de4b7ed02da11ae2463ff1639f6c0224b45958b6
[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 <linux/sched/mm.h>
21 #include <asm/unaligned.h>
22 #include "ctree.h"
23 #include "disk-io.h"
24 #include "transaction.h"
25 #include "btrfs_inode.h"
26 #include "volumes.h"
27 #include "print-tree.h"
28 #include "locking.h"
29 #include "tree-log.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "inode-map.h"
33 #include "check-integrity.h"
34 #include "rcu-string.h"
35 #include "dev-replace.h"
36 #include "raid56.h"
37 #include "sysfs.h"
38 #include "qgroup.h"
39 #include "compression.h"
40 #include "tree-checker.h"
41 #include "ref-verify.h"
42
43 #ifdef CONFIG_X86
44 #include <asm/cpufeature.h>
45 #endif
46
47 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
48                                  BTRFS_HEADER_FLAG_RELOC |\
49                                  BTRFS_SUPER_FLAG_ERROR |\
50                                  BTRFS_SUPER_FLAG_SEEDING |\
51                                  BTRFS_SUPER_FLAG_METADUMP |\
52                                  BTRFS_SUPER_FLAG_METADUMP_V2)
53
54 static const struct extent_io_ops btree_extent_io_ops;
55 static void end_workqueue_fn(struct btrfs_work *work);
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->root_key.objectid.  This ensures that all special purpose
130  * roots 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                 /*
284                  * Note: we don't need to check for the err == 1 case here, as
285                  * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
286                  * and 'min_len = 32' and the currently implemented mapping
287                  * algorithm we cannot cross a page boundary.
288                  */
289                 err = map_private_extent_buffer(buf, offset, 32,
290                                         &kaddr, &map_start, &map_len);
291                 if (err)
292                         return err;
293                 cur_len = min(len, map_len - (offset - map_start));
294                 crc = btrfs_csum_data(kaddr + offset - map_start,
295                                       crc, cur_len);
296                 len -= cur_len;
297                 offset += cur_len;
298         }
299         memset(result, 0, BTRFS_CSUM_SIZE);
300
301         btrfs_csum_final(crc, result);
302
303         if (verify) {
304                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
305                         u32 val;
306                         u32 found = 0;
307                         memcpy(&found, result, csum_size);
308
309                         read_extent_buffer(buf, &val, 0, csum_size);
310                         btrfs_warn_rl(fs_info,
311                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
312                                 fs_info->sb->s_id, buf->start,
313                                 val, found, btrfs_header_level(buf));
314                         return -EUCLEAN;
315                 }
316         } else {
317                 write_extent_buffer(buf, result, 0, csum_size);
318         }
319
320         return 0;
321 }
322
323 /*
324  * we can't consider a given block up to date unless the transid of the
325  * block matches the transid in the parent node's pointer.  This is how we
326  * detect blocks that either didn't get written at all or got written
327  * in the wrong place.
328  */
329 static int verify_parent_transid(struct extent_io_tree *io_tree,
330                                  struct extent_buffer *eb, u64 parent_transid,
331                                  int atomic)
332 {
333         struct extent_state *cached_state = NULL;
334         int ret;
335         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
336
337         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
338                 return 0;
339
340         if (atomic)
341                 return -EAGAIN;
342
343         if (need_lock) {
344                 btrfs_tree_read_lock(eb);
345                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
346         }
347
348         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
349                          &cached_state);
350         if (extent_buffer_uptodate(eb) &&
351             btrfs_header_generation(eb) == parent_transid) {
352                 ret = 0;
353                 goto out;
354         }
355         btrfs_err_rl(eb->fs_info,
356                 "parent transid verify failed on %llu wanted %llu found %llu",
357                         eb->start,
358                         parent_transid, btrfs_header_generation(eb));
359         ret = 1;
360
361         /*
362          * Things reading via commit roots that don't have normal protection,
363          * like send, can have a really old block in cache that may point at a
364          * block that has been freed and re-allocated.  So don't clear uptodate
365          * if we find an eb that is under IO (dirty/writeback) because we could
366          * end up reading in the stale data and then writing it back out and
367          * making everybody very sad.
368          */
369         if (!extent_buffer_under_io(eb))
370                 clear_extent_buffer_uptodate(eb);
371 out:
372         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
373                              &cached_state);
374         if (need_lock)
375                 btrfs_tree_read_unlock_blocking(eb);
376         return ret;
377 }
378
379 /*
380  * Return 0 if the superblock checksum type matches the checksum value of that
381  * algorithm. Pass the raw disk superblock data.
382  */
383 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
384                                   char *raw_disk_sb)
385 {
386         struct btrfs_super_block *disk_sb =
387                 (struct btrfs_super_block *)raw_disk_sb;
388         u16 csum_type = btrfs_super_csum_type(disk_sb);
389         int ret = 0;
390
391         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
392                 u32 crc = ~(u32)0;
393                 char result[sizeof(crc)];
394
395                 /*
396                  * The super_block structure does not span the whole
397                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
398                  * is filled with zeros and is included in the checksum.
399                  */
400                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
401                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
402                 btrfs_csum_final(crc, result);
403
404                 if (memcmp(raw_disk_sb, result, sizeof(result)))
405                         ret = 1;
406         }
407
408         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
409                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
410                                 csum_type);
411                 ret = 1;
412         }
413
414         return ret;
415 }
416
417 static int verify_level_key(struct btrfs_fs_info *fs_info,
418                             struct extent_buffer *eb, int level,
419                             struct btrfs_key *first_key, u64 parent_transid)
420 {
421         int found_level;
422         struct btrfs_key found_key;
423         int ret;
424
425         found_level = btrfs_header_level(eb);
426         if (found_level != level) {
427 #ifdef CONFIG_BTRFS_DEBUG
428                 WARN_ON(1);
429                 btrfs_err(fs_info,
430 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
431                           eb->start, level, found_level);
432 #endif
433                 return -EIO;
434         }
435
436         if (!first_key)
437                 return 0;
438
439         /*
440          * For live tree block (new tree blocks in current transaction),
441          * we need proper lock context to avoid race, which is impossible here.
442          * So we only checks tree blocks which is read from disk, whose
443          * generation <= fs_info->last_trans_committed.
444          */
445         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
446                 return 0;
447         if (found_level)
448                 btrfs_node_key_to_cpu(eb, &found_key, 0);
449         else
450                 btrfs_item_key_to_cpu(eb, &found_key, 0);
451         ret = btrfs_comp_cpu_keys(first_key, &found_key);
452
453 #ifdef CONFIG_BTRFS_DEBUG
454         if (ret) {
455                 WARN_ON(1);
456                 btrfs_err(fs_info,
457 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
458                           eb->start, parent_transid, first_key->objectid,
459                           first_key->type, first_key->offset,
460                           found_key.objectid, found_key.type,
461                           found_key.offset);
462         }
463 #endif
464         return ret;
465 }
466
467 /*
468  * helper to read a given tree block, doing retries as required when
469  * the checksums don't match and we have alternate mirrors to try.
470  *
471  * @parent_transid:     expected transid, skip check if 0
472  * @level:              expected level, mandatory check
473  * @first_key:          expected key of first slot, skip check if NULL
474  */
475 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
476                                           struct extent_buffer *eb,
477                                           u64 parent_transid, int level,
478                                           struct btrfs_key *first_key)
479 {
480         struct extent_io_tree *io_tree;
481         int failed = 0;
482         int ret;
483         int num_copies = 0;
484         int mirror_num = 0;
485         int failed_mirror = 0;
486
487         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
488         while (1) {
489                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
490                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
491                                                mirror_num);
492                 if (!ret) {
493                         if (verify_parent_transid(io_tree, eb,
494                                                    parent_transid, 0))
495                                 ret = -EIO;
496                         else if (verify_level_key(fs_info, eb, level,
497                                                   first_key, parent_transid))
498                                 ret = -EUCLEAN;
499                         else
500                                 break;
501                 }
502
503                 num_copies = btrfs_num_copies(fs_info,
504                                               eb->start, eb->len);
505                 if (num_copies == 1)
506                         break;
507
508                 if (!failed_mirror) {
509                         failed = 1;
510                         failed_mirror = eb->read_mirror;
511                 }
512
513                 mirror_num++;
514                 if (mirror_num == failed_mirror)
515                         mirror_num++;
516
517                 if (mirror_num > num_copies)
518                         break;
519         }
520
521         if (failed && !ret && failed_mirror)
522                 repair_eb_io_failure(fs_info, eb, failed_mirror);
523
524         return ret;
525 }
526
527 /*
528  * checksum a dirty tree block before IO.  This has extra checks to make sure
529  * we only fill in the checksum field in the first page of a multi-page block
530  */
531
532 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
533 {
534         u64 start = page_offset(page);
535         u64 found_start;
536         struct extent_buffer *eb;
537
538         eb = (struct extent_buffer *)page->private;
539         if (page != eb->pages[0])
540                 return 0;
541
542         found_start = btrfs_header_bytenr(eb);
543         /*
544          * Please do not consolidate these warnings into a single if.
545          * It is useful to know what went wrong.
546          */
547         if (WARN_ON(found_start != start))
548                 return -EUCLEAN;
549         if (WARN_ON(!PageUptodate(page)))
550                 return -EUCLEAN;
551
552         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
553                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
554
555         return csum_tree_block(fs_info, eb, 0);
556 }
557
558 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
559                                  struct extent_buffer *eb)
560 {
561         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
562         u8 fsid[BTRFS_FSID_SIZE];
563         int ret = 1;
564
565         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
566         while (fs_devices) {
567                 u8 *metadata_uuid;
568
569                 /*
570                  * Checking the incompat flag is only valid for the current
571                  * fs. For seed devices it's forbidden to have their uuid
572                  * changed so reading ->fsid in this case is fine
573                  */
574                 if (fs_devices == fs_info->fs_devices &&
575                     btrfs_fs_incompat(fs_info, METADATA_UUID))
576                         metadata_uuid = fs_devices->metadata_uuid;
577                 else
578                         metadata_uuid = fs_devices->fsid;
579
580                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
581                         ret = 0;
582                         break;
583                 }
584                 fs_devices = fs_devices->seed;
585         }
586         return ret;
587 }
588
589 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
590                                       u64 phy_offset, struct page *page,
591                                       u64 start, u64 end, int mirror)
592 {
593         u64 found_start;
594         int found_level;
595         struct extent_buffer *eb;
596         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
597         struct btrfs_fs_info *fs_info = root->fs_info;
598         int ret = 0;
599         int reads_done;
600
601         if (!page->private)
602                 goto out;
603
604         eb = (struct extent_buffer *)page->private;
605
606         /* the pending IO might have been the only thing that kept this buffer
607          * in memory.  Make sure we have a ref for all this other checks
608          */
609         extent_buffer_get(eb);
610
611         reads_done = atomic_dec_and_test(&eb->io_pages);
612         if (!reads_done)
613                 goto err;
614
615         eb->read_mirror = mirror;
616         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
617                 ret = -EIO;
618                 goto err;
619         }
620
621         found_start = btrfs_header_bytenr(eb);
622         if (found_start != eb->start) {
623                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
624                              eb->start, found_start);
625                 ret = -EIO;
626                 goto err;
627         }
628         if (check_tree_block_fsid(fs_info, eb)) {
629                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
630                              eb->start);
631                 ret = -EIO;
632                 goto err;
633         }
634         found_level = btrfs_header_level(eb);
635         if (found_level >= BTRFS_MAX_LEVEL) {
636                 btrfs_err(fs_info, "bad tree block level %d on %llu",
637                           (int)btrfs_header_level(eb), eb->start);
638                 ret = -EIO;
639                 goto err;
640         }
641
642         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
643                                        eb, found_level);
644
645         ret = csum_tree_block(fs_info, eb, 1);
646         if (ret)
647                 goto err;
648
649         /*
650          * If this is a leaf block and it is corrupt, set the corrupt bit so
651          * that we don't try and read the other copies of this block, just
652          * return -EIO.
