mm: compaction: introduce sync-light migration for use by compaction
[linux-2.6.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
47 static struct extent_io_ops btree_extent_io_ops;
48 static void end_workqueue_fn(struct btrfs_work *work);
49 static void free_fs_root(struct btrfs_root *root);
50 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
51                                     int read_only);
52 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
53 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
55                                       struct btrfs_root *root);
56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
59                                         struct extent_io_tree *dirty_pages,
60                                         int mark);
61 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
62                                        struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_root *root);
64
65 /*
66  * end_io_wq structs are used to do processing in task context when an IO is
67  * complete.  This is used during reads to verify checksums, and it is used
68  * by writes to insert metadata for new file extents after IO is complete.
69  */
70 struct end_io_wq {
71         struct bio *bio;
72         bio_end_io_t *end_io;
73         void *private;
74         struct btrfs_fs_info *info;
75         int error;
76         int metadata;
77         struct list_head list;
78         struct btrfs_work work;
79 };
80
81 /*
82  * async submit bios are used to offload expensive checksumming
83  * onto the worker threads.  They checksum file and metadata bios
84  * just before they are sent down the IO stack.
85  */
86 struct async_submit_bio {
87         struct inode *inode;
88         struct bio *bio;
89         struct list_head list;
90         extent_submit_bio_hook_t *submit_bio_start;
91         extent_submit_bio_hook_t *submit_bio_done;
92         int rw;
93         int mirror_num;
94         unsigned long bio_flags;
95         /*
96          * bio_offset is optional, can be used if the pages in the bio
97          * can't tell us where in the file the bio should go
98          */
99         u64 bio_offset;
100         struct btrfs_work work;
101 };
102
103 /*
104  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
105  * eb, the lockdep key is determined by the btrfs_root it belongs to and
106  * the level the eb occupies in the tree.
107  *
108  * Different roots are used for different purposes and may nest inside each
109  * other and they require separate keysets.  As lockdep keys should be
110  * static, assign keysets according to the purpose of the root as indicated
111  * by btrfs_root->objectid.  This ensures that all special purpose roots
112  * have separate keysets.
113  *
114  * Lock-nesting across peer nodes is always done with the immediate parent
115  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
116  * subclass to avoid triggering lockdep warning in such cases.
117  *
118  * The key is set by the readpage_end_io_hook after the buffer has passed
119  * csum validation but before the pages are unlocked.  It is also set by
120  * btrfs_init_new_buffer on freshly allocated blocks.
121  *
122  * We also add a check to make sure the highest level of the tree is the
123  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
124  * needs update as well.
125  */
126 #ifdef CONFIG_DEBUG_LOCK_ALLOC
127 # if BTRFS_MAX_LEVEL != 8
128 #  error
129 # endif
130
131 static struct btrfs_lockdep_keyset {
132         u64                     id;             /* root objectid */
133         const char              *name_stem;     /* lock name stem */
134         char                    names[BTRFS_MAX_LEVEL + 1][20];
135         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
136 } btrfs_lockdep_keysets[] = {
137         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
138         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
139         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
140         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
141         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
142         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
143         { .id = BTRFS_ORPHAN_OBJECTID,          .name_stem = "orphan"   },
144         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
145         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
146         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
147         { .id = 0,                              .name_stem = "tree"     },
148 };
149
150 void __init btrfs_init_lockdep(void)
151 {
152         int i, j;
153
154         /* initialize lockdep class names */
155         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
156                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
157
158                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
159                         snprintf(ks->names[j], sizeof(ks->names[j]),
160                                  "btrfs-%s-%02d", ks->name_stem, j);
161         }
162 }
163
164 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
165                                     int level)
166 {
167         struct btrfs_lockdep_keyset *ks;
168
169         BUG_ON(level >= ARRAY_SIZE(ks->keys));
170
171         /* find the matching keyset, id 0 is the default entry */
172         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
173                 if (ks->id == objectid)
174                         break;
175
176         lockdep_set_class_and_name(&eb->lock,
177                                    &ks->keys[level], ks->names[level]);
178 }
179
180 #endif
181
182 /*
183  * extents on the btree inode are pretty simple, there's one extent
184  * that covers the entire device
185  */
186 static struct extent_map *btree_get_extent(struct inode *inode,
187                 struct page *page, size_t pg_offset, u64 start, u64 len,
188                 int create)
189 {
190         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
191         struct extent_map *em;
192         int ret;
193
194         read_lock(&em_tree->lock);
195         em = lookup_extent_mapping(em_tree, start, len);
196         if (em) {
197                 em->bdev =
198                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
199                 read_unlock(&em_tree->lock);
200                 goto out;
201         }
202         read_unlock(&em_tree->lock);
203
204         em = alloc_extent_map();
205         if (!em) {
206                 em = ERR_PTR(-ENOMEM);
207                 goto out;
208         }
209         em->start = 0;
210         em->len = (u64)-1;
211         em->block_len = (u64)-1;
212         em->block_start = 0;
213         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
214
215         write_lock(&em_tree->lock);
216         ret = add_extent_mapping(em_tree, em);
217         if (ret == -EEXIST) {
218                 u64 failed_start = em->start;
219                 u64 failed_len = em->len;
220
221                 free_extent_map(em);
222                 em = lookup_extent_mapping(em_tree, start, len);
223                 if (em) {
224                         ret = 0;
225                 } else {
226                         em = lookup_extent_mapping(em_tree, failed_start,
227                                                    failed_len);
228                         ret = -EIO;
229                 }
230         } else if (ret) {
231                 free_extent_map(em);
232                 em = NULL;
233         }
234         write_unlock(&em_tree->lock);
235
236         if (ret)
237                 em = ERR_PTR(ret);
238 out:
239         return em;
240 }
241
242 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
243 {
244         return crc32c(seed, data, len);
245 }
246
247 void btrfs_csum_final(u32 crc, char *result)
248 {
249         put_unaligned_le32(~crc, result);
250 }
251
252 /*
253  * compute the csum for a btree block, and either verify it or write it
254  * into the csum field of the block.
255  */
256 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
257                            int verify)
258 {
259         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
260         char *result = NULL;
261         unsigned long len;
262         unsigned long cur_len;
263         unsigned long offset = BTRFS_CSUM_SIZE;
264         char *kaddr;
265         unsigned long map_start;
266         unsigned long map_len;
267         int err;
268         u32 crc = ~(u32)0;
269         unsigned long inline_result;
270
271         len = buf->len - offset;
272         while (len > 0) {
273                 err = map_private_extent_buffer(buf, offset, 32,
274                                         &kaddr, &map_start, &map_len);
275                 if (err)
276                         return 1;
277                 cur_len = min(len, map_len - (offset - map_start));
278                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
279                                       crc, cur_len);
280                 len -= cur_len;
281                 offset += cur_len;
282         }
283         if (csum_size > sizeof(inline_result)) {
284                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
285                 if (!result)
286                         return 1;
287         } else {
288                 result = (char *)&inline_result;
289         }
290
291         btrfs_csum_final(crc, result);
292
293         if (verify) {
294                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
295                         u32 val;
296                         u32 found = 0;
297                         memcpy(&found, result, csum_size);
298
299                         read_extent_buffer(buf, &val, 0, csum_size);
300                         printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
301                                        "failed on %llu wanted %X found %X "
302                                        "level %d\n",
303                                        root->fs_info->sb->s_id,
304                                        (unsigned long long)buf->start, val, found,
305                                        btrfs_header_level(buf));
306                         if (result != (char *)&inline_result)
307                                 kfree(result);
308                         return 1;
309                 }
310         } else {
311                 write_extent_buffer(buf, result, 0, csum_size);
312         }
313         if (result != (char *)&inline_result)
314                 kfree(result);
315         return 0;
316 }
317
318 /*
319  * we can't consider a given block up to date unless the transid of the
320  * block matches the transid in the parent node's pointer.  This is how we
321  * detect blocks that either didn't get written at all or got written
322  * in the wrong place.
323  */
324 static int verify_parent_transid(struct extent_io_tree *io_tree,
325                                  struct extent_buffer *eb, u64 parent_transid)
326 {
327         struct extent_state *cached_state = NULL;
328         int ret;
329
330         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
331                 return 0;
332
333         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
334                          0, &cached_state, GFP_NOFS);
335         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
336             btrfs_header_generation(eb) == parent_transid) {
337                 ret = 0;
338                 goto out;
339         }
340         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
341                        "found %llu\n",
342                        (unsigned long long)eb->start,
343                        (unsigned long long)parent_transid,
344                        (unsigned long long)btrfs_header_generation(eb));
345         ret = 1;
346         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
347 out:
348         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
349                              &cached_state, GFP_NOFS);
350         return ret;
351 }
352
353 /*
354  * helper to read a given tree block, doing retries as required when
355  * the checksums don't match and we have alternate mirrors to try.
356  */
357 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
358                                           struct extent_buffer *eb,
359                                           u64 start, u64 parent_transid)
360 {
361         struct extent_io_tree *io_tree;
362         int ret;
363         int num_copies = 0;
364         int mirror_num = 0;
365
366         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
367         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
368         while (1) {
369                 ret = read_extent_buffer_pages(io_tree, eb, start,
370                                                WAIT_COMPLETE,
371                                                btree_get_extent, mirror_num);
372                 if (!ret &&
373                     !verify_parent_transid(io_tree, eb, parent_transid))
374                         return ret;
375
376                 /*
377                  * This buffer's crc is fine, but its contents are corrupted, so
378                  * there is no reason to read the other copies, they won't be
379                  * any less wrong.
380                  */
381                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
382                         return ret;
383
384                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
385                                               eb->start, eb->len);
386                 if (num_copies == 1)
387                         return ret;
388
389                 mirror_num++;
390                 if (mirror_num > num_copies)
391                         return ret;
392         }
393         return -EIO;
394 }
395
396 /*
397  * checksum a dirty tree block before IO.  This has extra checks to make sure
398  * we only fill in the checksum field in the first page of a multi-page block
399  */
400
401 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
402 {
403         struct extent_io_tree *tree;
404         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
405         u64 found_start;
406         unsigned long len;
407         struct extent_buffer *eb;
408         int ret;
409
410         tree = &BTRFS_I(page->mapping->host)->io_tree;
411
412         if (page->private == EXTENT_PAGE_PRIVATE) {
413                 WARN_ON(1);
414                 goto out;
415         }
416         if (!page->private) {
417                 WARN_ON(1);
418                 goto out;
419         }
420         len = page->private >> 2;
421         WARN_ON(len == 0);
422
423         eb = alloc_extent_buffer(tree, start, len, page);
424         if (eb == NULL) {
425                 WARN_ON(1);
426                 goto out;
427         }
428         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
429                                              btrfs_header_generation(eb));
430         BUG_ON(ret);
431         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
432
433         found_start = btrfs_header_bytenr(eb);
434         if (found_start != start) {
435                 WARN_ON(1);
436                 goto err;
437         }
438         if (eb->first_page != page) {
439                 WARN_ON(1);
440                 goto err;
441         }
442         if (!PageUptodate(page)) {
443                 WARN_ON(1);
444                 goto err;
445         }
446         csum_tree_block(root, eb, 0);
447 err:
448         free_extent_buffer(eb);
449 out:
450         return 0;
451 }
452
453 static int check_tree_block_fsid(struct btrfs_root *root,
454                                  struct extent_buffer *eb)
455 {
456         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
457         u8 fsid[BTRFS_UUID_SIZE];
458         int ret = 1;
459
460         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
461                            BTRFS_FSID_SIZE);
462         while (fs_devices) {
463                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
464                         ret = 0;
465                         break;
466                 }
467                 fs_devices = fs_devices->seed;
468         }
469         return ret;
470 }
471
472 #define CORRUPT(reason, eb, root, slot)                         \
473         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
474                "root=%llu, slot=%d\n", reason,                  \
475                (unsigned long long)btrfs_header_bytenr(eb),     \
476                (unsigned long long)root->objectid, slot)
477
478 static noinline int check_leaf(struct btrfs_root *root,
479                                struct extent_buffer *leaf)
480 {
481         struct btrfs_key key;
482         struct btrfs_key leaf_key;
483         u32 nritems = btrfs_header_nritems(leaf);
484         int slot;
485
486         if (nritems == 0)
487                 return 0;
488
489         /* Check the 0 item */
490         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
491             BTRFS_LEAF_DATA_SIZE(root)) {
492                 CORRUPT("invalid item offset size pair", leaf, root, 0);
493                 return -EIO;
494         }
495
496         /*
497          * Check to make sure each items keys are in the correct order and their
498          * offsets make sense.  We only have to loop through nritems-1 because
499          * we check the current slot against the next slot, which verifies the
500          * next slot's offset+size makes sense and that the current's slot
501          * offset is correct.
502          */
503         for (slot = 0; slot < nritems - 1; slot++) {
504                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
505                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
506
507                 /* Make sure the keys are in the right order */
508                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
509                         CORRUPT("bad key order", leaf, root, slot);
510                         return -EIO;
511                 }
512
513                 /*
514                  * Make sure the offset and ends are right, remember that the
515                  * item data starts at the end of the leaf and grows towards the
516                  * front.
517                  */
518                 if (btrfs_item_offset_nr(leaf, slot) !=
519                         btrfs_item_end_nr(leaf, slot + 1)) {
520                         CORRUPT("slot offset bad", leaf, root, slot);
521                         return -EIO;
522                 }
523
524                 /*
525                  * Check to make sure that we don't point outside of the leaf,
526                  * just incase all the items are consistent to eachother, but
527                  * all point outside of the leaf.
