3830867e0225a0361b01e40229df01eaf12c433c
[sfrench/cifs-2.6.git] / fs / btrfs / disk-io.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5
6 #include <linux/fs.h>
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <asm/unaligned.h>
21 #include "ctree.h"
22 #include "disk-io.h"
23 #include "transaction.h"
24 #include "btrfs_inode.h"
25 #include "volumes.h"
26 #include "print-tree.h"
27 #include "locking.h"
28 #include "tree-log.h"
29 #include "free-space-cache.h"
30 #include "free-space-tree.h"
31 #include "inode-map.h"
32 #include "check-integrity.h"
33 #include "rcu-string.h"
34 #include "dev-replace.h"
35 #include "raid56.h"
36 #include "sysfs.h"
37 #include "qgroup.h"
38 #include "compression.h"
39 #include "tree-checker.h"
40 #include "ref-verify.h"
41
42 #ifdef CONFIG_X86
43 #include <asm/cpufeature.h>
44 #endif
45
46 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
47                                  BTRFS_HEADER_FLAG_RELOC |\
48                                  BTRFS_SUPER_FLAG_ERROR |\
49                                  BTRFS_SUPER_FLAG_SEEDING |\
50                                  BTRFS_SUPER_FLAG_METADUMP |\
51                                  BTRFS_SUPER_FLAG_METADUMP_V2)
52
53 static const struct extent_io_ops btree_extent_io_ops;
54 static void end_workqueue_fn(struct btrfs_work *work);
55 static void free_fs_root(struct btrfs_root *root);
56 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
57 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
58                                       struct btrfs_fs_info *fs_info);
59 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
60 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
61                                         struct extent_io_tree *dirty_pages,
62                                         int mark);
63 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
64                                        struct extent_io_tree *pinned_extents);
65 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
66 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
67
68 /*
69  * btrfs_end_io_wq structs are used to do processing in task context when an IO
70  * is complete.  This is used during reads to verify checksums, and it is used
71  * by writes to insert metadata for new file extents after IO is complete.
72  */
73 struct btrfs_end_io_wq {
74         struct bio *bio;
75         bio_end_io_t *end_io;
76         void *private;
77         struct btrfs_fs_info *info;
78         blk_status_t status;
79         enum btrfs_wq_endio_type metadata;
80         struct btrfs_work work;
81 };
82
83 static struct kmem_cache *btrfs_end_io_wq_cache;
84
85 int __init btrfs_end_io_wq_init(void)
86 {
87         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
88                                         sizeof(struct btrfs_end_io_wq),
89                                         0,
90                                         SLAB_MEM_SPREAD,
91                                         NULL);
92         if (!btrfs_end_io_wq_cache)
93                 return -ENOMEM;
94         return 0;
95 }
96
97 void __cold btrfs_end_io_wq_exit(void)
98 {
99         kmem_cache_destroy(btrfs_end_io_wq_cache);
100 }
101
102 /*
103  * async submit bios are used to offload expensive checksumming
104  * onto the worker threads.  They checksum file and metadata bios
105  * just before they are sent down the IO stack.
106  */
107 struct async_submit_bio {
108         void *private_data;
109         struct bio *bio;
110         extent_submit_bio_start_t *submit_bio_start;
111         int mirror_num;
112         /*
113          * bio_offset is optional, can be used if the pages in the bio
114          * can't tell us where in the file the bio should go
115          */
116         u64 bio_offset;
117         struct btrfs_work work;
118         blk_status_t status;
119 };
120
121 /*
122  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
123  * eb, the lockdep key is determined by the btrfs_root it belongs to and
124  * the level the eb occupies in the tree.
125  *
126  * Different roots are used for different purposes and may nest inside each
127  * other and they require separate keysets.  As lockdep keys should be
128  * static, assign keysets according to the purpose of the root as indicated
129  * by btrfs_root->objectid.  This ensures that all special purpose roots
130  * have separate keysets.
131  *
132  * Lock-nesting across peer nodes is always done with the immediate parent
133  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
134  * subclass to avoid triggering lockdep warning in such cases.
135  *
136  * The key is set by the readpage_end_io_hook after the buffer has passed
137  * csum validation but before the pages are unlocked.  It is also set by
138  * btrfs_init_new_buffer on freshly allocated blocks.
139  *
140  * We also add a check to make sure the highest level of the tree is the
141  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
142  * needs update as well.
143  */
144 #ifdef CONFIG_DEBUG_LOCK_ALLOC
145 # if BTRFS_MAX_LEVEL != 8
146 #  error
147 # endif
148
149 static struct btrfs_lockdep_keyset {
150         u64                     id;             /* root objectid */
151         const char              *name_stem;     /* lock name stem */
152         char                    names[BTRFS_MAX_LEVEL + 1][20];
153         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
154 } btrfs_lockdep_keysets[] = {
155         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
156         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
157         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
158         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
159         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
160         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
161         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
162         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
163         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
164         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
165         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
166         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
167         { .id = 0,                              .name_stem = "tree"     },
168 };
169
170 void __init btrfs_init_lockdep(void)
171 {
172         int i, j;
173
174         /* initialize lockdep class names */
175         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
176                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
177
178                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
179                         snprintf(ks->names[j], sizeof(ks->names[j]),
180                                  "btrfs-%s-%02d", ks->name_stem, j);
181         }
182 }
183
184 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
185                                     int level)
186 {
187         struct btrfs_lockdep_keyset *ks;
188
189         BUG_ON(level >= ARRAY_SIZE(ks->keys));
190
191         /* find the matching keyset, id 0 is the default entry */
192         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
193                 if (ks->id == objectid)
194                         break;
195
196         lockdep_set_class_and_name(&eb->lock,
197                                    &ks->keys[level], ks->names[level]);
198 }
199
200 #endif
201
202 /*
203  * extents on the btree inode are pretty simple, there's one extent
204  * that covers the entire device
205  */
206 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
207                 struct page *page, size_t pg_offset, u64 start, u64 len,
208                 int create)
209 {
210         struct btrfs_fs_info *fs_info = inode->root->fs_info;
211         struct extent_map_tree *em_tree = &inode->extent_tree;
212         struct extent_map *em;
213         int ret;
214
215         read_lock(&em_tree->lock);
216         em = lookup_extent_mapping(em_tree, start, len);
217         if (em) {
218                 em->bdev = fs_info->fs_devices->latest_bdev;
219                 read_unlock(&em_tree->lock);
220                 goto out;
221         }
222         read_unlock(&em_tree->lock);
223
224         em = alloc_extent_map();
225         if (!em) {
226                 em = ERR_PTR(-ENOMEM);
227                 goto out;
228         }
229         em->start = 0;
230         em->len = (u64)-1;
231         em->block_len = (u64)-1;
232         em->block_start = 0;
233         em->bdev = fs_info->fs_devices->latest_bdev;
234
235         write_lock(&em_tree->lock);
236         ret = add_extent_mapping(em_tree, em, 0);
237         if (ret == -EEXIST) {
238                 free_extent_map(em);
239                 em = lookup_extent_mapping(em_tree, start, len);
240                 if (!em)
241                         em = ERR_PTR(-EIO);
242         } else if (ret) {
243                 free_extent_map(em);
244                 em = ERR_PTR(ret);
245         }
246         write_unlock(&em_tree->lock);
247
248 out:
249         return em;
250 }
251
252 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
253 {
254         return crc32c(seed, data, len);
255 }
256
257 void btrfs_csum_final(u32 crc, u8 *result)
258 {
259         put_unaligned_le32(~crc, result);
260 }
261
262 /*
263  * compute the csum for a btree block, and either verify it or write it
264  * into the csum field of the block.
265  */
266 static int csum_tree_block(struct btrfs_fs_info *fs_info,
267                            struct extent_buffer *buf,
268                            int verify)
269 {
270         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
271         char result[BTRFS_CSUM_SIZE];
272         unsigned long len;
273         unsigned long cur_len;
274         unsigned long offset = BTRFS_CSUM_SIZE;
275         char *kaddr;
276         unsigned long map_start;
277         unsigned long map_len;
278         int err;
279         u32 crc = ~(u32)0;
280
281         len = buf->len - offset;
282         while (len > 0) {
283                 err = map_private_extent_buffer(buf, offset, 32,
284                                         &kaddr, &map_start, &map_len);
285                 if (err)
286                         return err;
287                 cur_len = min(len, map_len - (offset - map_start));
288                 crc = btrfs_csum_data(kaddr + offset - map_start,
289                                       crc, cur_len);
290                 len -= cur_len;
291                 offset += cur_len;
292         }
293         memset(result, 0, BTRFS_CSUM_SIZE);
294
295         btrfs_csum_final(crc, result);
296
297         if (verify) {
298                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
299                         u32 val;
300                         u32 found = 0;
301                         memcpy(&found, result, csum_size);
302
303                         read_extent_buffer(buf, &val, 0, csum_size);
304                         btrfs_warn_rl(fs_info,
305                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
306                                 fs_info->sb->s_id, buf->start,
307                                 val, found, btrfs_header_level(buf));
308                         return -EUCLEAN;
309                 }
310         } else {
311                 write_extent_buffer(buf, result, 0, csum_size);
312         }
313
314         return 0;
315 }
316
317 /*
318  * we can't consider a given block up to date unless the transid of the
319  * block matches the transid in the parent node's pointer.  This is how we
320  * detect blocks that either didn't get written at all or got written
321  * in the wrong place.
322  */
323 static int verify_parent_transid(struct extent_io_tree *io_tree,
324                                  struct extent_buffer *eb, u64 parent_transid,
325                                  int atomic)
326 {
327         struct extent_state *cached_state = NULL;
328         int ret;
329         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
330
331         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
332                 return 0;
333
334         if (atomic)
335                 return -EAGAIN;
336
337         if (need_lock) {
338                 btrfs_tree_read_lock(eb);
339                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
340         }
341
342         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
343                          &cached_state);
344         if (extent_buffer_uptodate(eb) &&
345             btrfs_header_generation(eb) == parent_transid) {
346                 ret = 0;
347                 goto out;
348         }
349         btrfs_err_rl(eb->fs_info,
350                 "parent transid verify failed on %llu wanted %llu found %llu",
351                         eb->start,
352                         parent_transid, btrfs_header_generation(eb));
353         ret = 1;
354
355         /*
356          * Things reading via commit roots that don't have normal protection,
357          * like send, can have a really old block in cache that may point at a
358          * block that has been freed and re-allocated.  So don't clear uptodate
359          * if we find an eb that is under IO (dirty/writeback) because we could
360          * end up reading in the stale data and then writing it back out and
361          * making everybody very sad.
362          */
363         if (!extent_buffer_under_io(eb))
364                 clear_extent_buffer_uptodate(eb);
365 out:
366         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
367                              &cached_state);
368         if (need_lock)
369                 btrfs_tree_read_unlock_blocking(eb);
370         return ret;
371 }
372
373 /*
374  * Return 0 if the superblock checksum type matches the checksum value of that
375  * algorithm. Pass the raw disk superblock data.
376  */
377 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
378                                   char *raw_disk_sb)
379 {
380         struct btrfs_super_block *disk_sb =
381                 (struct btrfs_super_block *)raw_disk_sb;
382         u16 csum_type = btrfs_super_csum_type(disk_sb);
383         int ret = 0;
384
385         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
386                 u32 crc = ~(u32)0;
387                 char result[sizeof(crc)];
388
389                 /*
390                  * The super_block structure does not span the whole
391                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
392                  * is filled with zeros and is included in the checksum.
393                  */
394                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
395                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
396                 btrfs_csum_final(crc, result);
397
398                 if (memcmp(raw_disk_sb, result, sizeof(result)))
399                         ret = 1;
400         }
401
402         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
403                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
404                                 csum_type);
405                 ret = 1;
406         }
407
408         return ret;
409 }
410
411 static int verify_level_key(struct btrfs_fs_info *fs_info,
412                             struct extent_buffer *eb, int level,
413                             struct btrfs_key *first_key, u64 parent_transid)
414 {
415         int found_level;
416         struct btrfs_key found_key;
417         int ret;
418
419         found_level = btrfs_header_level(eb);
420         if (found_level != level) {
421 #ifdef CONFIG_BTRFS_DEBUG
422                 WARN_ON(1);
423                 btrfs_err(fs_info,
424 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
425                           eb->start, level, found_level);
426 #endif
427                 return -EIO;
428         }
429
430         if (!first_key)
431                 return 0;
432
433         /*
434          * For live tree block (new tree blocks in current transaction),
435          * we need proper lock context to avoid race, which is impossible here.
436          * So we only checks tree blocks which is read from disk, whose
437          * generation <= fs_info->last_trans_committed.
438          */
439         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
440                 return 0;
441         if (found_level)
442                 btrfs_node_key_to_cpu(eb, &found_key, 0);
443         else
444                 btrfs_item_key_to_cpu(eb, &found_key, 0);
445         ret = btrfs_comp_cpu_keys(first_key, &found_key);
446
447 #ifdef CONFIG_BTRFS_DEBUG
448         if (ret) {
449                 WARN_ON(1);
450                 btrfs_err(fs_info,
451 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
452                           eb->start, parent_transid, first_key->objectid,
453                           first_key->type, first_key->offset,
454                           found_key.objectid, found_key.type,
455                           found_key.offset);
456         }
457 #endif
458         return ret;
459 }
460
461 /*
462  * helper to read a given tree block, doing retries as required when
463  * the checksums don't match and we have alternate mirrors to try.
464  *
465  * @parent_transid:     expected transid, skip check if 0
466  * @level:              expected level, mandatory check
467  * @first_key:          expected key of first slot, skip check if NULL
468  */
469 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
470                                           struct extent_buffer *eb,
471                                           u64 parent_transid, int level,
472                                           struct btrfs_key *first_key)
473 {
474         struct extent_io_tree *io_tree;
475         int failed = 0;
476         int ret;
477         int num_copies = 0;
478         int mirror_num = 0;
479         int failed_mirror = 0;
480
481         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
482         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
483         while (1) {
484                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
485                                                mirror_num);
486                 if (!ret) {
487                         if (verify_parent_transid(io_tree, eb,
488                                                    parent_transid, 0))
489                                 ret = -EIO;
490                         else if (verify_level_key(fs_info, eb, level,
491                                                   first_key, parent_transid))
492                                 ret = -EUCLEAN;
493                         else
494                                 break;
495                 }
496
497                 /*
498                  * This buffer's crc is fine, but its contents are corrupted, so
499                  * there is no reason to read the other copies, they won't be
500                  * any less wrong.