653          */
654         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
655                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
656                 ret = -EIO;
657         }
658
659         if (found_level > 0 && btrfs_check_node(fs_info, eb))
660                 ret = -EIO;
661
662         if (!ret)
663                 set_extent_buffer_uptodate(eb);
664 err:
665         if (reads_done &&
666             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
667                 btree_readahead_hook(eb, ret);
668
669         if (ret) {
670                 /*
671                  * our io error hook is going to dec the io pages
672                  * again, we have to make sure it has something
673                  * to decrement
674                  */
675                 atomic_inc(&eb->io_pages);
676                 clear_extent_buffer_uptodate(eb);
677         }
678         free_extent_buffer(eb);
679 out:
680         return ret;
681 }
682
683 static void end_workqueue_bio(struct bio *bio)
684 {
685         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
686         struct btrfs_fs_info *fs_info;
687         struct btrfs_workqueue *wq;
688         btrfs_work_func_t func;
689
690         fs_info = end_io_wq->info;
691         end_io_wq->status = bio->bi_status;
692
693         if (bio_op(bio) == REQ_OP_WRITE) {
694                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
695                         wq = fs_info->endio_meta_write_workers;
696                         func = btrfs_endio_meta_write_helper;
697                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
698                         wq = fs_info->endio_freespace_worker;
699                         func = btrfs_freespace_write_helper;
700                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
701                         wq = fs_info->endio_raid56_workers;
702                         func = btrfs_endio_raid56_helper;
703                 } else {
704                         wq = fs_info->endio_write_workers;
705                         func = btrfs_endio_write_helper;
706                 }
707         } else {
708                 if (unlikely(end_io_wq->metadata ==
709                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
710                         wq = fs_info->endio_repair_workers;
711                         func = btrfs_endio_repair_helper;
712                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
713                         wq = fs_info->endio_raid56_workers;
714                         func = btrfs_endio_raid56_helper;
715                 } else if (end_io_wq->metadata) {
716                         wq = fs_info->endio_meta_workers;
717                         func = btrfs_endio_meta_helper;
718                 } else {
719                         wq = fs_info->endio_workers;
720                         func = btrfs_endio_helper;
721                 }
722         }
723
724         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
725         btrfs_queue_work(wq, &end_io_wq->work);
726 }
727
728 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
729                         enum btrfs_wq_endio_type metadata)
730 {
731         struct btrfs_end_io_wq *end_io_wq;
732
733         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
734         if (!end_io_wq)
735                 return BLK_STS_RESOURCE;
736
737         end_io_wq->private = bio->bi_private;
738         end_io_wq->end_io = bio->bi_end_io;
739         end_io_wq->info = info;
740         end_io_wq->status = 0;
741         end_io_wq->bio = bio;
742         end_io_wq->metadata = metadata;
743
744         bio->bi_private = end_io_wq;
745         bio->bi_end_io = end_workqueue_bio;
746         return 0;
747 }
748
749 static void run_one_async_start(struct btrfs_work *work)
750 {
751         struct async_submit_bio *async;
752         blk_status_t ret;
753
754         async = container_of(work, struct  async_submit_bio, work);
755         ret = async->submit_bio_start(async->private_data, async->bio,
756                                       async->bio_offset);
757         if (ret)
758                 async->status = ret;
759 }
760
761 /*
762  * In order to insert checksums into the metadata in large chunks, we wait
763  * until bio submission time.   All the pages in the bio are checksummed and
764  * sums are attached onto the ordered extent record.
765  *
766  * At IO completion time the csums attached on the ordered extent record are
767  * inserted into the tree.
768  */
769 static void run_one_async_done(struct btrfs_work *work)
770 {
771         struct async_submit_bio *async;
772         struct inode *inode;
773         blk_status_t ret;
774
775         async = container_of(work, struct  async_submit_bio, work);
776         inode = async->private_data;
777
778         /* If an error occurred we just want to clean up the bio and move on */
779         if (async->status) {
780                 async->bio->bi_status = async->status;
781                 bio_endio(async->bio);
782                 return;
783         }
784
785         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
786                         async->mirror_num, 1);
787         if (ret) {
788                 async->bio->bi_status = ret;
789                 bio_endio(async->bio);
790         }
791 }
792
793 static void run_one_async_free(struct btrfs_work *work)
794 {
795         struct async_submit_bio *async;
796
797         async = container_of(work, struct  async_submit_bio, work);
798         kfree(async);
799 }
800
801 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
802                                  int mirror_num, unsigned long bio_flags,
803                                  u64 bio_offset, void *private_data,
804                                  extent_submit_bio_start_t *submit_bio_start)
805 {
806         struct async_submit_bio *async;
807
808         async = kmalloc(sizeof(*async), GFP_NOFS);
809         if (!async)
810                 return BLK_STS_RESOURCE;
811
812         async->private_data = private_data;
813         async->bio = bio;
814         async->mirror_num = mirror_num;
815         async->submit_bio_start = submit_bio_start;
816
817         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
818                         run_one_async_done, run_one_async_free);
819
820         async->bio_offset = bio_offset;
821
822         async->status = 0;
823
824         if (op_is_sync(bio->bi_opf))
825                 btrfs_set_work_high_priority(&async->work);
826
827         btrfs_queue_work(fs_info->workers, &async->work);
828         return 0;
829 }
830
831 static blk_status_t btree_csum_one_bio(struct bio *bio)
832 {
833         struct bio_vec *bvec;
834         struct btrfs_root *root;
835         int i, ret = 0;
836
837         ASSERT(!bio_flagged(bio, BIO_CLONED));
838         bio_for_each_segment_all(bvec, bio, i) {
839                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
840                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
841                 if (ret)
842                         break;
843         }
844
845         return errno_to_blk_status(ret);
846 }
847
848 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
849                                              u64 bio_offset)
850 {
851         /*
852          * when we're called for a write, we're already in the async
853          * submission context.  Just jump into btrfs_map_bio
854          */
855         return btree_csum_one_bio(bio);
856 }
857
858 static int check_async_write(struct btrfs_inode *bi)
859 {
860         if (atomic_read(&bi->sync_writers))
861                 return 0;
862 #ifdef CONFIG_X86
863         if (static_cpu_has(X86_FEATURE_XMM4_2))
864                 return 0;
865 #endif
866         return 1;
867 }
868
869 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
870                                           int mirror_num, unsigned long bio_flags,
871                                           u64 bio_offset)
872 {
873         struct inode *inode = private_data;
874         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
875         int async = check_async_write(BTRFS_I(inode));
876         blk_status_t ret;
877
878         if (bio_op(bio) != REQ_OP_WRITE) {
879                 /*
880                  * called for a read, do the setup so that checksum validation
881                  * can happen in the async kernel threads
882                  */
883                 ret = btrfs_bio_wq_end_io(fs_info, bio,
884                                           BTRFS_WQ_ENDIO_METADATA);
885                 if (ret)
886                         goto out_w_error;
887                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
888         } else if (!async) {
889                 ret = btree_csum_one_bio(bio);
890                 if (ret)
891                         goto out_w_error;
892                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
893         } else {
894                 /*
895                  * kthread helpers are used to submit writes so that
896                  * checksumming can happen in parallel across all CPUs
897                  */
898                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
899                                           bio_offset, private_data,
900                                           btree_submit_bio_start);
901         }
902
903         if (ret)
904                 goto out_w_error;
905         return 0;
906
907 out_w_error:
908         bio->bi_status = ret;
909         bio_endio(bio);
910         return ret;
911 }
912
913 #ifdef CONFIG_MIGRATION
914 static int btree_migratepage(struct address_space *mapping,
915                         struct page *newpage, struct page *page,
916                         enum migrate_mode mode)
917 {
918         /*
919          * we can't safely write a btree page from here,
920          * we haven't done the locking hook
921          */
922         if (PageDirty(page))
923                 return -EAGAIN;
924         /*
925          * Buffers may be managed in a filesystem specific way.
926          * We must have no buffers or drop them.
927          */
928         if (page_has_private(page) &&
929             !try_to_release_page(page, GFP_KERNEL))
930                 return -EAGAIN;
931         return migrate_page(mapping, newpage, page, mode);
932 }
933 #endif
934
935
936 static int btree_writepages(struct address_space *mapping,
937                             struct writeback_control *wbc)
938 {
939         struct btrfs_fs_info *fs_info;
940         int ret;
941
942         if (wbc->sync_mode == WB_SYNC_NONE) {
943
944                 if (wbc->for_kupdate)
945                         return 0;
946
947                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
948                 /* this is a bit racy, but that's ok */
949                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
950                                              BTRFS_DIRTY_METADATA_THRESH,
951                                              fs_info->dirty_metadata_batch);
952                 if (ret < 0)
953                         return 0;
954         }
955         return btree_write_cache_pages(mapping, wbc);
956 }
957
958 static int btree_readpage(struct file *file, struct page *page)
959 {
960         struct extent_io_tree *tree;
961         tree = &BTRFS_I(page->mapping->host)->io_tree;
962         return extent_read_full_page(tree, page, btree_get_extent, 0);
963 }
964
965 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
966 {
967         if (PageWriteback(page) || PageDirty(page))
968                 return 0;
969
970         return try_release_extent_buffer(page);
971 }
972
973 static void btree_invalidatepage(struct page *page, unsigned int offset,
974                                  unsigned int length)
975 {
976         struct extent_io_tree *tree;
977         tree = &BTRFS_I(page->mapping->host)->io_tree;
978         extent_invalidatepage(tree, page, offset);
979         btree_releasepage(page, GFP_NOFS);
980         if (PagePrivate(page)) {
981                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
982                            "page private not zero on page %llu",
983                            (unsigned long long)page_offset(page));
984                 ClearPagePrivate(page);
985                 set_page_private(page, 0);
986                 put_page(page);
987         }
988 }
989
990 static int btree_set_page_dirty(struct page *page)
991 {
992 #ifdef DEBUG
993         struct extent_buffer *eb;
994
995         BUG_ON(!PagePrivate(page));
996         eb = (struct extent_buffer *)page->private;
997         BUG_ON(!eb);
998         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
999         BUG_ON(!atomic_read(&eb->refs));
1000         btrfs_assert_tree_locked(eb);
1001 #endif
1002         return __set_page_dirty_nobuffers(page);
1003 }
1004
1005 static const struct address_space_operations btree_aops = {
1006         .readpage       = btree_readpage,
1007         .writepages     = btree_writepages,
1008         .releasepage    = btree_releasepage,
1009         .invalidatepage = btree_invalidatepage,
1010 #ifdef CONFIG_MIGRATION
1011         .migratepage    = btree_migratepage,
1012 #endif
1013         .set_page_dirty = btree_set_page_dirty,
1014 };
1015
1016 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1017 {
1018         struct extent_buffer *buf = NULL;
1019         struct inode *btree_inode = fs_info->btree_inode;
1020
1021         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1022         if (IS_ERR(buf))
1023                 return;
1024         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1025                                  buf, WAIT_NONE, 0);
1026         free_extent_buffer(buf);
1027 }
1028
1029 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1030                          int mirror_num, struct extent_buffer **eb)
1031 {
1032         struct extent_buffer *buf = NULL;
1033         struct inode *btree_inode = fs_info->btree_inode;
1034         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1035         int ret;
1036
1037         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1038         if (IS_ERR(buf))
1039                 return 0;
1040
1041         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1042
1043         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1044                                        mirror_num);
1045         if (ret) {
1046                 free_extent_buffer(buf);
1047                 return ret;
1048         }
1049
1050         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1051                 free_extent_buffer(buf);
1052                 return -EIO;
1053         } else if (extent_buffer_uptodate(buf)) {
1054                 *eb = buf;
1055         } else {
1056                 free_extent_buffer(buf);
1057         }
1058         return 0;
1059 }
1060
1061 struct extent_buffer *btrfs_find_create_tree_block(
1062                                                 struct btrfs_fs_info *fs_info,
1063                                                 u64 bytenr)
1064 {
1065         if (btrfs_is_testing(fs_info))
1066                 return alloc_test_extent_buffer(fs_info, bytenr);
1067         return alloc_extent_buffer(fs_info, bytenr);
1068 }
1069
1070
1071 int btrfs_write_tree_block(struct extent_buffer *buf)
1072 {
1073         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1074                                         buf->start + buf->len - 1);
1075 }
1076
1077 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1078 {
1079         filemap_fdatawait_range(buf->pages[0]->mapping,
1080                                 buf->start, buf->start + buf->len - 1);
1081 }
1082
1083 /*
1084  * Read tree block at logical address @bytenr and do variant basic but critical
1085  * verification.