528                  */
529                 if (btrfs_item_end_nr(leaf, slot) >
530                     BTRFS_LEAF_DATA_SIZE(root)) {
531                         CORRUPT("slot end outside of leaf", leaf, root, slot);
532                         return -EIO;
533                 }
534         }
535
536         return 0;
537 }
538
539 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
540                                struct extent_state *state)
541 {
542         struct extent_io_tree *tree;
543         u64 found_start;
544         int found_level;
545         unsigned long len;
546         struct extent_buffer *eb;
547         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
548         int ret = 0;
549
550         tree = &BTRFS_I(page->mapping->host)->io_tree;
551         if (page->private == EXTENT_PAGE_PRIVATE)
552                 goto out;
553         if (!page->private)
554                 goto out;
555
556         len = page->private >> 2;
557         WARN_ON(len == 0);
558
559         eb = alloc_extent_buffer(tree, start, len, page);
560         if (eb == NULL) {
561                 ret = -EIO;
562                 goto out;
563         }
564
565         found_start = btrfs_header_bytenr(eb);
566         if (found_start != start) {
567                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
568                                "%llu %llu\n",
569                                (unsigned long long)found_start,
570                                (unsigned long long)eb->start);
571                 ret = -EIO;
572                 goto err;
573         }
574         if (eb->first_page != page) {
575                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
576                        eb->first_page->index, page->index);
577                 WARN_ON(1);
578                 ret = -EIO;
579                 goto err;
580         }
581         if (check_tree_block_fsid(root, eb)) {
582                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
583                                (unsigned long long)eb->start);
584                 ret = -EIO;
585                 goto err;
586         }
587         found_level = btrfs_header_level(eb);
588
589         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
590                                        eb, found_level);
591
592         ret = csum_tree_block(root, eb, 1);
593         if (ret) {
594                 ret = -EIO;
595                 goto err;
596         }
597
598         /*
599          * If this is a leaf block and it is corrupt, set the corrupt bit so
600          * that we don't try and read the other copies of this block, just
601          * return -EIO.
602          */
603         if (found_level == 0 && check_leaf(root, eb)) {
604                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
605                 ret = -EIO;
606         }
607
608         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
609         end = eb->start + end - 1;
610 err:
611         if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
612                 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
613                 btree_readahead_hook(root, eb, eb->start, ret);
614         }
615
616         free_extent_buffer(eb);
617 out:
618         return ret;
619 }
620
621 static int btree_io_failed_hook(struct bio *failed_bio,
622                          struct page *page, u64 start, u64 end,
623                          int mirror_num, struct extent_state *state)
624 {
625         struct extent_io_tree *tree;
626         unsigned long len;
627         struct extent_buffer *eb;
628         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
629
630         tree = &BTRFS_I(page->mapping->host)->io_tree;
631         if (page->private == EXTENT_PAGE_PRIVATE)
632                 goto out;
633         if (!page->private)
634                 goto out;
635
636         len = page->private >> 2;
637         WARN_ON(len == 0);
638
639         eb = alloc_extent_buffer(tree, start, len, page);
640         if (eb == NULL)
641                 goto out;
642
643         if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
644                 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
645                 btree_readahead_hook(root, eb, eb->start, -EIO);
646         }
647         free_extent_buffer(eb);
648
649 out:
650         return -EIO;    /* we fixed nothing */
651 }
652
653 static void end_workqueue_bio(struct bio *bio, int err)
654 {
655         struct end_io_wq *end_io_wq = bio->bi_private;
656         struct btrfs_fs_info *fs_info;
657
658         fs_info = end_io_wq->info;
659         end_io_wq->error = err;
660         end_io_wq->work.func = end_workqueue_fn;
661         end_io_wq->work.flags = 0;
662
663         if (bio->bi_rw & REQ_WRITE) {
664                 if (end_io_wq->metadata == 1)
665                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
666                                            &end_io_wq->work);
667                 else if (end_io_wq->metadata == 2)
668                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
669                                            &end_io_wq->work);
670                 else
671                         btrfs_queue_worker(&fs_info->endio_write_workers,
672                                            &end_io_wq->work);
673         } else {
674                 if (end_io_wq->metadata)
675                         btrfs_queue_worker(&fs_info->endio_meta_workers,
676                                            &end_io_wq->work);
677                 else
678                         btrfs_queue_worker(&fs_info->endio_workers,
679                                            &end_io_wq->work);
680         }
681 }
682
683 /*
684  * For the metadata arg you want
685  *
686  * 0 - if data
687  * 1 - if normal metadta
688  * 2 - if writing to the free space cache area
689  */
690 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
691                         int metadata)
692 {
693         struct end_io_wq *end_io_wq;
694         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
695         if (!end_io_wq)
696                 return -ENOMEM;
697
698         end_io_wq->private = bio->bi_private;
699         end_io_wq->end_io = bio->bi_end_io;
700         end_io_wq->info = info;
701         end_io_wq->error = 0;
702         end_io_wq->bio = bio;
703         end_io_wq->metadata = metadata;
704
705         bio->bi_private = end_io_wq;
706         bio->bi_end_io = end_workqueue_bio;
707         return 0;
708 }
709
710 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
711 {
712         unsigned long limit = min_t(unsigned long,
713                                     info->workers.max_workers,
714                                     info->fs_devices->open_devices);
715         return 256 * limit;
716 }
717
718 static void run_one_async_start(struct btrfs_work *work)
719 {
720         struct async_submit_bio *async;
721
722         async = container_of(work, struct  async_submit_bio, work);
723         async->submit_bio_start(async->inode, async->rw, async->bio,
724                                async->mirror_num, async->bio_flags,
725                                async->bio_offset);
726 }
727
728 static void run_one_async_done(struct btrfs_work *work)
729 {
730         struct btrfs_fs_info *fs_info;
731         struct async_submit_bio *async;
732         int limit;
733
734         async = container_of(work, struct  async_submit_bio, work);
735         fs_info = BTRFS_I(async->inode)->root->fs_info;
736
737         limit = btrfs_async_submit_limit(fs_info);
738         limit = limit * 2 / 3;
739
740         atomic_dec(&fs_info->nr_async_submits);
741
742         if (atomic_read(&fs_info->nr_async_submits) < limit &&
743             waitqueue_active(&fs_info->async_submit_wait))
744                 wake_up(&fs_info->async_submit_wait);
745
746         async->submit_bio_done(async->inode, async->rw, async->bio,
747                                async->mirror_num, async->bio_flags,
748                                async->bio_offset);
749 }
750
751 static void run_one_async_free(struct btrfs_work *work)
752 {
753         struct async_submit_bio *async;
754
755         async = container_of(work, struct  async_submit_bio, work);
756         kfree(async);
757 }
758
759 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
760                         int rw, struct bio *bio, int mirror_num,
761                         unsigned long bio_flags,
762                         u64 bio_offset,
763                         extent_submit_bio_hook_t *submit_bio_start,
764                         extent_submit_bio_hook_t *submit_bio_done)
765 {
766         struct async_submit_bio *async;
767
768         async = kmalloc(sizeof(*async), GFP_NOFS);
769         if (!async)
770                 return -ENOMEM;
771
772         async->inode = inode;
773         async->rw = rw;
774         async->bio = bio;
775         async->mirror_num = mirror_num;
776         async->submit_bio_start = submit_bio_start;
777         async->submit_bio_done = submit_bio_done;
778
779         async->work.func = run_one_async_start;
780         async->work.ordered_func = run_one_async_done;
781         async->work.ordered_free = run_one_async_free;
782
783         async->work.flags = 0;
784         async->bio_flags = bio_flags;
785         async->bio_offset = bio_offset;
786
787         atomic_inc(&fs_info->nr_async_submits);
788
789         if (rw & REQ_SYNC)
790                 btrfs_set_work_high_prio(&async->work);
791
792         btrfs_queue_worker(&fs_info->workers, &async->work);
793
794         while (atomic_read(&fs_info->async_submit_draining) &&
795               atomic_read(&fs_info->nr_async_submits)) {
796                 wait_event(fs_info->async_submit_wait,
797                            (atomic_read(&fs_info->nr_async_submits) == 0));
798         }
799
800         return 0;
801 }
802
803 static int btree_csum_one_bio(struct bio *bio)
804 {
805         struct bio_vec *bvec = bio->bi_io_vec;
806         int bio_index = 0;
807         struct btrfs_root *root;
808
809         WARN_ON(bio->bi_vcnt <= 0);
810         while (bio_index < bio->bi_vcnt) {
811                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
812                 csum_dirty_buffer(root, bvec->bv_page);
813                 bio_index++;
814                 bvec++;
815         }
816         return 0;
817 }
818
819 static int __btree_submit_bio_start(struct inode *inode, int rw,
820                                     struct bio *bio, int mirror_num,
821                                     unsigned long bio_flags,
822                                     u64 bio_offset)
823 {
824         /*
825          * when we're called for a write, we're already in the async
826          * submission context.  Just jump into btrfs_map_bio
827          */
828         btree_csum_one_bio(bio);
829         return 0;
830 }
831
832 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
833                                  int mirror_num, unsigned long bio_flags,
834                                  u64 bio_offset)
835 {
836         /*
837          * when we're called for a write, we're already in the async
838          * submission context.  Just jump into btrfs_map_bio
839          */
840         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
841 }
842
843 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
844                                  int mirror_num, unsigned long bio_flags,
845                                  u64 bio_offset)
846 {
847         int ret;
848
849         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
850                                           bio, 1);
851         BUG_ON(ret);
852
853         if (!(rw & REQ_WRITE)) {
854                 /*
855                  * called for a read, do the setup so that checksum validation
856                  * can happen in the async kernel threads
857                  */
858                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
859                                      mirror_num, 0);
860         }
861
862         /*
863          * kthread helpers are used to submit writes so that checksumming
864          * can happen in parallel across all CPUs
865          */
866         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
867                                    inode, rw, bio, mirror_num, 0,
868                                    bio_offset,
869                                    __btree_submit_bio_start,
870                                    __btree_submit_bio_done);
871 }
872
873 #ifdef CONFIG_MIGRATION
874 static int btree_migratepage(struct address_space *mapping,
875                         struct page *newpage, struct page *page,
876                         enum migrate_mode mode)
877 {
878         /*
879          * we can't safely write a btree page from here,
880          * we haven't done the locking hook
881          */
882         if (PageDirty(page))
883                 return -EAGAIN;
884         /*
885          * Buffers may be managed in a filesystem specific way.
886          * We must have no buffers or drop them.
