ordered_sum,
715 ret = btrfs_csum_file_blocks(trans,
716 root->fs_info->csum_root,
718 list_del(&sums->list);
724 btrfs_release_path(path);
726 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
727 /* inline extents are easy, we just overwrite them */
728 ret = overwrite_item(trans, root, path, eb, slot, key);
733 inode_add_bytes(inode, nbytes);
734 ret = btrfs_update_inode(trans, root, inode);
742 * when cleaning up conflicts between the directory names in the
743 * subvolume, directory names in the log and directory names in the
744 * inode back references, we may have to unlink inodes from directories.
746 * This is a helper function to do the unlink of a specific directory
749 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
750 struct btrfs_root *root,
751 struct btrfs_path *path,
753 struct btrfs_dir_item *di)
758 struct extent_buffer *leaf;
759 struct btrfs_key location;
762 leaf = path->nodes[0];
764 btrfs_dir_item_key_to_cpu(leaf, di, &location);
765 name_len = btrfs_dir_name_len(leaf, di);
766 name = kmalloc(name_len, GFP_NOFS);
770 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
771 btrfs_release_path(path);
773 inode = read_one_inode(root, location.objectid);
779 ret = link_to_fixup_dir(trans, root, path, location.objectid);
783 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
787 ret = btrfs_run_delayed_items(trans, root);
795 * helper function to see if a given name and sequence number found
796 * in an inode back reference are already in a directory and correctly
797 * point to this inode
799 static noinline int inode_in_dir(struct btrfs_root *root,
800 struct btrfs_path *path,
801 u64 dirid, u64 objectid, u64 index,
802 const char *name, int name_len)
804 struct btrfs_dir_item *di;
805 struct btrfs_key location;
808 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
809 index, name, name_len, 0);
810 if (di && !IS_ERR(di)) {
811 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
812 if (location.objectid != objectid)
816 btrfs_release_path(path);
818 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
819 if (di && !IS_ERR(di)) {
820 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
821 if (location.objectid != objectid)
827 btrfs_release_path(path);
832 * helper function to check a log tree for a named back reference in
833 * an inode. This is used to decide if a back reference that is
834 * found in the subvolume conflicts with what we find in the log.
836 * inode backreferences may have multiple refs in a single item,
837 * during replay we process one reference at a time, and we don't
838 * want to delete valid links to a file from the subvolume if that
839 * link is also in the log.
841 static noinline int backref_in_log(struct btrfs_root *log,
842 struct btrfs_key *key,
844 char *name, int namelen)
846 struct btrfs_path *path;
847 struct btrfs_inode_ref *ref;
849 unsigned long ptr_end;
850 unsigned long name_ptr;
856 path = btrfs_alloc_path();
860 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
864 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
866 if (key->type == BTRFS_INODE_EXTREF_KEY) {
867 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
868 name, namelen, NULL))
874 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
875 ptr_end = ptr + item_size;
876 while (ptr < ptr_end) {
877 ref = (struct btrfs_inode_ref *)ptr;
878 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
879 if (found_name_len == namelen) {
880 name_ptr = (unsigned long)(ref + 1);
881 ret = memcmp_extent_buffer(path->nodes[0], name,
888 ptr = (unsigned long)(ref + 1) + found_name_len;
891 btrfs_free_path(path);
895 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
896 struct btrfs_root *root,
897 struct btrfs_path *path,
898 struct btrfs_root *log_root,
899 struct inode *dir, struct inode *inode,
900 struct extent_buffer *eb,
901 u64 inode_objectid, u64 parent_objectid,
902 u64 ref_index, char *name, int namelen,
908 struct extent_buffer *leaf;
909 struct btrfs_dir_item *di;
910 struct btrfs_key search_key;
911 struct btrfs_inode_extref *extref;
914 /* Search old style refs */
915 search_key.objectid = inode_objectid;
916 search_key.type = BTRFS_INODE_REF_KEY;
917 search_key.offset = parent_objectid;
918 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
920 struct btrfs_inode_ref *victim_ref;
922 unsigned long ptr_end;
924 leaf = path->nodes[0];
926 /* are we trying to overwrite a back ref for the root directory
927 * if so, just jump out, we're done
929 if (search_key.objectid == search_key.offset)
932 /* check all the names in this back reference to see
933 * if they are in the log. if so, we allow them to stay
934 * otherwise they must be unlinked as a conflict
936 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
937 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
938 while (ptr < ptr_end) {
939 victim_ref = (struct btrfs_inode_ref *)ptr;
940 victim_name_len = btrfs_inode_ref_name_len(leaf,
942 victim_name = kmalloc(victim_name_len, GFP_NOFS);
946 read_extent_buffer(leaf, victim_name,
947 (unsigned long)(victim_ref + 1),
950 if (!backref_in_log(log_root, &search_key,
955 btrfs_release_path(path);
957 ret = btrfs_unlink_inode(trans, root, dir,
963 ret = btrfs_run_delayed_items(trans, root);
971 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
975 * NOTE: we have searched root tree and checked the
976 * coresponding ref, it does not need to check again.
980 btrfs_release_path(path);
982 /* Same search but for extended refs */
983 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
984 inode_objectid, parent_objectid, 0,
986 if (!IS_ERR_OR_NULL(extref)) {
990 struct inode *victim_parent;
992 leaf = path->nodes[0];
994 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
995 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
997 while (cur_offset < item_size) {
998 extref = (struct btrfs_inode_extref *)base + cur_offset;
1000 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1002 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1005 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1008 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1011 search_key.objectid = inode_objectid;
1012 search_key.type = BTRFS_INODE_EXTREF_KEY;
1013 search_key.offset = btrfs_extref_hash(parent_objectid,
1017 if (!backref_in_log(log_root, &search_key,
1018 parent_objectid, victim_name,
1021 victim_parent = read_one_inode(root,
1023 if (victim_parent) {
1025 btrfs_release_path(path);
1027 ret = btrfs_unlink_inode(trans, root,
1033 ret = btrfs_run_delayed_items(
1036 iput(victim_parent);
1047 cur_offset += victim_name_len + sizeof(*extref);
1051 btrfs_release_path(path);
1053 /* look for a conflicting sequence number */
1054 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1055 ref_index, name, namelen, 0);
1056 if (di && !IS_ERR(di)) {
1057 ret = drop_one_dir_item(trans, root, path, dir, di);
1061 btrfs_release_path(path);
1063 /* look for a conflicing name */
1064 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1066 if (di && !IS_ERR(di)) {
1067 ret = drop_one_dir_item(trans, root, path, dir, di);
1071 btrfs_release_path(path);
1076 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1077 u32 *namelen, char **name, u64 *index,
1078 u64 *parent_objectid)
1080 struct btrfs_inode_extref *extref;
1082 extref = (struct btrfs_inode_extref *)ref_ptr;
1084 *namelen = btrfs_inode_extref_name_len(eb, extref);
1085 *name = kmalloc(*namelen, GFP_NOFS);
1089 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1092 *index = btrfs_inode_extref_index(eb, extref);
1093 if (parent_objectid)
1094 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1099 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1100 u32 *namelen, char **name, u64 *index)
1102 struct btrfs_inode_ref *ref;
1104 ref = (struct btrfs_inode_ref *)ref_ptr;
1106 *namelen = btrfs_inode_ref_name_len(eb, ref);
1107 *name = kmalloc(*namelen, GFP_NOFS);
1111 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1113 *index = btrfs_inode_ref_index(eb, ref);
1119 * replay one inode back reference item found in the log tree.
1120 * eb, slot and key refer to the buffer and key found in the log tree.
1121 * root is the destination we are replaying into, and path is for temp
1122 * use by this function. (it should be released on return).
1124 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1125 struct btrfs_root *root,
1126 struct btrfs_root *log,
1127 struct btrfs_path *path,
1128 struct extent_buffer *eb, int slot,
1129 struct btrfs_key *key)
1131 struct inode *dir = NULL;
1132 struct inode *inode = NULL;
1133 unsigned long ref_ptr;
1134 unsigned long ref_end;
1138 int search_done = 0;
1139 int log_ref_ver = 0;
1140 u64 parent_objectid;
1143 int ref_struct_size;
1145 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1146 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1148 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1149 struct btrfs_inode_extref *r;
1151 ref_struct_size = sizeof(struct btrfs_inode_extref);
1153 r = (struct btrfs_inode_extref *)ref_ptr;
1154 parent_objectid = btrfs_inode_extref_parent(eb, r);
1156 ref_struct_size = sizeof(struct btrfs_inode_ref);
1157 parent_objectid = key->offset;
1159 inode_objectid = key->objectid;
1162 * it is possible that we didn't log all the parent directories
1163 * for a given inode. If we don't find the dir, just don't
1164 * copy the back ref in. The link count fixup code will take
1167 dir = read_one_inode(root, parent_objectid);
1173 inode = read_one_inode(root, inode_objectid);
1179 while (ref_ptr < ref_end) {
1181 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1182 &ref_index, &parent_objectid);
1184 * parent object can change from one array
1188 dir = read_one_inode(root, parent_objectid);
1194 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1200 /* if we already have a perfect match, we're done */
1201 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
1202 ref_index, name, namelen)) {
1204 * look for a conflicting back reference in the
1205 * metadata. if we find one we have to unlink that name
1206 * of the file before we add our new link. Later on, we
1207 * overwrite any existing back reference, and we don't
1208 * want to create dangling pointers in the directory.
