~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/btrfs/disk-io.c

Version: ~ [ linux-5.10-rc5 ] ~ [ linux-5.9.10 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.79 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.159 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.208 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.245 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.245 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (C) 2007 Oracle.  All rights reserved.
  3  *
  4  * This program is free software; you can redistribute it and/or
  5  * modify it under the terms of the GNU General Public
  6  * License v2 as published by the Free Software Foundation.
  7  *
  8  * This program is distributed in the hope that it will be useful,
  9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 11  * General Public License for more details.
 12  *
 13  * You should have received a copy of the GNU General Public
 14  * License along with this program; if not, write to the
 15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 16  * Boston, MA 021110-1307, USA.
 17  */
 18 
 19 #include <linux/fs.h>
 20 #include <linux/blkdev.h>
 21 #include <linux/scatterlist.h>
 22 #include <linux/swap.h>
 23 #include <linux/radix-tree.h>
 24 #include <linux/writeback.h>
 25 #include <linux/buffer_head.h>
 26 #include <linux/workqueue.h>
 27 #include <linux/kthread.h>
 28 #include <linux/slab.h>
 29 #include <linux/migrate.h>
 30 #include <linux/ratelimit.h>
 31 #include <linux/uuid.h>
 32 #include <linux/semaphore.h>
 33 #include <asm/unaligned.h>
 34 #include "ctree.h"
 35 #include "disk-io.h"
 36 #include "hash.h"
 37 #include "transaction.h"
 38 #include "btrfs_inode.h"
 39 #include "volumes.h"
 40 #include "print-tree.h"
 41 #include "locking.h"
 42 #include "tree-log.h"
 43 #include "free-space-cache.h"
 44 #include "free-space-tree.h"
 45 #include "inode-map.h"
 46 #include "check-integrity.h"
 47 #include "rcu-string.h"
 48 #include "dev-replace.h"
 49 #include "raid56.h"
 50 #include "sysfs.h"
 51 #include "qgroup.h"
 52 #include "compression.h"
 53 #include "tree-checker.h"
 54 #include "ref-verify.h"
 55 
 56 #ifdef CONFIG_X86
 57 #include <asm/cpufeature.h>
 58 #endif
 59 
 60 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
 61                                  BTRFS_HEADER_FLAG_RELOC |\
 62                                  BTRFS_SUPER_FLAG_ERROR |\
 63                                  BTRFS_SUPER_FLAG_SEEDING |\
 64                                  BTRFS_SUPER_FLAG_METADUMP)
 65 
 66 static const struct extent_io_ops btree_extent_io_ops;
 67 static void end_workqueue_fn(struct btrfs_work *work);
 68 static void free_fs_root(struct btrfs_root *root);
 69 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info);
 70 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
 71 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
 72                                       struct btrfs_fs_info *fs_info);
 73 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
 74 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
 75                                         struct extent_io_tree *dirty_pages,
 76                                         int mark);
 77 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
 78                                        struct extent_io_tree *pinned_extents);
 79 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
 80 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
 81 
 82 /*
 83  * btrfs_end_io_wq structs are used to do processing in task context when an IO
 84  * is complete.  This is used during reads to verify checksums, and it is used
 85  * by writes to insert metadata for new file extents after IO is complete.
 86  */
 87 struct btrfs_end_io_wq {
 88         struct bio *bio;
 89         bio_end_io_t *end_io;
 90         void *private;
 91         struct btrfs_fs_info *info;
 92         blk_status_t status;
 93         enum btrfs_wq_endio_type metadata;
 94         struct btrfs_work work;
 95 };
 96 
 97 static struct kmem_cache *btrfs_end_io_wq_cache;
 98 
 99 int __init btrfs_end_io_wq_init(void)
100 {
101         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
102                                         sizeof(struct btrfs_end_io_wq),
103                                         0,
104                                         SLAB_MEM_SPREAD,
105                                         NULL);
106         if (!btrfs_end_io_wq_cache)
107                 return -ENOMEM;
108         return 0;
109 }
110 
111 void btrfs_end_io_wq_exit(void)
112 {
113         kmem_cache_destroy(btrfs_end_io_wq_cache);
114 }
115 
116 /*
117  * async submit bios are used to offload expensive checksumming
118  * onto the worker threads.  They checksum file and metadata bios
119  * just before they are sent down the IO stack.
120  */
121 struct async_submit_bio {
122         void *private_data;
123         struct btrfs_fs_info *fs_info;
124         struct bio *bio;
125         extent_submit_bio_hook_t *submit_bio_start;
126         extent_submit_bio_hook_t *submit_bio_done;
127         int mirror_num;
128         unsigned long bio_flags;
129         /*
130          * bio_offset is optional, can be used if the pages in the bio
131          * can't tell us where in the file the bio should go
132          */
133         u64 bio_offset;
134         struct btrfs_work work;
135         blk_status_t status;
136 };
137 
138 /*
139  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
140  * eb, the lockdep key is determined by the btrfs_root it belongs to and
141  * the level the eb occupies in the tree.
142  *
143  * Different roots are used for different purposes and may nest inside each
144  * other and they require separate keysets.  As lockdep keys should be
145  * static, assign keysets according to the purpose of the root as indicated
146  * by btrfs_root->objectid.  This ensures that all special purpose roots
147  * have separate keysets.
148  *
149  * Lock-nesting across peer nodes is always done with the immediate parent
150  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
151  * subclass to avoid triggering lockdep warning in such cases.
152  *
153  * The key is set by the readpage_end_io_hook after the buffer has passed
154  * csum validation but before the pages are unlocked.  It is also set by
155  * btrfs_init_new_buffer on freshly allocated blocks.
