Linux 3.8
[~shefty/rdma-dev.git] / fs / btrfs / disk-io.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
34 #include "compat.h"
35 #include "ctree.h"
36 #include "disk-io.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "volumes.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
42 #include "locking.h"
43 #include "tree-log.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
49
50 #ifdef CONFIG_X86
51 #include <asm/cpufeature.h>
52 #endif
53
54 static struct extent_io_ops btree_extent_io_ops;
55 static void end_workqueue_fn(struct btrfs_work *work);
56 static void free_fs_root(struct btrfs_root *root);
57 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
58                                     int read_only);
59 static void btrfs_destroy_ordered_operations(struct btrfs_root *root);
60 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
61 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
62                                       struct btrfs_root *root);
63 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
64 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
65 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
66                                         struct extent_io_tree *dirty_pages,
67                                         int mark);
68 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
69                                        struct extent_io_tree *pinned_extents);
70
71 /*
72  * end_io_wq structs are used to do processing in task context when an IO is
73  * complete.  This is used during reads to verify checksums, and it is used
74  * by writes to insert metadata for new file extents after IO is complete.
75  */
76 struct end_io_wq {
77         struct bio *bio;
78         bio_end_io_t *end_io;
79         void *private;
80         struct btrfs_fs_info *info;
81         int error;
82         int metadata;
83         struct list_head list;
84         struct btrfs_work work;
85 };
86
87 /*
88  * async submit bios are used to offload expensive checksumming
89  * onto the worker threads.  They checksum file and metadata bios
90  * just before they are sent down the IO stack.
91  */
92 struct async_submit_bio {
93         struct inode *inode;
94         struct bio *bio;
95         struct list_head list;
96         extent_submit_bio_hook_t *submit_bio_start;
97         extent_submit_bio_hook_t *submit_bio_done;
98         int rw;
99         int mirror_num;
100         unsigned long bio_flags;
101         /*
102          * bio_offset is optional, can be used if the pages in the bio
103          * can't tell us where in the file the bio should go
104          */
105         u64 bio_offset;
106         struct btrfs_work work;
107         int error;
108 };
109
110 /*
111  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
112  * eb, the lockdep key is determined by the btrfs_root it belongs to and
113  * the level the eb occupies in the tree.
114  *
115  * Different roots are used for different purposes and may nest inside each
116  * other and they require separate keysets.  As lockdep keys should be
117  * static, assign keysets according to the purpose of the root as indicated
118  * by btrfs_root->objectid.  This ensures that all special purpose roots
119  * have separate keysets.
120  *
121  * Lock-nesting across peer nodes is always done with the immediate parent
122  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
123  * subclass to avoid triggering lockdep warning in such cases.
124  *
125  * The key is set by the readpage_end_io_hook after the buffer has passed
126  * csum validation but before the pages are unlocked.  It is also set by
127  * btrfs_init_new_buffer on freshly allocated blocks.
128  *
129  * We also add a check to make sure the highest level of the tree is the
130  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
131  * needs update as well.
132  */
133 #ifdef CONFIG_DEBUG_LOCK_ALLOC
134 # if BTRFS_MAX_LEVEL != 8
135 #  error
136 # endif
137
138 static struct btrfs_lockdep_keyset {
139         u64                     id;             /* root objectid */
140         const char              *name_stem;     /* lock name stem */
141         char                    names[BTRFS_MAX_LEVEL + 1][20];
142         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
143 } btrfs_lockdep_keysets[] = {
144         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
145         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
146         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
147         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
148         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
149         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
150         { .id = BTRFS_ORPHAN_OBJECTID,          .name_stem = "orphan"   },
151         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
152         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
153         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
154         { .id = 0,                              .name_stem = "tree"     },
155 };
156
157 void __init btrfs_init_lockdep(void)
158 {
159         int i, j;
160
161         /* initialize lockdep class names */
162         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
163                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
164
165                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
166                         snprintf(ks->names[j], sizeof(ks->names[j]),
167                                  "btrfs-%s-%02d", ks->name_stem, j);
168         }
169 }
170
171 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
172                                     int level)
173 {
174         struct btrfs_lockdep_keyset *ks;
175
176         BUG_ON(level >= ARRAY_SIZE(ks->keys));
177
178         /* find the matching keyset, id 0 is the default entry */
179         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
180                 if (ks->id == objectid)
181                         break;
182
183         lockdep_set_class_and_name(&eb->lock,
184                                    &ks->keys[level], ks->names[level]);
185 }
186
187 #endif
188
189 /*
190  * extents on the btree inode are pretty simple, there's one extent
191  * that covers the entire device
192  */
193 static struct extent_map *btree_get_extent(struct inode *inode,
194                 struct page *page, size_t pg_offset, u64 start, u64 len,
195                 int create)
196 {
197         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
198         struct extent_map *em;
199         int ret;
200
201         read_lock(&em_tree->lock);
202         em = lookup_extent_mapping(em_tree, start, len);
203         if (em) {
204                 em->bdev =
205                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
206                 read_unlock(&em_tree->lock);
207                 goto out;
208         }
209         read_unlock(&em_tree->lock);
210
211         em = alloc_extent_map();
212         if (!em) {
213                 em = ERR_PTR(-ENOMEM);
214                 goto out;
215         }
216         em->start = 0;
217         em->len = (u64)-1;
218         em->block_len = (u64)-1;
219         em->block_start = 0;
220         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
221
222         write_lock(&em_tree->lock);
223         ret = add_extent_mapping(em_tree, em);
224         if (ret == -EEXIST) {
225                 free_extent_map(em);
226                 em = lookup_extent_mapping(em_tree, start, len);
227                 if (!em)
228                         em = ERR_PTR(-EIO);
229         } else if (ret) {
230                 free_extent_map(em);
231                 em = ERR_PTR(ret);
232         }
233         write_unlock(&em_tree->lock);
234
235 out:
236         return em;
237 }
238
239 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
240 {
241         return crc32c(seed, data, len);
242 }
243
244 void btrfs_csum_final(u32 crc, char *result)
245 {
246         put_unaligned_le32(~crc, result);
247 }
248
249 /*
250  * compute the csum for a btree block, and either verify it or write it
251  * into the csum field of the block.
252  */
253 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
254                            int verify)
255 {
256         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
257         char *result = NULL;
258         unsigned long len;
259         unsigned long cur_len;
260         unsigned long offset = BTRFS_CSUM_SIZE;
261         char *kaddr;
262         unsigned long map_start;
263         unsigned long map_len;
264         int err;
265         u32 crc = ~(u32)0;
266         unsigned long inline_result;
267
268         len = buf->len - offset;
269         while (len > 0) {
270                 err = map_private_extent_buffer(buf, offset, 32,
271                                         &kaddr, &map_start, &map_len);
272                 if (err)
273                         return 1;
274                 cur_len = min(len, map_len - (offset - map_start));
275                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
276                                       crc, cur_len);
277                 len -= cur_len;
278                 offset += cur_len;
279         }
280         if (csum_size > sizeof(inline_result)) {
281                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
282                 if (!result)
283                         return 1;
284         } else {
285                 result = (char *)&inline_result;
286         }
287
288         btrfs_csum_final(crc, result);
289
290         if (verify) {
291                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
292                         u32 val;
293                         u32 found = 0;
294                         memcpy(&found, result, csum_size);
295
296                         read_extent_buffer(buf, &val, 0, csum_size);
297                         printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
298                                        "failed on %llu wanted %X found %X "
299                                        "level %d\n",
300                                        root->fs_info->sb->s_id,
301                                        (unsigned long long)buf->start, val, found,
302                                        btrfs_header_level(buf));
303                         if (result != (char *)&inline_result)
304                                 kfree(result);
305                         return 1;
306                 }
307         } else {
308                 write_extent_buffer(buf, result, 0, csum_size);
309         }
310         if (result != (char *)&inline_result)
311                 kfree(result);
312         return 0;
313 }
314
315 /*
316  * we can't consider a given block up to date unless the transid of the
317  * block matches the transid in the parent node's pointer.  This is how we
318  * detect blocks that either didn't get written at all or got written
319  * in the wrong place.
320  */
321 static int verify_parent_transid(struct extent_io_tree *io_tree,
322                                  struct extent_buffer *eb, u64 parent_transid,
323                                  int atomic)
324 {
325         struct extent_state *cached_state = NULL;
326         int ret;
327
328         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
329                 return 0;
330
331         if (atomic)
332                 return -EAGAIN;
333
334         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
335                          0, &cached_state);
336         if (extent_buffer_uptodate(eb) &&
337             btrfs_header_generation(eb) == parent_transid) {
338                 ret = 0;
339                 goto out;
340         }
341         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
342                        "found %llu\n",
343                        (unsigned long long)eb->start,
344                        (unsigned long long)parent_transid,
345                        (unsigned long long)btrfs_header_generation(eb));
346         ret = 1;
347         clear_extent_buffer_uptodate(eb);
348 out:
349         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
350                              &cached_state, GFP_NOFS);
351         return ret;
352 }
353
354 /*
355  * helper to read a given tree block, doing retries as required when
356  * the checksums don't match and we have alternate mirrors to try.
357  */
358 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
359                                           struct extent_buffer *eb,
360                                           u64 start, u64 parent_transid)
361 {
362         struct extent_io_tree *io_tree;
363         int failed = 0;
364         int ret;
365         int num_copies = 0;
366         int mirror_num = 0;
367         int failed_mirror = 0;
368
369         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
370         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
371         while (1) {
372                 ret = read_extent_buffer_pages(io_tree, eb, start,
373                                                WAIT_COMPLETE,
374                                                btree_get_extent, mirror_num);
375                 if (!ret) {
376                         if (!verify_parent_transid(io_tree, eb,
377                                                    parent_transid, 0))
378                                 break;
379                         else
380                                 ret = -EIO;
381                 }
382
383                 /*
384                  * This buffer's crc is fine, but its contents are corrupted, so
385                  * there is no reason to read the other copies, they won't be
386                  * any less wrong.
387                  */
388                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
389                         break;
390
391                 num_copies = btrfs_num_copies(root->fs_info,
392                                               eb->start, eb->len);
393                 if (num_copies == 1)
394                         break;
395
396                 if (!failed_mirror) {
397                         failed = 1;
398                         failed_mirror = eb->read_mirror;
399                 }
400
401                 mirror_num++;
402                 if (mirror_num == failed_mirror)
403                         mirror_num++;
404
405                 if (mirror_num > num_copies)
406                         break;
407         }
408
409         if (failed && !ret && failed_mirror)
410                 repair_eb_io_failure(root, eb, failed_mirror);
411
412         return ret;
413 }
414
415 /*
416  * checksum a dirty tree block before IO.  This has extra checks to make sure
417  * we only fill in the checksum field in the first page of a multi-page block
418  */
419
420 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
421 {
422         struct extent_io_tree *tree;
423         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
424         u64 found_start;
425         struct extent_buffer *eb;
426
427         tree = &BTRFS_I(page->mapping->host)->io_tree;
428
429         eb = (struct extent_buffer *)page->private;
430         if (page != eb->pages[0])
431                 return 0;
432         found_start = btrfs_header_bytenr(eb);
433         if (found_start != start) {
434                 WARN_ON(1);
435                 return 0;
436         }
437         if (!PageUptodate(page)) {
438                 WARN_ON(1);
439                 return 0;
440         }
441         csum_tree_block(root, eb, 0);
442         return 0;
443 }
444
445 static int check_tree_block_fsid(struct btrfs_root *root,
446                                  struct extent_buffer *eb)
447 {
448         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
449         u8 fsid[BTRFS_UUID_SIZE];
450         int ret = 1;
451
452         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
453                            BTRFS_FSID_SIZE);
454         while (fs_devices) {
455                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
456                         ret = 0;
457                         break;
458                 }
459                 fs_devices = fs_devices->seed;
460         }
461         return ret;
462 }
463
464 #define CORRUPT(reason, eb, root, slot)                         \
465         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
466                "root=%llu, slot=%d\n", reason,                  \
467                (unsigned long long)btrfs_header_bytenr(eb),     \
468                (unsigned long long)root->objectid, slot)
469
470 static noinline int check_leaf(struct btrfs_root *root,
471                                struct extent_buffer *leaf)
472 {
473         struct btrfs_key key;
474         struct btrfs_key leaf_key;
475         u32 nritems = btrfs_header_nritems(leaf);
476         int slot;
477
478         if (nritems == 0)
479                 return 0;
480
481         /* Check the 0 item */
482         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
483             BTRFS_LEAF_DATA_SIZE(root)) {
484                 CORRUPT("invalid item offset size pair", leaf, root, 0);
485                 return -EIO;
486         }
487
488         /*
489          * Check to make sure each items keys are in the correct order and their
490          * offsets make sense.  We only have to loop through nritems-1 because
491          * we check the current slot against the next slot, which verifies the
492          * next slot's offset+size makes sense and that the current's slot
493          * offset is correct.
