2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/list_sort.h>
23 #include "transaction.h"
26 #include "print-tree.h"
32 /* magic values for the inode_only field in btrfs_log_inode:
34 * LOG_INODE_ALL means to log everything
35 * LOG_INODE_EXISTS means to log just enough to recreate the inode
38 #define LOG_INODE_ALL 0
39 #define LOG_INODE_EXISTS 1
42 * directory trouble cases
44 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
45 * log, we must force a full commit before doing an fsync of the directory
46 * where the unlink was done.
47 * ---> record transid of last unlink/rename per directory
51 * rename foo/some_dir foo2/some_dir
53 * fsync foo/some_dir/some_file
55 * The fsync above will unlink the original some_dir without recording
56 * it in its new location (foo2). After a crash, some_dir will be gone
57 * unless the fsync of some_file forces a full commit
59 * 2) we must log any new names for any file or dir that is in the fsync
60 * log. ---> check inode while renaming/linking.
62 * 2a) we must log any new names for any file or dir during rename
63 * when the directory they are being removed from was logged.
64 * ---> check inode and old parent dir during rename
66 * 2a is actually the more important variant. With the extra logging
67 * a crash might unlink the old name without recreating the new one
69 * 3) after a crash, we must go through any directories with a link count
70 * of zero and redo the rm -rf
77 * The directory f1 was fully removed from the FS, but fsync was never
78 * called on f1, only its parent dir. After a crash the rm -rf must
79 * be replayed. This must be able to recurse down the entire
80 * directory tree. The inode link count fixup code takes care of the
85 * stages for the tree walking. The first
86 * stage (0) is to only pin down the blocks we find
87 * the second stage (1) is to make sure that all the inodes
88 * we find in the log are created in the subvolume.
90 * The last stage is to deal with directories and links and extents
91 * and all the other fun semantics
93 #define LOG_WALK_PIN_ONLY 0
94 #define LOG_WALK_REPLAY_INODES 1
95 #define LOG_WALK_REPLAY_ALL 2
97 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode,
100 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root,
102 struct btrfs_path *path, u64 objectid);
103 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
104 struct btrfs_root *root,
105 struct btrfs_root *log,
106 struct btrfs_path *path,
107 u64 dirid, int del_all);
110 * tree logging is a special write ahead log used to make sure that
111 * fsyncs and O_SYNCs can happen without doing full tree commits.
113 * Full tree commits are expensive because they require commonly
114 * modified blocks to be recowed, creating many dirty pages in the
115 * extent tree an 4x-6x higher write load than ext3.
117 * Instead of doing a tree commit on every fsync, we use the
118 * key ranges and transaction ids to find items for a given file or directory
119 * that have changed in this transaction. Those items are copied into
120 * a special tree (one per subvolume root), that tree is written to disk
121 * and then the fsync is considered complete.
123 * After a crash, items are copied out of the log-tree back into the
124 * subvolume tree. Any file data extents found are recorded in the extent
125 * allocation tree, and the log-tree freed.
127 * The log tree is read three times, once to pin down all the extents it is
128 * using in ram and once, once to create all the inodes logged in the tree
129 * and once to do all the other items.
133 * start a sub transaction and setup the log tree
134 * this increments the log tree writer count to make the people
135 * syncing the tree wait for us to finish
137 static int start_log_trans(struct btrfs_trans_handle *trans,
138 struct btrfs_root *root)
143 mutex_lock(&root->log_mutex);
144 if (root->log_root) {
145 if (!root->log_start_pid) {
146 root->log_start_pid = current->pid;
147 root->log_multiple_pids = false;
148 } else if (root->log_start_pid != current->pid) {
149 root->log_multiple_pids = true;
152 atomic_inc(&root->log_batch);
153 atomic_inc(&root->log_writers);
154 mutex_unlock(&root->log_mutex);
157 root->log_multiple_pids = false;
158 root->log_start_pid = current->pid;
159 mutex_lock(&root->fs_info->tree_log_mutex);
160 if (!root->fs_info->log_root_tree) {
161 ret = btrfs_init_log_root_tree(trans, root->fs_info);
165 if (err == 0 && !root->log_root) {
166 ret = btrfs_add_log_tree(trans, root);
170 mutex_unlock(&root->fs_info->tree_log_mutex);
171 atomic_inc(&root->log_batch);
172 atomic_inc(&root->log_writers);
173 mutex_unlock(&root->log_mutex);
178 * returns 0 if there was a log transaction running and we were able
179 * to join, or returns -ENOENT if there were not transactions
182 static int join_running_log_trans(struct btrfs_root *root)
190 mutex_lock(&root->log_mutex);
191 if (root->log_root) {
193 atomic_inc(&root->log_writers);
195 mutex_unlock(&root->log_mutex);
200 * This either makes the current running log transaction wait
201 * until you call btrfs_end_log_trans() or it makes any future
202 * log transactions wait until you call btrfs_end_log_trans()
204 int btrfs_pin_log_trans(struct btrfs_root *root)
208 mutex_lock(&root->log_mutex);
209 atomic_inc(&root->log_writers);
210 mutex_unlock(&root->log_mutex);
215 * indicate we're done making changes to the log tree
216 * and wake up anyone waiting to do a sync
218 void btrfs_end_log_trans(struct btrfs_root *root)
220 if (atomic_dec_and_test(&root->log_writers)) {
222 if (waitqueue_active(&root->log_writer_wait))
223 wake_up(&root->log_writer_wait);
229 * the walk control struct is used to pass state down the chain when
230 * processing the log tree. The stage field tells us which part
231 * of the log tree processing we are currently doing. The others
232 * are state fields used for that specific part
234 struct walk_control {
235 /* should we free the extent on disk when done? This is used
236 * at transaction commit time while freeing a log tree
240 /* should we write out the extent buffer? This is used
241 * while flushing the log tree to disk during a sync
245 /* should we wait for the extent buffer io to finish? Also used
246 * while flushing the log tree to disk for a sync
250 /* pin only walk, we record which extents on disk belong to the
255 /* what stage of the replay code we're currently in */
258 /* the root we are currently replaying */
259 struct btrfs_root *replay_dest;
261 /* the trans handle for the current replay */
262 struct btrfs_trans_handle *trans;
264 /* the function that gets used to process blocks we find in the
265 * tree. Note the extent_buffer might not be up to date when it is
266 * passed in, and it must be checked or read if you need the data
269 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
270 struct walk_control *wc, u64 gen);
274 * process_func used to pin down extents, write them or wait on them
276 static int process_one_buffer(struct btrfs_root *log,
277 struct extent_buffer *eb,
278 struct walk_control *wc, u64 gen)
283 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
286 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
288 btrfs_write_tree_block(eb);
290 btrfs_wait_tree_block_writeback(eb);
296 * Item overwrite used by replay and tree logging. eb, slot and key all refer
297 * to the src data we are copying out.
299 * root is the tree we are copying into, and path is a scratch
300 * path for use in this function (it should be released on entry and
301 * will be released on exit).
303 * If the key is already in the destination tree the existing item is
304 * overwritten. If the existing item isn't big enough, it is extended.
305 * If it is too large, it is truncated.
307 * If the key isn't in the destination yet, a new item is inserted.
309 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
310 struct btrfs_root *root,
311 struct btrfs_path *path,
312 struct extent_buffer *eb, int slot,
313 struct btrfs_key *key)
317 u64 saved_i_size = 0;
318 int save_old_i_size = 0;
319 unsigned long src_ptr;
320 unsigned long dst_ptr;
321 int overwrite_root = 0;
322 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
324 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
327 item_size = btrfs_item_size_nr(eb, slot);
328 src_ptr = btrfs_item_ptr_offset(eb, slot);
330 /* look for the key in the destination tree */
331 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
338 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
340 if (dst_size != item_size)
343 if (item_size == 0) {
344 btrfs_release_path(path);
347 dst_copy = kmalloc(item_size, GFP_NOFS);
348 src_copy = kmalloc(item_size, GFP_NOFS);
349 if (!dst_copy || !src_copy) {
350 btrfs_release_path(path);
356 read_extent_buffer(eb, src_copy, src_ptr, item_size);
358 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
359 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
361 ret = memcmp(dst_copy, src_copy, item_size);
366 * they have the same contents, just return, this saves
367 * us from cowing blocks in the destination tree and doing
368 * extra writes that may not have been done by a previous
372 btrfs_release_path(path);
377 * We need to load the old nbytes into the inode so when we
378 * replay the extents we've logged we get the right nbytes.
381 struct btrfs_inode_item *item;
384 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
385 struct btrfs_inode_item);
386 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
387 item = btrfs_item_ptr(eb, slot,
388 struct btrfs_inode_item);
389 btrfs_set_inode_nbytes(eb, item, nbytes);
391 } else if (inode_item) {
392 struct btrfs_inode_item *item;
395 * New inode, set nbytes to 0 so that the nbytes comes out
396 * properly when we replay the extents.
398 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
399 btrfs_set_inode_nbytes(eb, item, 0);
402 btrfs_release_path(path);
403 /* try to insert the key into the destination tree */
404 ret = btrfs_insert_empty_item(trans, root, path,
407 /* make sure any existing item is the correct size */
408 if (ret == -EEXIST) {
410 found_size = btrfs_item_size_nr(path->nodes[0],
412 if (found_size > item_size)
413 btrfs_truncate_item(root, path, item_size, 1);
414 else if (found_size < item_size)
415 btrfs_extend_item(root, path,
416 item_size - found_size);
420 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
423 /* don't overwrite an existing inode if the generation number
424 * was logged as zero. This is done when the tree logging code
425 * is just logging an inode to make sure it exists after recovery.
427 * Also, don't overwrite i_size on directories during replay.
428 * log replay inserts and removes directory items based on the
429 * state of the tree found in the subvolume, and i_size is modified
432 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
433 struct btrfs_inode_item *src_item;
434 struct btrfs_inode_item *dst_item;
436 src_item = (struct btrfs_inode_item *)src_ptr;
437 dst_item = (struct btrfs_inode_item *)dst_ptr;
439 if (btrfs_inode_generation(eb, src_item) == 0)
442 if (overwrite_root &&
443 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
444 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
446 saved_i_size = btrfs_inode_size(path->nodes[0],
451 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
454 if (save_old_i_size) {
455 struct btrfs_inode_item *dst_item;
456 dst_item = (struct btrfs_inode_item *)dst_ptr;
457 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
460 /* make sure the generation is filled in */
461 if (key->type == BTRFS_INODE_ITEM_KEY) {
462 struct btrfs_inode_item *dst_item;
463 dst_item = (struct btrfs_inode_item *)dst_ptr;
464 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
465 btrfs_set_inode_generation(path->nodes[0], dst_item,
470 btrfs_mark_buffer_dirty(path->nodes[0]);
471 btrfs_release_path(path);
476 * simple helper to read an inode off the disk from a given root
477 * This can only be called for subvolume roots and not for the log
479 static noinline struct inode *read_one_inode(struct btrfs_root *root,
482 struct btrfs_key key;
485 key.objectid = objectid;
486 key.type = BTRFS_INODE_ITEM_KEY;
488 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
491 } else if (is_bad_inode(inode)) {
498 /* replays a single extent in 'eb' at 'slot' with 'key' into the
499 * subvolume 'root'. path is released on entry and should be released
502 * extents in the log tree have not been allocated out of the extent
503 * tree yet. So, this completes the allocation, taking a reference
504 * as required if the extent already exists or creating a new extent
505 * if it isn't in the extent allocation tree yet.
