]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - fs/btrfs/tree-log.c
Btrfs: only log the inode item if we can get away with it
[~shefty/rdma-dev.git] / fs / btrfs / tree-log.c
1 /*
2  * Copyright (C) 2008 Oracle.  All rights reserved.
3  *
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.
7  *
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.
12  *
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.
17  */
18
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/list_sort.h>
22 #include "ctree.h"
23 #include "transaction.h"
24 #include "disk-io.h"
25 #include "locking.h"
26 #include "print-tree.h"
27 #include "backref.h"
28 #include "compat.h"
29 #include "tree-log.h"
30 #include "hash.h"
31
32 /* magic values for the inode_only field in btrfs_log_inode:
33  *
34  * LOG_INODE_ALL means to log everything
35  * LOG_INODE_EXISTS means to log just enough to recreate the inode
36  * during log replay
37  */
38 #define LOG_INODE_ALL 0
39 #define LOG_INODE_EXISTS 1
40
41 /*
42  * directory trouble cases
43  *
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
48  *
49  * mkdir foo/some_dir
50  * normal commit
51  * rename foo/some_dir foo2/some_dir
52  * mkdir foo/some_dir
53  * fsync foo/some_dir/some_file
54  *
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
58  *
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.
61  *
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
65  *
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
68  *
69  * 3) after a crash, we must go through any directories with a link count
70  * of zero and redo the rm -rf
71  *
72  * mkdir f1/foo
73  * normal commit
74  * rm -rf f1/foo
75  * fsync(f1)
76  *
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
81  * ugly details.
82  */
83
84 /*
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.
89  *
90  * The last stage is to deal with directories and links and extents
91  * and all the other fun semantics
92  */
93 #define LOG_WALK_PIN_ONLY 0
94 #define LOG_WALK_REPLAY_INODES 1
95 #define LOG_WALK_REPLAY_ALL 2
96
97 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
98                              struct btrfs_root *root, struct inode *inode,
99                              int inode_only);
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);
108
109 /*
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.
112  *
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.
116  *
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.
122  *
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.
126  *
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.
130  */
131
132 /*
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
136  */
137 static int start_log_trans(struct btrfs_trans_handle *trans,
138                            struct btrfs_root *root)
139 {
140         int ret;
141         int err = 0;
142
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;
150                 }
151
152                 atomic_inc(&root->log_batch);
153                 atomic_inc(&root->log_writers);
154                 mutex_unlock(&root->log_mutex);
155                 return 0;
156         }
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);
162                 if (ret)
163                         err = ret;
164         }
165         if (err == 0 && !root->log_root) {
166                 ret = btrfs_add_log_tree(trans, root);
167                 if (ret)
168                         err = ret;
169         }
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);
174         return err;
175 }
176
177 /*
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
180  * in progress
181  */
182 static int join_running_log_trans(struct btrfs_root *root)
183 {
184         int ret = -ENOENT;
185
186         smp_mb();
187         if (!root->log_root)
188                 return -ENOENT;
189
190         mutex_lock(&root->log_mutex);
191         if (root->log_root) {
192                 ret = 0;
193                 atomic_inc(&root->log_writers);
194         }
195         mutex_unlock(&root->log_mutex);
196         return ret;
197 }
198
199 /*
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()
203  */
204 int btrfs_pin_log_trans(struct btrfs_root *root)
205 {
206         int ret = -ENOENT;
207
208         mutex_lock(&root->log_mutex);
209         atomic_inc(&root->log_writers);
210         mutex_unlock(&root->log_mutex);
211         return ret;
212 }
213
214 /*
215  * indicate we're done making changes to the log tree
216  * and wake up anyone waiting to do a sync
217  */
218 void btrfs_end_log_trans(struct btrfs_root *root)
219 {
220         if (atomic_dec_and_test(&root->log_writers)) {
221                 smp_mb();
222                 if (waitqueue_active(&root->log_writer_wait))
223                         wake_up(&root->log_writer_wait);
224         }
225 }
226
227
228 /*
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
233  */
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
237          */
238         int free;
239
240         /* should we write out the extent buffer?  This is used
241          * while flushing the log tree to disk during a sync
242          */
243         int write;
244
245         /* should we wait for the extent buffer io to finish?  Also used
246          * while flushing the log tree to disk for a sync
247          */
248         int wait;
249
250         /* pin only walk, we record which extents on disk belong to the
251          * log trees
252          */
253         int pin;
254
255         /* what stage of the replay code we're currently in */
256         int stage;
257
258         /* the root we are currently replaying */
259         struct btrfs_root *replay_dest;
260
261         /* the trans handle for the current replay */
262         struct btrfs_trans_handle *trans;
263
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
267          * inside it
268          */
269         int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
270                             struct walk_control *wc, u64 gen);
271 };
272
273 /*
274  * process_func used to pin down extents, write them or wait on them
275  */
276 static int process_one_buffer(struct btrfs_root *log,
277                               struct extent_buffer *eb,
278                               struct walk_control *wc, u64 gen)
279 {
280         if (wc->pin)
281                 btrfs_pin_extent_for_log_replay(wc->trans,
282                                                 log->fs_info->extent_root,
283                                                 eb->start, eb->len);
284
285         if (btrfs_buffer_uptodate(eb, gen, 0)) {
286                 if (wc->write)
287                         btrfs_write_tree_block(eb);
288                 if (wc->wait)
289                         btrfs_wait_tree_block_writeback(eb);
290         }
291         return 0;
292 }
293
294 /*
295  * Item overwrite used by replay and tree logging.  eb, slot and key all refer
296  * to the src data we are copying out.
297  *
298  * root is the tree we are copying into, and path is a scratch
299  * path for use in this function (it should be released on entry and
300  * will be released on exit).
301  *
302  * If the key is already in the destination tree the existing item is
303  * overwritten.  If the existing item isn't big enough, it is extended.
304  * If it is too large, it is truncated.
305  *
306  * If the key isn't in the destination yet, a new item is inserted.
307  */
308 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
309                                    struct btrfs_root *root,
310                                    struct btrfs_path *path,
311                                    struct extent_buffer *eb, int slot,
312                                    struct btrfs_key *key)
313 {
314         int ret;
315         u32 item_size;
316         u64 saved_i_size = 0;
317         int save_old_i_size = 0;
318         unsigned long src_ptr;
319         unsigned long dst_ptr;
320         int overwrite_root = 0;
321
322         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
323                 overwrite_root = 1;
324
325         item_size = btrfs_item_size_nr(eb, slot);
326         src_ptr = btrfs_item_ptr_offset(eb, slot);
327
328         /* look for the key in the destination tree */
329         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
330         if (ret == 0) {
331                 char *src_copy;
332                 char *dst_copy;
333                 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
334                                                   path->slots[0]);
335                 if (dst_size != item_size)
336                         goto insert;
337
338                 if (item_size == 0) {
339                         btrfs_release_path(path);
340                         return 0;
341                 }
342                 dst_copy = kmalloc(item_size, GFP_NOFS);
343                 src_copy = kmalloc(item_size, GFP_NOFS);
344                 if (!dst_copy || !src_copy) {
345                         btrfs_release_path(path);
346                         kfree(dst_copy);
347                         kfree(src_copy);
348                         return -ENOMEM;
349                 }
350
351                 read_extent_buffer(eb, src_copy, src_ptr, item_size);
352
353                 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
354                 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
355                                    item_size);
356                 ret = memcmp(dst_copy, src_copy, item_size);
357
358                 kfree(dst_copy);
359                 kfree(src_copy);
360                 /*
361                  * they have the same contents, just return, this saves
362                  * us from cowing blocks in the destination tree and doing
363                  * extra writes that may not have been done by a previous
364                  * sync
365                  */
366                 if (ret == 0) {
367                         btrfs_release_path(path);
368                         return 0;
369                 }
370
371         }
372 insert:
373         btrfs_release_path(path);
374         /* try to insert the key into the destination tree */
375         ret = btrfs_insert_empty_item(trans, root, path,
376                                       key, item_size);
377
378         /* make sure any existing item is the correct size */
379         if (ret == -EEXIST) {
380                 u32 found_size;
381                 found_size = btrfs_item_size_nr(path->nodes[0],
382                                                 path->slots[0]);
383                 if (found_size > item_size)
384                         btrfs_truncate_item(trans, root, path, item_size, 1);
385                 else if (found_size < item_size)
386                         btrfs_extend_item(trans, root, path,
387                                           item_size - found_size);
388         } else if (ret) {
389                 return ret;
390         }
391         dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
392                                         path->slots[0]);
393
394         /* don't overwrite an existing inode if the generation number
395          * was logged as zero.  This is done when the tree logging code
396          * is just logging an inode to make sure it exists after recovery.
397          *
398          * Also, don't overwrite i_size on directories during replay.
399          * log replay inserts and removes directory items based on the
400          * state of the tree found in the subvolume, and i_size is modified
401          * as it goes
402          */
403         if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
404                 struct btrfs_inode_item *src_item;
405                 struct btrfs_inode_item *dst_item;
406
407                 src_item = (struct btrfs_inode_item *)src_ptr;
408                 dst_item = (struct btrfs_inode_item *)dst_ptr;
409
410                 if (btrfs_inode_generation(eb, src_item) == 0)
411                         goto no_copy;
412
413                 if (overwrite_root &&
414                     S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
415                     S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
416                         save_old_i_size = 1;
417                         saved_i_size = btrfs_inode_size(path->nodes[0],
418                                                         dst_item);
419                 }
420         }
421
422         copy_extent_buffer(path->nodes[0], eb, dst_ptr,
423                            src_ptr, item_size);
424
425         if (save_old_i_size) {
426                 struct btrfs_inode_item *dst_item;
427                 dst_item = (struct btrfs_inode_item *)dst_ptr;
428                 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
429         }
430
431         /* make sure the generation is filled in */
432         if (key->type == BTRFS_INODE_ITEM_KEY) {
433                 struct btrfs_inode_item *dst_item;
434                 dst_item = (struct btrfs_inode_item *)dst_ptr;
435                 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
436                         btrfs_set_inode_generation(path->nodes[0], dst_item,
437                                                    trans->transid);
438                 }
439         }
440 no_copy:
441         btrfs_mark_buffer_dirty(path->nodes[0]);
442         btrfs_release_path(path);
443         return 0;
444 }
445
446 /*
447  * simple helper to read an inode off the disk from a given root
448  * This can only be called for subvolume roots and not for the log
449  */
450 static noinline struct inode *read_one_inode(struct btrfs_root *root,
451                                              u64 objectid)
452 {
453         struct btrfs_key key;
454         struct inode *inode;
455
456         key.objectid = objectid;
457         key.type = BTRFS_INODE_ITEM_KEY;
458         key.offset = 0;
459         inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
460         if (IS_ERR(inode)) {
461                 inode = NULL;
462         } else if (is_bad_inode(inode)) {
463                 iput(inode);
464                 inode = NULL;
465         }
466         return inode;
467 }
468
469 /* replays a single extent in 'eb' at 'slot' with 'key' into the
470  * subvolume 'root'.  path is released on entry and should be released
471  * on exit.
472  *
473  * extents in the log tree have not been allocated out of the extent
474  * tree yet.  So, this completes the allocation, taking a reference
475  * as required if the extent already exists or creating a new extent
476  * if it isn't in the extent allocation tree yet.
477  *
478  * The extent is inserted into the file, dropping any existing extents
479  * from the file that overlap the new one.
480  */
481 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
482                                       struct btrfs_root *root,
483                                       struct btrfs_path *path,
484                                       struct extent_buffer *eb, int slot,
485                                       struct btrfs_key *key)
486 {
487         int found_type;
488         u64 mask = root->sectorsize - 1;
489         u64 extent_end;
490         u64 start = key->offset;
491         u64 saved_nbytes;
492         struct btrfs_file_extent_item *item;
493         struct inode *inode = NULL;
494         unsigned long size;
495         int ret = 0;
496
497         item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
498         found_type = btrfs_file_extent_type(eb, item);
499
500         if (found_type == BTRFS_FILE_EXTENT_REG ||
501             found_type == BTRFS_FILE_EXTENT_PREALLOC)
502                 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
503         else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
504                 size = btrfs_file_extent_inline_len(eb, item);
505                 extent_end = (start + size + mask) & ~mask;
506         } else {
507                 ret = 0;
508                 goto out;
509         }
510
511         inode = read_one_inode(root, key->objectid);
512         if (!inode) {
513                 ret = -EIO;
514                 goto out;
515         }
516
517         /*
518          * first check to see if we already have this extent in the
519          * file.  This must be done before the btrfs_drop_extents run
520          * so we don't try to drop this extent.
