7aa9cd36bf1b496efd2a02e7cc86c1a08a29c663
[~shefty/rdma-dev.git] / fs / btrfs / disk-io.c
1 /*
2  * Copyright (C) 2007 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/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
34 #include "compat.h"
35 #include "ctree.h"
36 #include "disk-io.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "volumes.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
42 #include "locking.h"
43 #include "tree-log.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47
48 static struct extent_io_ops btree_extent_io_ops;
49 static void end_workqueue_fn(struct btrfs_work *work);
50 static void free_fs_root(struct btrfs_root *root);
51 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
52                                     int read_only);
53 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
54 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56                                       struct btrfs_root *root);
57 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
58 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
60                                         struct extent_io_tree *dirty_pages,
61                                         int mark);
62 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
63                                        struct extent_io_tree *pinned_extents);
64 static int btrfs_cleanup_transaction(struct btrfs_root *root);
65
66 /*
67  * end_io_wq structs are used to do processing in task context when an IO is
68  * complete.  This is used during reads to verify checksums, and it is used
69  * by writes to insert metadata for new file extents after IO is complete.
70  */
71 struct end_io_wq {
72         struct bio *bio;
73         bio_end_io_t *end_io;
74         void *private;
75         struct btrfs_fs_info *info;
76         int error;
77         int metadata;
78         struct list_head list;
79         struct btrfs_work work;
80 };
81
82 /*
83  * async submit bios are used to offload expensive checksumming
84  * onto the worker threads.  They checksum file and metadata bios
85  * just before they are sent down the IO stack.
86  */
87 struct async_submit_bio {
88         struct inode *inode;
89         struct bio *bio;
90         struct list_head list;
91         extent_submit_bio_hook_t *submit_bio_start;
92         extent_submit_bio_hook_t *submit_bio_done;
93         int rw;
94         int mirror_num;
95         unsigned long bio_flags;
96         /*
97          * bio_offset is optional, can be used if the pages in the bio
98          * can't tell us where in the file the bio should go
99          */
100         u64 bio_offset;
101         struct btrfs_work work;
102 };
103
104 /*
105  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
106  * eb, the lockdep key is determined by the btrfs_root it belongs to and
107  * the level the eb occupies in the tree.
108  *
109  * Different roots are used for different purposes and may nest inside each
110  * other and they require separate keysets.  As lockdep keys should be
111  * static, assign keysets according to the purpose of the root as indicated
112  * by btrfs_root->objectid.  This ensures that all special purpose roots
113  * have separate keysets.
114  *
115  * Lock-nesting across peer nodes is always done with the immediate parent
116  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
117  * subclass to avoid triggering lockdep warning in such cases.
118  *
119  * The key is set by the readpage_end_io_hook after the buffer has passed
120  * csum validation but before the pages are unlocked.  It is also set by
121  * btrfs_init_new_buffer on freshly allocated blocks.
122  *
123  * We also add a check to make sure the highest level of the tree is the
124  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
125  * needs update as well.
126  */
127 #ifdef CONFIG_DEBUG_LOCK_ALLOC
128 # if BTRFS_MAX_LEVEL != 8
129 #  error
130 # endif
131
132 static struct btrfs_lockdep_keyset {
133         u64                     id;             /* root objectid */
134         const char              *name_stem;     /* lock name stem */
135         char                    names[BTRFS_MAX_LEVEL + 1][20];
136         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
137 } btrfs_lockdep_keysets[] = {
138         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
139         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
140         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
141         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
142         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
143         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
144         { .id = BTRFS_ORPHAN_OBJECTID,          .name_stem = "orphan"   },
145         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
146         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
147         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
148         { .id = 0,                              .name_stem = "tree"     },
149 };
150
151 void __init btrfs_init_lockdep(void)
152 {
153         int i, j;
154
155         /* initialize lockdep class names */
156         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
157                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
158
159                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
160                         snprintf(ks->names[j], sizeof(ks->names[j]),
161                                  "btrfs-%s-%02d", ks->name_stem, j);
162         }
163 }
164
165 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
166                                     int level)
167 {
168         struct btrfs_lockdep_keyset *ks;
169
170         BUG_ON(level >= ARRAY_SIZE(ks->keys));
171
172         /* find the matching keyset, id 0 is the default entry */
173         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
174                 if (ks->id == objectid)
175                         break;
176
177         lockdep_set_class_and_name(&eb->lock,
178                                    &ks->keys[level], ks->names[level]);
179 }
180
181 #endif
182
183 /*
184  * extents on the btree inode are pretty simple, there's one extent
185  * that covers the entire device
186  */
187 static struct extent_map *btree_get_extent(struct inode *inode,
188                 struct page *page, size_t pg_offset, u64 start, u64 len,
189                 int create)
190 {
191         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
192         struct extent_map *em;
193         int ret;
194
195         read_lock(&em_tree->lock);
196         em = lookup_extent_mapping(em_tree, start, len);
197         if (em) {
198                 em->bdev =
199                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
200                 read_unlock(&em_tree->lock);
201                 goto out;
202         }
203         read_unlock(&em_tree->lock);
204
205         em = alloc_extent_map();
206         if (!em) {
207                 em = ERR_PTR(-ENOMEM);
208                 goto out;
209         }
210         em->start = 0;
211         em->len = (u64)-1;
212         em->block_len = (u64)-1;
213         em->block_start = 0;
214         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
215
216         write_lock(&em_tree->lock);
217         ret = add_extent_mapping(em_tree, em);
218         if (ret == -EEXIST) {
219                 u64 failed_start = em->start;
220                 u64 failed_len = em->len;
221
222                 free_extent_map(em);
223                 em = lookup_extent_mapping(em_tree, start, len);
224                 if (em) {
225                         ret = 0;
226                 } else {
227                         em = lookup_extent_mapping(em_tree, failed_start,
228                                                    failed_len);
229                         ret = -EIO;
230                 }
231         } else if (ret) {
232                 free_extent_map(em);
233                 em = NULL;
234         }
235         write_unlock(&em_tree->lock);
236
237         if (ret)
238                 em = ERR_PTR(ret);
239 out:
240         return em;
241 }
242
243 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
244 {
245         return crc32c(seed, data, len);
246 }
247
248 void btrfs_csum_final(u32 crc, char *result)
249 {
250         put_unaligned_le32(~crc, result);
251 }
252
253 /*
254  * compute the csum for a btree block, and either verify it or write it
255  * into the csum field of the block.
256  */
257 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
258                            int verify)
259 {
260         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
261         char *result = NULL;
262         unsigned long len;
263         unsigned long cur_len;
264         unsigned long offset = BTRFS_CSUM_SIZE;
265         char *kaddr;
266         unsigned long map_start;
267         unsigned long map_len;
268         int err;
269         u32 crc = ~(u32)0;
270         unsigned long inline_result;
271
272         len = buf->len - offset;
273         while (len > 0) {
274                 err = map_private_extent_buffer(buf, offset, 32,
275                                         &kaddr, &map_start, &map_len);
276                 if (err)
277                         return 1;
278                 cur_len = min(len, map_len - (offset - map_start));
279                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
280                                       crc, cur_len);
281                 len -= cur_len;
282                 offset += cur_len;
283         }
284         if (csum_size > sizeof(inline_result)) {
285                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
286                 if (!result)
287                         return 1;
288         } else {
289                 result = (char *)&inline_result;
290         }
291
292         btrfs_csum_final(crc, result);
293
294         if (verify) {
295                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
296                         u32 val;
297                         u32 found = 0;
298                         memcpy(&found, result, csum_size);
299
300                         read_extent_buffer(buf, &val, 0, csum_size);
301                         printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
302                                        "failed on %llu wanted %X found %X "
303                                        "level %d\n",
304                                        root->fs_info->sb->s_id,
305                                        (unsigned long long)buf->start, val, found,
306                                        btrfs_header_level(buf));
307                         if (result != (char *)&inline_result)
308                                 kfree(result);
309                         return 1;
310                 }
311         } else {
312                 write_extent_buffer(buf, result, 0, csum_size);
313         }
314         if (result != (char *)&inline_result)
315                 kfree(result);
316         return 0;
317 }
318
319 /*
320  * we can't consider a given block up to date unless the transid of the
321  * block matches the transid in the parent node's pointer.  This is how we
322  * detect blocks that either didn't get written at all or got written
323  * in the wrong place.
324  */
325 static int verify_parent_transid(struct extent_io_tree *io_tree,
326                                  struct extent_buffer *eb, u64 parent_transid)
327 {
328         struct extent_state *cached_state = NULL;
329         int ret;
330
331         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
332                 return 0;
333
334         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
335                          0, &cached_state, GFP_NOFS);
336         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
337             btrfs_header_generation(eb) == parent_transid) {
338                 ret = 0;
339                 goto out;
340         }
341         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
342                        "found %llu\n",
343                        (unsigned long long)eb->start,
344                        (unsigned long long)parent_transid,
345                        (unsigned long long)btrfs_header_generation(eb));
346         ret = 1;
347         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
348 out:
349         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
350                              &cached_state, GFP_NOFS);
351         return ret;
352 }
353
354 /*
355  * helper to read a given tree block, doing retries as required when
356  * the checksums don't match and we have alternate mirrors to try.
357  */
358 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
359                                           struct extent_buffer *eb,
360                                           u64 start, u64 parent_transid)
361 {
362         struct extent_io_tree *io_tree;
363         int ret;
364         int num_copies = 0;
365         int mirror_num = 0;
366
367         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
368         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
369         while (1) {
370                 ret = read_extent_buffer_pages(io_tree, eb, start,
371                                                WAIT_COMPLETE,
372                                                btree_get_extent, mirror_num);
373                 if (!ret &&
374                     !verify_parent_transid(io_tree, eb, parent_transid))
375                         return ret;
376
377                 /*
378                  * This buffer's crc is fine, but its contents are corrupted, so
379                  * there is no reason to read the other copies, they won't be
380                  * any less wrong.
381                  */
382                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
383                         return ret;
384
385                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
386                                               eb->start, eb->len);
387                 if (num_copies == 1)
388                         return ret;
389
390                 mirror_num++;
391                 if (mirror_num > num_copies)
392                         return ret;
393         }
394         return -EIO;
395 }
396
397 /*
398  * checksum a dirty tree block before IO.  This has extra checks to make sure
399  * we only fill in the checksum field in the first page of a multi-page block
400  */
401
402 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
403 {
404         struct extent_io_tree *tree;
405         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
406         u64 found_start;
407         unsigned long len;
408         struct extent_buffer *eb;
409         int ret;
410
411         tree = &BTRFS_I(page->mapping->host)->io_tree;
412
413         if (page->private == EXTENT_PAGE_PRIVATE) {
414                 WARN_ON(1);
415                 goto out;
416         }
417         if (!page->private) {
418                 WARN_ON(1);
419                 goto out;
420         }
421         len = page->private >> 2;
422         WARN_ON(len == 0);
423
424         eb = alloc_extent_buffer(tree, start, len, page);
425         if (eb == NULL) {
426                 WARN_ON(1);
427                 goto out;
428         }
429         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
430                                              btrfs_header_generation(eb));
431         BUG_ON(ret);
432         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
433
434         found_start = btrfs_header_bytenr(eb);
435         if (found_start != start) {
436                 WARN_ON(1);
437                 goto err;
438         }
439         if (eb->first_page != page) {
440                 WARN_ON(1);
441                 goto err;
442         }
443         if (!PageUptodate(page)) {
444                 WARN_ON(1);
445                 goto err;
446         }
447         csum_tree_block(root, eb, 0);
448 err:
449         free_extent_buffer(eb);
450 out:
451         return 0;
452 }
453
454 static int check_tree_block_fsid(struct btrfs_root *root,
455                                  struct extent_buffer *eb)
456 {
457         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
458         u8 fsid[BTRFS_UUID_SIZE];
459         int ret = 1;
460
461         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
462                            BTRFS_FSID_SIZE);
463         while (fs_devices) {
464                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
465                         ret = 0;
466                         break;
467                 }
468                 fs_devices = fs_devices->seed;
469         }
470         return ret;
471 }
472
473 #define CORRUPT(reason, eb, root, slot)                         \
474         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
475                "root=%llu, slot=%d\n", reason,                  \
476                (unsigned long long)btrfs_header_bytenr(eb),     \
477                (unsigned long long)root->objectid, slot)
478
479 static noinline int check_leaf(struct btrfs_root *root,
480                                struct extent_buffer *leaf)
481 {
482         struct btrfs_key key;
483         struct btrfs_key leaf_key;
484         u32 nritems = btrfs_header_nritems(leaf);
485         int slot;
486
487         if (nritems == 0)
488                 return 0;
489
490         /* Check the 0 item */
491         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
492             BTRFS_LEAF_DATA_SIZE(root)) {
493                 CORRUPT("invalid item offset size pair", leaf, root, 0);
494                 return -EIO;
495         }
496
497         /*
498          * Check to make sure each items keys are in the correct order and their
499          * offsets make sense.  We only have to loop through nritems-1 because
500          * we check the current slot against the next slot, which verifies the
501          * next slot's offset+size makes sense and that the current's slot
502          * offset is correct.