494          */
495         for (slot = 0; slot < nritems - 1; slot++) {
496                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
497                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
498
499                 /* Make sure the keys are in the right order */
500                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
501                         CORRUPT("bad key order", leaf, root, slot);
502                         return -EIO;
503                 }
504
505                 /*
506                  * Make sure the offset and ends are right, remember that the
507                  * item data starts at the end of the leaf and grows towards the
508                  * front.
509                  */
510                 if (btrfs_item_offset_nr(leaf, slot) !=
511                         btrfs_item_end_nr(leaf, slot + 1)) {
512                         CORRUPT("slot offset bad", leaf, root, slot);
513                         return -EIO;
514                 }
515
516                 /*
517                  * Check to make sure that we don't point outside of the leaf,
518                  * just incase all the items are consistent to eachother, but
519                  * all point outside of the leaf.
520                  */
521                 if (btrfs_item_end_nr(leaf, slot) >
522                     BTRFS_LEAF_DATA_SIZE(root)) {
523                         CORRUPT("slot end outside of leaf", leaf, root, slot);
524                         return -EIO;
525                 }
526         }
527
528         return 0;
529 }
530
531 struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
532                                        struct page *page, int max_walk)
533 {
534         struct extent_buffer *eb;
535         u64 start = page_offset(page);
536         u64 target = start;
537         u64 min_start;
538
539         if (start < max_walk)
540                 min_start = 0;
541         else
542                 min_start = start - max_walk;
543
544         while (start >= min_start) {
545                 eb = find_extent_buffer(tree, start, 0);
546                 if (eb) {
547                         /*
548                          * we found an extent buffer and it contains our page
549                          * horray!
550                          */
551                         if (eb->start <= target &&
552                             eb->start + eb->len > target)
553                                 return eb;
554
555                         /* we found an extent buffer that wasn't for us */
556                         free_extent_buffer(eb);
557                         return NULL;
558                 }
559                 if (start == 0)
560                         break;
561                 start -= PAGE_CACHE_SIZE;
562         }
563         return NULL;
564 }
565
566 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
567                                struct extent_state *state, int mirror)
568 {
569         struct extent_io_tree *tree;
570         u64 found_start;
571         int found_level;
572         struct extent_buffer *eb;
573         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
574         int ret = 0;
575         int reads_done;
576
577         if (!page->private)
578                 goto out;
579
580         tree = &BTRFS_I(page->mapping->host)->io_tree;
581         eb = (struct extent_buffer *)page->private;
582
583         /* the pending IO might have been the only thing that kept this buffer
584          * in memory.  Make sure we have a ref for all this other checks
585          */
586         extent_buffer_get(eb);
587
588         reads_done = atomic_dec_and_test(&eb->io_pages);
589         if (!reads_done)
590                 goto err;
591
592         eb->read_mirror = mirror;
593         if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
594                 ret = -EIO;
595                 goto err;
596         }
597
598         found_start = btrfs_header_bytenr(eb);
599         if (found_start != eb->start) {
600                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
601                                "%llu %llu\n",
602                                (unsigned long long)found_start,
603                                (unsigned long long)eb->start);
604                 ret = -EIO;
605                 goto err;
606         }
607         if (check_tree_block_fsid(root, eb)) {
608                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
609                                (unsigned long long)eb->start);
610                 ret = -EIO;
611                 goto err;
612         }
613         found_level = btrfs_header_level(eb);
614
615         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
616                                        eb, found_level);
617
618         ret = csum_tree_block(root, eb, 1);
619         if (ret) {
620                 ret = -EIO;
621                 goto err;
622         }
623
624         /*
625          * If this is a leaf block and it is corrupt, set the corrupt bit so
626          * that we don't try and read the other copies of this block, just
627          * return -EIO.
628          */
629         if (found_level == 0 && check_leaf(root, eb)) {
630                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
631                 ret = -EIO;
632         }
633
634         if (!ret)
635                 set_extent_buffer_uptodate(eb);
636 err:
637         if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
638                 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
639                 btree_readahead_hook(root, eb, eb->start, ret);
640         }
641
642         if (ret)
643                 clear_extent_buffer_uptodate(eb);
644         free_extent_buffer(eb);
645 out:
646         return ret;
647 }
648
649 static int btree_io_failed_hook(struct page *page, int failed_mirror)
650 {
651         struct extent_buffer *eb;
652         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
653
654         eb = (struct extent_buffer *)page->private;
655         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
656         eb->read_mirror = failed_mirror;
657         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
658                 btree_readahead_hook(root, eb, eb->start, -EIO);
659         return -EIO;    /* we fixed nothing */
660 }
661
662 static void end_workqueue_bio(struct bio *bio, int err)
663 {
664         struct end_io_wq *end_io_wq = bio->bi_private;
665         struct btrfs_fs_info *fs_info;
666
667         fs_info = end_io_wq->info;
668         end_io_wq->error = err;
669         end_io_wq->work.func = end_workqueue_fn;
670         end_io_wq->work.flags = 0;
671
672         if (bio->bi_rw & REQ_WRITE) {
673                 if (end_io_wq->metadata == 1)
674                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
675                                            &end_io_wq->work);
676                 else if (end_io_wq->metadata == 2)
677                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
678                                            &end_io_wq->work);
679                 else
680                         btrfs_queue_worker(&fs_info->endio_write_workers,
681                                            &end_io_wq->work);
682         } else {
683                 if (end_io_wq->metadata)
684                         btrfs_queue_worker(&fs_info->endio_meta_workers,
685                                            &end_io_wq->work);
686                 else
687                         btrfs_queue_worker(&fs_info->endio_workers,
688                                            &end_io_wq->work);
689         }
690 }
691
692 /*
693  * For the metadata arg you want
694  *
695  * 0 - if data
696  * 1 - if normal metadta
697  * 2 - if writing to the free space cache area
698  */
699 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
700                         int metadata)
701 {
702         struct end_io_wq *end_io_wq;
703         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
704         if (!end_io_wq)
705                 return -ENOMEM;
706
707         end_io_wq->private = bio->bi_private;
708         end_io_wq->end_io = bio->bi_end_io;
709         end_io_wq->info = info;
710         end_io_wq->error = 0;
711         end_io_wq->bio = bio;
712         end_io_wq->metadata = metadata;
713
714         bio->bi_private = end_io_wq;
715         bio->bi_end_io = end_workqueue_bio;
716         return 0;
717 }
718
719 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
720 {
721         unsigned long limit = min_t(unsigned long,
722                                     info->workers.max_workers,
723                                     info->fs_devices->open_devices);
724         return 256 * limit;
725 }
726
727 static void run_one_async_start(struct btrfs_work *work)
728 {
729         struct async_submit_bio *async;
730         int ret;
731
732         async = container_of(work, struct  async_submit_bio, work);
733         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
734                                       async->mirror_num, async->bio_flags,
735                                       async->bio_offset);
736         if (ret)
737                 async->error = ret;
738 }
739
740 static void run_one_async_done(struct btrfs_work *work)
741 {
742         struct btrfs_fs_info *fs_info;
743         struct async_submit_bio *async;
744         int limit;
745
746         async = container_of(work, struct  async_submit_bio, work);
747         fs_info = BTRFS_I(async->inode)->root->fs_info;
748
749         limit = btrfs_async_submit_limit(fs_info);
750         limit = limit * 2 / 3;
751
752         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
753             waitqueue_active(&fs_info->async_submit_wait))
754                 wake_up(&fs_info->async_submit_wait);
755
756         /* If an error occured we just want to clean up the bio and move on */
757         if (async->error) {
758                 bio_endio(async->bio, async->error);
759                 return;
760         }
761
762         async->submit_bio_done(async->inode, async->rw, async->bio,
763                                async->mirror_num, async->bio_flags,
764                                async->bio_offset);
765 }
766
767 static void run_one_async_free(struct btrfs_work *work)
768 {
769         struct async_submit_bio *async;
770
771         async = container_of(work, struct  async_submit_bio, work);
772         kfree(async);
773 }
774
775 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
776                         int rw, struct bio *bio, int mirror_num,
777                         unsigned long bio_flags,
778                         u64 bio_offset,
779                         extent_submit_bio_hook_t *submit_bio_start,
780                         extent_submit_bio_hook_t *submit_bio_done)
781 {
782         struct async_submit_bio *async;
783
784         async = kmalloc(sizeof(*async), GFP_NOFS);
785         if (!async)
786                 return -ENOMEM;
787
788         async->inode = inode;
789         async->rw = rw;
790         async->bio = bio;
791         async->mirror_num = mirror_num;
792         async->submit_bio_start = submit_bio_start;
793         async->submit_bio_done = submit_bio_done;
794
795         async->work.func = run_one_async_start;
796         async->work.ordered_func = run_one_async_done;
797         async->work.ordered_free = run_one_async_free;
798
799         async->work.flags = 0;
800         async->bio_flags = bio_flags;
801         async->bio_offset = bio_offset;
802
803         async->error = 0;
804
805         atomic_inc(&fs_info->nr_async_submits);
806
807         if (rw & REQ_SYNC)
808                 btrfs_set_work_high_prio(&async->work);
809
810         btrfs_queue_worker(&fs_info->workers, &async->work);
811
812         while (atomic_read(&fs_info->async_submit_draining) &&
813               atomic_read(&fs_info->nr_async_submits)) {
814                 wait_event(fs_info->async_submit_wait,
815                            (atomic_read(&fs_info->nr_async_submits) == 0));
816         }
817
818         return 0;
819 }
820
821 static int btree_csum_one_bio(struct bio *bio)
822 {
823         struct bio_vec *bvec = bio->bi_io_vec;
824         int bio_index = 0;
825         struct btrfs_root *root;
826         int ret = 0;
827
828         WARN_ON(bio->bi_vcnt <= 0);
829         while (bio_index < bio->bi_vcnt) {
830                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
831                 ret = csum_dirty_buffer(root, bvec->bv_page);
832                 if (ret)
833                         break;
834                 bio_index++;
835                 bvec++;
836         }
837         return ret;
838 }
839
840 static int __btree_submit_bio_start(struct inode *inode, int rw,
841                                     struct bio *bio, int mirror_num,
842                                     unsigned long bio_flags,
843                                     u64 bio_offset)
844 {
845         /*
846          * when we're called for a write, we're already in the async
847          * submission context.  Just jump into btrfs_map_bio
848          */
849         return btree_csum_one_bio(bio);
850 }
851
852 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
853                                  int mirror_num, unsigned long bio_flags,
854                                  u64 bio_offset)
855 {
856         int ret;
857
858         /*
859          * when we're called for a write, we're already in the async
860          * submission context.  Just jump into btrfs_map_bio
861          */
862         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
863         if (ret)
864                 bio_endio(bio, ret);
865         return ret;
866 }
867
868 static int check_async_write(struct inode *inode, unsigned long bio_flags)
869 {
870         if (bio_flags & EXTENT_BIO_TREE_LOG)
871                 return 0;
872 #ifdef CONFIG_X86
873         if (cpu_has_xmm4_2)
874                 return 0;
875 #endif
876         return 1;
877 }
878
879 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
880                                  int mirror_num, unsigned long bio_flags,
881                                  u64 bio_offset)
882 {
883         int async = check_async_write(inode, bio_flags);
884         int ret;
885
886         if (!(rw & REQ_WRITE)) {
887                 /*
888                  * called for a read, do the setup so that checksum validation
889                  * can happen in the async kernel threads
890                  */
891                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
892                                           bio, 1);
893                 if (ret)
894                         goto out_w_error;
895                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
896                                     mirror_num, 0);
897         } else if (!async) {
898                 ret = btree_csum_one_bio(bio);
899                 if (ret)
900                         goto out_w_error;
901                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
902                                     mirror_num, 0);
903         } else {
904                 /*
905                  * kthread helpers are used to submit writes so that
906                  * checksumming can happen in parallel across all CPUs
907                  */
908                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
909                                           inode, rw, bio, mirror_num, 0,
910                                           bio_offset,
911                                           __btree_submit_bio_start,
912                                           __btree_submit_bio_done);
913         }
914
915         if (ret) {
916 out_w_error:
917                 bio_endio(bio, ret);
918         }
919         return ret;
920 }
921
922 #ifdef CONFIG_MIGRATION
923 static int btree_migratepage(struct address_space *mapping,
924                         struct page *newpage, struct page *page,
925                         enum migrate_mode mode)
926 {
927         /*
928          * we can't safely write a btree page from here,
929          * we haven't done the locking hook
930          */
931         if (PageDirty(page))
932                 return -EAGAIN;
933         /*
934          * Buffers may be managed in a filesystem specific way.
935          * We must have no buffers or drop them.