521          */
522         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
523                                        start, 0);
524
525         if (ret == 0 &&
526             (found_type == BTRFS_FILE_EXTENT_REG ||
527              found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
528                 struct btrfs_file_extent_item cmp1;
529                 struct btrfs_file_extent_item cmp2;
530                 struct btrfs_file_extent_item *existing;
531                 struct extent_buffer *leaf;
532
533                 leaf = path->nodes[0];
534                 existing = btrfs_item_ptr(leaf, path->slots[0],
535                                           struct btrfs_file_extent_item);
536
537                 read_extent_buffer(eb, &cmp1, (unsigned long)item,
538                                    sizeof(cmp1));
539                 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
540                                    sizeof(cmp2));
541
542                 /*
543                  * we already have a pointer to this exact extent,
544                  * we don't have to do anything
545                  */
546                 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
547                         btrfs_release_path(path);
548                         goto out;
549                 }
550         }
551         btrfs_release_path(path);
552
553         saved_nbytes = inode_get_bytes(inode);
554         /* drop any overlapping extents */
555         ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
556         BUG_ON(ret);
557
558         if (found_type == BTRFS_FILE_EXTENT_REG ||
559             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
560                 u64 offset;
561                 unsigned long dest_offset;
562                 struct btrfs_key ins;
563
564                 ret = btrfs_insert_empty_item(trans, root, path, key,
565                                               sizeof(*item));
566                 BUG_ON(ret);
567                 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
568                                                     path->slots[0]);
569                 copy_extent_buffer(path->nodes[0], eb, dest_offset,
570                                 (unsigned long)item,  sizeof(*item));
571
572                 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
573                 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
574                 ins.type = BTRFS_EXTENT_ITEM_KEY;
575                 offset = key->offset - btrfs_file_extent_offset(eb, item);
576
577                 if (ins.objectid > 0) {
578                         u64 csum_start;
579                         u64 csum_end;
580                         LIST_HEAD(ordered_sums);
581                         /*
582                          * is this extent already allocated in the extent
583                          * allocation tree?  If so, just add a reference
584                          */
585                         ret = btrfs_lookup_extent(root, ins.objectid,
586                                                 ins.offset);
587                         if (ret == 0) {
588                                 ret = btrfs_inc_extent_ref(trans, root,
589                                                 ins.objectid, ins.offset,
590                                                 0, root->root_key.objectid,
591                                                 key->objectid, offset, 0);
592                                 BUG_ON(ret);
593                         } else {
594                                 /*
595                                  * insert the extent pointer in the extent
596                                  * allocation tree
597                                  */
598                                 ret = btrfs_alloc_logged_file_extent(trans,
599                                                 root, root->root_key.objectid,
600                                                 key->objectid, offset, &ins);
601                                 BUG_ON(ret);
602                         }
603                         btrfs_release_path(path);
604
605                         if (btrfs_file_extent_compression(eb, item)) {
606                                 csum_start = ins.objectid;
607                                 csum_end = csum_start + ins.offset;
608                         } else {
609                                 csum_start = ins.objectid +
610                                         btrfs_file_extent_offset(eb, item);
611                                 csum_end = csum_start +
612                                         btrfs_file_extent_num_bytes(eb, item);
613                         }
614
615                         ret = btrfs_lookup_csums_range(root->log_root,
616                                                 csum_start, csum_end - 1,
617                                                 &ordered_sums, 0);
618                         BUG_ON(ret);
619                         while (!list_empty(&ordered_sums)) {
620                                 struct btrfs_ordered_sum *sums;
621                                 sums = list_entry(ordered_sums.next,
622                                                 struct btrfs_ordered_sum,
623                                                 list);
624                                 ret = btrfs_csum_file_blocks(trans,
625                                                 root->fs_info->csum_root,
626                                                 sums);
627                                 BUG_ON(ret);
628                                 list_del(&sums->list);
629                                 kfree(sums);
630                         }
631                 } else {
632                         btrfs_release_path(path);
633                 }
634         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
635                 /* inline extents are easy, we just overwrite them */
636                 ret = overwrite_item(trans, root, path, eb, slot, key);
637                 BUG_ON(ret);
638         }
639
640         inode_set_bytes(inode, saved_nbytes);
641         ret = btrfs_update_inode(trans, root, inode);
642 out:
643         if (inode)
644                 iput(inode);
645         return ret;
646 }
647
648 /*
649  * when cleaning up conflicts between the directory names in the
650  * subvolume, directory names in the log and directory names in the
651  * inode back references, we may have to unlink inodes from directories.
652  *
653  * This is a helper function to do the unlink of a specific directory
654  * item
655  */
656 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
657                                       struct btrfs_root *root,
658                                       struct btrfs_path *path,
659                                       struct inode *dir,
660                                       struct btrfs_dir_item *di)
661 {
662         struct inode *inode;
663         char *name;
664         int name_len;
665         struct extent_buffer *leaf;
666         struct btrfs_key location;
667         int ret;
668
669         leaf = path->nodes[0];
670
671         btrfs_dir_item_key_to_cpu(leaf, di, &location);
672         name_len = btrfs_dir_name_len(leaf, di);
673         name = kmalloc(name_len, GFP_NOFS);
674         if (!name)
675                 return -ENOMEM;
676
677         read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
678         btrfs_release_path(path);
679
680         inode = read_one_inode(root, location.objectid);
681         if (!inode) {
682                 kfree(name);
683                 return -EIO;
684         }
685
686         ret = link_to_fixup_dir(trans, root, path, location.objectid);
687         BUG_ON(ret);
688
689         ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
690         BUG_ON(ret);
691         kfree(name);
692
693         iput(inode);
694
695         btrfs_run_delayed_items(trans, root);
696         return ret;
697 }
698
699 /*
700  * helper function to see if a given name and sequence number found
701  * in an inode back reference are already in a directory and correctly
702  * point to this inode
703  */
704 static noinline int inode_in_dir(struct btrfs_root *root,
705                                  struct btrfs_path *path,
706                                  u64 dirid, u64 objectid, u64 index,
707                                  const char *name, int name_len)
708 {
709         struct btrfs_dir_item *di;
710         struct btrfs_key location;
711         int match = 0;
712
713         di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
714                                          index, name, name_len, 0);
715         if (di && !IS_ERR(di)) {
716                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
717                 if (location.objectid != objectid)
718                         goto out;
719         } else
720                 goto out;
721         btrfs_release_path(path);
722
723         di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
724         if (di && !IS_ERR(di)) {
725                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
726                 if (location.objectid != objectid)
727                         goto out;
728         } else
729                 goto out;
730         match = 1;
731 out:
732         btrfs_release_path(path);
733         return match;
734 }
735
736 /*
737  * helper function to check a log tree for a named back reference in
738  * an inode.  This is used to decide if a back reference that is
739  * found in the subvolume conflicts with what we find in the log.
740  *
741  * inode backreferences may have multiple refs in a single item,
742  * during replay we process one reference at a time, and we don't
743  * want to delete valid links to a file from the subvolume if that
744  * link is also in the log.
745  */
746 static noinline int backref_in_log(struct btrfs_root *log,
747                                    struct btrfs_key *key,
748                                    u64 ref_objectid,
749                                    char *name, int namelen)
750 {
751         struct btrfs_path *path;
752         struct btrfs_inode_ref *ref;
753         unsigned long ptr;
754         unsigned long ptr_end;
755         unsigned long name_ptr;
756         int found_name_len;
757         int item_size;
758         int ret;
759         int match = 0;
760
761         path = btrfs_alloc_path();
762         if (!path)
763                 return -ENOMEM;
764
765         ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
766         if (ret != 0)
767                 goto out;
768
769         ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
770
771         if (key->type == BTRFS_INODE_EXTREF_KEY) {
772                 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
773                                                    name, namelen, NULL))
774                         match = 1;
775
776                 goto out;
777         }
778
779         item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
780         ptr_end = ptr + item_size;
781         while (ptr < ptr_end) {
782                 ref = (struct btrfs_inode_ref *)ptr;
783                 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
784                 if (found_name_len == namelen) {
785                         name_ptr = (unsigned long)(ref + 1);
786                         ret = memcmp_extent_buffer(path->nodes[0], name,
787                                                    name_ptr, namelen);
788                         if (ret == 0) {
789                                 match = 1;
790                                 goto out;
791                         }
792                 }
793                 ptr = (unsigned long)(ref + 1) + found_name_len;
794         }
795 out:
796         btrfs_free_path(path);
797         return match;
798 }
799
800 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
801                                   struct btrfs_root *root,
802                                   struct btrfs_path *path,
803                                   struct btrfs_root *log_root,
804                                   struct inode *dir, struct inode *inode,
805                                   struct extent_buffer *eb,
806                                   u64 inode_objectid, u64 parent_objectid,
807                                   u64 ref_index, char *name, int namelen,
808                                   int *search_done)
809 {
810         int ret;
811         char *victim_name;
812         int victim_name_len;
813         struct extent_buffer *leaf;
814         struct btrfs_dir_item *di;
815         struct btrfs_key search_key;
816         struct btrfs_inode_extref *extref;
817
818 again:
819         /* Search old style refs */
820         search_key.objectid = inode_objectid;
821         search_key.type = BTRFS_INODE_REF_KEY;
822         search_key.offset = parent_objectid;
823         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
824         if (ret == 0) {
825                 struct btrfs_inode_ref *victim_ref;
826                 unsigned long ptr;
827                 unsigned long ptr_end;
828
829                 leaf = path->nodes[0];
830
831                 /* are we trying to overwrite a back ref for the root directory
832                  * if so, just jump out, we're done
833                  */
834                 if (search_key.objectid == search_key.offset)
835                         return 1;
836
837                 /* check all the names in this back reference to see
838                  * if they are in the log.  if so, we allow them to stay
839                  * otherwise they must be unlinked as a conflict
840                  */
841                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
842                 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
843                 while (ptr < ptr_end) {
844                         victim_ref = (struct btrfs_inode_ref *)ptr;
845                         victim_name_len = btrfs_inode_ref_name_len(leaf,
846                                                                    victim_ref);
847                         victim_name = kmalloc(victim_name_len, GFP_NOFS);
848                         BUG_ON(!victim_name);
849
850                         read_extent_buffer(leaf, victim_name,
851                                            (unsigned long)(victim_ref + 1),
852                                            victim_name_len);
853
854                         if (!backref_in_log(log_root, &search_key,
855                                             parent_objectid,
856                                             victim_name,
857                                             victim_name_len)) {
858                                 btrfs_inc_nlink(inode);
859                                 btrfs_release_path(path);
860
861                                 ret = btrfs_unlink_inode(trans, root, dir,
862                                                          inode, victim_name,
863                                                          victim_name_len);
864                                 BUG_ON(ret);
865                                 btrfs_run_delayed_items(trans, root);
866                                 kfree(victim_name);
867                                 *search_done = 1;
868                                 goto again;
869                         }
870                         kfree(victim_name);
871
872                         ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
873                 }
874                 BUG_ON(ret);
875
876                 /*
877                  * NOTE: we have searched root tree and checked the
878                  * coresponding ref, it does not need to check again.