503          */
504         for (slot = 0; slot < nritems - 1; slot++) {
505                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
506                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
507
508                 /* Make sure the keys are in the right order */
509                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
510                         CORRUPT("bad key order", leaf, root, slot);
511                         return -EIO;
512                 }
513
514                 /*
515                  * Make sure the offset and ends are right, remember that the
516                  * item data starts at the end of the leaf and grows towards the
517                  * front.
518                  */
519                 if (btrfs_item_offset_nr(leaf, slot) !=
520                         btrfs_item_end_nr(leaf, slot + 1)) {
521                         CORRUPT("slot offset bad", leaf, root, slot);
522                         return -EIO;
523                 }
524
525                 /*
526                  * Check to make sure that we don't point outside of the leaf,
527                  * just incase all the items are consistent to eachother, but
528                  * all point outside of the leaf.
529                  */
530                 if (btrfs_item_end_nr(leaf, slot) >
531                     BTRFS_LEAF_DATA_SIZE(root)) {
532                         CORRUPT("slot end outside of leaf", leaf, root, slot);
533                         return -EIO;
534                 }
535         }
536
537         return 0;
538 }
539
540 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
541                                struct extent_state *state)
542 {
543         struct extent_io_tree *tree;
544         u64 found_start;
545         int found_level;
546         unsigned long len;
547         struct extent_buffer *eb;
548         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
549         int ret = 0;
550
551         tree = &BTRFS_I(page->mapping->host)->io_tree;
552         if (page->private == EXTENT_PAGE_PRIVATE)
553                 goto out;
554         if (!page->private)
555                 goto out;
556
557         len = page->private >> 2;
558         WARN_ON(len == 0);
559
560         eb = alloc_extent_buffer(tree, start, len, page);
561         if (eb == NULL) {
562                 ret = -EIO;
563                 goto out;
564         }
565
566         found_start = btrfs_header_bytenr(eb);
567         if (found_start != start) {
568                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
569                                "%llu %llu\n",
570                                (unsigned long long)found_start,
571                                (unsigned long long)eb->start);
572                 ret = -EIO;
573                 goto err;
574         }
575         if (eb->first_page != page) {
576                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
577                        eb->first_page->index, page->index);
578                 WARN_ON(1);
579                 ret = -EIO;
580                 goto err;
581         }
582         if (check_tree_block_fsid(root, eb)) {
583                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
584                                (unsigned long long)eb->start);
585                 ret = -EIO;
586                 goto err;
587         }
588         found_level = btrfs_header_level(eb);
589
590         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
591                                        eb, found_level);
592
593         ret = csum_tree_block(root, eb, 1);
594         if (ret) {
595                 ret = -EIO;
596                 goto err;
597         }
598
599         /*
600          * If this is a leaf block and it is corrupt, set the corrupt bit so
601          * that we don't try and read the other copies of this block, just
602          * return -EIO.
603          */
604         if (found_level == 0 && check_leaf(root, eb)) {
605                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
606                 ret = -EIO;
607         }
608
609         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
610         end = eb->start + end - 1;
611 err:
612         if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
613                 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
614                 btree_readahead_hook(root, eb, eb->start, ret);
615         }
616
617         free_extent_buffer(eb);
618 out:
619         return ret;
620 }
621
622 static int btree_io_failed_hook(struct bio *failed_bio,
623                          struct page *page, u64 start, u64 end,
624                          int mirror_num, struct extent_state *state)
625 {
626         struct extent_io_tree *tree;
627         unsigned long len;
628         struct extent_buffer *eb;
629         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
630
631         tree = &BTRFS_I(page->mapping->host)->io_tree;
632         if (page->private == EXTENT_PAGE_PRIVATE)
633                 goto out;
634         if (!page->private)
635                 goto out;
636
637         len = page->private >> 2;
638         WARN_ON(len == 0);
639
640         eb = alloc_extent_buffer(tree, start, len, page);
641         if (eb == NULL)
642                 goto out;
643
644         if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
645                 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
646                 btree_readahead_hook(root, eb, eb->start, -EIO);
647         }
648         free_extent_buffer(eb);
649
650 out:
651         return -EIO;    /* we fixed nothing */
652 }
653
654 static void end_workqueue_bio(struct bio *bio, int err)
655 {
656         struct end_io_wq *end_io_wq = bio->bi_private;
657         struct btrfs_fs_info *fs_info;
658
659         fs_info = end_io_wq->info;
660         end_io_wq->error = err;
661         end_io_wq->work.func = end_workqueue_fn;
662         end_io_wq->work.flags = 0;
663
664         if (bio->bi_rw & REQ_WRITE) {
665                 if (end_io_wq->metadata == 1)
666                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
667                                            &end_io_wq->work);
668                 else if (end_io_wq->metadata == 2)
669                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
670                                            &end_io_wq->work);
671                 else
672                         btrfs_queue_worker(&fs_info->endio_write_workers,
673                                            &end_io_wq->work);
674         } else {
675                 if (end_io_wq->metadata)
676                         btrfs_queue_worker(&fs_info->endio_meta_workers,
677                                            &end_io_wq->work);
678                 else
679                         btrfs_queue_worker(&fs_info->endio_workers,
680                                            &end_io_wq->work);
681         }
682 }
683
684 /*
685  * For the metadata arg you want
686  *
687  * 0 - if data
688  * 1 - if normal metadta
689  * 2 - if writing to the free space cache area
690  */
691 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
692                         int metadata)
693 {
694         struct end_io_wq *end_io_wq;
695         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
696         if (!end_io_wq)
697                 return -ENOMEM;
698
699         end_io_wq->private = bio->bi_private;
700         end_io_wq->end_io = bio->bi_end_io;
701         end_io_wq->info = info;
702         end_io_wq->error = 0;
703         end_io_wq->bio = bio;
704         end_io_wq->metadata = metadata;
705
706         bio->bi_private = end_io_wq;
707         bio->bi_end_io = end_workqueue_bio;
708         return 0;
709 }
710
711 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
712 {
713         unsigned long limit = min_t(unsigned long,
714                                     info->workers.max_workers,
715                                     info->fs_devices->open_devices);
716         return 256 * limit;
717 }
718
719 static void run_one_async_start(struct btrfs_work *work)
720 {
721         struct async_submit_bio *async;
722
723         async = container_of(work, struct  async_submit_bio, work);
724         async->submit_bio_start(async->inode, async->rw, async->bio,
725                                async->mirror_num, async->bio_flags,
726                                async->bio_offset);
727 }
728
729 static void run_one_async_done(struct btrfs_work *work)
730 {
731         struct btrfs_fs_info *fs_info;
732         struct async_submit_bio *async;
733         int limit;
734
735         async = container_of(work, struct  async_submit_bio, work);
736         fs_info = BTRFS_I(async->inode)->root->fs_info;
737
738         limit = btrfs_async_submit_limit(fs_info);
739         limit = limit * 2 / 3;
740
741         atomic_dec(&fs_info->nr_async_submits);
742
743         if (atomic_read(&fs_info->nr_async_submits) < limit &&
744             waitqueue_active(&fs_info->async_submit_wait))
745                 wake_up(&fs_info->async_submit_wait);
746
747         async->submit_bio_done(async->inode, async->rw, async->bio,
748                                async->mirror_num, async->bio_flags,
749                                async->bio_offset);
750 }
751
752 static void run_one_async_free(struct btrfs_work *work)
753 {
754         struct async_submit_bio *async;
755
756         async = container_of(work, struct  async_submit_bio, work);
757         kfree(async);
758 }
759
760 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
761                         int rw, struct bio *bio, int mirror_num,
762                         unsigned long bio_flags,
763                         u64 bio_offset,
764                         extent_submit_bio_hook_t *submit_bio_start,
765                         extent_submit_bio_hook_t *submit_bio_done)
766 {
767         struct async_submit_bio *async;
768
769         async = kmalloc(sizeof(*async), GFP_NOFS);
770         if (!async)
771                 return -ENOMEM;
772
773         async->inode = inode;
774         async->rw = rw;
775         async->bio = bio;
776         async->mirror_num = mirror_num;
777         async->submit_bio_start = submit_bio_start;
778         async->submit_bio_done = submit_bio_done;
779
780         async->work.func = run_one_async_start;
781         async->work.ordered_func = run_one_async_done;
782         async->work.ordered_free = run_one_async_free;
783
784         async->work.flags = 0;
785         async->bio_flags = bio_flags;
786         async->bio_offset = bio_offset;
787
788         atomic_inc(&fs_info->nr_async_submits);
789
790         if (rw & REQ_SYNC)
791                 btrfs_set_work_high_prio(&async->work);
792
793         btrfs_queue_worker(&fs_info->workers, &async->work);
794
795         while (atomic_read(&fs_info->async_submit_draining) &&
796               atomic_read(&fs_info->nr_async_submits)) {
797                 wait_event(fs_info->async_submit_wait,
798                            (atomic_read(&fs_info->nr_async_submits) == 0));
799         }
800
801         return 0;
802 }
803
804 static int btree_csum_one_bio(struct bio *bio)
805 {
806         struct bio_vec *bvec = bio->bi_io_vec;
807         int bio_index = 0;
808         struct btrfs_root *root;
809
810         WARN_ON(bio->bi_vcnt <= 0);
811         while (bio_index < bio->bi_vcnt) {
812                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
813                 csum_dirty_buffer(root, bvec->bv_page);
814                 bio_index++;
815                 bvec++;
816         }
817         return 0;
818 }
819
820 static int __btree_submit_bio_start(struct inode *inode, int rw,
821                                     struct bio *bio, int mirror_num,
822                                     unsigned long bio_flags,
823                                     u64 bio_offset)
824 {
825         /*
826          * when we're called for a write, we're already in the async
827          * submission context.  Just jump into btrfs_map_bio
828          */
829         btree_csum_one_bio(bio);
830         return 0;
831 }
832
833 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
834                                  int mirror_num, unsigned long bio_flags,
835                                  u64 bio_offset)
836 {
837         /*
838          * when we're called for a write, we're already in the async
839          * submission context.  Just jump into btrfs_map_bio
840          */
841         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
842 }
843
844 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
845                                  int mirror_num, unsigned long bio_flags,
846                                  u64 bio_offset)
847 {
848         int ret;
849
850         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
851                                           bio, 1);
852         BUG_ON(ret);
853
854         if (!(rw & REQ_WRITE)) {
855                 /*
856                  * called for a read, do the setup so that checksum validation
857                  * can happen in the async kernel threads
858                  */
859                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
860                                      mirror_num, 0);
861         }
862
863         /*
864          * kthread helpers are used to submit writes so that checksumming
865          * can happen in parallel across all CPUs
866          */
867         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
868                                    inode, rw, bio, mirror_num, 0,
869                                    bio_offset,
870                                    __btree_submit_bio_start,
871                                    __btree_submit_bio_done);
872 }
873
874 #ifdef CONFIG_MIGRATION
875 static int btree_migratepage(struct address_space *mapping,
876                         struct page *newpage, struct page *page,
877                         enum migrate_mode mode)
878 {
879         /*
880          * we can't safely write a btree page from here,
881          * we haven't done the locking hook
882          */
883         if (PageDirty(page))
884                 return -EAGAIN;
885         /*
886          * Buffers may be managed in a filesystem specific way.
887          * We must have no buffers or drop them.