501                  */
502                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags) ||
503                     ret == -EUCLEAN)
504                         break;
505
506                 num_copies = btrfs_num_copies(fs_info,
507                                               eb->start, eb->len);
508                 if (num_copies == 1)
509                         break;
510
511                 if (!failed_mirror) {
512                         failed = 1;
513                         failed_mirror = eb->read_mirror;
514                 }
515
516                 mirror_num++;
517                 if (mirror_num == failed_mirror)
518                         mirror_num++;
519
520                 if (mirror_num > num_copies)
521                         break;
522         }
523
524         if (failed && !ret && failed_mirror)
525                 repair_eb_io_failure(fs_info, eb, failed_mirror);
526
527         return ret;
528 }
529
530 /*
531  * checksum a dirty tree block before IO.  This has extra checks to make sure
532  * we only fill in the checksum field in the first page of a multi-page block
533  */
534
535 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
536 {
537         u64 start = page_offset(page);
538         u64 found_start;
539         struct extent_buffer *eb;
540
541         eb = (struct extent_buffer *)page->private;
542         if (page != eb->pages[0])
543                 return 0;
544
545         found_start = btrfs_header_bytenr(eb);
546         /*
547          * Please do not consolidate these warnings into a single if.
548          * It is useful to know what went wrong.
549          */
550         if (WARN_ON(found_start != start))
551                 return -EUCLEAN;
552         if (WARN_ON(!PageUptodate(page)))
553                 return -EUCLEAN;
554
555         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
556                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
557
558         return csum_tree_block(fs_info, eb, 0);
559 }
560
561 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
562                                  struct extent_buffer *eb)
563 {
564         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
565         u8 fsid[BTRFS_FSID_SIZE];
566         int ret = 1;
567
568         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
569         while (fs_devices) {
570                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
571                         ret = 0;
572                         break;
573                 }
574                 fs_devices = fs_devices->seed;
575         }
576         return ret;
577 }
578
579 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
580                                       u64 phy_offset, struct page *page,
581                                       u64 start, u64 end, int mirror)
582 {
583         u64 found_start;
584         int found_level;
585         struct extent_buffer *eb;
586         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
587         struct btrfs_fs_info *fs_info = root->fs_info;
588         int ret = 0;
589         int reads_done;
590
591         if (!page->private)
592                 goto out;
593
594         eb = (struct extent_buffer *)page->private;
595
596         /* the pending IO might have been the only thing that kept this buffer
597          * in memory.  Make sure we have a ref for all this other checks
598          */
599         extent_buffer_get(eb);
600
601         reads_done = atomic_dec_and_test(&eb->io_pages);
602         if (!reads_done)
603                 goto err;
604
605         eb->read_mirror = mirror;
606         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
607                 ret = -EIO;
608                 goto err;
609         }
610
611         found_start = btrfs_header_bytenr(eb);
612         if (found_start != eb->start) {
613                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
614                              eb->start, found_start);
615                 ret = -EIO;
616                 goto err;
617         }
618         if (check_tree_block_fsid(fs_info, eb)) {
619                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
620                              eb->start);
621                 ret = -EIO;
622                 goto err;
623         }
624         found_level = btrfs_header_level(eb);
625         if (found_level >= BTRFS_MAX_LEVEL) {
626                 btrfs_err(fs_info, "bad tree block level %d on %llu",
627                           (int)btrfs_header_level(eb), eb->start);
628                 ret = -EIO;
629                 goto err;
630         }
631
632         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
633                                        eb, found_level);
634
635         ret = csum_tree_block(fs_info, eb, 1);
636         if (ret)
637                 goto err;
638
639         /*
640          * If this is a leaf block and it is corrupt, set the corrupt bit so
641          * that we don't try and read the other copies of this block, just
642          * return -EIO.
643          */
644         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
645                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
646                 ret = -EIO;
647         }
648
649         if (found_level > 0 && btrfs_check_node(fs_info, eb))
650                 ret = -EIO;
651
652         if (!ret)
653                 set_extent_buffer_uptodate(eb);
654 err:
655         if (reads_done &&
656             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
657                 btree_readahead_hook(eb, ret);
658
659         if (ret) {
660                 /*
661                  * our io error hook is going to dec the io pages
662                  * again, we have to make sure it has something
663                  * to decrement
664                  */
665                 atomic_inc(&eb->io_pages);
666                 clear_extent_buffer_uptodate(eb);
667         }
668         free_extent_buffer(eb);
669 out:
670         return ret;
671 }
672
673 static int btree_io_failed_hook(struct page *page, int failed_mirror)
674 {
675         struct extent_buffer *eb;
676
677         eb = (struct extent_buffer *)page->private;
678         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
679         eb->read_mirror = failed_mirror;
680         atomic_dec(&eb->io_pages);
681         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
682                 btree_readahead_hook(eb, -EIO);
683         return -EIO;    /* we fixed nothing */
684 }
685
686 static void end_workqueue_bio(struct bio *bio)
687 {
688         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
689         struct btrfs_fs_info *fs_info;
690         struct btrfs_workqueue *wq;
691         btrfs_work_func_t func;
692
693         fs_info = end_io_wq->info;
694         end_io_wq->status = bio->bi_status;
695
696         if (bio_op(bio) == REQ_OP_WRITE) {
697                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
698                         wq = fs_info->endio_meta_write_workers;
699                         func = btrfs_endio_meta_write_helper;
700                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
701                         wq = fs_info->endio_freespace_worker;
702                         func = btrfs_freespace_write_helper;
703                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
704                         wq = fs_info->endio_raid56_workers;
705                         func = btrfs_endio_raid56_helper;
706                 } else {
707                         wq = fs_info->endio_write_workers;
708                         func = btrfs_endio_write_helper;
709                 }
710         } else {
711                 if (unlikely(end_io_wq->metadata ==
712                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
713                         wq = fs_info->endio_repair_workers;
714                         func = btrfs_endio_repair_helper;
715                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
716                         wq = fs_info->endio_raid56_workers;
717                         func = btrfs_endio_raid56_helper;
718                 } else if (end_io_wq->metadata) {
719                         wq = fs_info->endio_meta_workers;
720                         func = btrfs_endio_meta_helper;
721                 } else {
722                         wq = fs_info->endio_workers;
723                         func = btrfs_endio_helper;
724                 }
725         }
726
727         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
728         btrfs_queue_work(wq, &end_io_wq->work);
729 }
730
731 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
732                         enum btrfs_wq_endio_type metadata)
733 {
734         struct btrfs_end_io_wq *end_io_wq;
735
736         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
737         if (!end_io_wq)
738                 return BLK_STS_RESOURCE;
739
740         end_io_wq->private = bio->bi_private;
741         end_io_wq->end_io = bio->bi_end_io;
742         end_io_wq->info = info;
743         end_io_wq->status = 0;
744         end_io_wq->bio = bio;
745         end_io_wq->metadata = metadata;
746
747         bio->bi_private = end_io_wq;
748         bio->bi_end_io = end_workqueue_bio;
749         return 0;
750 }
751
752 static void run_one_async_start(struct btrfs_work *work)
753 {
754         struct async_submit_bio *async;
755         blk_status_t ret;
756
757         async = container_of(work, struct  async_submit_bio, work);
758         ret = async->submit_bio_start(async->private_data, async->bio,
759                                       async->bio_offset);
760         if (ret)
761                 async->status = ret;
762 }
763
764 static void run_one_async_done(struct btrfs_work *work)
765 {
766         struct async_submit_bio *async;
767
768         async = container_of(work, struct  async_submit_bio, work);
769
770         /* If an error occurred we just want to clean up the bio and move on */
771         if (async->status) {
772                 async->bio->bi_status = async->status;
773                 bio_endio(async->bio);
774                 return;
775         }
776
777         btrfs_submit_bio_done(async->private_data, async->bio, async->mirror_num);
778 }
779
780 static void run_one_async_free(struct btrfs_work *work)
781 {
782         struct async_submit_bio *async;
783
784         async = container_of(work, struct  async_submit_bio, work);
785         kfree(async);
786 }
787
788 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
789                                  int mirror_num, unsigned long bio_flags,
790                                  u64 bio_offset, void *private_data,
791                                  extent_submit_bio_start_t *submit_bio_start)
792 {
793         struct async_submit_bio *async;
794
795         async = kmalloc(sizeof(*async), GFP_NOFS);
796         if (!async)
797                 return BLK_STS_RESOURCE;
798
799         async->private_data = private_data;
800         async->bio = bio;
801         async->mirror_num = mirror_num;
802         async->submit_bio_start = submit_bio_start;
803
804         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
805                         run_one_async_done, run_one_async_free);
806
807         async->bio_offset = bio_offset;
808
809         async->status = 0;
810
811         if (op_is_sync(bio->bi_opf))
812                 btrfs_set_work_high_priority(&async->work);
813
814         btrfs_queue_work(fs_info->workers, &async->work);
815         return 0;
816 }
817
818 static blk_status_t btree_csum_one_bio(struct bio *bio)
819 {
820         struct bio_vec *bvec;
821         struct btrfs_root *root;
822         int i, ret = 0;
823
824         ASSERT(!bio_flagged(bio, BIO_CLONED));
825         bio_for_each_segment_all(bvec, bio, i) {
826                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
827                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
828                 if (ret)
829                         break;
830         }
831
832         return errno_to_blk_status(ret);
833 }
834
835 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
836                                              u64 bio_offset)
837 {
838         /*
839          * when we're called for a write, we're already in the async
840          * submission context.  Just jump into btrfs_map_bio
841          */
842         return btree_csum_one_bio(bio);
843 }
844
845 static int check_async_write(struct btrfs_inode *bi)
846 {
847         if (atomic_read(&bi->sync_writers))
848                 return 0;
849 #ifdef CONFIG_X86
850         if (static_cpu_has(X86_FEATURE_XMM4_2))
851                 return 0;
852 #endif
853         return 1;
854 }
855
856 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
857                                           int mirror_num, unsigned long bio_flags,
858                                           u64 bio_offset)
859 {
860         struct inode *inode = private_data;
861         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
862         int async = check_async_write(BTRFS_I(inode));
863         blk_status_t ret;
864
865         if (bio_op(bio) != REQ_OP_WRITE) {
866                 /*
867                  * called for a read, do the setup so that checksum validation
868                  * can happen in the async kernel threads
869                  */
870                 ret = btrfs_bio_wq_end_io(fs_info, bio,
871                                           BTRFS_WQ_ENDIO_METADATA);
872                 if (ret)
873                         goto out_w_error;
874                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
875         } else if (!async) {
876                 ret = btree_csum_one_bio(bio);
877                 if (ret)
878                         goto out_w_error;
879                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
880         } else {
881                 /*
882                  * kthread helpers are used to submit writes so that
883                  * checksumming can happen in parallel across all CPUs
884                  */
885                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
886                                           bio_offset, private_data,
887                                           btree_submit_bio_start);
888         }
889
890         if (ret)
891                 goto out_w_error;
892         return 0;
893
894 out_w_error:
895         bio->bi_status = ret;
896         bio_endio(bio);
897         return ret;
898 }
899
900 #ifdef CONFIG_MIGRATION
901 static int btree_migratepage(struct address_space *mapping,
902                         struct page *newpage, struct page *page,
903                         enum migrate_mode mode)
904 {
905         /*
906          * we can't safely write a btree page from here,
907          * we haven't done the locking hook
908          */
909         if (PageDirty(page))
910                 return -EAGAIN;
911         /*
912          * Buffers may be managed in a filesystem specific way.
913          * We must have no buffers or drop them.
914          */
915         if (page_has_private(page) &&
916             !try_to_release_page(page, GFP_KERNEL))
917                 return -EAGAIN;
918         return migrate_page(mapping, newpage, page, mode);
919 }
920 #endif
921
922
923 static int btree_writepages(struct address_space *mapping,
924                             struct writeback_control *wbc)
925 {
926         struct btrfs_fs_info *fs_info;
927         int ret;
928
929         if (wbc->sync_mode == WB_SYNC_NONE) {
930
931                 if (wbc->for_kupdate)
932                         return 0;
933
934                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
935                 /* this is a bit racy, but that's ok */
936                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
937                                              BTRFS_DIRTY_METADATA_THRESH,
938                                              fs_info->dirty_metadata_batch);
939                 if (ret < 0)
940                         return 0;
941         }
942         return btree_write_cache_pages(mapping, wbc);
943 }
944
945 static int btree_readpage(struct file *file, struct page *page)
946 {
947         struct extent_io_tree *tree;
948         tree = &BTRFS_I(page->mapping->host)->io_tree;
949         return extent_read_full_page(tree, page, btree_get_extent, 0);
950 }
951
952 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
953 {
954         if (PageWriteback(page) || PageDirty(page))
955                 return 0;
956
957         return try_release_extent_buffer(page);
958 }
959
960 static void btree_invalidatepage(struct page *page, unsigned int offset,
961                                  unsigned int length)
962 {
963         struct extent_io_tree *tree;
964         tree = &BTRFS_I(page->mapping->host)->io_tree;
965         extent_invalidatepage(tree, page, offset);
966         btree_releasepage(page, GFP_NOFS);
967         if (PagePrivate(page)) {
968                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
969                            "page private not zero on page %llu",
970                            (unsigned long long)page_offset(page));
971                 ClearPagePrivate(page);
972                 set_page_private(page, 0);
973                 put_page(page);
974         }
975 }
976
977 static int btree_set_page_dirty(struct page *page)
978 {
979 #ifdef DEBUG
980         struct extent_buffer *eb;
981
982         BUG_ON(!PagePrivate(page));
983         eb = (struct extent_buffer *)page->private;
984         BUG_ON(!eb);
985         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
986         BUG_ON(!atomic_read(&eb->refs));
987         btrfs_assert_tree_locked(eb);
988 #endif
989         return __set_page_dirty_nobuffers(page);
990 }
991
992 static const struct address_space_operations btree_aops = {
993         .readpage       = btree_readpage,
994         .writepages     = btree_writepages,
995         .releasepage    = btree_releasepage,
996         .invalidatepage = btree_invalidatepage,
997 #ifdef CONFIG_MIGRATION
998         .migratepage    = btree_migratepage,
999 #endif
1000         .set_page_dirty = btree_set_page_dirty,
1001 };
1002
1003 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1004 {
1005         struct extent_buffer *buf = NULL;
1006         struct inode *btree_inode = fs_info->btree_inode;
1007
1008         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1009         if (IS_ERR(buf))
1010                 return;
1011         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1012                                  buf, WAIT_NONE, 0);
1013         free_extent_buffer(buf);
1014 }
1015
1016 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1017                          int mirror_num, struct extent_buffer **eb)
1018 {
1019         struct extent_buffer *buf = NULL;
1020         struct inode *btree_inode = fs_info->btree_inode;
1021         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1022         int ret;
1023
1024         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1025         if (IS_ERR(buf))
1026                 return 0;
1027
1028         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1029
1030         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1031                                        mirror_num);
1032         if (ret) {
1033                 free_extent_buffer(buf);
1034                 return ret;
1035         }
1036
1037         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1038                 free_extent_buffer(buf);
1039                 return -EIO;
1040         } else if (extent_buffer_uptodate(buf)) {
1041                 *eb = buf;
1042         } else {
1043                 free_extent_buffer(buf);
1044         }
1045         return 0;
1046 }
1047
1048 struct extent_buffer *btrfs_find_create_tree_block(
1049                                                 struct btrfs_fs_info *fs_info,
1050                                                 u64 bytenr)
1051 {
1052         if (btrfs_is_testing(fs_info))
1053                 return alloc_test_extent_buffer(fs_info, bytenr);
1054         return alloc_extent_buffer(fs_info, bytenr);
1055 }
1056
1057
1058 int btrfs_write_tree_block(struct extent_buffer *buf)
1059 {
1060         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1061                                         buf->start + buf->len - 1);
1062 }
1063
1064 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1065 {
1066         filemap_fdatawait_range(buf->pages[0]->mapping,
1067                                 buf->start, buf->start + buf->len - 1);
1068 }
1069
1070 /*
1071  * Read tree block at logical address @bytenr and do variant basic but critical
1072  * verification.