936          */
937         if (page_has_private(page) &&
938             !try_to_release_page(page, GFP_KERNEL))
939                 return -EAGAIN;
940         return migrate_page(mapping, newpage, page, mode);
941 }
942 #endif
943
944
945 static int btree_writepages(struct address_space *mapping,
946                             struct writeback_control *wbc)
947 {
948         struct extent_io_tree *tree;
949         tree = &BTRFS_I(mapping->host)->io_tree;
950         if (wbc->sync_mode == WB_SYNC_NONE) {
951                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
952                 u64 num_dirty;
953                 unsigned long thresh = 32 * 1024 * 1024;
954
955                 if (wbc->for_kupdate)
956                         return 0;
957
958                 /* this is a bit racy, but that's ok */
959                 num_dirty = root->fs_info->dirty_metadata_bytes;
960                 if (num_dirty < thresh)
961                         return 0;
962         }
963         return btree_write_cache_pages(mapping, wbc);
964 }
965
966 static int btree_readpage(struct file *file, struct page *page)
967 {
968         struct extent_io_tree *tree;
969         tree = &BTRFS_I(page->mapping->host)->io_tree;
970         return extent_read_full_page(tree, page, btree_get_extent, 0);
971 }
972
973 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
974 {
975         if (PageWriteback(page) || PageDirty(page))
976                 return 0;
977         /*
978          * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
979          * slab allocation from alloc_extent_state down the callchain where
980          * it'd hit a BUG_ON as those flags are not allowed.
981          */
982         gfp_flags &= ~GFP_SLAB_BUG_MASK;
983
984         return try_release_extent_buffer(page, gfp_flags);
985 }
986
987 static void btree_invalidatepage(struct page *page, unsigned long offset)
988 {
989         struct extent_io_tree *tree;
990         tree = &BTRFS_I(page->mapping->host)->io_tree;
991         extent_invalidatepage(tree, page, offset);
992         btree_releasepage(page, GFP_NOFS);
993         if (PagePrivate(page)) {
994                 printk(KERN_WARNING "btrfs warning page private not zero "
995                        "on page %llu\n", (unsigned long long)page_offset(page));
996                 ClearPagePrivate(page);
997                 set_page_private(page, 0);
998                 page_cache_release(page);
999         }
1000 }
1001
1002 static int btree_set_page_dirty(struct page *page)
1003 {
1004 #ifdef DEBUG
1005         struct extent_buffer *eb;
1006
1007         BUG_ON(!PagePrivate(page));
1008         eb = (struct extent_buffer *)page->private;
1009         BUG_ON(!eb);
1010         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1011         BUG_ON(!atomic_read(&eb->refs));
1012         btrfs_assert_tree_locked(eb);
1013 #endif
1014         return __set_page_dirty_nobuffers(page);
1015 }
1016
1017 static const struct address_space_operations btree_aops = {
1018         .readpage       = btree_readpage,
1019         .writepages     = btree_writepages,
1020         .releasepage    = btree_releasepage,
1021         .invalidatepage = btree_invalidatepage,
1022 #ifdef CONFIG_MIGRATION
1023         .migratepage    = btree_migratepage,
1024 #endif
1025         .set_page_dirty = btree_set_page_dirty,
1026 };
1027
1028 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1029                          u64 parent_transid)
1030 {
1031         struct extent_buffer *buf = NULL;
1032         struct inode *btree_inode = root->fs_info->btree_inode;
1033         int ret = 0;
1034
1035         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1036         if (!buf)
1037                 return 0;
1038         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1039                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1040         free_extent_buffer(buf);
1041         return ret;
1042 }
1043
1044 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1045                          int mirror_num, struct extent_buffer **eb)
1046 {
1047         struct extent_buffer *buf = NULL;
1048         struct inode *btree_inode = root->fs_info->btree_inode;
1049         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1050         int ret;
1051
1052         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1053         if (!buf)
1054                 return 0;
1055
1056         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1057
1058         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1059                                        btree_get_extent, mirror_num);
1060         if (ret) {
1061                 free_extent_buffer(buf);
1062                 return ret;
1063         }
1064
1065         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1066                 free_extent_buffer(buf);
1067                 return -EIO;
1068         } else if (extent_buffer_uptodate(buf)) {
1069                 *eb = buf;
1070         } else {
1071                 free_extent_buffer(buf);
1072         }
1073         return 0;
1074 }
1075
1076 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1077                                             u64 bytenr, u32 blocksize)
1078 {
1079         struct inode *btree_inode = root->fs_info->btree_inode;
1080         struct extent_buffer *eb;
1081         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1082                                 bytenr, blocksize);
1083         return eb;
1084 }
1085
1086 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1087                                                  u64 bytenr, u32 blocksize)
1088 {
1089         struct inode *btree_inode = root->fs_info->btree_inode;
1090         struct extent_buffer *eb;
1091
1092         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1093                                  bytenr, blocksize);
1094         return eb;
1095 }
1096
1097
1098 int btrfs_write_tree_block(struct extent_buffer *buf)
1099 {
1100         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1101                                         buf->start + buf->len - 1);
1102 }
1103
1104 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1105 {
1106         return filemap_fdatawait_range(buf->pages[0]->mapping,
1107                                        buf->start, buf->start + buf->len - 1);
1108 }
1109
1110 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1111                                       u32 blocksize, u64 parent_transid)
1112 {
1113         struct extent_buffer *buf = NULL;
1114         int ret;
1115
1116         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1117         if (!buf)
1118                 return NULL;
1119
1120         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1121         return buf;
1122
1123 }
1124
1125 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1126                       struct extent_buffer *buf)
1127 {
1128         if (btrfs_header_generation(buf) ==
1129             root->fs_info->running_transaction->transid) {
1130                 btrfs_assert_tree_locked(buf);
1131
1132                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1133                         spin_lock(&root->fs_info->delalloc_lock);
1134                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1135                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1136                         else {
1137                                 spin_unlock(&root->fs_info->delalloc_lock);
1138                                 btrfs_panic(root->fs_info, -EOVERFLOW,
1139                                           "Can't clear %lu bytes from "
1140                                           " dirty_mdatadata_bytes (%llu)",
1141                                           buf->len,
1142                                           root->fs_info->dirty_metadata_bytes);
1143                         }
1144                         spin_unlock(&root->fs_info->delalloc_lock);
1145
1146                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1147                         btrfs_set_lock_blocking(buf);
1148                         clear_extent_buffer_dirty(buf);
1149                 }
1150         }
1151 }
1152
1153 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1154                          u32 stripesize, struct btrfs_root *root,
1155                          struct btrfs_fs_info *fs_info,
1156                          u64 objectid)
1157 {
1158         root->node = NULL;
1159         root->commit_root = NULL;
1160         root->sectorsize = sectorsize;
1161         root->nodesize = nodesize;
1162         root->leafsize = leafsize;
1163         root->stripesize = stripesize;
1164         root->ref_cows = 0;
1165         root->track_dirty = 0;
1166         root->in_radix = 0;
1167         root->orphan_item_inserted = 0;
1168         root->orphan_cleanup_state = 0;
1169
1170         root->objectid = objectid;
1171         root->last_trans = 0;
1172         root->highest_objectid = 0;
1173         root->name = NULL;
1174         root->inode_tree = RB_ROOT;
1175         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1176         root->block_rsv = NULL;
1177         root->orphan_block_rsv = NULL;
1178
1179         INIT_LIST_HEAD(&root->dirty_list);
1180         INIT_LIST_HEAD(&root->root_list);
1181         spin_lock_init(&root->orphan_lock);
1182         spin_lock_init(&root->inode_lock);
1183         spin_lock_init(&root->accounting_lock);
1184         mutex_init(&root->objectid_mutex);
1185         mutex_init(&root->log_mutex);
1186         init_waitqueue_head(&root->log_writer_wait);
1187         init_waitqueue_head(&root->log_commit_wait[0]);
1188         init_waitqueue_head(&root->log_commit_wait[1]);
1189         atomic_set(&root->log_commit[0], 0);
1190         atomic_set(&root->log_commit[1], 0);
1191         atomic_set(&root->log_writers, 0);
1192         atomic_set(&root->log_batch, 0);
1193         atomic_set(&root->orphan_inodes, 0);
1194         root->log_transid = 0;
1195         root->last_log_commit = 0;
1196         extent_io_tree_init(&root->dirty_log_pages,
1197                              fs_info->btree_inode->i_mapping);
1198
1199         memset(&root->root_key, 0, sizeof(root->root_key));
1200         memset(&root->root_item, 0, sizeof(root->root_item));
1201         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1202         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1203         root->defrag_trans_start = fs_info->generation;
1204         init_completion(&root->kobj_unregister);
1205         root->defrag_running = 0;
1206         root->root_key.objectid = objectid;
1207         root->anon_dev = 0;
1208
1209         spin_lock_init(&root->root_item_lock);
1210 }
1211
1212 static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1213                                             struct btrfs_fs_info *fs_info,
1214                                             u64 objectid,
1215                                             struct btrfs_root *root)
1216 {
1217         int ret;
1218         u32 blocksize;
1219         u64 generation;
1220
1221         __setup_root(tree_root->nodesize, tree_root->leafsize,
1222                      tree_root->sectorsize, tree_root->stripesize,
1223                      root, fs_info, objectid);
1224         ret = btrfs_find_last_root(tree_root, objectid,
1225                                    &root->root_item, &root->root_key);
1226         if (ret > 0)
1227                 return -ENOENT;
1228         else if (ret < 0)
1229                 return ret;
1230
1231         generation = btrfs_root_generation(&root->root_item);
1232         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1233         root->commit_root = NULL;
1234         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1235                                      blocksize, generation);
1236         if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1237                 free_extent_buffer(root->node);
1238                 root->node = NULL;
1239                 return -EIO;
1240         }
1241         root->commit_root = btrfs_root_node(root);
1242         return 0;
1243 }
1244
1245 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1246 {
1247         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1248         if (root)
1249                 root->fs_info = fs_info;
1250         return root;
1251 }
1252
1253 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1254                                      struct btrfs_fs_info *fs_info,
1255                                      u64 objectid)
1256 {
1257         struct extent_buffer *leaf;
1258         struct btrfs_root *tree_root = fs_info->tree_root;
1259         struct btrfs_root *root;
1260         struct btrfs_key key;
1261         int ret = 0;
1262         u64 bytenr;
1263
1264         root = btrfs_alloc_root(fs_info);
1265         if (!root)
1266                 return ERR_PTR(-ENOMEM);
1267
1268         __setup_root(tree_root->nodesize, tree_root->leafsize,
1269                      tree_root->sectorsize, tree_root->stripesize,
1270                      root, fs_info, objectid);
1271         root->root_key.objectid = objectid;
1272         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1273         root->root_key.offset = 0;
1274
1275         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1276                                       0, objectid, NULL, 0, 0, 0);
1277         if (IS_ERR(leaf)) {
1278                 ret = PTR_ERR(leaf);
1279                 goto fail;
1280         }
1281
1282         bytenr = leaf->start;
1283         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1284         btrfs_set_header_bytenr(leaf, leaf->start);
1285         btrfs_set_header_generation(leaf, trans->transid);
1286         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1287         btrfs_set_header_owner(leaf, objectid);
1288         root->node = leaf;
1289
1290         write_extent_buffer(leaf, fs_info->fsid,
1291                             (unsigned long)btrfs_header_fsid(leaf),
1292                             BTRFS_FSID_SIZE);
1293         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1294                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1295                             BTRFS_UUID_SIZE);
1296         btrfs_mark_buffer_dirty(leaf);
1297
1298         root->commit_root = btrfs_root_node(root);
1299         root->track_dirty = 1;
1300
1301
1302         root->root_item.flags = 0;
1303         root->root_item.byte_limit = 0;
1304         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1305         btrfs_set_root_generation(&root->root_item, trans->transid);
1306         btrfs_set_root_level(&root->root_item, 0);
1307         btrfs_set_root_refs(&root->root_item, 1);
1308         btrfs_set_root_used(&root->root_item, leaf->len);
1309         btrfs_set_root_last_snapshot(&root->root_item, 0);
1310         btrfs_set_root_dirid(&root->root_item, 0);
1311         root->root_item.drop_level = 0;
1312
1313         key.objectid = objectid;
1314         key.type = BTRFS_ROOT_ITEM_KEY;
1315         key.offset = 0;
1316         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1317         if (ret)
1318                 goto fail;
1319
1320         btrfs_tree_unlock(leaf);
1321
1322 fail:
1323         if (ret)
1324                 return ERR_PTR(ret);
1325
1326         return root;
1327 }
1328
1329 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1330                                          struct btrfs_fs_info *fs_info)
1331 {
1332         struct btrfs_root *root;
1333         struct btrfs_root *tree_root = fs_info->tree_root;
1334         struct extent_buffer *leaf;
1335
1336         root = btrfs_alloc_root(fs_info);
1337         if (!root)
1338                 return ERR_PTR(-ENOMEM);
1339
1340         __setup_root(tree_root->nodesize, tree_root->leafsize,
1341                      tree_root->sectorsize, tree_root->stripesize,
1342                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1343
1344         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1345         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1346         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1347         /*
1348          * log trees do not get reference counted because they go away
1349          * before a real commit is actually done.  They do store pointers
1350          * to file data extents, and those reference counts still get
1351          * updated (along with back refs to the log tree).