1086  *
1087  * @parent_transid:     expected transid of this tree block, skip check if 0
1088  * @level:              expected level, mandatory check
1089  * @first_key:          expected key in slot 0, skip check if NULL
1090  */
1091 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1092                                       u64 parent_transid, int level,
1093                                       struct btrfs_key *first_key)
1094 {
1095         struct extent_buffer *buf = NULL;
1096         int ret;
1097
1098         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1099         if (IS_ERR(buf))
1100                 return buf;
1101
1102         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1103                                              level, first_key);
1104         if (ret) {
1105                 free_extent_buffer(buf);
1106                 return ERR_PTR(ret);
1107         }
1108         return buf;
1109
1110 }
1111
1112 void clean_tree_block(struct btrfs_fs_info *fs_info,
1113                       struct extent_buffer *buf)
1114 {
1115         if (btrfs_header_generation(buf) ==
1116             fs_info->running_transaction->transid) {
1117                 btrfs_assert_tree_locked(buf);
1118
1119                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1120                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1121                                                  -buf->len,
1122                                                  fs_info->dirty_metadata_batch);
1123                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1124                         btrfs_set_lock_blocking(buf);
1125                         clear_extent_buffer_dirty(buf);
1126                 }
1127         }
1128 }
1129
1130 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1131 {
1132         struct btrfs_subvolume_writers *writers;
1133         int ret;
1134
1135         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1136         if (!writers)
1137                 return ERR_PTR(-ENOMEM);
1138
1139         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1140         if (ret < 0) {
1141                 kfree(writers);
1142                 return ERR_PTR(ret);
1143         }
1144
1145         init_waitqueue_head(&writers->wait);
1146         return writers;
1147 }
1148
1149 static void
1150 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1151 {
1152         percpu_counter_destroy(&writers->counter);
1153         kfree(writers);
1154 }
1155
1156 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1157                          u64 objectid)
1158 {
1159         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1160         root->node = NULL;
1161         root->commit_root = NULL;
1162         root->state = 0;
1163         root->orphan_cleanup_state = 0;
1164
1165         root->last_trans = 0;
1166         root->highest_objectid = 0;
1167         root->nr_delalloc_inodes = 0;
1168         root->nr_ordered_extents = 0;
1169         root->inode_tree = RB_ROOT;
1170         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1171         root->block_rsv = NULL;
1172
1173         INIT_LIST_HEAD(&root->dirty_list);
1174         INIT_LIST_HEAD(&root->root_list);
1175         INIT_LIST_HEAD(&root->delalloc_inodes);
1176         INIT_LIST_HEAD(&root->delalloc_root);
1177         INIT_LIST_HEAD(&root->ordered_extents);
1178         INIT_LIST_HEAD(&root->ordered_root);
1179         INIT_LIST_HEAD(&root->logged_list[0]);
1180         INIT_LIST_HEAD(&root->logged_list[1]);
1181         spin_lock_init(&root->inode_lock);
1182         spin_lock_init(&root->delalloc_lock);
1183         spin_lock_init(&root->ordered_extent_lock);
1184         spin_lock_init(&root->accounting_lock);
1185         spin_lock_init(&root->log_extents_lock[0]);
1186         spin_lock_init(&root->log_extents_lock[1]);
1187         spin_lock_init(&root->qgroup_meta_rsv_lock);
1188         mutex_init(&root->objectid_mutex);
1189         mutex_init(&root->log_mutex);
1190         mutex_init(&root->ordered_extent_mutex);
1191         mutex_init(&root->delalloc_mutex);
1192         init_waitqueue_head(&root->log_writer_wait);
1193         init_waitqueue_head(&root->log_commit_wait[0]);
1194         init_waitqueue_head(&root->log_commit_wait[1]);
1195         INIT_LIST_HEAD(&root->log_ctxs[0]);
1196         INIT_LIST_HEAD(&root->log_ctxs[1]);
1197         atomic_set(&root->log_commit[0], 0);
1198         atomic_set(&root->log_commit[1], 0);
1199         atomic_set(&root->log_writers, 0);
1200         atomic_set(&root->log_batch, 0);
1201         refcount_set(&root->refs, 1);
1202         atomic_set(&root->will_be_snapshotted, 0);
1203         atomic_set(&root->snapshot_force_cow, 0);
1204         atomic_set(&root->nr_swapfiles, 0);
1205         root->log_transid = 0;
1206         root->log_transid_committed = -1;
1207         root->last_log_commit = 0;
1208         if (!dummy)
1209                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1210
1211         memset(&root->root_key, 0, sizeof(root->root_key));
1212         memset(&root->root_item, 0, sizeof(root->root_item));
1213         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1214         if (!dummy)
1215                 root->defrag_trans_start = fs_info->generation;
1216         else
1217                 root->defrag_trans_start = 0;
1218         root->root_key.objectid = objectid;
1219         root->anon_dev = 0;
1220
1221         spin_lock_init(&root->root_item_lock);
1222 }
1223
1224 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1225                 gfp_t flags)
1226 {
1227         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1228         if (root)
1229                 root->fs_info = fs_info;
1230         return root;
1231 }
1232
1233 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1234 /* Should only be used by the testing infrastructure */
1235 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1236 {
1237         struct btrfs_root *root;
1238
1239         if (!fs_info)
1240                 return ERR_PTR(-EINVAL);
1241
1242         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1243         if (!root)
1244                 return ERR_PTR(-ENOMEM);
1245
1246         /* We don't use the stripesize in selftest, set it as sectorsize */
1247         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1248         root->alloc_bytenr = 0;
1249
1250         return root;
1251 }
1252 #endif
1253
1254 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1255                                      struct btrfs_fs_info *fs_info,
1256                                      u64 objectid)
1257 {
1258         struct extent_buffer *leaf;
1259         struct btrfs_root *tree_root = fs_info->tree_root;
1260         struct btrfs_root *root;
1261         struct btrfs_key key;
1262         unsigned int nofs_flag;
1263         int ret = 0;
1264         uuid_le uuid = NULL_UUID_LE;
1265
1266         /*
1267          * We're holding a transaction handle, so use a NOFS memory allocation
1268          * context to avoid deadlock if reclaim happens.
1269          */
1270         nofs_flag = memalloc_nofs_save();
1271         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1272         memalloc_nofs_restore(nofs_flag);
1273         if (!root)
1274                 return ERR_PTR(-ENOMEM);
1275
1276         __setup_root(root, fs_info, objectid);
1277         root->root_key.objectid = objectid;
1278         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1279         root->root_key.offset = 0;
1280
1281         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1282         if (IS_ERR(leaf)) {
1283                 ret = PTR_ERR(leaf);
1284                 leaf = NULL;
1285                 goto fail;
1286         }
1287
1288         root->node = leaf;
1289         btrfs_mark_buffer_dirty(leaf);
1290
1291         root->commit_root = btrfs_root_node(root);
1292         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1293
1294         root->root_item.flags = 0;
1295         root->root_item.byte_limit = 0;
1296         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1297         btrfs_set_root_generation(&root->root_item, trans->transid);
1298         btrfs_set_root_level(&root->root_item, 0);
1299         btrfs_set_root_refs(&root->root_item, 1);
1300         btrfs_set_root_used(&root->root_item, leaf->len);
1301         btrfs_set_root_last_snapshot(&root->root_item, 0);
1302         btrfs_set_root_dirid(&root->root_item, 0);
1303         if (is_fstree(objectid))
1304                 uuid_le_gen(&uuid);
1305         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1306         root->root_item.drop_level = 0;
1307
1308         key.objectid = objectid;
1309         key.type = BTRFS_ROOT_ITEM_KEY;
1310         key.offset = 0;
1311         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1312         if (ret)
1313                 goto fail;
1314
1315         btrfs_tree_unlock(leaf);
1316
1317         return root;
1318
1319 fail:
1320         if (leaf) {
1321                 btrfs_tree_unlock(leaf);
1322                 free_extent_buffer(root->commit_root);
1323                 free_extent_buffer(leaf);
1324         }
1325         kfree(root);
1326
1327         return ERR_PTR(ret);
1328 }
1329
1330 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1331                                          struct btrfs_fs_info *fs_info)
1332 {
1333         struct btrfs_root *root;
1334         struct extent_buffer *leaf;
1335
1336         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1337         if (!root)
1338                 return ERR_PTR(-ENOMEM);
1339
1340         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1341
1342         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1343         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1344         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1345
1346         /*
1347          * DON'T set REF_COWS for log trees
1348          *
1349          * log trees do not get reference counted because they go away
1350          * before a real commit is actually done.  They do store pointers
1351          * to file data extents, and those reference counts still get
1352          * updated (along with back refs to the log tree).
1353          */
1354
1355         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1356                         NULL, 0, 0, 0);
1357         if (IS_ERR(leaf)) {
1358                 kfree(root);
1359                 return ERR_CAST(leaf);
1360         }
1361
1362         root->node = leaf;
1363
1364         btrfs_mark_buffer_dirty(root->node);
1365         btrfs_tree_unlock(root->node);
1366         return root;
1367 }
1368
1369 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1370                              struct btrfs_fs_info *fs_info)
1371 {
1372         struct btrfs_root *log_root;
1373
1374         log_root = alloc_log_tree(trans, fs_info);
1375         if (IS_ERR(log_root))
1376                 return PTR_ERR(log_root);
1377         WARN_ON(fs_info->log_root_tree);
1378         fs_info->log_root_tree = log_root;
1379         return 0;
1380 }
1381
1382 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1383                        struct btrfs_root *root)
1384 {
1385         struct btrfs_fs_info *fs_info = root->fs_info;
1386         struct btrfs_root *log_root;
1387         struct btrfs_inode_item *inode_item;
1388
1389         log_root = alloc_log_tree(trans, fs_info);
1390         if (IS_ERR(log_root))
1391                 return PTR_ERR(log_root);
1392
1393         log_root->last_trans = trans->transid;
1394         log_root->root_key.offset = root->root_key.objectid;
1395
1396         inode_item = &log_root->root_item.inode;
1397         btrfs_set_stack_inode_generation(inode_item, 1);
1398         btrfs_set_stack_inode_size(inode_item, 3);
1399         btrfs_set_stack_inode_nlink(inode_item, 1);
1400         btrfs_set_stack_inode_nbytes(inode_item,
1401                                      fs_info->nodesize);
1402         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1403
1404         btrfs_set_root_node(&log_root->root_item, log_root->node);
1405
1406         WARN_ON(root->log_root);
1407         root->log_root = log_root;
1408         root->log_transid = 0;
1409         root->log_transid_committed = -1;
1410         root->last_log_commit = 0;
1411         return 0;
1412 }
1413
1414 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1415                                                struct btrfs_key *key)
1416 {
1417         struct btrfs_root *root;
1418         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1419         struct btrfs_path *path;
1420         u64 generation;
1421         int ret;
1422         int level;
1423
1424         path = btrfs_alloc_path();
1425         if (!path)
1426                 return ERR_PTR(-ENOMEM);
1427
1428         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1429         if (!root) {
1430                 ret = -ENOMEM;
1431                 goto alloc_fail;
1432         }
1433
1434         __setup_root(root, fs_info, key->objectid);
1435
1436         ret = btrfs_find_root(tree_root, key, path,
1437                               &root->root_item, &root->root_key);
1438         if (ret) {
1439                 if (ret > 0)
1440                         ret = -ENOENT;
1441                 goto find_fail;
1442         }
1443
1444         generation = btrfs_root_generation(&root->root_item);
1445         level = btrfs_root_level(&root->root_item);
1446         root->node = read_tree_block(fs_info,
1447                                      btrfs_root_bytenr(&root->root_item),
1448                                      generation, level, NULL);
1449         if (IS_ERR(root->node)) {
1450                 ret = PTR_ERR(root->node);
1451                 goto find_fail;
1452         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1453                 ret = -EIO;
1454                 free_extent_buffer(root->node);
1455                 goto find_fail;
1456         }
1457         root->commit_root = btrfs_root_node(root);
1458 out:
1459         btrfs_free_path(path);
1460         return root;
1461
1462 find_fail:
1463         kfree(root);
1464 alloc_fail:
1465         root = ERR_PTR(ret);
1466         goto out;
1467 }
1468
1469 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1470                                       struct btrfs_key *location)
1471 {
1472         struct btrfs_root *root;
1473
1474         root = btrfs_read_tree_root(tree_root, location);
1475         if (IS_ERR(root))
1476                 return root;
1477
1478         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1479                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1480                 btrfs_check_and_init_root_item(&root->root_item);
1481         }
1482
1483         return root;
1484 }
1485
1486 int btrfs_init_fs_root(struct btrfs_root *root)
1487 {
1488         int ret;
1489         struct btrfs_subvolume_writers *writers;
1490
1491         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1492         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1493                                         GFP_NOFS);
1494         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1495                 ret = -ENOMEM;
1496                 goto fail;
1497         }
1498
1499         writers = btrfs_alloc_subvolume_writers();
1500         if (IS_ERR(writers)) {
1501                 ret = PTR_ERR(writers);
1502                 goto fail;
1503         }
1504         root->subv_writers = writers;
1505
1506         btrfs_init_free_ino_ctl(root);
1507         spin_lock_init(&root->ino_cache_lock);
1508         init_waitqueue_head(&root->ino_cache_wait);
1509
1510         ret = get_anon_bdev(&root->anon_dev);
1511         if (ret)
1512                 goto fail;
1513
1514         mutex_lock(&root->objectid_mutex);
1515         ret = btrfs_find_highest_objectid(root,
1516                                         &root->highest_objectid);
1517         if (ret) {
1518                 mutex_unlock(&root->objectid_mutex);
1519                 goto fail;
1520         }
1521
1522         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1523
1524         mutex_unlock(&root->objectid_mutex);
1525
1526         return 0;
1527 fail:
1528         /* The caller is responsible to call btrfs_free_fs_root */
1529         return ret;
1530 }
1531
1532 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1533                                         u64 root_id)
1534 {
1535         struct btrfs_root *root;
1536
1537         spin_lock(&fs_info->fs_roots_radix_lock);
1538         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1539                                  (unsigned long)root_id);
1540         spin_unlock(&fs_info->fs_roots_radix_lock);
1541         return root;
1542 }
1543
1544 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1545                          struct btrfs_root *root)
1546 {
1547         int ret;
1548
1549         ret = radix_tree_preload(GFP_NOFS);
1550         if (ret)
1551                 return ret;
1552
1553         spin_lock(&fs_info->fs_roots_radix_lock);
1554         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1555                                 (unsigned long)root->root_key.objectid,
1556                                 root);
1557         if (ret == 0)
1558                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1559         spin_unlock(&fs_info->fs_roots_radix_lock);
1560         radix_tree_preload_end();
1561
1562         return ret;
1563 }
1564
1565 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1566                                      struct btrfs_key *location,
1567                                      bool check_ref)
1568 {
1569         struct btrfs_root *root;
1570         struct btrfs_path *path;
1571         struct btrfs_key key;
1572         int ret;
1573
1574         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1575                 return fs_info->tree_root;
1576         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1577                 return fs_info->extent_root;
1578         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1579                 return fs_info->chunk_root;
1580         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1581                 return fs_info->dev_root;
1582         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1583                 return fs_info->csum_root;
1584         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1585                 return fs_info->quota_root ? fs_info->quota_root :
1586                                              ERR_PTR(-ENOENT);
1587         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1588                 return fs_info->uuid_root ? fs_info->uuid_root :
1589                                             ERR_PTR(-ENOENT);
1590         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1591                 return fs_info->free_space_root ? fs_info->free_space_root :
1592                                                   ERR_PTR(-ENOENT);
1593 again:
1594         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1595         if (root) {
1596                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1597                         return ERR_PTR(-ENOENT);
1598                 return root;
1599         }
1600
1601         root = btrfs_read_fs_root(fs_info->tree_root, location);
1602         if (IS_ERR(root))
1603                 return root;
1604
1605         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1606                 ret = -ENOENT;
1607                 goto fail;
1608         }
1609
1610         ret = btrfs_init_fs_root(root);
1611         if (ret)
1612                 goto fail;
1613
1614         path = btrfs_alloc_path();
1615         if (!path) {
1616                 ret = -ENOMEM;
1617                 goto fail;
1618         }
1619         key.objectid = BTRFS_ORPHAN_OBJECTID;
1620         key.type = BTRFS_ORPHAN_ITEM_KEY;
1621         key.offset = location->objectid;
1622
1623         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1624         btrfs_free_path(path);
1625         if (ret < 0)
1626                 goto fail;
1627         if (ret == 0)
1628                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1629
1630         ret = btrfs_insert_fs_root(fs_info, root);
1631         if (ret) {
1632                 if (ret == -EEXIST) {
1633                         btrfs_free_fs_root(root);
1634                         goto again;
1635                 }
1636                 goto fail;
1637         }
1638         return root;
1639 fail:
1640         btrfs_free_fs_root(root);
1641         return ERR_PTR(ret);
1642 }
1643
1644 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1645 {
1646         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1647         int ret = 0;
1648         struct btrfs_device *device;
1649         struct backing_dev_info *bdi;
1650
1651         rcu_read_lock();
1652         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1653                 if (!device->bdev)
1654                         continue;
1655                 bdi = device->bdev->bd_bdi;
1656                 if (bdi_congested(bdi, bdi_bits)) {
1657                         ret = 1;
1658                         break;
1659                 }
1660         }
1661         rcu_read_unlock();
1662         return ret;
1663 }
1664
1665 /*
1666  * called by the kthread helper functions to finally call the bio end_io
1667  * functions.  This is where read checksum verification actually happens
1668  */
1669 static void end_workqueue_fn(struct btrfs_work *work)
1670 {
1671         struct bio *bio;
1672         struct btrfs_end_io_wq *end_io_wq;
1673
1674         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1675         bio = end_io_wq->bio;
1676
1677         bio->bi_status = end_io_wq->status;
1678         bio->bi_private = end_io_wq->private;
1679         bio->bi_end_io = end_io_wq->end_io;
1680         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1681         bio_endio(bio);
1682 }
1683
1684 static int cleaner_kthread(void *arg)
1685 {
1686         struct btrfs_root *root = arg;
1687         struct btrfs_fs_info *fs_info = root->fs_info;
1688         int again;
1689
1690         while (1) {
1691                 again = 0;
1692
1693                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1694
1695                 /* Make the cleaner go to sleep early. */
1696                 if (btrfs_need_cleaner_sleep(fs_info))
1697                         goto sleep;
1698
1699                 /*
1700                  * Do not do anything if we might cause open_ctree() to block
1701                  * before we have finished mounting the filesystem.