1073  *
1074  * @parent_transid:     expected transid of this tree block, skip check if 0
1075  * @level:              expected level, mandatory check
1076  * @first_key:          expected key in slot 0, skip check if NULL
1077  */
1078 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1079                                       u64 parent_transid, int level,
1080                                       struct btrfs_key *first_key)
1081 {
1082         struct extent_buffer *buf = NULL;
1083         int ret;
1084
1085         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1086         if (IS_ERR(buf))
1087                 return buf;
1088
1089         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1090                                              level, first_key);
1091         if (ret) {
1092                 free_extent_buffer(buf);
1093                 return ERR_PTR(ret);
1094         }
1095         return buf;
1096
1097 }
1098
1099 void clean_tree_block(struct btrfs_fs_info *fs_info,
1100                       struct extent_buffer *buf)
1101 {
1102         if (btrfs_header_generation(buf) ==
1103             fs_info->running_transaction->transid) {
1104                 btrfs_assert_tree_locked(buf);
1105
1106                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1107                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1108                                                  -buf->len,
1109                                                  fs_info->dirty_metadata_batch);
1110                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1111                         btrfs_set_lock_blocking(buf);
1112                         clear_extent_buffer_dirty(buf);
1113                 }
1114         }
1115 }
1116
1117 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1118 {
1119         struct btrfs_subvolume_writers *writers;
1120         int ret;
1121
1122         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1123         if (!writers)
1124                 return ERR_PTR(-ENOMEM);
1125
1126         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1127         if (ret < 0) {
1128                 kfree(writers);
1129                 return ERR_PTR(ret);
1130         }
1131
1132         init_waitqueue_head(&writers->wait);
1133         return writers;
1134 }
1135
1136 static void
1137 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1138 {
1139         percpu_counter_destroy(&writers->counter);
1140         kfree(writers);
1141 }
1142
1143 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1144                          u64 objectid)
1145 {
1146         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1147         root->node = NULL;
1148         root->commit_root = NULL;
1149         root->state = 0;
1150         root->orphan_cleanup_state = 0;
1151
1152         root->objectid = objectid;
1153         root->last_trans = 0;
1154         root->highest_objectid = 0;
1155         root->nr_delalloc_inodes = 0;
1156         root->nr_ordered_extents = 0;
1157         root->inode_tree = RB_ROOT;
1158         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1159         root->block_rsv = NULL;
1160
1161         INIT_LIST_HEAD(&root->dirty_list);
1162         INIT_LIST_HEAD(&root->root_list);
1163         INIT_LIST_HEAD(&root->delalloc_inodes);
1164         INIT_LIST_HEAD(&root->delalloc_root);
1165         INIT_LIST_HEAD(&root->ordered_extents);
1166         INIT_LIST_HEAD(&root->ordered_root);
1167         INIT_LIST_HEAD(&root->logged_list[0]);
1168         INIT_LIST_HEAD(&root->logged_list[1]);
1169         spin_lock_init(&root->inode_lock);
1170         spin_lock_init(&root->delalloc_lock);
1171         spin_lock_init(&root->ordered_extent_lock);
1172         spin_lock_init(&root->accounting_lock);
1173         spin_lock_init(&root->log_extents_lock[0]);
1174         spin_lock_init(&root->log_extents_lock[1]);
1175         spin_lock_init(&root->qgroup_meta_rsv_lock);
1176         mutex_init(&root->objectid_mutex);
1177         mutex_init(&root->log_mutex);
1178         mutex_init(&root->ordered_extent_mutex);
1179         mutex_init(&root->delalloc_mutex);
1180         init_waitqueue_head(&root->log_writer_wait);
1181         init_waitqueue_head(&root->log_commit_wait[0]);
1182         init_waitqueue_head(&root->log_commit_wait[1]);
1183         INIT_LIST_HEAD(&root->log_ctxs[0]);
1184         INIT_LIST_HEAD(&root->log_ctxs[1]);
1185         atomic_set(&root->log_commit[0], 0);
1186         atomic_set(&root->log_commit[1], 0);
1187         atomic_set(&root->log_writers, 0);
1188         atomic_set(&root->log_batch, 0);
1189         refcount_set(&root->refs, 1);
1190         atomic_set(&root->will_be_snapshotted, 0);
1191         root->log_transid = 0;
1192         root->log_transid_committed = -1;
1193         root->last_log_commit = 0;
1194         if (!dummy)
1195                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1196
1197         memset(&root->root_key, 0, sizeof(root->root_key));
1198         memset(&root->root_item, 0, sizeof(root->root_item));
1199         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1200         if (!dummy)
1201                 root->defrag_trans_start = fs_info->generation;
1202         else
1203                 root->defrag_trans_start = 0;
1204         root->root_key.objectid = objectid;
1205         root->anon_dev = 0;
1206
1207         spin_lock_init(&root->root_item_lock);
1208 }
1209
1210 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1211                 gfp_t flags)
1212 {
1213         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1214         if (root)
1215                 root->fs_info = fs_info;
1216         return root;
1217 }
1218
1219 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1220 /* Should only be used by the testing infrastructure */
1221 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1222 {
1223         struct btrfs_root *root;
1224
1225         if (!fs_info)
1226                 return ERR_PTR(-EINVAL);
1227
1228         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1229         if (!root)
1230                 return ERR_PTR(-ENOMEM);
1231
1232         /* We don't use the stripesize in selftest, set it as sectorsize */
1233         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1234         root->alloc_bytenr = 0;
1235
1236         return root;
1237 }
1238 #endif
1239
1240 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1241                                      struct btrfs_fs_info *fs_info,
1242                                      u64 objectid)
1243 {
1244         struct extent_buffer *leaf;
1245         struct btrfs_root *tree_root = fs_info->tree_root;
1246         struct btrfs_root *root;
1247         struct btrfs_key key;
1248         int ret = 0;
1249         uuid_le uuid = NULL_UUID_LE;
1250
1251         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1252         if (!root)
1253                 return ERR_PTR(-ENOMEM);
1254
1255         __setup_root(root, fs_info, objectid);
1256         root->root_key.objectid = objectid;
1257         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1258         root->root_key.offset = 0;
1259
1260         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1261         if (IS_ERR(leaf)) {
1262                 ret = PTR_ERR(leaf);
1263                 leaf = NULL;
1264                 goto fail;
1265         }
1266
1267         root->node = leaf;
1268         btrfs_mark_buffer_dirty(leaf);
1269
1270         root->commit_root = btrfs_root_node(root);
1271         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1272
1273         root->root_item.flags = 0;
1274         root->root_item.byte_limit = 0;
1275         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1276         btrfs_set_root_generation(&root->root_item, trans->transid);
1277         btrfs_set_root_level(&root->root_item, 0);
1278         btrfs_set_root_refs(&root->root_item, 1);
1279         btrfs_set_root_used(&root->root_item, leaf->len);
1280         btrfs_set_root_last_snapshot(&root->root_item, 0);
1281         btrfs_set_root_dirid(&root->root_item, 0);
1282         if (is_fstree(objectid))
1283                 uuid_le_gen(&uuid);
1284         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1285         root->root_item.drop_level = 0;
1286
1287         key.objectid = objectid;
1288         key.type = BTRFS_ROOT_ITEM_KEY;
1289         key.offset = 0;
1290         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1291         if (ret)
1292                 goto fail;
1293
1294         btrfs_tree_unlock(leaf);
1295
1296         return root;
1297
1298 fail:
1299         if (leaf) {
1300                 btrfs_tree_unlock(leaf);
1301                 free_extent_buffer(root->commit_root);
1302                 free_extent_buffer(leaf);
1303         }
1304         kfree(root);
1305
1306         return ERR_PTR(ret);
1307 }
1308
1309 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1310                                          struct btrfs_fs_info *fs_info)
1311 {
1312         struct btrfs_root *root;
1313         struct extent_buffer *leaf;
1314
1315         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1316         if (!root)
1317                 return ERR_PTR(-ENOMEM);
1318
1319         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1320
1321         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1322         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1323         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1324
1325         /*
1326          * DON'T set REF_COWS for log trees
1327          *
1328          * log trees do not get reference counted because they go away
1329          * before a real commit is actually done.  They do store pointers
1330          * to file data extents, and those reference counts still get
1331          * updated (along with back refs to the log tree).