156  *
157  * We also add a check to make sure the highest level of the tree is the
158  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
159  * needs update as well.
160  */
161 #ifdef CONFIG_DEBUG_LOCK_ALLOC
162 # if BTRFS_MAX_LEVEL != 8
163 #  error
164 # endif
165 
166 static struct btrfs_lockdep_keyset {
167         u64                     id;             /* root objectid */
168         const char              *name_stem;     /* lock name stem */
169         char                    names[BTRFS_MAX_LEVEL + 1][20];
170         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
171 } btrfs_lockdep_keysets[] = {
172         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
173         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
174         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
175         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
176         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
177         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
178         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
179         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
180         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
181         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
182         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
183         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
184         { .id = 0,                              .name_stem = "tree"     },
185 };
186 
187 void __init btrfs_init_lockdep(void)
188 {
189         int i, j;
190 
191         /* initialize lockdep class names */
192         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
193                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
194 
195                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
196                         snprintf(ks->names[j], sizeof(ks->names[j]),
197                                  "btrfs-%s-%02d", ks->name_stem, j);
198         }
199 }
200 
201 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
202                                     int level)
203 {
204         struct btrfs_lockdep_keyset *ks;
205 
206         BUG_ON(level >= ARRAY_SIZE(ks->keys));
207 
208         /* find the matching keyset, id 0 is the default entry */
209         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
210                 if (ks->id == objectid)
211                         break;
212 
213         lockdep_set_class_and_name(&eb->lock,
214                                    &ks->keys[level], ks->names[level]);
215 }
216 
217 #endif
218 
219 /*
220  * extents on the btree inode are pretty simple, there's one extent
221  * that covers the entire device
222  */
223 static struct extent_map *btree_get_extent(struct btrfs_inode *inode,
224                 struct page *page, size_t pg_offset, u64 start, u64 len,
225                 int create)
226 {
227         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
228         struct extent_map_tree *em_tree = &inode->extent_tree;
229         struct extent_map *em;
230         int ret;
231 
232         read_lock(&em_tree->lock);
233         em = lookup_extent_mapping(em_tree, start, len);
234         if (em) {
235                 em->bdev = fs_info->fs_devices->latest_bdev;
236                 read_unlock(&em_tree->lock);
237                 goto out;
238         }
239         read_unlock(&em_tree->lock);
240 
241         em = alloc_extent_map();
242         if (!em) {
243                 em = ERR_PTR(-ENOMEM);
244                 goto out;
245         }
246         em->start = 0;
247         em->len = (u64)-1;
248         em->block_len = (u64)-1;
249         em->block_start = 0;
250         em->bdev = fs_info->fs_devices->latest_bdev;
251 
252         write_lock(&em_tree->lock);
253         ret = add_extent_mapping(em_tree, em, 0);
254         if (ret == -EEXIST) {
255                 free_extent_map(em);
256                 em = lookup_extent_mapping(em_tree, start, len);
257                 if (!em)
258                         em = ERR_PTR(-EIO);
259         } else if (ret) {
260                 free_extent_map(em);
261                 em = ERR_PTR(ret);
262         }
263         write_unlock(&em_tree->lock);
264 
265 out:
266         return em;
267 }
268 
269 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
270 {
271         return btrfs_crc32c(seed, data, len);
272 }
273 
274 void btrfs_csum_final(u32 crc, u8 *result)
275 {
276         put_unaligned_le32(~crc, result);
277 }
278 
279 /*
280  * compute the csum for a btree block, and either verify it or write it
281  * into the csum field of the block.
282  */
283 static int csum_tree_block(struct btrfs_fs_info *fs_info,
284                            struct extent_buffer *buf,
285                            int verify)
286 {
287         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
288         char *result = NULL;
289         unsigned long len;
290         unsigned long cur_len;
291         unsigned long offset = BTRFS_CSUM_SIZE;
292         char *kaddr;
293         unsigned long map_start;
294         unsigned long map_len;
295         int err;
296         u32 crc = ~(u32)0;
297         unsigned long inline_result;
298 
299         len = buf->len - offset;
300         while (len > 0) {
301                 err = map_private_extent_buffer(buf, offset, 32,
302                                         &kaddr, &map_start, &map_len);
303                 if (err)
304                         return err;
305                 cur_len = min(len, map_len - (offset - map_start));
306                 crc = btrfs_csum_data(kaddr + offset - map_start,
307                                       crc, cur_len);
308                 len -= cur_len;
309                 offset += cur_len;
310         }
311         if (csum_size > sizeof(inline_result)) {
312                 result = kzalloc(csum_size, GFP_NOFS);
313                 if (!result)
314                         return -ENOMEM;
315         } else {
316                 result = (char *)&inline_result;
317         }
318 
319         btrfs_csum_final(crc, result);
320 
321         if (verify) {
322                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
323                         u32 val;
324                         u32 found = 0;
325                         memcpy(&found, result, csum_size);
326 
327                         read_extent_buffer(buf, &val, 0, csum_size);
328                         btrfs_warn_rl(fs_info,
329                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
330                                 fs_info->sb->s_id, buf->start,
331                                 val, found, btrfs_header_level(buf));
332                         if (result != (char *)&inline_result)
333                                 kfree(result);
334                         return -EUCLEAN;
335                 }
336         } else {
337                 write_extent_buffer(buf, result, 0, csum_size);
338         }
339         if (result != (char *)&inline_result)
340                 kfree(result);
341         return 0;
342 }
343 
344 /*
345  * we can't consider a given block up to date unless the transid of the
346  * block matches the transid in the parent node's pointer.  This is how we
347  * detect blocks that either didn't get written at all or got written
348  * in the wrong place.