879                  */
880                 *search_done = 1;
881         }
882         btrfs_release_path(path);
883
884         /* Same search but for extended refs */
885         extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
886                                            inode_objectid, parent_objectid, 0,
887                                            0);
888         if (!IS_ERR_OR_NULL(extref)) {
889                 u32 item_size;
890                 u32 cur_offset = 0;
891                 unsigned long base;
892                 struct inode *victim_parent;
893
894                 leaf = path->nodes[0];
895
896                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
897                 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
898
899                 while (cur_offset < item_size) {
900                         extref = (struct btrfs_inode_extref *)base + cur_offset;
901
902                         victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
903
904                         if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
905                                 goto next;
906
907                         victim_name = kmalloc(victim_name_len, GFP_NOFS);
908                         read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
909                                            victim_name_len);
910
911                         search_key.objectid = inode_objectid;
912                         search_key.type = BTRFS_INODE_EXTREF_KEY;
913                         search_key.offset = btrfs_extref_hash(parent_objectid,
914                                                               victim_name,
915                                                               victim_name_len);
916                         ret = 0;
917                         if (!backref_in_log(log_root, &search_key,
918                                             parent_objectid, victim_name,
919                                             victim_name_len)) {
920                                 ret = -ENOENT;
921                                 victim_parent = read_one_inode(root,
922                                                                parent_objectid);
923                                 if (victim_parent) {
924                                         btrfs_inc_nlink(inode);
925                                         btrfs_release_path(path);
926
927                                         ret = btrfs_unlink_inode(trans, root,
928                                                                  victim_parent,
929                                                                  inode,
930                                                                  victim_name,
931                                                                  victim_name_len);
932                                         btrfs_run_delayed_items(trans, root);
933                                 }
934                                 BUG_ON(ret);
935                                 iput(victim_parent);
936                                 kfree(victim_name);
937                                 *search_done = 1;
938                                 goto again;
939                         }
940                         kfree(victim_name);
941                         BUG_ON(ret);
942 next:
943                         cur_offset += victim_name_len + sizeof(*extref);
944                 }
945                 *search_done = 1;
946         }
947         btrfs_release_path(path);
948
949         /* look for a conflicting sequence number */
950         di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
951                                          ref_index, name, namelen, 0);
952         if (di && !IS_ERR(di)) {
953                 ret = drop_one_dir_item(trans, root, path, dir, di);
954                 BUG_ON(ret);
955         }
956         btrfs_release_path(path);
957
958         /* look for a conflicing name */
959         di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
960                                    name, namelen, 0);
961         if (di && !IS_ERR(di)) {
962                 ret = drop_one_dir_item(trans, root, path, dir, di);
963                 BUG_ON(ret);
964         }
965         btrfs_release_path(path);
966
967         return 0;
968 }
969
970 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
971                              u32 *namelen, char **name, u64 *index,
972                              u64 *parent_objectid)
973 {
974         struct btrfs_inode_extref *extref;
975
976         extref = (struct btrfs_inode_extref *)ref_ptr;
977
978         *namelen = btrfs_inode_extref_name_len(eb, extref);
979         *name = kmalloc(*namelen, GFP_NOFS);
980         if (*name == NULL)
981                 return -ENOMEM;
982
983         read_extent_buffer(eb, *name, (unsigned long)&extref->name,
984                            *namelen);
985
986         *index = btrfs_inode_extref_index(eb, extref);
987         if (parent_objectid)
988                 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
989
990         return 0;
991 }
992
993 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
994                           u32 *namelen, char **name, u64 *index)
995 {
996         struct btrfs_inode_ref *ref;
997
998         ref = (struct btrfs_inode_ref *)ref_ptr;
999
1000         *namelen = btrfs_inode_ref_name_len(eb, ref);
1001         *name = kmalloc(*namelen, GFP_NOFS);
1002         if (*name == NULL)
1003                 return -ENOMEM;
1004
1005         read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1006
1007         *index = btrfs_inode_ref_index(eb, ref);
1008
1009         return 0;
1010 }
1011
1012 /*
1013  * replay one inode back reference item found in the log tree.
1014  * eb, slot and key refer to the buffer and key found in the log tree.
1015  * root is the destination we are replaying into, and path is for temp
1016  * use by this function.  (it should be released on return).
1017  */
1018 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1019                                   struct btrfs_root *root,
1020                                   struct btrfs_root *log,
1021                                   struct btrfs_path *path,
1022                                   struct extent_buffer *eb, int slot,
1023                                   struct btrfs_key *key)
1024 {
1025         struct inode *dir;
1026         struct inode *inode;
1027         unsigned long ref_ptr;
1028         unsigned long ref_end;
1029         char *name;
1030         int namelen;
1031         int ret;
1032         int search_done = 0;
1033         int log_ref_ver = 0;
1034         u64 parent_objectid;
1035         u64 inode_objectid;
1036         u64 ref_index = 0;
1037         int ref_struct_size;
1038
1039         ref_ptr = btrfs_item_ptr_offset(eb, slot);
1040         ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1041
1042         if (key->type == BTRFS_INODE_EXTREF_KEY) {
1043                 struct btrfs_inode_extref *r;
1044
1045                 ref_struct_size = sizeof(struct btrfs_inode_extref);
1046                 log_ref_ver = 1;
1047                 r = (struct btrfs_inode_extref *)ref_ptr;
1048                 parent_objectid = btrfs_inode_extref_parent(eb, r);
1049         } else {
1050                 ref_struct_size = sizeof(struct btrfs_inode_ref);
1051                 parent_objectid = key->offset;
1052         }
1053         inode_objectid = key->objectid;
1054
1055         /*
1056          * it is possible that we didn't log all the parent directories
1057          * for a given inode.  If we don't find the dir, just don't
1058          * copy the back ref in.  The link count fixup code will take
1059          * care of the rest
1060          */
1061         dir = read_one_inode(root, parent_objectid);
1062         if (!dir)
1063                 return -ENOENT;
1064
1065         inode = read_one_inode(root, inode_objectid);
1066         if (!inode) {
1067                 iput(dir);
1068                 return -EIO;
1069         }
1070
1071         while (ref_ptr < ref_end) {
1072                 if (log_ref_ver) {
1073                         ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1074                                                 &ref_index, &parent_objectid);
1075                         /*
1076                          * parent object can change from one array
1077                          * item to another.
1078                          */
1079                         if (!dir)
1080                                 dir = read_one_inode(root, parent_objectid);
1081                         if (!dir)
1082                                 return -ENOENT;
1083                 } else {
1084                         ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1085                                              &ref_index);
1086                 }
1087                 if (ret)
1088                         return ret;
1089
1090                 /* if we already have a perfect match, we're done */
1091                 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
1092                                   ref_index, name, namelen)) {
1093                         /*
1094                          * look for a conflicting back reference in the
1095                          * metadata. if we find one we have to unlink that name
1096                          * of the file before we add our new link.  Later on, we
1097                          * overwrite any existing back reference, and we don't
1098                          * want to create dangling pointers in the directory.
1099                          */
1100
1101                         if (!search_done) {
1102                                 ret = __add_inode_ref(trans, root, path, log,
1103                                                       dir, inode, eb,
1104                                                       inode_objectid,
1105                                                       parent_objectid,
1106                                                       ref_index, name, namelen,
1107                                                       &search_done);
1108                                 if (ret == 1)
1109                                         goto out;
1110                                 BUG_ON(ret);
1111                         }
1112
1113                         /* insert our name */
1114                         ret = btrfs_add_link(trans, dir, inode, name, namelen,
1115                                              0, ref_index);
1116                         BUG_ON(ret);
1117
1118                         btrfs_update_inode(trans, root, inode);
1119                 }
1120
1121                 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1122                 kfree(name);
1123                 if (log_ref_ver) {
1124                         iput(dir);
1125                         dir = NULL;
1126                 }
1127         }
1128
1129         /* finally write the back reference in the inode */
1130         ret = overwrite_item(trans, root, path, eb, slot, key);
1131         BUG_ON(ret);
1132
1133 out:
1134         btrfs_release_path(path);
1135         iput(dir);
1136         iput(inode);
1137         return 0;
1138 }
1139
1140 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1141                               struct btrfs_root *root, u64 offset)
1142 {
1143         int ret;
1144         ret = btrfs_find_orphan_item(root, offset);
1145         if (ret > 0)
1146                 ret = btrfs_insert_orphan_item(trans, root, offset);
1147         return ret;
1148 }
1149
1150 static int count_inode_extrefs(struct btrfs_root *root,
1151                                struct inode *inode, struct btrfs_path *path)
1152 {
1153         int ret = 0;
1154         int name_len;
1155         unsigned int nlink = 0;
1156         u32 item_size;
1157         u32 cur_offset = 0;
1158         u64 inode_objectid = btrfs_ino(inode);
1159         u64 offset = 0;
1160         unsigned long ptr;
1161         struct btrfs_inode_extref *extref;
1162         struct extent_buffer *leaf;
1163
1164         while (1) {
1165                 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1166                                             &extref, &offset);
1167                 if (ret)
1168                         break;
1169
1170                 leaf = path->nodes[0];
1171                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1172                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1173
1174                 while (cur_offset < item_size) {
1175                         extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1176                         name_len = btrfs_inode_extref_name_len(leaf, extref);
1177
1178                         nlink++;
1179
1180                         cur_offset += name_len + sizeof(*extref);
1181                 }
1182
1183                 offset++;
1184                 btrfs_release_path(path);
1185         }
1186         btrfs_release_path(path);
1187
1188         if (ret < 0)
1189                 return ret;
1190         return nlink;
1191 }
1192
1193 static int count_inode_refs(struct btrfs_root *root,
1194                                struct inode *inode, struct btrfs_path *path)
1195 {
1196         int ret;
1197         struct btrfs_key key;
1198         unsigned int nlink = 0;
1199         unsigned long ptr;
1200         unsigned long ptr_end;
1201         int name_len;
1202         u64 ino = btrfs_ino(inode);
1203
1204         key.objectid = ino;
1205         key.type = BTRFS_INODE_REF_KEY;
1206         key.offset = (u64)-1;
1207
1208         while (1) {
1209                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1210                 if (ret < 0)
1211                         break;
1212                 if (ret > 0) {
1213                         if (path->slots[0] == 0)
1214                                 break;
1215                         path->slots[0]--;
1216                 }
1217                 btrfs_item_key_to_cpu(path->nodes[0], &key,
1218                                       path->slots[0]);
1219                 if (key.objectid != ino ||
1220                     key.type != BTRFS_INODE_REF_KEY)
1221                         break;
1222                 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1223                 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1224                                                    path->slots[0]);
1225                 while (ptr < ptr_end) {
1226                         struct btrfs_inode_ref *ref;
1227
1228                         ref = (struct btrfs_inode_ref *)ptr;
1229                         name_len = btrfs_inode_ref_name_len(path->nodes[0],
1230                                                             ref);
1231                         ptr = (unsigned long)(ref + 1) + name_len;
1232                         nlink++;
1233                 }
1234
1235                 if (key.offset == 0)
1236                         break;
1237                 key.offset--;
1238                 btrfs_release_path(path);
1239         }
1240         btrfs_release_path(path);
1241
1242         return nlink;
1243 }
1244
1245 /*
1246  * There are a few corners where the link count of the file can't
1247  * be properly maintained during replay.  So, instead of adding
1248  * lots of complexity to the log code, we just scan the backrefs
1249  * for any file that has been through replay.
1250  *
1251  * The scan will update the link count on the inode to reflect the
1252  * number of back refs found.  If it goes down to zero, the iput
1253  * will free the inode.