1212 ret = __add_inode_ref(trans, root, path, log,
1216 ref_index, name, namelen,
1225 /* insert our name */
1226 ret = btrfs_add_link(trans, dir, inode, name, namelen,
1231 btrfs_update_inode(trans, root, inode);
1234 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1243 /* finally write the back reference in the inode */
1244 ret = overwrite_item(trans, root, path, eb, slot, key);
1246 btrfs_release_path(path);
1253 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1254 struct btrfs_root *root, u64 offset)
1257 ret = btrfs_find_item(root, NULL, BTRFS_ORPHAN_OBJECTID,
1258 offset, BTRFS_ORPHAN_ITEM_KEY, NULL);
1260 ret = btrfs_insert_orphan_item(trans, root, offset);
1264 static int count_inode_extrefs(struct btrfs_root *root,
1265 struct inode *inode, struct btrfs_path *path)
1269 unsigned int nlink = 0;
1272 u64 inode_objectid = btrfs_ino(inode);
1275 struct btrfs_inode_extref *extref;
1276 struct extent_buffer *leaf;
1279 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1284 leaf = path->nodes[0];
1285 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1286 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1288 while (cur_offset < item_size) {
1289 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1290 name_len = btrfs_inode_extref_name_len(leaf, extref);
1294 cur_offset += name_len + sizeof(*extref);
1298 btrfs_release_path(path);
1300 btrfs_release_path(path);
1307 static int count_inode_refs(struct btrfs_root *root,
1308 struct inode *inode, struct btrfs_path *path)
1311 struct btrfs_key key;
1312 unsigned int nlink = 0;
1314 unsigned long ptr_end;
1316 u64 ino = btrfs_ino(inode);
1319 key.type = BTRFS_INODE_REF_KEY;
1320 key.offset = (u64)-1;
1323 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1327 if (path->slots[0] == 0)
1332 btrfs_item_key_to_cpu(path->nodes[0], &key,
1334 if (key.objectid != ino ||
1335 key.type != BTRFS_INODE_REF_KEY)
1337 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1338 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1340 while (ptr < ptr_end) {
1341 struct btrfs_inode_ref *ref;
1343 ref = (struct btrfs_inode_ref *)ptr;
1344 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1346 ptr = (unsigned long)(ref + 1) + name_len;
1350 if (key.offset == 0)
1352 if (path->slots[0] > 0) {
1357 btrfs_release_path(path);
1359 btrfs_release_path(path);
1365 * There are a few corners where the link count of the file can't
1366 * be properly maintained during replay. So, instead of adding
1367 * lots of complexity to the log code, we just scan the backrefs
1368 * for any file that has been through replay.
1370 * The scan will update the link count on the inode to reflect the
1371 * number of back refs found. If it goes down to zero, the iput
1372 * will free the inode.
1374 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1375 struct btrfs_root *root,
1376 struct inode *inode)
1378 struct btrfs_path *path;
1381 u64 ino = btrfs_ino(inode);
1383 path = btrfs_alloc_path();
1387 ret = count_inode_refs(root, inode, path);
1393 ret = count_inode_extrefs(root, inode, path);
1404 if (nlink != inode->i_nlink) {
1405 set_nlink(inode, nlink);
1406 btrfs_update_inode(trans, root, inode);
1408 BTRFS_I(inode)->index_cnt = (u64)-1;
1410 if (inode->i_nlink == 0) {
1411 if (S_ISDIR(inode->i_mode)) {
1412 ret = replay_dir_deletes(trans, root, NULL, path,
1417 ret = insert_orphan_item(trans, root, ino);
1421 btrfs_free_path(path);
1425 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1426 struct btrfs_root *root,
1427 struct btrfs_path *path)
1430 struct btrfs_key key;
1431 struct inode *inode;
1433 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1434 key.type = BTRFS_ORPHAN_ITEM_KEY;
1435 key.offset = (u64)-1;
1437 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1442 if (path->slots[0] == 0)
1447 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1448 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1449 key.type != BTRFS_ORPHAN_ITEM_KEY)
1452 ret = btrfs_del_item(trans, root, path);
1456 btrfs_release_path(path);
1457 inode = read_one_inode(root, key.offset);
1461 ret = fixup_inode_link_count(trans, root, inode);
1467 * fixup on a directory may create new entries,
1468 * make sure we always look for the highset possible
1471 key.offset = (u64)-1;
1475 btrfs_release_path(path);
1481 * record a given inode in the fixup dir so we can check its link
1482 * count when replay is done. The link count is incremented here
1483 * so the inode won't go away until we check it
1485 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1486 struct btrfs_root *root,
1487 struct btrfs_path *path,
1490 struct btrfs_key key;
1492 struct inode *inode;
1494 inode = read_one_inode(root, objectid);
1498 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1499 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1500 key.offset = objectid;
1502 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1504 btrfs_release_path(path);
1506 if (!inode->i_nlink)
1507 set_nlink(inode, 1);
1510 ret = btrfs_update_inode(trans, root, inode);
1511 } else if (ret == -EEXIST) {
1514 BUG(); /* Logic Error */
1522 * when replaying the log for a directory, we only insert names
1523 * for inodes that actually exist. This means an fsync on a directory
1524 * does not implicitly fsync all the new files in it
1526 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1527 struct btrfs_root *root,
1528 struct btrfs_path *path,
1529 u64 dirid, u64 index,
1530 char *name, int name_len, u8 type,
1531 struct btrfs_key *location)
1533 struct inode *inode;
1537 inode = read_one_inode(root, location->objectid);
1541 dir = read_one_inode(root, dirid);
1547 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1549 /* FIXME, put inode into FIXUP list */
1557 * take a single entry in a log directory item and replay it into
1560 * if a conflicting item exists in the subdirectory already,
1561 * the inode it points to is unlinked and put into the link count
1564 * If a name from the log points to a file or directory that does
1565 * not exist in the FS, it is skipped. fsyncs on directories
1566 * do not force down inodes inside that directory, just changes to the
1567 * names or unlinks in a directory.
1569 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1570 struct btrfs_root *root,
1571 struct btrfs_path *path,
1572 struct extent_buffer *eb,
1573 struct btrfs_dir_item *di,
1574 struct btrfs_key *key)
1578 struct btrfs_dir_item *dst_di;
1579 struct btrfs_key found_key;
1580 struct btrfs_key log_key;
1585 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1587 dir = read_one_inode(root, key->objectid);
1591 name_len = btrfs_dir_name_len(eb, di);
1592 name = kmalloc(name_len, GFP_NOFS);
1598 log_type = btrfs_dir_type(eb, di);
1599 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1602 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1603 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1608 btrfs_release_path(path);
1610 if (key->type == BTRFS_DIR_ITEM_KEY) {
1611 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1613 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1614 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1623 if (IS_ERR_OR_NULL(dst_di)) {
1624 /* we need a sequence number to insert, so we only
1625 * do inserts for the BTRFS_DIR_INDEX_KEY types
1627 if (key->type != BTRFS_DIR_INDEX_KEY)
1632 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1633 /* the existing item matches the logged item */
1634 if (found_key.objectid == log_key.objectid &&
1635 found_key.type == log_key.type &&
1636 found_key.offset == log_key.offset &&
1637 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1642 * don't drop the conflicting directory entry if the inode
1643 * for the new entry doesn't exist
1648 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1652 if (key->type == BTRFS_DIR_INDEX_KEY)
1655 btrfs_release_path(path);
1656 if (!ret && update_size) {
1657 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1658 ret = btrfs_update_inode(trans, root, dir);
1665 btrfs_release_path(path);
1666 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1667 name, name_len, log_type, &log_key);
1668 if (ret && ret != -ENOENT)
1670 update_size = false;
1676 * find all the names in a directory item and reconcile them into
1677 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1678 * one name in a directory item, but the same code gets used for
1679 * both directory index types
1681 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1682 struct btrfs_root *root,
1683 struct btrfs_path *path,
1684 struct extent_buffer *eb, int slot,
1685 struct btrfs_key *key)
1688 u32 item_size = btrfs_item_size_nr(eb, slot);
1689 struct btrfs_dir_item *di;
1692 unsigned long ptr_end;
1694 ptr = btrfs_item_ptr_offset(eb, slot);
1695 ptr_end = ptr + item_size;
1696 while (ptr < ptr_end) {
1697 di = (struct btrfs_dir_item *)ptr;
1698 if (verify_dir_item(root, eb, di))
1700 name_len = btrfs_dir_name_len(eb, di);
1701 ret = replay_one_name(trans, root, path, eb, di, key);
1704 ptr = (unsigned long)(di + 1);
1711 * directory replay has two parts. There are the standard directory
1712 * items in the log copied from the subvolume, and range items
1713 * created in the log while the subvolume was logged.
1715 * The range items tell us which parts of the key space the log
1716 * is authoritative for. During replay, if a key in the subvolume
1717 * directory is in a logged range item, but not actually in the log
1718 * that means it was deleted from the directory before the fsync
1719 * and should be removed.