349  */
350 static int verify_parent_transid(struct extent_io_tree *io_tree,
351                                  struct extent_buffer *eb, u64 parent_transid,
352                                  int atomic)
353 {
354         struct extent_state *cached_state = NULL;
355         int ret;
356         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
357 
358         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
359                 return 0;
360 
361         if (atomic)
362                 return -EAGAIN;
363 
364         if (need_lock) {
365                 btrfs_tree_read_lock(eb);
366                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
367         }
368 
369         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
370                          &cached_state);
371         if (extent_buffer_uptodate(eb) &&
372             btrfs_header_generation(eb) == parent_transid) {
373                 ret = 0;
374                 goto out;
375         }
376         btrfs_err_rl(eb->fs_info,
377                 "parent transid verify failed on %llu wanted %llu found %llu",
378                         eb->start,
379                         parent_transid, btrfs_header_generation(eb));
380         ret = 1;
381 
382         /*
383          * Things reading via commit roots that don't have normal protection,
384          * like send, can have a really old block in cache that may point at a
385          * block that has been freed and re-allocated.  So don't clear uptodate
386          * if we find an eb that is under IO (dirty/writeback) because we could
387          * end up reading in the stale data and then writing it back out and
388          * making everybody very sad.
389          */
390         if (!extent_buffer_under_io(eb))
391                 clear_extent_buffer_uptodate(eb);
392 out:
393         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
394                              &cached_state, GFP_NOFS);
395         if (need_lock)
396                 btrfs_tree_read_unlock_blocking(eb);
397         return ret;
398 }
399 
400 /*
401  * Return 0 if the superblock checksum type matches the checksum value of that
402  * algorithm. Pass the raw disk superblock data.
403  */
404 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
405                                   char *raw_disk_sb)
406 {
407         struct btrfs_super_block *disk_sb =
408                 (struct btrfs_super_block *)raw_disk_sb;
409         u16 csum_type = btrfs_super_csum_type(disk_sb);
410         int ret = 0;
411 
412         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
413                 u32 crc = ~(u32)0;
414                 const int csum_size = sizeof(crc);
415                 char result[csum_size];
416 
417                 /*
418                  * The super_block structure does not span the whole
419                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
420                  * is filled with zeros and is included in the checksum.
421                  */
422                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
423                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
424                 btrfs_csum_final(crc, result);
425 
426                 if (memcmp(raw_disk_sb, result, csum_size))
427                         ret = 1;
428         }
429 
430         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
431                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
432                                 csum_type);
433                 ret = 1;
434         }
435 
436         return ret;
437 }
438 
439 /*
440  * helper to read a given tree block, doing retries as required when
441  * the checksums don't match and we have alternate mirrors to try.
442  */
443 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
444                                           struct extent_buffer *eb,
445                                           u64 parent_transid)
446 {
447         struct extent_io_tree *io_tree;
448         int failed = 0;
449         int ret;
450         int num_copies = 0;
451         int mirror_num = 0;
452         int failed_mirror = 0;
453 
454         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
455         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
456         while (1) {
457                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
458                                                btree_get_extent, mirror_num);
459                 if (!ret) {
460                         if (!verify_parent_transid(io_tree, eb,
461                                                    parent_transid, 0))
462                                 break;
463                         else
464                                 ret = -EIO;
465                 }
466 
467                 /*
468                  * This buffer's crc is fine, but its contents are corrupted, so
469                  * there is no reason to read the other copies, they won't be
470                  * any less wrong.
471                  */
472                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
473                         break;
474 
475                 num_copies = btrfs_num_copies(fs_info,
476                                               eb->start, eb->len);
477                 if (num_copies == 1)
478                         break;
479 
480                 if (!failed_mirror) {
481                         failed = 1;
482                         failed_mirror = eb->read_mirror;
483                 }
484 
485                 mirror_num++;
486                 if (mirror_num == failed_mirror)
487                         mirror_num++;
488 
489                 if (mirror_num > num_copies)
490                         break;
491         }
492 
493         if (failed && !ret && failed_mirror)
494                 repair_eb_io_failure(fs_info, eb, failed_mirror);
495 
496         return ret;
497 }
498 
499 /*
500  * checksum a dirty tree block before IO.  This has extra checks to make sure
501  * we only fill in the checksum field in the first page of a multi-page block
502  */
503 
504 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
505 {
506         u64 start = page_offset(page);
507         u64 found_start;
508         struct extent_buffer *eb;
509 
510         eb = (struct extent_buffer *)page->private;
511         if (page != eb->pages[0])
512                 return 0;
513 
514         found_start = btrfs_header_bytenr(eb);
515         /*
516          * Please do not consolidate these warnings into a single if.
517          * It is useful to know what went wrong.