1254  */
1255 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1256                                            struct btrfs_root *root,
1257                                            struct inode *inode)
1258 {
1259         struct btrfs_path *path;
1260         int ret;
1261         u64 nlink = 0;
1262         u64 ino = btrfs_ino(inode);
1263
1264         path = btrfs_alloc_path();
1265         if (!path)
1266                 return -ENOMEM;
1267
1268         ret = count_inode_refs(root, inode, path);
1269         if (ret < 0)
1270                 goto out;
1271
1272         nlink = ret;
1273
1274         ret = count_inode_extrefs(root, inode, path);
1275         if (ret == -ENOENT)
1276                 ret = 0;
1277
1278         if (ret < 0)
1279                 goto out;
1280
1281         nlink += ret;
1282
1283         ret = 0;
1284
1285         if (nlink != inode->i_nlink) {
1286                 set_nlink(inode, nlink);
1287                 btrfs_update_inode(trans, root, inode);
1288         }
1289         BTRFS_I(inode)->index_cnt = (u64)-1;
1290
1291         if (inode->i_nlink == 0) {
1292                 if (S_ISDIR(inode->i_mode)) {
1293                         ret = replay_dir_deletes(trans, root, NULL, path,
1294                                                  ino, 1);
1295                         BUG_ON(ret);
1296                 }
1297                 ret = insert_orphan_item(trans, root, ino);
1298                 BUG_ON(ret);
1299         }
1300
1301 out:
1302         btrfs_free_path(path);
1303         return ret;
1304 }
1305
1306 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1307                                             struct btrfs_root *root,
1308                                             struct btrfs_path *path)
1309 {
1310         int ret;
1311         struct btrfs_key key;
1312         struct inode *inode;
1313
1314         key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1315         key.type = BTRFS_ORPHAN_ITEM_KEY;
1316         key.offset = (u64)-1;
1317         while (1) {
1318                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1319                 if (ret < 0)
1320                         break;
1321
1322                 if (ret == 1) {
1323                         if (path->slots[0] == 0)
1324                                 break;
1325                         path->slots[0]--;
1326                 }
1327
1328                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1329                 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1330                     key.type != BTRFS_ORPHAN_ITEM_KEY)
1331                         break;
1332
1333                 ret = btrfs_del_item(trans, root, path);
1334                 if (ret)
1335                         goto out;
1336
1337                 btrfs_release_path(path);
1338                 inode = read_one_inode(root, key.offset);
1339                 if (!inode)
1340                         return -EIO;
1341
1342                 ret = fixup_inode_link_count(trans, root, inode);
1343                 BUG_ON(ret);
1344
1345                 iput(inode);
1346
1347                 /*
1348                  * fixup on a directory may create new entries,
1349                  * make sure we always look for the highset possible
1350                  * offset
1351                  */
1352                 key.offset = (u64)-1;
1353         }
1354         ret = 0;
1355 out:
1356         btrfs_release_path(path);
1357         return ret;
1358 }
1359
1360
1361 /*
1362  * record a given inode in the fixup dir so we can check its link
1363  * count when replay is done.  The link count is incremented here
1364  * so the inode won't go away until we check it
1365  */
1366 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1367                                       struct btrfs_root *root,
1368                                       struct btrfs_path *path,
1369                                       u64 objectid)
1370 {
1371         struct btrfs_key key;
1372         int ret = 0;
1373         struct inode *inode;
1374
1375         inode = read_one_inode(root, objectid);
1376         if (!inode)
1377                 return -EIO;
1378
1379         key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1380         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1381         key.offset = objectid;
1382
1383         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1384
1385         btrfs_release_path(path);
1386         if (ret == 0) {
1387                 btrfs_inc_nlink(inode);
1388                 ret = btrfs_update_inode(trans, root, inode);
1389         } else if (ret == -EEXIST) {
1390                 ret = 0;
1391         } else {
1392                 BUG();
1393         }
1394         iput(inode);
1395
1396         return ret;
1397 }
1398
1399 /*
1400  * when replaying the log for a directory, we only insert names
1401  * for inodes that actually exist.  This means an fsync on a directory
1402  * does not implicitly fsync all the new files in it
1403  */
1404 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1405                                     struct btrfs_root *root,
1406                                     struct btrfs_path *path,
1407                                     u64 dirid, u64 index,
1408                                     char *name, int name_len, u8 type,
1409                                     struct btrfs_key *location)
1410 {
1411         struct inode *inode;
1412         struct inode *dir;
1413         int ret;
1414
1415         inode = read_one_inode(root, location->objectid);
1416         if (!inode)
1417                 return -ENOENT;
1418
1419         dir = read_one_inode(root, dirid);
1420         if (!dir) {
1421                 iput(inode);
1422                 return -EIO;
1423         }
1424         ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1425
1426         /* FIXME, put inode into FIXUP list */
1427
1428         iput(inode);
1429         iput(dir);
1430         return ret;
1431 }
1432
1433 /*
1434  * take a single entry in a log directory item and replay it into
1435  * the subvolume.
1436  *
1437  * if a conflicting item exists in the subdirectory already,
1438  * the inode it points to is unlinked and put into the link count
1439  * fix up tree.
1440  *
1441  * If a name from the log points to a file or directory that does
1442  * not exist in the FS, it is skipped.  fsyncs on directories
1443  * do not force down inodes inside that directory, just changes to the
1444  * names or unlinks in a directory.
1445  */
1446 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1447                                     struct btrfs_root *root,
1448                                     struct btrfs_path *path,
1449                                     struct extent_buffer *eb,
1450                                     struct btrfs_dir_item *di,
1451                                     struct btrfs_key *key)
1452 {
1453         char *name;
1454         int name_len;
1455         struct btrfs_dir_item *dst_di;
1456         struct btrfs_key found_key;
1457         struct btrfs_key log_key;
1458         struct inode *dir;
1459         u8 log_type;
1460         int exists;
1461         int ret;
1462
1463         dir = read_one_inode(root, key->objectid);
1464         if (!dir)
1465                 return -EIO;
1466
1467         name_len = btrfs_dir_name_len(eb, di);
1468         name = kmalloc(name_len, GFP_NOFS);
1469         if (!name)
1470                 return -ENOMEM;
1471
1472         log_type = btrfs_dir_type(eb, di);
1473         read_extent_buffer(eb, name, (unsigned long)(di + 1),
1474                    name_len);
1475
1476         btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1477         exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1478         if (exists == 0)
1479                 exists = 1;
1480         else
1481                 exists = 0;
1482         btrfs_release_path(path);
1483
1484         if (key->type == BTRFS_DIR_ITEM_KEY) {
1485                 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1486                                        name, name_len, 1);
1487         } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1488                 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1489                                                      key->objectid,
1490                                                      key->offset, name,
1491                                                      name_len, 1);
1492         } else {
1493                 BUG();
1494         }
1495         if (IS_ERR_OR_NULL(dst_di)) {
1496                 /* we need a sequence number to insert, so we only
1497                  * do inserts for the BTRFS_DIR_INDEX_KEY types
1498                  */
1499                 if (key->type != BTRFS_DIR_INDEX_KEY)
1500                         goto out;
1501                 goto insert;
1502         }
1503
1504         btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1505         /* the existing item matches the logged item */
1506         if (found_key.objectid == log_key.objectid &&
1507             found_key.type == log_key.type &&
1508             found_key.offset == log_key.offset &&
1509             btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1510                 goto out;
1511         }
1512
1513         /*
1514          * don't drop the conflicting directory entry if the inode
1515          * for the new entry doesn't exist
1516          */
1517         if (!exists)
1518                 goto out;
1519
1520         ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1521         BUG_ON(ret);
1522
1523         if (key->type == BTRFS_DIR_INDEX_KEY)
1524                 goto insert;
1525 out:
1526         btrfs_release_path(path);
1527         kfree(name);
1528         iput(dir);
1529         return 0;
1530
1531 insert:
1532         btrfs_release_path(path);
1533         ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1534                               name, name_len, log_type, &log_key);
1535
1536         BUG_ON(ret && ret != -ENOENT);
1537         goto out;
1538 }
1539
1540 /*
1541  * find all the names in a directory item and reconcile them into
1542  * the subvolume.  Only BTRFS_DIR_ITEM_KEY types will have more than
1543  * one name in a directory item, but the same code gets used for
1544  * both directory index types
1545  */
1546 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1547                                         struct btrfs_root *root,
1548                                         struct btrfs_path *path,
1549                                         struct extent_buffer *eb, int slot,
1550                                         struct btrfs_key *key)
1551 {
1552         int ret;
1553         u32 item_size = btrfs_item_size_nr(eb, slot);
1554         struct btrfs_dir_item *di;
1555         int name_len;
1556         unsigned long ptr;
1557         unsigned long ptr_end;
1558
1559         ptr = btrfs_item_ptr_offset(eb, slot);
1560         ptr_end = ptr + item_size;
1561         while (ptr < ptr_end) {
1562                 di = (struct btrfs_dir_item *)ptr;
1563                 if (verify_dir_item(root, eb, di))
1564                         return -EIO;
1565                 name_len = btrfs_dir_name_len(eb, di);
1566                 ret = replay_one_name(trans, root, path, eb, di, key);
1567                 BUG_ON(ret);
1568                 ptr = (unsigned long)(di + 1);
1569                 ptr += name_len;
1570         }
1571         return 0;
1572 }
1573
1574 /*
1575  * directory replay has two parts.  There are the standard directory
1576  * items in the log copied from the subvolume, and range items
1577  * created in the log while the subvolume was logged.
1578  *
1579  * The range items tell us which parts of the key space the log
1580  * is authoritative for.  During replay, if a key in the subvolume
1581  * directory is in a logged range item, but not actually in the log
1582  * that means it was deleted from the directory before the fsync
1583  * and should be removed.
1584  */
1585 static noinline int find_dir_range(struct btrfs_root *root,
1586                                    struct btrfs_path *path,
1587                                    u64 dirid, int key_type,
1588                                    u64 *start_ret, u64 *end_ret)
1589 {
1590         struct btrfs_key key;
1591         u64 found_end;
1592         struct btrfs_dir_log_item *item;
1593         int ret;
1594         int nritems;
1595
1596         if (*start_ret == (u64)-1)
1597                 return 1;
1598
1599         key.objectid = dirid;
1600         key.type = key_type;
1601         key.offset = *start_ret;
1602
1603         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1604         if (ret < 0)
1605                 goto out;
1606         if (ret > 0) {
1607                 if (path->slots[0] == 0)
1608                         goto out;
1609                 path->slots[0]--;
1610         }
1611         if (ret != 0)
1612                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1613
1614         if (key.type != key_type || key.objectid != dirid) {
1615                 ret = 1;
1616                 goto next;
1617         }
1618         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1619                               struct btrfs_dir_log_item);
1620         found_end = btrfs_dir_log_end(path->nodes[0], item);
1621
1622         if (*start_ret >= key.offset && *start_ret <= found_end) {
1623                 ret = 0;
1624                 *start_ret = key.offset;
1625                 *end_ret = found_end;
1626                 goto out;
1627         }
1628         ret = 1;
1629 next:
1630         /* check the next slot in the tree to see if it is a valid item */
1631         nritems = btrfs_header_nritems(path->nodes[0]);
1632         if (path->slots[0] >= nritems) {
1633                 ret = btrfs_next_leaf(root, path);
1634                 if (ret)
1635                         goto out;
1636         } else {
1637                 path->slots[0]++;
1638         }
1639
1640         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1641
1642         if (key.type != key_type || key.objectid != dirid) {
1643                 ret = 1;
1644                 goto out;
1645         }
1646         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1647                               struct btrfs_dir_log_item);
1648         found_end = btrfs_dir_log_end(path->nodes[0], item);
1649         *start_ret = key.offset;
1650         *end_ret = found_end;
1651         ret = 0;
1652 out:
1653         btrfs_release_path(path);
1654         return ret;
1655 }
1656
1657 /*
1658  * this looks for a given directory item in the log.  If the directory
1659  * item is not in the log, the item is removed and the inode it points
1660  * to is unlinked
1661  */
1662 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1663                                       struct btrfs_root *root,
1664                                       struct btrfs_root *log,
1665                                       struct btrfs_path *path,
1666                                       struct btrfs_path *log_path,
1667                                       struct inode *dir,
1668                                       struct btrfs_key *dir_key)
1669 {
1670         int ret;
1671         struct extent_buffer *eb;
1672         int slot;
1673         u32 item_size;
1674         struct btrfs_dir_item *di;
1675         struct btrfs_dir_item *log_di;
1676         int name_len;
1677         unsigned long ptr;
1678         unsigned long ptr_end;
1679         char *name;
1680         struct inode *inode;
1681         struct btrfs_key location;
1682
1683 again:
1684         eb = path->nodes[0];
1685         slot = path->slots[0];
1686         item_size = btrfs_item_size_nr(eb, slot);
1687         ptr = btrfs_item_ptr_offset(eb, slot);
1688         ptr_end = ptr + item_size;
1689         while (ptr < ptr_end) {
1690                 di = (struct btrfs_dir_item *)ptr;
1691                 if (verify_dir_item(root, eb, di)) {
1692                         ret = -EIO;
1693                         goto out;
1694                 }
1695
1696                 name_len = btrfs_dir_name_len(eb, di);
1697                 name = kmalloc(name_len, GFP_NOFS);
1698                 if (!name) {
1699                         ret = -ENOMEM;
1700                         goto out;
1701                 }
1702                 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1703                                   name_len);
1704                 log_di = NULL;
1705                 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1706                         log_di = btrfs_lookup_dir_item(trans, log, log_path,
1707                                                        dir_key->objectid,
1708                                                        name, name_len, 0);
1709                 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1710                         log_di = btrfs_lookup_dir_index_item(trans, log,
1711                                                      log_path,
1712                                                      dir_key->objectid,
1713                                                      dir_key->offset,
1714                                                      name, name_len, 0);
1715                 }
1716                 if (IS_ERR_OR_NULL(log_di)) {
1717                         btrfs_dir_item_key_to_cpu(eb, di, &location);
1718                         btrfs_release_path(path);
1719                         btrfs_release_path(log_path);
1720                         inode = read_one_inode(root, location.objectid);
1721                         if (!inode) {
1722                                 kfree(name);
1723                                 return -EIO;
1724                         }
1725
1726                         ret = link_to_fixup_dir(trans, root,
1727                                                 path, location.objectid);
1728                         BUG_ON(ret);
1729                         btrfs_inc_nlink(inode);
1730                         ret = btrfs_unlink_inode(trans, root, dir, inode,
1731                                                  name, name_len);
1732                         BUG_ON(ret);
1733
1734                         btrfs_run_delayed_items(trans, root);
1735
1736                         kfree(name);
1737                         iput(inode);
1738
1739                         /* there might still be more names under this key
1740                          * check and repeat if required
1741                          */
1742                         ret = btrfs_search_slot(NULL, root, dir_key, path,
1743                                                 0, 0);
1744                         if (ret == 0)
1745                                 goto again;
1746                         ret = 0;
1747                         goto out;
1748                 }
1749                 btrfs_release_path(log_path);
1750                 kfree(name);
1751
1752                 ptr = (unsigned long)(di + 1);
1753                 ptr += name_len;
1754         }
1755         ret = 0;
1756 out:
1757         btrfs_release_path(path);
1758         btrfs_release_path(log_path);
1759         return ret;
1760 }
1761
1762 /*
1763  * deletion replay happens before we copy any new directory items
1764  * out of the log or out of backreferences from inodes.  It
1765  * scans the log to find ranges of keys that log is authoritative for,
1766  * and then scans the directory to find items in those ranges that are
1767  * not present in the log.