1352          */
1353         root->ref_cows = 0;
1354
1355         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1356                                       BTRFS_TREE_LOG_OBJECTID, NULL,
1357                                       0, 0, 0);
1358         if (IS_ERR(leaf)) {
1359                 kfree(root);
1360                 return ERR_CAST(leaf);
1361         }
1362
1363         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1364         btrfs_set_header_bytenr(leaf, leaf->start);
1365         btrfs_set_header_generation(leaf, trans->transid);
1366         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1367         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1368         root->node = leaf;
1369
1370         write_extent_buffer(root->node, root->fs_info->fsid,
1371                             (unsigned long)btrfs_header_fsid(root->node),
1372                             BTRFS_FSID_SIZE);
1373         btrfs_mark_buffer_dirty(root->node);
1374         btrfs_tree_unlock(root->node);
1375         return root;
1376 }
1377
1378 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1379                              struct btrfs_fs_info *fs_info)
1380 {
1381         struct btrfs_root *log_root;
1382
1383         log_root = alloc_log_tree(trans, fs_info);
1384         if (IS_ERR(log_root))
1385                 return PTR_ERR(log_root);
1386         WARN_ON(fs_info->log_root_tree);
1387         fs_info->log_root_tree = log_root;
1388         return 0;
1389 }
1390
1391 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1392                        struct btrfs_root *root)
1393 {
1394         struct btrfs_root *log_root;
1395         struct btrfs_inode_item *inode_item;
1396
1397         log_root = alloc_log_tree(trans, root->fs_info);
1398         if (IS_ERR(log_root))
1399                 return PTR_ERR(log_root);
1400
1401         log_root->last_trans = trans->transid;
1402         log_root->root_key.offset = root->root_key.objectid;
1403
1404         inode_item = &log_root->root_item.inode;
1405         inode_item->generation = cpu_to_le64(1);
1406         inode_item->size = cpu_to_le64(3);
1407         inode_item->nlink = cpu_to_le32(1);
1408         inode_item->nbytes = cpu_to_le64(root->leafsize);
1409         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1410
1411         btrfs_set_root_node(&log_root->root_item, log_root->node);
1412
1413         WARN_ON(root->log_root);
1414         root->log_root = log_root;
1415         root->log_transid = 0;
1416         root->last_log_commit = 0;
1417         return 0;
1418 }
1419
1420 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1421                                                struct btrfs_key *location)
1422 {
1423         struct btrfs_root *root;
1424         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1425         struct btrfs_path *path;
1426         struct extent_buffer *l;
1427         u64 generation;
1428         u32 blocksize;
1429         int ret = 0;
1430         int slot;
1431
1432         root = btrfs_alloc_root(fs_info);
1433         if (!root)
1434                 return ERR_PTR(-ENOMEM);
1435         if (location->offset == (u64)-1) {
1436                 ret = find_and_setup_root(tree_root, fs_info,
1437                                           location->objectid, root);
1438                 if (ret) {
1439                         kfree(root);
1440                         return ERR_PTR(ret);
1441                 }
1442                 goto out;
1443         }
1444
1445         __setup_root(tree_root->nodesize, tree_root->leafsize,
1446                      tree_root->sectorsize, tree_root->stripesize,
1447                      root, fs_info, location->objectid);
1448
1449         path = btrfs_alloc_path();
1450         if (!path) {
1451                 kfree(root);
1452                 return ERR_PTR(-ENOMEM);
1453         }
1454         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1455         if (ret == 0) {
1456                 l = path->nodes[0];
1457                 slot = path->slots[0];
1458                 btrfs_read_root_item(tree_root, l, slot, &root->root_item);
1459                 memcpy(&root->root_key, location, sizeof(*location));
1460         }
1461         btrfs_free_path(path);
1462         if (ret) {
1463                 kfree(root);
1464                 if (ret > 0)
1465                         ret = -ENOENT;
1466                 return ERR_PTR(ret);
1467         }
1468
1469         generation = btrfs_root_generation(&root->root_item);
1470         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1471         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1472                                      blocksize, generation);
1473         root->commit_root = btrfs_root_node(root);
1474         BUG_ON(!root->node); /* -ENOMEM */
1475 out:
1476         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1477                 root->ref_cows = 1;
1478                 btrfs_check_and_init_root_item(&root->root_item);
1479         }
1480
1481         return root;
1482 }
1483
1484 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1485                                               struct btrfs_key *location)
1486 {
1487         struct btrfs_root *root;
1488         int ret;
1489
1490         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1491                 return fs_info->tree_root;
1492         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1493                 return fs_info->extent_root;
1494         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1495                 return fs_info->chunk_root;
1496         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1497                 return fs_info->dev_root;
1498         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1499                 return fs_info->csum_root;
1500         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1501                 return fs_info->quota_root ? fs_info->quota_root :
1502                                              ERR_PTR(-ENOENT);
1503 again:
1504         spin_lock(&fs_info->fs_roots_radix_lock);
1505         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1506                                  (unsigned long)location->objectid);
1507         spin_unlock(&fs_info->fs_roots_radix_lock);
1508         if (root)
1509                 return root;
1510
1511         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1512         if (IS_ERR(root))
1513                 return root;
1514
1515         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1516         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1517                                         GFP_NOFS);
1518         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1519                 ret = -ENOMEM;
1520                 goto fail;
1521         }
1522
1523         btrfs_init_free_ino_ctl(root);
1524         mutex_init(&root->fs_commit_mutex);
1525         spin_lock_init(&root->cache_lock);
1526         init_waitqueue_head(&root->cache_wait);
1527
1528         ret = get_anon_bdev(&root->anon_dev);
1529         if (ret)
1530                 goto fail;
1531
1532         if (btrfs_root_refs(&root->root_item) == 0) {
1533                 ret = -ENOENT;
1534                 goto fail;
1535         }
1536
1537         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1538         if (ret < 0)
1539                 goto fail;
1540         if (ret == 0)
1541                 root->orphan_item_inserted = 1;
1542
1543         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1544         if (ret)
1545                 goto fail;
1546
1547         spin_lock(&fs_info->fs_roots_radix_lock);
1548         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1549                                 (unsigned long)root->root_key.objectid,
1550                                 root);
1551         if (ret == 0)
1552                 root->in_radix = 1;
1553
1554         spin_unlock(&fs_info->fs_roots_radix_lock);
1555         radix_tree_preload_end();
1556         if (ret) {
1557                 if (ret == -EEXIST) {
1558                         free_fs_root(root);
1559                         goto again;
1560                 }
1561                 goto fail;
1562         }
1563
1564         ret = btrfs_find_dead_roots(fs_info->tree_root,
1565                                     root->root_key.objectid);
1566         WARN_ON(ret);
1567         return root;
1568 fail:
1569         free_fs_root(root);
1570         return ERR_PTR(ret);
1571 }
1572
1573 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1574 {
1575         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1576         int ret = 0;
1577         struct btrfs_device *device;
1578         struct backing_dev_info *bdi;
1579
1580         rcu_read_lock();
1581         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1582                 if (!device->bdev)
1583                         continue;
1584                 bdi = blk_get_backing_dev_info(device->bdev);
1585                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1586                         ret = 1;
1587                         break;
1588                 }
1589         }
1590         rcu_read_unlock();
1591         return ret;
1592 }
1593
1594 /*
1595  * If this fails, caller must call bdi_destroy() to get rid of the
1596  * bdi again.
1597  */
1598 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1599 {
1600         int err;
1601
1602         bdi->capabilities = BDI_CAP_MAP_COPY;
1603         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1604         if (err)
1605                 return err;
1606
1607         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1608         bdi->congested_fn       = btrfs_congested_fn;
1609         bdi->congested_data     = info;
1610         return 0;
1611 }
1612
1613 /*
1614  * called by the kthread helper functions to finally call the bio end_io
1615  * functions.  This is where read checksum verification actually happens
1616  */
1617 static void end_workqueue_fn(struct btrfs_work *work)
1618 {
1619         struct bio *bio;
1620         struct end_io_wq *end_io_wq;
1621         struct btrfs_fs_info *fs_info;
1622         int error;
1623
1624         end_io_wq = container_of(work, struct end_io_wq, work);
1625         bio = end_io_wq->bio;
1626         fs_info = end_io_wq->info;
1627
1628         error = end_io_wq->error;
1629         bio->bi_private = end_io_wq->private;
1630         bio->bi_end_io = end_io_wq->end_io;
1631         kfree(end_io_wq);
1632         bio_endio(bio, error);
1633 }
1634
1635 static int cleaner_kthread(void *arg)
1636 {
1637         struct btrfs_root *root = arg;
1638
1639         do {
1640                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1641                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1642                         btrfs_run_delayed_iputs(root);
1643                         btrfs_clean_old_snapshots(root);
1644                         mutex_unlock(&root->fs_info->cleaner_mutex);
1645                         btrfs_run_defrag_inodes(root->fs_info);
1646                 }
1647
1648                 if (!try_to_freeze()) {
1649                         set_current_state(TASK_INTERRUPTIBLE);
1650                         if (!kthread_should_stop())
1651                                 schedule();
1652                         __set_current_state(TASK_RUNNING);
1653                 }
1654         } while (!kthread_should_stop());
1655         return 0;
1656 }
1657
1658 static int transaction_kthread(void *arg)
1659 {
1660         struct btrfs_root *root = arg;
1661         struct btrfs_trans_handle *trans;
1662         struct btrfs_transaction *cur;
1663         u64 transid;
1664         unsigned long now;
1665         unsigned long delay;
1666         bool cannot_commit;
1667
1668         do {
1669                 cannot_commit = false;
1670                 delay = HZ * 30;
1671                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1672
1673                 spin_lock(&root->fs_info->trans_lock);
1674                 cur = root->fs_info->running_transaction;
1675                 if (!cur) {
1676                         spin_unlock(&root->fs_info->trans_lock);
1677                         goto sleep;
1678                 }
1679
1680                 now = get_seconds();
1681                 if (!cur->blocked &&
1682                     (now < cur->start_time || now - cur->start_time < 30)) {
1683                         spin_unlock(&root->fs_info->trans_lock);
1684                         delay = HZ * 5;
1685                         goto sleep;
1686                 }
1687                 transid = cur->transid;
1688                 spin_unlock(&root->fs_info->trans_lock);
1689
1690                 /* If the file system is aborted, this will always fail. */
1691                 trans = btrfs_attach_transaction(root);
1692                 if (IS_ERR(trans)) {
1693                         if (PTR_ERR(trans) != -ENOENT)
1694                                 cannot_commit = true;
1695                         goto sleep;
1696                 }
1697                 if (transid == trans->transid) {
1698                         btrfs_commit_transaction(trans, root);
1699                 } else {
1700                         btrfs_end_transaction(trans, root);
1701                 }
1702 sleep:
1703                 wake_up_process(root->fs_info->cleaner_kthread);
1704                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1705
1706                 if (!try_to_freeze()) {
1707                         set_current_state(TASK_INTERRUPTIBLE);
1708                         if (!kthread_should_stop() &&
1709                             (!btrfs_transaction_blocked(root->fs_info) ||
1710                              cannot_commit))
1711                                 schedule_timeout(delay);
1712                         __set_current_state(TASK_RUNNING);
1713                 }
1714         } while (!kthread_should_stop());
1715         return 0;
1716 }
1717
1718 /*
1719  * this will find the highest generation in the array of
1720  * root backups.  The index of the highest array is returned,
1721  * or -1 if we can't find anything.
1722  *
1723  * We check to make sure the array is valid by comparing the
1724  * generation of the latest  root in the array with the generation
1725  * in the super block.  If they don't match we pitch it.
1726  */
1727 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1728 {
1729         u64 cur;
1730         int newest_index = -1;
1731         struct btrfs_root_backup *root_backup;
1732         int i;
1733
1734         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1735                 root_backup = info->super_copy->super_roots + i;
1736                 cur = btrfs_backup_tree_root_gen(root_backup);
1737                 if (cur == newest_gen)
1738                         newest_index = i;
1739         }
1740
1741         /* check to see if we actually wrapped around */
1742         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1743                 root_backup = info->super_copy->super_roots;
1744                 cur = btrfs_backup_tree_root_gen(root_backup);
1745                 if (cur == newest_gen)
1746                         newest_index = 0;
1747         }
1748         return newest_index;
1749 }
1750
1751
1752 /*
1753  * find the oldest backup so we know where to store new entries
1754  * in the backup array.  This will set the backup_root_index
1755  * field in the fs_info struct
1756  */
1757 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1758                                      u64 newest_gen)
1759 {
1760         int newest_index = -1;
1761
1762         newest_index = find_newest_super_backup(info, newest_gen);
1763         /* if there was garbage in there, just move along */
1764         if (newest_index == -1) {
1765                 info->backup_root_index = 0;
1766         } else {
1767                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1768         }
1769 }
1770
1771 /*
1772  * copy all the root pointers into the super backup array.