1721 static noinline int find_dir_range(struct btrfs_root *root,
1722 struct btrfs_path *path,
1723 u64 dirid, int key_type,
1724 u64 *start_ret, u64 *end_ret)
1726 struct btrfs_key key;
1728 struct btrfs_dir_log_item *item;
1732 if (*start_ret == (u64)-1)
1735 key.objectid = dirid;
1736 key.type = key_type;
1737 key.offset = *start_ret;
1739 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1743 if (path->slots[0] == 0)
1748 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1750 if (key.type != key_type || key.objectid != dirid) {
1754 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1755 struct btrfs_dir_log_item);
1756 found_end = btrfs_dir_log_end(path->nodes[0], item);
1758 if (*start_ret >= key.offset && *start_ret <= found_end) {
1760 *start_ret = key.offset;
1761 *end_ret = found_end;
1766 /* check the next slot in the tree to see if it is a valid item */
1767 nritems = btrfs_header_nritems(path->nodes[0]);
1768 if (path->slots[0] >= nritems) {
1769 ret = btrfs_next_leaf(root, path);
1776 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1778 if (key.type != key_type || key.objectid != dirid) {
1782 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1783 struct btrfs_dir_log_item);
1784 found_end = btrfs_dir_log_end(path->nodes[0], item);
1785 *start_ret = key.offset;
1786 *end_ret = found_end;
1789 btrfs_release_path(path);
1794 * this looks for a given directory item in the log. If the directory
1795 * item is not in the log, the item is removed and the inode it points
1798 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1799 struct btrfs_root *root,
1800 struct btrfs_root *log,
1801 struct btrfs_path *path,
1802 struct btrfs_path *log_path,
1804 struct btrfs_key *dir_key)
1807 struct extent_buffer *eb;
1810 struct btrfs_dir_item *di;
1811 struct btrfs_dir_item *log_di;
1814 unsigned long ptr_end;
1816 struct inode *inode;
1817 struct btrfs_key location;
1820 eb = path->nodes[0];
1821 slot = path->slots[0];
1822 item_size = btrfs_item_size_nr(eb, slot);
1823 ptr = btrfs_item_ptr_offset(eb, slot);
1824 ptr_end = ptr + item_size;
1825 while (ptr < ptr_end) {
1826 di = (struct btrfs_dir_item *)ptr;
1827 if (verify_dir_item(root, eb, di)) {
1832 name_len = btrfs_dir_name_len(eb, di);
1833 name = kmalloc(name_len, GFP_NOFS);
1838 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1841 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1842 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1845 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1846 log_di = btrfs_lookup_dir_index_item(trans, log,
1852 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
1853 btrfs_dir_item_key_to_cpu(eb, di, &location);
1854 btrfs_release_path(path);
1855 btrfs_release_path(log_path);
1856 inode = read_one_inode(root, location.objectid);
1862 ret = link_to_fixup_dir(trans, root,
1863 path, location.objectid);
1871 ret = btrfs_unlink_inode(trans, root, dir, inode,
1874 ret = btrfs_run_delayed_items(trans, root);
1880 /* there might still be more names under this key
1881 * check and repeat if required
1883 ret = btrfs_search_slot(NULL, root, dir_key, path,
1889 } else if (IS_ERR(log_di)) {
1891 return PTR_ERR(log_di);
1893 btrfs_release_path(log_path);
1896 ptr = (unsigned long)(di + 1);
1901 btrfs_release_path(path);
1902 btrfs_release_path(log_path);
1907 * deletion replay happens before we copy any new directory items
1908 * out of the log or out of backreferences from inodes. It
1909 * scans the log to find ranges of keys that log is authoritative for,
1910 * and then scans the directory to find items in those ranges that are
1911 * not present in the log.
1913 * Anything we don't find in the log is unlinked and removed from the
1916 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1917 struct btrfs_root *root,
1918 struct btrfs_root *log,
1919 struct btrfs_path *path,
1920 u64 dirid, int del_all)
1924 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1926 struct btrfs_key dir_key;
1927 struct btrfs_key found_key;
1928 struct btrfs_path *log_path;
1931 dir_key.objectid = dirid;
1932 dir_key.type = BTRFS_DIR_ITEM_KEY;
1933 log_path = btrfs_alloc_path();
1937 dir = read_one_inode(root, dirid);
1938 /* it isn't an error if the inode isn't there, that can happen
1939 * because we replay the deletes before we copy in the inode item
1943 btrfs_free_path(log_path);
1951 range_end = (u64)-1;
1953 ret = find_dir_range(log, path, dirid, key_type,
1954 &range_start, &range_end);
1959 dir_key.offset = range_start;
1962 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1967 nritems = btrfs_header_nritems(path->nodes[0]);
1968 if (path->slots[0] >= nritems) {
1969 ret = btrfs_next_leaf(root, path);
1973 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1975 if (found_key.objectid != dirid ||
1976 found_key.type != dir_key.type)
1979 if (found_key.offset > range_end)
1982 ret = check_item_in_log(trans, root, log, path,
1987 if (found_key.offset == (u64)-1)
1989 dir_key.offset = found_key.offset + 1;
1991 btrfs_release_path(path);
1992 if (range_end == (u64)-1)
1994 range_start = range_end + 1;
1999 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2000 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2001 dir_key.type = BTRFS_DIR_INDEX_KEY;
2002 btrfs_release_path(path);
2006 btrfs_release_path(path);
2007 btrfs_free_path(log_path);
2013 * the process_func used to replay items from the log tree. This
2014 * gets called in two different stages. The first stage just looks
2015 * for inodes and makes sure they are all copied into the subvolume.
2017 * The second stage copies all the other item types from the log into
2018 * the subvolume. The two stage approach is slower, but gets rid of
2019 * lots of complexity around inodes referencing other inodes that exist
2020 * only in the log (references come from either directory items or inode
2023 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2024 struct walk_control *wc, u64 gen)
2027 struct btrfs_path *path;
2028 struct btrfs_root *root = wc->replay_dest;
2029 struct btrfs_key key;
2034 ret = btrfs_read_buffer(eb, gen);
2038 level = btrfs_header_level(eb);
2043 path = btrfs_alloc_path();
2047 nritems = btrfs_header_nritems(eb);
2048 for (i = 0; i < nritems; i++) {
2049 btrfs_item_key_to_cpu(eb, &key, i);
2051 /* inode keys are done during the first stage */
2052 if (key.type == BTRFS_INODE_ITEM_KEY &&
2053 wc->stage == LOG_WALK_REPLAY_INODES) {
2054 struct btrfs_inode_item *inode_item;
2057 inode_item = btrfs_item_ptr(eb, i,
2058 struct btrfs_inode_item);
2059 mode = btrfs_inode_mode(eb, inode_item);
2060 if (S_ISDIR(mode)) {
2061 ret = replay_dir_deletes(wc->trans,
2062 root, log, path, key.objectid, 0);
2066 ret = overwrite_item(wc->trans, root, path,
2071 /* for regular files, make sure corresponding
2072 * orhpan item exist. extents past the new EOF
2073 * will be truncated later by orphan cleanup.
2075 if (S_ISREG(mode)) {
2076 ret = insert_orphan_item(wc->trans, root,
2082 ret = link_to_fixup_dir(wc->trans, root,
2083 path, key.objectid);
2088 if (key.type == BTRFS_DIR_INDEX_KEY &&
2089 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2090 ret = replay_one_dir_item(wc->trans, root, path,
2096 if (wc->stage < LOG_WALK_REPLAY_ALL)
2099 /* these keys are simply copied */
2100 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2101 ret = overwrite_item(wc->trans, root, path,
2105 } else if (key.type == BTRFS_INODE_REF_KEY ||
2106 key.type == BTRFS_INODE_EXTREF_KEY) {
2107 ret = add_inode_ref(wc->trans, root, log, path,
2109 if (ret && ret != -ENOENT)
2112 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2113 ret = replay_one_extent(wc->trans, root, path,
2117 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2118 ret = replay_one_dir_item(wc->trans, root, path,
2124 btrfs_free_path(path);
2128 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2129 struct btrfs_root *root,
2130 struct btrfs_path *path, int *level,
2131 struct walk_control *wc)
2136 struct extent_buffer *next;
2137 struct extent_buffer *cur;
2138 struct extent_buffer *parent;
2142 WARN_ON(*level < 0);
2143 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2145 while (*level > 0) {
2146 WARN_ON(*level < 0);
2147 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2148 cur = path->nodes[*level];
2150 WARN_ON(btrfs_header_level(cur) != *level);
2152 if (path->slots[*level] >=
2153 btrfs_header_nritems(cur))
2156 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2157 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2158 blocksize = btrfs_level_size(root, *level - 1);
2160 parent = path->nodes[*level];
2161 root_owner = btrfs_header_owner(parent);
2163 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
2168 ret = wc->process_func(root, next, wc, ptr_gen);
2170 free_extent_buffer(next);
2174 path->slots[*level]++;
2176 ret = btrfs_read_buffer(next, ptr_gen);
2178 free_extent_buffer(next);
2183 btrfs_tree_lock(next);
2184 btrfs_set_lock_blocking(next);
2185 clean_tree_block(trans, root, next);
2186 btrfs_wait_tree_block_writeback(next);
2187 btrfs_tree_unlock(next);
2190 WARN_ON(root_owner !=
2191 BTRFS_TREE_LOG_OBJECTID);
2192 ret = btrfs_free_and_pin_reserved_extent(root,
2195 free_extent_buffer(next);
2199 free_extent_buffer(next);
2202 ret = btrfs_read_buffer(next, ptr_gen);
2204 free_extent_buffer(next);
2208 WARN_ON(*level <= 0);
2209 if (path->nodes[*level-1])
2210 free_extent_buffer(path->nodes[*level-1]);
2211 path->nodes[*level-1] = next;
2212 *level = btrfs_header_level(next);
2213 path->slots[*level] = 0;
2216 WARN_ON(*level < 0);
2217 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2219 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2225 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2226 struct btrfs_root *root,
2227 struct btrfs_path *path, int *level,
2228 struct walk_control *wc)
2235 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2236 slot = path->slots[i];
2237 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2240 WARN_ON(*level == 0);
2243 struct extent_buffer *parent;
2244 if (path->nodes[*level] == root->node)
2245 parent = path->nodes[*level];
2247 parent = path->nodes[*level + 1];
2249 root_owner = btrfs_header_owner(parent);
2250 ret = wc->process_func(root, path->nodes[*level], wc,
2251 btrfs_header_generation(path->nodes[*level]));
2256 struct extent_buffer *next;
2258 next = path->nodes[*level];
2261 btrfs_tree_lock(next);
2262 btrfs_set_lock_blocking(next);
2263 clean_tree_block(trans, root, next);
2264 btrfs_wait_tree_block_writeback(next);
2265 btrfs_tree_unlock(next);
2268 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2269 ret = btrfs_free_and_pin_reserved_extent(root,
2270 path->nodes[*level]->start,
2271 path->nodes[*level]->len);
2275 free_extent_buffer(path->nodes[*level]);
2276 path->nodes[*level] = NULL;
2284 * drop the reference count on the tree rooted at 'snap'. This traverses
2285 * the tree freeing any blocks that have a ref count of zero after being
2288 static int walk_log_tree(struct btrfs_trans_handle *trans,
2289 struct btrfs_root *log, struct walk_control *wc)
2294 struct btrfs_path *path;
2297 path = btrfs_alloc_path();
2301 level = btrfs_header_level(log->node);
2303 path->nodes[level] = log->node;
2304 extent_buffer_get(log->node);
2305 path->slots[level] = 0;
2308 wret = walk_down_log_tree(trans, log, path, &level, wc);
2316 wret = walk_up_log_tree(trans, log, path, &level, wc);
2325 /* was the root node processed? if not, catch it here */
2326 if (path->nodes[orig_level]) {
2327 ret = wc->process_func(log, path->nodes[orig_level], wc,
2328 btrfs_header_generation(path->nodes[orig_level]));
2332 struct extent_buffer *next;
2334 next = path->nodes[orig_level];
2337 btrfs_tree_lock(next);
2338 btrfs_set_lock_blocking(next);
2339 clean_tree_block(trans, log, next);
2340 btrfs_wait_tree_block_writeback(next);
2341 btrfs_tree_unlock(next);
2344 WARN_ON(log->root_key.objectid !=
2345 BTRFS_TREE_LOG_OBJECTID);
2346 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2354 btrfs_free_path(path);
2359 * helper function to update the item for a given subvolumes log root
2360 * in the tree of log roots
2362 static int update_log_root(struct btrfs_trans_handle *trans,
2363 struct btrfs_root *log)
2367 if (log->log_transid == 1) {
2368 /* insert root item on the first sync */
2369 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2370 &log->root_key, &log->root_item);
2372 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2373 &log->root_key, &log->root_item);
2378 static void wait_log_commit(struct btrfs_trans_handle *trans,
2379 struct btrfs_root *root, int transid)
2382 int index = transid % 2;
2385 * we only allow two pending log transactions at a time,
2386 * so we know that if ours is more than 2 older than the
2387 * current transaction, we're done
2390 prepare_to_wait(&root->log_commit_wait[index],
2391 &wait, TASK_UNINTERRUPTIBLE);
2392 mutex_unlock(&root->log_mutex);
2394 if (root->log_transid_committed < transid &&
2395 atomic_read(&root->log_commit[index]))
2398 finish_wait(&root->log_commit_wait[index], &wait);
2399 mutex_lock(&root->log_mutex);
2400 } while (root->log_transid_committed < transid &&
2401 atomic_read(&root->log_commit[index]));
2404 static void wait_for_writer(struct btrfs_trans_handle *trans,
2405 struct btrfs_root *root)
2409 while (atomic_read(&root->log_writers)) {
2410 prepare_to_wait(&root->log_writer_wait,
2411 &wait, TASK_UNINTERRUPTIBLE);
2412 mutex_unlock(&root->log_mutex);
2413 if (atomic_read(&root->log_writers))
2415 mutex_lock(&root->log_mutex);
2416 finish_wait(&root->log_writer_wait, &wait);
2420 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2421 struct btrfs_log_ctx *ctx)
2426 mutex_lock(&root->log_mutex);
2427 list_del_init(&ctx->list);
2428 mutex_unlock(&root->log_mutex);
2432 * Invoked in log mutex context, or be sure there is no other task which
2433 * can access the list.
