btrfs: remove max_active var from open_ctree
[sfrench/cifs-2.6.git] / fs / btrfs / disk-io.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <linux/error-injection.h>
34 #include <linux/crc32c.h>
35 #include <asm/unaligned.h>
36 #include "ctree.h"
37 #include "disk-io.h"
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "volumes.h"
41 #include "print-tree.h"
42 #include "locking.h"
43 #include "tree-log.h"
44 #include "free-space-cache.h"
45 #include "free-space-tree.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
50 #include "raid56.h"
51 #include "sysfs.h"
52 #include "qgroup.h"
53 #include "compression.h"
54 #include "tree-checker.h"
55 #include "ref-verify.h"
56
57 #ifdef CONFIG_X86
58 #include <asm/cpufeature.h>
59 #endif
60
61 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
62                                  BTRFS_HEADER_FLAG_RELOC |\
63                                  BTRFS_SUPER_FLAG_ERROR |\
64                                  BTRFS_SUPER_FLAG_SEEDING |\
65                                  BTRFS_SUPER_FLAG_METADUMP |\
66                                  BTRFS_SUPER_FLAG_METADUMP_V2)
67
68 static const struct extent_io_ops btree_extent_io_ops;
69 static void end_workqueue_fn(struct btrfs_work *work);
70 static void free_fs_root(struct btrfs_root *root);
71 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info);
72 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
73 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
74                                       struct btrfs_fs_info *fs_info);
75 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
76 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
77                                         struct extent_io_tree *dirty_pages,
78                                         int mark);
79 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
80                                        struct extent_io_tree *pinned_extents);
81 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
82 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
83
84 /*
85  * btrfs_end_io_wq structs are used to do processing in task context when an IO
86  * is complete.  This is used during reads to verify checksums, and it is used
87  * by writes to insert metadata for new file extents after IO is complete.
88  */
89 struct btrfs_end_io_wq {
90         struct bio *bio;
91         bio_end_io_t *end_io;
92         void *private;
93         struct btrfs_fs_info *info;
94         blk_status_t status;
95         enum btrfs_wq_endio_type metadata;
96         struct btrfs_work work;
97 };
98
99 static struct kmem_cache *btrfs_end_io_wq_cache;
100
101 int __init btrfs_end_io_wq_init(void)
102 {
103         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
104                                         sizeof(struct btrfs_end_io_wq),
105                                         0,
106                                         SLAB_MEM_SPREAD,
107                                         NULL);
108         if (!btrfs_end_io_wq_cache)
109                 return -ENOMEM;
110         return 0;
111 }
112
113 void __cold btrfs_end_io_wq_exit(void)
114 {
115         kmem_cache_destroy(btrfs_end_io_wq_cache);
116 }
117
118 /*
119  * async submit bios are used to offload expensive checksumming
120  * onto the worker threads.  They checksum file and metadata bios
121  * just before they are sent down the IO stack.
122  */
123 struct async_submit_bio {
124         void *private_data;
125         struct btrfs_fs_info *fs_info;
126         struct bio *bio;
127         extent_submit_bio_start_t *submit_bio_start;
128         extent_submit_bio_done_t *submit_bio_done;
129         int mirror_num;
130         unsigned long bio_flags;
131         /*
132          * bio_offset is optional, can be used if the pages in the bio
133          * can't tell us where in the file the bio should go
134          */
135         u64 bio_offset;
136         struct btrfs_work work;
137         blk_status_t status;
138 };
139
140 /*
141  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
142  * eb, the lockdep key is determined by the btrfs_root it belongs to and
143  * the level the eb occupies in the tree.
144  *
145  * Different roots are used for different purposes and may nest inside each
146  * other and they require separate keysets.  As lockdep keys should be
147  * static, assign keysets according to the purpose of the root as indicated
148  * by btrfs_root->objectid.  This ensures that all special purpose roots
149  * have separate keysets.
150  *
151  * Lock-nesting across peer nodes is always done with the immediate parent
152  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
153  * subclass to avoid triggering lockdep warning in such cases.
154  *
155  * The key is set by the readpage_end_io_hook after the buffer has passed
156  * csum validation but before the pages are unlocked.  It is also set by
157  * btrfs_init_new_buffer on freshly allocated blocks.
158  *
159  * We also add a check to make sure the highest level of the tree is the
160  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
161  * needs update as well.
162  */
163 #ifdef CONFIG_DEBUG_LOCK_ALLOC
164 # if BTRFS_MAX_LEVEL != 8
165 #  error
166 # endif
167
168 static struct btrfs_lockdep_keyset {
169         u64                     id;             /* root objectid */
170         const char              *name_stem;     /* lock name stem */
171         char                    names[BTRFS_MAX_LEVEL + 1][20];
172         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
173 } btrfs_lockdep_keysets[] = {
174         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
175         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
176         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
177         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
178         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
179         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
180         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
181         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
182         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
183         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
184         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
185         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
186         { .id = 0,                              .name_stem = "tree"     },
187 };
188
189 void __init btrfs_init_lockdep(void)
190 {
191         int i, j;
192
193         /* initialize lockdep class names */
194         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
195                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
196
197                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
198                         snprintf(ks->names[j], sizeof(ks->names[j]),
199                                  "btrfs-%s-%02d", ks->name_stem, j);
200         }
201 }
202
203 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
204                                     int level)
205 {
206         struct btrfs_lockdep_keyset *ks;
207
208         BUG_ON(level >= ARRAY_SIZE(ks->keys));
209
210         /* find the matching keyset, id 0 is the default entry */
211         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
212                 if (ks->id == objectid)
213                         break;
214
215         lockdep_set_class_and_name(&eb->lock,
216                                    &ks->keys[level], ks->names[level]);
217 }
218
219 #endif
220
221 /*
222  * extents on the btree inode are pretty simple, there's one extent
223  * that covers the entire device
224  */
225 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
226                 struct page *page, size_t pg_offset, u64 start, u64 len,
227                 int create)
228 {
229         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
230         struct extent_map_tree *em_tree = &inode->extent_tree;
231         struct extent_map *em;
232         int ret;
233
234         read_lock(&em_tree->lock);
235         em = lookup_extent_mapping(em_tree, start, len);
236         if (em) {
237                 em->bdev = fs_info->fs_devices->latest_bdev;
238                 read_unlock(&em_tree->lock);
239                 goto out;
240         }
241         read_unlock(&em_tree->lock);
242
243         em = alloc_extent_map();
244         if (!em) {
245                 em = ERR_PTR(-ENOMEM);
246                 goto out;
247         }
248         em->start = 0;
249         em->len = (u64)-1;
250         em->block_len = (u64)-1;
251         em->block_start = 0;
252         em->bdev = fs_info->fs_devices->latest_bdev;
253
254         write_lock(&em_tree->lock);
255         ret = add_extent_mapping(em_tree, em, 0);
256         if (ret == -EEXIST) {
257                 free_extent_map(em);
258                 em = lookup_extent_mapping(em_tree, start, len);
259                 if (!em)
260                         em = ERR_PTR(-EIO);
261         } else if (ret) {
262                 free_extent_map(em);
263                 em = ERR_PTR(ret);
264         }
265         write_unlock(&em_tree->lock);
266
267 out:
268         return em;
269 }
270
271 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
272 {
273         return crc32c(seed, data, len);
274 }
275
276 void btrfs_csum_final(u32 crc, u8 *result)
277 {
278         put_unaligned_le32(~crc, result);
279 }
280
281 /*
282  * compute the csum for a btree block, and either verify it or write it
283  * into the csum field of the block.
284  */
285 static int csum_tree_block(struct btrfs_fs_info *fs_info,
286                            struct extent_buffer *buf,
287                            int verify)
288 {
289         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
290         char result[BTRFS_CSUM_SIZE];
291         unsigned long len;
292         unsigned long cur_len;
293         unsigned long offset = BTRFS_CSUM_SIZE;
294         char *kaddr;
295         unsigned long map_start;
296         unsigned long map_len;
297         int err;
298         u32 crc = ~(u32)0;
299
300         len = buf->len - offset;
301         while (len > 0) {
302                 err = map_private_extent_buffer(buf, offset, 32,
303                                         &kaddr, &map_start, &map_len);
304                 if (err)
305                         return err;
306                 cur_len = min(len, map_len - (offset - map_start));
307                 crc = btrfs_csum_data(kaddr + offset - map_start,
308                                       crc, cur_len);
309                 len -= cur_len;
310                 offset += cur_len;
311         }
312         memset(result, 0, BTRFS_CSUM_SIZE);
313
314         btrfs_csum_final(crc, result);
315
316         if (verify) {
317                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
318                         u32 val;
319                         u32 found = 0;
320                         memcpy(&found, result, csum_size);
321
322                         read_extent_buffer(buf, &val, 0, csum_size);
323                         btrfs_warn_rl(fs_info,
324                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
325                                 fs_info->sb->s_id, buf->start,
326                                 val, found, btrfs_header_level(buf));
327                         return -EUCLEAN;
328                 }
329         } else {
330                 write_extent_buffer(buf, result, 0, csum_size);
331         }
332
333         return 0;
334 }
335
336 /*
337  * we can't consider a given block up to date unless the transid of the
338  * block matches the transid in the parent node's pointer.  This is how we
339  * detect blocks that either didn't get written at all or got written
340  * in the wrong place.
341  */
342 static int verify_parent_transid(struct extent_io_tree *io_tree,
343                                  struct extent_buffer *eb, u64 parent_transid,
344                                  int atomic)
345 {
346         struct extent_state *cached_state = NULL;
347         int ret;
348         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
349
350         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
351                 return 0;
352
353         if (atomic)
354                 return -EAGAIN;
355
356         if (need_lock) {
357                 btrfs_tree_read_lock(eb);
358                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
359         }
360
361         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
362                          &cached_state);
363         if (extent_buffer_uptodate(eb) &&
364             btrfs_header_generation(eb) == parent_transid) {
365                 ret = 0;
366                 goto out;
367         }
368         btrfs_err_rl(eb->fs_info,
369                 "parent transid verify failed on %llu wanted %llu found %llu",
370                         eb->start,
371                         parent_transid, btrfs_header_generation(eb));
372         ret = 1;
373
374         /*
375          * Things reading via commit roots that don't have normal protection,
376          * like send, can have a really old block in cache that may point at a
377          * block that has been freed and re-allocated.  So don't clear uptodate
378          * if we find an eb that is under IO (dirty/writeback) because we could
379          * end up reading in the stale data and then writing it back out and
380          * making everybody very sad.
381          */
382         if (!extent_buffer_under_io(eb))
383                 clear_extent_buffer_uptodate(eb);
384 out:
385         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
386                              &cached_state);
387         if (need_lock)
388                 btrfs_tree_read_unlock_blocking(eb);
389         return ret;
390 }
391
392 /*
393  * Return 0 if the superblock checksum type matches the checksum value of that
394  * algorithm. Pass the raw disk superblock data.
395  */
396 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
397                                   char *raw_disk_sb)
398 {
399         struct btrfs_super_block *disk_sb =
400                 (struct btrfs_super_block *)raw_disk_sb;
401         u16 csum_type = btrfs_super_csum_type(disk_sb);
402         int ret = 0;
403
404         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
405                 u32 crc = ~(u32)0;
406                 char result[sizeof(crc)];
407
408                 /*
409                  * The super_block structure does not span the whole
410                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
411                  * is filled with zeros and is included in the checksum.
412                  */
413                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
414                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
415                 btrfs_csum_final(crc, result);
416
417                 if (memcmp(raw_disk_sb, result, sizeof(result)))
418                         ret = 1;
419         }
420
421         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
422                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
423                                 csum_type);
424                 ret = 1;
425         }
426
427         return ret;
428 }
429
430 /*
431  * helper to read a given tree block, doing retries as required when
432  * the checksums don't match and we have alternate mirrors to try.
433  */
434 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
435                                           struct extent_buffer *eb,
436                                           u64 parent_transid)
437 {
438         struct extent_io_tree *io_tree;
439         int failed = 0;
440         int ret;
441         int num_copies = 0;
442         int mirror_num = 0;
443         int failed_mirror = 0;
444
445         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
446         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
447         while (1) {
448                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
449                                                mirror_num);
450                 if (!ret) {
451                         if (!verify_parent_transid(io_tree, eb,
452                                                    parent_transid, 0))
453                                 break;
454                         else
455                                 ret = -EIO;
456                 }
457
458                 /*
459                  * This buffer's crc is fine, but its contents are corrupted, so
460                  * there is no reason to read the other copies, they won't be
461                  * any less wrong.