1773  * this will bump the backup pointer by one when it is
1774  * done
1775  */
1776 static void backup_super_roots(struct btrfs_fs_info *info)
1777 {
1778         int next_backup;
1779         struct btrfs_root_backup *root_backup;
1780         int last_backup;
1781
1782         next_backup = info->backup_root_index;
1783         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1784                 BTRFS_NUM_BACKUP_ROOTS;
1785
1786         /*
1787          * just overwrite the last backup if we're at the same generation
1788          * this happens only at umount
1789          */
1790         root_backup = info->super_for_commit->super_roots + last_backup;
1791         if (btrfs_backup_tree_root_gen(root_backup) ==
1792             btrfs_header_generation(info->tree_root->node))
1793                 next_backup = last_backup;
1794
1795         root_backup = info->super_for_commit->super_roots + next_backup;
1796
1797         /*
1798          * make sure all of our padding and empty slots get zero filled
1799          * regardless of which ones we use today
1800          */
1801         memset(root_backup, 0, sizeof(*root_backup));
1802
1803         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1804
1805         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1806         btrfs_set_backup_tree_root_gen(root_backup,
1807                                btrfs_header_generation(info->tree_root->node));
1808
1809         btrfs_set_backup_tree_root_level(root_backup,
1810                                btrfs_header_level(info->tree_root->node));
1811
1812         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1813         btrfs_set_backup_chunk_root_gen(root_backup,
1814                                btrfs_header_generation(info->chunk_root->node));
1815         btrfs_set_backup_chunk_root_level(root_backup,
1816                                btrfs_header_level(info->chunk_root->node));
1817
1818         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1819         btrfs_set_backup_extent_root_gen(root_backup,
1820                                btrfs_header_generation(info->extent_root->node));
1821         btrfs_set_backup_extent_root_level(root_backup,
1822                                btrfs_header_level(info->extent_root->node));
1823
1824         /*
1825          * we might commit during log recovery, which happens before we set
1826          * the fs_root.  Make sure it is valid before we fill it in.
1827          */
1828         if (info->fs_root && info->fs_root->node) {
1829                 btrfs_set_backup_fs_root(root_backup,
1830                                          info->fs_root->node->start);
1831                 btrfs_set_backup_fs_root_gen(root_backup,
1832                                btrfs_header_generation(info->fs_root->node));
1833                 btrfs_set_backup_fs_root_level(root_backup,
1834                                btrfs_header_level(info->fs_root->node));
1835         }
1836
1837         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1838         btrfs_set_backup_dev_root_gen(root_backup,
1839                                btrfs_header_generation(info->dev_root->node));
1840         btrfs_set_backup_dev_root_level(root_backup,
1841                                        btrfs_header_level(info->dev_root->node));
1842
1843         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1844         btrfs_set_backup_csum_root_gen(root_backup,
1845                                btrfs_header_generation(info->csum_root->node));
1846         btrfs_set_backup_csum_root_level(root_backup,
1847                                btrfs_header_level(info->csum_root->node));
1848
1849         btrfs_set_backup_total_bytes(root_backup,
1850                              btrfs_super_total_bytes(info->super_copy));
1851         btrfs_set_backup_bytes_used(root_backup,
1852                              btrfs_super_bytes_used(info->super_copy));
1853         btrfs_set_backup_num_devices(root_backup,
1854                              btrfs_super_num_devices(info->super_copy));
1855
1856         /*
1857          * if we don't copy this out to the super_copy, it won't get remembered
1858          * for the next commit
1859          */
1860         memcpy(&info->super_copy->super_roots,
1861                &info->super_for_commit->super_roots,
1862                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1863 }
1864
1865 /*
1866  * this copies info out of the root backup array and back into
1867  * the in-memory super block.  It is meant to help iterate through
1868  * the array, so you send it the number of backups you've already
1869  * tried and the last backup index you used.
1870  *
1871  * this returns -1 when it has tried all the backups
1872  */
1873 static noinline int next_root_backup(struct btrfs_fs_info *info,
1874                                      struct btrfs_super_block *super,
1875                                      int *num_backups_tried, int *backup_index)
1876 {
1877         struct btrfs_root_backup *root_backup;
1878         int newest = *backup_index;
1879
1880         if (*num_backups_tried == 0) {
1881                 u64 gen = btrfs_super_generation(super);
1882
1883                 newest = find_newest_super_backup(info, gen);
1884                 if (newest == -1)
1885                         return -1;
1886
1887                 *backup_index = newest;
1888                 *num_backups_tried = 1;
1889         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1890                 /* we've tried all the backups, all done */
1891                 return -1;
1892         } else {
1893                 /* jump to the next oldest backup */
1894                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1895                         BTRFS_NUM_BACKUP_ROOTS;
1896                 *backup_index = newest;
1897                 *num_backups_tried += 1;
1898         }
1899         root_backup = super->super_roots + newest;
1900
1901         btrfs_set_super_generation(super,
1902                                    btrfs_backup_tree_root_gen(root_backup));
1903         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1904         btrfs_set_super_root_level(super,
1905                                    btrfs_backup_tree_root_level(root_backup));
1906         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1907
1908         /*
1909          * fixme: the total bytes and num_devices need to match or we should
1910          * need a fsck
1911          */
1912         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1913         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1914         return 0;
1915 }
1916
1917 /* helper to cleanup tree roots */
1918 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1919 {
1920         free_extent_buffer(info->tree_root->node);
1921         free_extent_buffer(info->tree_root->commit_root);
1922         free_extent_buffer(info->dev_root->node);
1923         free_extent_buffer(info->dev_root->commit_root);
1924         free_extent_buffer(info->extent_root->node);
1925         free_extent_buffer(info->extent_root->commit_root);
1926         free_extent_buffer(info->csum_root->node);
1927         free_extent_buffer(info->csum_root->commit_root);
1928         if (info->quota_root) {
1929                 free_extent_buffer(info->quota_root->node);
1930                 free_extent_buffer(info->quota_root->commit_root);
1931         }
1932
1933         info->tree_root->node = NULL;
1934         info->tree_root->commit_root = NULL;
1935         info->dev_root->node = NULL;
1936         info->dev_root->commit_root = NULL;
1937         info->extent_root->node = NULL;
1938         info->extent_root->commit_root = NULL;
1939         info->csum_root->node = NULL;
1940         info->csum_root->commit_root = NULL;
1941         if (info->quota_root) {
1942                 info->quota_root->node = NULL;
1943                 info->quota_root->commit_root = NULL;
1944         }
1945
1946         if (chunk_root) {
1947                 free_extent_buffer(info->chunk_root->node);
1948                 free_extent_buffer(info->chunk_root->commit_root);
1949                 info->chunk_root->node = NULL;
1950                 info->chunk_root->commit_root = NULL;
1951         }
1952 }
1953
1954
1955 int open_ctree(struct super_block *sb,
1956                struct btrfs_fs_devices *fs_devices,
1957                char *options)
1958 {
1959         u32 sectorsize;
1960         u32 nodesize;
1961         u32 leafsize;
1962         u32 blocksize;
1963         u32 stripesize;
1964         u64 generation;
1965         u64 features;
1966         struct btrfs_key location;
1967         struct buffer_head *bh;
1968         struct btrfs_super_block *disk_super;
1969         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1970         struct btrfs_root *tree_root;
1971         struct btrfs_root *extent_root;
1972         struct btrfs_root *csum_root;
1973         struct btrfs_root *chunk_root;
1974         struct btrfs_root *dev_root;
1975         struct btrfs_root *quota_root;
1976         struct btrfs_root *log_tree_root;
1977         int ret;
1978         int err = -EINVAL;
1979         int num_backups_tried = 0;
1980         int backup_index = 0;
1981
1982         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1983         extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1984         csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1985         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1986         dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1987         quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
1988
1989         if (!tree_root || !extent_root || !csum_root ||
1990             !chunk_root || !dev_root || !quota_root) {
1991                 err = -ENOMEM;
1992                 goto fail;
1993         }
1994
1995         ret = init_srcu_struct(&fs_info->subvol_srcu);
1996         if (ret) {
1997                 err = ret;
1998                 goto fail;
1999         }
2000
2001         ret = setup_bdi(fs_info, &fs_info->bdi);
2002         if (ret) {
2003                 err = ret;
2004                 goto fail_srcu;
2005         }
2006
2007         fs_info->btree_inode = new_inode(sb);
2008         if (!fs_info->btree_inode) {
2009                 err = -ENOMEM;
2010                 goto fail_bdi;
2011         }
2012
2013         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2014
2015         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2016         INIT_LIST_HEAD(&fs_info->trans_list);
2017         INIT_LIST_HEAD(&fs_info->dead_roots);
2018         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2019         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
2020         INIT_LIST_HEAD(&fs_info->ordered_operations);
2021         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2022         spin_lock_init(&fs_info->delalloc_lock);
2023         spin_lock_init(&fs_info->trans_lock);
2024         spin_lock_init(&fs_info->fs_roots_radix_lock);
2025         spin_lock_init(&fs_info->delayed_iput_lock);
2026         spin_lock_init(&fs_info->defrag_inodes_lock);
2027         spin_lock_init(&fs_info->free_chunk_lock);
2028         spin_lock_init(&fs_info->tree_mod_seq_lock);
2029         rwlock_init(&fs_info->tree_mod_log_lock);
2030         mutex_init(&fs_info->reloc_mutex);
2031
2032         init_completion(&fs_info->kobj_unregister);
2033         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2034         INIT_LIST_HEAD(&fs_info->space_info);
2035         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2036         btrfs_mapping_init(&fs_info->mapping_tree);
2037         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2038                              BTRFS_BLOCK_RSV_GLOBAL);
2039         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2040                              BTRFS_BLOCK_RSV_DELALLOC);
2041         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2042         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2043         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2044         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2045                              BTRFS_BLOCK_RSV_DELOPS);
2046         atomic_set(&fs_info->nr_async_submits, 0);
2047         atomic_set(&fs_info->async_delalloc_pages, 0);
2048         atomic_set(&fs_info->async_submit_draining, 0);
2049         atomic_set(&fs_info->nr_async_bios, 0);
2050         atomic_set(&fs_info->defrag_running, 0);
2051         atomic_set(&fs_info->tree_mod_seq, 0);
2052         fs_info->sb = sb;
2053         fs_info->max_inline = 8192 * 1024;
2054         fs_info->metadata_ratio = 0;
2055         fs_info->defrag_inodes = RB_ROOT;
2056         fs_info->trans_no_join = 0;
2057         fs_info->free_chunk_space = 0;
2058         fs_info->tree_mod_log = RB_ROOT;
2059
2060         /* readahead state */
2061         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2062         spin_lock_init(&fs_info->reada_lock);
2063
2064         fs_info->thread_pool_size = min_t(unsigned long,
2065                                           num_online_cpus() + 2, 8);
2066
2067         INIT_LIST_HEAD(&fs_info->ordered_extents);
2068         spin_lock_init(&fs_info->ordered_extent_lock);
2069         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2070                                         GFP_NOFS);
2071         if (!fs_info->delayed_root) {
2072                 err = -ENOMEM;
2073                 goto fail_iput;
2074         }
2075         btrfs_init_delayed_root(fs_info->delayed_root);
2076
2077         mutex_init(&fs_info->scrub_lock);
2078         atomic_set(&fs_info->scrubs_running, 0);
2079         atomic_set(&fs_info->scrub_pause_req, 0);
2080         atomic_set(&fs_info->scrubs_paused, 0);
2081         atomic_set(&fs_info->scrub_cancel_req, 0);
2082         init_waitqueue_head(&fs_info->scrub_pause_wait);
2083         init_rwsem(&fs_info->scrub_super_lock);
2084         fs_info->scrub_workers_refcnt = 0;
2085 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2086         fs_info->check_integrity_print_mask = 0;
2087 #endif
2088
2089         spin_lock_init(&fs_info->balance_lock);
2090         mutex_init(&fs_info->balance_mutex);
2091         atomic_set(&fs_info->balance_running, 0);
2092         atomic_set(&fs_info->balance_pause_req, 0);
2093         atomic_set(&fs_info->balance_cancel_req, 0);
2094         fs_info->balance_ctl = NULL;
2095         init_waitqueue_head(&fs_info->balance_wait_q);
2096
2097         sb->s_blocksize = 4096;
2098         sb->s_blocksize_bits = blksize_bits(4096);
2099         sb->s_bdi = &fs_info->bdi;
2100
2101         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2102         set_nlink(fs_info->btree_inode, 1);
2103         /*
2104          * we set the i_size on the btree inode to the max possible int.