888          */
889         if (page_has_private(page) &&
890             !try_to_release_page(page, GFP_KERNEL))
891                 return -EAGAIN;
892         return migrate_page(mapping, newpage, page, mode);
893 }
894 #endif
895
896 static int btree_writepage(struct page *page, struct writeback_control *wbc)
897 {
898         struct extent_io_tree *tree;
899         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
900         struct extent_buffer *eb;
901         int was_dirty;
902
903         tree = &BTRFS_I(page->mapping->host)->io_tree;
904         if (!(current->flags & PF_MEMALLOC)) {
905                 return extent_write_full_page(tree, page,
906                                               btree_get_extent, wbc);
907         }
908
909         redirty_page_for_writepage(wbc, page);
910         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
911         WARN_ON(!eb);
912
913         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
914         if (!was_dirty) {
915                 spin_lock(&root->fs_info->delalloc_lock);
916                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
917                 spin_unlock(&root->fs_info->delalloc_lock);
918         }
919         free_extent_buffer(eb);
920
921         unlock_page(page);
922         return 0;
923 }
924
925 static int btree_writepages(struct address_space *mapping,
926                             struct writeback_control *wbc)
927 {
928         struct extent_io_tree *tree;
929         tree = &BTRFS_I(mapping->host)->io_tree;
930         if (wbc->sync_mode == WB_SYNC_NONE) {
931                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
932                 u64 num_dirty;
933                 unsigned long thresh = 32 * 1024 * 1024;
934
935                 if (wbc->for_kupdate)
936                         return 0;
937
938                 /* this is a bit racy, but that's ok */
939                 num_dirty = root->fs_info->dirty_metadata_bytes;
940                 if (num_dirty < thresh)
941                         return 0;
942         }
943         return extent_writepages(tree, mapping, btree_get_extent, wbc);
944 }
945
946 static int btree_readpage(struct file *file, struct page *page)
947 {
948         struct extent_io_tree *tree;
949         tree = &BTRFS_I(page->mapping->host)->io_tree;
950         return extent_read_full_page(tree, page, btree_get_extent, 0);
951 }
952
953 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
954 {
955         struct extent_io_tree *tree;
956         struct extent_map_tree *map;
957         int ret;
958
959         if (PageWriteback(page) || PageDirty(page))
960                 return 0;
961
962         tree = &BTRFS_I(page->mapping->host)->io_tree;
963         map = &BTRFS_I(page->mapping->host)->extent_tree;
964
965         ret = try_release_extent_state(map, tree, page, gfp_flags);
966         if (!ret)
967                 return 0;
968
969         ret = try_release_extent_buffer(tree, page);
970         if (ret == 1) {
971                 ClearPagePrivate(page);
972                 set_page_private(page, 0);
973                 page_cache_release(page);
974         }
975
976         return ret;
977 }
978
979 static void btree_invalidatepage(struct page *page, unsigned long offset)
980 {
981         struct extent_io_tree *tree;
982         tree = &BTRFS_I(page->mapping->host)->io_tree;
983         extent_invalidatepage(tree, page, offset);
984         btree_releasepage(page, GFP_NOFS);
985         if (PagePrivate(page)) {
986                 printk(KERN_WARNING "btrfs warning page private not zero "
987                        "on page %llu\n", (unsigned long long)page_offset(page));
988                 ClearPagePrivate(page);
989                 set_page_private(page, 0);
990                 page_cache_release(page);
991         }
992 }
993
994 static const struct address_space_operations btree_aops = {
995         .readpage       = btree_readpage,
996         .writepage      = btree_writepage,
997         .writepages     = btree_writepages,
998         .releasepage    = btree_releasepage,
999         .invalidatepage = btree_invalidatepage,
1000 #ifdef CONFIG_MIGRATION
1001         .migratepage    = btree_migratepage,
1002 #endif
1003 };
1004
1005 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1006                          u64 parent_transid)
1007 {
1008         struct extent_buffer *buf = NULL;
1009         struct inode *btree_inode = root->fs_info->btree_inode;
1010         int ret = 0;
1011
1012         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1013         if (!buf)
1014                 return 0;
1015         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1016                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1017         free_extent_buffer(buf);
1018         return ret;
1019 }
1020
1021 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1022                          int mirror_num, struct extent_buffer **eb)
1023 {
1024         struct extent_buffer *buf = NULL;
1025         struct inode *btree_inode = root->fs_info->btree_inode;
1026         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1027         int ret;
1028
1029         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1030         if (!buf)
1031                 return 0;
1032
1033         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1034
1035         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1036                                        btree_get_extent, mirror_num);
1037         if (ret) {
1038                 free_extent_buffer(buf);
1039                 return ret;
1040         }
1041
1042         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1043                 free_extent_buffer(buf);
1044                 return -EIO;
1045         } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1046                 *eb = buf;
1047         } else {
1048                 free_extent_buffer(buf);
1049         }
1050         return 0;
1051 }
1052
1053 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1054                                             u64 bytenr, u32 blocksize)
1055 {
1056         struct inode *btree_inode = root->fs_info->btree_inode;
1057         struct extent_buffer *eb;
1058         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1059                                 bytenr, blocksize);
1060         return eb;
1061 }
1062
1063 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1064                                                  u64 bytenr, u32 blocksize)
1065 {
1066         struct inode *btree_inode = root->fs_info->btree_inode;
1067         struct extent_buffer *eb;
1068
1069         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1070                                  bytenr, blocksize, NULL);
1071         return eb;
1072 }
1073
1074
1075 int btrfs_write_tree_block(struct extent_buffer *buf)
1076 {
1077         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1078                                         buf->start + buf->len - 1);
1079 }
1080
1081 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1082 {
1083         return filemap_fdatawait_range(buf->first_page->mapping,
1084                                        buf->start, buf->start + buf->len - 1);
1085 }
1086
1087 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1088                                       u32 blocksize, u64 parent_transid)
1089 {
1090         struct extent_buffer *buf = NULL;
1091         int ret;
1092
1093         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1094         if (!buf)
1095                 return NULL;
1096
1097         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1098
1099         if (ret == 0)
1100                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1101         return buf;
1102
1103 }
1104
1105 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1106                      struct extent_buffer *buf)
1107 {
1108         struct inode *btree_inode = root->fs_info->btree_inode;
1109         if (btrfs_header_generation(buf) ==
1110             root->fs_info->running_transaction->transid) {
1111                 btrfs_assert_tree_locked(buf);
1112
1113                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1114                         spin_lock(&root->fs_info->delalloc_lock);
1115                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1116                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1117                         else
1118                                 WARN_ON(1);
1119                         spin_unlock(&root->fs_info->delalloc_lock);
1120                 }
1121
1122                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1123                 btrfs_set_lock_blocking(buf);
1124                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1125                                           buf);
1126         }
1127         return 0;
1128 }
1129
1130 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1131                         u32 stripesize, struct btrfs_root *root,
1132                         struct btrfs_fs_info *fs_info,
1133                         u64 objectid)
1134 {
1135         root->node = NULL;
1136         root->commit_root = NULL;
1137         root->sectorsize = sectorsize;
1138         root->nodesize = nodesize;
1139         root->leafsize = leafsize;
1140         root->stripesize = stripesize;
1141         root->ref_cows = 0;
1142         root->track_dirty = 0;
1143         root->in_radix = 0;
1144         root->orphan_item_inserted = 0;
1145         root->orphan_cleanup_state = 0;
1146
1147         root->objectid = objectid;
1148         root->last_trans = 0;
1149         root->highest_objectid = 0;
1150         root->name = NULL;
1151         root->inode_tree = RB_ROOT;
1152         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1153         root->block_rsv = NULL;
1154         root->orphan_block_rsv = NULL;
1155
1156         INIT_LIST_HEAD(&root->dirty_list);
1157         INIT_LIST_HEAD(&root->orphan_list);
1158         INIT_LIST_HEAD(&root->root_list);
1159         spin_lock_init(&root->orphan_lock);
1160         spin_lock_init(&root->inode_lock);
1161         spin_lock_init(&root->accounting_lock);
1162         mutex_init(&root->objectid_mutex);
1163         mutex_init(&root->log_mutex);
1164         init_waitqueue_head(&root->log_writer_wait);
1165         init_waitqueue_head(&root->log_commit_wait[0]);
1166         init_waitqueue_head(&root->log_commit_wait[1]);
1167         atomic_set(&root->log_commit[0], 0);
1168         atomic_set(&root->log_commit[1], 0);
1169         atomic_set(&root->log_writers, 0);
1170         root->log_batch = 0;
1171         root->log_transid = 0;
1172         root->last_log_commit = 0;
1173         extent_io_tree_init(&root->dirty_log_pages,
1174                              fs_info->btree_inode->i_mapping);
1175
1176         memset(&root->root_key, 0, sizeof(root->root_key));
1177         memset(&root->root_item, 0, sizeof(root->root_item));
1178         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1179         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1180         root->defrag_trans_start = fs_info->generation;
1181         init_completion(&root->kobj_unregister);
1182         root->defrag_running = 0;
1183         root->root_key.objectid = objectid;
1184         root->anon_dev = 0;
1185         return 0;
1186 }
1187
1188 static int find_and_setup_root(struct btrfs_root *tree_root,
1189                                struct btrfs_fs_info *fs_info,
1190                                u64 objectid,
1191                                struct btrfs_root *root)
1192 {
1193         int ret;
1194         u32 blocksize;
1195         u64 generation;
1196
1197         __setup_root(tree_root->nodesize, tree_root->leafsize,
1198                      tree_root->sectorsize, tree_root->stripesize,
1199                      root, fs_info, objectid);
1200         ret = btrfs_find_last_root(tree_root, objectid,
1201                                    &root->root_item, &root->root_key);
1202         if (ret > 0)
1203                 return -ENOENT;
1204         BUG_ON(ret);
1205
1206         generation = btrfs_root_generation(&root->root_item);
1207         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1208         root->commit_root = NULL;
1209         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1210                                      blocksize, generation);
1211         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1212                 free_extent_buffer(root->node);
1213                 root->node = NULL;
1214                 return -EIO;
1215         }
1216         root->commit_root = btrfs_root_node(root);
1217         return 0;
1218 }
1219
1220 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1221 {
1222         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1223         if (root)
1224                 root->fs_info = fs_info;
1225         return root;
1226 }
1227
1228 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1229                                          struct btrfs_fs_info *fs_info)
1230 {
1231         struct btrfs_root *root;
1232         struct btrfs_root *tree_root = fs_info->tree_root;
1233         struct extent_buffer *leaf;
1234
1235         root = btrfs_alloc_root(fs_info);
1236         if (!root)
1237                 return ERR_PTR(-ENOMEM);
1238
1239         __setup_root(tree_root->nodesize, tree_root->leafsize,
1240                      tree_root->sectorsize, tree_root->stripesize,
1241                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1242
1243         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1244         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1245         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1246         /*
1247          * log trees do not get reference counted because they go away
1248          * before a real commit is actually done.  They do store pointers
1249          * to file data extents, and those reference counts still get
1250          * updated (along with back refs to the log tree).
1251          */
1252         root->ref_cows = 0;
1253
1254         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1255                                       BTRFS_TREE_LOG_OBJECTID, NULL,
1256                                       0, 0, 0, 0);
1257         if (IS_ERR(leaf)) {
1258                 kfree(root);
1259                 return ERR_CAST(leaf);
1260         }
1261
1262         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1263         btrfs_set_header_bytenr(leaf, leaf->start);
1264         btrfs_set_header_generation(leaf, trans->transid);
1265         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1266         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1267         root->node = leaf;
1268
1269         write_extent_buffer(root->node, root->fs_info->fsid,
1270                             (unsigned long)btrfs_header_fsid(root->node),
1271                             BTRFS_FSID_SIZE);
1272         btrfs_mark_buffer_dirty(root->node);
1273         btrfs_tree_unlock(root->node);
1274         return root;
1275 }
1276
1277 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1278                              struct btrfs_fs_info *fs_info)
1279 {
1280         struct btrfs_root *log_root;
1281
1282         log_root = alloc_log_tree(trans, fs_info);
1283         if (IS_ERR(log_root))
1284                 return PTR_ERR(log_root);
1285         WARN_ON(fs_info->log_root_tree);
1286         fs_info->log_root_tree = log_root;
1287         return 0;
1288 }
1289
1290 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1291                        struct btrfs_root *root)
1292 {
1293         struct btrfs_root *log_root;
1294         struct btrfs_inode_item *inode_item;
1295
1296         log_root = alloc_log_tree(trans, root->fs_info);
1297         if (IS_ERR(log_root))
1298                 return PTR_ERR(log_root);
1299
1300         log_root->last_trans = trans->transid;
1301         log_root->root_key.offset = root->root_key.objectid;
1302
1303         inode_item = &log_root->root_item.inode;
1304         inode_item->generation = cpu_to_le64(1);
1305         inode_item->size = cpu_to_le64(3);
1306         inode_item->nlink = cpu_to_le32(1);
1307         inode_item->nbytes = cpu_to_le64(root->leafsize);
1308         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1309
1310         btrfs_set_root_node(&log_root->root_item, log_root->node);
1311
1312         WARN_ON(root->log_root);
1313         root->log_root = log_root;
1314         root->log_transid = 0;
1315         root->last_log_commit = 0;
1316         return 0;
1317 }
1318
1319 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1320                                                struct btrfs_key *location)
1321 {
1322         struct btrfs_root *root;
1323         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1324         struct btrfs_path *path;
1325         struct extent_buffer *l;
1326         u64 generation;
1327         u32 blocksize;
1328         int ret = 0;
1329
1330         root = btrfs_alloc_root(fs_info);
1331         if (!root)
1332                 return ERR_PTR(-ENOMEM);
1333         if (location->offset == (u64)-1) {
1334                 ret = find_and_setup_root(tree_root, fs_info,
1335                                           location->objectid, root);
1336                 if (ret) {
1337                         kfree(root);
1338                         return ERR_PTR(ret);
1339                 }
1340                 goto out;
1341         }
1342
1343         __setup_root(tree_root->nodesize, tree_root->leafsize,
1344                      tree_root->sectorsize, tree_root->stripesize,
1345                      root, fs_info, location->objectid);
1346
1347         path = btrfs_alloc_path();
1348         if (!path) {
1349                 kfree(root);
1350                 return ERR_PTR(-ENOMEM);
1351         }
1352         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1353         if (ret == 0) {
1354                 l = path->nodes[0];
1355                 read_extent_buffer(l, &root->root_item,
1356                                 btrfs_item_ptr_offset(l, path->slots[0]),
1357                                 sizeof(root->root_item));
1358                 memcpy(&root->root_key, location, sizeof(*location));
1359         }
1360         btrfs_free_path(path);
1361         if (ret) {
1362                 kfree(root);
1363                 if (ret > 0)
1364                         ret = -ENOENT;
1365                 return ERR_PTR(ret);
1366         }
1367
1368         generation = btrfs_root_generation(&root->root_item);
1369         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1370         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1371                                      blocksize, generation);
1372         root->commit_root = btrfs_root_node(root);
1373         BUG_ON(!root->node);
1374 out:
1375         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1376                 root->ref_cows = 1;
1377                 btrfs_check_and_init_root_item(&root->root_item);
1378         }
1379
1380         return root;
1381 }
1382
1383 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1384                                               struct btrfs_key *location)
1385 {
1386         struct btrfs_root *root;
1387         int ret;
1388
1389         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1390                 return fs_info->tree_root;
1391         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1392                 return fs_info->extent_root;
1393         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1394                 return fs_info->chunk_root;
1395         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1396                 return fs_info->dev_root;
1397         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1398                 return fs_info->csum_root;
1399 again:
1400         spin_lock(&fs_info->fs_roots_radix_lock);
1401         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1402                                  (unsigned long)location->objectid);
1403         spin_unlock(&fs_info->fs_roots_radix_lock);
1404         if (root)
1405                 return root;
1406
1407         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1408         if (IS_ERR(root))
1409                 return root;
1410
1411         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1412         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1413                                         GFP_NOFS);
1414         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1415                 ret = -ENOMEM;
1416                 goto fail;
1417         }
1418
1419         btrfs_init_free_ino_ctl(root);
1420         mutex_init(&root->fs_commit_mutex);
1421         spin_lock_init(&root->cache_lock);
1422         init_waitqueue_head(&root->cache_wait);
1423
1424         ret = get_anon_bdev(&root->anon_dev);
1425         if (ret)
1426                 goto fail;
1427
1428         if (btrfs_root_refs(&root->root_item) == 0) {
1429                 ret = -ENOENT;
1430                 goto fail;
1431         }
1432
1433         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1434         if (ret < 0)
1435                 goto fail;
1436         if (ret == 0)
1437                 root->orphan_item_inserted = 1;
1438
1439         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1440         if (ret)
1441                 goto fail;
1442
1443         spin_lock(&fs_info->fs_roots_radix_lock);
1444         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1445                                 (unsigned long)root->root_key.objectid,
1446                                 root);
1447         if (ret == 0)
1448                 root->in_radix = 1;
1449
1450         spin_unlock(&fs_info->fs_roots_radix_lock);
1451         radix_tree_preload_end();
1452         if (ret) {
1453                 if (ret == -EEXIST) {
1454                         free_fs_root(root);
1455                         goto again;
1456                 }
1457                 goto fail;
1458         }
1459
1460         ret = btrfs_find_dead_roots(fs_info->tree_root,
1461                                     root->root_key.objectid);
1462         WARN_ON(ret);
1463         return root;
1464 fail:
1465         free_fs_root(root);
1466         return ERR_PTR(ret);
1467 }
1468
1469 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1470 {
1471         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1472         int ret = 0;
1473         struct btrfs_device *device;
1474         struct backing_dev_info *bdi;
1475
1476         rcu_read_lock();
1477         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1478                 if (!device->bdev)
1479                         continue;
1480                 bdi = blk_get_backing_dev_info(device->bdev);
1481                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1482                         ret = 1;
1483                         break;
1484                 }
1485         }
1486         rcu_read_unlock();
1487         return ret;
1488 }
1489
1490 /*
1491  * If this fails, caller must call bdi_destroy() to get rid of the
1492  * bdi again.