887          */
888         if (page_has_private(page) &&
889             !try_to_release_page(page, GFP_KERNEL))
890                 return -EAGAIN;
891         return migrate_page(mapping, newpage, page, mode);
892 }
893 #endif
894
895 static int btree_writepage(struct page *page, struct writeback_control *wbc)
896 {
897         struct extent_io_tree *tree;
898         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
899         struct extent_buffer *eb;
900         int was_dirty;
901
902         tree = &BTRFS_I(page->mapping->host)->io_tree;
903         if (!(current->flags & PF_MEMALLOC)) {
904                 return extent_write_full_page(tree, page,
905                                               btree_get_extent, wbc);
906         }
907
908         redirty_page_for_writepage(wbc, page);
909         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
910         WARN_ON(!eb);
911
912         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
913         if (!was_dirty) {
914                 spin_lock(&root->fs_info->delalloc_lock);
915                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
916                 spin_unlock(&root->fs_info->delalloc_lock);
917         }
918         free_extent_buffer(eb);
919
920         unlock_page(page);
921         return 0;
922 }
923
924 static int btree_writepages(struct address_space *mapping,
925                             struct writeback_control *wbc)
926 {
927         struct extent_io_tree *tree;
928         tree = &BTRFS_I(mapping->host)->io_tree;
929         if (wbc->sync_mode == WB_SYNC_NONE) {
930                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
931                 u64 num_dirty;
932                 unsigned long thresh = 32 * 1024 * 1024;
933
934                 if (wbc->for_kupdate)
935                         return 0;
936
937                 /* this is a bit racy, but that's ok */
938                 num_dirty = root->fs_info->dirty_metadata_bytes;
939                 if (num_dirty < thresh)
940                         return 0;
941         }
942         return extent_writepages(tree, mapping, btree_get_extent, wbc);
943 }
944
945 static int btree_readpage(struct file *file, struct page *page)
946 {
947         struct extent_io_tree *tree;
948         tree = &BTRFS_I(page->mapping->host)->io_tree;
949         return extent_read_full_page(tree, page, btree_get_extent, 0);
950 }
951
952 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
953 {
954         struct extent_io_tree *tree;
955         struct extent_map_tree *map;
956         int ret;
957
958         if (PageWriteback(page) || PageDirty(page))
959                 return 0;
960
961         tree = &BTRFS_I(page->mapping->host)->io_tree;
962         map = &BTRFS_I(page->mapping->host)->extent_tree;
963
964         ret = try_release_extent_state(map, tree, page, gfp_flags);
965         if (!ret)
966                 return 0;
967
968         ret = try_release_extent_buffer(tree, page);
969         if (ret == 1) {
970                 ClearPagePrivate(page);
971                 set_page_private(page, 0);
972                 page_cache_release(page);
973         }
974
975         return ret;
976 }
977
978 static void btree_invalidatepage(struct page *page, unsigned long offset)
979 {
980         struct extent_io_tree *tree;
981         tree = &BTRFS_I(page->mapping->host)->io_tree;
982         extent_invalidatepage(tree, page, offset);
983         btree_releasepage(page, GFP_NOFS);
984         if (PagePrivate(page)) {
985                 printk(KERN_WARNING "btrfs warning page private not zero "
986                        "on page %llu\n", (unsigned long long)page_offset(page));
987                 ClearPagePrivate(page);
988                 set_page_private(page, 0);
989                 page_cache_release(page);
990         }
991 }
992
993 static const struct address_space_operations btree_aops = {
994         .readpage       = btree_readpage,
995         .writepage      = btree_writepage,
996         .writepages     = btree_writepages,
997         .releasepage    = btree_releasepage,
998         .invalidatepage = btree_invalidatepage,
999 #ifdef CONFIG_MIGRATION
1000         .migratepage    = btree_migratepage,
1001 #endif
1002 };
1003
1004 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1005                          u64 parent_transid)
1006 {
1007         struct extent_buffer *buf = NULL;
1008         struct inode *btree_inode = root->fs_info->btree_inode;
1009         int ret = 0;
1010
1011         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1012         if (!buf)
1013                 return 0;
1014         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1015                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1016         free_extent_buffer(buf);
1017         return ret;
1018 }
1019
1020 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1021                          int mirror_num, struct extent_buffer **eb)
1022 {
1023         struct extent_buffer *buf = NULL;
1024         struct inode *btree_inode = root->fs_info->btree_inode;
1025         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1026         int ret;
1027
1028         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1029         if (!buf)
1030                 return 0;
1031
1032         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1033
1034         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1035                                        btree_get_extent, mirror_num);
1036         if (ret) {
1037                 free_extent_buffer(buf);
1038                 return ret;
1039         }
1040
1041         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1042                 free_extent_buffer(buf);
1043                 return -EIO;
1044         } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1045                 *eb = buf;
1046         } else {
1047                 free_extent_buffer(buf);
1048         }
1049         return 0;
1050 }
1051
1052 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1053                                             u64 bytenr, u32 blocksize)
1054 {
1055         struct inode *btree_inode = root->fs_info->btree_inode;
1056         struct extent_buffer *eb;
1057         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1058                                 bytenr, blocksize);
1059         return eb;
1060 }
1061
1062 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1063                                                  u64 bytenr, u32 blocksize)
1064 {
1065         struct inode *btree_inode = root->fs_info->btree_inode;
1066         struct extent_buffer *eb;
1067
1068         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1069                                  bytenr, blocksize, NULL);
1070         return eb;
1071 }
1072
1073
1074 int btrfs_write_tree_block(struct extent_buffer *buf)
1075 {
1076         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1077                                         buf->start + buf->len - 1);
1078 }
1079
1080 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1081 {
1082         return filemap_fdatawait_range(buf->first_page->mapping,
1083                                        buf->start, buf->start + buf->len - 1);
1084 }
1085
1086 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1087                                       u32 blocksize, u64 parent_transid)
1088 {
1089         struct extent_buffer *buf = NULL;
1090         int ret;
1091
1092         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1093         if (!buf)
1094                 return NULL;
1095
1096         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1097
1098         if (ret == 0)
1099                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1100         return buf;
1101
1102 }
1103
1104 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1105                      struct extent_buffer *buf)
1106 {
1107         struct inode *btree_inode = root->fs_info->btree_inode;
1108         if (btrfs_header_generation(buf) ==
1109             root->fs_info->running_transaction->transid) {
1110                 btrfs_assert_tree_locked(buf);
1111
1112                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1113                         spin_lock(&root->fs_info->delalloc_lock);
1114                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1115                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1116                         else
1117                                 WARN_ON(1);
1118                         spin_unlock(&root->fs_info->delalloc_lock);
1119                 }
1120
1121                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1122                 btrfs_set_lock_blocking(buf);
1123                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1124                                           buf);
1125         }
1126         return 0;
1127 }
1128
1129 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1130                         u32 stripesize, struct btrfs_root *root,
1131                         struct btrfs_fs_info *fs_info,
1132                         u64 objectid)
1133 {
1134         root->node = NULL;
1135         root->commit_root = NULL;
1136         root->sectorsize = sectorsize;
1137         root->nodesize = nodesize;
1138         root->leafsize = leafsize;
1139         root->stripesize = stripesize;
1140         root->ref_cows = 0;
1141         root->track_dirty = 0;
1142         root->in_radix = 0;
1143         root->orphan_item_inserted = 0;
1144         root->orphan_cleanup_state = 0;
1145
1146         root->fs_info = fs_info;
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 *alloc_log_tree(struct btrfs_trans_handle *trans,
1221                                          struct btrfs_fs_info *fs_info)
1222 {
1223         struct btrfs_root *root;
1224         struct btrfs_root *tree_root = fs_info->tree_root;
1225         struct extent_buffer *leaf;
1226
1227         root = kzalloc(sizeof(*root), GFP_NOFS);
1228         if (!root)
1229                 return ERR_PTR(-ENOMEM);
1230
1231         __setup_root(tree_root->nodesize, tree_root->leafsize,
1232                      tree_root->sectorsize, tree_root->stripesize,
1233                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1234
1235         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1236         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1237         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1238         /*
1239          * log trees do not get reference counted because they go away
1240          * before a real commit is actually done.  They do store pointers
1241          * to file data extents, and those reference counts still get
1242          * updated (along with back refs to the log tree).
1243          */
1244         root->ref_cows = 0;
1245
1246         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1247                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1248         if (IS_ERR(leaf)) {
1249                 kfree(root);
1250                 return ERR_CAST(leaf);
1251         }
1252
1253         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1254         btrfs_set_header_bytenr(leaf, leaf->start);
1255         btrfs_set_header_generation(leaf, trans->transid);
1256         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1257         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1258         root->node = leaf;
1259
1260         write_extent_buffer(root->node, root->fs_info->fsid,
1261                             (unsigned long)btrfs_header_fsid(root->node),
1262                             BTRFS_FSID_SIZE);
1263         btrfs_mark_buffer_dirty(root->node);
1264         btrfs_tree_unlock(root->node);
1265         return root;
1266 }
1267
1268 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1269                              struct btrfs_fs_info *fs_info)
1270 {
1271         struct btrfs_root *log_root;
1272
1273         log_root = alloc_log_tree(trans, fs_info);
1274         if (IS_ERR(log_root))
1275                 return PTR_ERR(log_root);
1276         WARN_ON(fs_info->log_root_tree);
1277         fs_info->log_root_tree = log_root;
1278         return 0;
1279 }
1280
1281 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1282                        struct btrfs_root *root)
1283 {
1284         struct btrfs_root *log_root;
1285         struct btrfs_inode_item *inode_item;
1286
1287         log_root = alloc_log_tree(trans, root->fs_info);
1288         if (IS_ERR(log_root))
1289                 return PTR_ERR(log_root);
1290
1291         log_root->last_trans = trans->transid;
1292         log_root->root_key.offset = root->root_key.objectid;
1293
1294         inode_item = &log_root->root_item.inode;
1295         inode_item->generation = cpu_to_le64(1);
1296         inode_item->size = cpu_to_le64(3);
1297         inode_item->nlink = cpu_to_le32(1);
1298         inode_item->nbytes = cpu_to_le64(root->leafsize);
1299         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1300
1301         btrfs_set_root_node(&log_root->root_item, log_root->node);
1302
1303         WARN_ON(root->log_root);
1304         root->log_root = log_root;
1305         root->log_transid = 0;
1306         root->last_log_commit = 0;
1307         return 0;
1308 }
1309
1310 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1311                                                struct btrfs_key *location)
1312 {
1313         struct btrfs_root *root;
1314         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1315         struct btrfs_path *path;
1316         struct extent_buffer *l;
1317         u64 generation;
1318         u32 blocksize;
1319         int ret = 0;
1320
1321         root = kzalloc(sizeof(*root), GFP_NOFS);
1322         if (!root)
1323                 return ERR_PTR(-ENOMEM);
1324         if (location->offset == (u64)-1) {
1325                 ret = find_and_setup_root(tree_root, fs_info,
1326                                           location->objectid, root);
1327                 if (ret) {
1328                         kfree(root);
1329                         return ERR_PTR(ret);
1330                 }
1331                 goto out;
1332         }
1333
1334         __setup_root(tree_root->nodesize, tree_root->leafsize,
1335                      tree_root->sectorsize, tree_root->stripesize,
1336                      root, fs_info, location->objectid);
1337
1338         path = btrfs_alloc_path();
1339         if (!path) {
1340                 kfree(root);
1341                 return ERR_PTR(-ENOMEM);
1342         }
1343         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1344         if (ret == 0) {
1345                 l = path->nodes[0];
1346                 read_extent_buffer(l, &root->root_item,
1347                                 btrfs_item_ptr_offset(l, path->slots[0]),
1348                                 sizeof(root->root_item));
1349                 memcpy(&root->root_key, location, sizeof(*location));
1350         }
1351         btrfs_free_path(path);
1352         if (ret) {
1353                 kfree(root);
1354                 if (ret > 0)
1355                         ret = -ENOENT;
1356                 return ERR_PTR(ret);
1357         }
1358
1359         generation = btrfs_root_generation(&root->root_item);
1360         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1361         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1362                                      blocksize, generation);
1363         root->commit_root = btrfs_root_node(root);
1364         BUG_ON(!root->node);
1365 out:
1366         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1367                 root->ref_cows = 1;
1368                 btrfs_check_and_init_root_item(&root->root_item);
1369         }
1370
1371         return root;
1372 }
1373
1374 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1375                                               struct btrfs_key *location)
1376 {
1377         struct btrfs_root *root;
1378         int ret;
1379
1380         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1381                 return fs_info->tree_root;
1382         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1383                 return fs_info->extent_root;
1384         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1385                 return fs_info->chunk_root;
1386         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1387                 return fs_info->dev_root;
1388         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1389                 return fs_info->csum_root;
1390 again:
1391         spin_lock(&fs_info->fs_roots_radix_lock);
1392         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1393                                  (unsigned long)location->objectid);
1394         spin_unlock(&fs_info->fs_roots_radix_lock);
1395         if (root)
1396                 return root;
1397
1398         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1399         if (IS_ERR(root))
1400                 return root;
1401
1402         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1403         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1404                                         GFP_NOFS);
1405         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1406                 ret = -ENOMEM;
1407                 goto fail;
1408         }
1409
1410         btrfs_init_free_ino_ctl(root);
1411         mutex_init(&root->fs_commit_mutex);
1412         spin_lock_init(&root->cache_lock);
1413         init_waitqueue_head(&root->cache_wait);
1414
1415         ret = get_anon_bdev(&root->anon_dev);
1416         if (ret)
1417                 goto fail;
1418
1419         if (btrfs_root_refs(&root->root_item) == 0) {
1420                 ret = -ENOENT;
1421                 goto fail;
1422         }
1423
1424         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1425         if (ret < 0)
1426                 goto fail;
1427         if (ret == 0)
1428                 root->orphan_item_inserted = 1;
1429
1430         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1431         if (ret)
1432                 goto fail;
1433
1434         spin_lock(&fs_info->fs_roots_radix_lock);
1435         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1436                                 (unsigned long)root->root_key.objectid,
1437                                 root);
1438         if (ret == 0)
1439                 root->in_radix = 1;
1440
1441         spin_unlock(&fs_info->fs_roots_radix_lock);
1442         radix_tree_preload_end();
1443         if (ret) {
1444                 if (ret == -EEXIST) {
1445                         free_fs_root(root);
1446                         goto again;
1447                 }
1448                 goto fail;
1449         }
1450
1451         ret = btrfs_find_dead_roots(fs_info->tree_root,
1452                                     root->root_key.objectid);
1453         WARN_ON(ret);
1454         return root;
1455 fail:
1456         free_fs_root(root);
1457         return ERR_PTR(ret);
1458 }
1459
1460 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1461 {
1462         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1463         int ret = 0;
1464         struct btrfs_device *device;
1465         struct backing_dev_info *bdi;
1466
1467         rcu_read_lock();
1468         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1469                 if (!device->bdev)
1470                         continue;
1471                 bdi = blk_get_backing_dev_info(device->bdev);
1472                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1473                         ret = 1;
1474                         break;
1475                 }
1476         }
1477         rcu_read_unlock();
1478         return ret;
1479 }
1480
1481 /*
1482  * If this fails, caller must call bdi_destroy() to get rid of the
1483  * bdi again.