2435 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2436 int index, int error)
2438 struct btrfs_log_ctx *ctx;
2441 INIT_LIST_HEAD(&root->log_ctxs[index]);
2445 list_for_each_entry(ctx, &root->log_ctxs[index], list)
2446 ctx->log_ret = error;
2448 INIT_LIST_HEAD(&root->log_ctxs[index]);
2452 * btrfs_sync_log does sends a given tree log down to the disk and
2453 * updates the super blocks to record it. When this call is done,
2454 * you know that any inodes previously logged are safely on disk only
2457 * Any other return value means you need to call btrfs_commit_transaction.
2458 * Some of the edge cases for fsyncing directories that have had unlinks
2459 * or renames done in the past mean that sometimes the only safe
2460 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2461 * that has happened.
2463 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2464 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2470 struct btrfs_root *log = root->log_root;
2471 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2472 int log_transid = 0;
2473 struct btrfs_log_ctx root_log_ctx;
2474 struct blk_plug plug;
2476 mutex_lock(&root->log_mutex);
2477 log_transid = ctx->log_transid;
2478 if (root->log_transid_committed >= log_transid) {
2479 mutex_unlock(&root->log_mutex);
2480 return ctx->log_ret;
2483 index1 = log_transid % 2;
2484 if (atomic_read(&root->log_commit[index1])) {
2485 wait_log_commit(trans, root, log_transid);
2486 mutex_unlock(&root->log_mutex);
2487 return ctx->log_ret;
2489 ASSERT(log_transid == root->log_transid);
2490 atomic_set(&root->log_commit[index1], 1);
2492 /* wait for previous tree log sync to complete */
2493 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2494 wait_log_commit(trans, root, log_transid - 1);
2497 int batch = atomic_read(&root->log_batch);
2498 /* when we're on an ssd, just kick the log commit out */
2499 if (!btrfs_test_opt(root, SSD) &&
2500 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2501 mutex_unlock(&root->log_mutex);
2502 schedule_timeout_uninterruptible(1);
2503 mutex_lock(&root->log_mutex);
2505 wait_for_writer(trans, root);
2506 if (batch == atomic_read(&root->log_batch))
2510 /* bail out if we need to do a full commit */
2511 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2513 btrfs_free_logged_extents(log, log_transid);
2514 mutex_unlock(&root->log_mutex);
2518 if (log_transid % 2 == 0)
2519 mark = EXTENT_DIRTY;
2523 /* we start IO on all the marked extents here, but we don't actually
2524 * wait for them until later.
2526 blk_start_plug(&plug);
2527 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2529 blk_finish_plug(&plug);
2530 btrfs_abort_transaction(trans, root, ret);
2531 btrfs_free_logged_extents(log, log_transid);
2532 btrfs_set_log_full_commit(root->fs_info, trans);
2533 mutex_unlock(&root->log_mutex);
2537 btrfs_set_root_node(&log->root_item, log->node);
2539 root->log_transid++;
2540 log->log_transid = root->log_transid;
2541 root->log_start_pid = 0;
2543 * IO has been started, blocks of the log tree have WRITTEN flag set
2544 * in their headers. new modifications of the log will be written to
2545 * new positions. so it's safe to allow log writers to go in.
2547 mutex_unlock(&root->log_mutex);
2549 btrfs_init_log_ctx(&root_log_ctx);
2551 mutex_lock(&log_root_tree->log_mutex);
2552 atomic_inc(&log_root_tree->log_batch);
2553 atomic_inc(&log_root_tree->log_writers);
2555 index2 = log_root_tree->log_transid % 2;
2556 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2557 root_log_ctx.log_transid = log_root_tree->log_transid;
2559 mutex_unlock(&log_root_tree->log_mutex);
2561 ret = update_log_root(trans, log);
2563 mutex_lock(&log_root_tree->log_mutex);
2564 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2566 if (waitqueue_active(&log_root_tree->log_writer_wait))
2567 wake_up(&log_root_tree->log_writer_wait);
2571 if (!list_empty(&root_log_ctx.list))
2572 list_del_init(&root_log_ctx.list);
2574 blk_finish_plug(&plug);
2575 btrfs_set_log_full_commit(root->fs_info, trans);
2577 if (ret != -ENOSPC) {
2578 btrfs_abort_transaction(trans, root, ret);
2579 mutex_unlock(&log_root_tree->log_mutex);
2582 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2583 btrfs_free_logged_extents(log, log_transid);
2584 mutex_unlock(&log_root_tree->log_mutex);
2589 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2590 mutex_unlock(&log_root_tree->log_mutex);
2591 ret = root_log_ctx.log_ret;
2595 index2 = root_log_ctx.log_transid % 2;
2596 if (atomic_read(&log_root_tree->log_commit[index2])) {
2597 blk_finish_plug(&plug);
2598 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2599 wait_log_commit(trans, log_root_tree,
2600 root_log_ctx.log_transid);
2601 btrfs_free_logged_extents(log, log_transid);
2602 mutex_unlock(&log_root_tree->log_mutex);
2603 ret = root_log_ctx.log_ret;
2606 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
2607 atomic_set(&log_root_tree->log_commit[index2], 1);
2609 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2610 wait_log_commit(trans, log_root_tree,
2611 root_log_ctx.log_transid - 1);
2614 wait_for_writer(trans, log_root_tree);
2617 * now that we've moved on to the tree of log tree roots,
2618 * check the full commit flag again
2620 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2621 blk_finish_plug(&plug);
2622 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2623 btrfs_free_logged_extents(log, log_transid);
2624 mutex_unlock(&log_root_tree->log_mutex);
2626 goto out_wake_log_root;
2629 ret = btrfs_write_marked_extents(log_root_tree,
2630 &log_root_tree->dirty_log_pages,
2631 EXTENT_DIRTY | EXTENT_NEW);
2632 blk_finish_plug(&plug);
2634 btrfs_set_log_full_commit(root->fs_info, trans);
2635 btrfs_abort_transaction(trans, root, ret);
2636 btrfs_free_logged_extents(log, log_transid);
2637 mutex_unlock(&log_root_tree->log_mutex);
2638 goto out_wake_log_root;
2640 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2641 btrfs_wait_marked_extents(log_root_tree,
2642 &log_root_tree->dirty_log_pages,
2643 EXTENT_NEW | EXTENT_DIRTY);
2644 btrfs_wait_logged_extents(log, log_transid);
2646 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2647 log_root_tree->node->start);
2648 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2649 btrfs_header_level(log_root_tree->node));
2651 log_root_tree->log_transid++;
2652 mutex_unlock(&log_root_tree->log_mutex);
2655 * nobody else is going to jump in and write the the ctree
2656 * super here because the log_commit atomic below is protecting
2657 * us. We must be called with a transaction handle pinning
2658 * the running transaction open, so a full commit can't hop
2659 * in and cause problems either.
2661 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2663 btrfs_set_log_full_commit(root->fs_info, trans);
2664 btrfs_abort_transaction(trans, root, ret);
2665 goto out_wake_log_root;
2668 mutex_lock(&root->log_mutex);
2669 if (root->last_log_commit < log_transid)
2670 root->last_log_commit = log_transid;
2671 mutex_unlock(&root->log_mutex);
2675 * We needn't get log_mutex here because we are sure all
2676 * the other tasks are blocked.