507 * The extent is inserted into the file, dropping any existing extents
508 * from the file that overlap the new one.
510 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
511 struct btrfs_root *root,
512 struct btrfs_path *path,
513 struct extent_buffer *eb, int slot,
514 struct btrfs_key *key)
518 u64 start = key->offset;
520 struct btrfs_file_extent_item *item;
521 struct inode *inode = NULL;
525 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
526 found_type = btrfs_file_extent_type(eb, item);
528 if (found_type == BTRFS_FILE_EXTENT_REG ||
529 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
530 nbytes = btrfs_file_extent_num_bytes(eb, item);
531 extent_end = start + nbytes;
534 * We don't add to the inodes nbytes if we are prealloc or a
537 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
539 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
540 size = btrfs_file_extent_inline_len(eb, item);
541 nbytes = btrfs_file_extent_ram_bytes(eb, item);
542 extent_end = ALIGN(start + size, root->sectorsize);
548 inode = read_one_inode(root, key->objectid);
555 * first check to see if we already have this extent in the
556 * file. This must be done before the btrfs_drop_extents run
557 * so we don't try to drop this extent.
559 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
563 (found_type == BTRFS_FILE_EXTENT_REG ||
564 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
565 struct btrfs_file_extent_item cmp1;
566 struct btrfs_file_extent_item cmp2;
567 struct btrfs_file_extent_item *existing;
568 struct extent_buffer *leaf;
570 leaf = path->nodes[0];
571 existing = btrfs_item_ptr(leaf, path->slots[0],
572 struct btrfs_file_extent_item);
574 read_extent_buffer(eb, &cmp1, (unsigned long)item,
576 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
580 * we already have a pointer to this exact extent,
581 * we don't have to do anything
583 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
584 btrfs_release_path(path);
588 btrfs_release_path(path);
590 /* drop any overlapping extents */
591 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
595 if (found_type == BTRFS_FILE_EXTENT_REG ||
596 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
598 unsigned long dest_offset;
599 struct btrfs_key ins;
601 ret = btrfs_insert_empty_item(trans, root, path, key,
605 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
607 copy_extent_buffer(path->nodes[0], eb, dest_offset,
608 (unsigned long)item, sizeof(*item));
610 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
611 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
612 ins.type = BTRFS_EXTENT_ITEM_KEY;
613 offset = key->offset - btrfs_file_extent_offset(eb, item);
615 if (ins.objectid > 0) {
618 LIST_HEAD(ordered_sums);
620 * is this extent already allocated in the extent
621 * allocation tree? If so, just add a reference
623 ret = btrfs_lookup_extent(root, ins.objectid,
626 ret = btrfs_inc_extent_ref(trans, root,
627 ins.objectid, ins.offset,
628 0, root->root_key.objectid,
629 key->objectid, offset, 0);
634 * insert the extent pointer in the extent
637 ret = btrfs_alloc_logged_file_extent(trans,
638 root, root->root_key.objectid,
639 key->objectid, offset, &ins);
643 btrfs_release_path(path);
645 if (btrfs_file_extent_compression(eb, item)) {
646 csum_start = ins.objectid;
647 csum_end = csum_start + ins.offset;
649 csum_start = ins.objectid +
650 btrfs_file_extent_offset(eb, item);
651 csum_end = csum_start +
652 btrfs_file_extent_num_bytes(eb, item);
655 ret = btrfs_lookup_csums_range(root->log_root,
656 csum_start, csum_end - 1,
660 while (!list_empty(&ordered_sums)) {
661 struct btrfs_ordered_sum *sums;
662 sums = list_entry(ordered_sums.next,
663 struct btrfs_ordered_sum,
666 ret = btrfs_csum_file_blocks(trans,
667 root->fs_info->csum_root,
669 list_del(&sums->list);
675 btrfs_release_path(path);
677 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
678 /* inline extents are easy, we just overwrite them */
679 ret = overwrite_item(trans, root, path, eb, slot, key);
684 inode_add_bytes(inode, nbytes);
685 ret = btrfs_update_inode(trans, root, inode);
693 * when cleaning up conflicts between the directory names in the
694 * subvolume, directory names in the log and directory names in the
695 * inode back references, we may have to unlink inodes from directories.
697 * This is a helper function to do the unlink of a specific directory
700 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
701 struct btrfs_root *root,
702 struct btrfs_path *path,
704 struct btrfs_dir_item *di)
709 struct extent_buffer *leaf;
710 struct btrfs_key location;
713 leaf = path->nodes[0];
715 btrfs_dir_item_key_to_cpu(leaf, di, &location);
716 name_len = btrfs_dir_name_len(leaf, di);
717 name = kmalloc(name_len, GFP_NOFS);
721 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
722 btrfs_release_path(path);
724 inode = read_one_inode(root, location.objectid);
730 ret = link_to_fixup_dir(trans, root, path, location.objectid);
734 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
737 btrfs_run_delayed_items(trans, root);
745 * helper function to see if a given name and sequence number found
746 * in an inode back reference are already in a directory and correctly
747 * point to this inode
749 static noinline int inode_in_dir(struct btrfs_root *root,
750 struct btrfs_path *path,
751 u64 dirid, u64 objectid, u64 index,
752 const char *name, int name_len)
754 struct btrfs_dir_item *di;
755 struct btrfs_key location;
758 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
759 index, name, name_len, 0);
760 if (di && !IS_ERR(di)) {
761 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
762 if (location.objectid != objectid)
766 btrfs_release_path(path);
768 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
769 if (di && !IS_ERR(di)) {
770 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
771 if (location.objectid != objectid)
777 btrfs_release_path(path);
782 * helper function to check a log tree for a named back reference in
783 * an inode. This is used to decide if a back reference that is
784 * found in the subvolume conflicts with what we find in the log.
786 * inode backreferences may have multiple refs in a single item,
787 * during replay we process one reference at a time, and we don't
788 * want to delete valid links to a file from the subvolume if that
789 * link is also in the log.
791 static noinline int backref_in_log(struct btrfs_root *log,
792 struct btrfs_key *key,
794 char *name, int namelen)
796 struct btrfs_path *path;
797 struct btrfs_inode_ref *ref;
799 unsigned long ptr_end;
800 unsigned long name_ptr;
806 path = btrfs_alloc_path();
810 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
814 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
816 if (key->type == BTRFS_INODE_EXTREF_KEY) {
817 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
818 name, namelen, NULL))
824 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
825 ptr_end = ptr + item_size;
826 while (ptr < ptr_end) {
827 ref = (struct btrfs_inode_ref *)ptr;
828 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
829 if (found_name_len == namelen) {
830 name_ptr = (unsigned long)(ref + 1);
831 ret = memcmp_extent_buffer(path->nodes[0], name,
838 ptr = (unsigned long)(ref + 1) + found_name_len;
841 btrfs_free_path(path);
845 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
846 struct btrfs_root *root,
847 struct btrfs_path *path,
848 struct btrfs_root *log_root,
849 struct inode *dir, struct inode *inode,
850 struct extent_buffer *eb,
851 u64 inode_objectid, u64 parent_objectid,
852 u64 ref_index, char *name, int namelen,
858 struct extent_buffer *leaf;
859 struct btrfs_dir_item *di;
860 struct btrfs_key search_key;
861 struct btrfs_inode_extref *extref;
864 /* Search old style refs */
865 search_key.objectid = inode_objectid;
866 search_key.type = BTRFS_INODE_REF_KEY;
867 search_key.offset = parent_objectid;
868 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
870 struct btrfs_inode_ref *victim_ref;
872 unsigned long ptr_end;
874 leaf = path->nodes[0];
876 /* are we trying to overwrite a back ref for the root directory
877 * if so, just jump out, we're done
879 if (search_key.objectid == search_key.offset)
882 /* check all the names in this back reference to see
883 * if they are in the log. if so, we allow them to stay
884 * otherwise they must be unlinked as a conflict
886 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
887 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
888 while (ptr < ptr_end) {
889 victim_ref = (struct btrfs_inode_ref *)ptr;
890 victim_name_len = btrfs_inode_ref_name_len(leaf,
892 victim_name = kmalloc(victim_name_len, GFP_NOFS);
896 read_extent_buffer(leaf, victim_name,
897 (unsigned long)(victim_ref + 1),
900 if (!backref_in_log(log_root, &search_key,
904 btrfs_inc_nlink(inode);
905 btrfs_release_path(path);
907 ret = btrfs_unlink_inode(trans, root, dir,
913 btrfs_run_delayed_items(trans, root);
919 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
923 * NOTE: we have searched root tree and checked the
924 * coresponding ref, it does not need to check again.
928 btrfs_release_path(path);
930 /* Same search but for extended refs */
931 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
932 inode_objectid, parent_objectid, 0,
934 if (!IS_ERR_OR_NULL(extref)) {
938 struct inode *victim_parent;
940 leaf = path->nodes[0];
942 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
943 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
945 while (cur_offset < item_size) {
946 extref = (struct btrfs_inode_extref *)base + cur_offset;
948 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
950 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
953 victim_name = kmalloc(victim_name_len, GFP_NOFS);
956 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
959 search_key.objectid = inode_objectid;
960 search_key.type = BTRFS_INODE_EXTREF_KEY;
961 search_key.offset = btrfs_extref_hash(parent_objectid,
965 if (!backref_in_log(log_root, &search_key,
966 parent_objectid, victim_name,
969 victim_parent = read_one_inode(root,
972 btrfs_inc_nlink(inode);
973 btrfs_release_path(path);
975 ret = btrfs_unlink_inode(trans, root,
980 btrfs_run_delayed_items(trans, root);
993 cur_offset += victim_name_len + sizeof(*extref);
997 btrfs_release_path(path);
999 /* look for a conflicting sequence number */
1000 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1001 ref_index, name, namelen, 0);
1002 if (di && !IS_ERR(di)) {
1003 ret = drop_one_dir_item(trans, root, path, dir, di);
1007 btrfs_release_path(path);
1009 /* look for a conflicing name */
1010 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1012 if (di && !IS_ERR(di)) {
1013 ret = drop_one_dir_item(trans, root, path, dir, di);
1017 btrfs_release_path(path);
1022 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1023 u32 *namelen, char **name, u64 *index,
1024 u64 *parent_objectid)
1026 struct btrfs_inode_extref *extref;
1028 extref = (struct btrfs_inode_extref *)ref_ptr;
1030 *namelen = btrfs_inode_extref_name_len(eb, extref);
1031 *name = kmalloc(*namelen, GFP_NOFS);
1035 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1038 *index = btrfs_inode_extref_index(eb, extref);
1039 if (parent_objectid)
1040 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1045 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1046 u32 *namelen, char **name, u64 *index)
1048 struct btrfs_inode_ref *ref;
1050 ref = (struct btrfs_inode_ref *)ref_ptr;
1052 *namelen = btrfs_inode_ref_name_len(eb, ref);
1053 *name = kmalloc(*namelen, GFP_NOFS);
1057 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1059 *index = btrfs_inode_ref_index(eb, ref);
1065 * replay one inode back reference item found in the log tree.