1768  *
1769  * Anything we don't find in the log is unlinked and removed from the
1770  * directory.
1771  */
1772 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1773                                        struct btrfs_root *root,
1774                                        struct btrfs_root *log,
1775                                        struct btrfs_path *path,
1776                                        u64 dirid, int del_all)
1777 {
1778         u64 range_start;
1779         u64 range_end;
1780         int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1781         int ret = 0;
1782         struct btrfs_key dir_key;
1783         struct btrfs_key found_key;
1784         struct btrfs_path *log_path;
1785         struct inode *dir;
1786
1787         dir_key.objectid = dirid;
1788         dir_key.type = BTRFS_DIR_ITEM_KEY;
1789         log_path = btrfs_alloc_path();
1790         if (!log_path)
1791                 return -ENOMEM;
1792
1793         dir = read_one_inode(root, dirid);
1794         /* it isn't an error if the inode isn't there, that can happen
1795          * because we replay the deletes before we copy in the inode item
1796          * from the log
1797          */
1798         if (!dir) {
1799                 btrfs_free_path(log_path);
1800                 return 0;
1801         }
1802 again:
1803         range_start = 0;
1804         range_end = 0;
1805         while (1) {
1806                 if (del_all)
1807                         range_end = (u64)-1;
1808                 else {
1809                         ret = find_dir_range(log, path, dirid, key_type,
1810                                              &range_start, &range_end);
1811                         if (ret != 0)
1812                                 break;
1813                 }
1814
1815                 dir_key.offset = range_start;
1816                 while (1) {
1817                         int nritems;
1818                         ret = btrfs_search_slot(NULL, root, &dir_key, path,
1819                                                 0, 0);
1820                         if (ret < 0)
1821                                 goto out;
1822
1823                         nritems = btrfs_header_nritems(path->nodes[0]);
1824                         if (path->slots[0] >= nritems) {
1825                                 ret = btrfs_next_leaf(root, path);
1826                                 if (ret)
1827                                         break;
1828                         }
1829                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1830                                               path->slots[0]);
1831                         if (found_key.objectid != dirid ||
1832                             found_key.type != dir_key.type)
1833                                 goto next_type;
1834
1835                         if (found_key.offset > range_end)
1836                                 break;
1837
1838                         ret = check_item_in_log(trans, root, log, path,
1839                                                 log_path, dir,
1840                                                 &found_key);
1841                         BUG_ON(ret);
1842                         if (found_key.offset == (u64)-1)
1843                                 break;
1844                         dir_key.offset = found_key.offset + 1;
1845                 }
1846                 btrfs_release_path(path);
1847                 if (range_end == (u64)-1)
1848                         break;
1849                 range_start = range_end + 1;
1850         }
1851
1852 next_type:
1853         ret = 0;
1854         if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1855                 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1856                 dir_key.type = BTRFS_DIR_INDEX_KEY;
1857                 btrfs_release_path(path);
1858                 goto again;
1859         }
1860 out:
1861         btrfs_release_path(path);
1862         btrfs_free_path(log_path);
1863         iput(dir);
1864         return ret;
1865 }
1866
1867 /*
1868  * the process_func used to replay items from the log tree.  This
1869  * gets called in two different stages.  The first stage just looks
1870  * for inodes and makes sure they are all copied into the subvolume.
1871  *
1872  * The second stage copies all the other item types from the log into
1873  * the subvolume.  The two stage approach is slower, but gets rid of
1874  * lots of complexity around inodes referencing other inodes that exist
1875  * only in the log (references come from either directory items or inode
1876  * back refs).
1877  */
1878 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1879                              struct walk_control *wc, u64 gen)
1880 {
1881         int nritems;
1882         struct btrfs_path *path;
1883         struct btrfs_root *root = wc->replay_dest;
1884         struct btrfs_key key;
1885         int level;
1886         int i;
1887         int ret;
1888
1889         ret = btrfs_read_buffer(eb, gen);
1890         if (ret)
1891                 return ret;
1892
1893         level = btrfs_header_level(eb);
1894
1895         if (level != 0)
1896                 return 0;
1897
1898         path = btrfs_alloc_path();
1899         if (!path)
1900                 return -ENOMEM;
1901
1902         nritems = btrfs_header_nritems(eb);
1903         for (i = 0; i < nritems; i++) {
1904                 btrfs_item_key_to_cpu(eb, &key, i);
1905
1906                 /* inode keys are done during the first stage */
1907                 if (key.type == BTRFS_INODE_ITEM_KEY &&
1908                     wc->stage == LOG_WALK_REPLAY_INODES) {
1909                         struct btrfs_inode_item *inode_item;
1910                         u32 mode;
1911
1912                         inode_item = btrfs_item_ptr(eb, i,
1913                                             struct btrfs_inode_item);
1914                         mode = btrfs_inode_mode(eb, inode_item);
1915                         if (S_ISDIR(mode)) {
1916                                 ret = replay_dir_deletes(wc->trans,
1917                                          root, log, path, key.objectid, 0);
1918                                 BUG_ON(ret);
1919                         }
1920                         ret = overwrite_item(wc->trans, root, path,
1921                                              eb, i, &key);
1922                         BUG_ON(ret);
1923
1924                         /* for regular files, make sure corresponding
1925                          * orhpan item exist. extents past the new EOF
1926                          * will be truncated later by orphan cleanup.
1927                          */
1928                         if (S_ISREG(mode)) {
1929                                 ret = insert_orphan_item(wc->trans, root,
1930                                                          key.objectid);
1931                                 BUG_ON(ret);
1932                         }
1933
1934                         ret = link_to_fixup_dir(wc->trans, root,
1935                                                 path, key.objectid);
1936                         BUG_ON(ret);
1937                 }
1938                 if (wc->stage < LOG_WALK_REPLAY_ALL)
1939                         continue;
1940
1941                 /* these keys are simply copied */
1942                 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1943                         ret = overwrite_item(wc->trans, root, path,
1944                                              eb, i, &key);
1945                         BUG_ON(ret);
1946                 } else if (key.type == BTRFS_INODE_REF_KEY) {
1947                         ret = add_inode_ref(wc->trans, root, log, path,
1948                                             eb, i, &key);
1949                         BUG_ON(ret && ret != -ENOENT);
1950                 } else if (key.type == BTRFS_INODE_EXTREF_KEY) {
1951                         ret = add_inode_ref(wc->trans, root, log, path,
1952                                             eb, i, &key);
1953                         BUG_ON(ret && ret != -ENOENT);
1954                 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1955                         ret = replay_one_extent(wc->trans, root, path,
1956                                                 eb, i, &key);
1957                         BUG_ON(ret);
1958                 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1959                            key.type == BTRFS_DIR_INDEX_KEY) {
1960                         ret = replay_one_dir_item(wc->trans, root, path,
1961                                                   eb, i, &key);
1962                         BUG_ON(ret);
1963                 }
1964         }
1965         btrfs_free_path(path);
1966         return 0;
1967 }
1968
1969 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
1970                                    struct btrfs_root *root,
1971                                    struct btrfs_path *path, int *level,
1972                                    struct walk_control *wc)
1973 {
1974         u64 root_owner;
1975         u64 bytenr;
1976         u64 ptr_gen;
1977         struct extent_buffer *next;
1978         struct extent_buffer *cur;
1979         struct extent_buffer *parent;
1980         u32 blocksize;
1981         int ret = 0;
1982
1983         WARN_ON(*level < 0);
1984         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1985
1986         while (*level > 0) {
1987                 WARN_ON(*level < 0);
1988                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1989                 cur = path->nodes[*level];
1990
1991                 if (btrfs_header_level(cur) != *level)
1992                         WARN_ON(1);
1993
1994                 if (path->slots[*level] >=
1995                     btrfs_header_nritems(cur))
1996                         break;
1997
1998                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1999                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2000                 blocksize = btrfs_level_size(root, *level - 1);
2001
2002                 parent = path->nodes[*level];
2003                 root_owner = btrfs_header_owner(parent);
2004
2005                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
2006                 if (!next)
2007                         return -ENOMEM;
2008
2009                 if (*level == 1) {
2010                         ret = wc->process_func(root, next, wc, ptr_gen);
2011                         if (ret)
2012                                 return ret;
2013
2014                         path->slots[*level]++;
2015                         if (wc->free) {
2016                                 ret = btrfs_read_buffer(next, ptr_gen);
2017                                 if (ret) {
2018                                         free_extent_buffer(next);
2019                                         return ret;
2020                                 }
2021
2022                                 btrfs_tree_lock(next);
2023                                 btrfs_set_lock_blocking(next);
2024                                 clean_tree_block(trans, root, next);
2025                                 btrfs_wait_tree_block_writeback(next);
2026                                 btrfs_tree_unlock(next);
2027
2028                                 WARN_ON(root_owner !=
2029                                         BTRFS_TREE_LOG_OBJECTID);
2030                                 ret = btrfs_free_and_pin_reserved_extent(root,
2031                                                          bytenr, blocksize);
2032                                 BUG_ON(ret); /* -ENOMEM or logic errors */
2033                         }
2034                         free_extent_buffer(next);
2035                         continue;
2036                 }
2037                 ret = btrfs_read_buffer(next, ptr_gen);
2038                 if (ret) {
2039                         free_extent_buffer(next);
2040                         return ret;
2041                 }
2042
2043                 WARN_ON(*level <= 0);
2044                 if (path->nodes[*level-1])
2045                         free_extent_buffer(path->nodes[*level-1]);
2046                 path->nodes[*level-1] = next;
2047                 *level = btrfs_header_level(next);
2048                 path->slots[*level] = 0;
2049                 cond_resched();
2050         }
2051         WARN_ON(*level < 0);
2052         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2053
2054         path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2055
2056         cond_resched();
2057         return 0;
2058 }
2059
2060 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2061                                  struct btrfs_root *root,
2062                                  struct btrfs_path *path, int *level,
2063                                  struct walk_control *wc)
2064 {
2065         u64 root_owner;
2066         int i;
2067         int slot;
2068         int ret;
2069
2070         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2071                 slot = path->slots[i];
2072                 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2073                         path->slots[i]++;
2074                         *level = i;
2075                         WARN_ON(*level == 0);
2076                         return 0;
2077                 } else {
2078                         struct extent_buffer *parent;
2079                         if (path->nodes[*level] == root->node)
2080                                 parent = path->nodes[*level];
2081                         else
2082                                 parent = path->nodes[*level + 1];
2083
2084                         root_owner = btrfs_header_owner(parent);
2085                         ret = wc->process_func(root, path->nodes[*level], wc,
2086                                  btrfs_header_generation(path->nodes[*level]));
2087                         if (ret)
2088                                 return ret;
2089
2090                         if (wc->free) {
2091                                 struct extent_buffer *next;
2092
2093                                 next = path->nodes[*level];
2094
2095                                 btrfs_tree_lock(next);
2096                                 btrfs_set_lock_blocking(next);
2097                                 clean_tree_block(trans, root, next);
2098                                 btrfs_wait_tree_block_writeback(next);
2099                                 btrfs_tree_unlock(next);
2100
2101                                 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2102                                 ret = btrfs_free_and_pin_reserved_extent(root,
2103                                                 path->nodes[*level]->start,
2104                                                 path->nodes[*level]->len);
2105                                 BUG_ON(ret);
2106                         }
2107                         free_extent_buffer(path->nodes[*level]);
2108                         path->nodes[*level] = NULL;
2109                         *level = i + 1;
2110                 }
2111         }
2112         return 1;
2113 }
2114
2115 /*
2116  * drop the reference count on the tree rooted at 'snap'.  This traverses
2117  * the tree freeing any blocks that have a ref count of zero after being
2118  * decremented.