1332          */
1333
1334         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1335                         NULL, 0, 0, 0);
1336         if (IS_ERR(leaf)) {
1337                 kfree(root);
1338                 return ERR_CAST(leaf);
1339         }
1340
1341         root->node = leaf;
1342
1343         btrfs_mark_buffer_dirty(root->node);
1344         btrfs_tree_unlock(root->node);
1345         return root;
1346 }
1347
1348 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1349                              struct btrfs_fs_info *fs_info)
1350 {
1351         struct btrfs_root *log_root;
1352
1353         log_root = alloc_log_tree(trans, fs_info);
1354         if (IS_ERR(log_root))
1355                 return PTR_ERR(log_root);
1356         WARN_ON(fs_info->log_root_tree);
1357         fs_info->log_root_tree = log_root;
1358         return 0;
1359 }
1360
1361 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1362                        struct btrfs_root *root)
1363 {
1364         struct btrfs_fs_info *fs_info = root->fs_info;
1365         struct btrfs_root *log_root;
1366         struct btrfs_inode_item *inode_item;
1367
1368         log_root = alloc_log_tree(trans, fs_info);
1369         if (IS_ERR(log_root))
1370                 return PTR_ERR(log_root);
1371
1372         log_root->last_trans = trans->transid;
1373         log_root->root_key.offset = root->root_key.objectid;
1374
1375         inode_item = &log_root->root_item.inode;
1376         btrfs_set_stack_inode_generation(inode_item, 1);
1377         btrfs_set_stack_inode_size(inode_item, 3);
1378         btrfs_set_stack_inode_nlink(inode_item, 1);
1379         btrfs_set_stack_inode_nbytes(inode_item,
1380                                      fs_info->nodesize);
1381         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1382
1383         btrfs_set_root_node(&log_root->root_item, log_root->node);
1384
1385         WARN_ON(root->log_root);
1386         root->log_root = log_root;
1387         root->log_transid = 0;
1388         root->log_transid_committed = -1;
1389         root->last_log_commit = 0;
1390         return 0;
1391 }
1392
1393 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1394                                                struct btrfs_key *key)
1395 {
1396         struct btrfs_root *root;
1397         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1398         struct btrfs_path *path;
1399         u64 generation;
1400         int ret;
1401         int level;
1402
1403         path = btrfs_alloc_path();
1404         if (!path)
1405                 return ERR_PTR(-ENOMEM);
1406
1407         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1408         if (!root) {
1409                 ret = -ENOMEM;
1410                 goto alloc_fail;
1411         }
1412
1413         __setup_root(root, fs_info, key->objectid);
1414
1415         ret = btrfs_find_root(tree_root, key, path,
1416                               &root->root_item, &root->root_key);
1417         if (ret) {
1418                 if (ret > 0)
1419                         ret = -ENOENT;
1420                 goto find_fail;
1421         }
1422
1423         generation = btrfs_root_generation(&root->root_item);
1424         level = btrfs_root_level(&root->root_item);
1425         root->node = read_tree_block(fs_info,
1426                                      btrfs_root_bytenr(&root->root_item),
1427                                      generation, level, NULL);
1428         if (IS_ERR(root->node)) {
1429                 ret = PTR_ERR(root->node);
1430                 goto find_fail;
1431         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1432                 ret = -EIO;
1433                 free_extent_buffer(root->node);
1434                 goto find_fail;
1435         }
1436         root->commit_root = btrfs_root_node(root);
1437 out:
1438         btrfs_free_path(path);
1439         return root;
1440
1441 find_fail:
1442         kfree(root);
1443 alloc_fail:
1444         root = ERR_PTR(ret);
1445         goto out;
1446 }
1447
1448 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1449                                       struct btrfs_key *location)
1450 {
1451         struct btrfs_root *root;
1452
1453         root = btrfs_read_tree_root(tree_root, location);
1454         if (IS_ERR(root))
1455                 return root;
1456
1457         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1458                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1459                 btrfs_check_and_init_root_item(&root->root_item);
1460         }
1461
1462         return root;
1463 }
1464
1465 int btrfs_init_fs_root(struct btrfs_root *root)
1466 {
1467         int ret;
1468         struct btrfs_subvolume_writers *writers;
1469
1470         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1471         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1472                                         GFP_NOFS);
1473         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1474                 ret = -ENOMEM;
1475                 goto fail;
1476         }
1477
1478         writers = btrfs_alloc_subvolume_writers();
1479         if (IS_ERR(writers)) {
1480                 ret = PTR_ERR(writers);
1481                 goto fail;
1482         }
1483         root->subv_writers = writers;
1484
1485         btrfs_init_free_ino_ctl(root);
1486         spin_lock_init(&root->ino_cache_lock);
1487         init_waitqueue_head(&root->ino_cache_wait);
1488
1489         ret = get_anon_bdev(&root->anon_dev);
1490         if (ret)
1491                 goto fail;
1492
1493         mutex_lock(&root->objectid_mutex);
1494         ret = btrfs_find_highest_objectid(root,
1495                                         &root->highest_objectid);
1496         if (ret) {
1497                 mutex_unlock(&root->objectid_mutex);
1498                 goto fail;
1499         }
1500
1501         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1502
1503         mutex_unlock(&root->objectid_mutex);
1504
1505         return 0;
1506 fail:
1507         /* the caller is responsible to call free_fs_root */
1508         return ret;
1509 }
1510
1511 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1512                                         u64 root_id)
1513 {
1514         struct btrfs_root *root;
1515
1516         spin_lock(&fs_info->fs_roots_radix_lock);
1517         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1518                                  (unsigned long)root_id);
1519         spin_unlock(&fs_info->fs_roots_radix_lock);
1520         return root;
1521 }
1522
1523 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1524                          struct btrfs_root *root)
1525 {
1526         int ret;
1527
1528         ret = radix_tree_preload(GFP_NOFS);
1529         if (ret)
1530                 return ret;
1531
1532         spin_lock(&fs_info->fs_roots_radix_lock);
1533         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1534                                 (unsigned long)root->root_key.objectid,
1535                                 root);
1536         if (ret == 0)
1537                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1538         spin_unlock(&fs_info->fs_roots_radix_lock);
1539         radix_tree_preload_end();
1540
1541         return ret;
1542 }
1543
1544 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1545                                      struct btrfs_key *location,
1546                                      bool check_ref)
1547 {
1548         struct btrfs_root *root;
1549         struct btrfs_path *path;
1550         struct btrfs_key key;
1551         int ret;
1552
1553         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1554                 return fs_info->tree_root;
1555         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1556                 return fs_info->extent_root;
1557         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1558                 return fs_info->chunk_root;
1559         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1560                 return fs_info->dev_root;
1561         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1562                 return fs_info->csum_root;
1563         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1564                 return fs_info->quota_root ? fs_info->quota_root :
1565                                              ERR_PTR(-ENOENT);
1566         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1567                 return fs_info->uuid_root ? fs_info->uuid_root :
1568                                             ERR_PTR(-ENOENT);
1569         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1570                 return fs_info->free_space_root ? fs_info->free_space_root :
1571                                                   ERR_PTR(-ENOENT);
1572 again:
1573         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1574         if (root) {
1575                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1576                         return ERR_PTR(-ENOENT);
1577                 return root;
1578         }
1579
1580         root = btrfs_read_fs_root(fs_info->tree_root, location);
1581         if (IS_ERR(root))
1582                 return root;
1583
1584         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1585                 ret = -ENOENT;
1586                 goto fail;
1587         }
1588
1589         ret = btrfs_init_fs_root(root);
1590         if (ret)
1591                 goto fail;
1592
1593         path = btrfs_alloc_path();
1594         if (!path) {
1595                 ret = -ENOMEM;
1596                 goto fail;
1597         }
1598         key.objectid = BTRFS_ORPHAN_OBJECTID;
1599         key.type = BTRFS_ORPHAN_ITEM_KEY;
1600         key.offset = location->objectid;
1601
1602         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1603         btrfs_free_path(path);
1604         if (ret < 0)
1605                 goto fail;
1606         if (ret == 0)
1607                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1608
1609         ret = btrfs_insert_fs_root(fs_info, root);
1610         if (ret) {
1611                 if (ret == -EEXIST) {
1612                         free_fs_root(root);
1613                         goto again;
1614                 }
1615                 goto fail;
1616         }
1617         return root;
1618 fail:
1619         free_fs_root(root);
1620         return ERR_PTR(ret);
1621 }
1622
1623 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1624 {
1625         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1626         int ret = 0;
1627         struct btrfs_device *device;
1628         struct backing_dev_info *bdi;
1629
1630         rcu_read_lock();
1631         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1632                 if (!device->bdev)
1633                         continue;
1634                 bdi = device->bdev->bd_bdi;
1635                 if (bdi_congested(bdi, bdi_bits)) {
1636                         ret = 1;
1637                         break;
1638                 }
1639         }
1640         rcu_read_unlock();
1641         return ret;
1642 }
1643
1644 /*
1645  * called by the kthread helper functions to finally call the bio end_io
1646  * functions.  This is where read checksum verification actually happens
1647  */
1648 static void end_workqueue_fn(struct btrfs_work *work)
1649 {
1650         struct bio *bio;
1651         struct btrfs_end_io_wq *end_io_wq;
1652
1653         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1654         bio = end_io_wq->bio;
1655
1656         bio->bi_status = end_io_wq->status;
1657         bio->bi_private = end_io_wq->private;
1658         bio->bi_end_io = end_io_wq->end_io;
1659         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1660         bio_endio(bio);
1661 }
1662
1663 static int cleaner_kthread(void *arg)
1664 {
1665         struct btrfs_root *root = arg;
1666         struct btrfs_fs_info *fs_info = root->fs_info;
1667         int again;
1668         struct btrfs_trans_handle *trans;
1669
1670         do {
1671                 again = 0;
1672
1673                 /* Make the cleaner go to sleep early. */
1674                 if (btrfs_need_cleaner_sleep(fs_info))
1675                         goto sleep;
1676
1677                 /*
1678                  * Do not do anything if we might cause open_ctree() to block
1679                  * before we have finished mounting the filesystem.
1680                  */
1681                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1682                         goto sleep;
1683
1684                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1685                         goto sleep;
1686
1687                 /*
1688                  * Avoid the problem that we change the status of the fs
1689                  * during the above check and trylock.
1690                  */
1691                 if (btrfs_need_cleaner_sleep(fs_info)) {
1692                         mutex_unlock(&fs_info->cleaner_mutex);
1693                         goto sleep;
1694                 }
1695
1696                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1697                 btrfs_run_delayed_iputs(fs_info);
1698                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1699
1700                 again = btrfs_clean_one_deleted_snapshot(root);
1701                 mutex_unlock(&fs_info->cleaner_mutex);
1702
1703                 /*
1704                  * The defragger has dealt with the R/O remount and umount,
1705                  * needn't do anything special here.
1706                  */
1707                 btrfs_run_defrag_inodes(fs_info);
1708
1709                 /*
1710                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1711                  * with relocation (btrfs_relocate_chunk) and relocation
1712                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1713                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1714                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1715                  * unused block groups.
1716                  */
1717                 btrfs_delete_unused_bgs(fs_info);
1718 sleep:
1719                 if (!again) {
1720                         set_current_state(TASK_INTERRUPTIBLE);
1721                         if (!kthread_should_stop())
1722                                 schedule();
1723                         __set_current_state(TASK_RUNNING);
1724                 }
1725         } while (!kthread_should_stop());
1726
1727         /*
1728          * Transaction kthread is stopped before us and wakes us up.
1729          * However we might have started a new transaction and COWed some
1730          * tree blocks when deleting unused block groups for example. So
1731          * make sure we commit the transaction we started to have a clean
1732          * shutdown when evicting the btree inode - if it has dirty pages
1733          * when we do the final iput() on it, eviction will trigger a
1734          * writeback for it which will fail with null pointer dereferences
1735          * since work queues and other resources were already released and
1736          * destroyed by the time the iput/eviction/writeback is made.
1737          */
1738         trans = btrfs_attach_transaction(root);
1739         if (IS_ERR(trans)) {
1740                 if (PTR_ERR(trans) != -ENOENT)
1741                         btrfs_err(fs_info,
1742                                   "cleaner transaction attach returned %ld",
1743                                   PTR_ERR(trans));
1744         } else {
1745                 int ret;
1746
1747                 ret = btrfs_commit_transaction(trans);
1748                 if (ret)
1749                         btrfs_err(fs_info,
1750                                   "cleaner open transaction commit returned %d",
1751                                   ret);
1752         }
1753
1754         return 0;
1755 }
1756
1757 static int transaction_kthread(void *arg)
1758 {
1759         struct btrfs_root *root = arg;
1760         struct btrfs_fs_info *fs_info = root->fs_info;
1761         struct btrfs_trans_handle *trans;
1762         struct btrfs_transaction *cur;
1763         u64 transid;
1764         time64_t now;
1765         unsigned long delay;
1766         bool cannot_commit;
1767
1768         do {
1769                 cannot_commit = false;
1770                 delay = HZ * fs_info->commit_interval;
1771                 mutex_lock(&fs_info->transaction_kthread_mutex);
1772
1773                 spin_lock(&fs_info->trans_lock);
1774                 cur = fs_info->running_transaction;
1775                 if (!cur) {
1776                         spin_unlock(&fs_info->trans_lock);
1777                         goto sleep;
1778                 }
1779
1780                 now = ktime_get_seconds();
1781                 if (cur->state < TRANS_STATE_BLOCKED &&
1782                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1783                     (now < cur->start_time ||
1784                      now - cur->start_time < fs_info->commit_interval)) {
1785                         spin_unlock(&fs_info->trans_lock);
1786                         delay = HZ * 5;
1787                         goto sleep;
1788                 }
1789                 transid = cur->transid;
1790                 spin_unlock(&fs_info->trans_lock);
1791
1792                 /* If the file system is aborted, this will always fail. */
1793                 trans = btrfs_attach_transaction(root);
1794                 if (IS_ERR(trans)) {
1795                         if (PTR_ERR(trans) != -ENOENT)
1796                                 cannot_commit = true;
1797                         goto sleep;
1798                 }
1799                 if (transid == trans->transid) {
1800                         btrfs_commit_transaction(trans);
1801                 } else {
1802                         btrfs_end_transaction(trans);
1803                 }
1804 sleep:
1805                 wake_up_process(fs_info->cleaner_kthread);
1806                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1807
1808                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1809                                       &fs_info->fs_state)))
1810                         btrfs_cleanup_transaction(fs_info);
1811                 if (!kthread_should_stop() &&
1812                                 (!btrfs_transaction_blocked(fs_info) ||
1813                                  cannot_commit))
1814                         schedule_timeout_interruptible(delay);
1815         } while (!kthread_should_stop());
1816         return 0;
1817 }
1818
1819 /*
1820  * this will find the highest generation in the array of
1821  * root backups.  The index of the highest array is returned,
1822  * or -1 if we can't find anything.
1823  *
1824  * We check to make sure the array is valid by comparing the
1825  * generation of the latest  root in the array with the generation
1826  * in the super block.  If they don't match we pitch it.
1827  */
1828 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1829 {
1830         u64 cur;
1831         int newest_index = -1;
1832         struct btrfs_root_backup *root_backup;
1833         int i;
1834
1835         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1836                 root_backup = info->super_copy->super_roots + i;
1837                 cur = btrfs_backup_tree_root_gen(root_backup);
1838                 if (cur == newest_gen)
1839                         newest_index = i;
1840         }
1841
1842         /* check to see if we actually wrapped around */
1843         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1844                 root_backup = info->super_copy->super_roots;
1845                 cur = btrfs_backup_tree_root_gen(root_backup);
1846                 if (cur == newest_gen)
1847                         newest_index = 0;
1848         }
1849         return newest_index;
1850 }
1851
1852
1853 /*
1854  * find the oldest backup so we know where to store new entries
1855  * in the backup array.  This will set the backup_root_index
1856  * field in the fs_info struct
1857  */
1858 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1859                                      u64 newest_gen)
1860 {
1861         int newest_index = -1;
1862
1863         newest_index = find_newest_super_backup(info, newest_gen);
1864         /* if there was garbage in there, just move along */
1865         if (newest_index == -1) {
1866                 info->backup_root_index = 0;
1867         } else {
1868                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1869         }
1870 }
1871
1872 /*
1873  * copy all the root pointers into the super backup array.
1874  * this will bump the backup pointer by one when it is
1875  * done
1876  */
1877 static void backup_super_roots(struct btrfs_fs_info *info)
1878 {
1879         int next_backup;
1880         struct btrfs_root_backup *root_backup;
1881         int last_backup;
1882
1883         next_backup = info->backup_root_index;
1884         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1885                 BTRFS_NUM_BACKUP_ROOTS;
1886
1887         /*
1888          * just overwrite the last backup if we're at the same generation
1889          * this happens only at umount
1890          */
1891         root_backup = info->super_for_commit->super_roots + last_backup;
1892         if (btrfs_backup_tree_root_gen(root_backup) ==
1893             btrfs_header_generation(info->tree_root->node))
1894                 next_backup = last_backup;
1895
1896         root_backup = info->super_for_commit->super_roots + next_backup;
1897
1898         /*
1899          * make sure all of our padding and empty slots get zero filled
1900          * regardless of which ones we use today
1901          */
1902         memset(root_backup, 0, sizeof(*root_backup));
1903
1904         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1905
1906         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1907         btrfs_set_backup_tree_root_gen(root_backup,
1908                                btrfs_header_generation(info->tree_root->node));
1909
1910         btrfs_set_backup_tree_root_level(root_backup,
1911                                btrfs_header_level(info->tree_root->node));
1912
1913         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1914         btrfs_set_backup_chunk_root_gen(root_backup,
1915                                btrfs_header_generation(info->chunk_root->node));
1916         btrfs_set_backup_chunk_root_level(root_backup,
1917                                btrfs_header_level(info->chunk_root->node));
1918
1919         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1920         btrfs_set_backup_extent_root_gen(root_backup,
1921                                btrfs_header_generation(info->extent_root->node));
1922         btrfs_set_backup_extent_root_level(root_backup,
1923                                btrfs_header_level(info->extent_root->node));
1924
1925         /*
1926          * we might commit during log recovery, which happens before we set
1927          * the fs_root.  Make sure it is valid before we fill it in.