1066 * eb, slot and key refer to the buffer and key found in the log tree.
1067 * root is the destination we are replaying into, and path is for temp
1068 * use by this function. (it should be released on return).
1070 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1071 struct btrfs_root *root,
1072 struct btrfs_root *log,
1073 struct btrfs_path *path,
1074 struct extent_buffer *eb, int slot,
1075 struct btrfs_key *key)
1078 struct inode *inode;
1079 unsigned long ref_ptr;
1080 unsigned long ref_end;
1084 int search_done = 0;
1085 int log_ref_ver = 0;
1086 u64 parent_objectid;
1089 int ref_struct_size;
1091 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1092 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1094 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1095 struct btrfs_inode_extref *r;
1097 ref_struct_size = sizeof(struct btrfs_inode_extref);
1099 r = (struct btrfs_inode_extref *)ref_ptr;
1100 parent_objectid = btrfs_inode_extref_parent(eb, r);
1102 ref_struct_size = sizeof(struct btrfs_inode_ref);
1103 parent_objectid = key->offset;
1105 inode_objectid = key->objectid;
1108 * it is possible that we didn't log all the parent directories
1109 * for a given inode. If we don't find the dir, just don't
1110 * copy the back ref in. The link count fixup code will take
1113 dir = read_one_inode(root, parent_objectid);
1117 inode = read_one_inode(root, inode_objectid);
1123 while (ref_ptr < ref_end) {
1125 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1126 &ref_index, &parent_objectid);
1128 * parent object can change from one array
1132 dir = read_one_inode(root, parent_objectid);
1136 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1142 /* if we already have a perfect match, we're done */
1143 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
1144 ref_index, name, namelen)) {
1146 * look for a conflicting back reference in the
1147 * metadata. if we find one we have to unlink that name
1148 * of the file before we add our new link. Later on, we
1149 * overwrite any existing back reference, and we don't
1150 * want to create dangling pointers in the directory.
1154 ret = __add_inode_ref(trans, root, path, log,
1158 ref_index, name, namelen,
1168 /* insert our name */
1169 ret = btrfs_add_link(trans, dir, inode, name, namelen,
1174 btrfs_update_inode(trans, root, inode);
1177 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1185 /* finally write the back reference in the inode */
1186 ret = overwrite_item(trans, root, path, eb, slot, key);
1188 btrfs_release_path(path);
1194 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1195 struct btrfs_root *root, u64 offset)
1198 ret = btrfs_find_orphan_item(root, offset);
1200 ret = btrfs_insert_orphan_item(trans, root, offset);
1204 static int count_inode_extrefs(struct btrfs_root *root,
1205 struct inode *inode, struct btrfs_path *path)
1209 unsigned int nlink = 0;
1212 u64 inode_objectid = btrfs_ino(inode);
1215 struct btrfs_inode_extref *extref;
1216 struct extent_buffer *leaf;
1219 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1224 leaf = path->nodes[0];
1225 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1226 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1228 while (cur_offset < item_size) {
1229 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1230 name_len = btrfs_inode_extref_name_len(leaf, extref);
1234 cur_offset += name_len + sizeof(*extref);
1238 btrfs_release_path(path);
1240 btrfs_release_path(path);
1247 static int count_inode_refs(struct btrfs_root *root,
1248 struct inode *inode, struct btrfs_path *path)
1251 struct btrfs_key key;
1252 unsigned int nlink = 0;
1254 unsigned long ptr_end;
1256 u64 ino = btrfs_ino(inode);
1259 key.type = BTRFS_INODE_REF_KEY;
1260 key.offset = (u64)-1;
1263 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1267 if (path->slots[0] == 0)
1271 btrfs_item_key_to_cpu(path->nodes[0], &key,
1273 if (key.objectid != ino ||
1274 key.type != BTRFS_INODE_REF_KEY)
1276 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1277 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1279 while (ptr < ptr_end) {
1280 struct btrfs_inode_ref *ref;
1282 ref = (struct btrfs_inode_ref *)ptr;
1283 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1285 ptr = (unsigned long)(ref + 1) + name_len;
1289 if (key.offset == 0)
1292 btrfs_release_path(path);
1294 btrfs_release_path(path);
1300 * There are a few corners where the link count of the file can't
1301 * be properly maintained during replay. So, instead of adding
1302 * lots of complexity to the log code, we just scan the backrefs
1303 * for any file that has been through replay.
1305 * The scan will update the link count on the inode to reflect the
1306 * number of back refs found. If it goes down to zero, the iput
1307 * will free the inode.
1309 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1310 struct btrfs_root *root,
1311 struct inode *inode)
1313 struct btrfs_path *path;
1316 u64 ino = btrfs_ino(inode);
1318 path = btrfs_alloc_path();
1322 ret = count_inode_refs(root, inode, path);
1328 ret = count_inode_extrefs(root, inode, path);
1339 if (nlink != inode->i_nlink) {
1340 set_nlink(inode, nlink);
1341 btrfs_update_inode(trans, root, inode);
1343 BTRFS_I(inode)->index_cnt = (u64)-1;
1345 if (inode->i_nlink == 0) {
1346 if (S_ISDIR(inode->i_mode)) {
1347 ret = replay_dir_deletes(trans, root, NULL, path,
1352 ret = insert_orphan_item(trans, root, ino);
1356 btrfs_free_path(path);
1360 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1361 struct btrfs_root *root,
1362 struct btrfs_path *path)
1365 struct btrfs_key key;
1366 struct inode *inode;
1368 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1369 key.type = BTRFS_ORPHAN_ITEM_KEY;
1370 key.offset = (u64)-1;
1372 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1377 if (path->slots[0] == 0)
1382 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1383 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1384 key.type != BTRFS_ORPHAN_ITEM_KEY)
1387 ret = btrfs_del_item(trans, root, path);
1391 btrfs_release_path(path);
1392 inode = read_one_inode(root, key.offset);
1396 ret = fixup_inode_link_count(trans, root, inode);
1402 * fixup on a directory may create new entries,
1403 * make sure we always look for the highset possible
1406 key.offset = (u64)-1;
1410 btrfs_release_path(path);
1416 * record a given inode in the fixup dir so we can check its link
1417 * count when replay is done. The link count is incremented here
1418 * so the inode won't go away until we check it
1420 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1421 struct btrfs_root *root,
1422 struct btrfs_path *path,
1425 struct btrfs_key key;
1427 struct inode *inode;
1429 inode = read_one_inode(root, objectid);
1433 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1434 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1435 key.offset = objectid;
1437 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1439 btrfs_release_path(path);
1441 if (!inode->i_nlink)
1442 set_nlink(inode, 1);
1444 btrfs_inc_nlink(inode);
1445 ret = btrfs_update_inode(trans, root, inode);
1446 } else if (ret == -EEXIST) {
1449 BUG(); /* Logic Error */
1457 * when replaying the log for a directory, we only insert names
1458 * for inodes that actually exist. This means an fsync on a directory
1459 * does not implicitly fsync all the new files in it
1461 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1462 struct btrfs_root *root,
1463 struct btrfs_path *path,
1464 u64 dirid, u64 index,
1465 char *name, int name_len, u8 type,
1466 struct btrfs_key *location)
1468 struct inode *inode;
1472 inode = read_one_inode(root, location->objectid);
1476 dir = read_one_inode(root, dirid);
1481 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1483 /* FIXME, put inode into FIXUP list */
1491 * take a single entry in a log directory item and replay it into
1494 * if a conflicting item exists in the subdirectory already,
1495 * the inode it points to is unlinked and put into the link count
1498 * If a name from the log points to a file or directory that does
1499 * not exist in the FS, it is skipped. fsyncs on directories
1500 * do not force down inodes inside that directory, just changes to the
1501 * names or unlinks in a directory.
1503 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1504 struct btrfs_root *root,
1505 struct btrfs_path *path,
1506 struct extent_buffer *eb,
1507 struct btrfs_dir_item *di,
1508 struct btrfs_key *key)
1512 struct btrfs_dir_item *dst_di;
1513 struct btrfs_key found_key;
1514 struct btrfs_key log_key;
1520 dir = read_one_inode(root, key->objectid);
1524 name_len = btrfs_dir_name_len(eb, di);
1525 name = kmalloc(name_len, GFP_NOFS);
1529 log_type = btrfs_dir_type(eb, di);
1530 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1533 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1534 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1539 btrfs_release_path(path);
1541 if (key->type == BTRFS_DIR_ITEM_KEY) {
1542 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1544 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1545 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1554 if (IS_ERR_OR_NULL(dst_di)) {
1555 /* we need a sequence number to insert, so we only
1556 * do inserts for the BTRFS_DIR_INDEX_KEY types
1558 if (key->type != BTRFS_DIR_INDEX_KEY)
1563 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1564 /* the existing item matches the logged item */
1565 if (found_key.objectid == log_key.objectid &&
1566 found_key.type == log_key.type &&
1567 found_key.offset == log_key.offset &&
1568 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1573 * don't drop the conflicting directory entry if the inode
1574 * for the new entry doesn't exist
1579 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1583 if (key->type == BTRFS_DIR_INDEX_KEY)
1586 btrfs_release_path(path);
1592 btrfs_release_path(path);
1593 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1594 name, name_len, log_type, &log_key);
1595 if (ret && ret != -ENOENT)
1602 * find all the names in a directory item and reconcile them into
1603 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1604 * one name in a directory item, but the same code gets used for
1605 * both directory index types
1607 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1608 struct btrfs_root *root,
1609 struct btrfs_path *path,
1610 struct extent_buffer *eb, int slot,
1611 struct btrfs_key *key)
1614 u32 item_size = btrfs_item_size_nr(eb, slot);
1615 struct btrfs_dir_item *di;
1618 unsigned long ptr_end;
1620 ptr = btrfs_item_ptr_offset(eb, slot);
1621 ptr_end = ptr + item_size;
1622 while (ptr < ptr_end) {
1623 di = (struct btrfs_dir_item *)ptr;
1624 if (verify_dir_item(root, eb, di))
1626 name_len = btrfs_dir_name_len(eb, di);
1627 ret = replay_one_name(trans, root, path, eb, di, key);
1630 ptr = (unsigned long)(di + 1);
1637 * directory replay has two parts. There are the standard directory
1638 * items in the log copied from the subvolume, and range items
1639 * created in the log while the subvolume was logged.
1641 * The range items tell us which parts of the key space the log
1642 * is authoritative for. During replay, if a key in the subvolume
1643 * directory is in a logged range item, but not actually in the log
1644 * that means it was deleted from the directory before the fsync
1645 * and should be removed.