2119  */
2120 static int walk_log_tree(struct btrfs_trans_handle *trans,
2121                          struct btrfs_root *log, struct walk_control *wc)
2122 {
2123         int ret = 0;
2124         int wret;
2125         int level;
2126         struct btrfs_path *path;
2127         int i;
2128         int orig_level;
2129
2130         path = btrfs_alloc_path();
2131         if (!path)
2132                 return -ENOMEM;
2133
2134         level = btrfs_header_level(log->node);
2135         orig_level = level;
2136         path->nodes[level] = log->node;
2137         extent_buffer_get(log->node);
2138         path->slots[level] = 0;
2139
2140         while (1) {
2141                 wret = walk_down_log_tree(trans, log, path, &level, wc);
2142                 if (wret > 0)
2143                         break;
2144                 if (wret < 0) {
2145                         ret = wret;
2146                         goto out;
2147                 }
2148
2149                 wret = walk_up_log_tree(trans, log, path, &level, wc);
2150                 if (wret > 0)
2151                         break;
2152                 if (wret < 0) {
2153                         ret = wret;
2154                         goto out;
2155                 }
2156         }
2157
2158         /* was the root node processed? if not, catch it here */
2159         if (path->nodes[orig_level]) {
2160                 ret = wc->process_func(log, path->nodes[orig_level], wc,
2161                          btrfs_header_generation(path->nodes[orig_level]));
2162                 if (ret)
2163                         goto out;
2164                 if (wc->free) {
2165                         struct extent_buffer *next;
2166
2167                         next = path->nodes[orig_level];
2168
2169                         btrfs_tree_lock(next);
2170                         btrfs_set_lock_blocking(next);
2171                         clean_tree_block(trans, log, next);
2172                         btrfs_wait_tree_block_writeback(next);
2173                         btrfs_tree_unlock(next);
2174
2175                         WARN_ON(log->root_key.objectid !=
2176                                 BTRFS_TREE_LOG_OBJECTID);
2177                         ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2178                                                          next->len);
2179                         BUG_ON(ret); /* -ENOMEM or logic errors */
2180                 }
2181         }
2182
2183 out:
2184         for (i = 0; i <= orig_level; i++) {
2185                 if (path->nodes[i]) {
2186                         free_extent_buffer(path->nodes[i]);
2187                         path->nodes[i] = NULL;
2188                 }
2189         }
2190         btrfs_free_path(path);
2191         return ret;
2192 }
2193
2194 /*
2195  * helper function to update the item for a given subvolumes log root
2196  * in the tree of log roots
2197  */
2198 static int update_log_root(struct btrfs_trans_handle *trans,
2199                            struct btrfs_root *log)
2200 {
2201         int ret;
2202
2203         if (log->log_transid == 1) {
2204                 /* insert root item on the first sync */
2205                 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2206                                 &log->root_key, &log->root_item);
2207         } else {
2208                 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2209                                 &log->root_key, &log->root_item);
2210         }
2211         return ret;
2212 }
2213
2214 static int wait_log_commit(struct btrfs_trans_handle *trans,
2215                            struct btrfs_root *root, unsigned long transid)
2216 {
2217         DEFINE_WAIT(wait);
2218         int index = transid % 2;
2219
2220         /*
2221          * we only allow two pending log transactions at a time,
2222          * so we know that if ours is more than 2 older than the
2223          * current transaction, we're done
2224          */
2225         do {
2226                 prepare_to_wait(&root->log_commit_wait[index],
2227                                 &wait, TASK_UNINTERRUPTIBLE);
2228                 mutex_unlock(&root->log_mutex);
2229
2230                 if (root->fs_info->last_trans_log_full_commit !=
2231                     trans->transid && root->log_transid < transid + 2 &&
2232                     atomic_read(&root->log_commit[index]))
2233                         schedule();
2234
2235                 finish_wait(&root->log_commit_wait[index], &wait);
2236                 mutex_lock(&root->log_mutex);
2237         } while (root->fs_info->last_trans_log_full_commit !=
2238                  trans->transid && root->log_transid < transid + 2 &&
2239                  atomic_read(&root->log_commit[index]));
2240         return 0;
2241 }
2242
2243 static void wait_for_writer(struct btrfs_trans_handle *trans,
2244                             struct btrfs_root *root)
2245 {
2246         DEFINE_WAIT(wait);
2247         while (root->fs_info->last_trans_log_full_commit !=
2248                trans->transid && atomic_read(&root->log_writers)) {
2249                 prepare_to_wait(&root->log_writer_wait,
2250                                 &wait, TASK_UNINTERRUPTIBLE);
2251                 mutex_unlock(&root->log_mutex);
2252                 if (root->fs_info->last_trans_log_full_commit !=
2253                     trans->transid && atomic_read(&root->log_writers))
2254                         schedule();
2255                 mutex_lock(&root->log_mutex);
2256                 finish_wait(&root->log_writer_wait, &wait);
2257         }
2258 }
2259
2260 /*
2261  * btrfs_sync_log does sends a given tree log down to the disk and
2262  * updates the super blocks to record it.  When this call is done,
2263  * you know that any inodes previously logged are safely on disk only
2264  * if it returns 0.
2265  *
2266  * Any other return value means you need to call btrfs_commit_transaction.
2267  * Some of the edge cases for fsyncing directories that have had unlinks
2268  * or renames done in the past mean that sometimes the only safe
2269  * fsync is to commit the whole FS.  When btrfs_sync_log returns -EAGAIN,
2270  * that has happened.
2271  */
2272 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2273                    struct btrfs_root *root)
2274 {
2275         int index1;
2276         int index2;
2277         int mark;
2278         int ret;
2279         struct btrfs_root *log = root->log_root;
2280         struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2281         unsigned long log_transid = 0;
2282
2283         mutex_lock(&root->log_mutex);
2284         index1 = root->log_transid % 2;
2285         if (atomic_read(&root->log_commit[index1])) {
2286                 wait_log_commit(trans, root, root->log_transid);
2287                 mutex_unlock(&root->log_mutex);
2288                 return 0;
2289         }
2290         atomic_set(&root->log_commit[index1], 1);
2291
2292         /* wait for previous tree log sync to complete */
2293         if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2294                 wait_log_commit(trans, root, root->log_transid - 1);
2295         while (1) {
2296                 int batch = atomic_read(&root->log_batch);
2297                 /* when we're on an ssd, just kick the log commit out */
2298                 if (!btrfs_test_opt(root, SSD) && root->log_multiple_pids) {
2299                         mutex_unlock(&root->log_mutex);
2300                         schedule_timeout_uninterruptible(1);
2301                         mutex_lock(&root->log_mutex);
2302                 }
2303                 wait_for_writer(trans, root);
2304                 if (batch == atomic_read(&root->log_batch))
2305                         break;
2306         }
2307
2308         /* bail out if we need to do a full commit */
2309         if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2310                 ret = -EAGAIN;
2311                 mutex_unlock(&root->log_mutex);
2312                 goto out;
2313         }
2314
2315         log_transid = root->log_transid;
2316         if (log_transid % 2 == 0)
2317                 mark = EXTENT_DIRTY;
2318         else
2319                 mark = EXTENT_NEW;
2320
2321         /* we start IO on  all the marked extents here, but we don't actually
2322          * wait for them until later.
2323          */
2324         ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2325         if (ret) {
2326                 btrfs_abort_transaction(trans, root, ret);
2327                 mutex_unlock(&root->log_mutex);
2328                 goto out;
2329         }
2330
2331         btrfs_set_root_node(&log->root_item, log->node);
2332
2333         root->log_transid++;
2334         log->log_transid = root->log_transid;
2335         root->log_start_pid = 0;
2336         smp_mb();
2337         /*
2338          * IO has been started, blocks of the log tree have WRITTEN flag set
2339          * in their headers. new modifications of the log will be written to
2340          * new positions. so it's safe to allow log writers to go in.
2341          */
2342         mutex_unlock(&root->log_mutex);
2343
2344         mutex_lock(&log_root_tree->log_mutex);
2345         atomic_inc(&log_root_tree->log_batch);
2346         atomic_inc(&log_root_tree->log_writers);
2347         mutex_unlock(&log_root_tree->log_mutex);
2348
2349         ret = update_log_root(trans, log);
2350
2351         mutex_lock(&log_root_tree->log_mutex);
2352         if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2353                 smp_mb();
2354                 if (waitqueue_active(&log_root_tree->log_writer_wait))
2355                         wake_up(&log_root_tree->log_writer_wait);
2356         }
2357
2358         if (ret) {
2359                 if (ret != -ENOSPC) {
2360                         btrfs_abort_transaction(trans, root, ret);
2361                         mutex_unlock(&log_root_tree->log_mutex);
2362                         goto out;
2363                 }
2364                 root->fs_info->last_trans_log_full_commit = trans->transid;
2365                 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2366                 mutex_unlock(&log_root_tree->log_mutex);
2367                 ret = -EAGAIN;
2368                 goto out;
2369         }
2370
2371         index2 = log_root_tree->log_transid % 2;
2372         if (atomic_read(&log_root_tree->log_commit[index2])) {
2373                 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2374                 wait_log_commit(trans, log_root_tree,
2375                                 log_root_tree->log_transid);
2376                 mutex_unlock(&log_root_tree->log_mutex);
2377                 ret = 0;
2378                 goto out;
2379         }
2380         atomic_set(&log_root_tree->log_commit[index2], 1);
2381
2382         if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2383                 wait_log_commit(trans, log_root_tree,
2384                                 log_root_tree->log_transid - 1);
2385         }
2386
2387         wait_for_writer(trans, log_root_tree);
2388
2389         /*
2390          * now that we've moved on to the tree of log tree roots,
2391          * check the full commit flag again
2392          */
2393         if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2394                 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2395                 mutex_unlock(&log_root_tree->log_mutex);
2396                 ret = -EAGAIN;
2397                 goto out_wake_log_root;
2398         }
2399
2400         ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2401                                 &log_root_tree->dirty_log_pages,
2402                                 EXTENT_DIRTY | EXTENT_NEW);
2403         if (ret) {
2404                 btrfs_abort_transaction(trans, root, ret);
2405                 mutex_unlock(&log_root_tree->log_mutex);
2406                 goto out_wake_log_root;
2407         }
2408         btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2409
2410         btrfs_set_super_log_root(root->fs_info->super_for_commit,
2411                                 log_root_tree->node->start);
2412         btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2413                                 btrfs_header_level(log_root_tree->node));
2414
2415         log_root_tree->log_transid++;
2416         smp_mb();
2417
2418         mutex_unlock(&log_root_tree->log_mutex);
2419
2420         /*
2421          * nobody else is going to jump in and write the the ctree
2422          * super here because the log_commit atomic below is protecting
2423          * us.  We must be called with a transaction handle pinning
2424          * the running transaction open, so a full commit can't hop
2425          * in and cause problems either.