1702                  */
1703                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1704                         goto sleep;
1705
1706                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1707                         goto sleep;
1708
1709                 /*
1710                  * Avoid the problem that we change the status of the fs
1711                  * during the above check and trylock.
1712                  */
1713                 if (btrfs_need_cleaner_sleep(fs_info)) {
1714                         mutex_unlock(&fs_info->cleaner_mutex);
1715                         goto sleep;
1716                 }
1717
1718                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1719                 btrfs_run_delayed_iputs(fs_info);
1720                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1721
1722                 again = btrfs_clean_one_deleted_snapshot(root);
1723                 mutex_unlock(&fs_info->cleaner_mutex);
1724
1725                 /*
1726                  * The defragger has dealt with the R/O remount and umount,
1727                  * needn't do anything special here.
1728                  */
1729                 btrfs_run_defrag_inodes(fs_info);
1730
1731                 /*
1732                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1733                  * with relocation (btrfs_relocate_chunk) and relocation
1734                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1735                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1736                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1737                  * unused block groups.
1738                  */
1739                 btrfs_delete_unused_bgs(fs_info);
1740 sleep:
1741                 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1742                 if (kthread_should_park())
1743                         kthread_parkme();
1744                 if (kthread_should_stop())
1745                         return 0;
1746                 if (!again) {
1747                         set_current_state(TASK_INTERRUPTIBLE);
1748                         schedule();
1749                         __set_current_state(TASK_RUNNING);
1750                 }
1751         }
1752 }
1753
1754 static int transaction_kthread(void *arg)
1755 {
1756         struct btrfs_root *root = arg;
1757         struct btrfs_fs_info *fs_info = root->fs_info;
1758         struct btrfs_trans_handle *trans;
1759         struct btrfs_transaction *cur;
1760         u64 transid;
1761         time64_t now;
1762         unsigned long delay;
1763         bool cannot_commit;
1764
1765         do {
1766                 cannot_commit = false;
1767                 delay = HZ * fs_info->commit_interval;
1768                 mutex_lock(&fs_info->transaction_kthread_mutex);
1769
1770                 spin_lock(&fs_info->trans_lock);
1771                 cur = fs_info->running_transaction;
1772                 if (!cur) {
1773                         spin_unlock(&fs_info->trans_lock);
1774                         goto sleep;
1775                 }
1776
1777                 now = ktime_get_seconds();
1778                 if (cur->state < TRANS_STATE_BLOCKED &&
1779                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1780                     (now < cur->start_time ||
1781                      now - cur->start_time < fs_info->commit_interval)) {
1782                         spin_unlock(&fs_info->trans_lock);
1783                         delay = HZ * 5;
1784                         goto sleep;
1785                 }
1786                 transid = cur->transid;
1787                 spin_unlock(&fs_info->trans_lock);
1788
1789                 /* If the file system is aborted, this will always fail. */
1790                 trans = btrfs_attach_transaction(root);
1791                 if (IS_ERR(trans)) {
1792                         if (PTR_ERR(trans) != -ENOENT)
1793                                 cannot_commit = true;
1794                         goto sleep;
1795                 }
1796                 if (transid == trans->transid) {
1797                         btrfs_commit_transaction(trans);
1798                 } else {
1799                         btrfs_end_transaction(trans);
1800                 }
1801 sleep:
1802                 wake_up_process(fs_info->cleaner_kthread);
1803                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1804
1805                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1806                                       &fs_info->fs_state)))
1807                         btrfs_cleanup_transaction(fs_info);
1808                 if (!kthread_should_stop() &&
1809                                 (!btrfs_transaction_blocked(fs_info) ||
1810                                  cannot_commit))
1811                         schedule_timeout_interruptible(delay);
1812         } while (!kthread_should_stop());
1813         return 0;
1814 }
1815
1816 /*
1817  * this will find the highest generation in the array of
1818  * root backups.  The index of the highest array is returned,
1819  * or -1 if we can't find anything.
1820  *
1821  * We check to make sure the array is valid by comparing the
1822  * generation of the latest  root in the array with the generation
1823  * in the super block.  If they don't match we pitch it.
1824  */
1825 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1826 {
1827         u64 cur;
1828         int newest_index = -1;
1829         struct btrfs_root_backup *root_backup;
1830         int i;
1831
1832         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1833                 root_backup = info->super_copy->super_roots + i;
1834                 cur = btrfs_backup_tree_root_gen(root_backup);
1835                 if (cur == newest_gen)
1836                         newest_index = i;
1837         }
1838
1839         /* check to see if we actually wrapped around */
1840         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1841                 root_backup = info->super_copy->super_roots;
1842                 cur = btrfs_backup_tree_root_gen(root_backup);
1843                 if (cur == newest_gen)
1844                         newest_index = 0;
1845         }
1846         return newest_index;
1847 }
1848
1849
1850 /*
1851  * find the oldest backup so we know where to store new entries
1852  * in the backup array.  This will set the backup_root_index
1853  * field in the fs_info struct
1854  */
1855 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1856                                      u64 newest_gen)
1857 {
1858         int newest_index = -1;
1859
1860         newest_index = find_newest_super_backup(info, newest_gen);
1861         /* if there was garbage in there, just move along */
1862         if (newest_index == -1) {
1863                 info->backup_root_index = 0;
1864         } else {
1865                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1866         }
1867 }
1868
1869 /*
1870  * copy all the root pointers into the super backup array.
1871  * this will bump the backup pointer by one when it is
1872  * done
1873  */
1874 static void backup_super_roots(struct btrfs_fs_info *info)
1875 {
1876         int next_backup;
1877         struct btrfs_root_backup *root_backup;
1878         int last_backup;
1879
1880         next_backup = info->backup_root_index;
1881         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1882                 BTRFS_NUM_BACKUP_ROOTS;
1883
1884         /*
1885          * just overwrite the last backup if we're at the same generation
1886          * this happens only at umount
1887          */
1888         root_backup = info->super_for_commit->super_roots + last_backup;
1889         if (btrfs_backup_tree_root_gen(root_backup) ==
1890             btrfs_header_generation(info->tree_root->node))
1891                 next_backup = last_backup;
1892
1893         root_backup = info->super_for_commit->super_roots + next_backup;
1894
1895         /*
1896          * make sure all of our padding and empty slots get zero filled
1897          * regardless of which ones we use today
1898          */
1899         memset(root_backup, 0, sizeof(*root_backup));
1900
1901         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1902
1903         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1904         btrfs_set_backup_tree_root_gen(root_backup,
1905                                btrfs_header_generation(info->tree_root->node));
1906
1907         btrfs_set_backup_tree_root_level(root_backup,
1908                                btrfs_header_level(info->tree_root->node));
1909
1910         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1911         btrfs_set_backup_chunk_root_gen(root_backup,
1912                                btrfs_header_generation(info->chunk_root->node));
1913         btrfs_set_backup_chunk_root_level(root_backup,
1914                                btrfs_header_level(info->chunk_root->node));
1915
1916         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1917         btrfs_set_backup_extent_root_gen(root_backup,
1918                                btrfs_header_generation(info->extent_root->node));
1919         btrfs_set_backup_extent_root_level(root_backup,
1920                                btrfs_header_level(info->extent_root->node));
1921
1922         /*
1923          * we might commit during log recovery, which happens before we set
1924          * the fs_root.  Make sure it is valid before we fill it in.
1925          */
1926         if (info->fs_root && info->fs_root->node) {
1927                 btrfs_set_backup_fs_root(root_backup,
1928                                          info->fs_root->node->start);
1929                 btrfs_set_backup_fs_root_gen(root_backup,
1930                                btrfs_header_generation(info->fs_root->node));
1931                 btrfs_set_backup_fs_root_level(root_backup,
1932                                btrfs_header_level(info->fs_root->node));
1933         }
1934
1935         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1936         btrfs_set_backup_dev_root_gen(root_backup,
1937                                btrfs_header_generation(info->dev_root->node));
1938         btrfs_set_backup_dev_root_level(root_backup,
1939                                        btrfs_header_level(info->dev_root->node));
1940
1941         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1942         btrfs_set_backup_csum_root_gen(root_backup,
1943                                btrfs_header_generation(info->csum_root->node));
1944         btrfs_set_backup_csum_root_level(root_backup,
1945                                btrfs_header_level(info->csum_root->node));
1946
1947         btrfs_set_backup_total_bytes(root_backup,
1948                              btrfs_super_total_bytes(info->super_copy));
1949         btrfs_set_backup_bytes_used(root_backup,
1950                              btrfs_super_bytes_used(info->super_copy));
1951         btrfs_set_backup_num_devices(root_backup,
1952                              btrfs_super_num_devices(info->super_copy));
1953
1954         /*
1955          * if we don't copy this out to the super_copy, it won't get remembered
1956          * for the next commit
1957          */
1958         memcpy(&info->super_copy->super_roots,
1959                &info->super_for_commit->super_roots,
1960                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1961 }
1962
1963 /*
1964  * this copies info out of the root backup array and back into
1965  * the in-memory super block.  It is meant to help iterate through
1966  * the array, so you send it the number of backups you've already
1967  * tried and the last backup index you used.