1484  */
1485 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1486 {
1487         int err;
1488
1489         bdi->capabilities = BDI_CAP_MAP_COPY;
1490         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1491         if (err)
1492                 return err;
1493
1494         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1495         bdi->congested_fn       = btrfs_congested_fn;
1496         bdi->congested_data     = info;
1497         return 0;
1498 }
1499
1500 static int bio_ready_for_csum(struct bio *bio)
1501 {
1502         u64 length = 0;
1503         u64 buf_len = 0;
1504         u64 start = 0;
1505         struct page *page;
1506         struct extent_io_tree *io_tree = NULL;
1507         struct bio_vec *bvec;
1508         int i;
1509         int ret;
1510
1511         bio_for_each_segment(bvec, bio, i) {
1512                 page = bvec->bv_page;
1513                 if (page->private == EXTENT_PAGE_PRIVATE) {
1514                         length += bvec->bv_len;
1515                         continue;
1516                 }
1517                 if (!page->private) {
1518                         length += bvec->bv_len;
1519                         continue;
1520                 }
1521                 length = bvec->bv_len;
1522                 buf_len = page->private >> 2;
1523                 start = page_offset(page) + bvec->bv_offset;
1524                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1525         }
1526         /* are we fully contained in this bio? */
1527         if (buf_len <= length)
1528                 return 1;
1529
1530         ret = extent_range_uptodate(io_tree, start + length,
1531                                     start + buf_len - 1);
1532         return ret;
1533 }
1534
1535 /*
1536  * called by the kthread helper functions to finally call the bio end_io
1537  * functions.  This is where read checksum verification actually happens
1538  */
1539 static void end_workqueue_fn(struct btrfs_work *work)
1540 {
1541         struct bio *bio;
1542         struct end_io_wq *end_io_wq;
1543         struct btrfs_fs_info *fs_info;
1544         int error;
1545
1546         end_io_wq = container_of(work, struct end_io_wq, work);
1547         bio = end_io_wq->bio;
1548         fs_info = end_io_wq->info;
1549
1550         /* metadata bio reads are special because the whole tree block must
1551          * be checksummed at once.  This makes sure the entire block is in
1552          * ram and up to date before trying to verify things.  For
1553          * blocksize <= pagesize, it is basically a noop
1554          */
1555         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1556             !bio_ready_for_csum(bio)) {
1557                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1558                                    &end_io_wq->work);
1559                 return;
1560         }
1561         error = end_io_wq->error;
1562         bio->bi_private = end_io_wq->private;
1563         bio->bi_end_io = end_io_wq->end_io;
1564         kfree(end_io_wq);
1565         bio_endio(bio, error);
1566 }
1567
1568 static int cleaner_kthread(void *arg)
1569 {
1570         struct btrfs_root *root = arg;
1571
1572         do {
1573                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1574
1575                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1576                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1577                         btrfs_run_delayed_iputs(root);
1578                         btrfs_clean_old_snapshots(root);
1579                         mutex_unlock(&root->fs_info->cleaner_mutex);
1580                         btrfs_run_defrag_inodes(root->fs_info);
1581                 }
1582
1583                 if (!try_to_freeze()) {
1584                         set_current_state(TASK_INTERRUPTIBLE);
1585                         if (!kthread_should_stop())
1586                                 schedule();
1587                         __set_current_state(TASK_RUNNING);
1588                 }
1589         } while (!kthread_should_stop());
1590         return 0;
1591 }
1592
1593 static int transaction_kthread(void *arg)
1594 {
1595         struct btrfs_root *root = arg;
1596         struct btrfs_trans_handle *trans;
1597         struct btrfs_transaction *cur;
1598         u64 transid;
1599         unsigned long now;
1600         unsigned long delay;
1601         int ret;
1602
1603         do {
1604                 delay = HZ * 30;
1605                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1606                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1607
1608                 spin_lock(&root->fs_info->trans_lock);
1609                 cur = root->fs_info->running_transaction;
1610                 if (!cur) {
1611                         spin_unlock(&root->fs_info->trans_lock);
1612                         goto sleep;
1613                 }
1614
1615                 now = get_seconds();
1616                 if (!cur->blocked &&
1617                     (now < cur->start_time || now - cur->start_time < 30)) {
1618                         spin_unlock(&root->fs_info->trans_lock);
1619                         delay = HZ * 5;
1620                         goto sleep;
1621                 }
1622                 transid = cur->transid;
1623                 spin_unlock(&root->fs_info->trans_lock);
1624
1625                 trans = btrfs_join_transaction(root);
1626                 BUG_ON(IS_ERR(trans));
1627                 if (transid == trans->transid) {
1628                         ret = btrfs_commit_transaction(trans, root);
1629                         BUG_ON(ret);
1630                 } else {
1631                         btrfs_end_transaction(trans, root);
1632                 }
1633 sleep:
1634                 wake_up_process(root->fs_info->cleaner_kthread);
1635                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1636
1637                 if (!try_to_freeze()) {
1638                         set_current_state(TASK_INTERRUPTIBLE);
1639                         if (!kthread_should_stop() &&
1640                             !btrfs_transaction_blocked(root->fs_info))
1641                                 schedule_timeout(delay);
1642                         __set_current_state(TASK_RUNNING);
1643                 }
1644         } while (!kthread_should_stop());
1645         return 0;
1646 }
1647
1648 /*
1649  * this will find the highest generation in the array of
1650  * root backups.  The index of the highest array is returned,
1651  * or -1 if we can't find anything.
1652  *
1653  * We check to make sure the array is valid by comparing the
1654  * generation of the latest  root in the array with the generation
1655  * in the super block.  If they don't match we pitch it.
1656  */
1657 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1658 {
1659         u64 cur;
1660         int newest_index = -1;
1661         struct btrfs_root_backup *root_backup;
1662         int i;
1663
1664         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1665                 root_backup = info->super_copy->super_roots + i;
1666                 cur = btrfs_backup_tree_root_gen(root_backup);
1667                 if (cur == newest_gen)
1668                         newest_index = i;
1669         }
1670
1671         /* check to see if we actually wrapped around */
1672         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1673                 root_backup = info->super_copy->super_roots;
1674                 cur = btrfs_backup_tree_root_gen(root_backup);
1675                 if (cur == newest_gen)
1676                         newest_index = 0;
1677         }
1678         return newest_index;
1679 }
1680
1681
1682 /*
1683  * find the oldest backup so we know where to store new entries
1684  * in the backup array.  This will set the backup_root_index
1685  * field in the fs_info struct
1686  */
1687 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1688                                      u64 newest_gen)
1689 {
1690         int newest_index = -1;
1691
1692         newest_index = find_newest_super_backup(info, newest_gen);
1693         /* if there was garbage in there, just move along */
1694         if (newest_index == -1) {
1695                 info->backup_root_index = 0;
1696         } else {
1697                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1698         }
1699 }
1700
1701 /*
1702  * copy all the root pointers into the super backup array.
1703  * this will bump the backup pointer by one when it is
1704  * done
1705  */
1706 static void backup_super_roots(struct btrfs_fs_info *info)
1707 {
1708         int next_backup;
1709         struct btrfs_root_backup *root_backup;
1710         int last_backup;
1711
1712         next_backup = info->backup_root_index;
1713         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1714                 BTRFS_NUM_BACKUP_ROOTS;
1715
1716         /*
1717          * just overwrite the last backup if we're at the same generation
1718          * this happens only at umount
1719          */
1720         root_backup = info->super_for_commit->super_roots + last_backup;
1721         if (btrfs_backup_tree_root_gen(root_backup) ==
1722             btrfs_header_generation(info->tree_root->node))
1723                 next_backup = last_backup;
1724
1725         root_backup = info->super_for_commit->super_roots + next_backup;
1726
1727         /*
1728          * make sure all of our padding and empty slots get zero filled
1729          * regardless of which ones we use today
1730          */
1731         memset(root_backup, 0, sizeof(*root_backup));
1732
1733         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1734
1735         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1736         btrfs_set_backup_tree_root_gen(root_backup,
1737                                btrfs_header_generation(info->tree_root->node));
1738
1739         btrfs_set_backup_tree_root_level(root_backup,
1740                                btrfs_header_level(info->tree_root->node));
1741
1742         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1743         btrfs_set_backup_chunk_root_gen(root_backup,
1744                                btrfs_header_generation(info->chunk_root->node));
1745         btrfs_set_backup_chunk_root_level(root_backup,
1746                                btrfs_header_level(info->chunk_root->node));
1747
1748         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1749         btrfs_set_backup_extent_root_gen(root_backup,
1750                                btrfs_header_generation(info->extent_root->node));
1751         btrfs_set_backup_extent_root_level(root_backup,
1752                                btrfs_header_level(info->extent_root->node));
1753
1754         /*
1755          * we might commit during log recovery, which happens before we set
1756          * the fs_root.  Make sure it is valid before we fill it in.
1757          */
1758         if (info->fs_root && info->fs_root->node) {
1759                 btrfs_set_backup_fs_root(root_backup,
1760                                          info->fs_root->node->start);
1761                 btrfs_set_backup_fs_root_gen(root_backup,
1762                                btrfs_header_generation(info->fs_root->node));
1763                 btrfs_set_backup_fs_root_level(root_backup,
1764                                btrfs_header_level(info->fs_root->node));
1765         }
1766
1767         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1768         btrfs_set_backup_dev_root_gen(root_backup,
1769                                btrfs_header_generation(info->dev_root->node));
1770         btrfs_set_backup_dev_root_level(root_backup,
1771                                        btrfs_header_level(info->dev_root->node));
1772
1773         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1774         btrfs_set_backup_csum_root_gen(root_backup,
1775                                btrfs_header_generation(info->csum_root->node));
1776         btrfs_set_backup_csum_root_level(root_backup,
1777                                btrfs_header_level(info->csum_root->node));
1778
1779         btrfs_set_backup_total_bytes(root_backup,
1780                              btrfs_super_total_bytes(info->super_copy));
1781         btrfs_set_backup_bytes_used(root_backup,
1782                              btrfs_super_bytes_used(info->super_copy));
1783         btrfs_set_backup_num_devices(root_backup,
1784                              btrfs_super_num_devices(info->super_copy));
1785
1786         /*
1787          * if we don't copy this out to the super_copy, it won't get remembered
1788          * for the next commit
1789          */
1790         memcpy(&info->super_copy->super_roots,
1791                &info->super_for_commit->super_roots,
1792                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1793 }
1794
1795 /*
1796  * this copies info out of the root backup array and back into
1797  * the in-memory super block.  It is meant to help iterate through
1798  * the array, so you send it the number of backups you've already
1799  * tried and the last backup index you used.
1800  *
1801  * this returns -1 when it has tried all the backups
1802  */
1803 static noinline int next_root_backup(struct btrfs_fs_info *info,
1804                                      struct btrfs_super_block *super,
1805                                      int *num_backups_tried, int *backup_index)
1806 {
1807         struct btrfs_root_backup *root_backup;
1808         int newest = *backup_index;
1809
1810         if (*num_backups_tried == 0) {
1811                 u64 gen = btrfs_super_generation(super);
1812
1813                 newest = find_newest_super_backup(info, gen);
1814                 if (newest == -1)
1815                         return -1;
1816
1817                 *backup_index = newest;
1818                 *num_backups_tried = 1;
1819         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1820                 /* we've tried all the backups, all done */
1821                 return -1;
1822         } else {
1823                 /* jump to the next oldest backup */
1824                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1825                         BTRFS_NUM_BACKUP_ROOTS;
1826                 *backup_index = newest;
1827                 *num_backups_tried += 1;
1828         }
1829         root_backup = super->super_roots + newest;
1830
1831         btrfs_set_super_generation(super,
1832                                    btrfs_backup_tree_root_gen(root_backup));
1833         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1834         btrfs_set_super_root_level(super,
1835                                    btrfs_backup_tree_root_level(root_backup));
1836         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1837
1838         /*
1839          * fixme: the total bytes and num_devices need to match or we should
1840          * need a fsck
1841          */
1842         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1843         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1844         return 0;
1845 }
1846
1847 /* helper to cleanup tree roots */
1848 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1849 {
1850         free_extent_buffer(info->tree_root->node);
1851         free_extent_buffer(info->tree_root->commit_root);
1852         free_extent_buffer(info->dev_root->node);
1853         free_extent_buffer(info->dev_root->commit_root);
1854         free_extent_buffer(info->extent_root->node);
1855         free_extent_buffer(info->extent_root->commit_root);
1856         free_extent_buffer(info->csum_root->node);
1857         free_extent_buffer(info->csum_root->commit_root);
1858
1859         info->tree_root->node = NULL;
1860         info->tree_root->commit_root = NULL;
1861         info->dev_root->node = NULL;
1862         info->dev_root->commit_root = NULL;
1863         info->extent_root->node = NULL;
1864         info->extent_root->commit_root = NULL;
1865         info->csum_root->node = NULL;
1866         info->csum_root->commit_root = NULL;
1867
1868         if (chunk_root) {
1869                 free_extent_buffer(info->chunk_root->node);
1870                 free_extent_buffer(info->chunk_root->commit_root);
1871                 info->chunk_root->node = NULL;
1872                 info->chunk_root->commit_root = NULL;
1873         }
1874 }
1875
1876
1877 struct btrfs_root *open_ctree(struct super_block *sb,
1878                               struct btrfs_fs_devices *fs_devices,
1879                               char *options)
1880 {
1881         u32 sectorsize;
1882         u32 nodesize;
1883         u32 leafsize;
1884         u32 blocksize;
1885         u32 stripesize;
1886         u64 generation;
1887         u64 features;
1888         struct btrfs_key location;
1889         struct buffer_head *bh;
1890         struct btrfs_super_block *disk_super;
1891         struct btrfs_root *tree_root = btrfs_sb(sb);
1892         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1893         struct btrfs_root *extent_root;
1894         struct btrfs_root *csum_root;
1895         struct btrfs_root *chunk_root;
1896         struct btrfs_root *dev_root;
1897         struct btrfs_root *log_tree_root;
1898         int ret;
1899         int err = -EINVAL;
1900         int num_backups_tried = 0;
1901         int backup_index = 0;
1902
1903         extent_root = fs_info->extent_root =
1904                 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1905         csum_root = fs_info->csum_root =
1906                 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1907         chunk_root = fs_info->chunk_root =
1908                 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1909         dev_root = fs_info->dev_root =
1910                 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1911
1912         if (!extent_root || !csum_root || !chunk_root || !dev_root) {
1913                 err = -ENOMEM;
1914                 goto fail;
1915         }
1916
1917         ret = init_srcu_struct(&fs_info->subvol_srcu);
1918         if (ret) {
1919                 err = ret;
1920                 goto fail;
1921         }
1922
1923         ret = setup_bdi(fs_info, &fs_info->bdi);
1924         if (ret) {
1925                 err = ret;
1926                 goto fail_srcu;
1927         }
1928
1929         fs_info->btree_inode = new_inode(sb);
1930         if (!fs_info->btree_inode) {
1931                 err = -ENOMEM;
1932                 goto fail_bdi;
1933         }
1934
1935         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1936
1937         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1938         INIT_LIST_HEAD(&fs_info->trans_list);
1939         INIT_LIST_HEAD(&fs_info->dead_roots);
1940         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1941         INIT_LIST_HEAD(&fs_info->hashers);
1942         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1943         INIT_LIST_HEAD(&fs_info->ordered_operations);
1944         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1945         spin_lock_init(&fs_info->delalloc_lock);
1946         spin_lock_init(&fs_info->trans_lock);
1947         spin_lock_init(&fs_info->ref_cache_lock);
1948         spin_lock_init(&fs_info->fs_roots_radix_lock);
1949         spin_lock_init(&fs_info->delayed_iput_lock);
1950         spin_lock_init(&fs_info->defrag_inodes_lock);
1951         spin_lock_init(&fs_info->free_chunk_lock);
1952         mutex_init(&fs_info->reloc_mutex);
1953
1954         init_completion(&fs_info->kobj_unregister);
1955         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1956         INIT_LIST_HEAD(&fs_info->space_info);
1957         btrfs_mapping_init(&fs_info->mapping_tree);
1958         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1959         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1960         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1961         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1962         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1963         btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1964         atomic_set(&fs_info->nr_async_submits, 0);
1965         atomic_set(&fs_info->async_delalloc_pages, 0);
1966         atomic_set(&fs_info->async_submit_draining, 0);
1967         atomic_set(&fs_info->nr_async_bios, 0);
1968         atomic_set(&fs_info->defrag_running, 0);
1969         fs_info->sb = sb;
1970         fs_info->max_inline = 8192 * 1024;
1971         fs_info->metadata_ratio = 0;
1972         fs_info->defrag_inodes = RB_ROOT;
1973         fs_info->trans_no_join = 0;
1974         fs_info->free_chunk_space = 0;
1975
1976         /* readahead state */
1977         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1978         spin_lock_init(&fs_info->reada_lock);
1979
1980         fs_info->thread_pool_size = min_t(unsigned long,
1981                                           num_online_cpus() + 2, 8);
1982
1983         INIT_LIST_HEAD(&fs_info->ordered_extents);
1984         spin_lock_init(&fs_info->ordered_extent_lock);
1985         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1986                                         GFP_NOFS);
1987         if (!fs_info->delayed_root) {
1988                 err = -ENOMEM;
1989                 goto fail_iput;
1990         }
1991         btrfs_init_delayed_root(fs_info->delayed_root);
1992
1993         mutex_init(&fs_info->scrub_lock);
1994         atomic_set(&fs_info->scrubs_running, 0);
1995         atomic_set(&fs_info->scrub_pause_req, 0);
1996         atomic_set(&fs_info->scrubs_paused, 0);
1997         atomic_set(&fs_info->scrub_cancel_req, 0);
1998         init_waitqueue_head(&fs_info->scrub_pause_wait);
1999         init_rwsem(&fs_info->scrub_super_lock);
2000         fs_info->scrub_workers_refcnt = 0;
2001
2002         sb->s_blocksize = 4096;
2003         sb->s_blocksize_bits = blksize_bits(4096);
2004         sb->s_bdi = &fs_info->bdi;
2005
2006         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2007         set_nlink(fs_info->btree_inode, 1);
2008         /*
2009          * we set the i_size on the btree inode to the max possible int.