518          */
519         if (WARN_ON(found_start != start))
520                 return -EUCLEAN;
521         if (WARN_ON(!PageUptodate(page)))
522                 return -EUCLEAN;
523 
524         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
525                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
526 
527         return csum_tree_block(fs_info, eb, 0);
528 }
529 
530 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
531                                  struct extent_buffer *eb)
532 {
533         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
534         u8 fsid[BTRFS_FSID_SIZE];
535         int ret = 1;
536 
537         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
538         while (fs_devices) {
539                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
540                         ret = 0;
541                         break;
542                 }
543                 fs_devices = fs_devices->seed;
544         }
545         return ret;
546 }
547 
548 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
549                                       u64 phy_offset, struct page *page,
550                                       u64 start, u64 end, int mirror)
551 {
552         u64 found_start;
553         int found_level;
554         struct extent_buffer *eb;
555         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
556         struct btrfs_fs_info *fs_info = root->fs_info;
557         int ret = 0;
558         int reads_done;
559 
560         if (!page->private)
561                 goto out;
562 
563         eb = (struct extent_buffer *)page->private;
564 
565         /* the pending IO might have been the only thing that kept this buffer
566          * in memory.  Make sure we have a ref for all this other checks
567          */
568         extent_buffer_get(eb);
569 
570         reads_done = atomic_dec_and_test(&eb->io_pages);
571         if (!reads_done)
572                 goto err;
573 
574         eb->read_mirror = mirror;
575         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
576                 ret = -EIO;
577                 goto err;
578         }
579 
580         found_start = btrfs_header_bytenr(eb);
581         if (found_start != eb->start) {
582                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
583                              found_start, eb->start);
584                 ret = -EIO;
585                 goto err;
586         }
587         if (check_tree_block_fsid(fs_info, eb)) {
588                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
589                              eb->start);
590                 ret = -EIO;
591                 goto err;
592         }
593         found_level = btrfs_header_level(eb);
594         if (found_level >= BTRFS_MAX_LEVEL) {
595                 btrfs_err(fs_info, "bad tree block level %d",
596                           (int)btrfs_header_level(eb));
597                 ret = -EIO;
598                 goto err;
599         }
600 
601         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
602                                        eb, found_level);
603 
604         ret = csum_tree_block(fs_info, eb, 1);
605         if (ret)
606                 goto err;
607 
608         /*
609          * If this is a leaf block and it is corrupt, set the corrupt bit so
610          * that we don't try and read the other copies of this block, just
611          * return -EIO.
612          */
613         if (found_level == 0 && btrfs_check_leaf_full(root, eb)) {
614                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
615                 ret = -EIO;
616         }
617 
618         if (found_level > 0 && btrfs_check_node(root, eb))
619                 ret = -EIO;
620 
621         if (!ret)
622                 set_extent_buffer_uptodate(eb);
623 err:
624         if (reads_done &&
625             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
626                 btree_readahead_hook(eb, ret);
627 
628         if (ret) {
629                 /*
630                  * our io error hook is going to dec the io pages
631                  * again, we have to make sure it has something
632                  * to decrement
633                  */
634                 atomic_inc(&eb->io_pages);
635                 clear_extent_buffer_uptodate(eb);
636         }
637         free_extent_buffer(eb);
638 out:
639         return ret;
640 }
641 
642 static int btree_io_failed_hook(struct page *page, int failed_mirror)
643 {
644         struct extent_buffer *eb;
645 
646         eb = (struct extent_buffer *)page->private;
647         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
648         eb->read_mirror = failed_mirror;
649         atomic_dec(&eb->io_pages);
650         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
651                 btree_readahead_hook(eb, -EIO);
652         return -EIO;    /* we fixed nothing */
653 }
654 
655 static void end_workqueue_bio(struct bio *bio)
656 {
657         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
658         struct btrfs_fs_info *fs_info;
659         struct btrfs_workqueue *wq;
660         btrfs_work_func_t func;
661 
662         fs_info = end_io_wq->info;
663         end_io_wq->status = bio->bi_status;
664 
665         if (bio_op(bio) == REQ_OP_WRITE) {
666                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
667                         wq = fs_info->endio_meta_write_workers;
668                         func = btrfs_endio_meta_write_helper;
669                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
670                         wq = fs_info->endio_freespace_worker;
671                         func = btrfs_freespace_write_helper;
672                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
673                         wq = fs_info->endio_raid56_workers;
674                         func = btrfs_endio_raid56_helper;
675                 } else {
676                         wq = fs_info->endio_write_workers;
677                         func = btrfs_endio_write_helper;
678                 }
679         } else {
680                 if (unlikely(end_io_wq->metadata ==
681                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
682                         wq = fs_info->endio_repair_workers;
683                         func = btrfs_endio_repair_helper;
684                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
685                         wq = fs_info->endio_raid56_workers;
686                         func = btrfs_endio_raid56_helper;
687                 } else if (end_io_wq->metadata) {
688                         wq = fs_info->endio_meta_workers;
689                         func = btrfs_endio_meta_helper;
690                 } else {
691                         wq = fs_info->endio_workers;
692                         func = btrfs_endio_helper;
693                 }
694         }
695 
696         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
697         btrfs_queue_work(wq, &end_io_wq->work);
698 }
699 
700 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
701                         enum btrfs_wq_endio_type metadata)
702 {
703         struct btrfs_end_io_wq *end_io_wq;
704 
705         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
706         if (!end_io_wq)
707                 return BLK_STS_RESOURCE;
708 
709         end_io_wq->private = bio->bi_private;
710         end_io_wq->end_io = bio->bi_end_io;
711         end_io_wq->info = info;
712         end_io_wq->status = 0;
713         end_io_wq->bio = bio;