1968  *
1969  * this returns -1 when it has tried all the backups
1970  */
1971 static noinline int next_root_backup(struct btrfs_fs_info *info,
1972                                      struct btrfs_super_block *super,
1973                                      int *num_backups_tried, int *backup_index)
1974 {
1975         struct btrfs_root_backup *root_backup;
1976         int newest = *backup_index;
1977
1978         if (*num_backups_tried == 0) {
1979                 u64 gen = btrfs_super_generation(super);
1980
1981                 newest = find_newest_super_backup(info, gen);
1982                 if (newest == -1)
1983                         return -1;
1984
1985                 *backup_index = newest;
1986                 *num_backups_tried = 1;
1987         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1988                 /* we've tried all the backups, all done */
1989                 return -1;
1990         } else {
1991                 /* jump to the next oldest backup */
1992                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1993                         BTRFS_NUM_BACKUP_ROOTS;
1994                 *backup_index = newest;
1995                 *num_backups_tried += 1;
1996         }
1997         root_backup = super->super_roots + newest;
1998
1999         btrfs_set_super_generation(super,
2000                                    btrfs_backup_tree_root_gen(root_backup));
2001         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2002         btrfs_set_super_root_level(super,
2003                                    btrfs_backup_tree_root_level(root_backup));
2004         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2005
2006         /*
2007          * fixme: the total bytes and num_devices need to match or we should
2008          * need a fsck
2009          */
2010         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2011         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2012         return 0;
2013 }
2014
2015 /* helper to cleanup workers */
2016 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2017 {
2018         btrfs_destroy_workqueue(fs_info->fixup_workers);
2019         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2020         btrfs_destroy_workqueue(fs_info->workers);
2021         btrfs_destroy_workqueue(fs_info->endio_workers);
2022         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2023         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2024         btrfs_destroy_workqueue(fs_info->rmw_workers);
2025         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2026         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2027         btrfs_destroy_workqueue(fs_info->submit_workers);
2028         btrfs_destroy_workqueue(fs_info->delayed_workers);
2029         btrfs_destroy_workqueue(fs_info->caching_workers);
2030         btrfs_destroy_workqueue(fs_info->readahead_workers);
2031         btrfs_destroy_workqueue(fs_info->flush_workers);
2032         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2033         btrfs_destroy_workqueue(fs_info->extent_workers);
2034         /*
2035          * Now that all other work queues are destroyed, we can safely destroy
2036          * the queues used for metadata I/O, since tasks from those other work
2037          * queues can do metadata I/O operations.
2038          */
2039         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2040         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2041 }
2042
2043 static void free_root_extent_buffers(struct btrfs_root *root)
2044 {
2045         if (root) {
2046                 free_extent_buffer(root->node);
2047                 free_extent_buffer(root->commit_root);
2048                 root->node = NULL;
2049                 root->commit_root = NULL;
2050         }
2051 }
2052
2053 /* helper to cleanup tree roots */
2054 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2055 {
2056         free_root_extent_buffers(info->tree_root);
2057
2058         free_root_extent_buffers(info->dev_root);
2059         free_root_extent_buffers(info->extent_root);
2060         free_root_extent_buffers(info->csum_root);
2061         free_root_extent_buffers(info->quota_root);
2062         free_root_extent_buffers(info->uuid_root);
2063         if (chunk_root)
2064                 free_root_extent_buffers(info->chunk_root);
2065         free_root_extent_buffers(info->free_space_root);
2066 }
2067
2068 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2069 {
2070         int ret;
2071         struct btrfs_root *gang[8];
2072         int i;
2073
2074         while (!list_empty(&fs_info->dead_roots)) {
2075                 gang[0] = list_entry(fs_info->dead_roots.next,
2076                                      struct btrfs_root, root_list);
2077                 list_del(&gang[0]->root_list);
2078
2079                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2080                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2081                 } else {
2082                         free_extent_buffer(gang[0]->node);
2083                         free_extent_buffer(gang[0]->commit_root);
2084                         btrfs_put_fs_root(gang[0]);
2085                 }
2086         }
2087
2088         while (1) {
2089                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2090                                              (void **)gang, 0,
2091                                              ARRAY_SIZE(gang));
2092                 if (!ret)
2093                         break;
2094                 for (i = 0; i < ret; i++)
2095                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2096         }
2097
2098         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2099                 btrfs_free_log_root_tree(NULL, fs_info);
2100                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2101         }
2102 }
2103
2104 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2105 {
2106         mutex_init(&fs_info->scrub_lock);
2107         atomic_set(&fs_info->scrubs_running, 0);
2108         atomic_set(&fs_info->scrub_pause_req, 0);
2109         atomic_set(&fs_info->scrubs_paused, 0);
2110         atomic_set(&fs_info->scrub_cancel_req, 0);
2111         init_waitqueue_head(&fs_info->scrub_pause_wait);
2112         fs_info->scrub_workers_refcnt = 0;
2113 }
2114
2115 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2116 {
2117         spin_lock_init(&fs_info->balance_lock);
2118         mutex_init(&fs_info->balance_mutex);
2119         atomic_set(&fs_info->balance_pause_req, 0);
2120         atomic_set(&fs_info->balance_cancel_req, 0);
2121         fs_info->balance_ctl = NULL;
2122         init_waitqueue_head(&fs_info->balance_wait_q);
2123 }
2124
2125 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2126 {
2127         struct inode *inode = fs_info->btree_inode;
2128
2129         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2130         set_nlink(inode, 1);
2131         /*
2132          * we set the i_size on the btree inode to the max possible int.
2133          * the real end of the address space is determined by all of
2134          * the devices in the system
2135          */
2136         inode->i_size = OFFSET_MAX;
2137         inode->i_mapping->a_ops = &btree_aops;
2138
2139         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2140         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2141         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2142         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2143
2144         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2145
2146         BTRFS_I(inode)->root = fs_info->tree_root;
2147         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2148         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2149         btrfs_insert_inode_hash(inode);
2150 }
2151
2152 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2153 {
2154         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2155         init_rwsem(&fs_info->dev_replace.rwsem);
2156         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2157 }
2158
2159 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2160 {
2161         spin_lock_init(&fs_info->qgroup_lock);
2162         mutex_init(&fs_info->qgroup_ioctl_lock);
2163         fs_info->qgroup_tree = RB_ROOT;
2164         fs_info->qgroup_op_tree = RB_ROOT;
2165         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2166         fs_info->qgroup_seq = 1;
2167         fs_info->qgroup_ulist = NULL;
2168         fs_info->qgroup_rescan_running = false;
2169         mutex_init(&fs_info->qgroup_rescan_lock);
2170 }
2171
2172 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2173                 struct btrfs_fs_devices *fs_devices)
2174 {
2175         u32 max_active = fs_info->thread_pool_size;
2176         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2177
2178         fs_info->workers =
2179                 btrfs_alloc_workqueue(fs_info, "worker",
2180                                       flags | WQ_HIGHPRI, max_active, 16);
2181
2182         fs_info->delalloc_workers =
2183                 btrfs_alloc_workqueue(fs_info, "delalloc",
2184                                       flags, max_active, 2);
2185
2186         fs_info->flush_workers =
2187                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2188                                       flags, max_active, 0);
2189
2190         fs_info->caching_workers =
2191                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2192
2193         /*
2194          * a higher idle thresh on the submit workers makes it much more
2195          * likely that bios will be send down in a sane order to the
2196          * devices
2197          */
2198         fs_info->submit_workers =
2199                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2200                                       min_t(u64, fs_devices->num_devices,
2201                                             max_active), 64);
2202
2203         fs_info->fixup_workers =
2204                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2205
2206         /*
2207          * endios are largely parallel and should have a very
2208          * low idle thresh
2209          */
2210         fs_info->endio_workers =
2211                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2212         fs_info->endio_meta_workers =
2213                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2214                                       max_active, 4);
2215         fs_info->endio_meta_write_workers =
2216                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2217                                       max_active, 2);
2218         fs_info->endio_raid56_workers =
2219                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2220                                       max_active, 4);
2221         fs_info->endio_repair_workers =
2222                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2223         fs_info->rmw_workers =
2224                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2225         fs_info->endio_write_workers =
2226                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2227                                       max_active, 2);
2228         fs_info->endio_freespace_worker =
2229                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2230                                       max_active, 0);
2231         fs_info->delayed_workers =
2232                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2233                                       max_active, 0);
2234         fs_info->readahead_workers =
2235                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2236                                       max_active, 2);
2237         fs_info->qgroup_rescan_workers =
2238                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2239         fs_info->extent_workers =
2240                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2241                                       min_t(u64, fs_devices->num_devices,
2242                                             max_active), 8);
2243
2244         if (!(fs_info->workers && fs_info->delalloc_workers &&
2245               fs_info->submit_workers && fs_info->flush_workers &&
2246               fs_info->endio_workers && fs_info->endio_meta_workers &&
2247               fs_info->endio_meta_write_workers &&
2248               fs_info->endio_repair_workers &&
2249               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2250               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2251               fs_info->caching_workers && fs_info->readahead_workers &&
2252               fs_info->fixup_workers && fs_info->delayed_workers &&
2253               fs_info->extent_workers &&
2254               fs_info->qgroup_rescan_workers)) {
2255                 return -ENOMEM;
2256         }
2257
2258         return 0;
2259 }
2260
2261 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2262                             struct btrfs_fs_devices *fs_devices)
2263 {
2264         int ret;
2265         struct btrfs_root *log_tree_root;
2266         struct btrfs_super_block *disk_super = fs_info->super_copy;
2267         u64 bytenr = btrfs_super_log_root(disk_super);
2268         int level = btrfs_super_log_root_level(disk_super);
2269
2270         if (fs_devices->rw_devices == 0) {
2271                 btrfs_warn(fs_info, "log replay required on RO media");
2272                 return -EIO;
2273         }
2274
2275         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2276         if (!log_tree_root)
2277                 return -ENOMEM;
2278
2279         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2280
2281         log_tree_root->node = read_tree_block(fs_info, bytenr,
2282                                               fs_info->generation + 1,
2283                                               level, NULL);
2284         if (IS_ERR(log_tree_root->node)) {
2285                 btrfs_warn(fs_info, "failed to read log tree");
2286                 ret = PTR_ERR(log_tree_root->node);
2287                 kfree(log_tree_root);
2288                 return ret;
2289         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2290                 btrfs_err(fs_info, "failed to read log tree");
2291                 free_extent_buffer(log_tree_root->node);
2292                 kfree(log_tree_root);
2293                 return -EIO;
2294         }
2295         /* returns with log_tree_root freed on success */
2296         ret = btrfs_recover_log_trees(log_tree_root);
2297         if (ret) {
2298                 btrfs_handle_fs_error(fs_info, ret,
2299                                       "Failed to recover log tree");
2300                 free_extent_buffer(log_tree_root->node);
2301                 kfree(log_tree_root);
2302                 return ret;
2303         }
2304
2305         if (sb_rdonly(fs_info->sb)) {
2306                 ret = btrfs_commit_super(fs_info);
2307                 if (ret)
2308                         return ret;
2309         }
2310
2311         return 0;
2312 }
2313
2314 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2315 {
2316         struct btrfs_root *tree_root = fs_info->tree_root;
2317         struct btrfs_root *root;
2318         struct btrfs_key location;
2319         int ret;
2320
2321         BUG_ON(!fs_info->tree_root);
2322
2323         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2324         location.type = BTRFS_ROOT_ITEM_KEY;
2325         location.offset = 0;
2326
2327         root = btrfs_read_tree_root(tree_root, &location);
2328         if (IS_ERR(root)) {
2329                 ret = PTR_ERR(root);
2330                 goto out;
2331         }
2332         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2333         fs_info->extent_root = root;
2334
2335         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2336         root = btrfs_read_tree_root(tree_root, &location);
2337         if (IS_ERR(root)) {
2338                 ret = PTR_ERR(root);
2339                 goto out;
2340         }
2341         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2342         fs_info->dev_root = root;
2343         btrfs_init_devices_late(fs_info);
2344
2345         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2346         root = btrfs_read_tree_root(tree_root, &location);
2347         if (IS_ERR(root)) {
2348                 ret = PTR_ERR(root);
2349                 goto out;
2350         }
2351         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2352         fs_info->csum_root = root;
2353
2354         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2355         root = btrfs_read_tree_root(tree_root, &location);
2356         if (!IS_ERR(root)) {
2357                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2358                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2359                 fs_info->quota_root = root;
2360         }
2361
2362         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2363         root = btrfs_read_tree_root(tree_root, &location);
2364         if (IS_ERR(root)) {
2365                 ret = PTR_ERR(root);
2366                 if (ret != -ENOENT)
2367                         goto out;
2368         } else {
2369                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2370                 fs_info->uuid_root = root;
2371         }
2372
2373         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2374                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2375                 root = btrfs_read_tree_root(tree_root, &location);
2376                 if (IS_ERR(root)) {
2377                         ret = PTR_ERR(root);
2378                         goto out;
2379                 }
2380                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2381                 fs_info->free_space_root = root;
2382         }
2383
2384         return 0;
2385 out:
2386         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2387                    location.objectid, ret);
2388         return ret;
2389 }
2390
2391 /*
2392  * Real super block validation
2393  * NOTE: super csum type and incompat features will not be checked here.
2394  *
2395  * @sb:         super block to check
2396  * @mirror_num: the super block number to check its bytenr:
2397  *              0       the primary (1st) sb
2398  *              1, 2    2nd and 3rd backup copy
2399  *             -1       skip bytenr check
2400  */
2401 static int validate_super(struct btrfs_fs_info *fs_info,
2402                             struct btrfs_super_block *sb, int mirror_num)
2403 {
2404         u64 nodesize = btrfs_super_nodesize(sb);
2405         u64 sectorsize = btrfs_super_sectorsize(sb);
2406         int ret = 0;
2407
2408         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2409                 btrfs_err(fs_info, "no valid FS found");
2410                 ret = -EINVAL;
2411         }
2412         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2413                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2414                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2415                 ret = -EINVAL;
2416         }
2417         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2418                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2419                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2420                 ret = -EINVAL;
2421         }
2422         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2423                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2424                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2425                 ret = -EINVAL;
2426         }
2427         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2428                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2429                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2430                 ret = -EINVAL;
2431         }
2432
2433         /*
2434          * Check sectorsize and nodesize first, other check will need it.