2010          * the real end of the address space is determined by all of
2011          * the devices in the system
2012          */
2013         fs_info->btree_inode->i_size = OFFSET_MAX;
2014         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2015         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2016
2017         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2018         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2019                              fs_info->btree_inode->i_mapping);
2020         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2021
2022         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2023
2024         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2025         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2026                sizeof(struct btrfs_key));
2027         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
2028         insert_inode_hash(fs_info->btree_inode);
2029
2030         spin_lock_init(&fs_info->block_group_cache_lock);
2031         fs_info->block_group_cache_tree = RB_ROOT;
2032
2033         extent_io_tree_init(&fs_info->freed_extents[0],
2034                              fs_info->btree_inode->i_mapping);
2035         extent_io_tree_init(&fs_info->freed_extents[1],
2036                              fs_info->btree_inode->i_mapping);
2037         fs_info->pinned_extents = &fs_info->freed_extents[0];
2038         fs_info->do_barriers = 1;
2039
2040
2041         mutex_init(&fs_info->ordered_operations_mutex);
2042         mutex_init(&fs_info->tree_log_mutex);
2043         mutex_init(&fs_info->chunk_mutex);
2044         mutex_init(&fs_info->transaction_kthread_mutex);
2045         mutex_init(&fs_info->cleaner_mutex);
2046         mutex_init(&fs_info->volume_mutex);
2047         init_rwsem(&fs_info->extent_commit_sem);
2048         init_rwsem(&fs_info->cleanup_work_sem);
2049         init_rwsem(&fs_info->subvol_sem);
2050
2051         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2052         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2053
2054         init_waitqueue_head(&fs_info->transaction_throttle);
2055         init_waitqueue_head(&fs_info->transaction_wait);
2056         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2057         init_waitqueue_head(&fs_info->async_submit_wait);
2058
2059         __setup_root(4096, 4096, 4096, 4096, tree_root,
2060                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2061
2062         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2063         if (!bh) {
2064                 err = -EINVAL;
2065                 goto fail_alloc;
2066         }
2067
2068         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2069         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2070                sizeof(*fs_info->super_for_commit));
2071         brelse(bh);
2072
2073         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2074
2075         disk_super = fs_info->super_copy;
2076         if (!btrfs_super_root(disk_super))
2077                 goto fail_alloc;
2078
2079         /* check FS state, whether FS is broken. */
2080         fs_info->fs_state |= btrfs_super_flags(disk_super);
2081
2082         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2083
2084         /*
2085          * run through our array of backup supers and setup
2086          * our ring pointer to the oldest one
2087          */
2088         generation = btrfs_super_generation(disk_super);
2089         find_oldest_super_backup(fs_info, generation);
2090
2091         /*
2092          * In the long term, we'll store the compression type in the super
2093          * block, and it'll be used for per file compression control.
2094          */
2095         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2096
2097         ret = btrfs_parse_options(tree_root, options);
2098         if (ret) {
2099                 err = ret;
2100                 goto fail_alloc;
2101         }
2102
2103         features = btrfs_super_incompat_flags(disk_super) &
2104                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2105         if (features) {
2106                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2107                        "unsupported optional features (%Lx).\n",
2108                        (unsigned long long)features);
2109                 err = -EINVAL;
2110                 goto fail_alloc;
2111         }
2112
2113         features = btrfs_super_incompat_flags(disk_super);
2114         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2115         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2116                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2117         btrfs_set_super_incompat_flags(disk_super, features);
2118
2119         features = btrfs_super_compat_ro_flags(disk_super) &
2120                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2121         if (!(sb->s_flags & MS_RDONLY) && features) {
2122                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2123                        "unsupported option features (%Lx).\n",
2124                        (unsigned long long)features);
2125                 err = -EINVAL;
2126                 goto fail_alloc;
2127         }
2128
2129         btrfs_init_workers(&fs_info->generic_worker,
2130                            "genwork", 1, NULL);
2131
2132         btrfs_init_workers(&fs_info->workers, "worker",
2133                            fs_info->thread_pool_size,
2134                            &fs_info->generic_worker);
2135
2136         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2137                            fs_info->thread_pool_size,
2138                            &fs_info->generic_worker);
2139
2140         btrfs_init_workers(&fs_info->submit_workers, "submit",
2141                            min_t(u64, fs_devices->num_devices,
2142                            fs_info->thread_pool_size),
2143                            &fs_info->generic_worker);
2144
2145         btrfs_init_workers(&fs_info->caching_workers, "cache",
2146                            2, &fs_info->generic_worker);
2147
2148         /* a higher idle thresh on the submit workers makes it much more
2149          * likely that bios will be send down in a sane order to the
2150          * devices
2151          */
2152         fs_info->submit_workers.idle_thresh = 64;
2153
2154         fs_info->workers.idle_thresh = 16;
2155         fs_info->workers.ordered = 1;
2156
2157         fs_info->delalloc_workers.idle_thresh = 2;
2158         fs_info->delalloc_workers.ordered = 1;
2159
2160         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2161                            &fs_info->generic_worker);
2162         btrfs_init_workers(&fs_info->endio_workers, "endio",
2163                            fs_info->thread_pool_size,
2164                            &fs_info->generic_worker);
2165         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2166                            fs_info->thread_pool_size,
2167                            &fs_info->generic_worker);
2168         btrfs_init_workers(&fs_info->endio_meta_write_workers,
2169                            "endio-meta-write", fs_info->thread_pool_size,
2170                            &fs_info->generic_worker);
2171         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2172                            fs_info->thread_pool_size,
2173                            &fs_info->generic_worker);
2174         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2175                            1, &fs_info->generic_worker);
2176         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2177                            fs_info->thread_pool_size,
2178                            &fs_info->generic_worker);
2179         btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2180                            fs_info->thread_pool_size,
2181                            &fs_info->generic_worker);
2182
2183         /*
2184          * endios are largely parallel and should have a very
2185          * low idle thresh
2186          */
2187         fs_info->endio_workers.idle_thresh = 4;
2188         fs_info->endio_meta_workers.idle_thresh = 4;
2189
2190         fs_info->endio_write_workers.idle_thresh = 2;
2191         fs_info->endio_meta_write_workers.idle_thresh = 2;
2192         fs_info->readahead_workers.idle_thresh = 2;
2193
2194         /*
2195          * btrfs_start_workers can really only fail because of ENOMEM so just
2196          * return -ENOMEM if any of these fail.