462                  */
463                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
464                         break;
465
466                 num_copies = btrfs_num_copies(fs_info,
467                                               eb->start, eb->len);
468                 if (num_copies == 1)
469                         break;
470
471                 if (!failed_mirror) {
472                         failed = 1;
473                         failed_mirror = eb->read_mirror;
474                 }
475
476                 mirror_num++;
477                 if (mirror_num == failed_mirror)
478                         mirror_num++;
479
480                 if (mirror_num > num_copies)
481                         break;
482         }
483
484         if (failed && !ret && failed_mirror)
485                 repair_eb_io_failure(fs_info, eb, failed_mirror);
486
487         return ret;
488 }
489
490 /*
491  * checksum a dirty tree block before IO.  This has extra checks to make sure
492  * we only fill in the checksum field in the first page of a multi-page block
493  */
494
495 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
496 {
497         u64 start = page_offset(page);
498         u64 found_start;
499         struct extent_buffer *eb;
500
501         eb = (struct extent_buffer *)page->private;
502         if (page != eb->pages[0])
503                 return 0;
504
505         found_start = btrfs_header_bytenr(eb);
506         /*
507          * Please do not consolidate these warnings into a single if.
508          * It is useful to know what went wrong.
509          */
510         if (WARN_ON(found_start != start))
511                 return -EUCLEAN;
512         if (WARN_ON(!PageUptodate(page)))
513                 return -EUCLEAN;
514
515         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
516                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
517
518         return csum_tree_block(fs_info, eb, 0);
519 }
520
521 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
522                                  struct extent_buffer *eb)
523 {
524         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
525         u8 fsid[BTRFS_FSID_SIZE];
526         int ret = 1;
527
528         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
529         while (fs_devices) {
530                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
531                         ret = 0;
532                         break;
533                 }
534                 fs_devices = fs_devices->seed;
535         }
536         return ret;
537 }
538
539 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
540                                       u64 phy_offset, struct page *page,
541                                       u64 start, u64 end, int mirror)
542 {
543         u64 found_start;
544         int found_level;
545         struct extent_buffer *eb;
546         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
547         struct btrfs_fs_info *fs_info = root->fs_info;
548         int ret = 0;
549         int reads_done;
550
551         if (!page->private)
552                 goto out;
553
554         eb = (struct extent_buffer *)page->private;
555
556         /* the pending IO might have been the only thing that kept this buffer
557          * in memory.  Make sure we have a ref for all this other checks
558          */
559         extent_buffer_get(eb);
560
561         reads_done = atomic_dec_and_test(&eb->io_pages);
562         if (!reads_done)
563                 goto err;
564
565         eb->read_mirror = mirror;
566         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
567                 ret = -EIO;
568                 goto err;
569         }
570
571         found_start = btrfs_header_bytenr(eb);
572         if (found_start != eb->start) {
573                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
574                              found_start, eb->start);
575                 ret = -EIO;
576                 goto err;
577         }
578         if (check_tree_block_fsid(fs_info, eb)) {
579                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
580                              eb->start);
581                 ret = -EIO;
582                 goto err;
583         }
584         found_level = btrfs_header_level(eb);
585         if (found_level >= BTRFS_MAX_LEVEL) {
586                 btrfs_err(fs_info, "bad tree block level %d",
587                           (int)btrfs_header_level(eb));
588                 ret = -EIO;
589                 goto err;
590         }
591
592         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
593                                        eb, found_level);
594
595         ret = csum_tree_block(fs_info, eb, 1);
596         if (ret)
597                 goto err;
598
599         /*
600          * If this is a leaf block and it is corrupt, set the corrupt bit so
601          * that we don't try and read the other copies of this block, just
602          * return -EIO.
603          */
604         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
605                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
606                 ret = -EIO;
607         }
608
609         if (found_level > 0 && btrfs_check_node(fs_info, eb))
610                 ret = -EIO;
611
612         if (!ret)
613                 set_extent_buffer_uptodate(eb);
614 err:
615         if (reads_done &&
616             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
617                 btree_readahead_hook(eb, ret);
618
619         if (ret) {
620                 /*
621                  * our io error hook is going to dec the io pages
622                  * again, we have to make sure it has something
623                  * to decrement
624                  */
625                 atomic_inc(&eb->io_pages);
626                 clear_extent_buffer_uptodate(eb);
627         }
628         free_extent_buffer(eb);
629 out:
630         return ret;
631 }
632
633 static int btree_io_failed_hook(struct page *page, int failed_mirror)
634 {
635         struct extent_buffer *eb;
636
637         eb = (struct extent_buffer *)page->private;
638         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
639         eb->read_mirror = failed_mirror;
640         atomic_dec(&eb->io_pages);
641         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
642                 btree_readahead_hook(eb, -EIO);
643         return -EIO;    /* we fixed nothing */
644 }
645
646 static void end_workqueue_bio(struct bio *bio)
647 {
648         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
649         struct btrfs_fs_info *fs_info;
650         struct btrfs_workqueue *wq;
651         btrfs_work_func_t func;
652
653         fs_info = end_io_wq->info;
654         end_io_wq->status = bio->bi_status;
655
656         if (bio_op(bio) == REQ_OP_WRITE) {
657                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
658                         wq = fs_info->endio_meta_write_workers;
659                         func = btrfs_endio_meta_write_helper;
660                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
661                         wq = fs_info->endio_freespace_worker;
662                         func = btrfs_freespace_write_helper;
663                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
664                         wq = fs_info->endio_raid56_workers;
665                         func = btrfs_endio_raid56_helper;
666                 } else {
667                         wq = fs_info->endio_write_workers;
668                         func = btrfs_endio_write_helper;
669                 }
670         } else {
671                 if (unlikely(end_io_wq->metadata ==
672                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
673                         wq = fs_info->endio_repair_workers;
674                         func = btrfs_endio_repair_helper;
675                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
676                         wq = fs_info->endio_raid56_workers;
677                         func = btrfs_endio_raid56_helper;
678                 } else if (end_io_wq->metadata) {
679                         wq = fs_info->endio_meta_workers;
680                         func = btrfs_endio_meta_helper;
681                 } else {
682                         wq = fs_info->endio_workers;
683                         func = btrfs_endio_helper;
684                 }
685         }
686
687         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
688         btrfs_queue_work(wq, &end_io_wq->work);
689 }
690
691 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
692                         enum btrfs_wq_endio_type metadata)
693 {
694         struct btrfs_end_io_wq *end_io_wq;
695
696         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
697         if (!end_io_wq)
698                 return BLK_STS_RESOURCE;
699
700         end_io_wq->private = bio->bi_private;
701         end_io_wq->end_io = bio->bi_end_io;
702         end_io_wq->info = info;
703         end_io_wq->status = 0;
704         end_io_wq->bio = bio;
705         end_io_wq->metadata = metadata;
706
707         bio->bi_private = end_io_wq;
708         bio->bi_end_io = end_workqueue_bio;
709         return 0;
710 }
711
712 static void run_one_async_start(struct btrfs_work *work)
713 {
714         struct async_submit_bio *async;
715         blk_status_t ret;
716
717         async = container_of(work, struct  async_submit_bio, work);
718         ret = async->submit_bio_start(async->private_data, async->bio,
719                                       async->bio_offset);
720         if (ret)
721                 async->status = ret;
722 }
723
724 static void run_one_async_done(struct btrfs_work *work)
725 {
726         struct async_submit_bio *async;
727
728         async = container_of(work, struct  async_submit_bio, work);
729
730         /* If an error occurred we just want to clean up the bio and move on */
731         if (async->status) {
732                 async->bio->bi_status = async->status;
733                 bio_endio(async->bio);
734                 return;
735         }
736
737         async->submit_bio_done(async->private_data, async->bio, async->mirror_num);
738 }
739
740 static void run_one_async_free(struct btrfs_work *work)
741 {
742         struct async_submit_bio *async;
743
744         async = container_of(work, struct  async_submit_bio, work);
745         kfree(async);
746 }
747
748 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
749                                  int mirror_num, unsigned long bio_flags,
750                                  u64 bio_offset, void *private_data,
751                                  extent_submit_bio_start_t *submit_bio_start,
752                                  extent_submit_bio_done_t *submit_bio_done)
753 {
754         struct async_submit_bio *async;
755
756         async = kmalloc(sizeof(*async), GFP_NOFS);
757         if (!async)
758                 return BLK_STS_RESOURCE;
759
760         async->private_data = private_data;
761         async->fs_info = fs_info;
762         async->bio = bio;
763         async->mirror_num = mirror_num;
764         async->submit_bio_start = submit_bio_start;
765         async->submit_bio_done = submit_bio_done;
766
767         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
768                         run_one_async_done, run_one_async_free);
769
770         async->bio_flags = bio_flags;
771         async->bio_offset = bio_offset;
772
773         async->status = 0;
774
775         if (op_is_sync(bio->bi_opf))
776                 btrfs_set_work_high_priority(&async->work);
777
778         btrfs_queue_work(fs_info->workers, &async->work);
779         return 0;
780 }
781
782 static blk_status_t btree_csum_one_bio(struct bio *bio)
783 {
784         struct bio_vec *bvec;
785         struct btrfs_root *root;
786         int i, ret = 0;
787
788         ASSERT(!bio_flagged(bio, BIO_CLONED));
789         bio_for_each_segment_all(bvec, bio, i) {
790                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
791                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
792                 if (ret)
793                         break;
794         }
795
796         return errno_to_blk_status(ret);
797 }
798
799 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
800                                              u64 bio_offset)
801 {
802         /*
803          * when we're called for a write, we're already in the async
804          * submission context.  Just jump into btrfs_map_bio
805          */
806         return btree_csum_one_bio(bio);
807 }
808
809 static blk_status_t btree_submit_bio_done(void *private_data, struct bio *bio,
810                                             int mirror_num)
811 {
812         struct inode *inode = private_data;
813         blk_status_t ret;
814
815         /*
816          * when we're called for a write, we're already in the async
817          * submission context.  Just jump into btrfs_map_bio
818          */
819         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
820         if (ret) {
821                 bio->bi_status = ret;
822                 bio_endio(bio);
823         }
824         return ret;
825 }
826
827 static int check_async_write(struct btrfs_inode *bi)
828 {
829         if (atomic_read(&bi->sync_writers))
830                 return 0;
831 #ifdef CONFIG_X86
832         if (static_cpu_has(X86_FEATURE_XMM4_2))
833                 return 0;
834 #endif
835         return 1;
836 }
837
838 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
839                                           int mirror_num, unsigned long bio_flags,
840                                           u64 bio_offset)
841 {
842         struct inode *inode = private_data;
843         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
844         int async = check_async_write(BTRFS_I(inode));
845         blk_status_t ret;
846
847         if (bio_op(bio) != REQ_OP_WRITE) {
848                 /*
849                  * called for a read, do the setup so that checksum validation
850                  * can happen in the async kernel threads
851                  */
852                 ret = btrfs_bio_wq_end_io(fs_info, bio,
853                                           BTRFS_WQ_ENDIO_METADATA);
854                 if (ret)
855                         goto out_w_error;
856                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
857         } else if (!async) {
858                 ret = btree_csum_one_bio(bio);
859                 if (ret)
860                         goto out_w_error;
861                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
862         } else {
863                 /*
864                  * kthread helpers are used to submit writes so that
865                  * checksumming can happen in parallel across all CPUs
866                  */
867                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
868                                           bio_offset, private_data,
869                                           btree_submit_bio_start,
870                                           btree_submit_bio_done);
871         }
872
873         if (ret)
874                 goto out_w_error;
875         return 0;
876
877 out_w_error:
878         bio->bi_status = ret;
879         bio_endio(bio);
880         return ret;
881 }
882
883 #ifdef CONFIG_MIGRATION
884 static int btree_migratepage(struct address_space *mapping,
885                         struct page *newpage, struct page *page,
886                         enum migrate_mode mode)
887 {
888         /*
889          * we can't safely write a btree page from here,
890          * we haven't done the locking hook
891          */
892         if (PageDirty(page))
893                 return -EAGAIN;
894         /*
895          * Buffers may be managed in a filesystem specific way.
896          * We must have no buffers or drop them.