2435          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2436          */
2437         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2438             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2439                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2440                 ret = -EINVAL;
2441         }
2442         /* Only PAGE SIZE is supported yet */
2443         if (sectorsize != PAGE_SIZE) {
2444                 btrfs_err(fs_info,
2445                         "sectorsize %llu not supported yet, only support %lu",
2446                         sectorsize, PAGE_SIZE);
2447                 ret = -EINVAL;
2448         }
2449         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2450             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2451                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2452                 ret = -EINVAL;
2453         }
2454         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2455                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2456                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2457                 ret = -EINVAL;
2458         }
2459
2460         /* Root alignment check */
2461         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2462                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2463                            btrfs_super_root(sb));
2464                 ret = -EINVAL;
2465         }
2466         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2467                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2468                            btrfs_super_chunk_root(sb));
2469                 ret = -EINVAL;
2470         }
2471         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2472                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2473                            btrfs_super_log_root(sb));
2474                 ret = -EINVAL;
2475         }
2476
2477         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2478                    BTRFS_FSID_SIZE) != 0) {
2479                 btrfs_err(fs_info,
2480                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2481                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2482                 ret = -EINVAL;
2483         }
2484
2485         /*
2486          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2487          * done later
2488          */
2489         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2490                 btrfs_err(fs_info, "bytes_used is too small %llu",
2491                           btrfs_super_bytes_used(sb));
2492                 ret = -EINVAL;
2493         }
2494         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2495                 btrfs_err(fs_info, "invalid stripesize %u",
2496                           btrfs_super_stripesize(sb));
2497                 ret = -EINVAL;
2498         }
2499         if (btrfs_super_num_devices(sb) > (1UL << 31))
2500                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2501                            btrfs_super_num_devices(sb));
2502         if (btrfs_super_num_devices(sb) == 0) {
2503                 btrfs_err(fs_info, "number of devices is 0");
2504                 ret = -EINVAL;
2505         }
2506
2507         if (mirror_num >= 0 &&
2508             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2509                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2510                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2511                 ret = -EINVAL;
2512         }
2513
2514         /*
2515          * Obvious sys_chunk_array corruptions, it must hold at least one key
2516          * and one chunk
2517          */
2518         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2519                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2520                           btrfs_super_sys_array_size(sb),
2521                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2522                 ret = -EINVAL;
2523         }
2524         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2525                         + sizeof(struct btrfs_chunk)) {
2526                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2527                           btrfs_super_sys_array_size(sb),
2528                           sizeof(struct btrfs_disk_key)
2529                           + sizeof(struct btrfs_chunk));
2530                 ret = -EINVAL;
2531         }
2532
2533         /*
2534          * The generation is a global counter, we'll trust it more than the others
2535          * but it's still possible that it's the one that's wrong.
2536          */
2537         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2538                 btrfs_warn(fs_info,
2539                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2540                         btrfs_super_generation(sb),
2541                         btrfs_super_chunk_root_generation(sb));
2542         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2543             && btrfs_super_cache_generation(sb) != (u64)-1)
2544                 btrfs_warn(fs_info,
2545                         "suspicious: generation < cache_generation: %llu < %llu",
2546                         btrfs_super_generation(sb),
2547                         btrfs_super_cache_generation(sb));
2548
2549         return ret;
2550 }
2551
2552 /*
2553  * Validation of super block at mount time.
2554  * Some checks already done early at mount time, like csum type and incompat
2555  * flags will be skipped.
2556  */
2557 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2558 {
2559         return validate_super(fs_info, fs_info->super_copy, 0);
2560 }
2561
2562 /*
2563  * Validation of super block at write time.
2564  * Some checks like bytenr check will be skipped as their values will be
2565  * overwritten soon.
2566  * Extra checks like csum type and incompat flags will be done here.
2567  */
2568 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2569                                       struct btrfs_super_block *sb)
2570 {
2571         int ret;
2572
2573         ret = validate_super(fs_info, sb, -1);
2574         if (ret < 0)
2575                 goto out;
2576         if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2577                 ret = -EUCLEAN;
2578                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2579                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2580                 goto out;
2581         }
2582         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2583                 ret = -EUCLEAN;
2584                 btrfs_err(fs_info,
2585                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2586                           btrfs_super_incompat_flags(sb),
2587                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2588                 goto out;
2589         }
2590 out:
2591         if (ret < 0)
2592                 btrfs_err(fs_info,
2593                 "super block corruption detected before writing it to disk");
2594         return ret;
2595 }
2596
2597 int open_ctree(struct super_block *sb,
2598                struct btrfs_fs_devices *fs_devices,
2599                char *options)
2600 {
2601         u32 sectorsize;
2602         u32 nodesize;
2603         u32 stripesize;
2604         u64 generation;
2605         u64 features;
2606         struct btrfs_key location;
2607         struct buffer_head *bh;
2608         struct btrfs_super_block *disk_super;
2609         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2610         struct btrfs_root *tree_root;
2611         struct btrfs_root *chunk_root;
2612         int ret;
2613         int err = -EINVAL;
2614         int num_backups_tried = 0;
2615         int backup_index = 0;
2616         int clear_free_space_tree = 0;
2617         int level;
2618
2619         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2620         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2621         if (!tree_root || !chunk_root) {
2622                 err = -ENOMEM;
2623                 goto fail;
2624         }
2625
2626         ret = init_srcu_struct(&fs_info->subvol_srcu);
2627         if (ret) {
2628                 err = ret;
2629                 goto fail;
2630         }
2631
2632         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2633         if (ret) {
2634                 err = ret;
2635                 goto fail_srcu;
2636         }
2637         fs_info->dirty_metadata_batch = PAGE_SIZE *
2638                                         (1 + ilog2(nr_cpu_ids));
2639
2640         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2641         if (ret) {
2642                 err = ret;
2643                 goto fail_dirty_metadata_bytes;
2644         }
2645
2646         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2647                         GFP_KERNEL);
2648         if (ret) {
2649                 err = ret;
2650                 goto fail_delalloc_bytes;
2651         }
2652
2653         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2654         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2655         INIT_LIST_HEAD(&fs_info->trans_list);
2656         INIT_LIST_HEAD(&fs_info->dead_roots);
2657         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2658         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2659         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2660         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2661         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2662         spin_lock_init(&fs_info->delalloc_root_lock);
2663         spin_lock_init(&fs_info->trans_lock);
2664         spin_lock_init(&fs_info->fs_roots_radix_lock);
2665         spin_lock_init(&fs_info->delayed_iput_lock);
2666         spin_lock_init(&fs_info->defrag_inodes_lock);
2667         spin_lock_init(&fs_info->tree_mod_seq_lock);
2668         spin_lock_init(&fs_info->super_lock);
2669         spin_lock_init(&fs_info->qgroup_op_lock);
2670         spin_lock_init(&fs_info->buffer_lock);
2671         spin_lock_init(&fs_info->unused_bgs_lock);
2672         rwlock_init(&fs_info->tree_mod_log_lock);
2673         mutex_init(&fs_info->unused_bg_unpin_mutex);
2674         mutex_init(&fs_info->delete_unused_bgs_mutex);
2675         mutex_init(&fs_info->reloc_mutex);
2676         mutex_init(&fs_info->delalloc_root_mutex);
2677         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2678         seqlock_init(&fs_info->profiles_lock);
2679
2680         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2681         INIT_LIST_HEAD(&fs_info->space_info);
2682         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2683         INIT_LIST_HEAD(&fs_info->unused_bgs);
2684         btrfs_mapping_init(&fs_info->mapping_tree);
2685         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2686                              BTRFS_BLOCK_RSV_GLOBAL);
2687         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2688         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2689         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2690         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2691                              BTRFS_BLOCK_RSV_DELOPS);
2692         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2693                              BTRFS_BLOCK_RSV_DELREFS);
2694
2695         atomic_set(&fs_info->async_delalloc_pages, 0);
2696         atomic_set(&fs_info->defrag_running, 0);
2697         atomic_set(&fs_info->qgroup_op_seq, 0);
2698         atomic_set(&fs_info->reada_works_cnt, 0);
2699         atomic64_set(&fs_info->tree_mod_seq, 0);
2700         fs_info->sb = sb;
2701         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2702         fs_info->metadata_ratio = 0;
2703         fs_info->defrag_inodes = RB_ROOT;
2704         atomic64_set(&fs_info->free_chunk_space, 0);
2705         fs_info->tree_mod_log = RB_ROOT;
2706         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2707         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2708         /* readahead state */
2709         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2710         spin_lock_init(&fs_info->reada_lock);
2711         btrfs_init_ref_verify(fs_info);
2712
2713         fs_info->thread_pool_size = min_t(unsigned long,
2714                                           num_online_cpus() + 2, 8);
2715
2716         INIT_LIST_HEAD(&fs_info->ordered_roots);
2717         spin_lock_init(&fs_info->ordered_root_lock);
2718
2719         fs_info->btree_inode = new_inode(sb);
2720         if (!fs_info->btree_inode) {
2721                 err = -ENOMEM;
2722                 goto fail_bio_counter;
2723         }
2724         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2725
2726         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2727                                         GFP_KERNEL);
2728         if (!fs_info->delayed_root) {
2729                 err = -ENOMEM;
2730                 goto fail_iput;
2731         }
2732         btrfs_init_delayed_root(fs_info->delayed_root);
2733
2734         btrfs_init_scrub(fs_info);
2735 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2736         fs_info->check_integrity_print_mask = 0;
2737 #endif
2738         btrfs_init_balance(fs_info);
2739         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2740
2741         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2742         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2743
2744         btrfs_init_btree_inode(fs_info);
2745
2746         spin_lock_init(&fs_info->block_group_cache_lock);
2747         fs_info->block_group_cache_tree = RB_ROOT;
2748         fs_info->first_logical_byte = (u64)-1;
2749
2750         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2751         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2752         fs_info->pinned_extents = &fs_info->freed_extents[0];
2753         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2754
2755         mutex_init(&fs_info->ordered_operations_mutex);
2756         mutex_init(&fs_info->tree_log_mutex);
2757         mutex_init(&fs_info->chunk_mutex);
2758         mutex_init(&fs_info->transaction_kthread_mutex);
2759         mutex_init(&fs_info->cleaner_mutex);
2760         mutex_init(&fs_info->ro_block_group_mutex);
2761         init_rwsem(&fs_info->commit_root_sem);
2762         init_rwsem(&fs_info->cleanup_work_sem);
2763         init_rwsem(&fs_info->subvol_sem);
2764         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2765
2766         btrfs_init_dev_replace_locks(fs_info);
2767         btrfs_init_qgroup(fs_info);
2768
2769         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2770         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2771
2772         init_waitqueue_head(&fs_info->transaction_throttle);
2773         init_waitqueue_head(&fs_info->transaction_wait);
2774         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2775         init_waitqueue_head(&fs_info->async_submit_wait);
2776
2777         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2778
2779         /* Usable values until the real ones are cached from the superblock */
2780         fs_info->nodesize = 4096;
2781         fs_info->sectorsize = 4096;
2782         fs_info->stripesize = 4096;
2783
2784         spin_lock_init(&fs_info->swapfile_pins_lock);
2785         fs_info->swapfile_pins = RB_ROOT;
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         brelse(bh);
2824
2825         disk_super = fs_info->super_copy;
2826
2827         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2828                        BTRFS_FSID_SIZE));
2829
2830         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2831                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2832                                 fs_info->super_copy->metadata_uuid,
2833                                 BTRFS_FSID_SIZE));
2834         }
2835
2836         features = btrfs_super_flags(disk_super);
2837         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2838                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2839                 btrfs_set_super_flags(disk_super, features);
2840                 btrfs_info(fs_info,
2841                         "found metadata UUID change in progress flag, clearing");
2842         }
2843
2844         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2845                sizeof(*fs_info->super_for_commit));
2846
2847         ret = btrfs_validate_mount_super(fs_info);
2848         if (ret) {
2849                 btrfs_err(fs_info, "superblock contains fatal errors");
2850                 err = -EINVAL;
2851                 goto fail_alloc;
2852         }
2853
2854         if (!btrfs_super_root(disk_super))
2855                 goto fail_alloc;
2856
2857         /* check FS state, whether FS is broken. */
2858         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2859                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2860
2861         /*
2862          * run through our array of backup supers and setup
2863          * our ring pointer to the oldest one
2864          */
2865         generation = btrfs_super_generation(disk_super);
2866         find_oldest_super_backup(fs_info, generation);
2867
2868         /*
2869          * In the long term, we'll store the compression type in the super
2870          * block, and it'll be used for per file compression control.