2426          */
2427         btrfs_scrub_pause_super(root);
2428         ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2429         btrfs_scrub_continue_super(root);
2430         if (ret) {
2431                 btrfs_abort_transaction(trans, root, ret);
2432                 goto out_wake_log_root;
2433         }
2434
2435         mutex_lock(&root->log_mutex);
2436         if (root->last_log_commit < log_transid)
2437                 root->last_log_commit = log_transid;
2438         mutex_unlock(&root->log_mutex);
2439
2440 out_wake_log_root:
2441         atomic_set(&log_root_tree->log_commit[index2], 0);
2442         smp_mb();
2443         if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2444                 wake_up(&log_root_tree->log_commit_wait[index2]);
2445 out:
2446         atomic_set(&root->log_commit[index1], 0);
2447         smp_mb();
2448         if (waitqueue_active(&root->log_commit_wait[index1]))
2449                 wake_up(&root->log_commit_wait[index1]);
2450         return ret;
2451 }
2452
2453 static void free_log_tree(struct btrfs_trans_handle *trans,
2454                           struct btrfs_root *log)
2455 {
2456         int ret;
2457         u64 start;
2458         u64 end;
2459         struct walk_control wc = {
2460                 .free = 1,
2461                 .process_func = process_one_buffer
2462         };
2463
2464         ret = walk_log_tree(trans, log, &wc);
2465         BUG_ON(ret);
2466
2467         while (1) {
2468                 ret = find_first_extent_bit(&log->dirty_log_pages,
2469                                 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2470                                 NULL);
2471                 if (ret)
2472                         break;
2473
2474                 clear_extent_bits(&log->dirty_log_pages, start, end,
2475                                   EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2476         }
2477
2478         free_extent_buffer(log->node);
2479         kfree(log);
2480 }
2481
2482 /*
2483  * free all the extents used by the tree log.  This should be called
2484  * at commit time of the full transaction
2485  */
2486 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2487 {
2488         if (root->log_root) {
2489                 free_log_tree(trans, root->log_root);
2490                 root->log_root = NULL;
2491         }
2492         return 0;
2493 }
2494
2495 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2496                              struct btrfs_fs_info *fs_info)
2497 {
2498         if (fs_info->log_root_tree) {
2499                 free_log_tree(trans, fs_info->log_root_tree);
2500                 fs_info->log_root_tree = NULL;
2501         }
2502         return 0;
2503 }
2504
2505 /*
2506  * If both a file and directory are logged, and unlinks or renames are
2507  * mixed in, we have a few interesting corners:
2508  *
2509  * create file X in dir Y
2510  * link file X to X.link in dir Y
2511  * fsync file X
2512  * unlink file X but leave X.link
2513  * fsync dir Y
2514  *
2515  * After a crash we would expect only X.link to exist.  But file X
2516  * didn't get fsync'd again so the log has back refs for X and X.link.
2517  *
2518  * We solve this by removing directory entries and inode backrefs from the
2519  * log when a file that was logged in the current transaction is
2520  * unlinked.  Any later fsync will include the updated log entries, and
2521  * we'll be able to reconstruct the proper directory items from backrefs.
2522  *
2523  * This optimizations allows us to avoid relogging the entire inode
2524  * or the entire directory.
2525  */
2526 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2527                                  struct btrfs_root *root,
2528                                  const char *name, int name_len,
2529                                  struct inode *dir, u64 index)
2530 {
2531         struct btrfs_root *log;
2532         struct btrfs_dir_item *di;
2533         struct btrfs_path *path;
2534         int ret;
2535         int err = 0;
2536         int bytes_del = 0;
2537         u64 dir_ino = btrfs_ino(dir);
2538
2539         if (BTRFS_I(dir)->logged_trans < trans->transid)
2540                 return 0;
2541
2542         ret = join_running_log_trans(root);
2543         if (ret)
2544                 return 0;
2545
2546         mutex_lock(&BTRFS_I(dir)->log_mutex);
2547
2548         log = root->log_root;
2549         path = btrfs_alloc_path();
2550         if (!path) {
2551                 err = -ENOMEM;
2552                 goto out_unlock;
2553         }
2554
2555         di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
2556                                    name, name_len, -1);
2557         if (IS_ERR(di)) {
2558                 err = PTR_ERR(di);
2559                 goto fail;
2560         }
2561         if (di) {
2562                 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2563                 bytes_del += name_len;
2564                 BUG_ON(ret);
2565         }
2566         btrfs_release_path(path);
2567         di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
2568                                          index, name, name_len, -1);
2569         if (IS_ERR(di)) {
2570                 err = PTR_ERR(di);
2571                 goto fail;
2572         }
2573         if (di) {
2574                 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2575                 bytes_del += name_len;
2576                 BUG_ON(ret);
2577         }
2578
2579         /* update the directory size in the log to reflect the names
2580          * we have removed
2581          */
2582         if (bytes_del) {
2583                 struct btrfs_key key;
2584
2585                 key.objectid = dir_ino;
2586                 key.offset = 0;
2587                 key.type = BTRFS_INODE_ITEM_KEY;
2588                 btrfs_release_path(path);
2589
2590                 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2591                 if (ret < 0) {
2592                         err = ret;
2593                         goto fail;
2594                 }
2595                 if (ret == 0) {
2596                         struct btrfs_inode_item *item;
2597                         u64 i_size;
2598
2599                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2600                                               struct btrfs_inode_item);
2601                         i_size = btrfs_inode_size(path->nodes[0], item);
2602                         if (i_size > bytes_del)
2603                                 i_size -= bytes_del;
2604                         else
2605                                 i_size = 0;
2606                         btrfs_set_inode_size(path->nodes[0], item, i_size);
2607                         btrfs_mark_buffer_dirty(path->nodes[0]);
2608                 } else
2609                         ret = 0;
2610                 btrfs_release_path(path);
2611         }
2612 fail:
2613         btrfs_free_path(path);
2614 out_unlock:
2615         mutex_unlock(&BTRFS_I(dir)->log_mutex);
2616         if (ret == -ENOSPC) {
2617                 root->fs_info->last_trans_log_full_commit = trans->transid;
2618                 ret = 0;
2619         } else if (ret < 0)
2620                 btrfs_abort_transaction(trans, root, ret);
2621
2622         btrfs_end_log_trans(root);
2623
2624         return err;
2625 }
2626
2627 /* see comments for btrfs_del_dir_entries_in_log */
2628 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2629                                struct btrfs_root *root,
2630                                const char *name, int name_len,
2631                                struct inode *inode, u64 dirid)
2632 {
2633         struct btrfs_root *log;
2634         u64 index;
2635         int ret;
2636
2637         if (BTRFS_I(inode)->logged_trans < trans->transid)
2638                 return 0;
2639
2640         ret = join_running_log_trans(root);
2641         if (ret)
2642                 return 0;
2643         log = root->log_root;
2644         mutex_lock(&BTRFS_I(inode)->log_mutex);
2645
2646         ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
2647                                   dirid, &index);
2648         mutex_unlock(&BTRFS_I(inode)->log_mutex);
2649         if (ret == -ENOSPC) {
2650                 root->fs_info->last_trans_log_full_commit = trans->transid;
2651                 ret = 0;
2652         } else if (ret < 0 && ret != -ENOENT)
2653                 btrfs_abort_transaction(trans, root, ret);
2654         btrfs_end_log_trans(root);
2655
2656         return ret;
2657 }
2658
2659 /*
2660  * creates a range item in the log for 'dirid'.  first_offset and
2661  * last_offset tell us which parts of the key space the log should
2662  * be considered authoritative for.
2663  */
2664 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2665                                        struct btrfs_root *log,
2666                                        struct btrfs_path *path,
2667                                        int key_type, u64 dirid,
2668                                        u64 first_offset, u64 last_offset)
2669 {
2670         int ret;
2671         struct btrfs_key key;
2672         struct btrfs_dir_log_item *item;
2673
2674         key.objectid = dirid;
2675         key.offset = first_offset;
2676         if (key_type == BTRFS_DIR_ITEM_KEY)
2677                 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2678         else
2679                 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2680         ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2681         if (ret)
2682                 return ret;
2683
2684         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2685                               struct btrfs_dir_log_item);
2686         btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2687         btrfs_mark_buffer_dirty(path->nodes[0]);
2688         btrfs_release_path(path);
2689         return 0;
2690 }
2691
2692 /*
2693  * log all the items included in the current transaction for a given
2694  * directory.  This also creates the range items in the log tree required
2695  * to replay anything deleted before the fsync
2696  */
2697 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2698                           struct btrfs_root *root, struct inode *inode,
2699                           struct btrfs_path *path,
2700                           struct btrfs_path *dst_path, int key_type,
2701                           u64 min_offset, u64 *last_offset_ret)
2702 {
2703         struct btrfs_key min_key;
2704         struct btrfs_key max_key;
2705         struct btrfs_root *log = root->log_root;
2706         struct extent_buffer *src;
2707         int err = 0;
2708         int ret;
2709         int i;
2710         int nritems;
2711         u64 first_offset = min_offset;
2712         u64 last_offset = (u64)-1;
2713         u64 ino = btrfs_ino(inode);
2714
2715         log = root->log_root;
2716         max_key.objectid = ino;
2717         max_key.offset = (u64)-1;
2718         max_key.type = key_type;
2719
2720         min_key.objectid = ino;
2721         min_key.type = key_type;
2722         min_key.offset = min_offset;
2723
2724         path->keep_locks = 1;
2725
2726         ret = btrfs_search_forward(root, &min_key, &max_key,
2727                                    path, 0, trans->transid);
2728
2729         /*
2730          * we didn't find anything from this transaction, see if there
2731          * is anything at all
2732          */
2733         if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2734                 min_key.objectid = ino;
2735                 min_key.type = key_type;
2736                 min_key.offset = (u64)-1;
2737                 btrfs_release_path(path);
2738                 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2739                 if (ret < 0) {
2740                         btrfs_release_path(path);
2741                         return ret;
2742                 }
2743                 ret = btrfs_previous_item(root, path, ino, key_type);
2744
2745                 /* if ret == 0 there are items for this type,
2746                  * create a range to tell us the last key of this type.
2747                  * otherwise, there are no items in this directory after
2748                  * *min_offset, and we create a range to indicate that.