897          */
898         if (page_has_private(page) &&
899             !try_to_release_page(page, GFP_KERNEL))
900                 return -EAGAIN;
901         return migrate_page(mapping, newpage, page, mode);
902 }
903 #endif
904
905
906 static int btree_writepages(struct address_space *mapping,
907                             struct writeback_control *wbc)
908 {
909         struct btrfs_fs_info *fs_info;
910         int ret;
911
912         if (wbc->sync_mode == WB_SYNC_NONE) {
913
914                 if (wbc->for_kupdate)
915                         return 0;
916
917                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
918                 /* this is a bit racy, but that's ok */
919                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
920                                              BTRFS_DIRTY_METADATA_THRESH);
921                 if (ret < 0)
922                         return 0;
923         }
924         return btree_write_cache_pages(mapping, wbc);
925 }
926
927 static int btree_readpage(struct file *file, struct page *page)
928 {
929         struct extent_io_tree *tree;
930         tree = &BTRFS_I(page->mapping->host)->io_tree;
931         return extent_read_full_page(tree, page, btree_get_extent, 0);
932 }
933
934 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
935 {
936         if (PageWriteback(page) || PageDirty(page))
937                 return 0;
938
939         return try_release_extent_buffer(page);
940 }
941
942 static void btree_invalidatepage(struct page *page, unsigned int offset,
943                                  unsigned int length)
944 {
945         struct extent_io_tree *tree;
946         tree = &BTRFS_I(page->mapping->host)->io_tree;
947         extent_invalidatepage(tree, page, offset);
948         btree_releasepage(page, GFP_NOFS);
949         if (PagePrivate(page)) {
950                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
951                            "page private not zero on page %llu",
952                            (unsigned long long)page_offset(page));
953                 ClearPagePrivate(page);
954                 set_page_private(page, 0);
955                 put_page(page);
956         }
957 }
958
959 static int btree_set_page_dirty(struct page *page)
960 {
961 #ifdef DEBUG
962         struct extent_buffer *eb;
963
964         BUG_ON(!PagePrivate(page));
965         eb = (struct extent_buffer *)page->private;
966         BUG_ON(!eb);
967         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
968         BUG_ON(!atomic_read(&eb->refs));
969         btrfs_assert_tree_locked(eb);
970 #endif
971         return __set_page_dirty_nobuffers(page);
972 }
973
974 static const struct address_space_operations btree_aops = {
975         .readpage       = btree_readpage,
976         .writepages     = btree_writepages,
977         .releasepage    = btree_releasepage,
978         .invalidatepage = btree_invalidatepage,
979 #ifdef CONFIG_MIGRATION
980         .migratepage    = btree_migratepage,
981 #endif
982         .set_page_dirty = btree_set_page_dirty,
983 };
984
985 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
986 {
987         struct extent_buffer *buf = NULL;
988         struct inode *btree_inode = fs_info->btree_inode;
989
990         buf = btrfs_find_create_tree_block(fs_info, bytenr);
991         if (IS_ERR(buf))
992                 return;
993         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
994                                  buf, WAIT_NONE, 0);
995         free_extent_buffer(buf);
996 }
997
998 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
999                          int mirror_num, struct extent_buffer **eb)
1000 {
1001         struct extent_buffer *buf = NULL;
1002         struct inode *btree_inode = fs_info->btree_inode;
1003         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1004         int ret;
1005
1006         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1007         if (IS_ERR(buf))
1008                 return 0;
1009
1010         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1011
1012         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1013                                        mirror_num);
1014         if (ret) {
1015                 free_extent_buffer(buf);
1016                 return ret;
1017         }
1018
1019         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1020                 free_extent_buffer(buf);
1021                 return -EIO;
1022         } else if (extent_buffer_uptodate(buf)) {
1023                 *eb = buf;
1024         } else {
1025                 free_extent_buffer(buf);
1026         }
1027         return 0;
1028 }
1029
1030 struct extent_buffer *btrfs_find_create_tree_block(
1031                                                 struct btrfs_fs_info *fs_info,
1032                                                 u64 bytenr)
1033 {
1034         if (btrfs_is_testing(fs_info))
1035                 return alloc_test_extent_buffer(fs_info, bytenr);
1036         return alloc_extent_buffer(fs_info, bytenr);
1037 }
1038
1039
1040 int btrfs_write_tree_block(struct extent_buffer *buf)
1041 {
1042         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1043                                         buf->start + buf->len - 1);
1044 }
1045
1046 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1047 {
1048         filemap_fdatawait_range(buf->pages[0]->mapping,
1049                                 buf->start, buf->start + buf->len - 1);
1050 }
1051
1052 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1053                                       u64 parent_transid)
1054 {
1055         struct extent_buffer *buf = NULL;
1056         int ret;
1057
1058         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1059         if (IS_ERR(buf))
1060                 return buf;
1061
1062         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1063         if (ret) {
1064                 free_extent_buffer(buf);
1065                 return ERR_PTR(ret);
1066         }
1067         return buf;
1068
1069 }
1070
1071 void clean_tree_block(struct btrfs_fs_info *fs_info,
1072                       struct extent_buffer *buf)
1073 {
1074         if (btrfs_header_generation(buf) ==
1075             fs_info->running_transaction->transid) {
1076                 btrfs_assert_tree_locked(buf);
1077
1078                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1079                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1080                                                  -buf->len,
1081                                                  fs_info->dirty_metadata_batch);
1082                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1083                         btrfs_set_lock_blocking(buf);
1084                         clear_extent_buffer_dirty(buf);
1085                 }
1086         }
1087 }
1088
1089 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1090 {
1091         struct btrfs_subvolume_writers *writers;
1092         int ret;
1093
1094         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1095         if (!writers)
1096                 return ERR_PTR(-ENOMEM);
1097
1098         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1099         if (ret < 0) {
1100                 kfree(writers);
1101                 return ERR_PTR(ret);
1102         }
1103
1104         init_waitqueue_head(&writers->wait);
1105         return writers;
1106 }
1107
1108 static void
1109 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1110 {
1111         percpu_counter_destroy(&writers->counter);
1112         kfree(writers);
1113 }
1114
1115 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1116                          u64 objectid)
1117 {
1118         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1119         root->node = NULL;
1120         root->commit_root = NULL;
1121         root->state = 0;
1122         root->orphan_cleanup_state = 0;
1123
1124         root->objectid = objectid;
1125         root->last_trans = 0;
1126         root->highest_objectid = 0;
1127         root->nr_delalloc_inodes = 0;
1128         root->nr_ordered_extents = 0;
1129         root->name = NULL;
1130         root->inode_tree = RB_ROOT;
1131         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1132         root->block_rsv = NULL;
1133         root->orphan_block_rsv = NULL;
1134
1135         INIT_LIST_HEAD(&root->dirty_list);
1136         INIT_LIST_HEAD(&root->root_list);
1137         INIT_LIST_HEAD(&root->delalloc_inodes);
1138         INIT_LIST_HEAD(&root->delalloc_root);
1139         INIT_LIST_HEAD(&root->ordered_extents);
1140         INIT_LIST_HEAD(&root->ordered_root);
1141         INIT_LIST_HEAD(&root->logged_list[0]);
1142         INIT_LIST_HEAD(&root->logged_list[1]);
1143         spin_lock_init(&root->orphan_lock);
1144         spin_lock_init(&root->inode_lock);
1145         spin_lock_init(&root->delalloc_lock);
1146         spin_lock_init(&root->ordered_extent_lock);
1147         spin_lock_init(&root->accounting_lock);
1148         spin_lock_init(&root->log_extents_lock[0]);
1149         spin_lock_init(&root->log_extents_lock[1]);
1150         mutex_init(&root->objectid_mutex);
1151         mutex_init(&root->log_mutex);
1152         mutex_init(&root->ordered_extent_mutex);
1153         mutex_init(&root->delalloc_mutex);
1154         init_waitqueue_head(&root->log_writer_wait);
1155         init_waitqueue_head(&root->log_commit_wait[0]);
1156         init_waitqueue_head(&root->log_commit_wait[1]);
1157         INIT_LIST_HEAD(&root->log_ctxs[0]);
1158         INIT_LIST_HEAD(&root->log_ctxs[1]);
1159         atomic_set(&root->log_commit[0], 0);
1160         atomic_set(&root->log_commit[1], 0);
1161         atomic_set(&root->log_writers, 0);
1162         atomic_set(&root->log_batch, 0);
1163         atomic_set(&root->orphan_inodes, 0);
1164         refcount_set(&root->refs, 1);
1165         atomic_set(&root->will_be_snapshotted, 0);
1166         atomic64_set(&root->qgroup_meta_rsv, 0);
1167         root->log_transid = 0;
1168         root->log_transid_committed = -1;
1169         root->last_log_commit = 0;
1170         if (!dummy)
1171                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1172
1173         memset(&root->root_key, 0, sizeof(root->root_key));
1174         memset(&root->root_item, 0, sizeof(root->root_item));
1175         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1176         if (!dummy)
1177                 root->defrag_trans_start = fs_info->generation;
1178         else
1179                 root->defrag_trans_start = 0;
1180         root->root_key.objectid = objectid;
1181         root->anon_dev = 0;
1182
1183         spin_lock_init(&root->root_item_lock);
1184 }
1185
1186 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1187                 gfp_t flags)
1188 {
1189         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1190         if (root)
1191                 root->fs_info = fs_info;
1192         return root;
1193 }
1194
1195 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1196 /* Should only be used by the testing infrastructure */
1197 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1198 {
1199         struct btrfs_root *root;
1200
1201         if (!fs_info)
1202                 return ERR_PTR(-EINVAL);
1203
1204         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1205         if (!root)
1206                 return ERR_PTR(-ENOMEM);
1207
1208         /* We don't use the stripesize in selftest, set it as sectorsize */
1209         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1210         root->alloc_bytenr = 0;
1211
1212         return root;
1213 }
1214 #endif
1215
1216 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1217                                      struct btrfs_fs_info *fs_info,
1218                                      u64 objectid)
1219 {
1220         struct extent_buffer *leaf;
1221         struct btrfs_root *tree_root = fs_info->tree_root;
1222         struct btrfs_root *root;
1223         struct btrfs_key key;
1224         int ret = 0;
1225         uuid_le uuid = NULL_UUID_LE;
1226
1227         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1228         if (!root)
1229                 return ERR_PTR(-ENOMEM);
1230
1231         __setup_root(root, fs_info, objectid);
1232         root->root_key.objectid = objectid;
1233         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1234         root->root_key.offset = 0;
1235
1236         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1237         if (IS_ERR(leaf)) {
1238                 ret = PTR_ERR(leaf);
1239                 leaf = NULL;
1240                 goto fail;
1241         }
1242
1243         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1244         btrfs_set_header_bytenr(leaf, leaf->start);
1245         btrfs_set_header_generation(leaf, trans->transid);
1246         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1247         btrfs_set_header_owner(leaf, objectid);
1248         root->node = leaf;
1249
1250         write_extent_buffer_fsid(leaf, fs_info->fsid);
1251         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1252         btrfs_mark_buffer_dirty(leaf);
1253
1254         root->commit_root = btrfs_root_node(root);
1255         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1256
1257         root->root_item.flags = 0;
1258         root->root_item.byte_limit = 0;
1259         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1260         btrfs_set_root_generation(&root->root_item, trans->transid);
1261         btrfs_set_root_level(&root->root_item, 0);
1262         btrfs_set_root_refs(&root->root_item, 1);
1263         btrfs_set_root_used(&root->root_item, leaf->len);
1264         btrfs_set_root_last_snapshot(&root->root_item, 0);
1265         btrfs_set_root_dirid(&root->root_item, 0);
1266         if (is_fstree(objectid))
1267                 uuid_le_gen(&uuid);
1268         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1269         root->root_item.drop_level = 0;
1270
1271         key.objectid = objectid;
1272         key.type = BTRFS_ROOT_ITEM_KEY;
1273         key.offset = 0;
1274         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1275         if (ret)
1276                 goto fail;
1277
1278         btrfs_tree_unlock(leaf);
1279
1280         return root;
1281
1282 fail:
1283         if (leaf) {
1284                 btrfs_tree_unlock(leaf);
1285                 free_extent_buffer(root->commit_root);
1286                 free_extent_buffer(leaf);
1287         }
1288         kfree(root);
1289
1290         return ERR_PTR(ret);
1291 }
1292
1293 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1294                                          struct btrfs_fs_info *fs_info)
1295 {
1296         struct btrfs_root *root;
1297         struct extent_buffer *leaf;
1298
1299         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1300         if (!root)
1301                 return ERR_PTR(-ENOMEM);
1302
1303         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1304
1305         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1306         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1307         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1308
1309         /*
1310          * DON'T set REF_COWS for log trees
1311          *
1312          * log trees do not get reference counted because they go away
1313          * before a real commit is actually done.  They do store pointers
1314          * to file data extents, and those reference counts still get
1315          * updated (along with back refs to the log tree).