2871          */
2872         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2873
2874         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2875         if (ret) {
2876                 err = ret;
2877                 goto fail_alloc;
2878         }
2879
2880         features = btrfs_super_incompat_flags(disk_super) &
2881                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2882         if (features) {
2883                 btrfs_err(fs_info,
2884                     "cannot mount because of unsupported optional features (%llx)",
2885                     features);
2886                 err = -EINVAL;
2887                 goto fail_alloc;
2888         }
2889
2890         features = btrfs_super_incompat_flags(disk_super);
2891         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2892         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2893                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2894         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2895                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2896
2897         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2898                 btrfs_info(fs_info, "has skinny extents");
2899
2900         /*
2901          * flag our filesystem as having big metadata blocks if
2902          * they are bigger than the page size
2903          */
2904         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2905                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2906                         btrfs_info(fs_info,
2907                                 "flagging fs with big metadata feature");
2908                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2909         }
2910
2911         nodesize = btrfs_super_nodesize(disk_super);
2912         sectorsize = btrfs_super_sectorsize(disk_super);
2913         stripesize = sectorsize;
2914         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2915         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2916
2917         /* Cache block sizes */
2918         fs_info->nodesize = nodesize;
2919         fs_info->sectorsize = sectorsize;
2920         fs_info->stripesize = stripesize;
2921
2922         /*
2923          * mixed block groups end up with duplicate but slightly offset
2924          * extent buffers for the same range.  It leads to corruptions
2925          */
2926         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2927             (sectorsize != nodesize)) {
2928                 btrfs_err(fs_info,
2929 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2930                         nodesize, sectorsize);
2931                 goto fail_alloc;
2932         }
2933
2934         /*
2935          * Needn't use the lock because there is no other task which will
2936          * update the flag.
2937          */
2938         btrfs_set_super_incompat_flags(disk_super, features);
2939
2940         features = btrfs_super_compat_ro_flags(disk_super) &
2941                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2942         if (!sb_rdonly(sb) && features) {
2943                 btrfs_err(fs_info,
2944         "cannot mount read-write because of unsupported optional features (%llx)",
2945                        features);
2946                 err = -EINVAL;
2947                 goto fail_alloc;
2948         }
2949
2950         ret = btrfs_init_workqueues(fs_info, fs_devices);
2951         if (ret) {
2952                 err = ret;
2953                 goto fail_sb_buffer;
2954         }
2955
2956         sb->s_bdi->congested_fn = btrfs_congested_fn;
2957         sb->s_bdi->congested_data = fs_info;
2958         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2959         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2960         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2961         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2962
2963         sb->s_blocksize = sectorsize;
2964         sb->s_blocksize_bits = blksize_bits(sectorsize);
2965         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2966
2967         mutex_lock(&fs_info->chunk_mutex);
2968         ret = btrfs_read_sys_array(fs_info);
2969         mutex_unlock(&fs_info->chunk_mutex);
2970         if (ret) {
2971                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2972                 goto fail_sb_buffer;
2973         }
2974
2975         generation = btrfs_super_chunk_root_generation(disk_super);
2976         level = btrfs_super_chunk_root_level(disk_super);
2977
2978         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2979
2980         chunk_root->node = read_tree_block(fs_info,
2981                                            btrfs_super_chunk_root(disk_super),
2982                                            generation, level, NULL);
2983         if (IS_ERR(chunk_root->node) ||
2984             !extent_buffer_uptodate(chunk_root->node)) {
2985                 btrfs_err(fs_info, "failed to read chunk root");
2986                 if (!IS_ERR(chunk_root->node))
2987                         free_extent_buffer(chunk_root->node);
2988                 chunk_root->node = NULL;
2989                 goto fail_tree_roots;
2990         }
2991         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2992         chunk_root->commit_root = btrfs_root_node(chunk_root);
2993
2994         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2995            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2996
2997         ret = btrfs_read_chunk_tree(fs_info);
2998         if (ret) {
2999                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3000                 goto fail_tree_roots;
3001         }
3002
3003         /*
3004          * Keep the devid that is marked to be the target device for the
3005          * device replace procedure
3006          */
3007         btrfs_free_extra_devids(fs_devices, 0);
3008
3009         if (!fs_devices->latest_bdev) {
3010                 btrfs_err(fs_info, "failed to read devices");
3011                 goto fail_tree_roots;
3012         }
3013
3014 retry_root_backup:
3015         generation = btrfs_super_generation(disk_super);
3016         level = btrfs_super_root_level(disk_super);
3017
3018         tree_root->node = read_tree_block(fs_info,
3019                                           btrfs_super_root(disk_super),
3020                                           generation, level, NULL);
3021         if (IS_ERR(tree_root->node) ||
3022             !extent_buffer_uptodate(tree_root->node)) {
3023                 btrfs_warn(fs_info, "failed to read tree root");
3024                 if (!IS_ERR(tree_root->node))
3025                         free_extent_buffer(tree_root->node);
3026                 tree_root->node = NULL;
3027                 goto recovery_tree_root;
3028         }
3029
3030         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3031         tree_root->commit_root = btrfs_root_node(tree_root);
3032         btrfs_set_root_refs(&tree_root->root_item, 1);
3033
3034         mutex_lock(&tree_root->objectid_mutex);
3035         ret = btrfs_find_highest_objectid(tree_root,
3036                                         &tree_root->highest_objectid);
3037         if (ret) {
3038                 mutex_unlock(&tree_root->objectid_mutex);
3039                 goto recovery_tree_root;
3040         }
3041
3042         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3043
3044         mutex_unlock(&tree_root->objectid_mutex);
3045
3046         ret = btrfs_read_roots(fs_info);
3047         if (ret)
3048                 goto recovery_tree_root;
3049
3050         fs_info->generation = generation;
3051         fs_info->last_trans_committed = generation;
3052
3053         ret = btrfs_verify_dev_extents(fs_info);
3054         if (ret) {
3055                 btrfs_err(fs_info,
3056                           "failed to verify dev extents against chunks: %d",
3057                           ret);
3058                 goto fail_block_groups;
3059         }
3060         ret = btrfs_recover_balance(fs_info);
3061         if (ret) {
3062                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3063                 goto fail_block_groups;
3064         }
3065
3066         ret = btrfs_init_dev_stats(fs_info);
3067         if (ret) {
3068                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3069                 goto fail_block_groups;
3070         }
3071
3072         ret = btrfs_init_dev_replace(fs_info);
3073         if (ret) {
3074                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3075                 goto fail_block_groups;
3076         }
3077
3078         btrfs_free_extra_devids(fs_devices, 1);
3079
3080         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3081         if (ret) {
3082                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3083                                 ret);
3084                 goto fail_block_groups;
3085         }
3086
3087         ret = btrfs_sysfs_add_device(fs_devices);
3088         if (ret) {
3089                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3090                                 ret);
3091                 goto fail_fsdev_sysfs;
3092         }
3093
3094         ret = btrfs_sysfs_add_mounted(fs_info);
3095         if (ret) {
3096                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3097                 goto fail_fsdev_sysfs;
3098         }
3099
3100         ret = btrfs_init_space_info(fs_info);
3101         if (ret) {
3102                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3103                 goto fail_sysfs;
3104         }
3105
3106         ret = btrfs_read_block_groups(fs_info);
3107         if (ret) {
3108                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3109                 goto fail_sysfs;
3110         }
3111
3112         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3113                 btrfs_warn(fs_info,
3114                 "writable mount is not allowed due to too many missing devices");
3115                 goto fail_sysfs;
3116         }
3117
3118         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3119                                                "btrfs-cleaner");
3120         if (IS_ERR(fs_info->cleaner_kthread))
3121                 goto fail_sysfs;
3122
3123         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3124                                                    tree_root,
3125                                                    "btrfs-transaction");
3126         if (IS_ERR(fs_info->transaction_kthread))
3127                 goto fail_cleaner;
3128
3129         if (!btrfs_test_opt(fs_info, NOSSD) &&
3130             !fs_info->fs_devices->rotating) {
3131                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3132         }
3133
3134         /*
3135          * Mount does not set all options immediately, we can do it now and do
3136          * not have to wait for transaction commit
3137          */
3138         btrfs_apply_pending_changes(fs_info);
3139
3140 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3141         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3142                 ret = btrfsic_mount(fs_info, fs_devices,
3143                                     btrfs_test_opt(fs_info,
3144                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3145                                     1 : 0,
3146                                     fs_info->check_integrity_print_mask);
3147                 if (ret)
3148                         btrfs_warn(fs_info,
3149                                 "failed to initialize integrity check module: %d",
3150                                 ret);
3151         }
3152 #endif
3153         ret = btrfs_read_qgroup_config(fs_info);
3154         if (ret)
3155                 goto fail_trans_kthread;
3156
3157         if (btrfs_build_ref_tree(fs_info))
3158                 btrfs_err(fs_info, "couldn't build ref tree");
3159
3160         /* do not make disk changes in broken FS or nologreplay is given */
3161         if (btrfs_super_log_root(disk_super) != 0 &&
3162             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3163                 ret = btrfs_replay_log(fs_info, fs_devices);
3164                 if (ret) {
3165                         err = ret;
3166                         goto fail_qgroup;
3167                 }
3168         }
3169
3170         ret = btrfs_find_orphan_roots(fs_info);
3171         if (ret)
3172                 goto fail_qgroup;
3173
3174         if (!sb_rdonly(sb)) {
3175                 ret = btrfs_cleanup_fs_roots(fs_info);
3176                 if (ret)
3177                         goto fail_qgroup;
3178
3179                 mutex_lock(&fs_info->cleaner_mutex);
3180                 ret = btrfs_recover_relocation(tree_root);
3181                 mutex_unlock(&fs_info->cleaner_mutex);
3182                 if (ret < 0) {
3183                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3184                                         ret);
3185                         err = -EINVAL;
3186                         goto fail_qgroup;
3187                 }
3188         }
3189
3190         location.objectid = BTRFS_FS_TREE_OBJECTID;
3191         location.type = BTRFS_ROOT_ITEM_KEY;
3192         location.offset = 0;
3193
3194         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3195         if (IS_ERR(fs_info->fs_root)) {
3196                 err = PTR_ERR(fs_info->fs_root);
3197                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3198                 goto fail_qgroup;
3199         }
3200
3201         if (sb_rdonly(sb))
3202                 return 0;
3203
3204         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3205             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3206                 clear_free_space_tree = 1;
3207         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3208                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3209                 btrfs_warn(fs_info, "free space tree is invalid");
3210                 clear_free_space_tree = 1;
3211         }
3212
3213         if (clear_free_space_tree) {
3214                 btrfs_info(fs_info, "clearing free space tree");
3215                 ret = btrfs_clear_free_space_tree(fs_info);
3216                 if (ret) {
3217                         btrfs_warn(fs_info,
3218                                    "failed to clear free space tree: %d", ret);
3219                         close_ctree(fs_info);
3220                         return ret;
3221                 }
3222         }
3223
3224         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3225             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3226                 btrfs_info(fs_info, "creating free space tree");
3227                 ret = btrfs_create_free_space_tree(fs_info);
3228                 if (ret) {
3229                         btrfs_warn(fs_info,
3230                                 "failed to create free space tree: %d", ret);
3231                         close_ctree(fs_info);
3232                         return ret;
3233                 }
3234         }
3235
3236         down_read(&fs_info->cleanup_work_sem);
3237         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3238             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3239                 up_read(&fs_info->cleanup_work_sem);
3240                 close_ctree(fs_info);
3241                 return ret;
3242         }
3243         up_read(&fs_info->cleanup_work_sem);
3244
3245         ret = btrfs_resume_balance_async(fs_info);
3246         if (ret) {
3247                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3248                 close_ctree(fs_info);
3249                 return ret;
3250         }
3251
3252         ret = btrfs_resume_dev_replace_async(fs_info);
3253         if (ret) {
3254                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3255                 close_ctree(fs_info);
3256                 return ret;
3257         }
3258
3259         btrfs_qgroup_rescan_resume(fs_info);
3260
3261         if (!fs_info->uuid_root) {
3262                 btrfs_info(fs_info, "creating UUID tree");
3263                 ret = btrfs_create_uuid_tree(fs_info);
3264                 if (ret) {
3265                         btrfs_warn(fs_info,
3266                                 "failed to create the UUID tree: %d", ret);
3267                         close_ctree(fs_info);
3268                         return ret;
3269                 }
3270         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3271                    fs_info->generation !=
3272                                 btrfs_super_uuid_tree_generation(disk_super)) {
3273                 btrfs_info(fs_info, "checking UUID tree");
3274                 ret = btrfs_check_uuid_tree(fs_info);
3275                 if (ret) {
3276                         btrfs_warn(fs_info,
3277                                 "failed to check the UUID tree: %d", ret);
3278                         close_ctree(fs_info);
3279                         return ret;
3280                 }
3281         } else {
3282                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3283         }
3284         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3285
3286         /*
3287          * backuproot only affect mount behavior, and if open_ctree succeeded,
3288          * no need to keep the flag
3289          */
3290         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3291
3292         return 0;
3293
3294 fail_qgroup:
3295         btrfs_free_qgroup_config(fs_info);
3296 fail_trans_kthread:
3297         kthread_stop(fs_info->transaction_kthread);
3298         btrfs_cleanup_transaction(fs_info);
3299         btrfs_free_fs_roots(fs_info);
3300 fail_cleaner:
3301         kthread_stop(fs_info->cleaner_kthread);
3302
3303         /*
3304          * make sure we're done with the btree inode before we stop our
3305          * kthreads
3306          */
3307         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3308
3309 fail_sysfs:
3310         btrfs_sysfs_remove_mounted(fs_info);
3311
3312 fail_fsdev_sysfs:
3313         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3314
3315 fail_block_groups:
3316         btrfs_put_block_group_cache(fs_info);
3317
3318 fail_tree_roots:
3319         free_root_pointers(fs_info, 1);
3320         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3321
3322 fail_sb_buffer:
3323         btrfs_stop_all_workers(fs_info);
3324         btrfs_free_block_groups(fs_info);
3325 fail_alloc:
3326 fail_iput:
3327         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3328
3329         iput(fs_info->btree_inode);
3330 fail_bio_counter:
3331         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3332 fail_delalloc_bytes:
3333         percpu_counter_destroy(&fs_info->delalloc_bytes);
3334 fail_dirty_metadata_bytes:
3335         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3336 fail_srcu:
3337         cleanup_srcu_struct(&fs_info->subvol_srcu);
3338 fail:
3339         btrfs_free_stripe_hash_table(fs_info);
3340         btrfs_close_devices(fs_info->fs_devices);
3341         return err;
3342
3343 recovery_tree_root:
3344         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3345                 goto fail_tree_roots;
3346
3347         free_root_pointers(fs_info, 0);
3348
3349         /* don't use the log in recovery mode, it won't be valid */
3350         btrfs_set_super_log_root(disk_super, 0);
3351
3352         /* we can't trust the free space cache either */
3353         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3354
3355         ret = next_root_backup(fs_info, fs_info->super_copy,
3356                                &num_backups_tried, &backup_index);
3357         if (ret == -1)
3358                 goto fail_block_groups;
3359         goto retry_root_backup;
3360 }
3361 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3362
3363 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3364 {
3365         if (uptodate) {
3366                 set_buffer_uptodate(bh);
3367         } else {
3368                 struct btrfs_device *device = (struct btrfs_device *)
3369                         bh->b_private;
3370
3371                 btrfs_warn_rl_in_rcu(device->fs_info,
3372                                 "lost page write due to IO error on %s",
3373                                           rcu_str_deref(device->name));
3374                 /* note, we don't set_buffer_write_io_error because we have
3375                  * our own ways of dealing with the IO errors
3376                  */
3377                 clear_buffer_uptodate(bh);
3378                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3379         }
3380         unlock_buffer(bh);
3381         put_bh(bh);
3382 }
3383
3384 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3385                         struct buffer_head **bh_ret)
3386 {
3387         struct buffer_head *bh;
3388         struct btrfs_super_block *super;
3389         u64 bytenr;
3390
3391         bytenr = btrfs_sb_offset(copy_num);
3392         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3393                 return -EINVAL;
3394
3395         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3396         /*
3397          * If we fail to read from the underlying devices, as of now
3398          * the best option we have is to mark it EIO.