2105          * the real end of the address space is determined by all of
2106          * the devices in the system
2107          */
2108         fs_info->btree_inode->i_size = OFFSET_MAX;
2109         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2110         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2111
2112         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2113         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2114                              fs_info->btree_inode->i_mapping);
2115         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2116         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2117
2118         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2119
2120         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2121         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2122                sizeof(struct btrfs_key));
2123         set_bit(BTRFS_INODE_DUMMY,
2124                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2125         insert_inode_hash(fs_info->btree_inode);
2126
2127         spin_lock_init(&fs_info->block_group_cache_lock);
2128         fs_info->block_group_cache_tree = RB_ROOT;
2129
2130         extent_io_tree_init(&fs_info->freed_extents[0],
2131                              fs_info->btree_inode->i_mapping);
2132         extent_io_tree_init(&fs_info->freed_extents[1],
2133                              fs_info->btree_inode->i_mapping);
2134         fs_info->pinned_extents = &fs_info->freed_extents[0];
2135         fs_info->do_barriers = 1;
2136
2137
2138         mutex_init(&fs_info->ordered_operations_mutex);
2139         mutex_init(&fs_info->tree_log_mutex);
2140         mutex_init(&fs_info->chunk_mutex);
2141         mutex_init(&fs_info->transaction_kthread_mutex);
2142         mutex_init(&fs_info->cleaner_mutex);
2143         mutex_init(&fs_info->volume_mutex);
2144         init_rwsem(&fs_info->extent_commit_sem);
2145         init_rwsem(&fs_info->cleanup_work_sem);
2146         init_rwsem(&fs_info->subvol_sem);
2147         fs_info->dev_replace.lock_owner = 0;
2148         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2149         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2150         mutex_init(&fs_info->dev_replace.lock_management_lock);
2151         mutex_init(&fs_info->dev_replace.lock);
2152
2153         spin_lock_init(&fs_info->qgroup_lock);
2154         fs_info->qgroup_tree = RB_ROOT;
2155         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2156         fs_info->qgroup_seq = 1;
2157         fs_info->quota_enabled = 0;
2158         fs_info->pending_quota_state = 0;
2159
2160         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2161         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2162
2163         init_waitqueue_head(&fs_info->transaction_throttle);
2164         init_waitqueue_head(&fs_info->transaction_wait);
2165         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2166         init_waitqueue_head(&fs_info->async_submit_wait);
2167
2168         __setup_root(4096, 4096, 4096, 4096, tree_root,
2169                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2170
2171         invalidate_bdev(fs_devices->latest_bdev);
2172         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2173         if (!bh) {
2174                 err = -EINVAL;
2175                 goto fail_alloc;
2176         }
2177
2178         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2179         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2180                sizeof(*fs_info->super_for_commit));
2181         brelse(bh);
2182
2183         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2184
2185         disk_super = fs_info->super_copy;
2186         if (!btrfs_super_root(disk_super))
2187                 goto fail_alloc;
2188
2189         /* check FS state, whether FS is broken. */
2190         fs_info->fs_state |= btrfs_super_flags(disk_super);
2191
2192         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2193         if (ret) {
2194                 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2195                 err = ret;
2196                 goto fail_alloc;
2197         }
2198
2199         /*
2200          * run through our array of backup supers and setup
2201          * our ring pointer to the oldest one
2202          */
2203         generation = btrfs_super_generation(disk_super);
2204         find_oldest_super_backup(fs_info, generation);
2205
2206         /*
2207          * In the long term, we'll store the compression type in the super
2208          * block, and it'll be used for per file compression control.
2209          */
2210         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2211
2212         ret = btrfs_parse_options(tree_root, options);
2213         if (ret) {
2214                 err = ret;
2215                 goto fail_alloc;
2216         }
2217
2218         features = btrfs_super_incompat_flags(disk_super) &
2219                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2220         if (features) {
2221                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2222                        "unsupported optional features (%Lx).\n",
2223                        (unsigned long long)features);
2224                 err = -EINVAL;
2225                 goto fail_alloc;
2226         }
2227
2228         if (btrfs_super_leafsize(disk_super) !=
2229             btrfs_super_nodesize(disk_super)) {
2230                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2231                        "blocksizes don't match.  node %d leaf %d\n",
2232                        btrfs_super_nodesize(disk_super),
2233                        btrfs_super_leafsize(disk_super));
2234                 err = -EINVAL;
2235                 goto fail_alloc;
2236         }
2237         if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2238                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2239                        "blocksize (%d) was too large\n",
2240                        btrfs_super_leafsize(disk_super));
2241                 err = -EINVAL;
2242                 goto fail_alloc;
2243         }
2244
2245         features = btrfs_super_incompat_flags(disk_super);
2246         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2247         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2248                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2249
2250         /*
2251          * flag our filesystem as having big metadata blocks if
2252          * they are bigger than the page size
2253          */
2254         if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2255                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2256                         printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2257                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2258         }
2259
2260         nodesize = btrfs_super_nodesize(disk_super);
2261         leafsize = btrfs_super_leafsize(disk_super);
2262         sectorsize = btrfs_super_sectorsize(disk_super);
2263         stripesize = btrfs_super_stripesize(disk_super);
2264
2265         /*
2266          * mixed block groups end up with duplicate but slightly offset
2267          * extent buffers for the same range.  It leads to corruptions
2268          */
2269         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2270             (sectorsize != leafsize)) {
2271                 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2272                                 "are not allowed for mixed block groups on %s\n",
2273                                 sb->s_id);
2274                 goto fail_alloc;
2275         }
2276
2277         btrfs_set_super_incompat_flags(disk_super, features);
2278
2279         features = btrfs_super_compat_ro_flags(disk_super) &
2280                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2281         if (!(sb->s_flags & MS_RDONLY) && features) {
2282                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2283                        "unsupported option features (%Lx).\n",
2284                        (unsigned long long)features);
2285                 err = -EINVAL;
2286                 goto fail_alloc;
2287         }
2288
2289         btrfs_init_workers(&fs_info->generic_worker,
2290                            "genwork", 1, NULL);
2291
2292         btrfs_init_workers(&fs_info->workers, "worker",
2293                            fs_info->thread_pool_size,
2294                            &fs_info->generic_worker);
2295
2296         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2297                            fs_info->thread_pool_size,
2298                            &fs_info->generic_worker);
2299
2300         btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2301                            fs_info->thread_pool_size,
2302                            &fs_info->generic_worker);
2303
2304         btrfs_init_workers(&fs_info->submit_workers, "submit",
2305                            min_t(u64, fs_devices->num_devices,
2306                            fs_info->thread_pool_size),
2307                            &fs_info->generic_worker);
2308
2309         btrfs_init_workers(&fs_info->caching_workers, "cache",
2310                            2, &fs_info->generic_worker);
2311
2312         /* a higher idle thresh on the submit workers makes it much more
2313          * likely that bios will be send down in a sane order to the
2314          * devices
2315          */
2316         fs_info->submit_workers.idle_thresh = 64;
2317
2318         fs_info->workers.idle_thresh = 16;
2319         fs_info->workers.ordered = 1;
2320
2321         fs_info->delalloc_workers.idle_thresh = 2;
2322         fs_info->delalloc_workers.ordered = 1;
2323
2324         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2325                            &fs_info->generic_worker);
2326         btrfs_init_workers(&fs_info->endio_workers, "endio",
2327                            fs_info->thread_pool_size,
2328                            &fs_info->generic_worker);
2329         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2330                            fs_info->thread_pool_size,
2331                            &fs_info->generic_worker);
2332         btrfs_init_workers(&fs_info->endio_meta_write_workers,
2333                            "endio-meta-write", fs_info->thread_pool_size,
2334                            &fs_info->generic_worker);
2335         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2336                            fs_info->thread_pool_size,
2337                            &fs_info->generic_worker);
2338         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2339                            1, &fs_info->generic_worker);
2340         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2341                            fs_info->thread_pool_size,
2342                            &fs_info->generic_worker);
2343         btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2344                            fs_info->thread_pool_size,
2345                            &fs_info->generic_worker);
2346
2347         /*
2348          * endios are largely parallel and should have a very
2349          * low idle thresh
2350          */
2351         fs_info->endio_workers.idle_thresh = 4;
2352         fs_info->endio_meta_workers.idle_thresh = 4;
2353
2354         fs_info->endio_write_workers.idle_thresh = 2;
2355         fs_info->endio_meta_write_workers.idle_thresh = 2;
2356         fs_info->readahead_workers.idle_thresh = 2;
2357
2358         /*
2359          * btrfs_start_workers can really only fail because of ENOMEM so just
2360          * return -ENOMEM if any of these fail.
2361          */
2362         ret = btrfs_start_workers(&fs_info->workers);
2363         ret |= btrfs_start_workers(&fs_info->generic_worker);
2364         ret |= btrfs_start_workers(&fs_info->submit_workers);
2365         ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2366         ret |= btrfs_start_workers(&fs_info->fixup_workers);
2367         ret |= btrfs_start_workers(&fs_info->endio_workers);
2368         ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2369         ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2370         ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2371         ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2372         ret |= btrfs_start_workers(&fs_info->delayed_workers);
2373         ret |= btrfs_start_workers(&fs_info->caching_workers);
2374         ret |= btrfs_start_workers(&fs_info->readahead_workers);
2375         ret |= btrfs_start_workers(&fs_info->flush_workers);
2376         if (ret) {
2377                 err = -ENOMEM;
2378                 goto fail_sb_buffer;
2379         }
2380
2381         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2382         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2383                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2384
2385         tree_root->nodesize = nodesize;
2386         tree_root->leafsize = leafsize;
2387         tree_root->sectorsize = sectorsize;
2388         tree_root->stripesize = stripesize;
2389
2390         sb->s_blocksize = sectorsize;
2391         sb->s_blocksize_bits = blksize_bits(sectorsize);
2392
2393         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2394                     sizeof(disk_super->magic))) {
2395                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2396                 goto fail_sb_buffer;
2397         }
2398
2399         if (sectorsize != PAGE_SIZE) {
2400                 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2401                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2402                 goto fail_sb_buffer;
2403         }
2404
2405         mutex_lock(&fs_info->chunk_mutex);
2406         ret = btrfs_read_sys_array(tree_root);
2407         mutex_unlock(&fs_info->chunk_mutex);
2408         if (ret) {
2409                 printk(KERN_WARNING "btrfs: failed to read the system "
2410                        "array on %s\n", sb->s_id);
2411                 goto fail_sb_buffer;
2412         }
2413
2414         blocksize = btrfs_level_size(tree_root,
2415                                      btrfs_super_chunk_root_level(disk_super));
2416         generation = btrfs_super_chunk_root_generation(disk_super);
2417
2418         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2419                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2420
2421         chunk_root->node = read_tree_block(chunk_root,
2422                                            btrfs_super_chunk_root(disk_super),
2423                                            blocksize, generation);
2424         BUG_ON(!chunk_root->node); /* -ENOMEM */
2425         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2426                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2427                        sb->s_id);
2428                 goto fail_tree_roots;
2429         }
2430         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2431         chunk_root->commit_root = btrfs_root_node(chunk_root);
2432
2433         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2434            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2435            BTRFS_UUID_SIZE);
2436
2437         ret = btrfs_read_chunk_tree(chunk_root);
2438         if (ret) {
2439                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2440                        sb->s_id);
2441                 goto fail_tree_roots;
2442         }
2443
2444         /*
2445          * keep the device that is marked to be the target device for the
2446          * dev_replace procedure
2447          */
2448         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2449
2450         if (!fs_devices->latest_bdev) {
2451                 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2452                        sb->s_id);
2453                 goto fail_tree_roots;
2454         }
2455
2456 retry_root_backup:
2457         blocksize = btrfs_level_size(tree_root,
2458                                      btrfs_super_root_level(disk_super));
2459         generation = btrfs_super_generation(disk_super);
2460
2461         tree_root->node = read_tree_block(tree_root,
2462                                           btrfs_super_root(disk_super),
2463                                           blocksize, generation);
2464         if (!tree_root->node ||
2465             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2466                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2467                        sb->s_id);
2468
2469                 goto recovery_tree_root;
2470         }
2471
2472         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2473         tree_root->commit_root = btrfs_root_node(tree_root);
2474
2475         ret = find_and_setup_root(tree_root, fs_info,
2476                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2477         if (ret)
2478                 goto recovery_tree_root;
2479         extent_root->track_dirty = 1;
2480
2481         ret = find_and_setup_root(tree_root, fs_info,
2482                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2483         if (ret)
2484                 goto recovery_tree_root;
2485         dev_root->track_dirty = 1;
2486
2487         ret = find_and_setup_root(tree_root, fs_info,