1928          */
1929         if (info->fs_root && info->fs_root->node) {
1930                 btrfs_set_backup_fs_root(root_backup,
1931                                          info->fs_root->node->start);
1932                 btrfs_set_backup_fs_root_gen(root_backup,
1933                                btrfs_header_generation(info->fs_root->node));
1934                 btrfs_set_backup_fs_root_level(root_backup,
1935                                btrfs_header_level(info->fs_root->node));
1936         }
1937
1938         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1939         btrfs_set_backup_dev_root_gen(root_backup,
1940                                btrfs_header_generation(info->dev_root->node));
1941         btrfs_set_backup_dev_root_level(root_backup,
1942                                        btrfs_header_level(info->dev_root->node));
1943
1944         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1945         btrfs_set_backup_csum_root_gen(root_backup,
1946                                btrfs_header_generation(info->csum_root->node));
1947         btrfs_set_backup_csum_root_level(root_backup,
1948                                btrfs_header_level(info->csum_root->node));
1949
1950         btrfs_set_backup_total_bytes(root_backup,
1951                              btrfs_super_total_bytes(info->super_copy));
1952         btrfs_set_backup_bytes_used(root_backup,
1953                              btrfs_super_bytes_used(info->super_copy));
1954         btrfs_set_backup_num_devices(root_backup,
1955                              btrfs_super_num_devices(info->super_copy));
1956
1957         /*
1958          * if we don't copy this out to the super_copy, it won't get remembered
1959          * for the next commit
1960          */
1961         memcpy(&info->super_copy->super_roots,
1962                &info->super_for_commit->super_roots,
1963                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1964 }
1965
1966 /*
1967  * this copies info out of the root backup array and back into
1968  * the in-memory super block.  It is meant to help iterate through
1969  * the array, so you send it the number of backups you've already
1970  * tried and the last backup index you used.
1971  *
1972  * this returns -1 when it has tried all the backups
1973  */
1974 static noinline int next_root_backup(struct btrfs_fs_info *info,
1975                                      struct btrfs_super_block *super,
1976                                      int *num_backups_tried, int *backup_index)
1977 {
1978         struct btrfs_root_backup *root_backup;
1979         int newest = *backup_index;
1980
1981         if (*num_backups_tried == 0) {
1982                 u64 gen = btrfs_super_generation(super);
1983
1984                 newest = find_newest_super_backup(info, gen);
1985                 if (newest == -1)
1986                         return -1;
1987
1988                 *backup_index = newest;
1989                 *num_backups_tried = 1;
1990         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1991                 /* we've tried all the backups, all done */
1992                 return -1;
1993         } else {
1994                 /* jump to the next oldest backup */
1995                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1996                         BTRFS_NUM_BACKUP_ROOTS;
1997                 *backup_index = newest;
1998                 *num_backups_tried += 1;
1999         }
2000         root_backup = super->super_roots + newest;
2001
2002         btrfs_set_super_generation(super,
2003                                    btrfs_backup_tree_root_gen(root_backup));
2004         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2005         btrfs_set_super_root_level(super,
2006                                    btrfs_backup_tree_root_level(root_backup));
2007         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2008
2009         /*
2010          * fixme: the total bytes and num_devices need to match or we should
2011          * need a fsck
2012          */
2013         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2014         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2015         return 0;
2016 }
2017
2018 /* helper to cleanup workers */
2019 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2020 {
2021         btrfs_destroy_workqueue(fs_info->fixup_workers);
2022         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2023         btrfs_destroy_workqueue(fs_info->workers);
2024         btrfs_destroy_workqueue(fs_info->endio_workers);
2025         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2026         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2027         btrfs_destroy_workqueue(fs_info->rmw_workers);
2028         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2029         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2030         btrfs_destroy_workqueue(fs_info->submit_workers);
2031         btrfs_destroy_workqueue(fs_info->delayed_workers);
2032         btrfs_destroy_workqueue(fs_info->caching_workers);
2033         btrfs_destroy_workqueue(fs_info->readahead_workers);
2034         btrfs_destroy_workqueue(fs_info->flush_workers);
2035         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2036         btrfs_destroy_workqueue(fs_info->extent_workers);
2037         /*
2038          * Now that all other work queues are destroyed, we can safely destroy
2039          * the queues used for metadata I/O, since tasks from those other work
2040          * queues can do metadata I/O operations.
2041          */
2042         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2043         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2044 }
2045
2046 static void free_root_extent_buffers(struct btrfs_root *root)
2047 {
2048         if (root) {
2049                 free_extent_buffer(root->node);
2050                 free_extent_buffer(root->commit_root);
2051                 root->node = NULL;
2052                 root->commit_root = NULL;
2053         }
2054 }
2055
2056 /* helper to cleanup tree roots */
2057 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2058 {
2059         free_root_extent_buffers(info->tree_root);
2060
2061         free_root_extent_buffers(info->dev_root);
2062         free_root_extent_buffers(info->extent_root);
2063         free_root_extent_buffers(info->csum_root);
2064         free_root_extent_buffers(info->quota_root);
2065         free_root_extent_buffers(info->uuid_root);
2066         if (chunk_root)
2067                 free_root_extent_buffers(info->chunk_root);
2068         free_root_extent_buffers(info->free_space_root);
2069 }
2070
2071 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2072 {
2073         int ret;
2074         struct btrfs_root *gang[8];
2075         int i;
2076
2077         while (!list_empty(&fs_info->dead_roots)) {
2078                 gang[0] = list_entry(fs_info->dead_roots.next,
2079                                      struct btrfs_root, root_list);
2080                 list_del(&gang[0]->root_list);
2081
2082                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2083                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2084                 } else {
2085                         free_extent_buffer(gang[0]->node);
2086                         free_extent_buffer(gang[0]->commit_root);
2087                         btrfs_put_fs_root(gang[0]);
2088                 }
2089         }
2090
2091         while (1) {
2092                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2093                                              (void **)gang, 0,
2094                                              ARRAY_SIZE(gang));
2095                 if (!ret)
2096                         break;
2097                 for (i = 0; i < ret; i++)
2098                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2099         }
2100
2101         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2102                 btrfs_free_log_root_tree(NULL, fs_info);
2103                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2104         }
2105 }
2106
2107 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2108 {
2109         mutex_init(&fs_info->scrub_lock);
2110         atomic_set(&fs_info->scrubs_running, 0);
2111         atomic_set(&fs_info->scrub_pause_req, 0);
2112         atomic_set(&fs_info->scrubs_paused, 0);
2113         atomic_set(&fs_info->scrub_cancel_req, 0);
2114         init_waitqueue_head(&fs_info->scrub_pause_wait);
2115         fs_info->scrub_workers_refcnt = 0;
2116 }
2117
2118 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2119 {
2120         spin_lock_init(&fs_info->balance_lock);
2121         mutex_init(&fs_info->balance_mutex);
2122         atomic_set(&fs_info->balance_pause_req, 0);
2123         atomic_set(&fs_info->balance_cancel_req, 0);
2124         fs_info->balance_ctl = NULL;
2125         init_waitqueue_head(&fs_info->balance_wait_q);
2126 }
2127
2128 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2129 {
2130         struct inode *inode = fs_info->btree_inode;
2131
2132         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2133         set_nlink(inode, 1);
2134         /*
2135          * we set the i_size on the btree inode to the max possible int.
2136          * the real end of the address space is determined by all of
2137          * the devices in the system
2138          */
2139         inode->i_size = OFFSET_MAX;
2140         inode->i_mapping->a_ops = &btree_aops;
2141
2142         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2143         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2144         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2145         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2146
2147         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2148
2149         BTRFS_I(inode)->root = fs_info->tree_root;
2150         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2151         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2152         btrfs_insert_inode_hash(inode);
2153 }
2154
2155 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2156 {
2157         fs_info->dev_replace.lock_owner = 0;
2158         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2159         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2160         rwlock_init(&fs_info->dev_replace.lock);
2161         atomic_set(&fs_info->dev_replace.read_locks, 0);
2162         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2163         init_waitqueue_head(&fs_info->replace_wait);
2164         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2165 }
2166
2167 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2168 {
2169         spin_lock_init(&fs_info->qgroup_lock);
2170         mutex_init(&fs_info->qgroup_ioctl_lock);
2171         fs_info->qgroup_tree = RB_ROOT;
2172         fs_info->qgroup_op_tree = RB_ROOT;
2173         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2174         fs_info->qgroup_seq = 1;
2175         fs_info->qgroup_ulist = NULL;
2176         fs_info->qgroup_rescan_running = false;
2177         mutex_init(&fs_info->qgroup_rescan_lock);
2178 }
2179
2180 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2181                 struct btrfs_fs_devices *fs_devices)
2182 {
2183         u32 max_active = fs_info->thread_pool_size;
2184         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2185
2186         fs_info->workers =
2187                 btrfs_alloc_workqueue(fs_info, "worker",
2188                                       flags | WQ_HIGHPRI, max_active, 16);
2189
2190         fs_info->delalloc_workers =
2191                 btrfs_alloc_workqueue(fs_info, "delalloc",
2192                                       flags, max_active, 2);
2193
2194         fs_info->flush_workers =
2195                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2196                                       flags, max_active, 0);
2197
2198         fs_info->caching_workers =
2199                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2200
2201         /*
2202          * a higher idle thresh on the submit workers makes it much more
2203          * likely that bios will be send down in a sane order to the
2204          * devices
2205          */
2206         fs_info->submit_workers =
2207                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2208                                       min_t(u64, fs_devices->num_devices,
2209                                             max_active), 64);
2210
2211         fs_info->fixup_workers =
2212                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2213
2214         /*
2215          * endios are largely parallel and should have a very
2216          * low idle thresh
2217          */
2218         fs_info->endio_workers =
2219                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2220         fs_info->endio_meta_workers =
2221                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2222                                       max_active, 4);
2223         fs_info->endio_meta_write_workers =
2224                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2225                                       max_active, 2);
2226         fs_info->endio_raid56_workers =
2227                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2228                                       max_active, 4);
2229         fs_info->endio_repair_workers =
2230                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2231         fs_info->rmw_workers =
2232                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2233         fs_info->endio_write_workers =
2234                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2235                                       max_active, 2);
2236         fs_info->endio_freespace_worker =
2237                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2238                                       max_active, 0);
2239         fs_info->delayed_workers =
2240                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2241                                       max_active, 0);
2242         fs_info->readahead_workers =
2243                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2244                                       max_active, 2);
2245         fs_info->qgroup_rescan_workers =
2246                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2247         fs_info->extent_workers =
2248                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2249                                       min_t(u64, fs_devices->num_devices,
2250                                             max_active), 8);
2251
2252         if (!(fs_info->workers && fs_info->delalloc_workers &&
2253               fs_info->submit_workers && fs_info->flush_workers &&
2254               fs_info->endio_workers && fs_info->endio_meta_workers &&
2255               fs_info->endio_meta_write_workers &&
2256               fs_info->endio_repair_workers &&
2257               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2258               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2259               fs_info->caching_workers && fs_info->readahead_workers &&
2260               fs_info->fixup_workers && fs_info->delayed_workers &&
2261               fs_info->extent_workers &&
2262               fs_info->qgroup_rescan_workers)) {
2263                 return -ENOMEM;
2264         }
2265
2266         return 0;
2267 }
2268
2269 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2270                             struct btrfs_fs_devices *fs_devices)
2271 {
2272         int ret;
2273         struct btrfs_root *log_tree_root;
2274         struct btrfs_super_block *disk_super = fs_info->super_copy;
2275         u64 bytenr = btrfs_super_log_root(disk_super);
2276         int level = btrfs_super_log_root_level(disk_super);
2277
2278         if (fs_devices->rw_devices == 0) {
2279                 btrfs_warn(fs_info, "log replay required on RO media");
2280                 return -EIO;
2281         }
2282
2283         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2284         if (!log_tree_root)
2285                 return -ENOMEM;
2286
2287         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2288
2289         log_tree_root->node = read_tree_block(fs_info, bytenr,
2290                                               fs_info->generation + 1,
2291                                               level, NULL);
2292         if (IS_ERR(log_tree_root->node)) {
2293                 btrfs_warn(fs_info, "failed to read log tree");
2294                 ret = PTR_ERR(log_tree_root->node);
2295                 kfree(log_tree_root);
2296                 return ret;
2297         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2298                 btrfs_err(fs_info, "failed to read log tree");
2299                 free_extent_buffer(log_tree_root->node);
2300                 kfree(log_tree_root);
2301                 return -EIO;
2302         }
2303         /* returns with log_tree_root freed on success */
2304         ret = btrfs_recover_log_trees(log_tree_root);
2305         if (ret) {
2306                 btrfs_handle_fs_error(fs_info, ret,
2307                                       "Failed to recover log tree");
2308                 free_extent_buffer(log_tree_root->node);
2309                 kfree(log_tree_root);
2310                 return ret;
2311         }
2312
2313         if (sb_rdonly(fs_info->sb)) {
2314                 ret = btrfs_commit_super(fs_info);
2315                 if (ret)
2316                         return ret;
2317         }
2318
2319         return 0;
2320 }
2321
2322 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2323 {
2324         struct btrfs_root *tree_root = fs_info->tree_root;
2325         struct btrfs_root *root;
2326         struct btrfs_key location;
2327         int ret;
2328
2329         BUG_ON(!fs_info->tree_root);
2330
2331         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2332         location.type = BTRFS_ROOT_ITEM_KEY;
2333         location.offset = 0;
2334
2335         root = btrfs_read_tree_root(tree_root, &location);
2336         if (IS_ERR(root)) {
2337                 ret = PTR_ERR(root);
2338                 goto out;
2339         }
2340         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2341         fs_info->extent_root = root;
2342
2343         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2344         root = btrfs_read_tree_root(tree_root, &location);
2345         if (IS_ERR(root)) {
2346                 ret = PTR_ERR(root);
2347                 goto out;
2348         }
2349         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2350         fs_info->dev_root = root;
2351         btrfs_init_devices_late(fs_info);
2352
2353         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2354         root = btrfs_read_tree_root(tree_root, &location);
2355         if (IS_ERR(root)) {
2356                 ret = PTR_ERR(root);
2357                 goto out;
2358         }
2359         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2360         fs_info->csum_root = root;
2361
2362         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2363         root = btrfs_read_tree_root(tree_root, &location);
2364         if (!IS_ERR(root)) {
2365                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2366                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2367                 fs_info->quota_root = root;
2368         }
2369
2370         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2371         root = btrfs_read_tree_root(tree_root, &location);
2372         if (IS_ERR(root)) {
2373                 ret = PTR_ERR(root);
2374                 if (ret != -ENOENT)
2375                         goto out;
2376         } else {
2377                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2378                 fs_info->uuid_root = root;
2379         }
2380
2381         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2382                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2383                 root = btrfs_read_tree_root(tree_root, &location);
2384                 if (IS_ERR(root)) {
2385                         ret = PTR_ERR(root);
2386                         goto out;
2387                 }
2388                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2389                 fs_info->free_space_root = root;
2390         }
2391
2392         return 0;
2393 out:
2394         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2395                    location.objectid, ret);
2396         return ret;
2397 }
2398
2399 /*
2400  * Real super block validation
2401  * NOTE: super csum type and incompat features will not be checked here.