2197          */
2198         ret = btrfs_start_workers(&fs_info->workers);
2199         ret |= btrfs_start_workers(&fs_info->generic_worker);
2200         ret |= btrfs_start_workers(&fs_info->submit_workers);
2201         ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2202         ret |= btrfs_start_workers(&fs_info->fixup_workers);
2203         ret |= btrfs_start_workers(&fs_info->endio_workers);
2204         ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2205         ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2206         ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2207         ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2208         ret |= btrfs_start_workers(&fs_info->delayed_workers);
2209         ret |= btrfs_start_workers(&fs_info->caching_workers);
2210         ret |= btrfs_start_workers(&fs_info->readahead_workers);
2211         if (ret) {
2212                 ret = -ENOMEM;
2213                 goto fail_sb_buffer;
2214         }
2215
2216         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2217         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2218                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2219
2220         nodesize = btrfs_super_nodesize(disk_super);
2221         leafsize = btrfs_super_leafsize(disk_super);
2222         sectorsize = btrfs_super_sectorsize(disk_super);
2223         stripesize = btrfs_super_stripesize(disk_super);
2224         tree_root->nodesize = nodesize;
2225         tree_root->leafsize = leafsize;
2226         tree_root->sectorsize = sectorsize;
2227         tree_root->stripesize = stripesize;
2228
2229         sb->s_blocksize = sectorsize;
2230         sb->s_blocksize_bits = blksize_bits(sectorsize);
2231
2232         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2233                     sizeof(disk_super->magic))) {
2234                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2235                 goto fail_sb_buffer;
2236         }
2237
2238         mutex_lock(&fs_info->chunk_mutex);
2239         ret = btrfs_read_sys_array(tree_root);
2240         mutex_unlock(&fs_info->chunk_mutex);
2241         if (ret) {
2242                 printk(KERN_WARNING "btrfs: failed to read the system "
2243                        "array on %s\n", sb->s_id);
2244                 goto fail_sb_buffer;
2245         }
2246
2247         blocksize = btrfs_level_size(tree_root,
2248                                      btrfs_super_chunk_root_level(disk_super));
2249         generation = btrfs_super_chunk_root_generation(disk_super);
2250
2251         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2252                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2253
2254         chunk_root->node = read_tree_block(chunk_root,
2255                                            btrfs_super_chunk_root(disk_super),
2256                                            blocksize, generation);
2257         BUG_ON(!chunk_root->node);
2258         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2259                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2260                        sb->s_id);
2261                 goto fail_tree_roots;
2262         }
2263         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2264         chunk_root->commit_root = btrfs_root_node(chunk_root);
2265
2266         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2267            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2268            BTRFS_UUID_SIZE);
2269
2270         mutex_lock(&fs_info->chunk_mutex);
2271         ret = btrfs_read_chunk_tree(chunk_root);
2272         mutex_unlock(&fs_info->chunk_mutex);
2273         if (ret) {
2274                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2275                        sb->s_id);
2276                 goto fail_tree_roots;
2277         }
2278
2279         btrfs_close_extra_devices(fs_devices);
2280
2281 retry_root_backup:
2282         blocksize = btrfs_level_size(tree_root,
2283                                      btrfs_super_root_level(disk_super));
2284         generation = btrfs_super_generation(disk_super);
2285
2286         tree_root->node = read_tree_block(tree_root,
2287                                           btrfs_super_root(disk_super),
2288                                           blocksize, generation);
2289         if (!tree_root->node ||
2290             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2291                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2292                        sb->s_id);
2293
2294                 goto recovery_tree_root;
2295         }
2296
2297         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2298         tree_root->commit_root = btrfs_root_node(tree_root);
2299
2300         ret = find_and_setup_root(tree_root, fs_info,
2301                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2302         if (ret)
2303                 goto recovery_tree_root;
2304         extent_root->track_dirty = 1;
2305
2306         ret = find_and_setup_root(tree_root, fs_info,
2307                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2308         if (ret)
2309                 goto recovery_tree_root;
2310         dev_root->track_dirty = 1;
2311
2312         ret = find_and_setup_root(tree_root, fs_info,
2313                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2314         if (ret)
2315                 goto recovery_tree_root;
2316
2317         csum_root->track_dirty = 1;
2318
2319         fs_info->generation = generation;
2320         fs_info->last_trans_committed = generation;
2321         fs_info->data_alloc_profile = (u64)-1;
2322         fs_info->metadata_alloc_profile = (u64)-1;
2323         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2324
2325         ret = btrfs_init_space_info(fs_info);
2326         if (ret) {
2327                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2328                 goto fail_block_groups;
2329         }
2330
2331         ret = btrfs_read_block_groups(extent_root);
2332         if (ret) {
2333                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2334                 goto fail_block_groups;
2335         }
2336
2337         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2338                                                "btrfs-cleaner");
2339         if (IS_ERR(fs_info->cleaner_kthread))
2340                 goto fail_block_groups;
2341
2342         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2343                                                    tree_root,
2344                                                    "btrfs-transaction");
2345         if (IS_ERR(fs_info->transaction_kthread))
2346                 goto fail_cleaner;
2347
2348         if (!btrfs_test_opt(tree_root, SSD) &&
2349             !btrfs_test_opt(tree_root, NOSSD) &&
2350             !fs_info->fs_devices->rotating) {
2351                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2352                        "mode\n");
2353                 btrfs_set_opt(fs_info->mount_opt, SSD);
2354         }
2355
2356         /* do not make disk changes in broken FS */
2357         if (btrfs_super_log_root(disk_super) != 0 &&
2358             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2359                 u64 bytenr = btrfs_super_log_root(disk_super);
2360
2361                 if (fs_devices->rw_devices == 0) {
2362                         printk(KERN_WARNING "Btrfs log replay required "
2363                                "on RO media\n");
2364                         err = -EIO;
2365                         goto fail_trans_kthread;
2366                 }
2367                 blocksize =
2368                      btrfs_level_size(tree_root,
2369                                       btrfs_super_log_root_level(disk_super));
2370
2371                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2372                 if (!log_tree_root) {
2373                         err = -ENOMEM;
2374                         goto fail_trans_kthread;
2375                 }
2376
2377                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2378                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2379
2380                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2381                                                       blocksize,
2382                                                       generation + 1);
2383                 ret = btrfs_recover_log_trees(log_tree_root);
2384                 BUG_ON(ret);
2385
2386                 if (sb->s_flags & MS_RDONLY) {
2387                         ret =  btrfs_commit_super(tree_root);
2388                         BUG_ON(ret);
2389                 }
2390         }
2391
2392         ret = btrfs_find_orphan_roots(tree_root);
2393         BUG_ON(ret);
2394
2395         if (!(sb->s_flags & MS_RDONLY)) {
2396                 ret = btrfs_cleanup_fs_roots(fs_info);
2397                 BUG_ON(ret);
2398
2399                 ret = btrfs_recover_relocation(tree_root);
2400                 if (ret < 0) {
2401                         printk(KERN_WARNING
2402                                "btrfs: failed to recover relocation\n");
2403                         err = -EINVAL;
2404                         goto fail_trans_kthread;
2405                 }
2406         }
2407
2408         location.objectid = BTRFS_FS_TREE_OBJECTID;
2409         location.type = BTRFS_ROOT_ITEM_KEY;
2410         location.offset = (u64)-1;
2411
2412         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2413         if (!fs_info->fs_root)
2414                 goto fail_trans_kthread;
2415         if (IS_ERR(fs_info->fs_root)) {
2416                 err = PTR_ERR(fs_info->fs_root);
2417                 goto fail_trans_kthread;
2418         }
2419
2420         if (!(sb->s_flags & MS_RDONLY)) {
2421                 down_read(&fs_info->cleanup_work_sem);
2422                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2423                 if (!err)
2424                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2425                 up_read(&fs_info->cleanup_work_sem);
2426                 if (err) {
2427                         close_ctree(tree_root);
2428                         return ERR_PTR(err);
2429                 }
2430         }
2431
2432         return tree_root;
2433
2434 fail_trans_kthread:
2435         kthread_stop(fs_info->transaction_kthread);
2436 fail_cleaner:
2437         kthread_stop(fs_info->cleaner_kthread);
2438
2439         /*
2440          * make sure we're done with the btree inode before we stop our
2441          * kthreads
2442          */
2443         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2444         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2445
2446 fail_block_groups:
2447         btrfs_free_block_groups(fs_info);
2448
2449 fail_tree_roots:
2450         free_root_pointers(fs_info, 1);
2451
2452 fail_sb_buffer:
2453         btrfs_stop_workers(&fs_info->generic_worker);
2454         btrfs_stop_workers(&fs_info->readahead_workers);
2455         btrfs_stop_workers(&fs_info->fixup_workers);
2456         btrfs_stop_workers(&fs_info->delalloc_workers);
2457         btrfs_stop_workers(&fs_info->workers);
2458         btrfs_stop_workers(&fs_info->endio_workers);
2459         btrfs_stop_workers(&fs_info->endio_meta_workers);
2460         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2461         btrfs_stop_workers(&fs_info->endio_write_workers);
2462         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2463         btrfs_stop_workers(&fs_info->submit_workers);
2464         btrfs_stop_workers(&fs_info->delayed_workers);
2465         btrfs_stop_workers(&fs_info->caching_workers);
2466 fail_alloc:
2467 fail_iput:
2468         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2469
2470         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2471         iput(fs_info->btree_inode);
2472 fail_bdi:
2473         bdi_destroy(&fs_info->bdi);
2474 fail_srcu:
2475         cleanup_srcu_struct(&fs_info->subvol_srcu);
2476 fail:
2477         btrfs_close_devices(fs_info->fs_devices);
2478         free_fs_info(fs_info);
2479         return ERR_PTR(err);
2480
2481 recovery_tree_root:
2482         if (!btrfs_test_opt(tree_root, RECOVERY))
2483                 goto fail_tree_roots;
2484
2485         free_root_pointers(fs_info, 0);
2486
2487         /* don't use the log in recovery mode, it won't be valid */
2488         btrfs_set_super_log_root(disk_super, 0);
2489
2490         /* we can't trust the free space cache either */
2491         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2492
2493         ret = next_root_backup(fs_info, fs_info->super_copy,
2494                                &num_backups_tried, &backup_index);
2495         if (ret == -1)
2496                 goto fail_block_groups;
2497         goto retry_root_backup;
2498 }
2499
2500 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2501 {
2502         char b[BDEVNAME_SIZE];
2503
2504         if (uptodate) {
2505                 set_buffer_uptodate(bh);
2506         } else {
2507                 printk_ratelimited(KERN_WARNING "lost page write due to "
2508                                         "I/O error on %s\n",
2509                                        bdevname(bh->b_bdev, b));
2510                 /* note, we dont' set_buffer_write_io_error because we have
2511                  * our own ways of dealing with the IO errors
2512                  */
2513                 clear_buffer_uptodate(bh);
2514         }
2515         unlock_buffer(bh);
2516         put_bh(bh);
2517 }
2518
2519 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2520 {
2521         struct buffer_head *bh;
2522         struct buffer_head *latest = NULL;
2523         struct btrfs_super_block *super;
2524         int i;
2525         u64 transid = 0;
2526         u64 bytenr;
2527
2528         /* we would like to check all the supers, but that would make
2529          * a btrfs mount succeed after a mkfs from a different FS.
2530          * So, we need to add a special mount option to scan for
2531          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2532          */
2533         for (i = 0; i < 1; i++) {
2534                 bytenr = btrfs_sb_offset(i);
2535                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2536                         break;
2537                 bh = __bread(bdev, bytenr / 4096, 4096);
2538                 if (!bh)
2539                         continue;
2540
2541                 super = (struct btrfs_super_block *)bh->b_data;
2542                 if (btrfs_super_bytenr(super) != bytenr ||
2543                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2544                             sizeof(super->magic))) {
2545                         brelse(bh);
2546                         continue;
2547                 }
2548
2549                 if (!latest || btrfs_super_generation(super) > transid) {
2550                         brelse(latest);
2551                         latest = bh;
2552                         transid = btrfs_super_generation(super);
2553                 } else {
2554                         brelse(bh);
2555                 }
2556         }
2557         return latest;
2558 }
2559
2560 /*
2561  * this should be called twice, once with wait == 0 and
2562  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2563  * we write are pinned.
2564  *
2565  * They are released when wait == 1 is done.
2566  * max_mirrors must be the same for both runs, and it indicates how
2567  * many supers on this one device should be written.
2568  *
2569  * max_mirrors == 0 means to write them all.
2570  */
2571 static int write_dev_supers(struct btrfs_device *device,
2572                             struct btrfs_super_block *sb,
2573                             int do_barriers, int wait, int max_mirrors)
2574 {
2575         struct buffer_head *bh;
2576         int i;
2577         int ret;
2578         int errors = 0;
2579         u32 crc;
2580         u64 bytenr;
2581
2582         if (max_mirrors == 0)
2583                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2584
2585         for (i = 0; i < max_mirrors; i++) {
2586                 bytenr = btrfs_sb_offset(i);
2587                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2588                         break;
2589
2590                 if (wait) {
2591                         bh = __find_get_block(device->bdev, bytenr / 4096,
2592                                               BTRFS_SUPER_INFO_SIZE);
2593                         BUG_ON(!bh);
2594                         wait_on_buffer(bh);
2595                         if (!buffer_uptodate(bh))
2596                                 errors++;
2597
2598                         /* drop our reference */
2599                         brelse(bh);
2600
2601                         /* drop the reference from the wait == 0 run */
2602                         brelse(bh);
2603                         continue;
2604                 } else {
2605                         btrfs_set_super_bytenr(sb, bytenr);
2606
2607                         crc = ~(u32)0;
2608                         crc = btrfs_csum_data(NULL, (char *)sb +
2609                                               BTRFS_CSUM_SIZE, crc,
2610                                               BTRFS_SUPER_INFO_SIZE -
2611                                               BTRFS_CSUM_SIZE);
2612                         btrfs_csum_final(crc, sb->csum);
2613
2614                         /*
2615                          * one reference for us, and we leave it for the
2616                          * caller
2617                          */
2618                         bh = __getblk(device->bdev, bytenr / 4096,
2619                                       BTRFS_SUPER_INFO_SIZE);
2620                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2621
2622                         /* one reference for submit_bh */
2623                         get_bh(bh);
2624
2625                         set_buffer_uptodate(bh);
2626                         lock_buffer(bh);
2627                         bh->b_end_io = btrfs_end_buffer_write_sync;
2628                 }
2629
2630                 /*
2631                  * we fua the first super.  The others we allow
2632                  * to go down lazy.
2633                  */
2634                 ret = submit_bh(WRITE_FUA, bh);
2635                 if (ret)
2636                         errors++;
2637         }
2638         return errors < i ? 0 : -1;
2639 }
2640
2641 /*
2642  * endio for the write_dev_flush, this will wake anyone waiting
2643  * for the barrier when it is done
2644  */
2645 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2646 {
2647         if (err) {
2648                 if (err == -EOPNOTSUPP)
2649                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2650                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2651         }
2652         if (bio->bi_private)
2653                 complete(bio->bi_private);
2654         bio_put(bio);
2655 }
2656
2657 /*
2658  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
2659  * sent down.  With wait == 1, it waits for the previous flush.