1493  */
1494 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1495 {
1496         int err;
1497
1498         bdi->capabilities = BDI_CAP_MAP_COPY;
1499         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1500         if (err)
1501                 return err;
1502
1503         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1504         bdi->congested_fn       = btrfs_congested_fn;
1505         bdi->congested_data     = info;
1506         return 0;
1507 }
1508
1509 static int bio_ready_for_csum(struct bio *bio)
1510 {
1511         u64 length = 0;
1512         u64 buf_len = 0;
1513         u64 start = 0;
1514         struct page *page;
1515         struct extent_io_tree *io_tree = NULL;
1516         struct bio_vec *bvec;
1517         int i;
1518         int ret;
1519
1520         bio_for_each_segment(bvec, bio, i) {
1521                 page = bvec->bv_page;
1522                 if (page->private == EXTENT_PAGE_PRIVATE) {
1523                         length += bvec->bv_len;
1524                         continue;
1525                 }
1526                 if (!page->private) {
1527                         length += bvec->bv_len;
1528                         continue;
1529                 }
1530                 length = bvec->bv_len;
1531                 buf_len = page->private >> 2;
1532                 start = page_offset(page) + bvec->bv_offset;
1533                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1534         }
1535         /* are we fully contained in this bio? */
1536         if (buf_len <= length)
1537                 return 1;
1538
1539         ret = extent_range_uptodate(io_tree, start + length,
1540                                     start + buf_len - 1);
1541         return ret;
1542 }
1543
1544 /*
1545  * called by the kthread helper functions to finally call the bio end_io
1546  * functions.  This is where read checksum verification actually happens
1547  */
1548 static void end_workqueue_fn(struct btrfs_work *work)
1549 {
1550         struct bio *bio;
1551         struct end_io_wq *end_io_wq;
1552         struct btrfs_fs_info *fs_info;
1553         int error;
1554
1555         end_io_wq = container_of(work, struct end_io_wq, work);
1556         bio = end_io_wq->bio;
1557         fs_info = end_io_wq->info;
1558
1559         /* metadata bio reads are special because the whole tree block must
1560          * be checksummed at once.  This makes sure the entire block is in
1561          * ram and up to date before trying to verify things.  For
1562          * blocksize <= pagesize, it is basically a noop
1563          */
1564         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1565             !bio_ready_for_csum(bio)) {
1566                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1567                                    &end_io_wq->work);
1568                 return;
1569         }
1570         error = end_io_wq->error;
1571         bio->bi_private = end_io_wq->private;
1572         bio->bi_end_io = end_io_wq->end_io;
1573         kfree(end_io_wq);
1574         bio_endio(bio, error);
1575 }
1576
1577 static int cleaner_kthread(void *arg)
1578 {
1579         struct btrfs_root *root = arg;
1580
1581         do {
1582                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1583
1584                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1585                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1586                         btrfs_run_delayed_iputs(root);
1587                         btrfs_clean_old_snapshots(root);
1588                         mutex_unlock(&root->fs_info->cleaner_mutex);
1589                         btrfs_run_defrag_inodes(root->fs_info);
1590                 }
1591
1592                 if (!try_to_freeze()) {
1593                         set_current_state(TASK_INTERRUPTIBLE);
1594                         if (!kthread_should_stop())
1595                                 schedule();
1596                         __set_current_state(TASK_RUNNING);
1597                 }
1598         } while (!kthread_should_stop());
1599         return 0;
1600 }
1601
1602 static int transaction_kthread(void *arg)
1603 {
1604         struct btrfs_root *root = arg;
1605         struct btrfs_trans_handle *trans;
1606         struct btrfs_transaction *cur;
1607         u64 transid;
1608         unsigned long now;
1609         unsigned long delay;
1610         int ret;
1611
1612         do {
1613                 delay = HZ * 30;
1614                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1615                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1616
1617                 spin_lock(&root->fs_info->trans_lock);
1618                 cur = root->fs_info->running_transaction;
1619                 if (!cur) {
1620                         spin_unlock(&root->fs_info->trans_lock);
1621                         goto sleep;
1622                 }
1623
1624                 now = get_seconds();
1625                 if (!cur->blocked &&
1626                     (now < cur->start_time || now - cur->start_time < 30)) {
1627                         spin_unlock(&root->fs_info->trans_lock);
1628                         delay = HZ * 5;
1629                         goto sleep;
1630                 }
1631                 transid = cur->transid;
1632                 spin_unlock(&root->fs_info->trans_lock);
1633
1634                 trans = btrfs_join_transaction(root);
1635                 BUG_ON(IS_ERR(trans));
1636                 if (transid == trans->transid) {
1637                         ret = btrfs_commit_transaction(trans, root);
1638                         BUG_ON(ret);
1639                 } else {
1640                         btrfs_end_transaction(trans, root);
1641                 }
1642 sleep:
1643                 wake_up_process(root->fs_info->cleaner_kthread);
1644                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1645
1646                 if (!try_to_freeze()) {
1647                         set_current_state(TASK_INTERRUPTIBLE);
1648                         if (!kthread_should_stop() &&
1649                             !btrfs_transaction_blocked(root->fs_info))
1650                                 schedule_timeout(delay);
1651                         __set_current_state(TASK_RUNNING);
1652                 }
1653         } while (!kthread_should_stop());
1654         return 0;
1655 }
1656
1657 /*
1658  * this will find the highest generation in the array of
1659  * root backups.  The index of the highest array is returned,
1660  * or -1 if we can't find anything.
1661  *
1662  * We check to make sure the array is valid by comparing the
1663  * generation of the latest  root in the array with the generation
1664  * in the super block.  If they don't match we pitch it.
1665  */
1666 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1667 {
1668         u64 cur;
1669         int newest_index = -1;
1670         struct btrfs_root_backup *root_backup;
1671         int i;
1672
1673         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1674                 root_backup = info->super_copy->super_roots + i;
1675                 cur = btrfs_backup_tree_root_gen(root_backup);
1676                 if (cur == newest_gen)
1677                         newest_index = i;
1678         }
1679
1680         /* check to see if we actually wrapped around */
1681         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1682                 root_backup = info->super_copy->super_roots;
1683                 cur = btrfs_backup_tree_root_gen(root_backup);
1684                 if (cur == newest_gen)
1685                         newest_index = 0;
1686         }
1687         return newest_index;
1688 }
1689
1690
1691 /*
1692  * find the oldest backup so we know where to store new entries
1693  * in the backup array.  This will set the backup_root_index
1694  * field in the fs_info struct
1695  */
1696 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1697                                      u64 newest_gen)
1698 {
1699         int newest_index = -1;
1700
1701         newest_index = find_newest_super_backup(info, newest_gen);
1702         /* if there was garbage in there, just move along */
1703         if (newest_index == -1) {
1704                 info->backup_root_index = 0;
1705         } else {
1706                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1707         }
1708 }
1709
1710 /*
1711  * copy all the root pointers into the super backup array.
1712  * this will bump the backup pointer by one when it is
1713  * done
1714  */
1715 static void backup_super_roots(struct btrfs_fs_info *info)
1716 {
1717         int next_backup;
1718         struct btrfs_root_backup *root_backup;
1719         int last_backup;
1720
1721         next_backup = info->backup_root_index;
1722         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1723                 BTRFS_NUM_BACKUP_ROOTS;
1724
1725         /*
1726          * just overwrite the last backup if we're at the same generation
1727          * this happens only at umount
1728          */
1729         root_backup = info->super_for_commit->super_roots + last_backup;
1730         if (btrfs_backup_tree_root_gen(root_backup) ==
1731             btrfs_header_generation(info->tree_root->node))
1732                 next_backup = last_backup;
1733
1734         root_backup = info->super_for_commit->super_roots + next_backup;
1735
1736         /*
1737          * make sure all of our padding and empty slots get zero filled
1738          * regardless of which ones we use today
1739          */
1740         memset(root_backup, 0, sizeof(*root_backup));
1741
1742         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1743
1744         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1745         btrfs_set_backup_tree_root_gen(root_backup,
1746                                btrfs_header_generation(info->tree_root->node));
1747
1748         btrfs_set_backup_tree_root_level(root_backup,
1749                                btrfs_header_level(info->tree_root->node));
1750
1751         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1752         btrfs_set_backup_chunk_root_gen(root_backup,
1753                                btrfs_header_generation(info->chunk_root->node));
1754         btrfs_set_backup_chunk_root_level(root_backup,
1755                                btrfs_header_level(info->chunk_root->node));
1756
1757         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1758         btrfs_set_backup_extent_root_gen(root_backup,
1759                                btrfs_header_generation(info->extent_root->node));
1760         btrfs_set_backup_extent_root_level(root_backup,
1761                                btrfs_header_level(info->extent_root->node));
1762
1763         /*
1764          * we might commit during log recovery, which happens before we set
1765          * the fs_root.  Make sure it is valid before we fill it in.
1766          */
1767         if (info->fs_root && info->fs_root->node) {
1768                 btrfs_set_backup_fs_root(root_backup,
1769                                          info->fs_root->node->start);
1770                 btrfs_set_backup_fs_root_gen(root_backup,
1771                                btrfs_header_generation(info->fs_root->node));
1772                 btrfs_set_backup_fs_root_level(root_backup,
1773                                btrfs_header_level(info->fs_root->node));
1774         }
1775
1776         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1777         btrfs_set_backup_dev_root_gen(root_backup,
1778                                btrfs_header_generation(info->dev_root->node));
1779         btrfs_set_backup_dev_root_level(root_backup,
1780                                        btrfs_header_level(info->dev_root->node));
1781
1782         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1783         btrfs_set_backup_csum_root_gen(root_backup,
1784                                btrfs_header_generation(info->csum_root->node));
1785         btrfs_set_backup_csum_root_level(root_backup,
1786                                btrfs_header_level(info->csum_root->node));
1787
1788         btrfs_set_backup_total_bytes(root_backup,
1789                              btrfs_super_total_bytes(info->super_copy));
1790         btrfs_set_backup_bytes_used(root_backup,
1791                              btrfs_super_bytes_used(info->super_copy));
1792         btrfs_set_backup_num_devices(root_backup,
1793                              btrfs_super_num_devices(info->super_copy));
1794
1795         /*
1796          * if we don't copy this out to the super_copy, it won't get remembered
1797          * for the next commit
1798          */
1799         memcpy(&info->super_copy->super_roots,
1800                &info->super_for_commit->super_roots,
1801                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1802 }
1803
1804 /*
1805  * this copies info out of the root backup array and back into
1806  * the in-memory super block.  It is meant to help iterate through
1807  * the array, so you send it the number of backups you've already
1808  * tried and the last backup index you used.