2678 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2680 mutex_lock(&log_root_tree->log_mutex);
2681 log_root_tree->log_transid_committed++;
2682 atomic_set(&log_root_tree->log_commit[index2], 0);
2683 mutex_unlock(&log_root_tree->log_mutex);
2685 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2686 wake_up(&log_root_tree->log_commit_wait[index2]);
2689 btrfs_remove_all_log_ctxs(root, index1, ret);
2691 mutex_lock(&root->log_mutex);
2692 root->log_transid_committed++;
2693 atomic_set(&root->log_commit[index1], 0);
2694 mutex_unlock(&root->log_mutex);
2696 if (waitqueue_active(&root->log_commit_wait[index1]))
2697 wake_up(&root->log_commit_wait[index1]);
2701 static void free_log_tree(struct btrfs_trans_handle *trans,
2702 struct btrfs_root *log)
2707 struct walk_control wc = {
2709 .process_func = process_one_buffer
2712 ret = walk_log_tree(trans, log, &wc);
2713 /* I don't think this can happen but just in case */
2715 btrfs_abort_transaction(trans, log, ret);
2718 ret = find_first_extent_bit(&log->dirty_log_pages,
2719 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2724 clear_extent_bits(&log->dirty_log_pages, start, end,
2725 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2729 * We may have short-circuited the log tree with the full commit logic
2730 * and left ordered extents on our list, so clear these out to keep us
2731 * from leaking inodes and memory.
2733 btrfs_free_logged_extents(log, 0);
2734 btrfs_free_logged_extents(log, 1);
2736 free_extent_buffer(log->node);
2741 * free all the extents used by the tree log. This should be called
2742 * at commit time of the full transaction
2744 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2746 if (root->log_root) {
2747 free_log_tree(trans, root->log_root);
2748 root->log_root = NULL;
2753 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2754 struct btrfs_fs_info *fs_info)
2756 if (fs_info->log_root_tree) {
2757 free_log_tree(trans, fs_info->log_root_tree);
2758 fs_info->log_root_tree = NULL;
2764 * If both a file and directory are logged, and unlinks or renames are
2765 * mixed in, we have a few interesting corners:
2767 * create file X in dir Y
2768 * link file X to X.link in dir Y
2770 * unlink file X but leave X.link
2773 * After a crash we would expect only X.link to exist. But file X
2774 * didn't get fsync'd again so the log has back refs for X and X.link.
2776 * We solve this by removing directory entries and inode backrefs from the
2777 * log when a file that was logged in the current transaction is
2778 * unlinked. Any later fsync will include the updated log entries, and
2779 * we'll be able to reconstruct the proper directory items from backrefs.
2781 * This optimizations allows us to avoid relogging the entire inode
2782 * or the entire directory.
2784 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2785 struct btrfs_root *root,
2786 const char *name, int name_len,
2787 struct inode *dir, u64 index)
2789 struct btrfs_root *log;
2790 struct btrfs_dir_item *di;
2791 struct btrfs_path *path;
2795 u64 dir_ino = btrfs_ino(dir);
2797 if (BTRFS_I(dir)->logged_trans < trans->transid)
2800 ret = join_running_log_trans(root);
2804 mutex_lock(&BTRFS_I(dir)->log_mutex);
2806 log = root->log_root;
2807 path = btrfs_alloc_path();
2813 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
2814 name, name_len, -1);
2820 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2821 bytes_del += name_len;
2827 btrfs_release_path(path);
2828 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
2829 index, name, name_len, -1);
2835 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2836 bytes_del += name_len;
2843 /* update the directory size in the log to reflect the names
2847 struct btrfs_key key;
2849 key.objectid = dir_ino;
2851 key.type = BTRFS_INODE_ITEM_KEY;
2852 btrfs_release_path(path);
2854 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2860 struct btrfs_inode_item *item;
2863 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2864 struct btrfs_inode_item);
2865 i_size = btrfs_inode_size(path->nodes[0], item);
2866 if (i_size > bytes_del)
2867 i_size -= bytes_del;
2870 btrfs_set_inode_size(path->nodes[0], item, i_size);
2871 btrfs_mark_buffer_dirty(path->nodes[0]);
2874 btrfs_release_path(path);
2877 btrfs_free_path(path);
2879 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2880 if (ret == -ENOSPC) {
2881 btrfs_set_log_full_commit(root->fs_info, trans);
2884 btrfs_abort_transaction(trans, root, ret);
2886 btrfs_end_log_trans(root);
2891 /* see comments for btrfs_del_dir_entries_in_log */
2892 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2893 struct btrfs_root *root,
2894 const char *name, int name_len,
2895 struct inode *inode, u64 dirid)
2897 struct btrfs_root *log;
2901 if (BTRFS_I(inode)->logged_trans < trans->transid)
2904 ret = join_running_log_trans(root);
2907 log = root->log_root;
2908 mutex_lock(&BTRFS_I(inode)->log_mutex);
2910 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
2912 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2913 if (ret == -ENOSPC) {
2914 btrfs_set_log_full_commit(root->fs_info, trans);
2916 } else if (ret < 0 && ret != -ENOENT)
2917 btrfs_abort_transaction(trans, root, ret);
2918 btrfs_end_log_trans(root);
2924 * creates a range item in the log for 'dirid'. first_offset and
2925 * last_offset tell us which parts of the key space the log should
2926 * be considered authoritative for.
2928 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2929 struct btrfs_root *log,
2930 struct btrfs_path *path,
2931 int key_type, u64 dirid,
2932 u64 first_offset, u64 last_offset)
2935 struct btrfs_key key;
2936 struct btrfs_dir_log_item *item;
2938 key.objectid = dirid;
2939 key.offset = first_offset;
2940 if (key_type == BTRFS_DIR_ITEM_KEY)
2941 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2943 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2944 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2948 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2949 struct btrfs_dir_log_item);
2950 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2951 btrfs_mark_buffer_dirty(path->nodes[0]);
2952 btrfs_release_path(path);
2957 * log all the items included in the current transaction for a given
2958 * directory. This also creates the range items in the log tree required
2959 * to replay anything deleted before the fsync
2961 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2962 struct btrfs_root *root, struct inode *inode,
2963 struct btrfs_path *path,
2964 struct btrfs_path *dst_path, int key_type,
2965 u64 min_offset, u64 *last_offset_ret)
2967 struct btrfs_key min_key;
2968 struct btrfs_root *log = root->log_root;
2969 struct extent_buffer *src;
2974 u64 first_offset = min_offset;
2975 u64 last_offset = (u64)-1;
2976 u64 ino = btrfs_ino(inode);
2978 log = root->log_root;
2980 min_key.objectid = ino;
2981 min_key.type = key_type;
2982 min_key.offset = min_offset;
2984 path->keep_locks = 1;
2986 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
2989 * we didn't find anything from this transaction, see if there
2990 * is anything at all
2992 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2993 min_key.objectid = ino;
2994 min_key.type = key_type;
2995 min_key.offset = (u64)-1;
2996 btrfs_release_path(path);
2997 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2999 btrfs_release_path(path);
3002 ret = btrfs_previous_item(root, path, ino, key_type);
3004 /* if ret == 0 there are items for this type,
3005 * create a range to tell us the last key of this type.
3006 * otherwise, there are no items in this directory after
3007 * *min_offset, and we create a range to indicate that.
3010 struct btrfs_key tmp;
3011 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3013 if (key_type == tmp.type)
3014 first_offset = max(min_offset, tmp.offset) + 1;
3019 /* go backward to find any previous key */
3020 ret = btrfs_previous_item(root, path, ino, key_type);
3022 struct btrfs_key tmp;
3023 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3024 if (key_type == tmp.type) {
3025 first_offset = tmp.offset;
3026 ret = overwrite_item(trans, log, dst_path,
3027 path->nodes[0], path->slots[0],
3035 btrfs_release_path(path);
3037 /* find the first key from this transaction again */
3038 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3039 if (WARN_ON(ret != 0))
3043 * we have a block from this transaction, log every item in it
3044 * from our directory
3047 struct btrfs_key tmp;
3048 src = path->nodes[0];
3049 nritems = btrfs_header_nritems(src);
3050 for (i = path->slots[0]; i < nritems; i++) {
3051 btrfs_item_key_to_cpu(src, &min_key, i);
3053 if (min_key.objectid != ino || min_key.type != key_type)
3055 ret = overwrite_item(trans, log, dst_path, src, i,
3062 path->slots[0] = nritems;
3065 * look ahead to the next item and see if it is also
3066 * from this directory and from this transaction
3068 ret = btrfs_next_leaf(root, path);
3070 last_offset = (u64)-1;
3073 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3074 if (tmp.objectid != ino || tmp.type != key_type) {
3075 last_offset = (u64)-1;
3078 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3079 ret = overwrite_item(trans, log, dst_path,
3080 path->nodes[0], path->slots[0],
3085 last_offset = tmp.offset;
3090 btrfs_release_path(path);
3091 btrfs_release_path(dst_path);
3094 *last_offset_ret = last_offset;
3096 * insert the log range keys to indicate where the log
3099 ret = insert_dir_log_key(trans, log, path, key_type,
3100 ino, first_offset, last_offset);
3108 * logging directories is very similar to logging inodes, We find all the items
3109 * from the current transaction and write them to the log.
3111 * The recovery code scans the directory in the subvolume, and if it finds a
3112 * key in the range logged that is not present in the log tree, then it means
3113 * that dir entry was unlinked during the transaction.
3115 * In order for that scan to work, we must include one key smaller than
3116 * the smallest logged by this transaction and one key larger than the largest
3117 * key logged by this transaction.