714         end_io_wq->metadata = metadata;
715 
716         bio->bi_private = end_io_wq;
717         bio->bi_end_io = end_workqueue_bio;
718         return 0;
719 }
720 
721 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
722 {
723         unsigned long limit = min_t(unsigned long,
724                                     info->thread_pool_size,
725                                     info->fs_devices->open_devices);
726         return 256 * limit;
727 }
728 
729 static void run_one_async_start(struct btrfs_work *work)
730 {
731         struct async_submit_bio *async;
732         blk_status_t ret;
733 
734         async = container_of(work, struct  async_submit_bio, work);
735         ret = async->submit_bio_start(async->private_data, async->bio,
736                                       async->mirror_num, async->bio_flags,
737                                       async->bio_offset);
738         if (ret)
739                 async->status = ret;
740 }
741 
742 static void run_one_async_done(struct btrfs_work *work)
743 {
744         struct async_submit_bio *async;
745 
746         async = container_of(work, struct  async_submit_bio, work);
747 
748         /* If an error occurred we just want to clean up the bio and move on */
749         if (async->status) {
750                 async->bio->bi_status = async->status;
751                 bio_endio(async->bio);
752                 return;
753         }
754 
755         async->submit_bio_done(async->private_data, async->bio, async->mirror_num,
756                                async->bio_flags, async->bio_offset);
757 }
758 
759 static void run_one_async_free(struct btrfs_work *work)
760 {
761         struct async_submit_bio *async;
762 
763         async = container_of(work, struct  async_submit_bio, work);
764         kfree(async);
765 }
766 
767 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
768                                  int mirror_num, unsigned long bio_flags,
769                                  u64 bio_offset, void *private_data,
770                                  extent_submit_bio_hook_t *submit_bio_start,
771                                  extent_submit_bio_hook_t *submit_bio_done)
772 {
773         struct async_submit_bio *async;
774 
775         async = kmalloc(sizeof(*async), GFP_NOFS);
776         if (!async)
777                 return BLK_STS_RESOURCE;
778 
779         async->private_data = private_data;
780         async->fs_info = fs_info;
781         async->bio = bio;
782         async->mirror_num = mirror_num;
783         async->submit_bio_start = submit_bio_start;
784         async->submit_bio_done = submit_bio_done;
785 
786         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
787                         run_one_async_done, run_one_async_free);
788 
789         async->bio_flags = bio_flags;
790         async->bio_offset = bio_offset;
791 
792         async->status = 0;
793 
794         if (op_is_sync(bio->bi_opf))
795                 btrfs_set_work_high_priority(&async->work);
796 
797         btrfs_queue_work(fs_info->workers, &async->work);
798         return 0;
799 }
800 
801 static blk_status_t btree_csum_one_bio(struct bio *bio)
802 {
803         struct bio_vec *bvec;
804         struct btrfs_root *root;
805         int i, ret = 0;
806 
807         ASSERT(!bio_flagged(bio, BIO_CLONED));
808         bio_for_each_segment_all(bvec, bio, i) {
809                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
810                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
811                 if (ret)
812                         break;
813         }
814 
815         return errno_to_blk_status(ret);
816 }
817 
818 static blk_status_t __btree_submit_bio_start(void *private_data, struct bio *bio,
819                                              int mirror_num, unsigned long bio_flags,
820                                              u64 bio_offset)
821 {
822         /*
823          * when we're called for a write, we're already in the async
824          * submission context.  Just jump into btrfs_map_bio
825          */
826         return btree_csum_one_bio(bio);
827 }
828 
829 static blk_status_t __btree_submit_bio_done(void *private_data, struct bio *bio,
830                                             int mirror_num, unsigned long bio_flags,
831                                             u64 bio_offset)
832 {
833         struct inode *inode = private_data;
834         blk_status_t ret;
835 
836         /*
837          * when we're called for a write, we're already in the async
838          * submission context.  Just jump into btrfs_map_bio
839          */
840         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
841         if (ret) {
842                 bio->bi_status = ret;
843                 bio_endio(bio);
844         }
845         return ret;
846 }
847 
848 static int check_async_write(struct btrfs_inode *bi)
849 {
850         if (atomic_read(&bi->sync_writers))
851                 return 0;
852 #ifdef CONFIG_X86
853         if (static_cpu_has(X86_FEATURE_XMM4_2))
854                 return 0;
855 #endif
856         return 1;
857 }
858 
859 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
860                                           int mirror_num, unsigned long bio_flags,
861                                           u64 bio_offset)
862 {
863         struct inode *inode = private_data;
864         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
865         int async = check_async_write(BTRFS_I(inode));
866         blk_status_t ret;
867 
868         if (bio_op(bio) != REQ_OP_WRITE) {
869                 /*
870                  * called for a read, do the setup so that checksum validation
871                  * can happen in the async kernel threads
872                  */
873                 ret = btrfs_bio_wq_end_io(fs_info, bio,
874                                           BTRFS_WQ_ENDIO_METADATA);
875                 if (ret)
876                         goto out_w_error;
877                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
878         } else if (!async) {
879                 ret = btree_csum_one_bio(bio);
880                 if (ret)
881                         goto out_w_error;
882                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
883         } else {
884                 /*
885                  * kthread helpers are used to submit writes so that
886                  * checksumming can happen in parallel across all CPUs
887                  */
888                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
889                                           bio_offset, private_data,
890                                           __btree_submit_bio_start,
891                                           __btree_submit_bio_done);
892         }
893 
894         if (ret)
895                 goto out_w_error;
896         return 0;
897 
898 out_w_error:
899         bio->bi_status = ret;
900         bio_endio(bio);
901         return ret;
902 }
903 
904 #ifdef CONFIG_MIGRATION
905 static int btree_migratepage(struct address_space *mapping,
906                         struct page *newpage, struct page *page,
907                         enum migrate_mode mode)
908 {
909         /*
910          * we can't safely write a btree page from here,
911          * we haven't done the locking hook
912          */
913         if (PageDirty(page))
914                 return -EAGAIN;
915         /*
916          * Buffers may be managed in a filesystem specific way.
917          * We must have no buffers or drop them.