2660  *
2661  * any device where the flush fails with eopnotsupp are flagged as not-barrier
2662  * capable
2663  */
2664 static int write_dev_flush(struct btrfs_device *device, int wait)
2665 {
2666         struct bio *bio;
2667         int ret = 0;
2668
2669         if (device->nobarriers)
2670                 return 0;
2671
2672         if (wait) {
2673                 bio = device->flush_bio;
2674                 if (!bio)
2675                         return 0;
2676
2677                 wait_for_completion(&device->flush_wait);
2678
2679                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2680                         printk("btrfs: disabling barriers on dev %s\n",
2681                                device->name);
2682                         device->nobarriers = 1;
2683                 }
2684                 if (!bio_flagged(bio, BIO_UPTODATE)) {
2685                         ret = -EIO;
2686                 }
2687
2688                 /* drop the reference from the wait == 0 run */
2689                 bio_put(bio);
2690                 device->flush_bio = NULL;
2691
2692                 return ret;
2693         }
2694
2695         /*
2696          * one reference for us, and we leave it for the
2697          * caller
2698          */
2699         device->flush_bio = NULL;;
2700         bio = bio_alloc(GFP_NOFS, 0);
2701         if (!bio)
2702                 return -ENOMEM;
2703
2704         bio->bi_end_io = btrfs_end_empty_barrier;
2705         bio->bi_bdev = device->bdev;
2706         init_completion(&device->flush_wait);
2707         bio->bi_private = &device->flush_wait;
2708         device->flush_bio = bio;
2709
2710         bio_get(bio);
2711         submit_bio(WRITE_FLUSH, bio);
2712
2713         return 0;
2714 }
2715
2716 /*
2717  * send an empty flush down to each device in parallel,
2718  * then wait for them
2719  */
2720 static int barrier_all_devices(struct btrfs_fs_info *info)
2721 {
2722         struct list_head *head;
2723         struct btrfs_device *dev;
2724         int errors = 0;
2725         int ret;
2726
2727         /* send down all the barriers */
2728         head = &info->fs_devices->devices;
2729         list_for_each_entry_rcu(dev, head, dev_list) {
2730                 if (!dev->bdev) {
2731                         errors++;
2732                         continue;
2733                 }
2734                 if (!dev->in_fs_metadata || !dev->writeable)
2735                         continue;
2736
2737                 ret = write_dev_flush(dev, 0);
2738                 if (ret)
2739                         errors++;
2740         }
2741
2742         /* wait for all the barriers */
2743         list_for_each_entry_rcu(dev, head, dev_list) {
2744                 if (!dev->bdev) {
2745                         errors++;
2746                         continue;
2747                 }
2748                 if (!dev->in_fs_metadata || !dev->writeable)
2749                         continue;
2750
2751                 ret = write_dev_flush(dev, 1);
2752                 if (ret)
2753                         errors++;
2754         }
2755         if (errors)
2756                 return -EIO;
2757         return 0;
2758 }
2759
2760 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2761 {
2762         struct list_head *head;
2763         struct btrfs_device *dev;
2764         struct btrfs_super_block *sb;
2765         struct btrfs_dev_item *dev_item;
2766         int ret;
2767         int do_barriers;
2768         int max_errors;
2769         int total_errors = 0;
2770         u64 flags;
2771
2772         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2773         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2774         backup_super_roots(root->fs_info);
2775
2776         sb = root->fs_info->super_for_commit;
2777         dev_item = &sb->dev_item;
2778
2779         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2780         head = &root->fs_info->fs_devices->devices;
2781
2782         if (do_barriers)
2783                 barrier_all_devices(root->fs_info);
2784
2785         list_for_each_entry_rcu(dev, head, dev_list) {
2786                 if (!dev->bdev) {
2787                         total_errors++;
2788                         continue;
2789                 }
2790                 if (!dev->in_fs_metadata || !dev->writeable)
2791                         continue;
2792
2793                 btrfs_set_stack_device_generation(dev_item, 0);
2794                 btrfs_set_stack_device_type(dev_item, dev->type);
2795                 btrfs_set_stack_device_id(dev_item, dev->devid);
2796                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2797                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2798                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2799                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2800                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2801                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2802                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2803
2804                 flags = btrfs_super_flags(sb);
2805                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2806
2807                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2808                 if (ret)
2809                         total_errors++;
2810         }
2811         if (total_errors > max_errors) {
2812                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2813                        total_errors);
2814                 BUG();
2815         }
2816
2817         total_errors = 0;
2818         list_for_each_entry_rcu(dev, head, dev_list) {
2819                 if (!dev->bdev)
2820                         continue;
2821                 if (!dev->in_fs_metadata || !dev->writeable)
2822                         continue;
2823
2824                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2825                 if (ret)
2826                         total_errors++;
2827         }
2828         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2829         if (total_errors > max_errors) {
2830                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2831                        total_errors);
2832                 BUG();
2833         }
2834         return 0;
2835 }
2836
2837 int write_ctree_super(struct btrfs_trans_handle *trans,
2838                       struct btrfs_root *root, int max_mirrors)
2839 {
2840         int ret;
2841
2842         ret = write_all_supers(root, max_mirrors);
2843         return ret;
2844 }
2845
2846 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2847 {
2848         spin_lock(&fs_info->fs_roots_radix_lock);
2849         radix_tree_delete(&fs_info->fs_roots_radix,
2850                           (unsigned long)root->root_key.objectid);
2851         spin_unlock(&fs_info->fs_roots_radix_lock);
2852
2853         if (btrfs_root_refs(&root->root_item) == 0)
2854                 synchronize_srcu(&fs_info->subvol_srcu);
2855
2856         __btrfs_remove_free_space_cache(root->free_ino_pinned);
2857         __btrfs_remove_free_space_cache(root->free_ino_ctl);
2858         free_fs_root(root);
2859         return 0;
2860 }
2861
2862 static void free_fs_root(struct btrfs_root *root)
2863 {
2864         iput(root->cache_inode);
2865         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2866         if (root->anon_dev)
2867                 free_anon_bdev(root->anon_dev);
2868         free_extent_buffer(root->node);
2869         free_extent_buffer(root->commit_root);
2870         kfree(root->free_ino_ctl);
2871         kfree(root->free_ino_pinned);
2872         kfree(root->name);
2873         kfree(root);
2874 }
2875
2876 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2877 {
2878         int ret;
2879         struct btrfs_root *gang[8];
2880         int i;
2881
2882         while (!list_empty(&fs_info->dead_roots)) {
2883                 gang[0] = list_entry(fs_info->dead_roots.next,
2884                                      struct btrfs_root, root_list);
2885                 list_del(&gang[0]->root_list);
2886
2887                 if (gang[0]->in_radix) {
2888                         btrfs_free_fs_root(fs_info, gang[0]);
2889                 } else {
2890                         free_extent_buffer(gang[0]->node);
2891                         free_extent_buffer(gang[0]->commit_root);
2892                         kfree(gang[0]);
2893                 }
2894         }
2895
2896         while (1) {
2897                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2898                                              (void **)gang, 0,
2899                                              ARRAY_SIZE(gang));
2900                 if (!ret)
2901                         break;
2902                 for (i = 0; i < ret; i++)
2903                         btrfs_free_fs_root(fs_info, gang[i]);
2904         }
2905         return 0;
2906 }
2907
2908 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2909 {
2910         u64 root_objectid = 0;
2911         struct btrfs_root *gang[8];
2912         int i;
2913         int ret;
2914
2915         while (1) {
2916                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2917                                              (void **)gang, root_objectid,
2918                                              ARRAY_SIZE(gang));
2919                 if (!ret)
2920                         break;
2921
2922                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2923                 for (i = 0; i < ret; i++) {
2924                         int err;
2925
2926                         root_objectid = gang[i]->root_key.objectid;
2927                         err = btrfs_orphan_cleanup(gang[i]);
2928                         if (err)
2929                                 return err;
2930                 }
2931                 root_objectid++;
2932         }
2933         return 0;
2934 }
2935
2936 int btrfs_commit_super(struct btrfs_root *root)
2937 {
2938         struct btrfs_trans_handle *trans;
2939         int ret;
2940
2941         mutex_lock(&root->fs_info->cleaner_mutex);
2942         btrfs_run_delayed_iputs(root);
2943         btrfs_clean_old_snapshots(root);
2944         mutex_unlock(&root->fs_info->cleaner_mutex);
2945
2946         /* wait until ongoing cleanup work done */
2947         down_write(&root->fs_info->cleanup_work_sem);
2948         up_write(&root->fs_info->cleanup_work_sem);
2949
2950         trans = btrfs_join_transaction(root);
2951         if (IS_ERR(trans))
2952                 return PTR_ERR(trans);
2953         ret = btrfs_commit_transaction(trans, root);
2954         BUG_ON(ret);
2955         /* run commit again to drop the original snapshot */
2956         trans = btrfs_join_transaction(root);
2957         if (IS_ERR(trans))
2958                 return PTR_ERR(trans);
2959         btrfs_commit_transaction(trans, root);
2960         ret = btrfs_write_and_wait_transaction(NULL, root);
2961         BUG_ON(ret);
2962
2963         ret = write_ctree_super(NULL, root, 0);
2964         return ret;
2965 }
2966
2967 int close_ctree(struct btrfs_root *root)
2968 {
2969         struct btrfs_fs_info *fs_info = root->fs_info;
2970         int ret;
2971
2972         fs_info->closing = 1;
2973         smp_mb();
2974
2975         btrfs_scrub_cancel(root);
2976
2977         /* wait for any defraggers to finish */
2978         wait_event(fs_info->transaction_wait,
2979                    (atomic_read(&fs_info->defrag_running) == 0));
2980
2981         /* clear out the rbtree of defraggable inodes */
2982         btrfs_run_defrag_inodes(root->fs_info);
2983
2984         /*
2985          * Here come 2 situations when btrfs is broken to flip readonly:
2986          *
2987          * 1. when btrfs flips readonly somewhere else before
2988          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2989          * and btrfs will skip to write sb directly to keep
2990          * ERROR state on disk.
2991          *
2992          * 2. when btrfs flips readonly just in btrfs_commit_super,
2993          * and in such case, btrfs cannot write sb via btrfs_commit_super,
2994          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2995          * btrfs will cleanup all FS resources first and write sb then.
2996          */
2997         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2998                 ret = btrfs_commit_super(root);
2999                 if (ret)
3000                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3001         }
3002
3003         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3004                 ret = btrfs_error_commit_super(root);
3005                 if (ret)
3006                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3007         }
3008
3009         btrfs_put_block_group_cache(fs_info);
3010
3011         kthread_stop(root->fs_info->transaction_kthread);
3012         kthread_stop(root->fs_info->cleaner_kthread);
3013
3014         fs_info->closing = 2;
3015         smp_mb();
3016
3017         if (fs_info->delalloc_bytes) {
3018                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3019                        (unsigned long long)fs_info->delalloc_bytes);
3020         }
3021         if (fs_info->total_ref_cache_size) {
3022                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3023                        (unsigned long long)fs_info->total_ref_cache_size);
3024         }
3025
3026         free_extent_buffer(fs_info->extent_root->node);
3027         free_extent_buffer(fs_info->extent_root->commit_root);
3028         free_extent_buffer(fs_info->tree_root->node);
3029         free_extent_buffer(fs_info->tree_root->commit_root);
3030         free_extent_buffer(root->fs_info->chunk_root->node);
3031         free_extent_buffer(root->fs_info->chunk_root->commit_root);
3032         free_extent_buffer(root->fs_info->dev_root->node);
3033         free_extent_buffer(root->fs_info->dev_root->commit_root);
3034         free_extent_buffer(root->fs_info->csum_root->node);
3035         free_extent_buffer(root->fs_info->csum_root->commit_root);
3036
3037         btrfs_free_block_groups(root->fs_info);
3038
3039         del_fs_roots(fs_info);
3040
3041         iput(fs_info->btree_inode);
3042
3043         btrfs_stop_workers(&fs_info->generic_worker);
3044         btrfs_stop_workers(&fs_info->fixup_workers);
3045         btrfs_stop_workers(&fs_info->delalloc_workers);
3046         btrfs_stop_workers(&fs_info->workers);
3047         btrfs_stop_workers(&fs_info->endio_workers);
3048         btrfs_stop_workers(&fs_info->endio_meta_workers);
3049         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3050         btrfs_stop_workers(&fs_info->endio_write_workers);
3051         btrfs_stop_workers(&fs_info->endio_freespace_worker);
3052         btrfs_stop_workers(&fs_info->submit_workers);
3053         btrfs_stop_workers(&fs_info->delayed_workers);
3054         btrfs_stop_workers(&fs_info->caching_workers);
3055         btrfs_stop_workers(&fs_info->readahead_workers);
3056
3057         btrfs_close_devices(fs_info->fs_devices);
3058         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3059
3060         bdi_destroy(&fs_info->bdi);
3061         cleanup_srcu_struct(&fs_info->subvol_srcu);
3062
3063         free_fs_info(fs_info);
3064
3065         return 0;
3066 }
3067
3068 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
3069 {
3070         int ret;
3071         struct inode *btree_inode = buf->first_page->mapping->host;
3072
3073         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
3074                                      NULL);
3075         if (!ret)
3076                 return ret;
3077
3078         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3079                                     parent_transid);
3080         return !ret;
3081 }
3082
3083 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3084 {
3085         struct inode *btree_inode = buf->first_page->mapping->host;
3086         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
3087                                           buf);
3088 }
3089
3090 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3091 {
3092         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3093         u64 transid = btrfs_header_generation(buf);
3094         struct inode *btree_inode = root->fs_info->btree_inode;
3095         int was_dirty;
3096
3097         btrfs_assert_tree_locked(buf);
3098         if (transid != root->fs_info->generation) {
3099                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3100                        "found %llu running %llu\n",
3101                         (unsigned long long)buf->start,
3102                         (unsigned long long)transid,
3103                         (unsigned long long)root->fs_info->generation);
3104                 WARN_ON(1);
3105         }
3106         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
3107                                             buf);
3108         if (!was_dirty) {
3109                 spin_lock(&root->fs_info->delalloc_lock);