2402  *
2403  * @sb:         super block to check
2404  * @mirror_num: the super block number to check its bytenr:
2405  *              0       the primary (1st) sb
2406  *              1, 2    2nd and 3rd backup copy
2407  *             -1       skip bytenr check
2408  */
2409 static int validate_super(struct btrfs_fs_info *fs_info,
2410                             struct btrfs_super_block *sb, int mirror_num)
2411 {
2412         u64 nodesize = btrfs_super_nodesize(sb);
2413         u64 sectorsize = btrfs_super_sectorsize(sb);
2414         int ret = 0;
2415
2416         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2417                 btrfs_err(fs_info, "no valid FS found");
2418                 ret = -EINVAL;
2419         }
2420         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2421                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2422                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2423                 ret = -EINVAL;
2424         }
2425         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2426                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2427                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2428                 ret = -EINVAL;
2429         }
2430         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2431                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2432                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2433                 ret = -EINVAL;
2434         }
2435         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2436                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2437                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2438                 ret = -EINVAL;
2439         }
2440
2441         /*
2442          * Check sectorsize and nodesize first, other check will need it.
2443          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2444          */
2445         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2446             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2447                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2448                 ret = -EINVAL;
2449         }
2450         /* Only PAGE SIZE is supported yet */
2451         if (sectorsize != PAGE_SIZE) {
2452                 btrfs_err(fs_info,
2453                         "sectorsize %llu not supported yet, only support %lu",
2454                         sectorsize, PAGE_SIZE);
2455                 ret = -EINVAL;
2456         }
2457         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2458             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2459                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2460                 ret = -EINVAL;
2461         }
2462         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2463                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2464                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2465                 ret = -EINVAL;
2466         }
2467
2468         /* Root alignment check */
2469         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2470                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2471                            btrfs_super_root(sb));
2472                 ret = -EINVAL;
2473         }
2474         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2475                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2476                            btrfs_super_chunk_root(sb));
2477                 ret = -EINVAL;
2478         }
2479         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2480                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2481                            btrfs_super_log_root(sb));
2482                 ret = -EINVAL;
2483         }
2484
2485         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
2486                 btrfs_err(fs_info,
2487                            "dev_item UUID does not match fsid: %pU != %pU",
2488                            fs_info->fsid, sb->dev_item.fsid);
2489                 ret = -EINVAL;
2490         }
2491
2492         /*
2493          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2494          * done later
2495          */
2496         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2497                 btrfs_err(fs_info, "bytes_used is too small %llu",
2498                           btrfs_super_bytes_used(sb));
2499                 ret = -EINVAL;
2500         }
2501         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2502                 btrfs_err(fs_info, "invalid stripesize %u",
2503                           btrfs_super_stripesize(sb));
2504                 ret = -EINVAL;
2505         }
2506         if (btrfs_super_num_devices(sb) > (1UL << 31))
2507                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2508                            btrfs_super_num_devices(sb));
2509         if (btrfs_super_num_devices(sb) == 0) {
2510                 btrfs_err(fs_info, "number of devices is 0");
2511                 ret = -EINVAL;
2512         }
2513
2514         if (mirror_num >= 0 &&
2515             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2516                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2517                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2518                 ret = -EINVAL;
2519         }
2520
2521         /*
2522          * Obvious sys_chunk_array corruptions, it must hold at least one key
2523          * and one chunk
2524          */
2525         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2526                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2527                           btrfs_super_sys_array_size(sb),
2528                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2529                 ret = -EINVAL;
2530         }
2531         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2532                         + sizeof(struct btrfs_chunk)) {
2533                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2534                           btrfs_super_sys_array_size(sb),
2535                           sizeof(struct btrfs_disk_key)
2536                           + sizeof(struct btrfs_chunk));
2537                 ret = -EINVAL;
2538         }
2539
2540         /*
2541          * The generation is a global counter, we'll trust it more than the others
2542          * but it's still possible that it's the one that's wrong.
2543          */
2544         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2545                 btrfs_warn(fs_info,
2546                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2547                         btrfs_super_generation(sb),
2548                         btrfs_super_chunk_root_generation(sb));
2549         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2550             && btrfs_super_cache_generation(sb) != (u64)-1)
2551                 btrfs_warn(fs_info,
2552                         "suspicious: generation < cache_generation: %llu < %llu",
2553                         btrfs_super_generation(sb),
2554                         btrfs_super_cache_generation(sb));
2555
2556         return ret;
2557 }
2558
2559 /*
2560  * Validation of super block at mount time.
2561  * Some checks already done early at mount time, like csum type and incompat
2562  * flags will be skipped.
2563  */
2564 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2565 {
2566         return validate_super(fs_info, fs_info->super_copy, 0);
2567 }
2568
2569 /*
2570  * Validation of super block at write time.
2571  * Some checks like bytenr check will be skipped as their values will be
2572  * overwritten soon.
2573  * Extra checks like csum type and incompat flags will be done here.
2574  */
2575 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2576                                       struct btrfs_super_block *sb)
2577 {
2578         int ret;
2579
2580         ret = validate_super(fs_info, sb, -1);
2581         if (ret < 0)
2582                 goto out;
2583         if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2584                 ret = -EUCLEAN;
2585                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2586                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2587                 goto out;
2588         }
2589         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2590                 ret = -EUCLEAN;
2591                 btrfs_err(fs_info,
2592                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2593                           btrfs_super_incompat_flags(sb),
2594                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2595                 goto out;
2596         }
2597 out:
2598         if (ret < 0)
2599                 btrfs_err(fs_info,
2600                 "super block corruption detected before writing it to disk");
2601         return ret;
2602 }
2603
2604 int open_ctree(struct super_block *sb,
2605                struct btrfs_fs_devices *fs_devices,
2606                char *options)
2607 {
2608         u32 sectorsize;
2609         u32 nodesize;
2610         u32 stripesize;
2611         u64 generation;
2612         u64 features;
2613         struct btrfs_key location;
2614         struct buffer_head *bh;
2615         struct btrfs_super_block *disk_super;
2616         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2617         struct btrfs_root *tree_root;
2618         struct btrfs_root *chunk_root;
2619         int ret;
2620         int err = -EINVAL;
2621         int num_backups_tried = 0;
2622         int backup_index = 0;
2623         int clear_free_space_tree = 0;
2624         int level;
2625
2626         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2627         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2628         if (!tree_root || !chunk_root) {
2629                 err = -ENOMEM;
2630                 goto fail;
2631         }
2632
2633         ret = init_srcu_struct(&fs_info->subvol_srcu);
2634         if (ret) {
2635                 err = ret;
2636                 goto fail;
2637         }
2638
2639         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2640         if (ret) {
2641                 err = ret;
2642                 goto fail_srcu;
2643         }
2644         fs_info->dirty_metadata_batch = PAGE_SIZE *
2645                                         (1 + ilog2(nr_cpu_ids));
2646
2647         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2648         if (ret) {
2649                 err = ret;
2650                 goto fail_dirty_metadata_bytes;
2651         }
2652
2653         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2654         if (ret) {
2655                 err = ret;
2656                 goto fail_delalloc_bytes;
2657         }
2658
2659         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2660         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2661         INIT_LIST_HEAD(&fs_info->trans_list);
2662         INIT_LIST_HEAD(&fs_info->dead_roots);
2663         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2664         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2665         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2666         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2667         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2668         spin_lock_init(&fs_info->delalloc_root_lock);
2669         spin_lock_init(&fs_info->trans_lock);
2670         spin_lock_init(&fs_info->fs_roots_radix_lock);
2671         spin_lock_init(&fs_info->delayed_iput_lock);
2672         spin_lock_init(&fs_info->defrag_inodes_lock);
2673         spin_lock_init(&fs_info->tree_mod_seq_lock);
2674         spin_lock_init(&fs_info->super_lock);
2675         spin_lock_init(&fs_info->qgroup_op_lock);
2676         spin_lock_init(&fs_info->buffer_lock);
2677         spin_lock_init(&fs_info->unused_bgs_lock);
2678         rwlock_init(&fs_info->tree_mod_log_lock);
2679         mutex_init(&fs_info->unused_bg_unpin_mutex);
2680         mutex_init(&fs_info->delete_unused_bgs_mutex);
2681         mutex_init(&fs_info->reloc_mutex);
2682         mutex_init(&fs_info->delalloc_root_mutex);
2683         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2684         seqlock_init(&fs_info->profiles_lock);
2685
2686         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2687         INIT_LIST_HEAD(&fs_info->space_info);
2688         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2689         INIT_LIST_HEAD(&fs_info->unused_bgs);
2690         btrfs_mapping_init(&fs_info->mapping_tree);
2691         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2692                              BTRFS_BLOCK_RSV_GLOBAL);
2693         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2694         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2695         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2696         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2697                              BTRFS_BLOCK_RSV_DELOPS);
2698         atomic_set(&fs_info->async_delalloc_pages, 0);
2699         atomic_set(&fs_info->defrag_running, 0);
2700         atomic_set(&fs_info->qgroup_op_seq, 0);
2701         atomic_set(&fs_info->reada_works_cnt, 0);
2702         atomic64_set(&fs_info->tree_mod_seq, 0);
2703         fs_info->sb = sb;
2704         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2705         fs_info->metadata_ratio = 0;
2706         fs_info->defrag_inodes = RB_ROOT;
2707         atomic64_set(&fs_info->free_chunk_space, 0);
2708         fs_info->tree_mod_log = RB_ROOT;
2709         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2710         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2711         /* readahead state */
2712         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2713         spin_lock_init(&fs_info->reada_lock);
2714         btrfs_init_ref_verify(fs_info);
2715
2716         fs_info->thread_pool_size = min_t(unsigned long,
2717                                           num_online_cpus() + 2, 8);
2718
2719         INIT_LIST_HEAD(&fs_info->ordered_roots);
2720         spin_lock_init(&fs_info->ordered_root_lock);
2721
2722         fs_info->btree_inode = new_inode(sb);
2723         if (!fs_info->btree_inode) {
2724                 err = -ENOMEM;
2725                 goto fail_bio_counter;
2726         }
2727         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2728
2729         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2730                                         GFP_KERNEL);
2731         if (!fs_info->delayed_root) {
2732                 err = -ENOMEM;
2733                 goto fail_iput;
2734         }
2735         btrfs_init_delayed_root(fs_info->delayed_root);
2736
2737         btrfs_init_scrub(fs_info);
2738 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2739         fs_info->check_integrity_print_mask = 0;
2740 #endif
2741         btrfs_init_balance(fs_info);
2742         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2743
2744         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2745         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2746
2747         btrfs_init_btree_inode(fs_info);
2748
2749         spin_lock_init(&fs_info->block_group_cache_lock);
2750         fs_info->block_group_cache_tree = RB_ROOT;
2751         fs_info->first_logical_byte = (u64)-1;
2752
2753         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2754         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2755         fs_info->pinned_extents = &fs_info->freed_extents[0];
2756         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2757
2758         mutex_init(&fs_info->ordered_operations_mutex);
2759         mutex_init(&fs_info->tree_log_mutex);
2760         mutex_init(&fs_info->chunk_mutex);
2761         mutex_init(&fs_info->transaction_kthread_mutex);
2762         mutex_init(&fs_info->cleaner_mutex);
2763         mutex_init(&fs_info->ro_block_group_mutex);
2764         init_rwsem(&fs_info->commit_root_sem);
2765         init_rwsem(&fs_info->cleanup_work_sem);
2766         init_rwsem(&fs_info->subvol_sem);
2767         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2768
2769         btrfs_init_dev_replace_locks(fs_info);
2770         btrfs_init_qgroup(fs_info);
2771
2772         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2773         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2774
2775         init_waitqueue_head(&fs_info->transaction_throttle);
2776         init_waitqueue_head(&fs_info->transaction_wait);
2777         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2778         init_waitqueue_head(&fs_info->async_submit_wait);
2779
2780         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2781
2782         /* Usable values until the real ones are cached from the superblock */
2783         fs_info->nodesize = 4096;
2784         fs_info->sectorsize = 4096;
2785         fs_info->stripesize = 4096;
2786
2787         ret = btrfs_alloc_stripe_hash_table(fs_info);
2788         if (ret) {
2789                 err = ret;
2790                 goto fail_alloc;
2791         }
2792
2793         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2794
2795         invalidate_bdev(fs_devices->latest_bdev);
2796
2797         /*
2798          * Read super block and check the signature bytes only
2799          */
2800         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2801         if (IS_ERR(bh)) {
2802                 err = PTR_ERR(bh);
2803                 goto fail_alloc;
2804         }
2805
2806         /*
2807          * We want to check superblock checksum, the type is stored inside.