1809  *
1810  * this returns -1 when it has tried all the backups
1811  */
1812 static noinline int next_root_backup(struct btrfs_fs_info *info,
1813                                      struct btrfs_super_block *super,
1814                                      int *num_backups_tried, int *backup_index)
1815 {
1816         struct btrfs_root_backup *root_backup;
1817         int newest = *backup_index;
1818
1819         if (*num_backups_tried == 0) {
1820                 u64 gen = btrfs_super_generation(super);
1821
1822                 newest = find_newest_super_backup(info, gen);
1823                 if (newest == -1)
1824                         return -1;
1825
1826                 *backup_index = newest;
1827                 *num_backups_tried = 1;
1828         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1829                 /* we've tried all the backups, all done */
1830                 return -1;
1831         } else {
1832                 /* jump to the next oldest backup */
1833                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1834                         BTRFS_NUM_BACKUP_ROOTS;
1835                 *backup_index = newest;
1836                 *num_backups_tried += 1;
1837         }
1838         root_backup = super->super_roots + newest;
1839
1840         btrfs_set_super_generation(super,
1841                                    btrfs_backup_tree_root_gen(root_backup));
1842         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1843         btrfs_set_super_root_level(super,
1844                                    btrfs_backup_tree_root_level(root_backup));
1845         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1846
1847         /*
1848          * fixme: the total bytes and num_devices need to match or we should
1849          * need a fsck
1850          */
1851         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1852         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1853         return 0;
1854 }
1855
1856 /* helper to cleanup tree roots */
1857 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1858 {
1859         free_extent_buffer(info->tree_root->node);
1860         free_extent_buffer(info->tree_root->commit_root);
1861         free_extent_buffer(info->dev_root->node);
1862         free_extent_buffer(info->dev_root->commit_root);
1863         free_extent_buffer(info->extent_root->node);
1864         free_extent_buffer(info->extent_root->commit_root);
1865         free_extent_buffer(info->csum_root->node);
1866         free_extent_buffer(info->csum_root->commit_root);
1867
1868         info->tree_root->node = NULL;
1869         info->tree_root->commit_root = NULL;
1870         info->dev_root->node = NULL;
1871         info->dev_root->commit_root = NULL;
1872         info->extent_root->node = NULL;
1873         info->extent_root->commit_root = NULL;
1874         info->csum_root->node = NULL;
1875         info->csum_root->commit_root = NULL;
1876
1877         if (chunk_root) {
1878                 free_extent_buffer(info->chunk_root->node);
1879                 free_extent_buffer(info->chunk_root->commit_root);
1880                 info->chunk_root->node = NULL;
1881                 info->chunk_root->commit_root = NULL;
1882         }
1883 }
1884
1885
1886 int open_ctree(struct super_block *sb,
1887                struct btrfs_fs_devices *fs_devices,
1888                char *options)
1889 {
1890         u32 sectorsize;
1891         u32 nodesize;
1892         u32 leafsize;
1893         u32 blocksize;
1894         u32 stripesize;
1895         u64 generation;
1896         u64 features;
1897         struct btrfs_key location;
1898         struct buffer_head *bh;
1899         struct btrfs_super_block *disk_super;
1900         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1901         struct btrfs_root *tree_root;
1902         struct btrfs_root *extent_root;
1903         struct btrfs_root *csum_root;
1904         struct btrfs_root *chunk_root;
1905         struct btrfs_root *dev_root;
1906         struct btrfs_root *log_tree_root;
1907         int ret;
1908         int err = -EINVAL;
1909         int num_backups_tried = 0;
1910         int backup_index = 0;
1911
1912         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1913         extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1914         csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1915         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1916         dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1917
1918         if (!tree_root || !extent_root || !csum_root ||
1919             !chunk_root || !dev_root) {
1920                 err = -ENOMEM;
1921                 goto fail;
1922         }
1923
1924         ret = init_srcu_struct(&fs_info->subvol_srcu);
1925         if (ret) {
1926                 err = ret;
1927                 goto fail;
1928         }
1929
1930         ret = setup_bdi(fs_info, &fs_info->bdi);
1931         if (ret) {
1932                 err = ret;
1933                 goto fail_srcu;
1934         }
1935
1936         fs_info->btree_inode = new_inode(sb);
1937         if (!fs_info->btree_inode) {
1938                 err = -ENOMEM;
1939                 goto fail_bdi;
1940         }
1941
1942         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1943
1944         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1945         INIT_LIST_HEAD(&fs_info->trans_list);
1946         INIT_LIST_HEAD(&fs_info->dead_roots);
1947         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1948         INIT_LIST_HEAD(&fs_info->hashers);
1949         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1950         INIT_LIST_HEAD(&fs_info->ordered_operations);
1951         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1952         spin_lock_init(&fs_info->delalloc_lock);
1953         spin_lock_init(&fs_info->trans_lock);
1954         spin_lock_init(&fs_info->ref_cache_lock);
1955         spin_lock_init(&fs_info->fs_roots_radix_lock);
1956         spin_lock_init(&fs_info->delayed_iput_lock);
1957         spin_lock_init(&fs_info->defrag_inodes_lock);
1958         spin_lock_init(&fs_info->free_chunk_lock);
1959         mutex_init(&fs_info->reloc_mutex);
1960
1961         init_completion(&fs_info->kobj_unregister);
1962         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1963         INIT_LIST_HEAD(&fs_info->space_info);
1964         btrfs_mapping_init(&fs_info->mapping_tree);
1965         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1966         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1967         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1968         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1969         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1970         btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1971         atomic_set(&fs_info->nr_async_submits, 0);
1972         atomic_set(&fs_info->async_delalloc_pages, 0);
1973         atomic_set(&fs_info->async_submit_draining, 0);
1974         atomic_set(&fs_info->nr_async_bios, 0);
1975         atomic_set(&fs_info->defrag_running, 0);
1976         fs_info->sb = sb;
1977         fs_info->max_inline = 8192 * 1024;
1978         fs_info->metadata_ratio = 0;
1979         fs_info->defrag_inodes = RB_ROOT;
1980         fs_info->trans_no_join = 0;
1981         fs_info->free_chunk_space = 0;
1982
1983         /* readahead state */
1984         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1985         spin_lock_init(&fs_info->reada_lock);
1986
1987         fs_info->thread_pool_size = min_t(unsigned long,
1988                                           num_online_cpus() + 2, 8);
1989
1990         INIT_LIST_HEAD(&fs_info->ordered_extents);
1991         spin_lock_init(&fs_info->ordered_extent_lock);
1992         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1993                                         GFP_NOFS);
1994         if (!fs_info->delayed_root) {
1995                 err = -ENOMEM;
1996                 goto fail_iput;
1997         }
1998         btrfs_init_delayed_root(fs_info->delayed_root);
1999
2000         mutex_init(&fs_info->scrub_lock);
2001         atomic_set(&fs_info->scrubs_running, 0);
2002         atomic_set(&fs_info->scrub_pause_req, 0);
2003         atomic_set(&fs_info->scrubs_paused, 0);
2004         atomic_set(&fs_info->scrub_cancel_req, 0);
2005         init_waitqueue_head(&fs_info->scrub_pause_wait);
2006         init_rwsem(&fs_info->scrub_super_lock);
2007         fs_info->scrub_workers_refcnt = 0;
2008 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2009         fs_info->check_integrity_print_mask = 0;
2010 #endif
2011
2012         spin_lock_init(&fs_info->balance_lock);
2013         mutex_init(&fs_info->balance_mutex);
2014         atomic_set(&fs_info->balance_running, 0);
2015         atomic_set(&fs_info->balance_pause_req, 0);
2016         atomic_set(&fs_info->balance_cancel_req, 0);
2017         fs_info->balance_ctl = NULL;
2018         init_waitqueue_head(&fs_info->balance_wait_q);
2019
2020         sb->s_blocksize = 4096;
2021         sb->s_blocksize_bits = blksize_bits(4096);
2022         sb->s_bdi = &fs_info->bdi;
2023
2024         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2025         set_nlink(fs_info->btree_inode, 1);
2026         /*
2027          * we set the i_size on the btree inode to the max possible int.
2028          * the real end of the address space is determined by all of
2029          * the devices in the system
2030          */
2031         fs_info->btree_inode->i_size = OFFSET_MAX;
2032         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2033         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2034
2035         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2036         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2037                              fs_info->btree_inode->i_mapping);
2038         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2039
2040         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2041
2042         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2043         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2044                sizeof(struct btrfs_key));
2045         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
2046         insert_inode_hash(fs_info->btree_inode);
2047
2048         spin_lock_init(&fs_info->block_group_cache_lock);
2049         fs_info->block_group_cache_tree = RB_ROOT;
2050
2051         extent_io_tree_init(&fs_info->freed_extents[0],
2052                              fs_info->btree_inode->i_mapping);
2053         extent_io_tree_init(&fs_info->freed_extents[1],
2054                              fs_info->btree_inode->i_mapping);
2055         fs_info->pinned_extents = &fs_info->freed_extents[0];
2056         fs_info->do_barriers = 1;
2057
2058
2059         mutex_init(&fs_info->ordered_operations_mutex);
2060         mutex_init(&fs_info->tree_log_mutex);
2061         mutex_init(&fs_info->chunk_mutex);
2062         mutex_init(&fs_info->transaction_kthread_mutex);
2063         mutex_init(&fs_info->cleaner_mutex);
2064         mutex_init(&fs_info->volume_mutex);
2065         init_rwsem(&fs_info->extent_commit_sem);
2066         init_rwsem(&fs_info->cleanup_work_sem);
2067         init_rwsem(&fs_info->subvol_sem);
2068
2069         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2070         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2071
2072         init_waitqueue_head(&fs_info->transaction_throttle);
2073         init_waitqueue_head(&fs_info->transaction_wait);
2074         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2075         init_waitqueue_head(&fs_info->async_submit_wait);
2076
2077         __setup_root(4096, 4096, 4096, 4096, tree_root,
2078                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2079
2080         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2081         if (!bh) {
2082                 err = -EINVAL;
2083                 goto fail_alloc;
2084         }
2085
2086         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2087         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2088                sizeof(*fs_info->super_for_commit));
2089         brelse(bh);
2090
2091         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2092
2093         disk_super = fs_info->super_copy;
2094         if (!btrfs_super_root(disk_super))
2095                 goto fail_alloc;
2096
2097         /* check FS state, whether FS is broken. */
2098         fs_info->fs_state |= btrfs_super_flags(disk_super);
2099
2100         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2101
2102         /*
2103          * run through our array of backup supers and setup
2104          * our ring pointer to the oldest one
2105          */
2106         generation = btrfs_super_generation(disk_super);
2107         find_oldest_super_backup(fs_info, generation);
2108
2109         /*
2110          * In the long term, we'll store the compression type in the super
2111          * block, and it'll be used for per file compression control.