1316          */
1317
1318         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1319                         NULL, 0, 0, 0);
1320         if (IS_ERR(leaf)) {
1321                 kfree(root);
1322                 return ERR_CAST(leaf);
1323         }
1324
1325         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1326         btrfs_set_header_bytenr(leaf, leaf->start);
1327         btrfs_set_header_generation(leaf, trans->transid);
1328         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1329         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1330         root->node = leaf;
1331
1332         write_extent_buffer_fsid(root->node, fs_info->fsid);
1333         btrfs_mark_buffer_dirty(root->node);
1334         btrfs_tree_unlock(root->node);
1335         return root;
1336 }
1337
1338 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1339                              struct btrfs_fs_info *fs_info)
1340 {
1341         struct btrfs_root *log_root;
1342
1343         log_root = alloc_log_tree(trans, fs_info);
1344         if (IS_ERR(log_root))
1345                 return PTR_ERR(log_root);
1346         WARN_ON(fs_info->log_root_tree);
1347         fs_info->log_root_tree = log_root;
1348         return 0;
1349 }
1350
1351 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1352                        struct btrfs_root *root)
1353 {
1354         struct btrfs_fs_info *fs_info = root->fs_info;
1355         struct btrfs_root *log_root;
1356         struct btrfs_inode_item *inode_item;
1357
1358         log_root = alloc_log_tree(trans, fs_info);
1359         if (IS_ERR(log_root))
1360                 return PTR_ERR(log_root);
1361
1362         log_root->last_trans = trans->transid;
1363         log_root->root_key.offset = root->root_key.objectid;
1364
1365         inode_item = &log_root->root_item.inode;
1366         btrfs_set_stack_inode_generation(inode_item, 1);
1367         btrfs_set_stack_inode_size(inode_item, 3);
1368         btrfs_set_stack_inode_nlink(inode_item, 1);
1369         btrfs_set_stack_inode_nbytes(inode_item,
1370                                      fs_info->nodesize);
1371         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1372
1373         btrfs_set_root_node(&log_root->root_item, log_root->node);
1374
1375         WARN_ON(root->log_root);
1376         root->log_root = log_root;
1377         root->log_transid = 0;
1378         root->log_transid_committed = -1;
1379         root->last_log_commit = 0;
1380         return 0;
1381 }
1382
1383 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1384                                                struct btrfs_key *key)
1385 {
1386         struct btrfs_root *root;
1387         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1388         struct btrfs_path *path;
1389         u64 generation;
1390         int ret;
1391
1392         path = btrfs_alloc_path();
1393         if (!path)
1394                 return ERR_PTR(-ENOMEM);
1395
1396         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1397         if (!root) {
1398                 ret = -ENOMEM;
1399                 goto alloc_fail;
1400         }
1401
1402         __setup_root(root, fs_info, key->objectid);
1403
1404         ret = btrfs_find_root(tree_root, key, path,
1405                               &root->root_item, &root->root_key);
1406         if (ret) {
1407                 if (ret > 0)
1408                         ret = -ENOENT;
1409                 goto find_fail;
1410         }
1411
1412         generation = btrfs_root_generation(&root->root_item);
1413         root->node = read_tree_block(fs_info,
1414                                      btrfs_root_bytenr(&root->root_item),
1415                                      generation);
1416         if (IS_ERR(root->node)) {
1417                 ret = PTR_ERR(root->node);
1418                 goto find_fail;
1419         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1420                 ret = -EIO;
1421                 free_extent_buffer(root->node);
1422                 goto find_fail;
1423         }
1424         root->commit_root = btrfs_root_node(root);
1425 out:
1426         btrfs_free_path(path);
1427         return root;
1428
1429 find_fail:
1430         kfree(root);
1431 alloc_fail:
1432         root = ERR_PTR(ret);
1433         goto out;
1434 }
1435
1436 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1437                                       struct btrfs_key *location)
1438 {
1439         struct btrfs_root *root;
1440
1441         root = btrfs_read_tree_root(tree_root, location);
1442         if (IS_ERR(root))
1443                 return root;
1444
1445         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1446                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1447                 btrfs_check_and_init_root_item(&root->root_item);
1448         }
1449
1450         return root;
1451 }
1452
1453 int btrfs_init_fs_root(struct btrfs_root *root)
1454 {
1455         int ret;
1456         struct btrfs_subvolume_writers *writers;
1457
1458         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1459         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1460                                         GFP_NOFS);
1461         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1462                 ret = -ENOMEM;
1463                 goto fail;
1464         }
1465
1466         writers = btrfs_alloc_subvolume_writers();
1467         if (IS_ERR(writers)) {
1468                 ret = PTR_ERR(writers);
1469                 goto fail;
1470         }
1471         root->subv_writers = writers;
1472
1473         btrfs_init_free_ino_ctl(root);
1474         spin_lock_init(&root->ino_cache_lock);
1475         init_waitqueue_head(&root->ino_cache_wait);
1476
1477         ret = get_anon_bdev(&root->anon_dev);
1478         if (ret)
1479                 goto fail;
1480
1481         mutex_lock(&root->objectid_mutex);
1482         ret = btrfs_find_highest_objectid(root,
1483                                         &root->highest_objectid);
1484         if (ret) {
1485                 mutex_unlock(&root->objectid_mutex);
1486                 goto fail;
1487         }
1488
1489         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1490
1491         mutex_unlock(&root->objectid_mutex);
1492
1493         return 0;
1494 fail:
1495         /* the caller is responsible to call free_fs_root */
1496         return ret;
1497 }
1498
1499 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1500                                         u64 root_id)
1501 {
1502         struct btrfs_root *root;
1503
1504         spin_lock(&fs_info->fs_roots_radix_lock);
1505         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1506                                  (unsigned long)root_id);
1507         spin_unlock(&fs_info->fs_roots_radix_lock);
1508         return root;
1509 }
1510
1511 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1512                          struct btrfs_root *root)
1513 {
1514         int ret;
1515
1516         ret = radix_tree_preload(GFP_NOFS);
1517         if (ret)
1518                 return ret;
1519
1520         spin_lock(&fs_info->fs_roots_radix_lock);
1521         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1522                                 (unsigned long)root->root_key.objectid,
1523                                 root);
1524         if (ret == 0)
1525                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1526         spin_unlock(&fs_info->fs_roots_radix_lock);
1527         radix_tree_preload_end();
1528
1529         return ret;
1530 }
1531
1532 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1533                                      struct btrfs_key *location,
1534                                      bool check_ref)
1535 {
1536         struct btrfs_root *root;
1537         struct btrfs_path *path;
1538         struct btrfs_key key;
1539         int ret;
1540
1541         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1542                 return fs_info->tree_root;
1543         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1544                 return fs_info->extent_root;
1545         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1546                 return fs_info->chunk_root;
1547         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1548                 return fs_info->dev_root;
1549         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1550                 return fs_info->csum_root;
1551         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1552                 return fs_info->quota_root ? fs_info->quota_root :
1553                                              ERR_PTR(-ENOENT);
1554         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1555                 return fs_info->uuid_root ? fs_info->uuid_root :
1556                                             ERR_PTR(-ENOENT);
1557         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1558                 return fs_info->free_space_root ? fs_info->free_space_root :
1559                                                   ERR_PTR(-ENOENT);
1560 again:
1561         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1562         if (root) {
1563                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1564                         return ERR_PTR(-ENOENT);
1565                 return root;
1566         }
1567
1568         root = btrfs_read_fs_root(fs_info->tree_root, location);
1569         if (IS_ERR(root))
1570                 return root;
1571
1572         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1573                 ret = -ENOENT;
1574                 goto fail;
1575         }
1576
1577         ret = btrfs_init_fs_root(root);
1578         if (ret)
1579                 goto fail;
1580
1581         path = btrfs_alloc_path();
1582         if (!path) {
1583                 ret = -ENOMEM;
1584                 goto fail;
1585         }
1586         key.objectid = BTRFS_ORPHAN_OBJECTID;
1587         key.type = BTRFS_ORPHAN_ITEM_KEY;
1588         key.offset = location->objectid;
1589
1590         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1591         btrfs_free_path(path);
1592         if (ret < 0)
1593                 goto fail;
1594         if (ret == 0)
1595                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1596
1597         ret = btrfs_insert_fs_root(fs_info, root);
1598         if (ret) {
1599                 if (ret == -EEXIST) {
1600                         free_fs_root(root);
1601                         goto again;
1602                 }
1603                 goto fail;
1604         }
1605         return root;
1606 fail:
1607         free_fs_root(root);
1608         return ERR_PTR(ret);
1609 }
1610
1611 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1612 {
1613         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1614         int ret = 0;
1615         struct btrfs_device *device;
1616         struct backing_dev_info *bdi;
1617
1618         rcu_read_lock();
1619         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1620                 if (!device->bdev)
1621                         continue;
1622                 bdi = device->bdev->bd_bdi;
1623                 if (bdi_congested(bdi, bdi_bits)) {
1624                         ret = 1;
1625                         break;
1626                 }
1627         }
1628         rcu_read_unlock();
1629         return ret;
1630 }
1631
1632 /*
1633  * called by the kthread helper functions to finally call the bio end_io
1634  * functions.  This is where read checksum verification actually happens
1635  */
1636 static void end_workqueue_fn(struct btrfs_work *work)
1637 {
1638         struct bio *bio;
1639         struct btrfs_end_io_wq *end_io_wq;
1640
1641         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1642         bio = end_io_wq->bio;
1643
1644         bio->bi_status = end_io_wq->status;
1645         bio->bi_private = end_io_wq->private;
1646         bio->bi_end_io = end_io_wq->end_io;
1647         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1648         bio_endio(bio);
1649 }
1650
1651 static int cleaner_kthread(void *arg)
1652 {
1653         struct btrfs_root *root = arg;
1654         struct btrfs_fs_info *fs_info = root->fs_info;
1655         int again;
1656         struct btrfs_trans_handle *trans;
1657
1658         do {
1659                 again = 0;
1660
1661                 /* Make the cleaner go to sleep early. */
1662                 if (btrfs_need_cleaner_sleep(fs_info))
1663                         goto sleep;
1664
1665                 /*
1666                  * Do not do anything if we might cause open_ctree() to block
1667                  * before we have finished mounting the filesystem.
1668                  */
1669                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1670                         goto sleep;
1671
1672                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1673                         goto sleep;
1674
1675                 /*
1676                  * Avoid the problem that we change the status of the fs
1677                  * during the above check and trylock.
1678                  */
1679                 if (btrfs_need_cleaner_sleep(fs_info)) {
1680                         mutex_unlock(&fs_info->cleaner_mutex);
1681                         goto sleep;
1682                 }
1683
1684                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1685                 btrfs_run_delayed_iputs(fs_info);
1686                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1687
1688                 again = btrfs_clean_one_deleted_snapshot(root);
1689                 mutex_unlock(&fs_info->cleaner_mutex);
1690
1691                 /*
1692                  * The defragger has dealt with the R/O remount and umount,
1693                  * needn't do anything special here.
1694                  */
1695                 btrfs_run_defrag_inodes(fs_info);
1696
1697                 /*
1698                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1699                  * with relocation (btrfs_relocate_chunk) and relocation
1700                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1701                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1702                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1703                  * unused block groups.
1704                  */
1705                 btrfs_delete_unused_bgs(fs_info);
1706 sleep:
1707                 if (!again) {
1708                         set_current_state(TASK_INTERRUPTIBLE);
1709                         if (!kthread_should_stop())
1710                                 schedule();
1711                         __set_current_state(TASK_RUNNING);
1712                 }
1713         } while (!kthread_should_stop());
1714
1715         /*
1716          * Transaction kthread is stopped before us and wakes us up.
1717          * However we might have started a new transaction and COWed some
1718          * tree blocks when deleting unused block groups for example. So
1719          * make sure we commit the transaction we started to have a clean
1720          * shutdown when evicting the btree inode - if it has dirty pages
1721          * when we do the final iput() on it, eviction will trigger a
1722          * writeback for it which will fail with null pointer dereferences
1723          * since work queues and other resources were already released and
1724          * destroyed by the time the iput/eviction/writeback is made.