918          */
919         if (page_has_private(page) &&
920             !try_to_release_page(page, GFP_KERNEL))
921                 return -EAGAIN;
922         return migrate_page(mapping, newpage, page, mode);
923 }
924 #endif
925 
926 
927 static int btree_writepages(struct address_space *mapping,
928                             struct writeback_control *wbc)
929 {
930         struct btrfs_fs_info *fs_info;
931         int ret;
932 
933         if (wbc->sync_mode == WB_SYNC_NONE) {
934 
935                 if (wbc->for_kupdate)
936                         return 0;
937 
938                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
939                 /* this is a bit racy, but that's ok */
940                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
941                                              BTRFS_DIRTY_METADATA_THRESH);
942                 if (ret < 0)
943                         return 0;
944         }
945         return btree_write_cache_pages(mapping, wbc);
946 }
947 
948 static int btree_readpage(struct file *file, struct page *page)
949 {
950         struct extent_io_tree *tree;
951         tree = &BTRFS_I(page->mapping->host)->io_tree;
952         return extent_read_full_page(tree, page, btree_get_extent, 0);
953 }
954 
955 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
956 {
957         if (PageWriteback(page) || PageDirty(page))
958                 return 0;
959 
960         return try_release_extent_buffer(page);
961 }
962 
963 static void btree_invalidatepage(struct page *page, unsigned int offset,
964                                  unsigned int length)
965 {
966         struct extent_io_tree *tree;
967         tree = &BTRFS_I(page->mapping->host)->io_tree;
968         extent_invalidatepage(tree, page, offset);
969         btree_releasepage(page, GFP_NOFS);
970         if (PagePrivate(page)) {
971                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
972                            "page private not zero on page %llu",
973                            (unsigned long long)page_offset(page));
974                 ClearPagePrivate(page);
975                 set_page_private(page, 0);
976                 put_page(page);
977         }
978 }
979 
980 static int btree_set_page_dirty(struct page *page)
981 {
982 #ifdef DEBUG
983         struct extent_buffer *eb;
984 
985         BUG_ON(!PagePrivate(page));
986         eb = (struct extent_buffer *)page->private;
987         BUG_ON(!eb);
988         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
989         BUG_ON(!atomic_read(&eb->refs));
990         btrfs_assert_tree_locked(eb);
991 #endif
992         return __set_page_dirty_nobuffers(page);
993 }
994 
995 static const struct address_space_operations btree_aops = {
996         .readpage       = btree_readpage,
997         .writepages     = btree_writepages,
998         .releasepage    = btree_releasepage,
999         .invalidatepage = btree_invalidatepage,
1000 #ifdef CONFIG_MIGRATION
1001         .migratepage    = btree_migratepage,
1002 #endif
1003         .set_page_dirty = btree_set_page_dirty,
1004 };
1005 
1006 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1007 {
1008         struct extent_buffer *buf = NULL;
1009         struct inode *btree_inode = fs_info->btree_inode;
1010 
1011         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1012         if (IS_ERR(buf))
1013                 return;
1014         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1015                                  buf, WAIT_NONE, btree_get_extent, 0);
1016         free_extent_buffer(buf);
1017 }
1018 
1019 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1020                          int mirror_num, struct extent_buffer **eb)
1021 {
1022         struct extent_buffer *buf = NULL;
1023         struct inode *btree_inode = fs_info->btree_inode;
1024         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1025         int ret;
1026 
1027         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1028         if (IS_ERR(buf))
1029                 return 0;
1030 
1031         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1032 
1033         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1034                                        btree_get_extent, mirror_num);
1035         if (ret) {
1036                 free_extent_buffer(buf);
1037                 return ret;
1038         }
1039 
1040         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1041                 free_extent_buffer(buf);
1042                 return -EIO;
1043         } else if (extent_buffer_uptodate(buf)) {
1044                 *eb = buf;
1045         } else {
1046                 free_extent_buffer(buf);
1047         }
1048         return 0;
1049 }
1050 
1051 struct extent_buffer *btrfs_find_create_tree_block(
1052                                                 struct btrfs_fs_info *fs_info,
1053                                                 u64 bytenr)
1054 {
1055         if (btrfs_is_testing(fs_info))
1056                 return alloc_test_extent_buffer(fs_info, bytenr);
1057         return alloc_extent_buffer(fs_info, bytenr);
1058 }
1059 
1060 
1061 int btrfs_write_tree_block(struct extent_buffer *buf)
1062 {
1063         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1064                                         buf->start + buf->len - 1);
1065 }
1066 
1067 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1068 {
1069         filemap_fdatawait_range(buf->pages[0]->mapping,
1070                                 buf->start, buf->start + buf->len - 1);
1071 }
1072 
1073 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1074                                       u64 parent_transid)
1075 {
1076         struct extent_buffer *buf = NULL;
1077         int ret;
1078 
1079         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1080         if (IS_ERR(buf))
1081                 return buf;
1082 
1083         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1084         if (ret) {
1085                 free_extent_buffer(buf);
1086                 return ERR_PTR(ret);
1087         }
1088         return buf;
1089 
1090 }
1091 
1092 void clean_tree_block(struct btrfs_fs_info *fs_info,
1093                       struct extent_buffer *buf)
1094 {
1095         if (btrfs_header_generation(buf) ==
1096             fs_info->running_transaction->transid) {
1097                 btrfs_assert_tree_locked(buf);
1098 
1099                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1100                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1101                                                  -buf->len,
1102                                                  fs_info->dirty_metadata_batch);
1103                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1104                         btrfs_set_lock_blocking(buf);
1105                         clear_extent_buffer_dirty(buf);
1106                 }
1107         }
1108 }
1109 
1110 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1111 {
1112         struct btrfs_subvolume_writers *writers;
1113         int ret;
1114 
1115         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1116         if (!writers)
1117                 return ERR_PTR(-ENOMEM);
1118 
1119         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1120         if (ret < 0) {
1121                 kfree(writers);
1122                 return ERR_PTR(ret);
1123         }
1124 
1125         init_waitqueue_head(&writers->wait);
1126         return writers;
1127 }
1128 
1129 static void
1130 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1131 {
1132         percpu_counter_destroy(&writers->counter);
1133         kfree(writers);
1134 }
1135 
1136 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1137                          u64 objectid)
1138 {