2808          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2809          */
2810         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2811                 btrfs_err(fs_info, "superblock checksum mismatch");
2812                 err = -EINVAL;
2813                 brelse(bh);
2814                 goto fail_alloc;
2815         }
2816
2817         /*
2818          * super_copy is zeroed at allocation time and we never touch the
2819          * following bytes up to INFO_SIZE, the checksum is calculated from
2820          * the whole block of INFO_SIZE
2821          */
2822         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2823         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2824                sizeof(*fs_info->super_for_commit));
2825         brelse(bh);
2826
2827         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2828
2829         ret = btrfs_validate_mount_super(fs_info);
2830         if (ret) {
2831                 btrfs_err(fs_info, "superblock contains fatal errors");
2832                 err = -EINVAL;
2833                 goto fail_alloc;
2834         }
2835
2836         disk_super = fs_info->super_copy;
2837         if (!btrfs_super_root(disk_super))
2838                 goto fail_alloc;
2839
2840         /* check FS state, whether FS is broken. */
2841         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2842                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2843
2844         /*
2845          * run through our array of backup supers and setup
2846          * our ring pointer to the oldest one
2847          */
2848         generation = btrfs_super_generation(disk_super);
2849         find_oldest_super_backup(fs_info, generation);
2850
2851         /*
2852          * In the long term, we'll store the compression type in the super
2853          * block, and it'll be used for per file compression control.
2854          */
2855         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2856
2857         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2858         if (ret) {
2859                 err = ret;
2860                 goto fail_alloc;
2861         }
2862
2863         features = btrfs_super_incompat_flags(disk_super) &
2864                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2865         if (features) {
2866                 btrfs_err(fs_info,
2867                     "cannot mount because of unsupported optional features (%llx)",
2868                     features);
2869                 err = -EINVAL;
2870                 goto fail_alloc;
2871         }
2872
2873         features = btrfs_super_incompat_flags(disk_super);
2874         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2875         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2876                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2877         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2878                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2879
2880         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2881                 btrfs_info(fs_info, "has skinny extents");
2882
2883         /*
2884          * flag our filesystem as having big metadata blocks if
2885          * they are bigger than the page size
2886          */
2887         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2888                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2889                         btrfs_info(fs_info,
2890                                 "flagging fs with big metadata feature");
2891                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2892         }
2893
2894         nodesize = btrfs_super_nodesize(disk_super);
2895         sectorsize = btrfs_super_sectorsize(disk_super);
2896         stripesize = sectorsize;
2897         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2898         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2899
2900         /* Cache block sizes */
2901         fs_info->nodesize = nodesize;
2902         fs_info->sectorsize = sectorsize;
2903         fs_info->stripesize = stripesize;
2904
2905         /*
2906          * mixed block groups end up with duplicate but slightly offset
2907          * extent buffers for the same range.  It leads to corruptions
2908          */
2909         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2910             (sectorsize != nodesize)) {
2911                 btrfs_err(fs_info,
2912 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2913                         nodesize, sectorsize);
2914                 goto fail_alloc;
2915         }
2916
2917         /*
2918          * Needn't use the lock because there is no other task which will
2919          * update the flag.
2920          */
2921         btrfs_set_super_incompat_flags(disk_super, features);
2922
2923         features = btrfs_super_compat_ro_flags(disk_super) &
2924                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2925         if (!sb_rdonly(sb) && features) {
2926                 btrfs_err(fs_info,
2927         "cannot mount read-write because of unsupported optional features (%llx)",
2928                        features);
2929                 err = -EINVAL;
2930                 goto fail_alloc;
2931         }
2932
2933         ret = btrfs_init_workqueues(fs_info, fs_devices);
2934         if (ret) {
2935                 err = ret;
2936                 goto fail_sb_buffer;
2937         }
2938
2939         sb->s_bdi->congested_fn = btrfs_congested_fn;
2940         sb->s_bdi->congested_data = fs_info;
2941         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2942         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2943         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2944         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2945
2946         sb->s_blocksize = sectorsize;
2947         sb->s_blocksize_bits = blksize_bits(sectorsize);
2948         memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2949
2950         mutex_lock(&fs_info->chunk_mutex);
2951         ret = btrfs_read_sys_array(fs_info);
2952         mutex_unlock(&fs_info->chunk_mutex);
2953         if (ret) {
2954                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2955                 goto fail_sb_buffer;
2956         }
2957
2958         generation = btrfs_super_chunk_root_generation(disk_super);
2959         level = btrfs_super_chunk_root_level(disk_super);
2960
2961         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2962
2963         chunk_root->node = read_tree_block(fs_info,
2964                                            btrfs_super_chunk_root(disk_super),
2965                                            generation, level, NULL);
2966         if (IS_ERR(chunk_root->node) ||
2967             !extent_buffer_uptodate(chunk_root->node)) {
2968                 btrfs_err(fs_info, "failed to read chunk root");
2969                 if (!IS_ERR(chunk_root->node))
2970                         free_extent_buffer(chunk_root->node);
2971                 chunk_root->node = NULL;
2972                 goto fail_tree_roots;
2973         }
2974         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2975         chunk_root->commit_root = btrfs_root_node(chunk_root);
2976
2977         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2978            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2979
2980         ret = btrfs_read_chunk_tree(fs_info);
2981         if (ret) {
2982                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2983                 goto fail_tree_roots;
2984         }
2985
2986         /*
2987          * Keep the devid that is marked to be the target device for the
2988          * device replace procedure
2989          */
2990         btrfs_free_extra_devids(fs_devices, 0);
2991
2992         if (!fs_devices->latest_bdev) {
2993                 btrfs_err(fs_info, "failed to read devices");
2994                 goto fail_tree_roots;
2995         }
2996
2997 retry_root_backup:
2998         generation = btrfs_super_generation(disk_super);
2999         level = btrfs_super_root_level(disk_super);
3000
3001         tree_root->node = read_tree_block(fs_info,
3002                                           btrfs_super_root(disk_super),
3003                                           generation, level, NULL);
3004         if (IS_ERR(tree_root->node) ||
3005             !extent_buffer_uptodate(tree_root->node)) {
3006                 btrfs_warn(fs_info, "failed to read tree root");
3007                 if (!IS_ERR(tree_root->node))
3008                         free_extent_buffer(tree_root->node);
3009                 tree_root->node = NULL;
3010                 goto recovery_tree_root;
3011         }
3012
3013         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3014         tree_root->commit_root = btrfs_root_node(tree_root);
3015         btrfs_set_root_refs(&tree_root->root_item, 1);
3016
3017         mutex_lock(&tree_root->objectid_mutex);
3018         ret = btrfs_find_highest_objectid(tree_root,
3019                                         &tree_root->highest_objectid);
3020         if (ret) {
3021                 mutex_unlock(&tree_root->objectid_mutex);
3022                 goto recovery_tree_root;
3023         }
3024
3025         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3026
3027         mutex_unlock(&tree_root->objectid_mutex);
3028
3029         ret = btrfs_read_roots(fs_info);
3030         if (ret)
3031                 goto recovery_tree_root;
3032
3033         fs_info->generation = generation;
3034         fs_info->last_trans_committed = generation;
3035
3036         ret = btrfs_recover_balance(fs_info);
3037         if (ret) {
3038                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3039                 goto fail_block_groups;
3040         }
3041
3042         ret = btrfs_init_dev_stats(fs_info);
3043         if (ret) {
3044                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3045                 goto fail_block_groups;
3046         }
3047
3048         ret = btrfs_init_dev_replace(fs_info);
3049         if (ret) {
3050                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3051                 goto fail_block_groups;
3052         }
3053
3054         btrfs_free_extra_devids(fs_devices, 1);
3055
3056         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3057         if (ret) {
3058                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3059                                 ret);
3060                 goto fail_block_groups;
3061         }
3062
3063         ret = btrfs_sysfs_add_device(fs_devices);
3064         if (ret) {
3065                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3066                                 ret);
3067                 goto fail_fsdev_sysfs;
3068         }
3069
3070         ret = btrfs_sysfs_add_mounted(fs_info);
3071         if (ret) {
3072                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3073                 goto fail_fsdev_sysfs;
3074         }
3075
3076         ret = btrfs_init_space_info(fs_info);
3077         if (ret) {
3078                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3079                 goto fail_sysfs;
3080         }
3081
3082         ret = btrfs_read_block_groups(fs_info);
3083         if (ret) {
3084                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3085                 goto fail_sysfs;
3086         }
3087
3088         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3089                 btrfs_warn(fs_info,
3090                 "writeable mount is not allowed due to too many missing devices");
3091                 goto fail_sysfs;
3092         }
3093
3094         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3095                                                "btrfs-cleaner");
3096         if (IS_ERR(fs_info->cleaner_kthread))
3097                 goto fail_sysfs;
3098
3099         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3100                                                    tree_root,
3101                                                    "btrfs-transaction");
3102         if (IS_ERR(fs_info->transaction_kthread))
3103                 goto fail_cleaner;
3104
3105         if (!btrfs_test_opt(fs_info, NOSSD) &&
3106             !fs_info->fs_devices->rotating) {
3107                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3108         }
3109
3110         /*
3111          * Mount does not set all options immediately, we can do it now and do
3112          * not have to wait for transaction commit
3113          */
3114         btrfs_apply_pending_changes(fs_info);
3115
3116 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3117         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3118                 ret = btrfsic_mount(fs_info, fs_devices,
3119                                     btrfs_test_opt(fs_info,
3120                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3121                                     1 : 0,
3122                                     fs_info->check_integrity_print_mask);
3123                 if (ret)
3124                         btrfs_warn(fs_info,
3125                                 "failed to initialize integrity check module: %d",
3126                                 ret);
3127         }
3128 #endif
3129         ret = btrfs_read_qgroup_config(fs_info);
3130         if (ret)
3131                 goto fail_trans_kthread;
3132
3133         if (btrfs_build_ref_tree(fs_info))
3134                 btrfs_err(fs_info, "couldn't build ref tree");
3135
3136         /* do not make disk changes in broken FS or nologreplay is given */
3137         if (btrfs_super_log_root(disk_super) != 0 &&
3138             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3139                 ret = btrfs_replay_log(fs_info, fs_devices);
3140                 if (ret) {
3141                         err = ret;
3142                         goto fail_qgroup;
3143                 }
3144         }
3145
3146         ret = btrfs_find_orphan_roots(fs_info);
3147         if (ret)
3148                 goto fail_qgroup;
3149
3150         if (!sb_rdonly(sb)) {
3151                 ret = btrfs_cleanup_fs_roots(fs_info);
3152                 if (ret)
3153                         goto fail_qgroup;
3154
3155                 mutex_lock(&fs_info->cleaner_mutex);
3156                 ret = btrfs_recover_relocation(tree_root);
3157                 mutex_unlock(&fs_info->cleaner_mutex);
3158                 if (ret < 0) {
3159                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3160                                         ret);
3161                         err = -EINVAL;
3162                         goto fail_qgroup;
3163                 }
3164         }
3165
3166         location.objectid = BTRFS_FS_TREE_OBJECTID;
3167         location.type = BTRFS_ROOT_ITEM_KEY;
3168         location.offset = 0;
3169
3170         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3171         if (IS_ERR(fs_info->fs_root)) {
3172                 err = PTR_ERR(fs_info->fs_root);
3173                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3174                 goto fail_qgroup;
3175         }
3176
3177         if (sb_rdonly(sb))
3178                 return 0;
3179
3180         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3181             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3182                 clear_free_space_tree = 1;
3183         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3184                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3185                 btrfs_warn(fs_info, "free space tree is invalid");
3186                 clear_free_space_tree = 1;
3187         }
3188
3189         if (clear_free_space_tree) {
3190                 btrfs_info(fs_info, "clearing free space tree");
3191                 ret = btrfs_clear_free_space_tree(fs_info);
3192                 if (ret) {
3193                         btrfs_warn(fs_info,
3194                                    "failed to clear free space tree: %d", ret);
3195                         close_ctree(fs_info);
3196                         return ret;
3197                 }
3198         }
3199
3200         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3201             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3202                 btrfs_info(fs_info, "creating free space tree");
3203                 ret = btrfs_create_free_space_tree(fs_info);
3204                 if (ret) {
3205                         btrfs_warn(fs_info,
3206                                 "failed to create free space tree: %d", ret);
3207                         close_ctree(fs_info);
3208                         return ret;
3209                 }
3210         }
3211
3212         down_read(&fs_info->cleanup_work_sem);
3213         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3214             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3215                 up_read(&fs_info->cleanup_work_sem);
3216                 close_ctree(fs_info);
3217                 return ret;
3218         }
3219         up_read(&fs_info->cleanup_work_sem);
3220
3221         ret = btrfs_resume_balance_async(fs_info);
3222         if (ret) {
3223                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3224                 close_ctree(fs_info);
3225                 return ret;
3226         }
3227
3228         ret = btrfs_resume_dev_replace_async(fs_info);
3229         if (ret) {
3230                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3231                 close_ctree(fs_info);
3232                 return ret;
3233         }
3234
3235         btrfs_qgroup_rescan_resume(fs_info);
3236
3237         if (!fs_info->uuid_root) {
3238                 btrfs_info(fs_info, "creating UUID tree");
3239                 ret = btrfs_create_uuid_tree(fs_info);
3240                 if (ret) {
3241                         btrfs_warn(fs_info,
3242                                 "failed to create the UUID tree: %d", ret);
3243                         close_ctree(fs_info);
3244                         return ret;
3245                 }
3246         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3247                    fs_info->generation !=
3248                                 btrfs_super_uuid_tree_generation(disk_super)) {
3249                 btrfs_info(fs_info, "checking UUID tree");
3250                 ret = btrfs_check_uuid_tree(fs_info);
3251                 if (ret) {
3252                         btrfs_warn(fs_info,
3253                                 "failed to check the UUID tree: %d", ret);
3254                         close_ctree(fs_info);
3255                         return ret;
3256                 }
3257         } else {
3258                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3259         }
3260         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3261
3262         /*
3263          * backuproot only affect mount behavior, and if open_ctree succeeded,
3264          * no need to keep the flag
3265          */
3266         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3267
3268         return 0;
3269
3270 fail_qgroup:
3271         btrfs_free_qgroup_config(fs_info);
3272 fail_trans_kthread:
3273         kthread_stop(fs_info->transaction_kthread);
3274         btrfs_cleanup_transaction(fs_info);
3275         btrfs_free_fs_roots(fs_info);
3276 fail_cleaner:
3277         kthread_stop(fs_info->cleaner_kthread);
3278
3279         /*
3280          * make sure we're done with the btree inode before we stop our
3281          * kthreads
3282          */
3283         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3284
3285 fail_sysfs:
3286         btrfs_sysfs_remove_mounted(fs_info);
3287
3288 fail_fsdev_sysfs:
3289         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3290
3291 fail_block_groups:
3292         btrfs_put_block_group_cache(fs_info);
3293
3294 fail_tree_roots:
3295         free_root_pointers(fs_info, 1);
3296         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3297
3298 fail_sb_buffer:
3299         btrfs_stop_all_workers(fs_info);
3300         btrfs_free_block_groups(fs_info);
3301 fail_alloc:
3302 fail_iput:
3303         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3304
3305         iput(fs_info->btree_inode);
3306 fail_bio_counter:
3307         percpu_counter_destroy(&fs_info->bio_counter);
3308 fail_delalloc_bytes:
3309         percpu_counter_destroy(&fs_info->delalloc_bytes);
3310 fail_dirty_metadata_bytes:
3311         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3312 fail_srcu:
3313         cleanup_srcu_struct(&fs_info->subvol_srcu);
3314 fail:
3315         btrfs_free_stripe_hash_table(fs_info);
3316         btrfs_close_devices(fs_info->fs_devices);
3317         return err;
3318
3319 recovery_tree_root:
3320         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3321                 goto fail_tree_roots;
3322
3323         free_root_pointers(fs_info, 0);
3324
3325         /* don't use the log in recovery mode, it won't be valid */
3326         btrfs_set_super_log_root(disk_super, 0);
3327
3328         /* we can't trust the free space cache either */
3329         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3330
3331         ret = next_root_backup(fs_info, fs_info->super_copy,
3332                                &num_backups_tried, &backup_index);
3333         if (ret == -1)
3334                 goto fail_block_groups;
3335         goto retry_root_backup;
3336 }
3337 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3338
3339 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3340 {
3341         if (uptodate) {
3342                 set_buffer_uptodate(bh);
3343         } else {
3344                 struct btrfs_device *device = (struct btrfs_device *)
3345                         bh->b_private;
3346
3347                 btrfs_warn_rl_in_rcu(device->fs_info,
3348                                 "lost page write due to IO error on %s",
3349                                           rcu_str_deref(device->name));
3350                 /* note, we don't set_buffer_write_io_error because we have
3351                  * our own ways of dealing with the IO errors
3352                  */
3353                 clear_buffer_uptodate(bh);
3354                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3355         }
3356         unlock_buffer(bh);
3357         put_bh(bh);
3358 }
3359
3360 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3361                         struct buffer_head **bh_ret)
3362 {
3363         struct buffer_head *bh;
3364         struct btrfs_super_block *super;
3365         u64 bytenr;
3366
3367         bytenr = btrfs_sb_offset(copy_num);
3368         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3369                 return -EINVAL;
3370
3371         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3372         /*
3373          * If we fail to read from the underlying devices, as of now
3374          * the best option we have is to mark it EIO.