1725          */
1726         trans = btrfs_attach_transaction(root);
1727         if (IS_ERR(trans)) {
1728                 if (PTR_ERR(trans) != -ENOENT)
1729                         btrfs_err(fs_info,
1730                                   "cleaner transaction attach returned %ld",
1731                                   PTR_ERR(trans));
1732         } else {
1733                 int ret;
1734
1735                 ret = btrfs_commit_transaction(trans);
1736                 if (ret)
1737                         btrfs_err(fs_info,
1738                                   "cleaner open transaction commit returned %d",
1739                                   ret);
1740         }
1741
1742         return 0;
1743 }
1744
1745 static int transaction_kthread(void *arg)
1746 {
1747         struct btrfs_root *root = arg;
1748         struct btrfs_fs_info *fs_info = root->fs_info;
1749         struct btrfs_trans_handle *trans;
1750         struct btrfs_transaction *cur;
1751         u64 transid;
1752         unsigned long now;
1753         unsigned long delay;
1754         bool cannot_commit;
1755
1756         do {
1757                 cannot_commit = false;
1758                 delay = HZ * fs_info->commit_interval;
1759                 mutex_lock(&fs_info->transaction_kthread_mutex);
1760
1761                 spin_lock(&fs_info->trans_lock);
1762                 cur = fs_info->running_transaction;
1763                 if (!cur) {
1764                         spin_unlock(&fs_info->trans_lock);
1765                         goto sleep;
1766                 }
1767
1768                 now = get_seconds();
1769                 if (cur->state < TRANS_STATE_BLOCKED &&
1770                     (now < cur->start_time ||
1771                      now - cur->start_time < fs_info->commit_interval)) {
1772                         spin_unlock(&fs_info->trans_lock);
1773                         delay = HZ * 5;
1774                         goto sleep;
1775                 }
1776                 transid = cur->transid;
1777                 spin_unlock(&fs_info->trans_lock);
1778
1779                 /* If the file system is aborted, this will always fail. */
1780                 trans = btrfs_attach_transaction(root);
1781                 if (IS_ERR(trans)) {
1782                         if (PTR_ERR(trans) != -ENOENT)
1783                                 cannot_commit = true;
1784                         goto sleep;
1785                 }
1786                 if (transid == trans->transid) {
1787                         btrfs_commit_transaction(trans);
1788                 } else {
1789                         btrfs_end_transaction(trans);
1790                 }
1791 sleep:
1792                 wake_up_process(fs_info->cleaner_kthread);
1793                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1794
1795                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1796                                       &fs_info->fs_state)))
1797                         btrfs_cleanup_transaction(fs_info);
1798                 if (!kthread_should_stop() &&
1799                                 (!btrfs_transaction_blocked(fs_info) ||
1800                                  cannot_commit))
1801                         schedule_timeout_interruptible(delay);
1802         } while (!kthread_should_stop());
1803         return 0;
1804 }
1805
1806 /*
1807  * this will find the highest generation in the array of
1808  * root backups.  The index of the highest array is returned,
1809  * or -1 if we can't find anything.
1810  *
1811  * We check to make sure the array is valid by comparing the
1812  * generation of the latest  root in the array with the generation
1813  * in the super block.  If they don't match we pitch it.
1814  */
1815 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1816 {
1817         u64 cur;
1818         int newest_index = -1;
1819         struct btrfs_root_backup *root_backup;
1820         int i;
1821
1822         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1823                 root_backup = info->super_copy->super_roots + i;
1824                 cur = btrfs_backup_tree_root_gen(root_backup);
1825                 if (cur == newest_gen)
1826                         newest_index = i;
1827         }
1828
1829         /* check to see if we actually wrapped around */
1830         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1831                 root_backup = info->super_copy->super_roots;
1832                 cur = btrfs_backup_tree_root_gen(root_backup);
1833                 if (cur == newest_gen)
1834                         newest_index = 0;
1835         }
1836         return newest_index;
1837 }
1838
1839
1840 /*
1841  * find the oldest backup so we know where to store new entries
1842  * in the backup array.  This will set the backup_root_index
1843  * field in the fs_info struct
1844  */
1845 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1846                                      u64 newest_gen)
1847 {
1848         int newest_index = -1;
1849
1850         newest_index = find_newest_super_backup(info, newest_gen);
1851         /* if there was garbage in there, just move along */
1852         if (newest_index == -1) {
1853                 info->backup_root_index = 0;
1854         } else {
1855                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1856         }
1857 }
1858
1859 /*
1860  * copy all the root pointers into the super backup array.
1861  * this will bump the backup pointer by one when it is
1862  * done
1863  */
1864 static void backup_super_roots(struct btrfs_fs_info *info)
1865 {
1866         int next_backup;
1867         struct btrfs_root_backup *root_backup;
1868         int last_backup;
1869
1870         next_backup = info->backup_root_index;
1871         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1872                 BTRFS_NUM_BACKUP_ROOTS;
1873
1874         /*
1875          * just overwrite the last backup if we're at the same generation
1876          * this happens only at umount
1877          */
1878         root_backup = info->super_for_commit->super_roots + last_backup;
1879         if (btrfs_backup_tree_root_gen(root_backup) ==
1880             btrfs_header_generation(info->tree_root->node))
1881                 next_backup = last_backup;
1882
1883         root_backup = info->super_for_commit->super_roots + next_backup;
1884
1885         /*
1886          * make sure all of our padding and empty slots get zero filled
1887          * regardless of which ones we use today
1888          */
1889         memset(root_backup, 0, sizeof(*root_backup));
1890
1891         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1892
1893         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1894         btrfs_set_backup_tree_root_gen(root_backup,
1895                                btrfs_header_generation(info->tree_root->node));
1896
1897         btrfs_set_backup_tree_root_level(root_backup,
1898                                btrfs_header_level(info->tree_root->node));
1899
1900         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1901         btrfs_set_backup_chunk_root_gen(root_backup,
1902                                btrfs_header_generation(info->chunk_root->node));
1903         btrfs_set_backup_chunk_root_level(root_backup,
1904                                btrfs_header_level(info->chunk_root->node));
1905
1906         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1907         btrfs_set_backup_extent_root_gen(root_backup,
1908                                btrfs_header_generation(info->extent_root->node));
1909         btrfs_set_backup_extent_root_level(root_backup,
1910                                btrfs_header_level(info->extent_root->node));
1911
1912         /*
1913          * we might commit during log recovery, which happens before we set
1914          * the fs_root.  Make sure it is valid before we fill it in.
1915          */
1916         if (info->fs_root && info->fs_root->node) {
1917                 btrfs_set_backup_fs_root(root_backup,
1918                                          info->fs_root->node->start);
1919                 btrfs_set_backup_fs_root_gen(root_backup,
1920                                btrfs_header_generation(info->fs_root->node));
1921                 btrfs_set_backup_fs_root_level(root_backup,
1922                                btrfs_header_level(info->fs_root->node));
1923         }
1924
1925         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1926         btrfs_set_backup_dev_root_gen(root_backup,
1927                                btrfs_header_generation(info->dev_root->node));
1928         btrfs_set_backup_dev_root_level(root_backup,
1929                                        btrfs_header_level(info->dev_root->node));
1930
1931         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1932         btrfs_set_backup_csum_root_gen(root_backup,
1933                                btrfs_header_generation(info->csum_root->node));
1934         btrfs_set_backup_csum_root_level(root_backup,
1935                                btrfs_header_level(info->csum_root->node));
1936
1937         btrfs_set_backup_total_bytes(root_backup,
1938                              btrfs_super_total_bytes(info->super_copy));
1939         btrfs_set_backup_bytes_used(root_backup,
1940                              btrfs_super_bytes_used(info->super_copy));
1941         btrfs_set_backup_num_devices(root_backup,
1942                              btrfs_super_num_devices(info->super_copy));
1943
1944         /*
1945          * if we don't copy this out to the super_copy, it won't get remembered
1946          * for the next commit
1947          */
1948         memcpy(&info->super_copy->super_roots,
1949                &info->super_for_commit->super_roots,
1950                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1951 }
1952
1953 /*
1954  * this copies info out of the root backup array and back into
1955  * the in-memory super block.  It is meant to help iterate through
1956  * the array, so you send it the number of backups you've already
1957  * tried and the last backup index you used.
1958  *
1959  * this returns -1 when it has tried all the backups
1960  */
1961 static noinline int next_root_backup(struct btrfs_fs_info *info,
1962                                      struct btrfs_super_block *super,
1963                                      int *num_backups_tried, int *backup_index)
1964 {
1965         struct btrfs_root_backup *root_backup;
1966         int newest = *backup_index;
1967
1968         if (*num_backups_tried == 0) {
1969                 u64 gen = btrfs_super_generation(super);
1970
1971                 newest = find_newest_super_backup(info, gen);
1972                 if (newest == -1)
1973                         return -1;
1974
1975                 *backup_index = newest;
1976                 *num_backups_tried = 1;
1977         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1978                 /* we've tried all the backups, all done */
1979                 return -1;
1980         } else {
1981                 /* jump to the next oldest backup */
1982                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1983                         BTRFS_NUM_BACKUP_ROOTS;
1984                 *backup_index = newest;
1985                 *num_backups_tried += 1;
1986         }
1987         root_backup = super->super_roots + newest;
1988
1989         btrfs_set_super_generation(super,
1990                                    btrfs_backup_tree_root_gen(root_backup));
1991         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1992         btrfs_set_super_root_level(super,
1993                                    btrfs_backup_tree_root_level(root_backup));
1994         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1995
1996         /*
1997          * fixme: the total bytes and num_devices need to match or we should
1998          * need a fsck
1999          */
2000         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2001         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2002         return 0;
2003 }
2004
2005 /* helper to cleanup workers */
2006 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2007 {
2008         btrfs_destroy_workqueue(fs_info->fixup_workers);
2009         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2010         btrfs_destroy_workqueue(fs_info->workers);
2011         btrfs_destroy_workqueue(fs_info->endio_workers);
2012         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2013         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2014         btrfs_destroy_workqueue(fs_info->rmw_workers);
2015         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2016         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2017         btrfs_destroy_workqueue(fs_info->submit_workers);
2018         btrfs_destroy_workqueue(fs_info->delayed_workers);
2019         btrfs_destroy_workqueue(fs_info->caching_workers);
2020         btrfs_destroy_workqueue(fs_info->readahead_workers);
2021         btrfs_destroy_workqueue(fs_info->flush_workers);
2022         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2023         btrfs_destroy_workqueue(fs_info->extent_workers);
2024         /*
2025          * Now that all other work queues are destroyed, we can safely destroy
2026          * the queues used for metadata I/O, since tasks from those other work
2027          * queues can do metadata I/O operations.
2028          */
2029         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2030         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2031 }
2032
2033 static void free_root_extent_buffers(struct btrfs_root *root)
2034 {
2035         if (root) {
2036                 free_extent_buffer(root->node);
2037                 free_extent_buffer(root->commit_root);
2038                 root->node = NULL;
2039                 root->commit_root = NULL;
2040         }
2041 }
2042
2043 /* helper to cleanup tree roots */
2044 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2045 {
2046         free_root_extent_buffers(info->tree_root);
2047
2048         free_root_extent_buffers(info->dev_root);
2049         free_root_extent_buffers(info->extent_root);
2050         free_root_extent_buffers(info->csum_root);
2051         free_root_extent_buffers(info->quota_root);
2052         free_root_extent_buffers(info->uuid_root);
2053         if (chunk_root)
2054                 free_root_extent_buffers(info->chunk_root);
2055         free_root_extent_buffers(info->free_space_root);
2056 }
2057
2058 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2059 {
2060         int ret;
2061         struct btrfs_root *gang[8];
2062         int i;
2063
2064         while (!list_empty(&fs_info->dead_roots)) {
2065                 gang[0] = list_entry(fs_info->dead_roots.next,
2066                                      struct btrfs_root, root_list);
2067                 list_del(&gang[0]->root_list);
2068
2069                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2070                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2071                 } else {
2072                         free_extent_buffer(gang[0]->node);
2073                         free_extent_buffer(gang[0]->commit_root);
2074                         btrfs_put_fs_root(gang[0]);
2075                 }
2076         }
2077
2078         while (1) {
2079                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2080                                              (void **)gang, 0,
2081                                              ARRAY_SIZE(gang));
2082                 if (!ret)
2083                         break;
2084                 for (i = 0; i < ret; i++)
2085                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2086         }
2087
2088         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2089                 btrfs_free_log_root_tree(NULL, fs_info);
2090                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2091         }
2092 }
2093
2094 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2095 {
2096         mutex_init(&fs_info->scrub_lock);
2097         atomic_set(&fs_info->scrubs_running, 0);
2098         atomic_set(&fs_info->scrub_pause_req, 0);
2099         atomic_set(&fs_info->scrubs_paused, 0);
2100         atomic_set(&fs_info->scrub_cancel_req, 0);
2101         init_waitqueue_head(&fs_info->scrub_pause_wait);
2102         fs_info->scrub_workers_refcnt = 0;
2103 }
2104
2105 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2106 {
2107         spin_lock_init(&fs_info->balance_lock);
2108         mutex_init(&fs_info->balance_mutex);
2109         atomic_set(&fs_info->balance_running, 0);
2110         atomic_set(&fs_info->balance_pause_req, 0);
2111         atomic_set(&fs_info->balance_cancel_req, 0);
2112         fs_info->balance_ctl = NULL;
2113         init_waitqueue_head(&fs_info->balance_wait_q);
2114 }
2115
2116 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2117 {
2118         struct inode *inode = fs_info->btree_inode;
2119
2120         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2121         set_nlink(inode, 1);
2122         /*
2123          * we set the i_size on the btree inode to the max possible int.