1139         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1140         root->node = NULL;
1141         root->commit_root = NULL;
1142         root->state = 0;
1143         root->orphan_cleanup_state = 0;
1144 
1145         root->objectid = objectid;
1146         root->last_trans = 0;
1147         root->highest_objectid = 0;
1148         root->nr_delalloc_inodes = 0;
1149         root->nr_ordered_extents = 0;
1150         root->name = NULL;
1151         root->inode_tree = RB_ROOT;
1152         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1153         root->block_rsv = NULL;
1154         root->orphan_block_rsv = NULL;
1155 
1156         INIT_LIST_HEAD(&root->dirty_list);
1157         INIT_LIST_HEAD(&root->root_list);
1158         INIT_LIST_HEAD(&root->delalloc_inodes);
1159         INIT_LIST_HEAD(&root->delalloc_root);
1160         INIT_LIST_HEAD(&root->ordered_extents);
1161         INIT_LIST_HEAD(&root->ordered_root);
1162         INIT_LIST_HEAD(&root->logged_list[0]);
1163         INIT_LIST_HEAD(&root->logged_list[1]);
1164         spin_lock_init(&root->orphan_lock);
1165         spin_lock_init(&root->inode_lock);
1166         spin_lock_init(&root->delalloc_lock);
1167         spin_lock_init(&root->ordered_extent_lock);
1168         spin_lock_init(&root->accounting_lock);
1169         spin_lock_init(&root->log_extents_lock[0]);
1170         spin_lock_init(&root->log_extents_lock[1]);
1171         mutex_init(&root->objectid_mutex);
1172         mutex_init(&root->log_mutex);
1173         mutex_init(&root->ordered_extent_mutex);
1174         mutex_init(&root->delalloc_mutex);
1175         init_waitqueue_head(&root->log_writer_wait);
1176         init_waitqueue_head(&root->log_commit_wait[0]);
1177         init_waitqueue_head(&root->log_commit_wait[1]);
1178         INIT_LIST_HEAD(&root->log_ctxs[0]);
1179         INIT_LIST_HEAD(&root->log_ctxs[1]);
1180         atomic_set(&root->log_commit[0], 0);
1181         atomic_set(&root->log_commit[1], 0);
1182         atomic_set(&root->log_writers, 0);
1183         atomic_set(&root->log_batch, 0);
1184         atomic_set(&root->orphan_inodes, 0);
1185         refcount_set(&root->refs, 1);
1186         atomic_set(&root->will_be_snapshotted, 0);
1187         atomic64_set(&root->qgroup_meta_rsv, 0);
1188         root->log_transid = 0;
1189         root->log_transid_committed = -1;
1190         root->last_log_commit = 0;
1191         if (!dummy)
1192                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1193 
1194         memset(&root->root_key, 0, sizeof(root->root_key));
1195         memset(&root->root_item, 0, sizeof(root->root_item));
1196         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1197         if (!dummy)
1198                 root->defrag_trans_start = fs_info->generation;
1199         else
1200                 root->defrag_trans_start = 0;
1201         root->root_key.objectid = objectid;
1202         root->anon_dev = 0;
1203 
1204         spin_lock_init(&root->root_item_lock);
1205 }
1206 
1207 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1208                 gfp_t flags)
1209 {
1210         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1211         if (root)
1212                 root->fs_info = fs_info;
1213         return root;
1214 }
1215 
1216 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1217 /* Should only be used by the testing infrastructure */
1218 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1219 {
1220         struct btrfs_root *root;
1221 
1222         if (!fs_info)
1223                 return ERR_PTR(-EINVAL);
1224 
1225         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1226         if (!root)
1227                 return ERR_PTR(-ENOMEM);
1228 
1229         /* We don't use the stripesize in selftest, set it as sectorsize */
1230         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1231         root->alloc_bytenr = 0;
1232 
1233         return root;
1234 }
1235 #endif
1236 
1237 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1238                                      struct btrfs_fs_info *fs_info,
1239                                      u64 objectid)
1240 {
1241         struct extent_buffer *leaf;
1242         struct btrfs_root *tree_root = fs_info->tree_root;
1243         struct btrfs_root *root;
1244         struct btrfs_key key;
1245         int ret = 0;
1246         uuid_le uuid;
1247 
1248         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1249         if (!root)
1250                 return ERR_PTR(-ENOMEM);
1251 
1252         __setup_root(root, fs_info, objectid);
1253         root->root_key.objectid = objectid;
1254         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1255         root->root_key.offset = 0;
1256 
1257         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1258         if (IS_ERR(leaf)) {
1259                 ret = PTR_ERR(leaf);
1260                 leaf = NULL;
1261                 goto fail;
1262         }
1263 
1264         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1265         btrfs_set_header_bytenr(leaf, leaf->start);
1266         btrfs_set_header_generation(leaf, trans->transid);
1267         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1268         btrfs_set_header_owner(leaf, objectid);
1269         root->node = leaf;
1270 
1271         write_extent_buffer_fsid(leaf, fs_info->fsid);
1272         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1273         btrfs_mark_buffer_dirty(leaf);
1274 
1275         root->commit_root = btrfs_root_node(root);
1276         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1277 
1278         root->root_item.flags = 0;
1279         root->root_item.byte_limit = 0;
1280         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1281         btrfs_set_root_generation(&root->root_item, trans->transid);
1282         btrfs_set_root_level(&root->root_item, 0);
1283         btrfs_set_root_refs(&root->root_item, 1);
1284         btrfs_set_root_used(&root->root_item, leaf->len);
1285         btrfs_set_root_last_snapshot(&root->root_item, 0);
1286         btrfs_set_root_dirid(&root->root_item, 0);
1287         uuid_le_gen(&uuid);
1288         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1289         root->root_item.drop_level = 0;
1290 
1291         key.objectid = objectid;
1292         key.type = BTRFS_ROOT_ITEM_KEY;
1293         key.offset = 0;
1294         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1295         if (ret)
1296                 goto fail;
1297 
1298         btrfs_tree_unlock(leaf);
1299 
1300         return root;
1301 
1302 fail:
1303         if (leaf) {
1304                 btrfs_tree_unlock(leaf);
1305                 free_extent_buffer(root->commit_root);
1306                 free_extent_buffer(leaf);
1307         }
1308         kfree(root);
1309 
1310         return ERR_PTR(ret);
1311 }
1312 
1313 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1314                                          struct btrfs_fs_info *fs_info)
1315 {
1316         struct btrfs_root *root;
1317         struct extent_buffer *leaf;
1318 
1319         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1320         if (!root)
1321                 return ERR_PTR(-ENOMEM);
1322 
1323         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1324 
1325         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1326         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1327         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1328 
1329         /*
1330          * DON'T set REF_COWS for log trees
1331          *
1332          * log trees do not get reference counted because they go away
1333          * before a real commit is actually done.  They do store pointers
1334          * to file data extents, and those reference counts still get
1335          * updated (along with back refs to the log tree).