2749                  */
2750                 if (ret == 0) {
2751                         struct btrfs_key tmp;
2752                         btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2753                                               path->slots[0]);
2754                         if (key_type == tmp.type)
2755                                 first_offset = max(min_offset, tmp.offset) + 1;
2756                 }
2757                 goto done;
2758         }
2759
2760         /* go backward to find any previous key */
2761         ret = btrfs_previous_item(root, path, ino, key_type);
2762         if (ret == 0) {
2763                 struct btrfs_key tmp;
2764                 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2765                 if (key_type == tmp.type) {
2766                         first_offset = tmp.offset;
2767                         ret = overwrite_item(trans, log, dst_path,
2768                                              path->nodes[0], path->slots[0],
2769                                              &tmp);
2770                         if (ret) {
2771                                 err = ret;
2772                                 goto done;
2773                         }
2774                 }
2775         }
2776         btrfs_release_path(path);
2777
2778         /* find the first key from this transaction again */
2779         ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2780         if (ret != 0) {
2781                 WARN_ON(1);
2782                 goto done;
2783         }
2784
2785         /*
2786          * we have a block from this transaction, log every item in it
2787          * from our directory
2788          */
2789         while (1) {
2790                 struct btrfs_key tmp;
2791                 src = path->nodes[0];
2792                 nritems = btrfs_header_nritems(src);
2793                 for (i = path->slots[0]; i < nritems; i++) {
2794                         btrfs_item_key_to_cpu(src, &min_key, i);
2795
2796                         if (min_key.objectid != ino || min_key.type != key_type)
2797                                 goto done;
2798                         ret = overwrite_item(trans, log, dst_path, src, i,
2799                                              &min_key);
2800                         if (ret) {
2801                                 err = ret;
2802                                 goto done;
2803                         }
2804                 }
2805                 path->slots[0] = nritems;
2806
2807                 /*
2808                  * look ahead to the next item and see if it is also
2809                  * from this directory and from this transaction
2810                  */
2811                 ret = btrfs_next_leaf(root, path);
2812                 if (ret == 1) {
2813                         last_offset = (u64)-1;
2814                         goto done;
2815                 }
2816                 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2817                 if (tmp.objectid != ino || tmp.type != key_type) {
2818                         last_offset = (u64)-1;
2819                         goto done;
2820                 }
2821                 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2822                         ret = overwrite_item(trans, log, dst_path,
2823                                              path->nodes[0], path->slots[0],
2824                                              &tmp);
2825                         if (ret)
2826                                 err = ret;
2827                         else
2828                                 last_offset = tmp.offset;
2829                         goto done;
2830                 }
2831         }
2832 done:
2833         btrfs_release_path(path);
2834         btrfs_release_path(dst_path);
2835
2836         if (err == 0) {
2837                 *last_offset_ret = last_offset;
2838                 /*
2839                  * insert the log range keys to indicate where the log
2840                  * is valid
2841                  */
2842                 ret = insert_dir_log_key(trans, log, path, key_type,
2843                                          ino, first_offset, last_offset);
2844                 if (ret)
2845                         err = ret;
2846         }
2847         return err;
2848 }
2849
2850 /*
2851  * logging directories is very similar to logging inodes, We find all the items
2852  * from the current transaction and write them to the log.
2853  *
2854  * The recovery code scans the directory in the subvolume, and if it finds a
2855  * key in the range logged that is not present in the log tree, then it means
2856  * that dir entry was unlinked during the transaction.
2857  *
2858  * In order for that scan to work, we must include one key smaller than
2859  * the smallest logged by this transaction and one key larger than the largest
2860  * key logged by this transaction.
2861  */
2862 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2863                           struct btrfs_root *root, struct inode *inode,
2864                           struct btrfs_path *path,
2865                           struct btrfs_path *dst_path)
2866 {
2867         u64 min_key;
2868         u64 max_key;
2869         int ret;
2870         int key_type = BTRFS_DIR_ITEM_KEY;
2871
2872 again:
2873         min_key = 0;
2874         max_key = 0;
2875         while (1) {
2876                 ret = log_dir_items(trans, root, inode, path,
2877                                     dst_path, key_type, min_key,
2878                                     &max_key);
2879                 if (ret)
2880                         return ret;
2881                 if (max_key == (u64)-1)
2882                         break;
2883                 min_key = max_key + 1;
2884         }
2885
2886         if (key_type == BTRFS_DIR_ITEM_KEY) {
2887                 key_type = BTRFS_DIR_INDEX_KEY;
2888                 goto again;
2889         }
2890         return 0;
2891 }
2892
2893 /*
2894  * a helper function to drop items from the log before we relog an
2895  * inode.  max_key_type indicates the highest item type to remove.
2896  * This cannot be run for file data extents because it does not
2897  * free the extents they point to.
2898  */
2899 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2900                                   struct btrfs_root *log,
2901                                   struct btrfs_path *path,
2902                                   u64 objectid, int max_key_type)
2903 {
2904         int ret;
2905         struct btrfs_key key;
2906         struct btrfs_key found_key;
2907         int start_slot;
2908
2909         key.objectid = objectid;
2910         key.type = max_key_type;
2911         key.offset = (u64)-1;
2912
2913         while (1) {
2914                 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2915                 BUG_ON(ret == 0);
2916                 if (ret < 0)
2917                         break;
2918
2919                 if (path->slots[0] == 0)
2920                         break;
2921
2922                 path->slots[0]--;
2923                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2924                                       path->slots[0]);
2925
2926                 if (found_key.objectid != objectid)
2927                         break;
2928
2929                 found_key.offset = 0;
2930                 found_key.type = 0;
2931                 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
2932                                        &start_slot);
2933
2934                 ret = btrfs_del_items(trans, log, path, start_slot,
2935                                       path->slots[0] - start_slot + 1);
2936                 /*
2937                  * If start slot isn't 0 then we don't need to re-search, we've
2938                  * found the last guy with the objectid in this tree.
2939                  */
2940                 if (ret || start_slot != 0)
2941                         break;
2942                 btrfs_release_path(path);
2943         }
2944         btrfs_release_path(path);
2945         if (ret > 0)
2946                 ret = 0;
2947         return ret;
2948 }
2949
2950 static void fill_inode_item(struct btrfs_trans_handle *trans,
2951                             struct extent_buffer *leaf,
2952                             struct btrfs_inode_item *item,
2953                             struct inode *inode, int log_inode_only)
2954 {
2955         btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2956         btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2957         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2958         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2959
2960         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2961                                inode->i_atime.tv_sec);
2962         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2963                                 inode->i_atime.tv_nsec);
2964
2965         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2966                                inode->i_mtime.tv_sec);
2967         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2968                                 inode->i_mtime.tv_nsec);
2969
2970         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2971                                inode->i_ctime.tv_sec);
2972         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2973                                 inode->i_ctime.tv_nsec);
2974
2975         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2976
2977         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2978         btrfs_set_inode_transid(leaf, item, trans->transid);
2979         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2980         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2981         btrfs_set_inode_block_group(leaf, item, 0);
2982
2983         if (log_inode_only) {
2984                 /* set the generation to zero so the recover code
2985                  * can tell the difference between an logging
2986                  * just to say 'this inode exists' and a logging
2987                  * to say 'update this inode with these values'
2988                  */
2989                 btrfs_set_inode_generation(leaf, item, 0);
2990                 btrfs_set_inode_size(leaf, item, 0);
2991         } else {
2992                 btrfs_set_inode_generation(leaf, item,
2993                                            BTRFS_I(inode)->generation);
2994                 btrfs_set_inode_size(leaf, item, inode->i_size);
2995         }
2996
2997 }
2998
2999 static noinline int copy_items(struct btrfs_trans_handle *trans,
3000                                struct inode *inode,
3001                                struct btrfs_path *dst_path,
3002                                struct extent_buffer *src,
3003                                int start_slot, int nr, int inode_only)
3004 {
3005         unsigned long src_offset;
3006         unsigned long dst_offset;
3007         struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3008         struct btrfs_file_extent_item *extent;
3009         struct btrfs_inode_item *inode_item;
3010         int ret;
3011         struct btrfs_key *ins_keys;
3012         u32 *ins_sizes;
3013         char *ins_data;
3014         int i;
3015         struct list_head ordered_sums;
3016         int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3017
3018         INIT_LIST_HEAD(&ordered_sums);
3019
3020         ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3021                            nr * sizeof(u32), GFP_NOFS);
3022         if (!ins_data)
3023                 return -ENOMEM;
3024
3025         ins_sizes = (u32 *)ins_data;
3026         ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3027
3028         for (i = 0; i < nr; i++) {
3029                 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3030                 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3031         }
3032         ret = btrfs_insert_empty_items(trans, log, dst_path,
3033                                        ins_keys, ins_sizes, nr);
3034         if (ret) {
3035                 kfree(ins_data);
3036                 return ret;
3037         }
3038
3039         for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3040                 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3041                                                    dst_path->slots[0]);
3042
3043                 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3044
3045                 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3046                         inode_item = btrfs_item_ptr(dst_path->nodes[0],
3047                                                     dst_path->slots[0],
3048                                                     struct btrfs_inode_item);
3049                         fill_inode_item(trans, dst_path->nodes[0], inode_item,
3050                                         inode, inode_only == LOG_INODE_EXISTS);
3051                 } else {
3052                         copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3053                                            src_offset, ins_sizes[i]);
3054                 }
3055
3056                 /* take a reference on file data extents so that truncates
3057                  * or deletes of this inode don't have to relog the inode
3058                  * again
3059                  */
3060                 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY &&
3061                     !skip_csum) {
3062                         int found_type;
3063                         extent = btrfs_item_ptr(src, start_slot + i,
3064                                                 struct btrfs_file_extent_item);
3065
3066                         if (btrfs_file_extent_generation(src, extent) < trans->transid)
3067                                 continue;
3068
3069                         found_type = btrfs_file_extent_type(src, extent);
3070                         if (found_type == BTRFS_FILE_EXTENT_REG) {
3071                                 u64 ds, dl, cs, cl;
3072                                 ds = btrfs_file_extent_disk_bytenr(src,
3073                                                                 extent);
3074                                 /* ds == 0 is a hole */
3075                                 if (ds == 0)
3076                                         continue;
3077
3078                                 dl = btrfs_file_extent_disk_num_bytes(src,
3079                                                                 extent);
3080                                 cs = btrfs_file_extent_offset(src, extent);
3081                                 cl = btrfs_file_extent_num_bytes(src,
3082                                                                 extent);
3083                                 if (btrfs_file_extent_compression(src,
3084                                                                   extent)) {
3085                                         cs = 0;
3086                                         cl = dl;
3087                                 }
3088
3089                                 ret = btrfs_lookup_csums_range(
3090                                                 log->fs_info->csum_root,
3091                                                 ds + cs, ds + cs + cl - 1,
3092                                                 &ordered_sums, 0);
3093                                 BUG_ON(ret);
3094                         }
3095                 }
3096         }
3097
3098         btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3099         btrfs_release_path(dst_path);
3100         kfree(ins_data);
3101
3102         /*
3103          * we have to do this after the loop above to avoid changing the
3104          * log tree while trying to change the log tree.
3105          */
3106         ret = 0;
3107         while (!list_empty(&ordered_sums)) {
3108                 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3109                                                    struct btrfs_ordered_sum,
3110                                                    list);
3111                 if (!ret)
3112                         ret = btrfs_csum_file_blocks(trans, log, sums);
3113                 list_del(&sums->list);
3114                 kfree(sums);
3115         }
3116         return ret;
3117 }
3118
3119 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3120 {
3121         struct extent_map *em1, *em2;
3122
3123         em1 = list_entry(a, struct extent_map, list);
3124         em2 = list_entry(b, struct extent_map, list);
3125
3126         if (em1->start < em2->start)
3127                 return -1;
3128         else if (em1->start > em2->start)
3129                 return 1;
3130         return 0;
3131 }
3132
3133 struct log_args {
3134         struct extent_buffer *src;
3135         u64 next_offset;
3136         int start_slot;
3137         int nr;
3138 };
3139
3140 static int log_one_extent(struct btrfs_trans_handle *trans,
3141                           struct inode *inode, struct btrfs_root *root,
3142                           struct extent_map *em, struct btrfs_path *path,
3143                           struct btrfs_path *dst_path, struct log_args *args)
3144 {
3145         struct btrfs_root *log = root->log_root;
3146         struct btrfs_key key;
3147         u64 start = em->mod_start;
3148         u64 search_start = start;
3149         u64 len = em->mod_len;
3150         u64 num_bytes;
3151         int nritems;
3152         int ret;
3153
3154         if (BTRFS_I(inode)->logged_trans == trans->transid) {
3155                 ret = __btrfs_drop_extents(trans, log, inode, dst_path, start,
3156                                            start + len, NULL, 0);
3157                 if (ret)
3158                         return ret;
3159         }
3160
3161         while (len) {
3162                 if (args->nr)
3163                         goto next_slot;
3164 again:
3165                 key.objectid = btrfs_ino(inode);
3166                 key.type = BTRFS_EXTENT_DATA_KEY;
3167                 key.offset = search_start;
3168
3169                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3170                 if (ret < 0)
3171                         return ret;
3172
3173                 if (ret) {
3174                         /*
3175                          * A rare case were we can have an em for a section of a