2124          * the real end of the address space is determined by all of
2125          * the devices in the system
2126          */
2127         inode->i_size = OFFSET_MAX;
2128         inode->i_mapping->a_ops = &btree_aops;
2129
2130         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2131         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2132         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2133         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2134
2135         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2136
2137         BTRFS_I(inode)->root = fs_info->tree_root;
2138         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2139         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2140         btrfs_insert_inode_hash(inode);
2141 }
2142
2143 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2144 {
2145         fs_info->dev_replace.lock_owner = 0;
2146         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2147         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2148         rwlock_init(&fs_info->dev_replace.lock);
2149         atomic_set(&fs_info->dev_replace.read_locks, 0);
2150         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2151         init_waitqueue_head(&fs_info->replace_wait);
2152         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2153 }
2154
2155 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2156 {
2157         spin_lock_init(&fs_info->qgroup_lock);
2158         mutex_init(&fs_info->qgroup_ioctl_lock);
2159         fs_info->qgroup_tree = RB_ROOT;
2160         fs_info->qgroup_op_tree = RB_ROOT;
2161         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2162         fs_info->qgroup_seq = 1;
2163         fs_info->qgroup_ulist = NULL;
2164         fs_info->qgroup_rescan_running = false;
2165         mutex_init(&fs_info->qgroup_rescan_lock);
2166 }
2167
2168 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2169                 struct btrfs_fs_devices *fs_devices)
2170 {
2171         u32 max_active = fs_info->thread_pool_size;
2172         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2173
2174         fs_info->workers =
2175                 btrfs_alloc_workqueue(fs_info, "worker",
2176                                       flags | WQ_HIGHPRI, max_active, 16);
2177
2178         fs_info->delalloc_workers =
2179                 btrfs_alloc_workqueue(fs_info, "delalloc",
2180                                       flags, max_active, 2);
2181
2182         fs_info->flush_workers =
2183                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2184                                       flags, max_active, 0);
2185
2186         fs_info->caching_workers =
2187                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2188
2189         /*
2190          * a higher idle thresh on the submit workers makes it much more
2191          * likely that bios will be send down in a sane order to the
2192          * devices
2193          */
2194         fs_info->submit_workers =
2195                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2196                                       min_t(u64, fs_devices->num_devices,
2197                                             max_active), 64);
2198
2199         fs_info->fixup_workers =
2200                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2201
2202         /*
2203          * endios are largely parallel and should have a very
2204          * low idle thresh
2205          */
2206         fs_info->endio_workers =
2207                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2208         fs_info->endio_meta_workers =
2209                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2210                                       max_active, 4);
2211         fs_info->endio_meta_write_workers =
2212                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2213                                       max_active, 2);
2214         fs_info->endio_raid56_workers =
2215                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2216                                       max_active, 4);
2217         fs_info->endio_repair_workers =
2218                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2219         fs_info->rmw_workers =
2220                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2221         fs_info->endio_write_workers =
2222                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2223                                       max_active, 2);
2224         fs_info->endio_freespace_worker =
2225                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2226                                       max_active, 0);
2227         fs_info->delayed_workers =
2228                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2229                                       max_active, 0);
2230         fs_info->readahead_workers =
2231                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2232                                       max_active, 2);
2233         fs_info->qgroup_rescan_workers =
2234                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2235         fs_info->extent_workers =
2236                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2237                                       min_t(u64, fs_devices->num_devices,
2238                                             max_active), 8);
2239
2240         if (!(fs_info->workers && fs_info->delalloc_workers &&
2241               fs_info->submit_workers && fs_info->flush_workers &&
2242               fs_info->endio_workers && fs_info->endio_meta_workers &&
2243               fs_info->endio_meta_write_workers &&
2244               fs_info->endio_repair_workers &&
2245               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2246               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2247               fs_info->caching_workers && fs_info->readahead_workers &&
2248               fs_info->fixup_workers && fs_info->delayed_workers &&
2249               fs_info->extent_workers &&
2250               fs_info->qgroup_rescan_workers)) {
2251                 return -ENOMEM;
2252         }
2253
2254         return 0;
2255 }
2256
2257 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2258                             struct btrfs_fs_devices *fs_devices)
2259 {
2260         int ret;
2261         struct btrfs_root *log_tree_root;
2262         struct btrfs_super_block *disk_super = fs_info->super_copy;
2263         u64 bytenr = btrfs_super_log_root(disk_super);
2264
2265         if (fs_devices->rw_devices == 0) {
2266                 btrfs_warn(fs_info, "log replay required on RO media");
2267                 return -EIO;
2268         }
2269
2270         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2271         if (!log_tree_root)
2272                 return -ENOMEM;
2273
2274         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2275
2276         log_tree_root->node = read_tree_block(fs_info, bytenr,
2277                                               fs_info->generation + 1);
2278         if (IS_ERR(log_tree_root->node)) {
2279                 btrfs_warn(fs_info, "failed to read log tree");
2280                 ret = PTR_ERR(log_tree_root->node);
2281                 kfree(log_tree_root);
2282                 return ret;
2283         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2284                 btrfs_err(fs_info, "failed to read log tree");
2285                 free_extent_buffer(log_tree_root->node);
2286                 kfree(log_tree_root);
2287                 return -EIO;
2288         }
2289         /* returns with log_tree_root freed on success */
2290         ret = btrfs_recover_log_trees(log_tree_root);
2291         if (ret) {
2292                 btrfs_handle_fs_error(fs_info, ret,
2293                                       "Failed to recover log tree");
2294                 free_extent_buffer(log_tree_root->node);
2295                 kfree(log_tree_root);
2296                 return ret;
2297         }
2298
2299         if (sb_rdonly(fs_info->sb)) {
2300                 ret = btrfs_commit_super(fs_info);
2301                 if (ret)
2302                         return ret;
2303         }
2304
2305         return 0;
2306 }
2307
2308 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2309 {
2310         struct btrfs_root *tree_root = fs_info->tree_root;
2311         struct btrfs_root *root;
2312         struct btrfs_key location;
2313         int ret;
2314
2315         BUG_ON(!fs_info->tree_root);
2316
2317         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2318         location.type = BTRFS_ROOT_ITEM_KEY;
2319         location.offset = 0;
2320
2321         root = btrfs_read_tree_root(tree_root, &location);
2322         if (IS_ERR(root))
2323                 return PTR_ERR(root);
2324         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2325         fs_info->extent_root = root;
2326
2327         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2328         root = btrfs_read_tree_root(tree_root, &location);
2329         if (IS_ERR(root))
2330                 return PTR_ERR(root);
2331         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2332         fs_info->dev_root = root;
2333         btrfs_init_devices_late(fs_info);
2334
2335         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2336         root = btrfs_read_tree_root(tree_root, &location);
2337         if (IS_ERR(root))
2338                 return PTR_ERR(root);
2339         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2340         fs_info->csum_root = root;
2341
2342         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2343         root = btrfs_read_tree_root(tree_root, &location);
2344         if (!IS_ERR(root)) {
2345                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2346                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2347                 fs_info->quota_root = root;
2348         }
2349
2350         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2351         root = btrfs_read_tree_root(tree_root, &location);
2352         if (IS_ERR(root)) {
2353                 ret = PTR_ERR(root);
2354                 if (ret != -ENOENT)
2355                         return ret;
2356         } else {
2357                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2358                 fs_info->uuid_root = root;
2359         }
2360
2361         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2362                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2363                 root = btrfs_read_tree_root(tree_root, &location);
2364                 if (IS_ERR(root))
2365                         return PTR_ERR(root);
2366                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2367                 fs_info->free_space_root = root;
2368         }
2369
2370         return 0;
2371 }
2372
2373 int open_ctree(struct super_block *sb,
2374                struct btrfs_fs_devices *fs_devices,
2375                char *options)
2376 {
2377         u32 sectorsize;
2378         u32 nodesize;
2379         u32 stripesize;
2380         u64 generation;
2381         u64 features;
2382         struct btrfs_key location;
2383         struct buffer_head *bh;
2384         struct btrfs_super_block *disk_super;
2385         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2386         struct btrfs_root *tree_root;
2387         struct btrfs_root *chunk_root;
2388         int ret;
2389         int err = -EINVAL;
2390         int num_backups_tried = 0;
2391         int backup_index = 0;
2392         int clear_free_space_tree = 0;
2393
2394         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2395         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2396         if (!tree_root || !chunk_root) {
2397                 err = -ENOMEM;
2398                 goto fail;
2399         }
2400
2401         ret = init_srcu_struct(&fs_info->subvol_srcu);
2402         if (ret) {
2403                 err = ret;
2404                 goto fail;
2405         }
2406
2407         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2408         if (ret) {
2409                 err = ret;
2410                 goto fail_srcu;
2411         }
2412         fs_info->dirty_metadata_batch = PAGE_SIZE *
2413                                         (1 + ilog2(nr_cpu_ids));
2414
2415         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2416         if (ret) {
2417                 err = ret;
2418                 goto fail_dirty_metadata_bytes;
2419         }
2420
2421         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2422         if (ret) {
2423                 err = ret;
2424                 goto fail_delalloc_bytes;
2425         }
2426
2427         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2428         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2429         INIT_LIST_HEAD(&fs_info->trans_list);
2430         INIT_LIST_HEAD(&fs_info->dead_roots);
2431         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2432         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2433         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2434         spin_lock_init(&fs_info->delalloc_root_lock);
2435         spin_lock_init(&fs_info->trans_lock);
2436         spin_lock_init(&fs_info->fs_roots_radix_lock);
2437         spin_lock_init(&fs_info->delayed_iput_lock);
2438         spin_lock_init(&fs_info->defrag_inodes_lock);
2439         spin_lock_init(&fs_info->tree_mod_seq_lock);
2440         spin_lock_init(&fs_info->super_lock);
2441         spin_lock_init(&fs_info->qgroup_op_lock);
2442         spin_lock_init(&fs_info->buffer_lock);
2443         spin_lock_init(&fs_info->unused_bgs_lock);
2444         rwlock_init(&fs_info->tree_mod_log_lock);
2445         mutex_init(&fs_info->unused_bg_unpin_mutex);
2446         mutex_init(&fs_info->delete_unused_bgs_mutex);
2447         mutex_init(&fs_info->reloc_mutex);
2448         mutex_init(&fs_info->delalloc_root_mutex);
2449         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2450         seqlock_init(&fs_info->profiles_lock);
2451
2452         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2453         INIT_LIST_HEAD(&fs_info->space_info);
2454         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2455         INIT_LIST_HEAD(&fs_info->unused_bgs);
2456         btrfs_mapping_init(&fs_info->mapping_tree);
2457         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2458                              BTRFS_BLOCK_RSV_GLOBAL);
2459         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2460         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2461         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2462         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2463                              BTRFS_BLOCK_RSV_DELOPS);
2464         atomic_set(&fs_info->async_delalloc_pages, 0);
2465         atomic_set(&fs_info->defrag_running, 0);
2466         atomic_set(&fs_info->qgroup_op_seq, 0);
2467         atomic_set(&fs_info->reada_works_cnt, 0);
2468         atomic64_set(&fs_info->tree_mod_seq, 0);
2469         fs_info->sb = sb;
2470         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2471         fs_info->metadata_ratio = 0;
2472         fs_info->defrag_inodes = RB_ROOT;
2473         atomic64_set(&fs_info->free_chunk_space, 0);
2474         fs_info->tree_mod_log = RB_ROOT;
2475         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2476         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2477         /* readahead state */
2478         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2479         spin_lock_init(&fs_info->reada_lock);
2480         btrfs_init_ref_verify(fs_info);
2481
2482         fs_info->thread_pool_size = min_t(unsigned long,
2483                                           num_online_cpus() + 2, 8);
2484
2485         INIT_LIST_HEAD(&fs_info->ordered_roots);
2486         spin_lock_init(&fs_info->ordered_root_lock);
2487
2488         fs_info->btree_inode = new_inode(sb);
2489         if (!fs_info->btree_inode) {
2490                 err = -ENOMEM;
2491                 goto fail_bio_counter;
2492         }
2493         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2494
2495         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2496                                         GFP_KERNEL);
2497         if (!fs_info->delayed_root) {
2498                 err = -ENOMEM;
2499                 goto fail_iput;
2500         }
2501         btrfs_init_delayed_root(fs_info->delayed_root);
2502
2503         btrfs_init_scrub(fs_info);
2504 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2505         fs_info->check_integrity_print_mask = 0;
2506 #endif
2507         btrfs_init_balance(fs_info);
2508         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2509
2510         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2511         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2512
2513         btrfs_init_btree_inode(fs_info);
2514
2515         spin_lock_init(&fs_info->block_group_cache_lock);
2516         fs_info->block_group_cache_tree = RB_ROOT;
2517         fs_info->first_logical_byte = (u64)-1;
2518
2519         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2520         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2521         fs_info->pinned_extents = &fs_info->freed_extents[0];
2522         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2523
2524         mutex_init(&fs_info->ordered_operations_mutex);
2525         mutex_init(&fs_info->tree_log_mutex);
2526         mutex_init(&fs_info->chunk_mutex);
2527         mutex_init(&fs_info->transaction_kthread_mutex);
2528         mutex_init(&fs_info->cleaner_mutex);
2529         mutex_init(&fs_info->volume_mutex);
2530         mutex_init(&fs_info->ro_block_group_mutex);
2531         init_rwsem(&fs_info->commit_root_sem);
2532         init_rwsem(&fs_info->cleanup_work_sem);
2533         init_rwsem(&fs_info->subvol_sem);
2534         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2535
2536         btrfs_init_dev_replace_locks(fs_info);
2537         btrfs_init_qgroup(fs_info);
2538
2539         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2540         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2541
2542         init_waitqueue_head(&fs_info->transaction_throttle);
2543         init_waitqueue_head(&fs_info->transaction_wait);
2544         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2545         init_waitqueue_head(&fs_info->async_submit_wait);
2546
2547         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2548
2549         /* Usable values until the real ones are cached from the superblock */
2550         fs_info->nodesize = 4096;
2551         fs_info->sectorsize = 4096;
2552         fs_info->stripesize = 4096;
2553
2554         ret = btrfs_alloc_stripe_hash_table(fs_info);
2555         if (ret) {
2556                 err = ret;
2557                 goto fail_alloc;
2558         }
2559
2560         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2561
2562         invalidate_bdev(fs_devices->latest_bdev);
2563
2564         /*
2565          * Read super block and check the signature bytes only
2566          */
2567         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2568         if (IS_ERR(bh)) {
2569                 err = PTR_ERR(bh);
2570                 goto fail_alloc;
2571         }
2572
2573         /*
2574          * We want to check superblock checksum, the type is stored inside.