3119 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3120 struct btrfs_root *root, struct inode *inode,
3121 struct btrfs_path *path,
3122 struct btrfs_path *dst_path)
3127 int key_type = BTRFS_DIR_ITEM_KEY;
3133 ret = log_dir_items(trans, root, inode, path,
3134 dst_path, key_type, min_key,
3138 if (max_key == (u64)-1)
3140 min_key = max_key + 1;
3143 if (key_type == BTRFS_DIR_ITEM_KEY) {
3144 key_type = BTRFS_DIR_INDEX_KEY;
3151 * a helper function to drop items from the log before we relog an
3152 * inode. max_key_type indicates the highest item type to remove.
3153 * This cannot be run for file data extents because it does not
3154 * free the extents they point to.
3156 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3157 struct btrfs_root *log,
3158 struct btrfs_path *path,
3159 u64 objectid, int max_key_type)
3162 struct btrfs_key key;
3163 struct btrfs_key found_key;
3166 key.objectid = objectid;
3167 key.type = max_key_type;
3168 key.offset = (u64)-1;
3171 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3172 BUG_ON(ret == 0); /* Logic error */
3176 if (path->slots[0] == 0)
3180 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3183 if (found_key.objectid != objectid)
3186 found_key.offset = 0;
3188 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3191 ret = btrfs_del_items(trans, log, path, start_slot,
3192 path->slots[0] - start_slot + 1);
3194 * If start slot isn't 0 then we don't need to re-search, we've
3195 * found the last guy with the objectid in this tree.
3197 if (ret || start_slot != 0)
3199 btrfs_release_path(path);
3201 btrfs_release_path(path);
3207 static void fill_inode_item(struct btrfs_trans_handle *trans,
3208 struct extent_buffer *leaf,
3209 struct btrfs_inode_item *item,
3210 struct inode *inode, int log_inode_only)
3212 struct btrfs_map_token token;
3214 btrfs_init_map_token(&token);
3216 if (log_inode_only) {
3217 /* set the generation to zero so the recover code
3218 * can tell the difference between an logging
3219 * just to say 'this inode exists' and a logging
3220 * to say 'update this inode with these values'
3222 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3223 btrfs_set_token_inode_size(leaf, item, 0, &token);
3225 btrfs_set_token_inode_generation(leaf, item,
3226 BTRFS_I(inode)->generation,
3228 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3231 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3232 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3233 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3234 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3236 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3237 inode->i_atime.tv_sec, &token);
3238 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3239 inode->i_atime.tv_nsec, &token);
3241 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3242 inode->i_mtime.tv_sec, &token);
3243 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3244 inode->i_mtime.tv_nsec, &token);
3246 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3247 inode->i_ctime.tv_sec, &token);
3248 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3249 inode->i_ctime.tv_nsec, &token);
3251 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3254 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3255 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3256 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3257 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3258 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3261 static int log_inode_item(struct btrfs_trans_handle *trans,
3262 struct btrfs_root *log, struct btrfs_path *path,
3263 struct inode *inode)
3265 struct btrfs_inode_item *inode_item;
3268 ret = btrfs_insert_empty_item(trans, log, path,
3269 &BTRFS_I(inode)->location,
3270 sizeof(*inode_item));
3271 if (ret && ret != -EEXIST)
3273 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3274 struct btrfs_inode_item);
3275 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0);
3276 btrfs_release_path(path);
3280 static noinline int copy_items(struct btrfs_trans_handle *trans,
3281 struct inode *inode,
3282 struct btrfs_path *dst_path,
3283 struct btrfs_path *src_path, u64 *last_extent,
3284 int start_slot, int nr, int inode_only)
3286 unsigned long src_offset;
3287 unsigned long dst_offset;
3288 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3289 struct btrfs_file_extent_item *extent;
3290 struct btrfs_inode_item *inode_item;
3291 struct extent_buffer *src = src_path->nodes[0];
3292 struct btrfs_key first_key, last_key, key;
3294 struct btrfs_key *ins_keys;
3298 struct list_head ordered_sums;
3299 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3300 bool has_extents = false;
3301 bool need_find_last_extent = true;
3304 INIT_LIST_HEAD(&ordered_sums);
3306 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3307 nr * sizeof(u32), GFP_NOFS);
3311 first_key.objectid = (u64)-1;
3313 ins_sizes = (u32 *)ins_data;
3314 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3316 for (i = 0; i < nr; i++) {
3317 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3318 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3320 ret = btrfs_insert_empty_items(trans, log, dst_path,
3321 ins_keys, ins_sizes, nr);
3327 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3328 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3329 dst_path->slots[0]);
3331 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3333 if ((i == (nr - 1)))
3334 last_key = ins_keys[i];
3336 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3337 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3339 struct btrfs_inode_item);
3340 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3341 inode, inode_only == LOG_INODE_EXISTS);
3343 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3344 src_offset, ins_sizes[i]);
3348 * We set need_find_last_extent here in case we know we were
3349 * processing other items and then walk into the first extent in
3350 * the inode. If we don't hit an extent then nothing changes,
3351 * we'll do the last search the next time around.
3353 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3355 if (first_key.objectid == (u64)-1)
3356 first_key = ins_keys[i];
3358 need_find_last_extent = false;
3361 /* take a reference on file data extents so that truncates
3362 * or deletes of this inode don't have to relog the inode
3365 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY &&
3368 extent = btrfs_item_ptr(src, start_slot + i,
3369 struct btrfs_file_extent_item);
3371 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3374 found_type = btrfs_file_extent_type(src, extent);
3375 if (found_type == BTRFS_FILE_EXTENT_REG) {
3377 ds = btrfs_file_extent_disk_bytenr(src,
3379 /* ds == 0 is a hole */
3383 dl = btrfs_file_extent_disk_num_bytes(src,
3385 cs = btrfs_file_extent_offset(src, extent);
3386 cl = btrfs_file_extent_num_bytes(src,
3388 if (btrfs_file_extent_compression(src,
3394 ret = btrfs_lookup_csums_range(
3395 log->fs_info->csum_root,
3396 ds + cs, ds + cs + cl - 1,
3399 btrfs_release_path(dst_path);
3407 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3408 btrfs_release_path(dst_path);
3412 * we have to do this after the loop above to avoid changing the
3413 * log tree while trying to change the log tree.
3416 while (!list_empty(&ordered_sums)) {
3417 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3418 struct btrfs_ordered_sum,
3421 ret = btrfs_csum_file_blocks(trans, log, sums);
3422 list_del(&sums->list);
3429 if (need_find_last_extent && *last_extent == first_key.offset) {
3431 * We don't have any leafs between our current one and the one
3432 * we processed before that can have file extent items for our
3433 * inode (and have a generation number smaller than our current
3436 need_find_last_extent = false;
3440 * Because we use btrfs_search_forward we could skip leaves that were
3441 * not modified and then assume *last_extent is valid when it really
3442 * isn't. So back up to the previous leaf and read the end of the last
3443 * extent before we go and fill in holes.
3445 if (need_find_last_extent) {
3448 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3453 if (src_path->slots[0])
3454 src_path->slots[0]--;
3455 src = src_path->nodes[0];
3456 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3457 if (key.objectid != btrfs_ino(inode) ||
3458 key.type != BTRFS_EXTENT_DATA_KEY)
3460 extent = btrfs_item_ptr(src, src_path->slots[0],
3461 struct btrfs_file_extent_item);
3462 if (btrfs_file_extent_type(src, extent) ==
3463 BTRFS_FILE_EXTENT_INLINE) {
3464 len = btrfs_file_extent_inline_len(src,
3467 *last_extent = ALIGN(key.offset + len,
3470 len = btrfs_file_extent_num_bytes(src, extent);
3471 *last_extent = key.offset + len;
3475 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3476 * things could have happened
3478 * 1) A merge could have happened, so we could currently be on a leaf
3479 * that holds what we were copying in the first place.
3480 * 2) A split could have happened, and now not all of the items we want
3481 * are on the same leaf.
3483 * So we need to adjust how we search for holes, we need to drop the
3484 * path and re-search for the first extent key we found, and then walk
3485 * forward until we hit the last one we copied.
3487 if (need_find_last_extent) {
3488 /* btrfs_prev_leaf could return 1 without releasing the path */
3489 btrfs_release_path(src_path);
3490 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3495 src = src_path->nodes[0];
3496 i = src_path->slots[0];
3502 * Ok so here we need to go through and fill in any holes we may have
3503 * to make sure that holes are punched for those areas in case they had
3504 * extents previously.