1647 static noinline int find_dir_range(struct btrfs_root *root,
1648 struct btrfs_path *path,
1649 u64 dirid, int key_type,
1650 u64 *start_ret, u64 *end_ret)
1652 struct btrfs_key key;
1654 struct btrfs_dir_log_item *item;
1658 if (*start_ret == (u64)-1)
1661 key.objectid = dirid;
1662 key.type = key_type;
1663 key.offset = *start_ret;
1665 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1669 if (path->slots[0] == 0)
1674 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1676 if (key.type != key_type || key.objectid != dirid) {
1680 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1681 struct btrfs_dir_log_item);
1682 found_end = btrfs_dir_log_end(path->nodes[0], item);
1684 if (*start_ret >= key.offset && *start_ret <= found_end) {
1686 *start_ret = key.offset;
1687 *end_ret = found_end;
1692 /* check the next slot in the tree to see if it is a valid item */
1693 nritems = btrfs_header_nritems(path->nodes[0]);
1694 if (path->slots[0] >= nritems) {
1695 ret = btrfs_next_leaf(root, path);
1702 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1704 if (key.type != key_type || key.objectid != dirid) {
1708 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1709 struct btrfs_dir_log_item);
1710 found_end = btrfs_dir_log_end(path->nodes[0], item);
1711 *start_ret = key.offset;
1712 *end_ret = found_end;
1715 btrfs_release_path(path);
1720 * this looks for a given directory item in the log. If the directory
1721 * item is not in the log, the item is removed and the inode it points
1724 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1725 struct btrfs_root *root,
1726 struct btrfs_root *log,
1727 struct btrfs_path *path,
1728 struct btrfs_path *log_path,
1730 struct btrfs_key *dir_key)
1733 struct extent_buffer *eb;
1736 struct btrfs_dir_item *di;
1737 struct btrfs_dir_item *log_di;
1740 unsigned long ptr_end;
1742 struct inode *inode;
1743 struct btrfs_key location;
1746 eb = path->nodes[0];
1747 slot = path->slots[0];
1748 item_size = btrfs_item_size_nr(eb, slot);
1749 ptr = btrfs_item_ptr_offset(eb, slot);
1750 ptr_end = ptr + item_size;
1751 while (ptr < ptr_end) {
1752 di = (struct btrfs_dir_item *)ptr;
1753 if (verify_dir_item(root, eb, di)) {
1758 name_len = btrfs_dir_name_len(eb, di);
1759 name = kmalloc(name_len, GFP_NOFS);
1764 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1767 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1768 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1771 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1772 log_di = btrfs_lookup_dir_index_item(trans, log,
1778 if (IS_ERR_OR_NULL(log_di)) {
1779 btrfs_dir_item_key_to_cpu(eb, di, &location);
1780 btrfs_release_path(path);
1781 btrfs_release_path(log_path);
1782 inode = read_one_inode(root, location.objectid);
1788 ret = link_to_fixup_dir(trans, root,
1789 path, location.objectid);
1796 btrfs_inc_nlink(inode);
1797 ret = btrfs_unlink_inode(trans, root, dir, inode,
1800 btrfs_run_delayed_items(trans, root);
1806 /* there might still be more names under this key
1807 * check and repeat if required
1809 ret = btrfs_search_slot(NULL, root, dir_key, path,
1816 btrfs_release_path(log_path);
1819 ptr = (unsigned long)(di + 1);
1824 btrfs_release_path(path);
1825 btrfs_release_path(log_path);
1830 * deletion replay happens before we copy any new directory items
1831 * out of the log or out of backreferences from inodes. It
1832 * scans the log to find ranges of keys that log is authoritative for,
1833 * and then scans the directory to find items in those ranges that are
1834 * not present in the log.
1836 * Anything we don't find in the log is unlinked and removed from the
1839 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1840 struct btrfs_root *root,
1841 struct btrfs_root *log,
1842 struct btrfs_path *path,
1843 u64 dirid, int del_all)
1847 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1849 struct btrfs_key dir_key;
1850 struct btrfs_key found_key;
1851 struct btrfs_path *log_path;
1854 dir_key.objectid = dirid;
1855 dir_key.type = BTRFS_DIR_ITEM_KEY;
1856 log_path = btrfs_alloc_path();
1860 dir = read_one_inode(root, dirid);
1861 /* it isn't an error if the inode isn't there, that can happen
1862 * because we replay the deletes before we copy in the inode item
1866 btrfs_free_path(log_path);
1874 range_end = (u64)-1;
1876 ret = find_dir_range(log, path, dirid, key_type,
1877 &range_start, &range_end);
1882 dir_key.offset = range_start;
1885 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1890 nritems = btrfs_header_nritems(path->nodes[0]);
1891 if (path->slots[0] >= nritems) {
1892 ret = btrfs_next_leaf(root, path);
1896 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1898 if (found_key.objectid != dirid ||
1899 found_key.type != dir_key.type)
1902 if (found_key.offset > range_end)
1905 ret = check_item_in_log(trans, root, log, path,
1910 if (found_key.offset == (u64)-1)
1912 dir_key.offset = found_key.offset + 1;
1914 btrfs_release_path(path);
1915 if (range_end == (u64)-1)
1917 range_start = range_end + 1;
1922 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1923 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1924 dir_key.type = BTRFS_DIR_INDEX_KEY;
1925 btrfs_release_path(path);
1929 btrfs_release_path(path);
1930 btrfs_free_path(log_path);
1936 * the process_func used to replay items from the log tree. This
1937 * gets called in two different stages. The first stage just looks
1938 * for inodes and makes sure they are all copied into the subvolume.
1940 * The second stage copies all the other item types from the log into
1941 * the subvolume. The two stage approach is slower, but gets rid of
1942 * lots of complexity around inodes referencing other inodes that exist
1943 * only in the log (references come from either directory items or inode
1946 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1947 struct walk_control *wc, u64 gen)
1950 struct btrfs_path *path;
1951 struct btrfs_root *root = wc->replay_dest;
1952 struct btrfs_key key;
1957 ret = btrfs_read_buffer(eb, gen);
1961 level = btrfs_header_level(eb);
1966 path = btrfs_alloc_path();
1970 nritems = btrfs_header_nritems(eb);
1971 for (i = 0; i < nritems; i++) {
1972 btrfs_item_key_to_cpu(eb, &key, i);
1974 /* inode keys are done during the first stage */
1975 if (key.type == BTRFS_INODE_ITEM_KEY &&
1976 wc->stage == LOG_WALK_REPLAY_INODES) {
1977 struct btrfs_inode_item *inode_item;
1980 inode_item = btrfs_item_ptr(eb, i,
1981 struct btrfs_inode_item);
1982 mode = btrfs_inode_mode(eb, inode_item);
1983 if (S_ISDIR(mode)) {
1984 ret = replay_dir_deletes(wc->trans,
1985 root, log, path, key.objectid, 0);
1989 ret = overwrite_item(wc->trans, root, path,
1994 /* for regular files, make sure corresponding
1995 * orhpan item exist. extents past the new EOF
1996 * will be truncated later by orphan cleanup.
1998 if (S_ISREG(mode)) {
1999 ret = insert_orphan_item(wc->trans, root,
2005 ret = link_to_fixup_dir(wc->trans, root,
2006 path, key.objectid);
2010 if (wc->stage < LOG_WALK_REPLAY_ALL)
2013 /* these keys are simply copied */
2014 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2015 ret = overwrite_item(wc->trans, root, path,
2019 } else if (key.type == BTRFS_INODE_REF_KEY ||
2020 key.type == BTRFS_INODE_EXTREF_KEY) {
2021 ret = add_inode_ref(wc->trans, root, log, path,
2023 if (ret && ret != -ENOENT)
2026 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2027 ret = replay_one_extent(wc->trans, root, path,
2031 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
2032 key.type == BTRFS_DIR_INDEX_KEY) {
2033 ret = replay_one_dir_item(wc->trans, root, path,
2039 btrfs_free_path(path);
2043 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2044 struct btrfs_root *root,
2045 struct btrfs_path *path, int *level,
2046 struct walk_control *wc)
2051 struct extent_buffer *next;
2052 struct extent_buffer *cur;
2053 struct extent_buffer *parent;
2057 WARN_ON(*level < 0);
2058 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2060 while (*level > 0) {
2061 WARN_ON(*level < 0);
2062 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2063 cur = path->nodes[*level];
2065 if (btrfs_header_level(cur) != *level)
2068 if (path->slots[*level] >=
2069 btrfs_header_nritems(cur))
2072 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2073 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2074 blocksize = btrfs_level_size(root, *level - 1);
2076 parent = path->nodes[*level];
2077 root_owner = btrfs_header_owner(parent);
2079 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
2084 ret = wc->process_func(root, next, wc, ptr_gen);
2086 free_extent_buffer(next);
2090 path->slots[*level]++;
2092 ret = btrfs_read_buffer(next, ptr_gen);
2094 free_extent_buffer(next);
2098 btrfs_tree_lock(next);
2099 btrfs_set_lock_blocking(next);
2100 clean_tree_block(trans, root, next);
2101 btrfs_wait_tree_block_writeback(next);
2102 btrfs_tree_unlock(next);
2104 WARN_ON(root_owner !=
2105 BTRFS_TREE_LOG_OBJECTID);
2106 ret = btrfs_free_and_pin_reserved_extent(root,
2109 free_extent_buffer(next);
2113 free_extent_buffer(next);
2116 ret = btrfs_read_buffer(next, ptr_gen);
2118 free_extent_buffer(next);
2122 WARN_ON(*level <= 0);
2123 if (path->nodes[*level-1])
2124 free_extent_buffer(path->nodes[*level-1]);
2125 path->nodes[*level-1] = next;
2126 *level = btrfs_header_level(next);
2127 path->slots[*level] = 0;
2130 WARN_ON(*level < 0);
2131 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2133 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2139 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2140 struct btrfs_root *root,
2141 struct btrfs_path *path, int *level,
2142 struct walk_control *wc)
2149 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2150 slot = path->slots[i];
2151 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2154 WARN_ON(*level == 0);
2157 struct extent_buffer *parent;
2158 if (path->nodes[*level] == root->node)
2159 parent = path->nodes[*level];
2161 parent = path->nodes[*level + 1];
2163 root_owner = btrfs_header_owner(parent);
2164 ret = wc->process_func(root, path->nodes[*level], wc,
2165 btrfs_header_generation(path->nodes[*level]));
2170 struct extent_buffer *next;
2172 next = path->nodes[*level];
2174 btrfs_tree_lock(next);
2175 btrfs_set_lock_blocking(next);
2176 clean_tree_block(trans, root, next);
2177 btrfs_wait_tree_block_writeback(next);
2178 btrfs_tree_unlock(next);
2180 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2181 ret = btrfs_free_and_pin_reserved_extent(root,
2182 path->nodes[*level]->start,
2183 path->nodes[*level]->len);
2187 free_extent_buffer(path->nodes[*level]);
2188 path->nodes[*level] = NULL;
2196 * drop the reference count on the tree rooted at 'snap'. This traverses
2197 * the tree freeing any blocks that have a ref count of zero after being