2575          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2576          */
2577         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2578                 btrfs_err(fs_info, "superblock checksum mismatch");
2579                 err = -EINVAL;
2580                 brelse(bh);
2581                 goto fail_alloc;
2582         }
2583
2584         /*
2585          * super_copy is zeroed at allocation time and we never touch the
2586          * following bytes up to INFO_SIZE, the checksum is calculated from
2587          * the whole block of INFO_SIZE
2588          */
2589         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2590         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2591                sizeof(*fs_info->super_for_commit));
2592         brelse(bh);
2593
2594         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2595
2596         ret = btrfs_check_super_valid(fs_info);
2597         if (ret) {
2598                 btrfs_err(fs_info, "superblock contains fatal errors");
2599                 err = -EINVAL;
2600                 goto fail_alloc;
2601         }
2602
2603         disk_super = fs_info->super_copy;
2604         if (!btrfs_super_root(disk_super))
2605                 goto fail_alloc;
2606
2607         /* check FS state, whether FS is broken. */
2608         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2609                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2610
2611         /*
2612          * run through our array of backup supers and setup
2613          * our ring pointer to the oldest one
2614          */
2615         generation = btrfs_super_generation(disk_super);
2616         find_oldest_super_backup(fs_info, generation);
2617
2618         /*
2619          * In the long term, we'll store the compression type in the super
2620          * block, and it'll be used for per file compression control.
2621          */
2622         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2623
2624         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2625         if (ret) {
2626                 err = ret;
2627                 goto fail_alloc;
2628         }
2629
2630         features = btrfs_super_incompat_flags(disk_super) &
2631                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2632         if (features) {
2633                 btrfs_err(fs_info,
2634                     "cannot mount because of unsupported optional features (%llx)",
2635                     features);
2636                 err = -EINVAL;
2637                 goto fail_alloc;
2638         }
2639
2640         features = btrfs_super_incompat_flags(disk_super);
2641         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2642         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2643                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2644         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2645                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2646
2647         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2648                 btrfs_info(fs_info, "has skinny extents");
2649
2650         /*
2651          * flag our filesystem as having big metadata blocks if
2652          * they are bigger than the page size
2653          */
2654         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2655                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2656                         btrfs_info(fs_info,
2657                                 "flagging fs with big metadata feature");
2658                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2659         }
2660
2661         nodesize = btrfs_super_nodesize(disk_super);
2662         sectorsize = btrfs_super_sectorsize(disk_super);
2663         stripesize = sectorsize;
2664         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2665         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2666
2667         /* Cache block sizes */
2668         fs_info->nodesize = nodesize;
2669         fs_info->sectorsize = sectorsize;
2670         fs_info->stripesize = stripesize;
2671
2672         /*
2673          * mixed block groups end up with duplicate but slightly offset
2674          * extent buffers for the same range.  It leads to corruptions
2675          */
2676         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2677             (sectorsize != nodesize)) {
2678                 btrfs_err(fs_info,
2679 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2680                         nodesize, sectorsize);
2681                 goto fail_alloc;
2682         }
2683
2684         /*
2685          * Needn't use the lock because there is no other task which will
2686          * update the flag.
2687          */
2688         btrfs_set_super_incompat_flags(disk_super, features);
2689
2690         features = btrfs_super_compat_ro_flags(disk_super) &
2691                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2692         if (!sb_rdonly(sb) && features) {
2693                 btrfs_err(fs_info,
2694         "cannot mount read-write because of unsupported optional features (%llx)",
2695                        features);
2696                 err = -EINVAL;
2697                 goto fail_alloc;
2698         }
2699
2700         ret = btrfs_init_workqueues(fs_info, fs_devices);
2701         if (ret) {
2702                 err = ret;
2703                 goto fail_sb_buffer;
2704         }
2705
2706         sb->s_bdi->congested_fn = btrfs_congested_fn;
2707         sb->s_bdi->congested_data = fs_info;
2708         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2709         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2710         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2711         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2712
2713         sb->s_blocksize = sectorsize;
2714         sb->s_blocksize_bits = blksize_bits(sectorsize);
2715         memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2716
2717         mutex_lock(&fs_info->chunk_mutex);
2718         ret = btrfs_read_sys_array(fs_info);
2719         mutex_unlock(&fs_info->chunk_mutex);
2720         if (ret) {
2721                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2722                 goto fail_sb_buffer;
2723         }
2724
2725         generation = btrfs_super_chunk_root_generation(disk_super);
2726
2727         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2728
2729         chunk_root->node = read_tree_block(fs_info,
2730                                            btrfs_super_chunk_root(disk_super),
2731                                            generation);
2732         if (IS_ERR(chunk_root->node) ||
2733             !extent_buffer_uptodate(chunk_root->node)) {
2734                 btrfs_err(fs_info, "failed to read chunk root");
2735                 if (!IS_ERR(chunk_root->node))
2736                         free_extent_buffer(chunk_root->node);
2737                 chunk_root->node = NULL;
2738                 goto fail_tree_roots;
2739         }
2740         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2741         chunk_root->commit_root = btrfs_root_node(chunk_root);
2742
2743         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2744            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2745
2746         ret = btrfs_read_chunk_tree(fs_info);
2747         if (ret) {
2748                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2749                 goto fail_tree_roots;
2750         }
2751
2752         /*
2753          * Keep the devid that is marked to be the target device for the
2754          * device replace procedure
2755          */
2756         btrfs_free_extra_devids(fs_devices, 0);
2757
2758         if (!fs_devices->latest_bdev) {
2759                 btrfs_err(fs_info, "failed to read devices");
2760                 goto fail_tree_roots;
2761         }
2762
2763 retry_root_backup:
2764         generation = btrfs_super_generation(disk_super);
2765
2766         tree_root->node = read_tree_block(fs_info,
2767                                           btrfs_super_root(disk_super),
2768                                           generation);
2769         if (IS_ERR(tree_root->node) ||
2770             !extent_buffer_uptodate(tree_root->node)) {
2771                 btrfs_warn(fs_info, "failed to read tree root");
2772                 if (!IS_ERR(tree_root->node))
2773                         free_extent_buffer(tree_root->node);
2774                 tree_root->node = NULL;
2775                 goto recovery_tree_root;
2776         }
2777
2778         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2779         tree_root->commit_root = btrfs_root_node(tree_root);
2780         btrfs_set_root_refs(&tree_root->root_item, 1);
2781
2782         mutex_lock(&tree_root->objectid_mutex);
2783         ret = btrfs_find_highest_objectid(tree_root,
2784                                         &tree_root->highest_objectid);
2785         if (ret) {
2786                 mutex_unlock(&tree_root->objectid_mutex);
2787                 goto recovery_tree_root;
2788         }
2789
2790         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2791
2792         mutex_unlock(&tree_root->objectid_mutex);
2793
2794         ret = btrfs_read_roots(fs_info);
2795         if (ret)
2796                 goto recovery_tree_root;
2797
2798         fs_info->generation = generation;
2799         fs_info->last_trans_committed = generation;
2800
2801         ret = btrfs_recover_balance(fs_info);
2802         if (ret) {
2803                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2804                 goto fail_block_groups;
2805         }
2806
2807         ret = btrfs_init_dev_stats(fs_info);
2808         if (ret) {
2809                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2810                 goto fail_block_groups;
2811         }
2812
2813         ret = btrfs_init_dev_replace(fs_info);
2814         if (ret) {
2815                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2816                 goto fail_block_groups;
2817         }
2818
2819         btrfs_free_extra_devids(fs_devices, 1);
2820
2821         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2822         if (ret) {
2823                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2824                                 ret);
2825                 goto fail_block_groups;
2826         }
2827
2828         ret = btrfs_sysfs_add_device(fs_devices);
2829         if (ret) {
2830                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2831                                 ret);
2832                 goto fail_fsdev_sysfs;
2833         }
2834
2835         ret = btrfs_sysfs_add_mounted(fs_info);
2836         if (ret) {
2837                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2838                 goto fail_fsdev_sysfs;
2839         }
2840
2841         ret = btrfs_init_space_info(fs_info);
2842         if (ret) {
2843                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2844                 goto fail_sysfs;
2845         }
2846
2847         ret = btrfs_read_block_groups(fs_info);
2848         if (ret) {
2849                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2850                 goto fail_sysfs;
2851         }
2852
2853         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
2854                 btrfs_warn(fs_info,
2855                 "writeable mount is not allowed due to too many missing devices");
2856                 goto fail_sysfs;
2857         }
2858
2859         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2860                                                "btrfs-cleaner");
2861         if (IS_ERR(fs_info->cleaner_kthread))
2862                 goto fail_sysfs;
2863
2864         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2865                                                    tree_root,
2866                                                    "btrfs-transaction");
2867         if (IS_ERR(fs_info->transaction_kthread))
2868                 goto fail_cleaner;
2869
2870         if (!btrfs_test_opt(fs_info, NOSSD) &&
2871             !fs_info->fs_devices->rotating) {
2872                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2873         }
2874
2875         /*
2876          * Mount does not set all options immediately, we can do it now and do
2877          * not have to wait for transaction commit
2878          */
2879         btrfs_apply_pending_changes(fs_info);
2880
2881 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2882         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2883                 ret = btrfsic_mount(fs_info, fs_devices,
2884                                     btrfs_test_opt(fs_info,
2885                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2886                                     1 : 0,
2887                                     fs_info->check_integrity_print_mask);
2888                 if (ret)
2889                         btrfs_warn(fs_info,
2890                                 "failed to initialize integrity check module: %d",
2891                                 ret);
2892         }
2893 #endif
2894         ret = btrfs_read_qgroup_config(fs_info);
2895         if (ret)
2896                 goto fail_trans_kthread;
2897
2898         if (btrfs_build_ref_tree(fs_info))
2899                 btrfs_err(fs_info, "couldn't build ref tree");
2900
2901         /* do not make disk changes in broken FS or nologreplay is given */
2902         if (btrfs_super_log_root(disk_super) != 0 &&
2903             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2904                 ret = btrfs_replay_log(fs_info, fs_devices);
2905                 if (ret) {
2906                         err = ret;
2907                         goto fail_qgroup;
2908                 }
2909         }
2910
2911         ret = btrfs_find_orphan_roots(fs_info);
2912         if (ret)
2913                 goto fail_qgroup;
2914
2915         if (!sb_rdonly(sb)) {