3110                 root->fs_info->dirty_metadata_bytes += buf->len;
3111                 spin_unlock(&root->fs_info->delalloc_lock);
3112         }
3113 }
3114
3115 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3116 {
3117         /*
3118          * looks as though older kernels can get into trouble with
3119          * this code, they end up stuck in balance_dirty_pages forever
3120          */
3121         u64 num_dirty;
3122         unsigned long thresh = 32 * 1024 * 1024;
3123
3124         if (current->flags & PF_MEMALLOC)
3125                 return;
3126
3127         btrfs_balance_delayed_items(root);
3128
3129         num_dirty = root->fs_info->dirty_metadata_bytes;
3130
3131         if (num_dirty > thresh) {
3132                 balance_dirty_pages_ratelimited_nr(
3133                                    root->fs_info->btree_inode->i_mapping, 1);
3134         }
3135         return;
3136 }
3137
3138 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3139 {
3140         /*
3141          * looks as though older kernels can get into trouble with
3142          * this code, they end up stuck in balance_dirty_pages forever
3143          */
3144         u64 num_dirty;
3145         unsigned long thresh = 32 * 1024 * 1024;
3146
3147         if (current->flags & PF_MEMALLOC)
3148                 return;
3149
3150         num_dirty = root->fs_info->dirty_metadata_bytes;
3151
3152         if (num_dirty > thresh) {
3153                 balance_dirty_pages_ratelimited_nr(
3154                                    root->fs_info->btree_inode->i_mapping, 1);
3155         }
3156         return;
3157 }
3158
3159 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3160 {
3161         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3162         int ret;
3163         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3164         if (ret == 0)
3165                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
3166         return ret;
3167 }
3168
3169 static int btree_lock_page_hook(struct page *page, void *data,
3170                                 void (*flush_fn)(void *))
3171 {
3172         struct inode *inode = page->mapping->host;
3173         struct btrfs_root *root = BTRFS_I(inode)->root;
3174         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3175         struct extent_buffer *eb;
3176         unsigned long len;
3177         u64 bytenr = page_offset(page);
3178
3179         if (page->private == EXTENT_PAGE_PRIVATE)
3180                 goto out;
3181
3182         len = page->private >> 2;
3183         eb = find_extent_buffer(io_tree, bytenr, len);
3184         if (!eb)
3185                 goto out;
3186
3187         if (!btrfs_try_tree_write_lock(eb)) {
3188                 flush_fn(data);
3189                 btrfs_tree_lock(eb);
3190         }
3191         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3192
3193         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3194                 spin_lock(&root->fs_info->delalloc_lock);
3195                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3196                         root->fs_info->dirty_metadata_bytes -= eb->len;
3197                 else
3198                         WARN_ON(1);
3199                 spin_unlock(&root->fs_info->delalloc_lock);
3200         }
3201
3202         btrfs_tree_unlock(eb);
3203         free_extent_buffer(eb);
3204 out:
3205         if (!trylock_page(page)) {
3206                 flush_fn(data);
3207                 lock_page(page);
3208         }
3209         return 0;
3210 }
3211
3212 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3213                               int read_only)
3214 {
3215         if (read_only)
3216                 return;
3217
3218         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3219                 printk(KERN_WARNING "warning: mount fs with errors, "
3220                        "running btrfsck is recommended\n");
3221 }
3222
3223 int btrfs_error_commit_super(struct btrfs_root *root)
3224 {
3225         int ret;
3226
3227         mutex_lock(&root->fs_info->cleaner_mutex);
3228         btrfs_run_delayed_iputs(root);
3229         mutex_unlock(&root->fs_info->cleaner_mutex);
3230
3231         down_write(&root->fs_info->cleanup_work_sem);
3232         up_write(&root->fs_info->cleanup_work_sem);
3233
3234         /* cleanup FS via transaction */
3235         btrfs_cleanup_transaction(root);
3236
3237         ret = write_ctree_super(NULL, root, 0);
3238
3239         return ret;
3240 }
3241
3242 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3243 {
3244         struct btrfs_inode *btrfs_inode;
3245         struct list_head splice;
3246
3247         INIT_LIST_HEAD(&splice);
3248
3249         mutex_lock(&root->fs_info->ordered_operations_mutex);
3250         spin_lock(&root->fs_info->ordered_extent_lock);
3251
3252         list_splice_init(&root->fs_info->ordered_operations, &splice);
3253         while (!list_empty(&splice)) {
3254                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3255                                          ordered_operations);
3256
3257                 list_del_init(&btrfs_inode->ordered_operations);
3258
3259                 btrfs_invalidate_inodes(btrfs_inode->root);
3260         }
3261
3262         spin_unlock(&root->fs_info->ordered_extent_lock);
3263         mutex_unlock(&root->fs_info->ordered_operations_mutex);
3264
3265         return 0;
3266 }
3267
3268 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3269 {
3270         struct list_head splice;
3271         struct btrfs_ordered_extent *ordered;
3272         struct inode *inode;
3273
3274         INIT_LIST_HEAD(&splice);
3275
3276         spin_lock(&root->fs_info->ordered_extent_lock);
3277
3278         list_splice_init(&root->fs_info->ordered_extents, &splice);
3279         while (!list_empty(&splice)) {
3280                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3281                                      root_extent_list);
3282
3283                 list_del_init(&ordered->root_extent_list);
3284                 atomic_inc(&ordered->refs);
3285
3286                 /* the inode may be getting freed (in sys_unlink path). */
3287                 inode = igrab(ordered->inode);
3288
3289                 spin_unlock(&root->fs_info->ordered_extent_lock);
3290                 if (inode)
3291                         iput(inode);
3292
3293                 atomic_set(&ordered->refs, 1);
3294                 btrfs_put_ordered_extent(ordered);
3295
3296                 spin_lock(&root->fs_info->ordered_extent_lock);
3297         }
3298
3299         spin_unlock(&root->fs_info->ordered_extent_lock);
3300
3301         return 0;
3302 }
3303
3304 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3305                                       struct btrfs_root *root)
3306 {
3307         struct rb_node *node;
3308         struct btrfs_delayed_ref_root *delayed_refs;
3309         struct btrfs_delayed_ref_node *ref;
3310         int ret = 0;
3311
3312         delayed_refs = &trans->delayed_refs;
3313
3314         spin_lock(&delayed_refs->lock);
3315         if (delayed_refs->num_entries == 0) {
3316                 spin_unlock(&delayed_refs->lock);
3317                 printk(KERN_INFO "delayed_refs has NO entry\n");
3318                 return ret;
3319         }
3320
3321         node = rb_first(&delayed_refs->root);
3322         while (node) {
3323                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3324                 node = rb_next(node);
3325
3326                 ref->in_tree = 0;
3327                 rb_erase(&ref->rb_node, &delayed_refs->root);
3328                 delayed_refs->num_entries--;
3329
3330                 atomic_set(&ref->refs, 1);
3331                 if (btrfs_delayed_ref_is_head(ref)) {
3332                         struct btrfs_delayed_ref_head *head;
3333
3334                         head = btrfs_delayed_node_to_head(ref);
3335                         mutex_lock(&head->mutex);
3336                         kfree(head->extent_op);
3337                         delayed_refs->num_heads--;
3338                         if (list_empty(&head->cluster))
3339                                 delayed_refs->num_heads_ready--;
3340                         list_del_init(&head->cluster);
3341                         mutex_unlock(&head->mutex);
3342                 }
3343
3344                 spin_unlock(&delayed_refs->lock);
3345                 btrfs_put_delayed_ref(ref);
3346
3347                 cond_resched();
3348                 spin_lock(&delayed_refs->lock);
3349         }
3350
3351         spin_unlock(&delayed_refs->lock);
3352
3353         return ret;
3354 }
3355
3356 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3357 {
3358         struct btrfs_pending_snapshot *snapshot;
3359         struct list_head splice;
3360
3361         INIT_LIST_HEAD(&splice);
3362
3363         list_splice_init(&t->pending_snapshots, &splice);
3364
3365         while (!list_empty(&splice)) {
3366                 snapshot = list_entry(splice.next,
3367                                       struct btrfs_pending_snapshot,
3368                                       list);
3369
3370                 list_del_init(&snapshot->list);
3371
3372                 kfree(snapshot);
3373         }
3374
3375         return 0;
3376 }
3377
3378 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3379 {
3380         struct btrfs_inode *btrfs_inode;
3381         struct list_head splice;
3382
3383         INIT_LIST_HEAD(&splice);
3384
3385         spin_lock(&root->fs_info->delalloc_lock);
3386         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3387
3388         while (!list_empty(&splice)) {
3389                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3390                                     delalloc_inodes);
3391
3392                 list_del_init(&btrfs_inode->delalloc_inodes);
3393
3394                 btrfs_invalidate_inodes(btrfs_inode->root);
3395         }
3396
3397         spin_unlock(&root->fs_info->delalloc_lock);
3398
3399         return 0;
3400 }
3401
3402 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3403                                         struct extent_io_tree *dirty_pages,
3404                                         int mark)
3405 {
3406         int ret;
3407         struct page *page;
3408         struct inode *btree_inode = root->fs_info->btree_inode;
3409         struct extent_buffer *eb;
3410         u64 start = 0;
3411         u64 end;
3412         u64 offset;
3413         unsigned long index;
3414
3415         while (1) {
3416                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3417                                             mark);
3418                 if (ret)
3419                         break;
3420
3421                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3422                 while (start <= end) {
3423                         index = start >> PAGE_CACHE_SHIFT;
3424                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3425                         page = find_get_page(btree_inode->i_mapping, index);
3426                         if (!page)
3427                                 continue;
3428                         offset = page_offset(page);
3429
3430                         spin_lock(&dirty_pages->buffer_lock);
3431                         eb = radix_tree_lookup(
3432                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3433                                                offset >> PAGE_CACHE_SHIFT);
3434                         spin_unlock(&dirty_pages->buffer_lock);
3435                         if (eb) {
3436                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3437                                                          &eb->bflags);
3438                                 atomic_set(&eb->refs, 1);
3439                         }
3440                         if (PageWriteback(page))
3441                                 end_page_writeback(page);
3442
3443                         lock_page(page);
3444                         if (PageDirty(page)) {
3445                                 clear_page_dirty_for_io(page);
3446                                 spin_lock_irq(&page->mapping->tree_lock);
3447                                 radix_tree_tag_clear(&page->mapping->page_tree,
3448                                                         page_index(page),
3449                                                         PAGECACHE_TAG_DIRTY);
3450                                 spin_unlock_irq(&page->mapping->tree_lock);
3451                         }
3452
3453                         page->mapping->a_ops->invalidatepage(page, 0);
3454                         unlock_page(page);
3455                 }
3456         }
3457
3458         return ret;
3459 }
3460
3461 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3462                                        struct extent_io_tree *pinned_extents)
3463 {
3464         struct extent_io_tree *unpin;
3465         u64 start;
3466         u64 end;
3467         int ret;
3468
3469         unpin = pinned_extents;
3470         while (1) {
3471                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3472                                             EXTENT_DIRTY);
3473                 if (ret)
3474                         break;
3475
3476                 /* opt_discard */
3477                 if (btrfs_test_opt(root, DISCARD))
3478                         ret = btrfs_error_discard_extent(root, start,
3479                                                          end + 1 - start,
3480                                                          NULL);
3481
3482                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3483                 btrfs_error_unpin_extent_range(root, start, end);
3484                 cond_resched();
3485         }
3486
3487         return 0;
3488 }
3489
3490 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3491 {
3492         struct btrfs_transaction *t;
3493         LIST_HEAD(list);
3494
3495         WARN_ON(1);
3496
3497         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3498
3499         spin_lock(&root->fs_info->trans_lock);
3500         list_splice_init(&root->fs_info->trans_list, &list);
3501         root->fs_info->trans_no_join = 1;
3502         spin_unlock(&root->fs_info->trans_lock);
3503
3504         while (!list_empty(&list)) {
3505                 t = list_entry(list.next, struct btrfs_transaction, list);
3506                 if (!t)
3507                         break;
3508
3509                 btrfs_destroy_ordered_operations(root);
3510
3511                 btrfs_destroy_ordered_extents(root);
3512
3513                 btrfs_destroy_delayed_refs(t, root);
3514
3515                 btrfs_block_rsv_release(root,
3516                                         &root->fs_info->trans_block_rsv,
3517                                         t->dirty_pages.dirty_bytes);
3518
3519                 /* FIXME: cleanup wait for commit */
3520                 t->in_commit = 1;
3521                 t->blocked = 1;
3522                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3523                         wake_up(&root->fs_info->transaction_blocked_wait);
3524
3525                 t->blocked = 0;
3526                 if (waitqueue_active(&root->fs_info->transaction_wait))
3527                         wake_up(&root->fs_info->transaction_wait);
3528
3529                 t->commit_done = 1;
3530                 if (waitqueue_active(&t->commit_wait))
3531                         wake_up(&t->commit_wait);
3532
3533                 btrfs_destroy_pending_snapshots(t);
3534
3535                 btrfs_destroy_delalloc_inodes(root);
3536
3537                 spin_lock(&root->fs_info->trans_lock);
3538                 root->fs_info->running_transaction = NULL;
3539                 spin_unlock(&root->fs_info->trans_lock);
3540
3541                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3542                                              EXTENT_DIRTY);
3543
3544                 btrfs_destroy_pinned_extent(root,
3545                                             root->fs_info->pinned_extents);
3546
3547                 atomic_set(&t->use_count, 0);
3548                 list_del_init(&t->list);
3549                 memset(t, 0, sizeof(*t));
3550                 kmem_cache_free(btrfs_transaction_cachep, t);
3551         }
3552
3553         spin_lock(&root->fs_info->trans_lock);
3554         root->fs_info->trans_no_join = 0;
3555         spin_unlock(&root->fs_info->trans_lock);
3556         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3557
3558         return 0;
3559 }
3560
3561 static struct extent_io_ops btree_extent_io_ops = {
3562         .write_cache_pages_lock_hook = btree_lock_page_hook,
3563         .readpage_end_io_hook = btree_readpage_end_io_hook,
3564         .readpage_io_failed_hook = btree_io_failed_hook,
3565         .submit_bio_hook = btree_submit_bio_hook,
3566         /* note we're sharing with inode.c for the merge bio hook */
3567         .merge_bio_hook = btrfs_merge_bio_hook,
3568 };