2200 static int walk_log_tree(struct btrfs_trans_handle *trans,
2201 struct btrfs_root *log, struct walk_control *wc)
2206 struct btrfs_path *path;
2209 path = btrfs_alloc_path();
2213 level = btrfs_header_level(log->node);
2215 path->nodes[level] = log->node;
2216 extent_buffer_get(log->node);
2217 path->slots[level] = 0;
2220 wret = walk_down_log_tree(trans, log, path, &level, wc);
2228 wret = walk_up_log_tree(trans, log, path, &level, wc);
2237 /* was the root node processed? if not, catch it here */
2238 if (path->nodes[orig_level]) {
2239 ret = wc->process_func(log, path->nodes[orig_level], wc,
2240 btrfs_header_generation(path->nodes[orig_level]));
2244 struct extent_buffer *next;
2246 next = path->nodes[orig_level];
2248 btrfs_tree_lock(next);
2249 btrfs_set_lock_blocking(next);
2250 clean_tree_block(trans, log, next);
2251 btrfs_wait_tree_block_writeback(next);
2252 btrfs_tree_unlock(next);
2254 WARN_ON(log->root_key.objectid !=
2255 BTRFS_TREE_LOG_OBJECTID);
2256 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2264 btrfs_free_path(path);
2269 * helper function to update the item for a given subvolumes log root
2270 * in the tree of log roots
2272 static int update_log_root(struct btrfs_trans_handle *trans,
2273 struct btrfs_root *log)
2277 if (log->log_transid == 1) {
2278 /* insert root item on the first sync */
2279 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2280 &log->root_key, &log->root_item);
2282 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2283 &log->root_key, &log->root_item);
2288 static int wait_log_commit(struct btrfs_trans_handle *trans,
2289 struct btrfs_root *root, unsigned long transid)
2292 int index = transid % 2;
2295 * we only allow two pending log transactions at a time,
2296 * so we know that if ours is more than 2 older than the
2297 * current transaction, we're done
2300 prepare_to_wait(&root->log_commit_wait[index],
2301 &wait, TASK_UNINTERRUPTIBLE);
2302 mutex_unlock(&root->log_mutex);
2304 if (root->fs_info->last_trans_log_full_commit !=
2305 trans->transid && root->log_transid < transid + 2 &&
2306 atomic_read(&root->log_commit[index]))
2309 finish_wait(&root->log_commit_wait[index], &wait);
2310 mutex_lock(&root->log_mutex);
2311 } while (root->fs_info->last_trans_log_full_commit !=
2312 trans->transid && root->log_transid < transid + 2 &&
2313 atomic_read(&root->log_commit[index]));
2317 static void wait_for_writer(struct btrfs_trans_handle *trans,
2318 struct btrfs_root *root)
2321 while (root->fs_info->last_trans_log_full_commit !=
2322 trans->transid && atomic_read(&root->log_writers)) {
2323 prepare_to_wait(&root->log_writer_wait,
2324 &wait, TASK_UNINTERRUPTIBLE);
2325 mutex_unlock(&root->log_mutex);
2326 if (root->fs_info->last_trans_log_full_commit !=
2327 trans->transid && atomic_read(&root->log_writers))
2329 mutex_lock(&root->log_mutex);
2330 finish_wait(&root->log_writer_wait, &wait);
2335 * btrfs_sync_log does sends a given tree log down to the disk and
2336 * updates the super blocks to record it. When this call is done,
2337 * you know that any inodes previously logged are safely on disk only
2340 * Any other return value means you need to call btrfs_commit_transaction.
2341 * Some of the edge cases for fsyncing directories that have had unlinks
2342 * or renames done in the past mean that sometimes the only safe
2343 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2344 * that has happened.
2346 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2347 struct btrfs_root *root)
2353 struct btrfs_root *log = root->log_root;
2354 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2355 unsigned long log_transid = 0;
2357 mutex_lock(&root->log_mutex);
2358 log_transid = root->log_transid;
2359 index1 = root->log_transid % 2;
2360 if (atomic_read(&root->log_commit[index1])) {
2361 wait_log_commit(trans, root, root->log_transid);
2362 mutex_unlock(&root->log_mutex);
2365 atomic_set(&root->log_commit[index1], 1);
2367 /* wait for previous tree log sync to complete */
2368 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2369 wait_log_commit(trans, root, root->log_transid - 1);
2371 int batch = atomic_read(&root->log_batch);
2372 /* when we're on an ssd, just kick the log commit out */
2373 if (!btrfs_test_opt(root, SSD) && root->log_multiple_pids) {
2374 mutex_unlock(&root->log_mutex);
2375 schedule_timeout_uninterruptible(1);
2376 mutex_lock(&root->log_mutex);
2378 wait_for_writer(trans, root);
2379 if (batch == atomic_read(&root->log_batch))
2383 /* bail out if we need to do a full commit */
2384 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2386 btrfs_free_logged_extents(log, log_transid);
2387 mutex_unlock(&root->log_mutex);
2391 if (log_transid % 2 == 0)
2392 mark = EXTENT_DIRTY;
2396 /* we start IO on all the marked extents here, but we don't actually
2397 * wait for them until later.
2399 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2401 btrfs_abort_transaction(trans, root, ret);
2402 btrfs_free_logged_extents(log, log_transid);
2403 mutex_unlock(&root->log_mutex);
2407 btrfs_set_root_node(&log->root_item, log->node);
2409 root->log_transid++;
2410 log->log_transid = root->log_transid;
2411 root->log_start_pid = 0;
2414 * IO has been started, blocks of the log tree have WRITTEN flag set
2415 * in their headers. new modifications of the log will be written to
2416 * new positions. so it's safe to allow log writers to go in.
2418 mutex_unlock(&root->log_mutex);
2420 mutex_lock(&log_root_tree->log_mutex);
2421 atomic_inc(&log_root_tree->log_batch);
2422 atomic_inc(&log_root_tree->log_writers);
2423 mutex_unlock(&log_root_tree->log_mutex);
2425 ret = update_log_root(trans, log);
2427 mutex_lock(&log_root_tree->log_mutex);
2428 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2430 if (waitqueue_active(&log_root_tree->log_writer_wait))
2431 wake_up(&log_root_tree->log_writer_wait);
2435 if (ret != -ENOSPC) {
2436 btrfs_abort_transaction(trans, root, ret);
2437 mutex_unlock(&log_root_tree->log_mutex);
2440 root->fs_info->last_trans_log_full_commit = trans->transid;
2441 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2442 btrfs_free_logged_extents(log, log_transid);
2443 mutex_unlock(&log_root_tree->log_mutex);
2448 index2 = log_root_tree->log_transid % 2;
2449 if (atomic_read(&log_root_tree->log_commit[index2])) {
2450 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2451 wait_log_commit(trans, log_root_tree,
2452 log_root_tree->log_transid);
2453 btrfs_free_logged_extents(log, log_transid);
2454 mutex_unlock(&log_root_tree->log_mutex);
2458 atomic_set(&log_root_tree->log_commit[index2], 1);
2460 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2461 wait_log_commit(trans, log_root_tree,
2462 log_root_tree->log_transid - 1);
2465 wait_for_writer(trans, log_root_tree);
2468 * now that we've moved on to the tree of log tree roots,
2469 * check the full commit flag again
2471 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2472 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2473 btrfs_free_logged_extents(log, log_transid);
2474 mutex_unlock(&log_root_tree->log_mutex);
2476 goto out_wake_log_root;
2479 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2480 &log_root_tree->dirty_log_pages,
2481 EXTENT_DIRTY | EXTENT_NEW);
2483 btrfs_abort_transaction(trans, root, ret);
2484 btrfs_free_logged_extents(log, log_transid);
2485 mutex_unlock(&log_root_tree->log_mutex);
2486 goto out_wake_log_root;
2488 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2489 btrfs_wait_logged_extents(log, log_transid);
2491 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2492 log_root_tree->node->start);
2493 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2494 btrfs_header_level(log_root_tree->node));
2496 log_root_tree->log_transid++;
2499 mutex_unlock(&log_root_tree->log_mutex);
2502 * nobody else is going to jump in and write the the ctree
2503 * super here because the log_commit atomic below is protecting
2504 * us. We must be called with a transaction handle pinning
2505 * the running transaction open, so a full commit can't hop
2506 * in and cause problems either.
2508 btrfs_scrub_pause_super(root);
2509 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2510 btrfs_scrub_continue_super(root);
2512 btrfs_abort_transaction(trans, root, ret);
2513 goto out_wake_log_root;
2516 mutex_lock(&root->log_mutex);
2517 if (root->last_log_commit < log_transid)
2518 root->last_log_commit = log_transid;
2519 mutex_unlock(&root->log_mutex);
2522 atomic_set(&log_root_tree->log_commit[index2], 0);
2524 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2525 wake_up(&log_root_tree->log_commit_wait[index2]);
2527 atomic_set(&root->log_commit[index1], 0);
2529 if (waitqueue_active(&root->log_commit_wait[index1]))
2530 wake_up(&root->log_commit_wait[index1]);
2534 static void free_log_tree(struct btrfs_trans_handle *trans,
2535 struct btrfs_root *log)
2540 struct walk_control wc = {
2542 .process_func = process_one_buffer
2546 ret = walk_log_tree(trans, log, &wc);
2548 /* I don't think this can happen but just in case */
2550 btrfs_abort_transaction(trans, log, ret);
2554 ret = find_first_extent_bit(&log->dirty_log_pages,
2555 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2560 clear_extent_bits(&log->dirty_log_pages, start, end,
2561 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2565 * We may have short-circuited the log tree with the full commit logic
2566 * and left ordered extents on our list, so clear these out to keep us
2567 * from leaking inodes and memory.
2569 btrfs_free_logged_extents(log, 0);
2570 btrfs_free_logged_extents(log, 1);
2572 free_extent_buffer(log->node);
2577 * free all the extents used by the tree log. This should be called
2578 * at commit time of the full transaction
2580 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2582 if (root->log_root) {
2583 free_log_tree(trans, root->log_root);
2584 root->log_root = NULL;
2589 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2590 struct btrfs_fs_info *fs_info)
2592 if (fs_info->log_root_tree) {
2593 free_log_tree(trans, fs_info->log_root_tree);
2594 fs_info->log_root_tree = NULL;
2600 * If both a file and directory are logged, and unlinks or renames are
2601 * mixed in, we have a few interesting corners:
2603 * create file X in dir Y
2604 * link file X to X.link in dir Y
2606 * unlink file X but leave X.link
2609 * After a crash we would expect only X.link to exist. But file X
2610 * didn't get fsync'd again so the log has back refs for X and X.link.
2612 * We solve this by removing directory entries and inode backrefs from the
2613 * log when a file that was logged in the current transaction is
2614 * unlinked. Any later fsync will include the updated log entries, and
2615 * we'll be able to reconstruct the proper directory items from backrefs.