3375          */
3376         if (!bh)
3377                 return -EIO;
3378
3379         super = (struct btrfs_super_block *)bh->b_data;
3380         if (btrfs_super_bytenr(super) != bytenr ||
3381                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3382                 brelse(bh);
3383                 return -EINVAL;
3384         }
3385
3386         *bh_ret = bh;
3387         return 0;
3388 }
3389
3390
3391 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3392 {
3393         struct buffer_head *bh;
3394         struct buffer_head *latest = NULL;
3395         struct btrfs_super_block *super;
3396         int i;
3397         u64 transid = 0;
3398         int ret = -EINVAL;
3399
3400         /* we would like to check all the supers, but that would make
3401          * a btrfs mount succeed after a mkfs from a different FS.
3402          * So, we need to add a special mount option to scan for
3403          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3404          */
3405         for (i = 0; i < 1; i++) {
3406                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3407                 if (ret)
3408                         continue;
3409
3410                 super = (struct btrfs_super_block *)bh->b_data;
3411
3412                 if (!latest || btrfs_super_generation(super) > transid) {
3413                         brelse(latest);
3414                         latest = bh;
3415                         transid = btrfs_super_generation(super);
3416                 } else {
3417                         brelse(bh);
3418                 }
3419         }
3420
3421         if (!latest)
3422                 return ERR_PTR(ret);
3423
3424         return latest;
3425 }
3426
3427 /*
3428  * Write superblock @sb to the @device. Do not wait for completion, all the
3429  * buffer heads we write are pinned.
3430  *
3431  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3432  * the expected device size at commit time. Note that max_mirrors must be
3433  * same for write and wait phases.
3434  *
3435  * Return number of errors when buffer head is not found or submission fails.
3436  */
3437 static int write_dev_supers(struct btrfs_device *device,
3438                             struct btrfs_super_block *sb, int max_mirrors)
3439 {
3440         struct buffer_head *bh;
3441         int i;
3442         int ret;
3443         int errors = 0;
3444         u32 crc;
3445         u64 bytenr;
3446         int op_flags;
3447
3448         if (max_mirrors == 0)
3449                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3450
3451         for (i = 0; i < max_mirrors; i++) {
3452                 bytenr = btrfs_sb_offset(i);
3453                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3454                     device->commit_total_bytes)
3455                         break;
3456
3457                 btrfs_set_super_bytenr(sb, bytenr);
3458
3459                 crc = ~(u32)0;
3460                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3461                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3462                 btrfs_csum_final(crc, sb->csum);
3463
3464                 /* One reference for us, and we leave it for the caller */
3465                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3466                               BTRFS_SUPER_INFO_SIZE);
3467                 if (!bh) {
3468                         btrfs_err(device->fs_info,
3469                             "couldn't get super buffer head for bytenr %llu",
3470                             bytenr);
3471                         errors++;
3472                         continue;
3473                 }
3474
3475                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3476
3477                 /* one reference for submit_bh */
3478                 get_bh(bh);
3479
3480                 set_buffer_uptodate(bh);
3481                 lock_buffer(bh);
3482                 bh->b_end_io = btrfs_end_buffer_write_sync;
3483                 bh->b_private = device;
3484
3485                 /*
3486                  * we fua the first super.  The others we allow
3487                  * to go down lazy.
3488                  */
3489                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3490                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3491                         op_flags |= REQ_FUA;
3492                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3493                 if (ret)
3494                         errors++;
3495         }
3496         return errors < i ? 0 : -1;
3497 }
3498
3499 /*
3500  * Wait for write completion of superblocks done by write_dev_supers,
3501  * @max_mirrors same for write and wait phases.
3502  *
3503  * Return number of errors when buffer head is not found or not marked up to
3504  * date.
3505  */
3506 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3507 {
3508         struct buffer_head *bh;
3509         int i;
3510         int errors = 0;
3511         bool primary_failed = false;
3512         u64 bytenr;
3513
3514         if (max_mirrors == 0)
3515                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3516
3517         for (i = 0; i < max_mirrors; i++) {
3518                 bytenr = btrfs_sb_offset(i);
3519                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3520                     device->commit_total_bytes)
3521                         break;
3522
3523                 bh = __find_get_block(device->bdev,
3524                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3525                                       BTRFS_SUPER_INFO_SIZE);
3526                 if (!bh) {
3527                         errors++;
3528                         if (i == 0)
3529                                 primary_failed = true;
3530                         continue;
3531                 }
3532                 wait_on_buffer(bh);
3533                 if (!buffer_uptodate(bh)) {
3534                         errors++;
3535                         if (i == 0)
3536                                 primary_failed = true;
3537                 }
3538
3539                 /* drop our reference */
3540                 brelse(bh);
3541
3542                 /* drop the reference from the writing run */
3543                 brelse(bh);
3544         }
3545
3546         /* log error, force error return */
3547         if (primary_failed) {
3548                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3549                           device->devid);
3550                 return -1;
3551         }
3552
3553         return errors < i ? 0 : -1;
3554 }
3555
3556 /*
3557  * endio for the write_dev_flush, this will wake anyone waiting
3558  * for the barrier when it is done
3559  */
3560 static void btrfs_end_empty_barrier(struct bio *bio)
3561 {
3562         complete(bio->bi_private);
3563 }
3564
3565 /*
3566  * Submit a flush request to the device if it supports it. Error handling is
3567  * done in the waiting counterpart.
3568  */
3569 static void write_dev_flush(struct btrfs_device *device)
3570 {
3571         struct request_queue *q = bdev_get_queue(device->bdev);
3572         struct bio *bio = device->flush_bio;
3573
3574         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3575                 return;
3576
3577         bio_reset(bio);
3578         bio->bi_end_io = btrfs_end_empty_barrier;
3579         bio_set_dev(bio, device->bdev);
3580         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3581         init_completion(&device->flush_wait);
3582         bio->bi_private = &device->flush_wait;
3583
3584         btrfsic_submit_bio(bio);
3585         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3586 }
3587
3588 /*
3589  * If the flush bio has been submitted by write_dev_flush, wait for it.
3590  */
3591 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3592 {
3593         struct bio *bio = device->flush_bio;
3594
3595         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3596                 return BLK_STS_OK;
3597
3598         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3599         wait_for_completion_io(&device->flush_wait);
3600
3601         return bio->bi_status;
3602 }
3603
3604 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3605 {
3606         if (!btrfs_check_rw_degradable(fs_info, NULL))
3607                 return -EIO;
3608         return 0;
3609 }
3610
3611 /*
3612  * send an empty flush down to each device in parallel,
3613  * then wait for them
3614  */
3615 static int barrier_all_devices(struct btrfs_fs_info *info)
3616 {
3617         struct list_head *head;
3618         struct btrfs_device *dev;
3619         int errors_wait = 0;
3620         blk_status_t ret;
3621
3622         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3623         /* send down all the barriers */
3624         head = &info->fs_devices->devices;
3625         list_for_each_entry(dev, head, dev_list) {
3626                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3627                         continue;
3628                 if (!dev->bdev)
3629                         continue;
3630                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3631                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3632                         continue;
3633
3634                 write_dev_flush(dev);
3635                 dev->last_flush_error = BLK_STS_OK;
3636         }
3637
3638         /* wait for all the barriers */
3639         list_for_each_entry(dev, head, dev_list) {
3640                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3641                         continue;
3642                 if (!dev->bdev) {
3643                         errors_wait++;
3644                         continue;
3645                 }
3646                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3647                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3648                         continue;
3649
3650                 ret = wait_dev_flush(dev);
3651                 if (ret) {
3652                         dev->last_flush_error = ret;
3653                         btrfs_dev_stat_inc_and_print(dev,
3654                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3655                         errors_wait++;
3656                 }
3657         }
3658
3659         if (errors_wait) {
3660                 /*
3661                  * At some point we need the status of all disks
3662                  * to arrive at the volume status. So error checking
3663                  * is being pushed to a separate loop.
3664                  */
3665                 return check_barrier_error(info);
3666         }
3667         return 0;
3668 }
3669
3670 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3671 {
3672         int raid_type;
3673         int min_tolerated = INT_MAX;
3674
3675         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3676             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3677                 min_tolerated = min(min_tolerated,
3678                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3679                                     tolerated_failures);
3680
3681         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3682                 if (raid_type == BTRFS_RAID_SINGLE)
3683                         continue;
3684                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3685                         continue;
3686                 min_tolerated = min(min_tolerated,
3687                                     btrfs_raid_array[raid_type].
3688                                     tolerated_failures);
3689         }
3690
3691         if (min_tolerated == INT_MAX) {
3692                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3693                 min_tolerated = 0;
3694         }
3695
3696         return min_tolerated;
3697 }
3698
3699 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3700 {
3701         struct list_head *head;
3702         struct btrfs_device *dev;
3703         struct btrfs_super_block *sb;
3704         struct btrfs_dev_item *dev_item;
3705         int ret;
3706         int do_barriers;
3707         int max_errors;
3708         int total_errors = 0;
3709         u64 flags;
3710
3711         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3712
3713         /*
3714          * max_mirrors == 0 indicates we're from commit_transaction,
3715          * not from fsync where the tree roots in fs_info have not
3716          * been consistent on disk.
3717          */
3718         if (max_mirrors == 0)
3719                 backup_super_roots(fs_info);
3720
3721         sb = fs_info->super_for_commit;
3722         dev_item = &sb->dev_item;
3723
3724         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3725         head = &fs_info->fs_devices->devices;
3726         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3727
3728         if (do_barriers) {
3729                 ret = barrier_all_devices(fs_info);
3730                 if (ret) {
3731                         mutex_unlock(
3732                                 &fs_info->fs_devices->device_list_mutex);
3733                         btrfs_handle_fs_error(fs_info, ret,
3734                                               "errors while submitting device barriers.");
3735                         return ret;
3736                 }
3737         }
3738
3739         list_for_each_entry(dev, head, dev_list) {
3740                 if (!dev->bdev) {
3741                         total_errors++;
3742                         continue;
3743                 }
3744                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3745                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3746                         continue;
3747
3748                 btrfs_set_stack_device_generation(dev_item, 0);
3749                 btrfs_set_stack_device_type(dev_item, dev->type);
3750                 btrfs_set_stack_device_id(dev_item, dev->devid);
3751                 btrfs_set_stack_device_total_bytes(dev_item,
3752                                                    dev->commit_total_bytes);
3753                 btrfs_set_stack_device_bytes_used(dev_item,
3754                                                   dev->commit_bytes_used);
3755                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3756                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3757                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3758                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3759                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3760
3761                 flags = btrfs_super_flags(sb);
3762                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3763
3764                 ret = btrfs_validate_write_super(fs_info, sb);
3765                 if (ret < 0) {
3766                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3767                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
3768                                 "unexpected superblock corruption detected");
3769                         return -EUCLEAN;
3770                 }
3771
3772                 ret = write_dev_supers(dev, sb, max_mirrors);
3773                 if (ret)
3774                         total_errors++;
3775         }
3776         if (total_errors > max_errors) {
3777                 btrfs_err(fs_info, "%d errors while writing supers",
3778                           total_errors);
3779                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3780
3781                 /* FUA is masked off if unsupported and can't be the reason */