3510 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3511 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3515 src = src_path->nodes[0];
3519 btrfs_item_key_to_cpu(src, &key, i);
3520 if (!btrfs_comp_cpu_keys(&key, &last_key))
3522 if (key.objectid != btrfs_ino(inode) ||
3523 key.type != BTRFS_EXTENT_DATA_KEY) {
3527 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3528 if (btrfs_file_extent_type(src, extent) ==
3529 BTRFS_FILE_EXTENT_INLINE) {
3530 len = btrfs_file_extent_inline_len(src, i, extent);
3531 extent_end = ALIGN(key.offset + len, log->sectorsize);
3533 len = btrfs_file_extent_num_bytes(src, extent);
3534 extent_end = key.offset + len;
3538 if (*last_extent == key.offset) {
3539 *last_extent = extent_end;
3542 offset = *last_extent;
3543 len = key.offset - *last_extent;
3544 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3545 offset, 0, 0, len, 0, len, 0,
3549 *last_extent = extent_end;
3552 * Need to let the callers know we dropped the path so they should
3555 if (!ret && need_find_last_extent)
3560 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3562 struct extent_map *em1, *em2;
3564 em1 = list_entry(a, struct extent_map, list);
3565 em2 = list_entry(b, struct extent_map, list);
3567 if (em1->start < em2->start)
3569 else if (em1->start > em2->start)
3574 static int log_one_extent(struct btrfs_trans_handle *trans,
3575 struct inode *inode, struct btrfs_root *root,
3576 struct extent_map *em, struct btrfs_path *path,
3577 struct list_head *logged_list)
3579 struct btrfs_root *log = root->log_root;
3580 struct btrfs_file_extent_item *fi;
3581 struct extent_buffer *leaf;
3582 struct btrfs_ordered_extent *ordered;
3583 struct list_head ordered_sums;
3584 struct btrfs_map_token token;
3585 struct btrfs_key key;
3586 u64 mod_start = em->mod_start;
3587 u64 mod_len = em->mod_len;
3590 u64 extent_offset = em->start - em->orig_start;
3593 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3594 int extent_inserted = 0;
3596 INIT_LIST_HEAD(&ordered_sums);
3597 btrfs_init_map_token(&token);
3599 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
3600 em->start + em->len, NULL, 0, 1,
3601 sizeof(*fi), &extent_inserted);
3605 if (!extent_inserted) {
3606 key.objectid = btrfs_ino(inode);
3607 key.type = BTRFS_EXTENT_DATA_KEY;
3608 key.offset = em->start;
3610 ret = btrfs_insert_empty_item(trans, log, path, &key,
3615 leaf = path->nodes[0];
3616 fi = btrfs_item_ptr(leaf, path->slots[0],
3617 struct btrfs_file_extent_item);
3619 btrfs_set_token_file_extent_generation(leaf, fi, em->generation,
3621 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3623 btrfs_set_token_file_extent_type(leaf, fi,
3624 BTRFS_FILE_EXTENT_PREALLOC,
3627 btrfs_set_token_file_extent_type(leaf, fi,
3628 BTRFS_FILE_EXTENT_REG,
3630 if (em->block_start == EXTENT_MAP_HOLE)
3634 block_len = max(em->block_len, em->orig_block_len);
3635 if (em->compress_type != BTRFS_COMPRESS_NONE) {
3636 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3639 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3641 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
3642 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3644 extent_offset, &token);
3645 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3648 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
3649 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
3653 btrfs_set_token_file_extent_offset(leaf, fi,
3654 em->start - em->orig_start,
3656 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
3657 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
3658 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
3660 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
3661 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
3662 btrfs_mark_buffer_dirty(leaf);
3664 btrfs_release_path(path);
3673 * First check and see if our csums are on our outstanding ordered
3676 list_for_each_entry(ordered, logged_list, log_list) {
3677 struct btrfs_ordered_sum *sum;
3682 if (ordered->file_offset + ordered->len <= mod_start ||
3683 mod_start + mod_len <= ordered->file_offset)
3687 * We are going to copy all the csums on this ordered extent, so
3688 * go ahead and adjust mod_start and mod_len in case this
3689 * ordered extent has already been logged.
3691 if (ordered->file_offset > mod_start) {
3692 if (ordered->file_offset + ordered->len >=
3693 mod_start + mod_len)
3694 mod_len = ordered->file_offset - mod_start;
3696 * If we have this case
3698 * |--------- logged extent ---------|
3699 * |----- ordered extent ----|
3701 * Just don't mess with mod_start and mod_len, we'll
3702 * just end up logging more csums than we need and it
3706 if (ordered->file_offset + ordered->len <
3707 mod_start + mod_len) {
3708 mod_len = (mod_start + mod_len) -
3709 (ordered->file_offset + ordered->len);
3710 mod_start = ordered->file_offset +
3718 * To keep us from looping for the above case of an ordered
3719 * extent that falls inside of the logged extent.
3721 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3725 if (ordered->csum_bytes_left) {
3726 btrfs_start_ordered_extent(inode, ordered, 0);
3727 wait_event(ordered->wait,
3728 ordered->csum_bytes_left == 0);
3731 list_for_each_entry(sum, &ordered->list, list) {
3732 ret = btrfs_csum_file_blocks(trans, log, sum);
3740 if (!mod_len || ret)
3743 if (em->compress_type) {
3745 csum_len = block_len;
3747 csum_offset = mod_start - em->start;
3751 /* block start is already adjusted for the file extent offset. */
3752 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
3753 em->block_start + csum_offset,
3754 em->block_start + csum_offset +
3755 csum_len - 1, &ordered_sums, 0);
3759 while (!list_empty(&ordered_sums)) {
3760 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3761 struct btrfs_ordered_sum,
3764 ret = btrfs_csum_file_blocks(trans, log, sums);
3765 list_del(&sums->list);
3772 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
3773 struct btrfs_root *root,
3774 struct inode *inode,
3775 struct btrfs_path *path,
3776 struct list_head *logged_list)
3778 struct extent_map *em, *n;
3779 struct list_head extents;
3780 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3785 INIT_LIST_HEAD(&extents);
3787 write_lock(&tree->lock);
3788 test_gen = root->fs_info->last_trans_committed;
3790 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
3791 list_del_init(&em->list);
3794 * Just an arbitrary number, this can be really CPU intensive
3795 * once we start getting a lot of extents, and really once we
3796 * have a bunch of extents we just want to commit since it will
3799 if (++num > 32768) {
3800 list_del_init(&tree->modified_extents);
3805 if (em->generation <= test_gen)
3807 /* Need a ref to keep it from getting evicted from cache */
3808 atomic_inc(&em->refs);
3809 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
3810 list_add_tail(&em->list, &extents);
3814 list_sort(NULL, &extents, extent_cmp);
3817 while (!list_empty(&extents)) {
3818 em = list_entry(extents.next, struct extent_map, list);
3820 list_del_init(&em->list);
3823 * If we had an error we just need to delete everybody from our
3827 clear_em_logging(tree, em);
3828 free_extent_map(em);
3832 write_unlock(&tree->lock);
3834 ret = log_one_extent(trans, inode, root, em, path, logged_list);
3835 write_lock(&tree->lock);
3836 clear_em_logging(tree, em);
3837 free_extent_map(em);
3839 WARN_ON(!list_empty(&extents));
3840 write_unlock(&tree->lock);
3842 btrfs_release_path(path);
3846 /* log a single inode in the tree log.
3847 * At least one parent directory for this inode must exist in the tree
3848 * or be logged already.
3850 * Any items from this inode changed by the current transaction are copied
3851 * to the log tree. An extra reference is taken on any extents in this
3852 * file, allowing us to avoid a whole pile of corner cases around logging
3853 * blocks that have been removed from the tree.
3855 * See LOG_INODE_ALL and related defines for a description of what inode_only
3858 * This handles both files and directories.
3860 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
3861 struct btrfs_root *root, struct inode *inode,
3864 struct btrfs_path *path;
3865 struct btrfs_path *dst_path;
3866 struct btrfs_key min_key;
3867 struct btrfs_key max_key;
3868 struct btrfs_root *log = root->log_root;
3869 struct extent_buffer *src = NULL;
3870 LIST_HEAD(logged_list);
3871 u64 last_extent = 0;
3875 int ins_start_slot = 0;
3877 bool fast_search = false;
3878 u64 ino = btrfs_ino(inode);
3880 path = btrfs_alloc_path();
3883 dst_path = btrfs_alloc_path();
3885 btrfs_free_path(path);
3889 min_key.objectid = ino;
3890 min_key.type = BTRFS_INODE_ITEM_KEY;
3893 max_key.objectid = ino;
3896 /* today the code can only do partial logging of directories */
3897 if (S_ISDIR(inode->i_mode) ||
3898 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3899 &BTRFS_I(inode)->runtime_flags) &&
3900 inode_only == LOG_INODE_EXISTS))
3901 max_key.type = BTRFS_XATTR_ITEM_KEY;
3903 max_key.type = (u8)-1;
3904 max_key.offset = (u64)-1;
3906 /* Only run delayed items if we are a dir or a new file */
3907 if (S_ISDIR(inode->i_mode) ||
3908 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) {
3909 ret = btrfs_commit_inode_delayed_items(trans, inode);
3911 btrfs_free_path(path);
3912 btrfs_free_path(dst_path);
3917 mutex_lock(&BTRFS_I(inode)->log_mutex);
3919 btrfs_get_logged_extents(inode, &logged_list);
3922 * a brute force approach to making sure we get the most uptodate
3923 * copies of everything.
3925 if (S_ISDIR(inode->i_mode)) {
3926 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
3928 if (inode_only == LOG_INODE_EXISTS)
3929 max_key_type = BTRFS_XATTR_ITEM_KEY;
3930 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
3932 if (test_and_clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3933 &BTRFS_I(inode)->runtime_flags)) {
3934 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3935 &BTRFS_I(inode)->runtime_flags);
3936 ret = btrfs_truncate_inode_items(trans, log,
3938 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3939 &BTRFS_I(inode)->runtime_flags) ||
3940 inode_only == LOG_INODE_EXISTS) {
3941 if (inode_only == LOG_INODE_ALL)
3943 max_key.type = BTRFS_XATTR_ITEM_KEY;
3944 ret = drop_objectid_items(trans, log, path, ino,
3947 if (inode_only == LOG_INODE_ALL)
3949 ret = log_inode_item(trans, log, dst_path, inode);
3962 path->keep_locks = 1;
3966 ret = btrfs_search_forward(root, &min_key,
3967 path, trans->transid);
3971 /* note, ins_nr might be > 0 here, cleanup outside the loop */
3972 if (min_key.objectid != ino)
3974 if (min_key.type > max_key.type)
3977 src = path->nodes[0];
3978 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
3981 } else if (!ins_nr) {
3982 ins_start_slot = path->slots[0];
3987 ret = copy_items(trans, inode, dst_path, path, &last_extent,
3988 ins_start_slot, ins_nr, inode_only);
3994 btrfs_release_path(path);
3998 ins_start_slot = path->slots[0];
4001 nritems = btrfs_header_nritems(path->nodes[0]);
4003 if (path->slots[0] < nritems) {
4004 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4009 ret = copy_items(trans, inode, dst_path, path,
4010 &last_extent, ins_start_slot,
4011 ins_nr, inode_only);
4019 btrfs_release_path(path);
4021 if (min_key.offset < (u64)-1) {
4023 } else if (min_key.type < max_key.type) {
4031 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4032 ins_start_slot, ins_nr, inode_only);
4042 btrfs_release_path(path);
4043 btrfs_release_path(dst_path);
4045 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
4051 } else if (inode_only == LOG_INODE_ALL) {
4052 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
4053 struct extent_map *em, *n;
4055 write_lock(&tree->lock);
4056 list_for_each_entry_safe(em, n, &tree->modified_extents, list)
4057 list_del_init(&em->list);
4058 write_unlock(&tree->lock);
4061 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
4062 ret = log_directory_changes(trans, root, inode, path, dst_path);
4068 BTRFS_I(inode)->logged_trans = trans->transid;
4069 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
4072 btrfs_put_logged_extents(&logged_list);
4074 btrfs_submit_logged_extents(&logged_list, log);
4075 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4077 btrfs_free_path(path);
4078 btrfs_free_path(dst_path);
4083 * follow the dentry parent pointers up the chain and see if any
4084 * of the directories in it require a full commit before they can
4085 * be logged. Returns zero if nothing special needs to be done or 1 if
4086 * a full commit is required.