3399          */
3400         if (!bh)
3401                 return -EIO;
3402
3403         super = (struct btrfs_super_block *)bh->b_data;
3404         if (btrfs_super_bytenr(super) != bytenr ||
3405                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3406                 brelse(bh);
3407                 return -EINVAL;
3408         }
3409
3410         *bh_ret = bh;
3411         return 0;
3412 }
3413
3414
3415 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3416 {
3417         struct buffer_head *bh;
3418         struct buffer_head *latest = NULL;
3419         struct btrfs_super_block *super;
3420         int i;
3421         u64 transid = 0;
3422         int ret = -EINVAL;
3423
3424         /* we would like to check all the supers, but that would make
3425          * a btrfs mount succeed after a mkfs from a different FS.
3426          * So, we need to add a special mount option to scan for
3427          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3428          */
3429         for (i = 0; i < 1; i++) {
3430                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3431                 if (ret)
3432                         continue;
3433
3434                 super = (struct btrfs_super_block *)bh->b_data;
3435
3436                 if (!latest || btrfs_super_generation(super) > transid) {
3437                         brelse(latest);
3438                         latest = bh;
3439                         transid = btrfs_super_generation(super);
3440                 } else {
3441                         brelse(bh);
3442                 }
3443         }
3444
3445         if (!latest)
3446                 return ERR_PTR(ret);
3447
3448         return latest;
3449 }
3450
3451 /*
3452  * Write superblock @sb to the @device. Do not wait for completion, all the
3453  * buffer heads we write are pinned.
3454  *
3455  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3456  * the expected device size at commit time. Note that max_mirrors must be
3457  * same for write and wait phases.
3458  *
3459  * Return number of errors when buffer head is not found or submission fails.
3460  */
3461 static int write_dev_supers(struct btrfs_device *device,
3462                             struct btrfs_super_block *sb, int max_mirrors)
3463 {
3464         struct buffer_head *bh;
3465         int i;
3466         int ret;
3467         int errors = 0;
3468         u32 crc;
3469         u64 bytenr;
3470         int op_flags;
3471
3472         if (max_mirrors == 0)
3473                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3474
3475         for (i = 0; i < max_mirrors; i++) {
3476                 bytenr = btrfs_sb_offset(i);
3477                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3478                     device->commit_total_bytes)
3479                         break;
3480
3481                 btrfs_set_super_bytenr(sb, bytenr);
3482
3483                 crc = ~(u32)0;
3484                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3485                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3486                 btrfs_csum_final(crc, sb->csum);
3487
3488                 /* One reference for us, and we leave it for the caller */
3489                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3490                               BTRFS_SUPER_INFO_SIZE);
3491                 if (!bh) {
3492                         btrfs_err(device->fs_info,
3493                             "couldn't get super buffer head for bytenr %llu",
3494                             bytenr);
3495                         errors++;
3496                         continue;
3497                 }
3498
3499                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3500
3501                 /* one reference for submit_bh */
3502                 get_bh(bh);
3503
3504                 set_buffer_uptodate(bh);
3505                 lock_buffer(bh);
3506                 bh->b_end_io = btrfs_end_buffer_write_sync;
3507                 bh->b_private = device;
3508
3509                 /*
3510                  * we fua the first super.  The others we allow
3511                  * to go down lazy.
3512                  */
3513                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3514                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3515                         op_flags |= REQ_FUA;
3516                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3517                 if (ret)
3518                         errors++;
3519         }
3520         return errors < i ? 0 : -1;
3521 }
3522
3523 /*
3524  * Wait for write completion of superblocks done by write_dev_supers,
3525  * @max_mirrors same for write and wait phases.
3526  *
3527  * Return number of errors when buffer head is not found or not marked up to
3528  * date.
3529  */
3530 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3531 {
3532         struct buffer_head *bh;
3533         int i;
3534         int errors = 0;
3535         bool primary_failed = false;
3536         u64 bytenr;
3537
3538         if (max_mirrors == 0)
3539                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3540
3541         for (i = 0; i < max_mirrors; i++) {
3542                 bytenr = btrfs_sb_offset(i);
3543                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3544                     device->commit_total_bytes)
3545                         break;
3546
3547                 bh = __find_get_block(device->bdev,
3548                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3549                                       BTRFS_SUPER_INFO_SIZE);
3550                 if (!bh) {
3551                         errors++;
3552                         if (i == 0)
3553                                 primary_failed = true;
3554                         continue;
3555                 }
3556                 wait_on_buffer(bh);
3557                 if (!buffer_uptodate(bh)) {
3558                         errors++;
3559                         if (i == 0)
3560                                 primary_failed = true;
3561                 }
3562
3563                 /* drop our reference */
3564                 brelse(bh);
3565
3566                 /* drop the reference from the writing run */
3567                 brelse(bh);
3568         }
3569
3570         /* log error, force error return */
3571         if (primary_failed) {
3572                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3573                           device->devid);
3574                 return -1;
3575         }
3576
3577         return errors < i ? 0 : -1;
3578 }
3579
3580 /*
3581  * endio for the write_dev_flush, this will wake anyone waiting
3582  * for the barrier when it is done
3583  */
3584 static void btrfs_end_empty_barrier(struct bio *bio)
3585 {
3586         complete(bio->bi_private);
3587 }
3588
3589 /*
3590  * Submit a flush request to the device if it supports it. Error handling is
3591  * done in the waiting counterpart.
3592  */
3593 static void write_dev_flush(struct btrfs_device *device)
3594 {
3595         struct request_queue *q = bdev_get_queue(device->bdev);
3596         struct bio *bio = device->flush_bio;
3597
3598         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3599                 return;
3600
3601         bio_reset(bio);
3602         bio->bi_end_io = btrfs_end_empty_barrier;
3603         bio_set_dev(bio, device->bdev);
3604         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3605         init_completion(&device->flush_wait);
3606         bio->bi_private = &device->flush_wait;
3607
3608         btrfsic_submit_bio(bio);
3609         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3610 }
3611
3612 /*
3613  * If the flush bio has been submitted by write_dev_flush, wait for it.
3614  */
3615 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3616 {
3617         struct bio *bio = device->flush_bio;
3618
3619         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3620                 return BLK_STS_OK;
3621
3622         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3623         wait_for_completion_io(&device->flush_wait);
3624
3625         return bio->bi_status;
3626 }
3627
3628 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3629 {
3630         if (!btrfs_check_rw_degradable(fs_info, NULL))
3631                 return -EIO;
3632         return 0;
3633 }
3634
3635 /*
3636  * send an empty flush down to each device in parallel,
3637  * then wait for them
3638  */
3639 static int barrier_all_devices(struct btrfs_fs_info *info)
3640 {
3641         struct list_head *head;
3642         struct btrfs_device *dev;
3643         int errors_wait = 0;
3644         blk_status_t ret;
3645
3646         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3647         /* send down all the barriers */
3648         head = &info->fs_devices->devices;
3649         list_for_each_entry(dev, head, dev_list) {
3650                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3651                         continue;
3652                 if (!dev->bdev)
3653                         continue;
3654                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3655                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3656                         continue;
3657
3658                 write_dev_flush(dev);
3659                 dev->last_flush_error = BLK_STS_OK;
3660         }
3661
3662         /* wait for all the barriers */
3663         list_for_each_entry(dev, head, dev_list) {
3664                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3665                         continue;
3666                 if (!dev->bdev) {
3667                         errors_wait++;
3668                         continue;
3669                 }
3670                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3671                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3672                         continue;
3673
3674                 ret = wait_dev_flush(dev);
3675                 if (ret) {
3676                         dev->last_flush_error = ret;
3677                         btrfs_dev_stat_inc_and_print(dev,
3678                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3679                         errors_wait++;
3680                 }
3681         }
3682
3683         if (errors_wait) {
3684                 /*
3685                  * At some point we need the status of all disks
3686                  * to arrive at the volume status. So error checking
3687                  * is being pushed to a separate loop.
3688                  */
3689                 return check_barrier_error(info);
3690         }
3691         return 0;
3692 }
3693
3694 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3695 {
3696         int raid_type;
3697         int min_tolerated = INT_MAX;
3698
3699         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3700             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3701                 min_tolerated = min(min_tolerated,
3702                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3703                                     tolerated_failures);
3704
3705         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3706                 if (raid_type == BTRFS_RAID_SINGLE)
3707                         continue;
3708                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3709                         continue;
3710                 min_tolerated = min(min_tolerated,
3711                                     btrfs_raid_array[raid_type].
3712                                     tolerated_failures);
3713         }
3714
3715         if (min_tolerated == INT_MAX) {
3716                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3717                 min_tolerated = 0;
3718         }
3719
3720         return min_tolerated;
3721 }
3722
3723 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3724 {
3725         struct list_head *head;
3726         struct btrfs_device *dev;
3727         struct btrfs_super_block *sb;
3728         struct btrfs_dev_item *dev_item;
3729         int ret;
3730         int do_barriers;
3731         int max_errors;
3732         int total_errors = 0;
3733         u64 flags;
3734
3735         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3736
3737         /*
3738          * max_mirrors == 0 indicates we're from commit_transaction,
3739          * not from fsync where the tree roots in fs_info have not
3740          * been consistent on disk.
3741          */
3742         if (max_mirrors == 0)
3743                 backup_super_roots(fs_info);
3744
3745         sb = fs_info->super_for_commit;
3746         dev_item = &sb->dev_item;
3747
3748         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3749         head = &fs_info->fs_devices->devices;
3750         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3751
3752         if (do_barriers) {
3753                 ret = barrier_all_devices(fs_info);
3754                 if (ret) {
3755                         mutex_unlock(
3756                                 &fs_info->fs_devices->device_list_mutex);
3757                         btrfs_handle_fs_error(fs_info, ret,
3758                                               "errors while submitting device barriers.");
3759                         return ret;
3760                 }
3761         }
3762
3763         list_for_each_entry(dev, head, dev_list) {
3764                 if (!dev->bdev) {
3765                         total_errors++;
3766                         continue;
3767                 }
3768                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3769                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3770                         continue;
3771
3772                 btrfs_set_stack_device_generation(dev_item, 0);
3773                 btrfs_set_stack_device_type(dev_item, dev->type);
3774                 btrfs_set_stack_device_id(dev_item, dev->devid);
3775                 btrfs_set_stack_device_total_bytes(dev_item,
3776                                                    dev->commit_total_bytes);
3777                 btrfs_set_stack_device_bytes_used(dev_item,
3778                                                   dev->commit_bytes_used);
3779                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3780                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3781                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3782                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3783                 memcpy(dev_item->fsid, dev->fs_dev