2112          */
2113         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2114
2115         ret = btrfs_parse_options(tree_root, options);
2116         if (ret) {
2117                 err = ret;
2118                 goto fail_alloc;
2119         }
2120
2121         features = btrfs_super_incompat_flags(disk_super) &
2122                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2123         if (features) {
2124                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2125                        "unsupported optional features (%Lx).\n",
2126                        (unsigned long long)features);
2127                 err = -EINVAL;
2128                 goto fail_alloc;
2129         }
2130
2131         features = btrfs_super_incompat_flags(disk_super);
2132         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2133         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2134                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2135         btrfs_set_super_incompat_flags(disk_super, features);
2136
2137         features = btrfs_super_compat_ro_flags(disk_super) &
2138                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2139         if (!(sb->s_flags & MS_RDONLY) && features) {
2140                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2141                        "unsupported option features (%Lx).\n",
2142                        (unsigned long long)features);
2143                 err = -EINVAL;
2144                 goto fail_alloc;
2145         }
2146
2147         btrfs_init_workers(&fs_info->generic_worker,
2148                            "genwork", 1, NULL);
2149
2150         btrfs_init_workers(&fs_info->workers, "worker",
2151                            fs_info->thread_pool_size,
2152                            &fs_info->generic_worker);
2153
2154         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2155                            fs_info->thread_pool_size,
2156                            &fs_info->generic_worker);
2157
2158         btrfs_init_workers(&fs_info->submit_workers, "submit",
2159                            min_t(u64, fs_devices->num_devices,
2160                            fs_info->thread_pool_size),
2161                            &fs_info->generic_worker);
2162
2163         btrfs_init_workers(&fs_info->caching_workers, "cache",
2164                            2, &fs_info->generic_worker);
2165
2166         /* a higher idle thresh on the submit workers makes it much more
2167          * likely that bios will be send down in a sane order to the
2168          * devices
2169          */
2170         fs_info->submit_workers.idle_thresh = 64;
2171
2172         fs_info->workers.idle_thresh = 16;
2173         fs_info->workers.ordered = 1;
2174
2175         fs_info->delalloc_workers.idle_thresh = 2;
2176         fs_info->delalloc_workers.ordered = 1;
2177
2178         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2179                            &fs_info->generic_worker);
2180         btrfs_init_workers(&fs_info->endio_workers, "endio",
2181                            fs_info->thread_pool_size,
2182                            &fs_info->generic_worker);
2183         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2184                            fs_info->thread_pool_size,
2185                            &fs_info->generic_worker);
2186         btrfs_init_workers(&fs_info->endio_meta_write_workers,
2187                            "endio-meta-write", fs_info->thread_pool_size,
2188                            &fs_info->generic_worker);
2189         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2190                            fs_info->thread_pool_size,
2191                            &fs_info->generic_worker);
2192         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2193                            1, &fs_info->generic_worker);
2194         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2195                            fs_info->thread_pool_size,
2196                            &fs_info->generic_worker);
2197         btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2198                            fs_info->thread_pool_size,
2199                            &fs_info->generic_worker);
2200
2201         /*
2202          * endios are largely parallel and should have a very
2203          * low idle thresh
2204          */
2205         fs_info->endio_workers.idle_thresh = 4;
2206         fs_info->endio_meta_workers.idle_thresh = 4;
2207
2208         fs_info->endio_write_workers.idle_thresh = 2;
2209         fs_info->endio_meta_write_workers.idle_thresh = 2;
2210         fs_info->readahead_workers.idle_thresh = 2;
2211
2212         /*
2213          * btrfs_start_workers can really only fail because of ENOMEM so just
2214          * return -ENOMEM if any of these fail.
2215          */
2216         ret = btrfs_start_workers(&fs_info->workers);
2217         ret |= btrfs_start_workers(&fs_info->generic_worker);
2218         ret |= btrfs_start_workers(&fs_info->submit_workers);
2219         ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2220         ret |= btrfs_start_workers(&fs_info->fixup_workers);
2221         ret |= btrfs_start_workers(&fs_info->endio_workers);
2222         ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2223         ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2224         ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2225         ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2226         ret |= btrfs_start_workers(&fs_info->delayed_workers);
2227         ret |= btrfs_start_workers(&fs_info->caching_workers);
2228         ret |= btrfs_start_workers(&fs_info->readahead_workers);
2229         if (ret) {
2230                 ret = -ENOMEM;
2231                 goto fail_sb_buffer;
2232         }
2233
2234         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2235         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2236                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2237
2238         nodesize = btrfs_super_nodesize(disk_super);
2239         leafsize = btrfs_super_leafsize(disk_super);
2240         sectorsize = btrfs_super_sectorsize(disk_super);
2241         stripesize = btrfs_super_stripesize(disk_super);
2242         tree_root->nodesize = nodesize;
2243         tree_root->leafsize = leafsize;
2244         tree_root->sectorsize = sectorsize;
2245         tree_root->stripesize = stripesize;
2246
2247         sb->s_blocksize = sectorsize;
2248         sb->s_blocksize_bits = blksize_bits(sectorsize);
2249
2250         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2251                     sizeof(disk_super->magic))) {
2252                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2253                 goto fail_sb_buffer;
2254         }
2255
2256         mutex_lock(&fs_info->chunk_mutex);
2257         ret = btrfs_read_sys_array(tree_root);
2258         mutex_unlock(&fs_info->chunk_mutex);
2259         if (ret) {
2260                 printk(KERN_WARNING "btrfs: failed to read the system "
2261                        "array on %s\n", sb->s_id);
2262                 goto fail_sb_buffer;
2263         }
2264
2265         blocksize = btrfs_level_size(tree_root,
2266                                      btrfs_super_chunk_root_level(disk_super));
2267         generation = btrfs_super_chunk_root_generation(disk_super);
2268
2269         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2270                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2271
2272         chunk_root->node = read_tree_block(chunk_root,
2273                                            btrfs_super_chunk_root(disk_super),
2274                                            blocksize, generation);
2275         BUG_ON(!chunk_root->node);
2276         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2277                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2278                        sb->s_id);
2279                 goto fail_tree_roots;
2280         }
2281         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2282         chunk_root->commit_root = btrfs_root_node(chunk_root);
2283
2284         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2285            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2286            BTRFS_UUID_SIZE);
2287
2288         ret = btrfs_read_chunk_tree(chunk_root);
2289         if (ret) {
2290                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2291                        sb->s_id);
2292                 goto fail_tree_roots;
2293         }
2294
2295         btrfs_close_extra_devices(fs_devices);
2296
2297 retry_root_backup:
2298         blocksize = btrfs_level_size(tree_root,
2299                                      btrfs_super_root_level(disk_super));
2300         generation = btrfs_super_generation(disk_super);
2301
2302         tree_root->node = read_tree_block(tree_root,
2303                                           btrfs_super_root(disk_super),
2304                                           blocksize, generation);
2305         if (!tree_root->node ||
2306             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2307                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2308                        sb->s_id);
2309
2310                 goto recovery_tree_root;
2311         }
2312
2313         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2314         tree_root->commit_root = btrfs_root_node(tree_root);
2315
2316         ret = find_and_setup_root(tree_root, fs_info,
2317                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2318         if (ret)
2319                 goto recovery_tree_root;
2320         extent_root->track_dirty = 1;
2321
2322         ret = find_and_setup_root(tree_root, fs_info,
2323                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2324         if (ret)
2325                 goto recovery_tree_root;
2326         dev_root->track_dirty = 1;
2327
2328         ret = find_and_setup_root(tree_root, fs_info,
2329                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2330         if (ret)
2331                 goto recovery_tree_root;
2332
2333         csum_root->track_dirty = 1;
2334
2335         fs_info->generation = generation;
2336         fs_info->last_trans_committed = generation;
2337
2338         ret = btrfs_init_space_info(fs_info);
2339         if (ret) {
2340                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2341                 goto fail_block_groups;
2342         }
2343
2344         ret = btrfs_read_block_groups(extent_root);
2345         if (ret) {
2346                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2347                 goto fail_block_groups;
2348         }
2349
2350         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2351                                                "btrfs-cleaner");
2352         if (IS_ERR(fs_info->cleaner_kthread))
2353                 goto fail_block_groups;
2354
2355         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2356                                                    tree_root,
2357                                                    "btrfs-transaction");
2358         if (IS_ERR(fs_info->transaction_kthread))
2359                 goto fail_cleaner;
2360
2361         if (!btrfs_test_opt(tree_root, SSD) &&
2362             !btrfs_test_opt(tree_root, NOSSD) &&
2363             !fs_info->fs_devices->rotating) {
2364                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2365                        "mode\n");
2366                 btrfs_set_opt(fs_info->mount_opt, SSD);
2367         }
2368
2369 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2370         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2371                 ret = btrfsic_mount(tree_root, fs_devices,
2372                                     btrfs_test_opt(tree_root,
2373                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2374                                     1 : 0,
2375                                     fs_info->check_integrity_print_mask);
2376                 if (ret)
2377                         printk(KERN_WARNING "btrfs: failed to initialize"
2378                                " integrity check module %s\n", sb->s_id);
2379         }
2380 #endif
2381
2382         /* do not make disk changes in broken FS */
2383         if (btrfs_super_log_root(disk_super) != 0 &&
2384             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2385                 u64 bytenr = btrfs_super_log_root(disk_super);
2386
2387                 if (fs_devices->rw_devices == 0) {
2388                         printk(KERN_WARNING "Btrfs log replay required "
2389                                "on RO media\n");
2390                         err = -EIO;
2391                         goto fail_trans_kthread;
2392                 }
2393                 blocksize =
2394                      btrfs_level_size(tree_root,
2395                                       btrfs_super_log_root_level(disk_super));
2396
2397                 log_tree_root = btrfs_alloc_root(fs_info);
2398                 if (!log_tree_root) {
2399                         err = -ENOMEM;
2400                         goto fail_trans_kthread;
2401                 }
2402
2403                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2404                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2405
2406                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2407                                                       blocksize,
2408                                                       generation + 1);
2409                 ret = btrfs_recover_log_trees(log_tree_root);
2410                 BUG_ON(ret);
2411
2412                 if (sb->s_flags & MS_RDONLY) {
2413                         ret =  btrfs_commit_super(tree_root);
2414                         BUG_ON(ret);
2415                 }
2416         }
2417
2418         ret = btrfs_find_orphan_roots(tree_root);
2419         BUG_ON(ret);
2420
2421         if (!(sb->s_flags & MS_RDONLY)) {
2422                 ret = btrfs_cleanup_fs_roots(fs_info);
2423                 BUG_ON(ret);
2424
2425                 ret = btrfs_recover_relocation(tree_root);
2426                 if (ret < 0) {
2427                         printk(KERN_WARNING
2428                                "btrfs: failed to recover relocation\n");
2429                         err = -EINVAL;
2430                         goto fail_trans_kthread;
2431                 }
2432         }
2433
2434         location.objectid = BTRFS_FS_TREE_OBJECTID;
2435         location.type = BTRFS_ROOT_ITEM_KEY;
2436         location.offset = (u64)-1;
2437
2438         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2439         if (!fs_info->fs_root)
2440                 goto fail_trans_kthread;
2441         if (IS_ERR(fs_info->fs_root)) {
2442                 err = PTR_ERR(fs_info->fs_root);
2443                 goto fail_trans_kthread;
2444         }
2445
2446         if (!(sb->s_flags & MS_RDONLY)) {
2447                 down_read(&fs_info->cleanup_work_sem);
2448                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2449                 if (!err)
2450                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2451                 up_read(&fs_info->cleanup_work_sem);
2452
2453                 if (!err)
2454                         err = btrfs_recover_balance(fs_info->tree_root);
2455
2456                 if (err) {
2457                         close_ctree(tree_root);
2458                         return err;
2459                 }
2460         }
2461
2462         return 0;
2463
2464 fail_trans_kthread:
2465         kthread_stop(fs_info->transaction_kthread);
2466 fail_cleaner:
2467         kthread_stop(fs_info->cleaner_kthread);
2468
2469         /*
2470          * make sure we're done with the btree inode before we stop our
2471          * kthreads
2472          */
2473         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2474         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2475
2476 fail_block_groups:
2477         btrfs_free_block_groups(fs_info);
2478
2479 fail_tree_roots:
2480         free_root_pointers(fs_info, 1);
2481
2482 fail_sb_buffer:
2483         btrfs_stop_workers(&fs_info->generic_worker);
2484         btrfs_stop_workers(&fs_info->readahead_workers);
2485         btrfs_stop_workers(&fs_info->fixup_workers);
2486         btrfs_stop_workers(&fs_info->delalloc_workers);
2487         btrfs_stop_workers(&fs_info->workers);
2488         btrfs_stop_workers(&fs_info->endio_workers);
2489         btrfs_stop_workers(&fs_info->endio_meta_workers);
2490         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2491         btrfs_stop_workers(&fs_info->endio_write_workers);
2492         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2493         btrfs_stop_workers(&fs_info->submit_workers);
2494         btrfs_stop_workers(&fs_info->delayed_workers);
2495         btrfs_stop_workers(&fs_info->caching_workers);
2496 fail_alloc:
2497 fail_iput:
2498         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2499
2500         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2501         iput(fs_info->btree_inode);
2502 fail_bdi:
2503         bdi_destroy(&fs_info->bdi);
2504 fail_srcu:
2505         cleanup_srcu_struct(&fs_info->subvol_srcu);
2506 fail:
2507         btrfs_close_devices(fs_info->fs_devices);
2508         return err;
2509
2510 recovery_tree_root:
2511         if (!btrfs_test_opt(tree_root, RECOVERY))
2512                 goto fail_tree_roots;
2513
2514         free_root_pointers(fs_info, 0);
2515
2516         /* don't use the log in recovery mode, it won't be valid */
2517         btrfs_set_super_log_root(disk_super, 0);
2518
2519         /* we can't trust the free space cache either */
2520         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2521
2522         ret = next_root_backup(fs_info, fs_info->super_copy,
2523                                &num_backups_tried, &backup_index);
2524         if (ret == -1)
2525                 goto fail_block_groups;
2526         goto retry_root_backup;
2527 }
2528
2529 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2530 {
2531         char b[BDEVNAME_SIZE];
2532
2533         if (uptodate) {
2534                 set_buffer_uptodate(bh);
2535         } else {
2536                 printk_ratelimited(KERN_WARNING "lost page write due to "
2537                                         "I/O error on %s\n",
2538                                        bdevname(bh->b_bdev, b));
2539                 /* note, we dont' set_buffer_write_io_error because we have
2540                  * our own ways of dealing with the IO errors
2541                  */
2542                 clear_buffer_uptodate(bh);
2543         }
2544         unlock_buffer(bh);
2545         put_bh(bh);
2546 }
2547
2548 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2549 {
2550         struct buffer_head *bh;
2551         struct buffer_head *latest = NULL;
2552         struct btrfs_super_block *super;
2553         int i;
2554         u64 transid = 0;
2555         u64 bytenr;
2556
2557         /* we would like to check all the supers, but that would make
2558          * a btrfs mount succeed after a mkfs from a different FS.