2916                 ret = btrfs_cleanup_fs_roots(fs_info);
2917                 if (ret)
2918                         goto fail_qgroup;
2919
2920                 mutex_lock(&fs_info->cleaner_mutex);
2921                 ret = btrfs_recover_relocation(tree_root);
2922                 mutex_unlock(&fs_info->cleaner_mutex);
2923                 if (ret < 0) {
2924                         btrfs_warn(fs_info, "failed to recover relocation: %d",
2925                                         ret);
2926                         err = -EINVAL;
2927                         goto fail_qgroup;
2928                 }
2929         }
2930
2931         location.objectid = BTRFS_FS_TREE_OBJECTID;
2932         location.type = BTRFS_ROOT_ITEM_KEY;
2933         location.offset = 0;
2934
2935         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2936         if (IS_ERR(fs_info->fs_root)) {
2937                 err = PTR_ERR(fs_info->fs_root);
2938                 goto fail_qgroup;
2939         }
2940
2941         if (sb_rdonly(sb))
2942                 return 0;
2943
2944         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2945             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2946                 clear_free_space_tree = 1;
2947         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2948                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2949                 btrfs_warn(fs_info, "free space tree is invalid");
2950                 clear_free_space_tree = 1;
2951         }
2952
2953         if (clear_free_space_tree) {
2954                 btrfs_info(fs_info, "clearing free space tree");
2955                 ret = btrfs_clear_free_space_tree(fs_info);
2956                 if (ret) {
2957                         btrfs_warn(fs_info,
2958                                    "failed to clear free space tree: %d", ret);
2959                         close_ctree(fs_info);
2960                         return ret;
2961                 }
2962         }
2963
2964         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
2965             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2966                 btrfs_info(fs_info, "creating free space tree");
2967                 ret = btrfs_create_free_space_tree(fs_info);
2968                 if (ret) {
2969                         btrfs_warn(fs_info,
2970                                 "failed to create free space tree: %d", ret);
2971                         close_ctree(fs_info);
2972                         return ret;
2973                 }
2974         }
2975
2976         down_read(&fs_info->cleanup_work_sem);
2977         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2978             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2979                 up_read(&fs_info->cleanup_work_sem);
2980                 close_ctree(fs_info);
2981                 return ret;
2982         }
2983         up_read(&fs_info->cleanup_work_sem);
2984
2985         ret = btrfs_resume_balance_async(fs_info);
2986         if (ret) {
2987                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
2988                 close_ctree(fs_info);
2989                 return ret;
2990         }
2991
2992         ret = btrfs_resume_dev_replace_async(fs_info);
2993         if (ret) {
2994                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
2995                 close_ctree(fs_info);
2996                 return ret;
2997         }
2998
2999         btrfs_qgroup_rescan_resume(fs_info);
3000
3001         if (!fs_info->uuid_root) {
3002                 btrfs_info(fs_info, "creating UUID tree");
3003                 ret = btrfs_create_uuid_tree(fs_info);
3004                 if (ret) {
3005                         btrfs_warn(fs_info,
3006                                 "failed to create the UUID tree: %d", ret);
3007                         close_ctree(fs_info);
3008                         return ret;
3009                 }
3010         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3011                    fs_info->generation !=
3012                                 btrfs_super_uuid_tree_generation(disk_super)) {
3013                 btrfs_info(fs_info, "checking UUID tree");
3014                 ret = btrfs_check_uuid_tree(fs_info);
3015                 if (ret) {
3016                         btrfs_warn(fs_info,
3017                                 "failed to check the UUID tree: %d", ret);
3018                         close_ctree(fs_info);
3019                         return ret;
3020                 }
3021         } else {
3022                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3023         }
3024         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3025
3026         /*
3027          * backuproot only affect mount behavior, and if open_ctree succeeded,
3028          * no need to keep the flag
3029          */
3030         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3031
3032         return 0;
3033
3034 fail_qgroup:
3035         btrfs_free_qgroup_config(fs_info);
3036 fail_trans_kthread:
3037         kthread_stop(fs_info->transaction_kthread);
3038         btrfs_cleanup_transaction(fs_info);
3039         btrfs_free_fs_roots(fs_info);
3040 fail_cleaner:
3041         kthread_stop(fs_info->cleaner_kthread);
3042
3043         /*
3044          * make sure we're done with the btree inode before we stop our
3045          * kthreads
3046          */
3047         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3048
3049 fail_sysfs:
3050         btrfs_sysfs_remove_mounted(fs_info);
3051
3052 fail_fsdev_sysfs:
3053         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3054
3055 fail_block_groups:
3056         btrfs_put_block_group_cache(fs_info);
3057
3058 fail_tree_roots:
3059         free_root_pointers(fs_info, 1);
3060         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3061
3062 fail_sb_buffer:
3063         btrfs_stop_all_workers(fs_info);
3064         btrfs_free_block_groups(fs_info);
3065 fail_alloc:
3066 fail_iput:
3067         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3068
3069         iput(fs_info->btree_inode);
3070 fail_bio_counter:
3071         percpu_counter_destroy(&fs_info->bio_counter);
3072 fail_delalloc_bytes:
3073         percpu_counter_destroy(&fs_info->delalloc_bytes);
3074 fail_dirty_metadata_bytes:
3075         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3076 fail_srcu:
3077         cleanup_srcu_struct(&fs_info->subvol_srcu);
3078 fail:
3079         btrfs_free_stripe_hash_table(fs_info);
3080         btrfs_close_devices(fs_info->fs_devices);
3081         return err;
3082
3083 recovery_tree_root:
3084         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3085                 goto fail_tree_roots;
3086
3087         free_root_pointers(fs_info, 0);
3088
3089         /* don't use the log in recovery mode, it won't be valid */
3090         btrfs_set_super_log_root(disk_super, 0);
3091
3092         /* we can't trust the free space cache either */
3093         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3094
3095         ret = next_root_backup(fs_info, fs_info->super_copy,
3096                                &num_backups_tried, &backup_index);
3097         if (ret == -1)
3098                 goto fail_block_groups;
3099         goto retry_root_backup;
3100 }
3101 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3102
3103 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3104 {
3105         if (uptodate) {
3106                 set_buffer_uptodate(bh);
3107         } else {
3108                 struct btrfs_device *device = (struct btrfs_device *)
3109                         bh->b_private;
3110
3111                 btrfs_warn_rl_in_rcu(device->fs_info,
3112                                 "lost page write due to IO error on %s",
3113                                           rcu_str_deref(device->name));
3114                 /* note, we don't set_buffer_write_io_error because we have
3115                  * our own ways of dealing with the IO errors
3116                  */
3117                 clear_buffer_uptodate(bh);
3118                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3119         }
3120         unlock_buffer(bh);
3121         put_bh(bh);
3122 }
3123
3124 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3125                         struct buffer_head **bh_ret)
3126 {
3127         struct buffer_head *bh;
3128         struct btrfs_super_block *super;
3129         u64 bytenr;
3130
3131         bytenr = btrfs_sb_offset(copy_num);
3132         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3133                 return -EINVAL;
3134
3135         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3136         /*
3137          * If we fail to read from the underlying devices, as of now
3138          * the best option we have is to mark it EIO.
3139          */
3140         if (!bh)
3141                 return -EIO;
3142
3143         super = (struct btrfs_super_block *)bh->b_data;
3144         if (btrfs_super_bytenr(super) != bytenr ||
3145                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3146                 brelse(bh);
3147                 return -EINVAL;
3148         }
3149
3150         *bh_ret = bh;
3151         return 0;
3152 }
3153
3154
3155 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3156 {
3157         struct buffer_head *bh;
3158         struct buffer_head *latest = NULL;
3159         struct btrfs_super_block *super;
3160         int i;
3161         u64 transid = 0;
3162         int ret = -EINVAL;
3163
3164         /* we would like to check all the supers, but that would make
3165          * a btrfs mount succeed after a mkfs from a different FS.
3166          * So, we need to add a special mount option to scan for
3167          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3168          */
3169         for (i = 0; i < 1; i++) {
3170                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3171                 if (ret)
3172                         continue;
3173
3174                 super = (struct btrfs_super_block *)bh->b_data;
3175
3176                 if (!latest || btrfs_super_generation(super) > transid) {
3177                         brelse(latest);
3178                         latest = bh;
3179                         transid = btrfs_super_generation(super);
3180                 } else {
3181                         brelse(bh);
3182                 }
3183         }
3184
3185         if (!latest)
3186                 return ERR_PTR(ret);
3187
3188         return latest;
3189 }
3190
3191 /*
3192  * Write superblock @sb to the @device. Do not wait for completion, all the
3193  * buffer heads we write are pinned.
3194  *
3195  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3196  * the expected device size at commit time. Note that max_mirrors must be
3197  * same for write and wait phases.
3198  *
3199  * Return number of errors when buffer head is not found or submission fails.
3200  */
3201 static int write_dev_supers(struct btrfs_device *device,
3202                             struct btrfs_super_block *sb, int max_mirrors)
3203 {
3204         struct buffer_head *bh;
3205         int i;
3206         int ret;
3207         int errors = 0;
3208         u32 crc;
3209         u64 bytenr;
3210         int op_flags;
3211
3212         if (max_mirrors == 0)
3213                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3214
3215         for (i = 0; i < max_mirrors; i++) {
3216                 bytenr = btrfs_sb_offset(i);
3217                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3218                     device->commit_total_bytes)
3219                         break;
3220
3221                 btrfs_set_super_bytenr(sb, bytenr);
3222
3223                 crc = ~(u32)0;
3224                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3225                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3226                 btrfs_csum_final(crc, sb->csum);
3227
3228                 /* One reference for us, and we leave it for the caller */
3229                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3230                               BTRFS_SUPER_INFO_SIZE);
3231                 if (!bh) {
3232                         btrfs_err(device->fs_info,
3233                             "couldn't get super buffer head for bytenr %llu",
3234                             bytenr);
3235                         errors++;
3236                         continue;
3237                 }
3238
3239                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3240
3241                 /* one reference for submit_bh */
3242                 get_bh(bh);
3243
3244                 set_buffer_uptodate(bh);
3245                 lock_buffer(bh);
3246                 bh->b_end_io = btrfs_end_buffer_write_sync;
3247                 bh->b_private = device;
3248
3249                 /*
3250                  * we fua the first super.  The others we allow
3251                  * to go down lazy.
3252                  */
3253                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3254                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3255                         op_flags |= REQ_FUA;
3256                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3257                 if (ret)
3258                         errors++;
3259         }
3260         return errors < i ? 0 : -1;
3261 }
3262
3263 /*
3264  * Wait for write completion of superblocks done by write_dev_supers,
3265  * @max_mirrors same for write and wait phases.
3266  *
3267  * Return number of errors when buffer head is not found or not marked up to
3268  * date.
3269  */
3270 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3271 {
3272         struct buffer_head *bh;
3273         int i;
3274         int errors = 0;
3275         bool primary_failed = false;
3276         u64 bytenr;
3277
3278         if (max_mirrors == 0)
3279                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3280
3281         for (i = 0; i < max_mirrors; i++) {