2617 * This optimizations allows us to avoid relogging the entire inode
2618 * or the entire directory.
2620 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2621 struct btrfs_root *root,
2622 const char *name, int name_len,
2623 struct inode *dir, u64 index)
2625 struct btrfs_root *log;
2626 struct btrfs_dir_item *di;
2627 struct btrfs_path *path;
2631 u64 dir_ino = btrfs_ino(dir);
2633 if (BTRFS_I(dir)->logged_trans < trans->transid)
2636 ret = join_running_log_trans(root);
2640 mutex_lock(&BTRFS_I(dir)->log_mutex);
2642 log = root->log_root;
2643 path = btrfs_alloc_path();
2649 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
2650 name, name_len, -1);
2656 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2657 bytes_del += name_len;
2663 btrfs_release_path(path);
2664 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
2665 index, name, name_len, -1);
2671 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2672 bytes_del += name_len;
2679 /* update the directory size in the log to reflect the names
2683 struct btrfs_key key;
2685 key.objectid = dir_ino;
2687 key.type = BTRFS_INODE_ITEM_KEY;
2688 btrfs_release_path(path);
2690 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2696 struct btrfs_inode_item *item;
2699 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2700 struct btrfs_inode_item);
2701 i_size = btrfs_inode_size(path->nodes[0], item);
2702 if (i_size > bytes_del)
2703 i_size -= bytes_del;
2706 btrfs_set_inode_size(path->nodes[0], item, i_size);
2707 btrfs_mark_buffer_dirty(path->nodes[0]);
2710 btrfs_release_path(path);
2713 btrfs_free_path(path);
2715 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2716 if (ret == -ENOSPC) {
2717 root->fs_info->last_trans_log_full_commit = trans->transid;
2720 btrfs_abort_transaction(trans, root, ret);
2722 btrfs_end_log_trans(root);
2727 /* see comments for btrfs_del_dir_entries_in_log */
2728 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2729 struct btrfs_root *root,
2730 const char *name, int name_len,
2731 struct inode *inode, u64 dirid)
2733 struct btrfs_root *log;
2737 if (BTRFS_I(inode)->logged_trans < trans->transid)
2740 ret = join_running_log_trans(root);
2743 log = root->log_root;
2744 mutex_lock(&BTRFS_I(inode)->log_mutex);
2746 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
2748 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2749 if (ret == -ENOSPC) {
2750 root->fs_info->last_trans_log_full_commit = trans->transid;
2752 } else if (ret < 0 && ret != -ENOENT)
2753 btrfs_abort_transaction(trans, root, ret);
2754 btrfs_end_log_trans(root);
2760 * creates a range item in the log for 'dirid'. first_offset and
2761 * last_offset tell us which parts of the key space the log should
2762 * be considered authoritative for.
2764 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *log,
2766 struct btrfs_path *path,
2767 int key_type, u64 dirid,
2768 u64 first_offset, u64 last_offset)
2771 struct btrfs_key key;
2772 struct btrfs_dir_log_item *item;
2774 key.objectid = dirid;
2775 key.offset = first_offset;
2776 if (key_type == BTRFS_DIR_ITEM_KEY)
2777 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2779 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2780 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2784 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2785 struct btrfs_dir_log_item);
2786 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2787 btrfs_mark_buffer_dirty(path->nodes[0]);
2788 btrfs_release_path(path);
2793 * log all the items included in the current transaction for a given
2794 * directory. This also creates the range items in the log tree required
2795 * to replay anything deleted before the fsync
2797 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2798 struct btrfs_root *root, struct inode *inode,
2799 struct btrfs_path *path,
2800 struct btrfs_path *dst_path, int key_type,
2801 u64 min_offset, u64 *last_offset_ret)
2803 struct btrfs_key min_key;
2804 struct btrfs_key max_key;
2805 struct btrfs_root *log = root->log_root;
2806 struct extent_buffer *src;
2811 u64 first_offset = min_offset;
2812 u64 last_offset = (u64)-1;
2813 u64 ino = btrfs_ino(inode);
2815 log = root->log_root;
2816 max_key.objectid = ino;
2817 max_key.offset = (u64)-1;
2818 max_key.type = key_type;
2820 min_key.objectid = ino;
2821 min_key.type = key_type;
2822 min_key.offset = min_offset;
2824 path->keep_locks = 1;
2826 ret = btrfs_search_forward(root, &min_key, &max_key,
2827 path, trans->transid);
2830 * we didn't find anything from this transaction, see if there
2831 * is anything at all
2833 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2834 min_key.objectid = ino;
2835 min_key.type = key_type;
2836 min_key.offset = (u64)-1;
2837 btrfs_release_path(path);
2838 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2840 btrfs_release_path(path);
2843 ret = btrfs_previous_item(root, path, ino, key_type);
2845 /* if ret == 0 there are items for this type,
2846 * create a range to tell us the last key of this type.
2847 * otherwise, there are no items in this directory after
2848 * *min_offset, and we create a range to indicate that.
2851 struct btrfs_key tmp;
2852 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2854 if (key_type == tmp.type)
2855 first_offset = max(min_offset, tmp.offset) + 1;
2860 /* go backward to find any previous key */
2861 ret = btrfs_previous_item(root, path, ino, key_type);
2863 struct btrfs_key tmp;
2864 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2865 if (key_type == tmp.type) {
2866 first_offset = tmp.offset;
2867 ret = overwrite_item(trans, log, dst_path,
2868 path->nodes[0], path->slots[0],
2876 btrfs_release_path(path);
2878 /* find the first key from this transaction again */
2879 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2886 * we have a block from this transaction, log every item in it
2887 * from our directory
2890 struct btrfs_key tmp;
2891 src = path->nodes[0];
2892 nritems = btrfs_header_nritems(src);
2893 for (i = path->slots[0]; i < nritems; i++) {
2894 btrfs_item_key_to_cpu(src, &min_key, i);
2896 if (min_key.objectid != ino || min_key.type != key_type)
2898 ret = overwrite_item(trans, log, dst_path, src, i,
2905 path->slots[0] = nritems;
2908 * look ahead to the next item and see if it is also
2909 * from this directory and from this transaction
2911 ret = btrfs_next_leaf(root, path);
2913 last_offset = (u64)-1;
2916 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2917 if (tmp.objectid != ino || tmp.type != key_type) {
2918 last_offset = (u64)-1;
2921 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2922 ret = overwrite_item(trans, log, dst_path,
2923 path->nodes[0], path->slots[0],
2928 last_offset = tmp.offset;
2933 btrfs_release_path(path);
2934 btrfs_release_path(dst_path);
2937 *last_offset_ret = last_offset;
2939 * insert the log range keys to indicate where the log
2942 ret = insert_dir_log_key(trans, log, path, key_type,
2943 ino, first_offset, last_offset);
2951 * logging directories is very similar to logging inodes, We find all the items
2952 * from the current transaction and write them to the log.
2954 * The recovery code scans the directory in the subvolume, and if it finds a
2955 * key in the range logged that is not present in the log tree, then it means
2956 * that dir entry was unlinked during the transaction.
2958 * In order for that scan to work, we must include one key smaller than
2959 * the smallest logged by this transaction and one key larger than the largest
2960 * key logged by this transaction.
2962 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2963 struct btrfs_root *root, struct inode *inode,
2964 struct btrfs_path *path,
2965 struct btrfs_path *dst_path)
2970 int key_type = BTRFS_DIR_ITEM_KEY;
2976 ret = log_dir_items(trans, root, inode, path,
2977 dst_path, key_type, min_key,
2981 if (max_key == (u64)-1)
2983 min_key = max_key + 1;
2986 if (key_type == BTRFS_DIR_ITEM_KEY) {
2987 key_type = BTRFS_DIR_INDEX_KEY;
2994 * a helper function to drop items from the log before we relog an
2995 * inode. max_key_type indicates the highest item type to remove.
2996 * This cannot be run for file data extents because it does not
2997 * free the extents they point to.
2999 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3000 struct btrfs_root *log,
3001 struct btrfs_path *path,
3002 u64 objectid, int max_key_type)
3005 struct btrfs_key key;
3006 struct btrfs_key found_key;
3009 key.objectid = objectid;
3010 key.type = max_key_type;
3011 key.offset = (u64)-1;
3014 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3015 BUG_ON(ret == 0); /* Logic error */
3019 if (path->slots[0] == 0)
3023 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3026 if (found_key.objectid != objectid)
3029 found_key.offset = 0;
3031 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3034 ret = btrfs_del_items(trans, log, path, start_slot,
3035 path->slots[0] - start_slot + 1);
3037 * If start slot isn't 0 then we don't need to re-search, we've
3038 * found the last guy with the objectid in this tree.
3040 if (ret || start_slot != 0)
3042 btrfs_release_path(path);
3044 btrfs_release_path(path);
3050 static void fill_inode_item(struct btrfs_trans_handle *trans,
3051 struct extent_buffer *leaf,
3052 struct btrfs_inode_item *item,
3053 struct inode *inode, int log_inode_only)
3055 struct btrfs_map_token token;
3057 btrfs_init_map_token(&token);
3059 if (log_inode_only) {
3060 /* set the generation to zero so the recover code
3061 * can tell the difference between an logging
3062 * just to say 'this inode exists' and a logging
3063 * to say 'update this inode with these values'
3065 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3066 btrfs_set_token_inode_size(leaf, item, 0, &token);
3068 btrfs_set_token_inode_generation(leaf, item,
3069 BTRFS_I(inode)->generation,
3071 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3074 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3075 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3076 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3077 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3079 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3080 inode->i_atime.tv_sec, &token);
3081 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3082 inode->i_atime.tv_nsec, &token);
3084 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3085 inode->i_mtime.tv_sec, &token);
3086 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3087 inode->i_mtime.tv_nsec, &token);
3089 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3090 inode->i_ctime.tv_sec, &token);
3091 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3092 inode->i_ctime.tv_nsec, &token);
3094 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3097 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3098 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3099 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3100 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3101 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3104 static int log_inode_item(struct btrfs_trans_handle *trans,
3105 struct btrfs_root *log, struct btrfs_path *path,
3106 struct inode *inode)
3108 struct btrfs_inode_item *inode_item;
3109 struct btrfs_key key;
3112 memcpy(&key, &BTRFS_I(inode)->location, sizeof(key));
3113 ret = btrfs_insert_empty_item(trans, log, path, &key,
3114 sizeof(*inode_item));
3115 if (ret && ret != -EEXIST)
3117 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3118 struct btrfs_inode_item);
3119 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0);
3120 btrfs_release_path(path);
3124 static noinline int copy_items(struct btrfs_trans_handle *trans,
3125 struct inode *inode,
3126 struct btrfs_path *dst_path,
3127 struct extent_buffer *src,
3128 int start_slot, int nr, int inode_only)
3130 unsigned long src_offset;
3131 unsigned long dst_offset;
3132 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3133 struct btrfs_file_extent_item *extent;
3134 struct btrfs_inode_item *inode_item;
3136 struct btrfs_key *ins_keys;
3140 struct list_head ordered_sums;
3141 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3143 INIT_LIST_HEAD(&ordered_sums);
3145 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3146 nr * sizeof(u32), GFP_NOFS);
3150 ins_sizes = (u32 *)ins_data;
3151 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3153 for (i = 0; i < nr; i++) {
3154 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3155 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3157 ret = btrfs_insert_empty_items(trans, log, dst_path,
3158 ins_keys, ins_sizes, nr);
3164 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3165 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3166 dst_path->slots[0]);
3168 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3170 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3171 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3173 struct btrfs_inode_item);
3174 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3175 inode, inode_only == LOG_INODE_EXISTS);
3177 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3178 src_offset, ins_sizes[i]);
3181 /* take a reference on file data extents so that truncates
3182 * or deletes of this inode don't have to relog the inode
3185 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY &&
3188 extent = btrfs_item_ptr(src, start_slot + i,
3189 struct btrfs_file_extent_item);
3191 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3194 found_type = btrfs_file_extent_type(src, extent);
3195 if (found_type == BTRFS_FILE_EXTENT_REG) {
3197 ds = btrfs_file_extent_disk_bytenr(src,
3199 /* ds == 0 is a hole */
3203 dl = btrfs_file_extent_disk_num_bytes(src,
3205 cs = btrfs_file_extent_offset(src, extent);
3206 cl = btrfs_file_extent_num_bytes(src,
3208 if (btrfs_file_extent_compression(src,
3214 ret = btrfs_lookup_csums_range(
3215 log->fs_info->csum_root,
3216 ds + cs, ds + cs + cl - 1,
3219 btrfs_release_path(dst_path);
3227 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3228 btrfs_release_path(dst_path);
3232 * we have to do this after the loop above to avoid changing the
3233 * log tree while trying to change the log tree.