2488                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2489         if (ret)
2490                 goto recovery_tree_root;
2491         csum_root->track_dirty = 1;
2492
2493         ret = find_and_setup_root(tree_root, fs_info,
2494                                   BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2495         if (ret) {
2496                 kfree(quota_root);
2497                 quota_root = fs_info->quota_root = NULL;
2498         } else {
2499                 quota_root->track_dirty = 1;
2500                 fs_info->quota_enabled = 1;
2501                 fs_info->pending_quota_state = 1;
2502         }
2503
2504         fs_info->generation = generation;
2505         fs_info->last_trans_committed = generation;
2506
2507         ret = btrfs_recover_balance(fs_info);
2508         if (ret) {
2509                 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2510                 goto fail_block_groups;
2511         }
2512
2513         ret = btrfs_init_dev_stats(fs_info);
2514         if (ret) {
2515                 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2516                        ret);
2517                 goto fail_block_groups;
2518         }
2519
2520         ret = btrfs_init_dev_replace(fs_info);
2521         if (ret) {
2522                 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2523                 goto fail_block_groups;
2524         }
2525
2526         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2527
2528         ret = btrfs_init_space_info(fs_info);
2529         if (ret) {
2530                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2531                 goto fail_block_groups;
2532         }
2533
2534         ret = btrfs_read_block_groups(extent_root);
2535         if (ret) {
2536                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2537                 goto fail_block_groups;
2538         }
2539         fs_info->num_tolerated_disk_barrier_failures =
2540                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2541         if (fs_info->fs_devices->missing_devices >
2542              fs_info->num_tolerated_disk_barrier_failures &&
2543             !(sb->s_flags & MS_RDONLY)) {
2544                 printk(KERN_WARNING
2545                        "Btrfs: too many missing devices, writeable mount is not allowed\n");
2546                 goto fail_block_groups;
2547         }
2548
2549         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2550                                                "btrfs-cleaner");
2551         if (IS_ERR(fs_info->cleaner_kthread))
2552                 goto fail_block_groups;
2553
2554         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2555                                                    tree_root,
2556                                                    "btrfs-transaction");
2557         if (IS_ERR(fs_info->transaction_kthread))
2558                 goto fail_cleaner;
2559
2560         if (!btrfs_test_opt(tree_root, SSD) &&
2561             !btrfs_test_opt(tree_root, NOSSD) &&
2562             !fs_info->fs_devices->rotating) {
2563                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2564                        "mode\n");
2565                 btrfs_set_opt(fs_info->mount_opt, SSD);
2566         }
2567
2568 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2569         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2570                 ret = btrfsic_mount(tree_root, fs_devices,
2571                                     btrfs_test_opt(tree_root,
2572                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2573                                     1 : 0,
2574                                     fs_info->check_integrity_print_mask);
2575                 if (ret)
2576                         printk(KERN_WARNING "btrfs: failed to initialize"
2577                                " integrity check module %s\n", sb->s_id);
2578         }
2579 #endif
2580         ret = btrfs_read_qgroup_config(fs_info);
2581         if (ret)
2582                 goto fail_trans_kthread;
2583
2584         /* do not make disk changes in broken FS */
2585         if (btrfs_super_log_root(disk_super) != 0) {
2586                 u64 bytenr = btrfs_super_log_root(disk_super);
2587
2588                 if (fs_devices->rw_devices == 0) {
2589                         printk(KERN_WARNING "Btrfs log replay required "
2590                                "on RO media\n");
2591                         err = -EIO;
2592                         goto fail_qgroup;
2593                 }
2594                 blocksize =
2595                      btrfs_level_size(tree_root,
2596                                       btrfs_super_log_root_level(disk_super));
2597
2598                 log_tree_root = btrfs_alloc_root(fs_info);
2599                 if (!log_tree_root) {
2600                         err = -ENOMEM;
2601                         goto fail_qgroup;
2602                 }
2603
2604                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2605                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2606
2607                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2608                                                       blocksize,
2609                                                       generation + 1);
2610                 /* returns with log_tree_root freed on success */
2611                 ret = btrfs_recover_log_trees(log_tree_root);
2612                 if (ret) {
2613                         btrfs_error(tree_root->fs_info, ret,
2614                                     "Failed to recover log tree");
2615                         free_extent_buffer(log_tree_root->node);
2616                         kfree(log_tree_root);
2617                         goto fail_trans_kthread;
2618                 }
2619
2620                 if (sb->s_flags & MS_RDONLY) {
2621                         ret = btrfs_commit_super(tree_root);
2622                         if (ret)
2623                                 goto fail_trans_kthread;
2624                 }
2625         }
2626
2627         ret = btrfs_find_orphan_roots(tree_root);
2628         if (ret)
2629                 goto fail_trans_kthread;
2630
2631         if (!(sb->s_flags & MS_RDONLY)) {
2632                 ret = btrfs_cleanup_fs_roots(fs_info);
2633                 if (ret)
2634                         goto fail_trans_kthread;
2635
2636                 ret = btrfs_recover_relocation(tree_root);
2637                 if (ret < 0) {
2638                         printk(KERN_WARNING
2639                                "btrfs: failed to recover relocation\n");
2640                         err = -EINVAL;
2641                         goto fail_qgroup;
2642                 }
2643         }
2644
2645         location.objectid = BTRFS_FS_TREE_OBJECTID;
2646         location.type = BTRFS_ROOT_ITEM_KEY;
2647         location.offset = (u64)-1;
2648
2649         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2650         if (!fs_info->fs_root)
2651                 goto fail_qgroup;
2652         if (IS_ERR(fs_info->fs_root)) {
2653                 err = PTR_ERR(fs_info->fs_root);
2654                 goto fail_qgroup;
2655         }
2656
2657         if (sb->s_flags & MS_RDONLY)
2658                 return 0;
2659
2660         down_read(&fs_info->cleanup_work_sem);
2661         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2662             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2663                 up_read(&fs_info->cleanup_work_sem);
2664                 close_ctree(tree_root);
2665                 return ret;
2666         }
2667         up_read(&fs_info->cleanup_work_sem);
2668
2669         ret = btrfs_resume_balance_async(fs_info);
2670         if (ret) {
2671                 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2672                 close_ctree(tree_root);
2673                 return ret;
2674         }
2675
2676         ret = btrfs_resume_dev_replace_async(fs_info);
2677         if (ret) {
2678                 pr_warn("btrfs: failed to resume dev_replace\n");
2679                 close_ctree(tree_root);
2680                 return ret;
2681         }
2682
2683         return 0;
2684
2685 fail_qgroup:
2686         btrfs_free_qgroup_config(fs_info);
2687 fail_trans_kthread:
2688         kthread_stop(fs_info->transaction_kthread);
2689 fail_cleaner:
2690         kthread_stop(fs_info->cleaner_kthread);
2691
2692         /*
2693          * make sure we're done with the btree inode before we stop our
2694          * kthreads
2695          */
2696         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2697         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2698
2699 fail_block_groups:
2700         btrfs_free_block_groups(fs_info);
2701
2702 fail_tree_roots:
2703         free_root_pointers(fs_info, 1);
2704
2705 fail_sb_buffer:
2706         btrfs_stop_workers(&fs_info->generic_worker);
2707         btrfs_stop_workers(&fs_info->readahead_workers);
2708         btrfs_stop_workers(&fs_info->fixup_workers);
2709         btrfs_stop_workers(&fs_info->delalloc_workers);
2710         btrfs_stop_workers(&fs_info->workers);
2711         btrfs_stop_workers(&fs_info->endio_workers);
2712         btrfs_stop_workers(&fs_info->endio_meta_workers);
2713         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2714         btrfs_stop_workers(&fs_info->endio_write_workers);
2715         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2716         btrfs_stop_workers(&fs_info->submit_workers);
2717         btrfs_stop_workers(&fs_info->delayed_workers);
2718         btrfs_stop_workers(&fs_info->caching_workers);
2719         btrfs_stop_workers(&fs_info->flush_workers);
2720 fail_alloc:
2721 fail_iput:
2722         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2723
2724         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2725         iput(fs_info->btree_inode);
2726 fail_bdi:
2727         bdi_destroy(&fs_info->bdi);
2728 fail_srcu:
2729         cleanup_srcu_struct(&fs_info->subvol_srcu);
2730 fail:
2731         btrfs_close_devices(fs_info->fs_devices);
2732         return err;
2733
2734 recovery_tree_root:
2735         if (!btrfs_test_opt(tree_root, RECOVERY))
2736                 goto fail_tree_roots;
2737
2738         free_root_pointers(fs_info, 0);
2739
2740         /* don't use the log in recovery mode, it won't be valid */
2741         btrfs_set_super_log_root(disk_super, 0);
2742
2743         /* we can't trust the free space cache either */
2744         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2745
2746         ret = next_root_backup(fs_info, fs_info->super_copy,
2747                                &num_backups_tried, &backup_index);
2748         if (ret == -1)
2749                 goto fail_block_groups;
2750         goto retry_root_backup;
2751 }
2752
2753 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2754 {
2755         if (uptodate) {
2756                 set_buffer_uptodate(bh);
2757         } else {
2758                 struct btrfs_device *device = (struct btrfs_device *)
2759                         bh->b_private;
2760
2761                 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2762                                           "I/O error on %s\n",
2763                                           rcu_str_deref(device->name));
2764                 /* note, we dont' set_buffer_write_io_error because we have
2765                  * our own ways of dealing with the IO errors
2766                  */
2767                 clear_buffer_uptodate(bh);
2768                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2769         }
2770         unlock_buffer(bh);
2771         put_bh(bh);
2772 }
2773
2774 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2775 {
2776         struct buffer_head *bh;
2777         struct buffer_head *latest = NULL;
2778         struct btrfs_super_block *super;
2779         int i;
2780         u64 transid = 0;
2781         u64 bytenr;
2782
2783         /* we would like to check all the supers, but that would make
2784          * a btrfs mount succeed after a mkfs from a different FS.
2785          * So, we need to add a special mount option to scan for
2786          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2787          */
2788         for (i = 0; i < 1; i++) {
2789                 bytenr = btrfs_sb_offset(i);
2790                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2791                         break;
2792                 bh = __bread(bdev, bytenr / 4096, 4096);
2793                 if (!bh)
2794                         continue;
2795
2796                 super = (struct btrfs_super_block *)bh->b_data;
2797                 if (btrfs_super_bytenr(super) != bytenr ||
2798                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2799                             sizeof(super->magic))) {
2800                         brelse(bh);
2801                         continue;
2802                 }
2803
2804                 if (!latest || btrfs_super_generation(super) > transid) {
2805                         brelse(latest);
2806                         latest = bh;
2807                         transid = btrfs_super_generation(super);
2808                 } else {
2809                         brelse(bh);
2810                 }
2811         }
2812         return latest;
2813 }
2814
2815 /*
2816  * this should be called twice, once with wait == 0 and
2817  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2818  * we write are pinned.
2819  *
2820  * They are released when wait == 1 is done.
2821  * max_mirrors must be the same for both runs, and it indicates how
2822  * many supers on this one device should be written.
2823  *
2824  * max_mirrors == 0 means to write them all.
2825  */
2826 static int write_dev_supers(struct btrfs_device *device,
2827                             struct btrfs_super_block *sb,
2828                             int do_barriers, int wait, int max_mirrors)
2829 {
2830         struct buffer_head *bh;
2831         int i;
2832         int ret;
2833         int errors = 0;
2834         u32 crc;
2835         u64 bytenr;
2836
2837         if (max_mirrors == 0)
2838                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2839
2840         for (i = 0; i < max_mirrors; i++) {
2841                 bytenr = btrfs_sb_offset(i);
2842                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2843                         break;
2844
2845                 if (wait) {
2846                         bh = __find_get_block(device->bdev, bytenr / 4096,
2847                                               BTRFS_SUPER_INFO_SIZE);
2848                         BUG_ON(!bh);
2849                         wait_on_buffer(bh);
2850                         if (!buffer_uptodate(bh))
2851                                 errors++;
2852
2853                         /* drop our reference */
2854                         brelse(bh);
2855
2856                         /* drop the reference from the wait == 0 run */
2857                         brelse(bh);
2858                         continue;
2859                 } else {
2860                         btrfs_set_super_bytenr(sb, bytenr);
2861
2862                         crc = ~(u32)0;
2863                         crc = btrfs_csum_data(NULL, (char *)sb +
2864                                               BTRFS_CSUM_SIZE, crc,
2865                                               BTRFS_SUPER_INFO_SIZE -
2866                                               BTRFS_CSUM_SIZE);
2867                         btrfs_csum_final(crc, sb->csum);
2868
2869                         /*
2870                          * one reference for us, and we leave it for the
2871                          * caller
2872                          */
2873                         bh = __getblk(device->bdev, bytenr / 4096,
2874                                       BTRFS_SUPER_INFO_SIZE);
2875                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2876
2877                         /* one reference for submit_bh */
2878                         get_bh(bh);
2879
2880                         set_buffer_uptodate(bh);
2881                         lock_buffer(bh);
2882                         bh->b_end_io = btrfs_end_buffer_write_sync;
2883                         bh->b_private = device;
2884                 }
2885
2886                 /*
2887                  * we fua the first super.  The others we allow
2888                  * to go down lazy.
2889                  */
2890                 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2891                 if (ret)
2892                         errors++;
2893         }
2894         return errors < i ? 0 : -1;
2895 }
2896
2897 /*
2898  * endio for the write_dev_flush, this will wake anyone waiting
2899  * for the barrier when it is done
2900  */
2901 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2902 {
2903         if (err) {
2904                 if (err == -EOPNOTSUPP)
2905                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2906                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2907         }
2908         if (bio->bi_private)
2909                 complete(bio->bi_private);
2910         bio_put(bio);
2911 }
2912
2913 /*
2914  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
2915  * sent down.  With wait == 1, it waits for the previous flush.