2559          * So, we need to add a special mount option to scan for
2560          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2561          */
2562         for (i = 0; i < 1; i++) {
2563                 bytenr = btrfs_sb_offset(i);
2564                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2565                         break;
2566                 bh = __bread(bdev, bytenr / 4096, 4096);
2567                 if (!bh)
2568                         continue;
2569
2570                 super = (struct btrfs_super_block *)bh->b_data;
2571                 if (btrfs_super_bytenr(super) != bytenr ||
2572                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2573                             sizeof(super->magic))) {
2574                         brelse(bh);
2575                         continue;
2576                 }
2577
2578                 if (!latest || btrfs_super_generation(super) > transid) {
2579                         brelse(latest);
2580                         latest = bh;
2581                         transid = btrfs_super_generation(super);
2582                 } else {
2583                         brelse(bh);
2584                 }
2585         }
2586         return latest;
2587 }
2588
2589 /*
2590  * this should be called twice, once with wait == 0 and
2591  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2592  * we write are pinned.
2593  *
2594  * They are released when wait == 1 is done.
2595  * max_mirrors must be the same for both runs, and it indicates how
2596  * many supers on this one device should be written.
2597  *
2598  * max_mirrors == 0 means to write them all.
2599  */
2600 static int write_dev_supers(struct btrfs_device *device,
2601                             struct btrfs_super_block *sb,
2602                             int do_barriers, int wait, int max_mirrors)
2603 {
2604         struct buffer_head *bh;
2605         int i;
2606         int ret;
2607         int errors = 0;
2608         u32 crc;
2609         u64 bytenr;
2610
2611         if (max_mirrors == 0)
2612                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2613
2614         for (i = 0; i < max_mirrors; i++) {
2615                 bytenr = btrfs_sb_offset(i);
2616                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2617                         break;
2618
2619                 if (wait) {
2620                         bh = __find_get_block(device->bdev, bytenr / 4096,
2621                                               BTRFS_SUPER_INFO_SIZE);
2622                         BUG_ON(!bh);
2623                         wait_on_buffer(bh);
2624                         if (!buffer_uptodate(bh))
2625                                 errors++;
2626
2627                         /* drop our reference */
2628                         brelse(bh);
2629
2630                         /* drop the reference from the wait == 0 run */
2631                         brelse(bh);
2632                         continue;
2633                 } else {
2634                         btrfs_set_super_bytenr(sb, bytenr);
2635
2636                         crc = ~(u32)0;
2637                         crc = btrfs_csum_data(NULL, (char *)sb +
2638                                               BTRFS_CSUM_SIZE, crc,
2639                                               BTRFS_SUPER_INFO_SIZE -
2640                                               BTRFS_CSUM_SIZE);
2641                         btrfs_csum_final(crc, sb->csum);
2642
2643                         /*
2644                          * one reference for us, and we leave it for the
2645                          * caller
2646                          */
2647                         bh = __getblk(device->bdev, bytenr / 4096,
2648                                       BTRFS_SUPER_INFO_SIZE);
2649                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2650
2651                         /* one reference for submit_bh */
2652                         get_bh(bh);
2653
2654                         set_buffer_uptodate(bh);
2655                         lock_buffer(bh);
2656                         bh->b_end_io = btrfs_end_buffer_write_sync;
2657                 }
2658
2659                 /*
2660                  * we fua the first super.  The others we allow
2661                  * to go down lazy.
2662                  */
2663                 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2664                 if (ret)
2665                         errors++;
2666         }
2667         return errors < i ? 0 : -1;
2668 }
2669
2670 /*
2671  * endio for the write_dev_flush, this will wake anyone waiting
2672  * for the barrier when it is done
2673  */
2674 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2675 {
2676         if (err) {
2677                 if (err == -EOPNOTSUPP)
2678                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2679                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2680         }
2681         if (bio->bi_private)
2682                 complete(bio->bi_private);
2683         bio_put(bio);
2684 }
2685
2686 /*
2687  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
2688  * sent down.  With wait == 1, it waits for the previous flush.
2689  *
2690  * any device where the flush fails with eopnotsupp are flagged as not-barrier
2691  * capable
2692  */
2693 static int write_dev_flush(struct btrfs_device *device, int wait)
2694 {
2695         struct bio *bio;
2696         int ret = 0;
2697
2698         if (device->nobarriers)
2699                 return 0;
2700
2701         if (wait) {
2702                 bio = device->flush_bio;
2703                 if (!bio)
2704                         return 0;
2705
2706                 wait_for_completion(&device->flush_wait);
2707
2708                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2709                         printk("btrfs: disabling barriers on dev %s\n",
2710                                device->name);
2711                         device->nobarriers = 1;
2712                 }
2713                 if (!bio_flagged(bio, BIO_UPTODATE)) {
2714                         ret = -EIO;
2715                 }
2716
2717                 /* drop the reference from the wait == 0 run */
2718                 bio_put(bio);
2719                 device->flush_bio = NULL;
2720
2721                 return ret;
2722         }
2723
2724         /*
2725          * one reference for us, and we leave it for the
2726          * caller
2727          */
2728         device->flush_bio = NULL;;
2729         bio = bio_alloc(GFP_NOFS, 0);
2730         if (!bio)
2731                 return -ENOMEM;
2732
2733         bio->bi_end_io = btrfs_end_empty_barrier;
2734         bio->bi_bdev = device->bdev;
2735         init_completion(&device->flush_wait);
2736         bio->bi_private = &device->flush_wait;
2737         device->flush_bio = bio;
2738
2739         bio_get(bio);
2740         btrfsic_submit_bio(WRITE_FLUSH, bio);
2741
2742         return 0;
2743 }
2744
2745 /*
2746  * send an empty flush down to each device in parallel,
2747  * then wait for them
2748  */
2749 static int barrier_all_devices(struct btrfs_fs_info *info)
2750 {
2751         struct list_head *head;
2752         struct btrfs_device *dev;
2753         int errors = 0;
2754         int ret;
2755
2756         /* send down all the barriers */
2757         head = &info->fs_devices->devices;
2758         list_for_each_entry_rcu(dev, head, dev_list) {
2759                 if (!dev->bdev) {
2760                         errors++;
2761                         continue;
2762                 }
2763                 if (!dev->in_fs_metadata || !dev->writeable)
2764                         continue;
2765
2766                 ret = write_dev_flush(dev, 0);
2767                 if (ret)
2768                         errors++;
2769         }
2770
2771         /* wait for all the barriers */
2772         list_for_each_entry_rcu(dev, head, dev_list) {
2773                 if (!dev->bdev) {
2774                         errors++;
2775                         continue;
2776                 }
2777                 if (!dev->in_fs_metadata || !dev->writeable)
2778                         continue;
2779
2780                 ret = write_dev_flush(dev, 1);
2781                 if (ret)
2782                         errors++;
2783         }
2784         if (errors)
2785                 return -EIO;
2786         return 0;
2787 }
2788
2789 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2790 {
2791         struct list_head *head;
2792         struct btrfs_device *dev;
2793         struct btrfs_super_block *sb;
2794         struct btrfs_dev_item *dev_item;
2795         int ret;
2796         int do_barriers;
2797         int max_errors;
2798         int total_errors = 0;
2799         u64 flags;
2800
2801         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2802         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2803         backup_super_roots(root->fs_info);
2804
2805         sb = root->fs_info->super_for_commit;
2806         dev_item = &sb->dev_item;
2807
2808         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2809         head = &root->fs_info->fs_devices->devices;
2810
2811         if (do_barriers)
2812                 barrier_all_devices(root->fs_info);
2813
2814         list_for_each_entry_rcu(dev, head, dev_list) {
2815                 if (!dev->bdev) {
2816                         total_errors++;
2817                         continue;
2818                 }
2819                 if (!dev->in_fs_metadata || !dev->writeable)
2820                         continue;
2821
2822                 btrfs_set_stack_device_generation(dev_item, 0);
2823                 btrfs_set_stack_device_type(dev_item, dev->type);
2824                 btrfs_set_stack_device_id(dev_item, dev->devid);
2825                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2826                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2827                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2828                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2829                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2830                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2831                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2832
2833                 flags = btrfs_super_flags(sb);
2834                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2835
2836                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2837                 if (ret)
2838                         total_errors++;
2839         }
2840         if (total_errors > max_errors) {
2841                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2842                        total_errors);
2843                 BUG();
2844         }
2845
2846         total_errors = 0;
2847         list_for_each_entry_rcu(dev, head, dev_list) {
2848                 if (!dev->bdev)
2849                         continue;
2850                 if (!dev->in_fs_metadata || !dev->writeable)
2851                         continue;
2852
2853                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2854                 if (ret)
2855                         total_errors++;
2856         }
2857         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2858         if (total_errors > max_errors) {
2859                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2860                        total_errors);
2861                 BUG();
2862         }
2863         return 0;
2864 }
2865
2866 int write_ctree_super(struct btrfs_trans_handle *trans,
2867                       struct btrfs_root *root, int max_mirrors)
2868 {
2869         int ret;
2870
2871         ret = write_all_supers(root, max_mirrors);
2872         return ret;
2873 }
2874
2875 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2876 {
2877         spin_lock(&fs_info->fs_roots_radix_lock);
2878         radix_tree_delete(&fs_info->fs_roots_radix,
2879                           (unsigned long)root->root_key.objectid);
2880         spin_unlock(&fs_info->fs_roots_radix_lock);
2881
2882         if (btrfs_root_refs(&root->root_item) == 0)
2883                 synchronize_srcu(&fs_info->subvol_srcu);
2884
2885         __btrfs_remove_free_space_cache(root->free_ino_pinned);
2886         __btrfs_remove_free_space_cache(root->free_ino_ctl);
2887         free_fs_root(root);
2888         return 0;
2889 }
2890
2891 static void free_fs_root(struct btrfs_root *root)
2892 {
2893         iput(root->cache_inode);
2894         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2895         if (root->anon_dev)
2896                 free_anon_bdev(root->anon_dev);
2897         free_extent_buffer(root->node);
2898         free_extent_buffer(root->commit_root);
2899         kfree(root->free_ino_ctl);
2900         kfree(root->free_ino_pinned);
2901         kfree(root->name);
2902         kfree(root);
2903 }
2904
2905 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2906 {
2907         int ret;
2908         struct btrfs_root *gang[8];
2909         int i;
2910
2911         while (!list_empty(&fs_info->dead_roots)) {
2912                 gang[0] = list_entry(fs_info->dead_roots.next,
2913                                      struct btrfs_root, root_list);
2914                 list_del(&gang[0]->root_list);
2915
2916                 if (gang[0]->in_radix) {
2917                         btrfs_free_fs_root(fs_info, gang[0]);
2918                 } else {
2919                         free_extent_buffer(gang[0]->node);
2920                         free_extent_buffer(gang[0]->commit_root);
2921                         kfree(gang[0]);
2922                 }
2923         }
2924
2925         while (1) {
2926                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2927                                              (void **)gang, 0,
2928                                              ARRAY_SIZE(gang));
2929                 if (!ret)
2930                         break;
2931                 for (i = 0; i < ret; i++)
2932                         btrfs_free_fs_root(fs_info, gang[i]);
2933         }
2934         return 0;
2935 }
2936
2937 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2938 {
2939         u64 root_objectid = 0;
2940         struct btrfs_root *gang[8];
2941         int i;
2942         int ret;
2943
2944         while (1) {
2945                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2946                                              (void **)gang, root_objectid,
2947                                              ARRAY_SIZE(gang));
2948                 if (!ret)
2949                         break;
2950
2951                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2952                 for (i = 0; i < ret; i++) {
2953                         int err;
2954
2955                         root_objectid = gang[i]->root_key.objectid;
2956                         err = btrfs_orphan_cleanup(gang[i]);
2957                         if (err)
2958                                 return err;
2959                 }
2960                 root_objectid++;
2961         }
2962         return 0;
2963 }
2964
2965 int btrfs_commit_super(struct btrfs_root *root)
2966 {
2967         struct btrfs_trans_handle *trans;
2968         int ret;
2969
2970         mutex_lock(&root->fs_info->cleaner_mutex);
2971         btrfs_run_delayed_iputs(root);
2972         btrfs_clean_old_snapshots(root);
2973         mutex_unlock(&root->fs_info->cleaner_mutex);
2974
2975         /* wait until ongoing cleanup work done */
2976         down_write(&root->fs_info->cleanup_work_sem);
2977         up_write(&root->fs_info->cleanup_work_sem);
2978
2979         trans = btrfs_join_transaction(root);
2980         if (IS_ERR(trans))
2981                 return PTR_ERR(trans);
2982         ret = btrfs_commit_transaction(trans, root);
2983         BUG_ON(ret);
2984         /* run commit again to drop the original snapshot */
2985         trans = btrfs_join_transaction(root);
2986         if (IS_ERR(trans))
2987                 return PTR_ERR(trans);
2988         btrfs_commit_transaction(trans, root);
2989         ret = btrfs_write_and_wait_transaction(NULL, root);
2990         BUG_ON(ret);
2991
2992         ret = write_ctree_super(NULL, root, 0);
2993         return ret;
2994 }
2995
2996 int close_ctree(struct btrfs_root *root)
2997 {
2998         struct btrfs_fs_info *fs_info = root->fs_info;
2999         int ret;
3000
3001         fs_info->closing = 1;
3002         smp_mb();
3003
3004         /* pause restriper - we want to resume on mount */
3005         btrfs_pause_balance(root->fs_info);
3006
3007         btrfs_scrub_cancel(root);
3008
3009         /* wait for any defraggers to finish */
3010         wait_event(fs_info->transaction_wait,
3011                    (atomic_read(&fs_info->defrag_running) == 0));
3012
3013         /* clear out the rbtree of defraggable inodes */
3014         btrfs_run_defrag_inodes(fs_info);
3015
3016         /*
3017          * Here come 2 situations when btrfs is broken to flip readonly:
3018          *
3019          * 1. when btrfs flips readonly somewhere else before
3020          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3021          * and btrfs will skip to write sb directly to keep
3022          * ERROR state on disk.