3236 while (!list_empty(&ordered_sums)) {
3237 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3238 struct btrfs_ordered_sum,
3241 ret = btrfs_csum_file_blocks(trans, log, sums);
3242 list_del(&sums->list);
3248 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3250 struct extent_map *em1, *em2;
3252 em1 = list_entry(a, struct extent_map, list);
3253 em2 = list_entry(b, struct extent_map, list);
3255 if (em1->start < em2->start)
3257 else if (em1->start > em2->start)
3262 static int log_one_extent(struct btrfs_trans_handle *trans,
3263 struct inode *inode, struct btrfs_root *root,
3264 struct extent_map *em, struct btrfs_path *path)
3266 struct btrfs_root *log = root->log_root;
3267 struct btrfs_file_extent_item *fi;
3268 struct extent_buffer *leaf;
3269 struct btrfs_ordered_extent *ordered;
3270 struct list_head ordered_sums;
3271 struct btrfs_map_token token;
3272 struct btrfs_key key;
3273 u64 mod_start = em->mod_start;
3274 u64 mod_len = em->mod_len;
3277 u64 extent_offset = em->start - em->orig_start;
3280 int index = log->log_transid % 2;
3281 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3283 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
3284 em->start + em->len, NULL, 0);
3288 INIT_LIST_HEAD(&ordered_sums);
3289 btrfs_init_map_token(&token);
3290 key.objectid = btrfs_ino(inode);
3291 key.type = BTRFS_EXTENT_DATA_KEY;
3292 key.offset = em->start;
3294 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*fi));
3297 leaf = path->nodes[0];
3298 fi = btrfs_item_ptr(leaf, path->slots[0],
3299 struct btrfs_file_extent_item);
3301 btrfs_set_token_file_extent_generation(leaf, fi, em->generation,
3303 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3305 btrfs_set_token_file_extent_type(leaf, fi,
3306 BTRFS_FILE_EXTENT_PREALLOC,
3309 btrfs_set_token_file_extent_type(leaf, fi,
3310 BTRFS_FILE_EXTENT_REG,
3312 if (em->block_start == 0)
3316 block_len = max(em->block_len, em->orig_block_len);
3317 if (em->compress_type != BTRFS_COMPRESS_NONE) {
3318 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3321 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3323 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
3324 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3326 extent_offset, &token);
3327 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3330 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
3331 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
3335 btrfs_set_token_file_extent_offset(leaf, fi,
3336 em->start - em->orig_start,
3338 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
3339 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
3340 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
3342 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
3343 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
3344 btrfs_mark_buffer_dirty(leaf);
3346 btrfs_release_path(path);
3354 if (em->compress_type) {
3356 csum_len = block_len;
3360 * First check and see if our csums are on our outstanding ordered
3364 spin_lock_irq(&log->log_extents_lock[index]);
3365 list_for_each_entry(ordered, &log->logged_list[index], log_list) {
3366 struct btrfs_ordered_sum *sum;
3371 if (ordered->inode != inode)
3374 if (ordered->file_offset + ordered->len <= mod_start ||
3375 mod_start + mod_len <= ordered->file_offset)
3379 * We are going to copy all the csums on this ordered extent, so
3380 * go ahead and adjust mod_start and mod_len in case this
3381 * ordered extent has already been logged.
3383 if (ordered->file_offset > mod_start) {
3384 if (ordered->file_offset + ordered->len >=
3385 mod_start + mod_len)
3386 mod_len = ordered->file_offset - mod_start;
3388 * If we have this case
3390 * |--------- logged extent ---------|
3391 * |----- ordered extent ----|
3393 * Just don't mess with mod_start and mod_len, we'll
3394 * just end up logging more csums than we need and it
3398 if (ordered->file_offset + ordered->len <
3399 mod_start + mod_len) {
3400 mod_len = (mod_start + mod_len) -
3401 (ordered->file_offset + ordered->len);
3402 mod_start = ordered->file_offset +
3410 * To keep us from looping for the above case of an ordered
3411 * extent that falls inside of the logged extent.
3413 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3416 atomic_inc(&ordered->refs);
3417 spin_unlock_irq(&log->log_extents_lock[index]);
3419 * we've dropped the lock, we must either break or
3420 * start over after this.
3423 wait_event(ordered->wait, ordered->csum_bytes_left == 0);
3425 list_for_each_entry(sum, &ordered->list, list) {
3426 ret = btrfs_csum_file_blocks(trans, log, sum);
3428 btrfs_put_ordered_extent(ordered);
3432 btrfs_put_ordered_extent(ordered);
3436 spin_unlock_irq(&log->log_extents_lock[index]);
3439 if (!mod_len || ret)
3442 csum_offset = mod_start - em->start;
3445 /* block start is already adjusted for the file extent offset. */
3446 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
3447 em->block_start + csum_offset,
3448 em->block_start + csum_offset +
3449 csum_len - 1, &ordered_sums, 0);
3453 while (!list_empty(&ordered_sums)) {
3454 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3455 struct btrfs_ordered_sum,
3458 ret = btrfs_csum_file_blocks(trans, log, sums);
3459 list_del(&sums->list);
3466 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
3467 struct btrfs_root *root,
3468 struct inode *inode,
3469 struct btrfs_path *path)
3471 struct extent_map *em, *n;
3472 struct list_head extents;
3473 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3478 INIT_LIST_HEAD(&extents);
3480 write_lock(&tree->lock);
3481 test_gen = root->fs_info->last_trans_committed;
3483 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
3484 list_del_init(&em->list);
3487 * Just an arbitrary number, this can be really CPU intensive
3488 * once we start getting a lot of extents, and really once we
3489 * have a bunch of extents we just want to commit since it will
3492 if (++num > 32768) {
3493 list_del_init(&tree->modified_extents);
3498 if (em->generation <= test_gen)
3500 /* Need a ref to keep it from getting evicted from cache */
3501 atomic_inc(&em->refs);
3502 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
3503 list_add_tail(&em->list, &extents);
3507 list_sort(NULL, &extents, extent_cmp);
3510 while (!list_empty(&extents)) {
3511 em = list_entry(extents.next, struct extent_map, list);
3513 list_del_init(&em->list);
3516 * If we had an error we just need to delete everybody from our
3520 clear_em_logging(tree, em);
3521 free_extent_map(em);
3525 write_unlock(&tree->lock);
3527 ret = log_one_extent(trans, inode, root, em, path);
3528 write_lock(&tree->lock);
3529 clear_em_logging(tree, em);
3530 free_extent_map(em);
3532 WARN_ON(!list_empty(&extents));
3533 write_unlock(&tree->lock);
3535 btrfs_release_path(path);
3539 /* log a single inode in the tree log.
3540 * At least one parent directory for this inode must exist in the tree
3541 * or be logged already.
3543 * Any items from this inode changed by the current transaction are copied
3544 * to the log tree. An extra reference is taken on any extents in this
3545 * file, allowing us to avoid a whole pile of corner cases around logging
3546 * blocks that have been removed from the tree.
3548 * See LOG_INODE_ALL and related defines for a description of what inode_only
3551 * This handles both files and directories.
3553 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
3554 struct btrfs_root *root, struct inode *inode,
3557 struct btrfs_path *path;
3558 struct btrfs_path *dst_path;
3559 struct btrfs_key min_key;
3560 struct btrfs_key max_key;
3561 struct btrfs_root *log = root->log_root;
3562 struct extent_buffer *src = NULL;
3566 int ins_start_slot = 0;
3568 bool fast_search = false;
3569 u64 ino = btrfs_ino(inode);
3571 path = btrfs_alloc_path();
3574 dst_path = btrfs_alloc_path();
3576 btrfs_free_path(path);
3580 min_key.objectid = ino;
3581 min_key.type = BTRFS_INODE_ITEM_KEY;
3584 max_key.objectid = ino;
3587 /* today the code can only do partial logging of directories */
3588 if (S_ISDIR(inode->i_mode) ||
3589 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3590 &BTRFS_I(inode)->runtime_flags) &&
3591 inode_only == LOG_INODE_EXISTS))
3592 max_key.type = BTRFS_XATTR_ITEM_KEY;
3594 max_key.type = (u8)-1;
3595 max_key.offset = (u64)-1;
3597 /* Only run delayed items if we are a dir or a new file */
3598 if (S_ISDIR(inode->i_mode) ||
3599 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) {
3600 ret = btrfs_commit_inode_delayed_items(trans, inode);
3602 btrfs_free_path(path);
3603 btrfs_free_path(dst_path);
3608 mutex_lock(&BTRFS_I(inode)->log_mutex);
3610 btrfs_get_logged_extents(log, inode);
3613 * a brute force approach to making sure we get the most uptodate
3614 * copies of everything.