2916  *
2917  * any device where the flush fails with eopnotsupp are flagged as not-barrier
2918  * capable
2919  */
2920 static int write_dev_flush(struct btrfs_device *device, int wait)
2921 {
2922         struct bio *bio;
2923         int ret = 0;
2924
2925         if (device->nobarriers)
2926                 return 0;
2927
2928         if (wait) {
2929                 bio = device->flush_bio;
2930                 if (!bio)
2931                         return 0;
2932
2933                 wait_for_completion(&device->flush_wait);
2934
2935                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2936                         printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2937                                       rcu_str_deref(device->name));
2938                         device->nobarriers = 1;
2939                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
2940                         ret = -EIO;
2941                         btrfs_dev_stat_inc_and_print(device,
2942                                 BTRFS_DEV_STAT_FLUSH_ERRS);
2943                 }
2944
2945                 /* drop the reference from the wait == 0 run */
2946                 bio_put(bio);
2947                 device->flush_bio = NULL;
2948
2949                 return ret;
2950         }
2951
2952         /*
2953          * one reference for us, and we leave it for the
2954          * caller
2955          */
2956         device->flush_bio = NULL;
2957         bio = bio_alloc(GFP_NOFS, 0);
2958         if (!bio)
2959                 return -ENOMEM;
2960
2961         bio->bi_end_io = btrfs_end_empty_barrier;
2962         bio->bi_bdev = device->bdev;
2963         init_completion(&device->flush_wait);
2964         bio->bi_private = &device->flush_wait;
2965         device->flush_bio = bio;
2966
2967         bio_get(bio);
2968         btrfsic_submit_bio(WRITE_FLUSH, bio);
2969
2970         return 0;
2971 }
2972
2973 /*
2974  * send an empty flush down to each device in parallel,
2975  * then wait for them
2976  */
2977 static int barrier_all_devices(struct btrfs_fs_info *info)
2978 {
2979         struct list_head *head;
2980         struct btrfs_device *dev;
2981         int errors_send = 0;
2982         int errors_wait = 0;
2983         int ret;
2984
2985         /* send down all the barriers */
2986         head = &info->fs_devices->devices;
2987         list_for_each_entry_rcu(dev, head, dev_list) {
2988                 if (!dev->bdev) {
2989                         errors_send++;
2990                         continue;
2991                 }
2992                 if (!dev->in_fs_metadata || !dev->writeable)
2993                         continue;
2994
2995                 ret = write_dev_flush(dev, 0);
2996                 if (ret)
2997                         errors_send++;
2998         }
2999
3000         /* wait for all the barriers */
3001         list_for_each_entry_rcu(dev, head, dev_list) {
3002                 if (!dev->bdev) {
3003                         errors_wait++;
3004                         continue;
3005                 }
3006                 if (!dev->in_fs_metadata || !dev->writeable)
3007                         continue;
3008
3009                 ret = write_dev_flush(dev, 1);
3010                 if (ret)
3011                         errors_wait++;
3012         }
3013         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3014             errors_wait > info->num_tolerated_disk_barrier_failures)
3015                 return -EIO;
3016         return 0;
3017 }
3018
3019 int btrfs_calc_num_tolerated_disk_barrier_failures(
3020         struct btrfs_fs_info *fs_info)
3021 {
3022         struct btrfs_ioctl_space_info space;
3023         struct btrfs_space_info *sinfo;
3024         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3025                        BTRFS_BLOCK_GROUP_SYSTEM,
3026                        BTRFS_BLOCK_GROUP_METADATA,
3027                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3028         int num_types = 4;
3029         int i;
3030         int c;
3031         int num_tolerated_disk_barrier_failures =
3032                 (int)fs_info->fs_devices->num_devices;
3033
3034         for (i = 0; i < num_types; i++) {
3035                 struct btrfs_space_info *tmp;
3036
3037                 sinfo = NULL;
3038                 rcu_read_lock();
3039                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3040                         if (tmp->flags == types[i]) {
3041                                 sinfo = tmp;
3042                                 break;
3043                         }
3044                 }
3045                 rcu_read_unlock();
3046
3047                 if (!sinfo)
3048                         continue;
3049
3050                 down_read(&sinfo->groups_sem);
3051                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3052                         if (!list_empty(&sinfo->block_groups[c])) {
3053                                 u64 flags;
3054
3055                                 btrfs_get_block_group_info(
3056                                         &sinfo->block_groups[c], &space);
3057                                 if (space.total_bytes == 0 ||
3058                                     space.used_bytes == 0)
3059                                         continue;
3060                                 flags = space.flags;
3061                                 /*
3062                                  * return
3063                                  * 0: if dup, single or RAID0 is configured for
3064                                  *    any of metadata, system or data, else
3065                                  * 1: if RAID5 is configured, or if RAID1 or
3066                                  *    RAID10 is configured and only two mirrors
3067                                  *    are used, else
3068                                  * 2: if RAID6 is configured, else
3069                                  * num_mirrors - 1: if RAID1 or RAID10 is
3070                                  *                  configured and more than
3071                                  *                  2 mirrors are used.
3072                                  */
3073                                 if (num_tolerated_disk_barrier_failures > 0 &&
3074                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3075                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3076                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3077                                       == 0)))
3078                                         num_tolerated_disk_barrier_failures = 0;
3079                                 else if (num_tolerated_disk_barrier_failures > 1
3080                                          &&
3081                                          (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3082                                                    BTRFS_BLOCK_GROUP_RAID10)))
3083                                         num_tolerated_disk_barrier_failures = 1;
3084                         }
3085                 }
3086                 up_read(&sinfo->groups_sem);
3087         }
3088
3089         return num_tolerated_disk_barrier_failures;
3090 }
3091
3092 int write_all_supers(struct btrfs_root *root, int max_mirrors)
3093 {
3094         struct list_head *head;
3095         struct btrfs_device *dev;
3096         struct btrfs_super_block *sb;
3097         struct btrfs_dev_item *dev_item;
3098         int ret;
3099         int do_barriers;
3100         int max_errors;
3101         int total_errors = 0;
3102         u64 flags;
3103
3104         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3105         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3106         backup_super_roots(root->fs_info);
3107
3108         sb = root->fs_info->super_for_commit;
3109         dev_item = &sb->dev_item;
3110
3111         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3112         head = &root->fs_info->fs_devices->devices;
3113
3114         if (do_barriers) {
3115                 ret = barrier_all_devices(root->fs_info);
3116                 if (ret) {
3117                         mutex_unlock(
3118                                 &root->fs_info->fs_devices->device_list_mutex);
3119                         btrfs_error(root->fs_info, ret,
3120                                     "errors while submitting device barriers.");
3121                         return ret;
3122                 }
3123         }
3124
3125         list_for_each_entry_rcu(dev, head, dev_list) {
3126                 if (!dev->bdev) {
3127                         total_errors++;
3128                         continue;
3129                 }
3130                 if (!dev->in_fs_metadata || !dev->writeable)
3131                         continue;
3132
3133                 btrfs_set_stack_device_generation(dev_item, 0);
3134                 btrfs_set_stack_device_type(dev_item, dev->type);
3135                 btrfs_set_stack_device_id(dev_item, dev->devid);
3136                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3137                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3138                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3139                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3140                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3141                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3142                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3143
3144                 flags = btrfs_super_flags(sb);
3145                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3146
3147                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3148                 if (ret)
3149                         total_errors++;
3150         }
3151         if (total_errors > max_errors) {
3152                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3153                        total_errors);
3154
3155                 /* This shouldn't happen. FUA is masked off if unsupported */
3156                 BUG();
3157         }
3158
3159         total_errors = 0;
3160         list_for_each_entry_rcu(dev, head, dev_list) {
3161                 if (!dev->bdev)
3162                         continue;
3163                 if (!dev->in_fs_metadata || !dev->writeable)
3164                         continue;
3165
3166                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3167                 if (ret)
3168                         total_errors++;
3169         }
3170         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3171         if (total_errors > max_errors) {
3172                 btrfs_error(root->fs_info, -EIO,
3173                             "%d errors while writing supers", total_errors);
3174                 return -EIO;
3175         }
3176         return 0;
3177 }
3178
3179 int write_ctree_super(struct btrfs_trans_handle *trans,
3180                       struct btrfs_root *root, int max_mirrors)
3181 {
3182         int ret;
3183
3184         ret = write_all_supers(root, max_mirrors);
3185         return ret;
3186 }
3187
3188 void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3189 {
3190         spin_lock(&fs_info->fs_roots_radix_lock);
3191         radix_tree_delete(&fs_info->fs_roots_radix,
3192                           (unsigned long)root->root_key.objectid);
3193         spin_unlock(&fs_info->fs_roots_radix_lock);
3194
3195         if (btrfs_root_refs(&root->root_item) == 0)
3196                 synchronize_srcu(&fs_info->subvol_srcu);
3197
3198         __btrfs_remove_free_space_cache(root->free_ino_pinned);
3199         __btrfs_remove_free_space_cache(root->free_ino_ctl);
3200         free_fs_root(root);
3201 }
3202
3203 static void free_fs_root(struct btrfs_root *root)
3204 {
3205         iput(root->cache_inode);
3206         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3207         if (root->anon_dev)
3208                 free_anon_bdev(root->anon_dev);
3209         free_extent_buffer(root->node);
3210         free_extent_buffer(root->commit_root);
3211         kfree(root->free_ino_ctl);
3212         kfree(root->free_ino_pinned);
3213         kfree(root->name);
3214         kfree(root);
3215 }
3216
3217 static void del_fs_roots(struct btrfs_fs_info *fs_info)
3218 {
3219         int ret;
3220         struct btrfs_root *gang[8];
3221         int i;
3222
3223         while (!list_empty(&fs_info->dead_roots)) {
3224                 gang[0] = list_entry(fs_info->dead_roots.next,
3225                                      struct btrfs_root, root_list);
3226                 list_del(&gang[0]->root_list);
3227
3228                 if (gang[0]->in_radix) {
3229                         btrfs_free_fs_root(fs_info, gang[0]);
3230                 } else {
3231                         free_extent_buffer(gang[0]->node);
3232                         free_extent_buffer(gang[0]->commit_root);
3233                         kfree(gang[0]);
3234                 }
3235         }
3236
3237         while (1) {
3238                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3239                                              (void **)gang, 0,
3240                                              ARRAY_SIZE(gang));
3241                 if (!ret)
3242                         break;
3243                 for (i = 0; i < ret; i++)
3244                         btrfs_free_fs_root(fs_info, gang[i]);
3245         }
3246 }
3247
3248 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3249 {
3250         u64 root_objectid = 0;
3251         struct btrfs_root *gang[8];
3252         int i;
3253         int ret;
3254
3255         while (1) {
3256                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3257                                              (void **)gang, root_objectid,
3258                                              ARRAY_SIZE(gang));
3259                 if (!ret)
3260                         break;
3261
3262                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3263                 for (i = 0; i < ret; i++) {
3264                         int err;
3265
3266                         root_objectid = gang[i]->root_key.objectid;
3267                         err = btrfs_orphan_cleanup(gang[i]);
3268                         if (err)
3269                                 return err;
3270                 }
3271                 root_objectid++;
3272         }
3273         return 0;
3274 }
3275
3276 int btrfs_commit_super(struct btrfs_root *root)
3277 {
3278         struct btrfs_trans_handle *trans;
3279         int ret;
3280
3281         mutex_lock(&root->fs_info->cleaner_mutex);
3282         btrfs_run_delayed_iputs(root);
3283         btrfs_clean_old_snapshots(root);
3284         mutex_unlock(&root->fs_info->cleaner_mutex);
3285
3286         /* wait until ongoing cleanup work done */
3287         down_write(&root->fs_info->cleanup_work_sem);
3288         up_write(&root->fs_info->cleanup_work_sem);
3289
3290         trans = btrfs_join_transaction(root);
3291         if (IS_ERR(trans))
3292                 return PTR_ERR(trans);
3293         ret = btrfs_commit_transaction(trans, root);
3294         if (ret)
3295                 return ret;
3296         /* run commit again to drop the original snapshot */
3297         trans = btrfs_join_transaction(root);
3298         if (IS_ERR(trans))
3299                 return PTR_ERR(trans);
3300         ret = btrfs_commit_transaction(trans, root);
3301         if (ret)
3302                 return ret;
3303         ret = btrfs_write_and_wait_transaction(NULL, root);
3304         if (ret) {
3305                 btrfs_error(root->fs_info, ret,
3306                             "Failed to sync btree inode to disk.");
3307                 return ret;
3308         }
3309
3310         ret = write_ctree_super(NULL, root, 0);
3311         return ret;
3312 }
3313
3314 int close_ctree(struct btrfs_root *root)
3315 {
3316         struct btrfs_fs_info *fs_info = root->fs_info;
3317         int ret;
3318
3319         fs_info->closing = 1;
3320         smp_mb();
3321
3322         /* pause restriper - we want to resume on mount */
3323         btrfs_pause_balance(fs_info);
3324
3325         btrfs_dev_replace_suspend_for_unmount(fs_info);