4088 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
4089 struct inode *inode,
4090 struct dentry *parent,
4091 struct super_block *sb,
4095 struct btrfs_root *root;
4096 struct dentry *old_parent = NULL;
4097 struct inode *orig_inode = inode;
4100 * for regular files, if its inode is already on disk, we don't
4101 * have to worry about the parents at all. This is because
4102 * we can use the last_unlink_trans field to record renames
4103 * and other fun in this file.
4105 if (S_ISREG(inode->i_mode) &&
4106 BTRFS_I(inode)->generation <= last_committed &&
4107 BTRFS_I(inode)->last_unlink_trans <= last_committed)
4110 if (!S_ISDIR(inode->i_mode)) {
4111 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
4113 inode = parent->d_inode;
4118 * If we are logging a directory then we start with our inode,
4119 * not our parents inode, so we need to skipp setting the
4120 * logged_trans so that further down in the log code we don't
4121 * think this inode has already been logged.
4123 if (inode != orig_inode)
4124 BTRFS_I(inode)->logged_trans = trans->transid;
4127 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
4128 root = BTRFS_I(inode)->root;
4131 * make sure any commits to the log are forced
4132 * to be full commits
4134 btrfs_set_log_full_commit(root->fs_info, trans);
4139 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
4142 if (IS_ROOT(parent))
4145 parent = dget_parent(parent);
4147 old_parent = parent;
4148 inode = parent->d_inode;
4157 * helper function around btrfs_log_inode to make sure newly created
4158 * parent directories also end up in the log. A minimal inode and backref
4159 * only logging is done of any parent directories that are older than
4160 * the last committed transaction
4162 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
4163 struct btrfs_root *root, struct inode *inode,
4164 struct dentry *parent, int exists_only,
4165 struct btrfs_log_ctx *ctx)
4167 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
4168 struct super_block *sb;
4169 struct dentry *old_parent = NULL;
4171 u64 last_committed = root->fs_info->last_trans_committed;
4175 if (btrfs_test_opt(root, NOTREELOG)) {
4181 * The prev transaction commit doesn't complete, we need do
4182 * full commit by ourselves.
4184 if (root->fs_info->last_trans_log_full_commit >
4185 root->fs_info->last_trans_committed) {
4190 if (root != BTRFS_I(inode)->root ||
4191 btrfs_root_refs(&root->root_item) == 0) {
4196 ret = check_parent_dirs_for_sync(trans, inode, parent,
4197 sb, last_committed);
4201 if (btrfs_inode_in_log(inode, trans->transid)) {
4202 ret = BTRFS_NO_LOG_SYNC;
4206 ret = start_log_trans(trans, root, ctx);
4210 ret = btrfs_log_inode(trans, root, inode, inode_only);
4215 * for regular files, if its inode is already on disk, we don't
4216 * have to worry about the parents at all. This is because
4217 * we can use the last_unlink_trans field to record renames
4218 * and other fun in this file.
4220 if (S_ISREG(inode->i_mode) &&
4221 BTRFS_I(inode)->generation <= last_committed &&
4222 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
4227 inode_only = LOG_INODE_EXISTS;
4229 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
4232 inode = parent->d_inode;
4233 if (root != BTRFS_I(inode)->root)
4236 if (BTRFS_I(inode)->generation >
4237 root->fs_info->last_trans_committed) {
4238 ret = btrfs_log_inode(trans, root, inode, inode_only);
4242 if (IS_ROOT(parent))
4245 parent = dget_parent(parent);
4247 old_parent = parent;
4253 btrfs_set_log_full_commit(root->fs_info, trans);
4258 btrfs_remove_log_ctx(root, ctx);
4259 btrfs_end_log_trans(root);
4265 * it is not safe to log dentry if the chunk root has added new
4266 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
4267 * If this returns 1, you must commit the transaction to safely get your
4270 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
4271 struct btrfs_root *root, struct dentry *dentry,
4272 struct btrfs_log_ctx *ctx)
4274 struct dentry *parent = dget_parent(dentry);
4277 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent,
4285 * should be called during mount to recover any replay any log trees
4288 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
4291 struct btrfs_path *path;
4292 struct btrfs_trans_handle *trans;
4293 struct btrfs_key key;
4294 struct btrfs_key found_key;
4295 struct btrfs_key tmp_key;
4296 struct btrfs_root *log;
4297 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
4298 struct walk_control wc = {
4299 .process_func = process_one_buffer,
4303 path = btrfs_alloc_path();
4307 fs_info->log_root_recovering = 1;
4309 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4310 if (IS_ERR(trans)) {
4311 ret = PTR_ERR(trans);
4318 ret = walk_log_tree(trans, log_root_tree, &wc);
4320 btrfs_error(fs_info, ret, "Failed to pin buffers while "
4321 "recovering log root tree.");
4326 key.objectid = BTRFS_TREE_LOG_OBJECTID;
4327 key.offset = (u64)-1;
4328 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
4331 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
4334 btrfs_error(fs_info, ret,
4335 "Couldn't find tree log root.");
4339 if (path->slots[0] == 0)
4343 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4345 btrfs_release_path(path);
4346 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4349 log = btrfs_read_fs_root(log_root_tree, &found_key);
4352 btrfs_error(fs_info, ret,
4353 "Couldn't read tree log root.");
4357 tmp_key.objectid = found_key.offset;
4358 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
4359 tmp_key.offset = (u64)-1;
4361 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
4362 if (IS_ERR(wc.replay_dest)) {
4363 ret = PTR_ERR(wc.replay_dest);
4364 free_extent_buffer(log->node);
4365 free_extent_buffer(log->commit_root);
4367 btrfs_error(fs_info, ret, "Couldn't read target root "
4368 "for tree log recovery.");
4372 wc.replay_dest->log_root = log;
4373 btrfs_record_root_in_trans(trans, wc.replay_dest);
4374 ret = walk_log_tree(trans, log, &wc);
4376 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
4377 ret = fixup_inode_link_counts(trans, wc.replay_dest,
4381 key.offset = found_key.offset - 1;
4382 wc.replay_dest->log_root = NULL;
4383 free_extent_buffer(log->node);
4384 free_extent_buffer(log->commit_root);
4390 if (found_key.offset == 0)
4393 btrfs_release_path(path);
4395 /* step one is to pin it all, step two is to replay just inodes */
4398 wc.process_func = replay_one_buffer;
4399 wc.stage = LOG_WALK_REPLAY_INODES;
4402 /* step three is to replay everything */
4403 if (wc.stage < LOG_WALK_REPLAY_ALL) {
4408 btrfs_free_path(path);
4410 /* step 4: commit the transaction, which also unpins the blocks */
4411 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
4415 free_extent_buffer(log_root_tree->node);
4416 log_root_tree->log_root = NULL;
4417 fs_info->log_root_recovering = 0;
4418 kfree(log_root_tree);
4423 btrfs_end_transaction(wc.trans, fs_info->tree_root);
4424 btrfs_free_path(path);
4429 * there are some corner cases where we want to force a full
4430 * commit instead of allowing a directory to be logged.
4432 * They revolve around files there were unlinked from the directory, and
4433 * this function updates the parent directory so that a full commit is
4434 * properly done if it is fsync'd later after the unlinks are done.
4436 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
4437 struct inode *dir, struct inode *inode,
4441 * when we're logging a file, if it hasn't been renamed
4442 * or unlinked, and its inode is fully committed on disk,
4443 * we don't have to worry about walking up the directory chain
4444 * to log its parents.
4446 * So, we use the last_unlink_trans field to put this transid
4447 * into the file. When the file is logged we check it and
4448 * don't log the parents if the file is fully on disk.
4450 if (S_ISREG(inode->i_mode))
4451 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4454 * if this directory was already logged any new
4455 * names for this file/dir will get recorded
4458 if (BTRFS_I(dir)->logged_trans == trans->transid)
4462 * if the inode we're about to unlink was logged,
4463 * the log will be properly updated for any new names
4465 if (BTRFS_I(inode)->logged_trans == trans->transid)
4469 * when renaming files across directories, if the directory
4470 * there we're unlinking from gets fsync'd later on, there's
4471 * no way to find the destination directory later and fsync it
4472 * properly. So, we have to be conservative and force commits
4473 * so the new name gets discovered.
4478 /* we can safely do the unlink without any special recording */
4482 BTRFS_I(dir)->last_unlink_trans = trans->transid;
4486 * Call this after adding a new name for a file and it will properly
4487 * update the log to reflect the new name.
4489 * It will return zero if all goes well, and it will return 1 if a
4490 * full transaction commit is required.
4492 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
4493 struct inode *inode, struct inode *old_dir,
4494 struct dentry *parent)
4496 struct btrfs_root * root = BTRFS_I(inode)->root;
4499 * this will force the logging code to walk the dentry chain
4502 if (S_ISREG(inode->i_mode))
4503 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4506 * if this inode hasn't been logged and directory we're renaming it
4507 * from hasn't been logged, we don't need to log it
4509 if (BTRFS_I(inode)->logged_trans <=
4510 root->fs_info->last_trans_committed &&
4511 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
4512 root->fs_info->last_trans_committed))
4515 return btrfs_log_inode_parent(trans, root, inode, parent, 1, NULL);