3023          *
3024          * 2. when btrfs flips readonly just in btrfs_commit_super,
3025          * and in such case, btrfs cannot write sb via btrfs_commit_super,
3026          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3027          * btrfs will cleanup all FS resources first and write sb then.
3028          */
3029         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3030                 ret = btrfs_commit_super(root);
3031                 if (ret)
3032                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3033         }
3034
3035         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3036                 ret = btrfs_error_commit_super(root);
3037                 if (ret)
3038                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3039         }
3040
3041         btrfs_put_block_group_cache(fs_info);
3042
3043         kthread_stop(fs_info->transaction_kthread);
3044         kthread_stop(fs_info->cleaner_kthread);
3045
3046         fs_info->closing = 2;
3047         smp_mb();
3048
3049         if (fs_info->delalloc_bytes) {
3050                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3051                        (unsigned long long)fs_info->delalloc_bytes);
3052         }
3053         if (fs_info->total_ref_cache_size) {
3054                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3055                        (unsigned long long)fs_info->total_ref_cache_size);
3056         }
3057
3058         free_extent_buffer(fs_info->extent_root->node);
3059         free_extent_buffer(fs_info->extent_root->commit_root);
3060         free_extent_buffer(fs_info->tree_root->node);
3061         free_extent_buffer(fs_info->tree_root->commit_root);
3062         free_extent_buffer(fs_info->chunk_root->node);
3063         free_extent_buffer(fs_info->chunk_root->commit_root);
3064         free_extent_buffer(fs_info->dev_root->node);
3065         free_extent_buffer(fs_info->dev_root->commit_root);
3066         free_extent_buffer(fs_info->csum_root->node);
3067         free_extent_buffer(fs_info->csum_root->commit_root);
3068
3069         btrfs_free_block_groups(fs_info);
3070
3071         del_fs_roots(fs_info);
3072
3073         iput(fs_info->btree_inode);
3074
3075         btrfs_stop_workers(&fs_info->generic_worker);
3076         btrfs_stop_workers(&fs_info->fixup_workers);
3077         btrfs_stop_workers(&fs_info->delalloc_workers);
3078         btrfs_stop_workers(&fs_info->workers);
3079         btrfs_stop_workers(&fs_info->endio_workers);
3080         btrfs_stop_workers(&fs_info->endio_meta_workers);
3081         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3082         btrfs_stop_workers(&fs_info->endio_write_workers);
3083         btrfs_stop_workers(&fs_info->endio_freespace_worker);
3084         btrfs_stop_workers(&fs_info->submit_workers);
3085         btrfs_stop_workers(&fs_info->delayed_workers);
3086         btrfs_stop_workers(&fs_info->caching_workers);
3087         btrfs_stop_workers(&fs_info->readahead_workers);
3088
3089 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3090         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3091                 btrfsic_unmount(root, fs_info->fs_devices);
3092 #endif
3093
3094         btrfs_close_devices(fs_info->fs_devices);
3095         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3096
3097         bdi_destroy(&fs_info->bdi);
3098         cleanup_srcu_struct(&fs_info->subvol_srcu);
3099
3100         return 0;
3101 }
3102
3103 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
3104 {
3105         int ret;
3106         struct inode *btree_inode = buf->first_page->mapping->host;
3107
3108         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
3109                                      NULL);
3110         if (!ret)
3111                 return ret;
3112
3113         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3114                                     parent_transid);
3115         return !ret;
3116 }
3117
3118 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3119 {
3120         struct inode *btree_inode = buf->first_page->mapping->host;
3121         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
3122                                           buf);
3123 }
3124
3125 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3126 {
3127         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3128         u64 transid = btrfs_header_generation(buf);
3129         struct inode *btree_inode = root->fs_info->btree_inode;
3130         int was_dirty;
3131
3132         btrfs_assert_tree_locked(buf);
3133         if (transid != root->fs_info->generation) {
3134                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3135                        "found %llu running %llu\n",
3136                         (unsigned long long)buf->start,
3137                         (unsigned long long)transid,
3138                         (unsigned long long)root->fs_info->generation);
3139                 WARN_ON(1);
3140         }
3141         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
3142                                             buf);
3143         if (!was_dirty) {
3144                 spin_lock(&root->fs_info->delalloc_lock);
3145                 root->fs_info->dirty_metadata_bytes += buf->len;
3146                 spin_unlock(&root->fs_info->delalloc_lock);
3147         }
3148 }
3149
3150 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3151 {
3152         /*
3153          * looks as though older kernels can get into trouble with
3154          * this code, they end up stuck in balance_dirty_pages forever
3155          */
3156         u64 num_dirty;
3157         unsigned long thresh = 32 * 1024 * 1024;
3158
3159         if (current->flags & PF_MEMALLOC)
3160                 return;
3161
3162         btrfs_balance_delayed_items(root);
3163
3164         num_dirty = root->fs_info->dirty_metadata_bytes;
3165
3166         if (num_dirty > thresh) {
3167                 balance_dirty_pages_ratelimited_nr(
3168                                    root->fs_info->btree_inode->i_mapping, 1);
3169         }
3170         return;
3171 }
3172
3173 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3174 {
3175         /*
3176          * looks as though older kernels can get into trouble with
3177          * this code, they end up stuck in balance_dirty_pages forever
3178          */
3179         u64 num_dirty;
3180         unsigned long thresh = 32 * 1024 * 1024;
3181
3182         if (current->flags & PF_MEMALLOC)
3183                 return;
3184
3185         num_dirty = root->fs_info->dirty_metadata_bytes;
3186
3187         if (num_dirty > thresh) {
3188                 balance_dirty_pages_ratelimited_nr(
3189                                    root->fs_info->btree_inode->i_mapping, 1);
3190         }
3191         return;
3192 }
3193
3194 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3195 {
3196         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3197         int ret;
3198         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3199         if (ret == 0)
3200                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
3201         return ret;
3202 }
3203
3204 static int btree_lock_page_hook(struct page *page, void *data,
3205                                 void (*flush_fn)(void *))
3206 {
3207         struct inode *inode = page->mapping->host;
3208         struct btrfs_root *root = BTRFS_I(inode)->root;
3209         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3210         struct extent_buffer *eb;
3211         unsigned long len;
3212         u64 bytenr = page_offset(page);
3213
3214         if (page->private == EXTENT_PAGE_PRIVATE)
3215                 goto out;
3216
3217         len = page->private >> 2;
3218         eb = find_extent_buffer(io_tree, bytenr, len);
3219         if (!eb)
3220                 goto out;
3221
3222         if (!btrfs_try_tree_write_lock(eb)) {
3223                 flush_fn(data);
3224                 btrfs_tree_lock(eb);
3225         }
3226         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3227
3228         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3229                 spin_lock(&root->fs_info->delalloc_lock);
3230                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3231                         root->fs_info->dirty_metadata_bytes -= eb->len;
3232                 else
3233                         WARN_ON(1);
3234                 spin_unlock(&root->fs_info->delalloc_lock);
3235         }
3236
3237         btrfs_tree_unlock(eb);
3238         free_extent_buffer(eb);
3239 out:
3240         if (!trylock_page(page)) {
3241                 flush_fn(data);
3242                 lock_page(page);
3243         }
3244         return 0;
3245 }
3246
3247 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3248                               int read_only)
3249 {
3250         if (read_only)
3251                 return;
3252
3253         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3254                 printk(KERN_WARNING "warning: mount fs with errors, "
3255                        "running btrfsck is recommended\n");
3256 }
3257
3258 int btrfs_error_commit_super(struct btrfs_root *root)
3259 {
3260         int ret;
3261
3262         mutex_lock(&root->fs_info->cleaner_mutex);
3263         btrfs_run_delayed_iputs(root);
3264         mutex_unlock(&root->fs_info->cleaner_mutex);
3265
3266         down_write(&root->fs_info->cleanup_work_sem);
3267         up_write(&root->fs_info->cleanup_work_sem);
3268
3269         /* cleanup FS via transaction */
3270         btrfs_cleanup_transaction(root);
3271
3272         ret = write_ctree_super(NULL, root, 0);
3273
3274         return ret;
3275 }
3276
3277 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3278 {
3279         struct btrfs_inode *btrfs_inode;
3280         struct list_head splice;
3281
3282         INIT_LIST_HEAD(&splice);
3283
3284         mutex_lock(&root->fs_info->ordered_operations_mutex);
3285         spin_lock(&root->fs_info->ordered_extent_lock);
3286
3287         list_splice_init(&root->fs_info->ordered_operations, &splice);
3288         while (!list_empty(&splice)) {
3289                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3290                                          ordered_operations);
3291
3292                 list_del_init(&btrfs_inode->ordered_operations);
3293
3294                 btrfs_invalidate_inodes(btrfs_inode->root);
3295         }
3296
3297         spin_unlock(&root->fs_info->ordered_extent_lock);
3298         mutex_unlock(&root->fs_info->ordered_operations_mutex);
3299
3300         return 0;
3301 }
3302
3303 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3304 {
3305         struct list_head splice;
3306         struct btrfs_ordered_extent *ordered;
3307         struct inode *inode;
3308
3309         INIT_LIST_HEAD(&splice);
3310
3311         spin_lock(&root->fs_info->ordered_extent_lock);
3312
3313         list_splice_init(&root->fs_info->ordered_extents, &splice);
3314         while (!list_empty(&splice)) {
3315                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3316                                      root_extent_list);
3317
3318                 list_del_init(&ordered->root_extent_list);
3319                 atomic_inc(&ordered->refs);
3320
3321                 /* the inode may be getting freed (in sys_unlink path). */
3322                 inode = igrab(ordered->inode);
3323
3324                 spin_unlock(&root->fs_info->ordered_extent_lock);
3325                 if (inode)
3326                         iput(inode);
3327
3328                 atomic_set(&ordered->refs, 1);
3329                 btrfs_put_ordered_extent(ordered);
3330
3331                 spin_lock(&root->fs_info->ordered_extent_lock);
3332         }
3333
3334         spin_unlock(&root->fs_info->ordered_extent_lock);
3335
3336         return 0;
3337 }
3338
3339 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3340                                       struct btrfs_root *root)
3341 {
3342         struct rb_node *node;
3343         struct btrfs_delayed_ref_root *delayed_refs;
3344         struct btrfs_delayed_ref_node *ref;
3345         int ret = 0;
3346
3347         delayed_refs = &trans->delayed_refs;
3348
3349         spin_lock(&delayed_refs->lock);
3350         if (delayed_refs->num_entries == 0) {
3351                 spin_unlock(&delayed_refs->lock);
3352                 printk(KERN_INFO "delayed_refs has NO entry\n");
3353                 return ret;
3354         }
3355
3356         node = rb_first(&delayed_refs->root);
3357         while (node) {
3358                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3359                 node = rb_next(node);
3360
3361                 ref->in_tree = 0;
3362                 rb_erase(&ref->rb_node, &delayed_refs->root);
3363                 delayed_refs->num_entries--;
3364
3365                 atomic_set(&ref->refs, 1);
3366                 if (btrfs_delayed_ref_is_head(ref)) {
3367                         struct btrfs_delayed_ref_head *head;
3368
3369                         head = btrfs_delayed_node_to_head(ref);
3370                         mutex_lock(&head->mutex);
3371                         kfree(head->extent_op);
3372                         delayed_refs->num_heads--;
3373                         if (list_empty(&head->cluster))
3374                                 delayed_refs->num_heads_ready--;
3375                         list_del_init(&head->cluster);
3376                         mutex_unlock(&head->mutex);
3377  &nb