3616 if (S_ISDIR(inode->i_mode)) {
3617 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
3619 if (inode_only == LOG_INODE_EXISTS)
3620 max_key_type = BTRFS_XATTR_ITEM_KEY;
3621 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
3623 if (test_and_clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3624 &BTRFS_I(inode)->runtime_flags)) {
3625 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3626 &BTRFS_I(inode)->runtime_flags);
3627 ret = btrfs_truncate_inode_items(trans, log,
3629 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3630 &BTRFS_I(inode)->runtime_flags)) {
3631 if (inode_only == LOG_INODE_ALL)
3633 max_key.type = BTRFS_XATTR_ITEM_KEY;
3634 ret = drop_objectid_items(trans, log, path, ino,
3637 if (inode_only == LOG_INODE_ALL)
3639 ret = log_inode_item(trans, log, dst_path, inode);
3652 path->keep_locks = 1;
3656 ret = btrfs_search_forward(root, &min_key, &max_key,
3657 path, trans->transid);
3661 /* note, ins_nr might be > 0 here, cleanup outside the loop */
3662 if (min_key.objectid != ino)
3664 if (min_key.type > max_key.type)
3667 src = path->nodes[0];
3668 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
3671 } else if (!ins_nr) {
3672 ins_start_slot = path->slots[0];
3677 ret = copy_items(trans, inode, dst_path, src, ins_start_slot,
3678 ins_nr, inode_only);
3684 ins_start_slot = path->slots[0];
3687 nritems = btrfs_header_nritems(path->nodes[0]);
3689 if (path->slots[0] < nritems) {
3690 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
3695 ret = copy_items(trans, inode, dst_path, src,
3697 ins_nr, inode_only);
3704 btrfs_release_path(path);
3706 if (min_key.offset < (u64)-1)
3708 else if (min_key.type < (u8)-1)
3710 else if (min_key.objectid < (u64)-1)
3716 ret = copy_items(trans, inode, dst_path, src, ins_start_slot,
3717 ins_nr, inode_only);
3727 btrfs_release_path(dst_path);
3728 ret = btrfs_log_changed_extents(trans, root, inode, dst_path);
3734 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3735 struct extent_map *em, *n;
3737 write_lock(&tree->lock);
3738 list_for_each_entry_safe(em, n, &tree->modified_extents, list)
3739 list_del_init(&em->list);
3740 write_unlock(&tree->lock);
3743 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
3744 btrfs_release_path(path);
3745 btrfs_release_path(dst_path);
3746 ret = log_directory_changes(trans, root, inode, path, dst_path);
3752 BTRFS_I(inode)->logged_trans = trans->transid;
3753 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
3756 btrfs_free_logged_extents(log, log->log_transid);
3757 mutex_unlock(&BTRFS_I(inode)->log_mutex);
3759 btrfs_free_path(path);
3760 btrfs_free_path(dst_path);
3765 * follow the dentry parent pointers up the chain and see if any
3766 * of the directories in it require a full commit before they can
3767 * be logged. Returns zero if nothing special needs to be done or 1 if
3768 * a full commit is required.
3770 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
3771 struct inode *inode,
3772 struct dentry *parent,
3773 struct super_block *sb,
3777 struct btrfs_root *root;
3778 struct dentry *old_parent = NULL;
3781 * for regular files, if its inode is already on disk, we don't
3782 * have to worry about the parents at all. This is because
3783 * we can use the last_unlink_trans field to record renames
3784 * and other fun in this file.
3786 if (S_ISREG(inode->i_mode) &&
3787 BTRFS_I(inode)->generation <= last_committed &&
3788 BTRFS_I(inode)->last_unlink_trans <= last_committed)
3791 if (!S_ISDIR(inode->i_mode)) {
3792 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3794 inode = parent->d_inode;
3798 BTRFS_I(inode)->logged_trans = trans->transid;
3801 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
3802 root = BTRFS_I(inode)->root;
3805 * make sure any commits to the log are forced
3806 * to be full commits
3808 root->fs_info->last_trans_log_full_commit =
3814 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3817 if (IS_ROOT(parent))
3820 parent = dget_parent(parent);
3822 old_parent = parent;
3823 inode = parent->d_inode;
3832 * helper function around btrfs_log_inode to make sure newly created
3833 * parent directories also end up in the log. A minimal inode and backref
3834 * only logging is done of any parent directories that are older than
3835 * the last committed transaction
3837 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
3838 struct btrfs_root *root, struct inode *inode,
3839 struct dentry *parent, int exists_only)
3841 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
3842 struct super_block *sb;
3843 struct dentry *old_parent = NULL;
3845 u64 last_committed = root->fs_info->last_trans_committed;
3849 if (btrfs_test_opt(root, NOTREELOG)) {
3854 if (root->fs_info->last_trans_log_full_commit >
3855 root->fs_info->last_trans_committed) {
3860 if (root != BTRFS_I(inode)->root ||
3861 btrfs_root_refs(&root->root_item) == 0) {
3866 ret = check_parent_dirs_for_sync(trans, inode, parent,
3867 sb, last_committed);
3871 if (btrfs_inode_in_log(inode, trans->transid)) {
3872 ret = BTRFS_NO_LOG_SYNC;
3876 ret = start_log_trans(trans, root);
3880 ret = btrfs_log_inode(trans, root, inode, inode_only);
3885 * for regular files, if its inode is already on disk, we don't
3886 * have to worry about the parents at all. This is because
3887 * we can use the last_unlink_trans field to record renames
3888 * and other fun in this file.
3890 if (S_ISREG(inode->i_mode) &&
3891 BTRFS_I(inode)->generation <= last_committed &&
3892 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3897 inode_only = LOG_INODE_EXISTS;
3899 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3902 inode = parent->d_inode;
3903 if (root != BTRFS_I(inode)->root)
3906 if (BTRFS_I(inode)->generation >
3907 root->fs_info->last_trans_committed) {
3908 ret = btrfs_log_inode(trans, root, inode, inode_only);
3912 if (IS_ROOT(parent))
3915 parent = dget_parent(parent);
3917 old_parent = parent;
3923 root->fs_info->last_trans_log_full_commit = trans->transid;
3926 btrfs_end_log_trans(root);
3932 * it is not safe to log dentry if the chunk root has added new
3933 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3934 * If this returns 1, you must commit the transaction to safely get your
3937 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3938 struct btrfs_root *root, struct dentry *dentry)
3940 struct dentry *parent = dget_parent(dentry);
3943 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
3950 * should be called during mount to recover any replay any log trees
3953 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3956 struct btrfs_path *path;
3957 struct btrfs_trans_handle *trans;
3958 struct btrfs_key key;
3959 struct btrfs_key found_key;
3960 struct btrfs_key tmp_key;
3961 struct btrfs_root *log;
3962 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3963 struct walk_control wc = {
3964 .process_func = process_one_buffer,
3968 path = btrfs_alloc_path();
3972 fs_info->log_root_recovering = 1;
3974 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3975 if (IS_ERR(trans)) {
3976 ret = PTR_ERR(trans);
3983 ret = walk_log_tree(trans, log_root_tree, &wc);
3985 btrfs_error(fs_info, ret, "Failed to pin buffers while "
3986 "recovering log root tree.");
3991 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3992 key.offset = (u64)-1;
3993 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3996 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3999 btrfs_error(fs_info, ret,
4000 "Couldn't find tree log root.");
4004 if (path->slots[0] == 0)
4008 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4010 btrfs_release_path(path);
4011 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4014 log = btrfs_read_fs_root(log_root_tree, &found_key);
4017 btrfs_error(fs_info, ret,
4018 "Couldn't read tree log root.");
4022 tmp_key.objectid = found_key.offset;
4023 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
4024 tmp_key.offset = (u64)-1;
4026 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
4027 if (IS_ERR(wc.replay_dest)) {
4028 ret = PTR_ERR(wc.replay_dest);
4029 free_extent_buffer(log->node);
4030 free_extent_buffer(log->commit_root);
4032 btrfs_error(fs_info, ret, "Couldn't read target root "
4033 "for tree log recovery.");
4037 wc.replay_dest->log_root = log;
4038 btrfs_record_root_in_trans(trans, wc.replay_dest);
4039 ret = walk_log_tree(trans, log, &wc);
4041 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
4042 ret = fixup_inode_link_counts(trans, wc.replay_dest,
4046 key.offset = found_key.offset - 1;
4047 wc.replay_dest->log_root = NULL;
4048 free_extent_buffer(log->node);
4049 free_extent_buffer(log->commit_root);
4055 if (found_key.offset == 0)
4058 btrfs_release_path(path);
4060 /* step one is to pin it all, step two is to replay just inodes */
4063 wc.process_func = replay_one_buffer;
4064 wc.stage = LOG_WALK_REPLAY_INODES;
4067 /* step three is to replay everything */
4068 if (wc.stage < LOG_WALK_REPLAY_ALL) {
4073 btrfs_free_path(path);
4075 /* step 4: commit the transaction, which also unpins the blocks */
4076 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
4080 free_extent_buffer(log_root_tree->node);
4081 log_root_tree->log_root = NULL;
4082 fs_info->log_root_recovering = 0;
4083 kfree(log_root_tree);
4088 btrfs_end_transaction(wc.trans, fs_info->tree_root);
4089 btrfs_free_path(path);
4094 * there are some corner cases where we want to force a full
4095 * commit instead of allowing a directory to be logged.
4097 * They revolve around files there were unlinked from the directory, and
4098 * this function updates the parent directory so that a full commit is
4099 * properly done if it is fsync'd later after the unlinks are done.
4101 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
4102 struct inode *dir, struct inode *inode,
4106 * when we're logging a file, if it hasn't been renamed
4107 * or unlinked, and its inode is fully committed on disk,
4108 * we don't have to worry about walking up the directory chain
4109 * to log its parents.
4111 * So, we use the last_unlink_trans field to put this transid
4112 * into the file. When the file is logged we check it and
4113 * don't log the parents if the file is fully on disk.
4115 if (S_ISREG(inode->i_mode))
4116 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4119 * if this directory was already logged any new
4120 * names for this file/dir will get recorded
4123 if (BTRFS_I(dir)->logged_trans == trans->transid)
4127 * if the inode we're about to unlink was logged,
4128 * the log will be properly updated for any new names
4130 if (BTRFS_I(inode)->logged_trans == trans->transid)
4134 * when renaming files across directories, if the directory
4135 * there we're unlinking from gets fsync'd later on, there's
4136 * no way to find the destination directory later and fsync it
4137 * properly. So, we have to be conservative and force commits
4138 * so the new name gets discovered.
4143 /* we can safely do the unlink without any special recording */
4147 BTRFS_I(dir)->last_unlink_trans = trans->transid;
4151 * Call this after adding a new name for a file and it will properly
4152 * update the log to reflect the new name.
4154 * It will return zero if all goes well, and it will return 1 if a
4155 * full transaction commit is required.
4157 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
4158 struct inode *inode, struct inode *old_dir,
4159 struct dentry *parent)
4161 struct btrfs_root * root = BTRFS_I(inode)->root;
4164 * this will force the logging code to walk the dentry chain
4167 if (S_ISREG(inode->i_mode))
4168 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4171 * if this inode hasn't been logged and directory we're renaming it
4172 * from hasn't been logged, we don't need to log it
4174 if (BTRFS_I(inode)->logged_trans <=
4175 root->fs_info->last_trans_committed &&
4176 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
4177 root->fs_info->last_trans_committed))
4180 return btrfs_log_inode_parent(trans, root, inode, parent, 1);