1336          */
1337 
1338         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1339                         NULL, 0, 0, 0);
1340         if (IS_ERR(leaf)) {
1341                 kfree(root);
1342                 return ERR_CAST(leaf);
1343         }
1344 
1345         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1346         btrfs_set_header_bytenr(leaf, leaf->start);
1347         btrfs_set_header_generation(leaf, trans->transid);
1348         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1349         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1350         root->node = leaf;
1351 
1352         write_extent_buffer_fsid(root->node, fs_info->fsid);
1353         btrfs_mark_buffer_dirty(root->node);
1354         btrfs_tree_unlock(root->node);
1355         return root;
1356 }
1357 
1358 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1359                              struct btrfs_fs_info *fs_info)
1360 {
1361         struct btrfs_root *log_root;
1362 
1363         log_root = alloc_log_tree(trans, fs_info);
1364         if (IS_ERR(log_root))
1365                 return PTR_ERR(log_root);
1366         WARN_ON(fs_info->log_root_tree);
1367         fs_info->log_root_tree = log_root;
1368         return 0;
1369 }
1370 
1371 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1372                        struct btrfs_root *root)
1373 {
1374         struct btrfs_fs_info *fs_info = root->fs_info;
1375         struct btrfs_root *log_root;
1376         struct btrfs_inode_item *inode_item;
1377 
1378         log_root = alloc_log_tree(trans, fs_info);
1379         if (IS_ERR(log_root))
1380                 return PTR_ERR(log_root);
1381 
1382         log_root->last_trans = trans->transid;
1383         log_root->root_key.offset = root->root_key.objectid;
1384 
1385         inode_item = &log_root->root_item.inode;
1386         btrfs_set_stack_inode_generation(inode_item, 1);
1387         btrfs_set_stack_inode_size(inode_item, 3);
1388         btrfs_set_stack_inode_nlink(inode_item, 1);
1389         btrfs_set_stack_inode_nbytes(inode_item,
1390                                      fs_info->nodesize);
1391         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1392 
1393         btrfs_set_root_node(&log_root->root_item, log_root->node);
1394 
1395         WARN_ON(root->log_root);
1396         root->log_root = log_root;
1397         root->log_transid = 0;
1398         root->log_transid_committed = -1;
1399         root->last_log_commit = 0;
1400         return 0;
1401 }
1402 
1403 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1404                                                struct btrfs_key *key)
1405 {
1406         struct btrfs_root *root;
1407         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1408         struct btrfs_path *path;
1409         u64 generation;
1410         int ret;
1411 
1412         path = btrfs_alloc_path();
1413         if (!path)
1414                 return ERR_PTR(-ENOMEM);
1415 
1416         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1417         if (!root) {
1418                 ret = -ENOMEM;
1419                 goto alloc_fail;
1420         }
1421 
1422         __setup_root(root, fs_info, key->objectid);
1423 
1424         ret = btrfs_find_root(tree_root, key, path,
1425                               &root->root_item, &root->root_key);
1426         if (ret) {
1427                 if (ret > 0)
1428                         ret = -ENOENT;
1429                 goto find_fail;
1430         }
1431 
1432         generation = btrfs_root_generation(&root->root_item);
1433         root->node = read_tree_block(fs_info,
1434                                      btrfs_root_bytenr(&root->root_item),
1435                                      generation);
1436         if (IS_ERR(root->node)) {
1437                 ret = PTR_ERR(root->node);
1438                 goto find_fail;
1439         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1440                 ret = -EIO;
1441                 free_extent_buffer(root->node);
1442                 goto find_fail;
1443         }
1444         root->commit_root = btrfs_root_node(root);
1445 out:
1446         btrfs_free_path(path);
1447         return root;
1448 
1449 find_fail:
1450         kfree(root);
1451 alloc_fail:
1452         root = ERR_PTR(ret);
1453         goto out;
1454 }
1455 
1456 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1457                                       struct btrfs_key *location)
1458 {
1459         struct btrfs_root *root;
1460 
1461         root = btrfs_read_tree_root(tree_root, location);
1462         if (IS_ERR(root))
1463                 return root;
1464 
1465         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1466                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1467                 btrfs_check_and_init_root_item(&root->root_item);
1468         }
1469 
1470         return root;
1471 }
1472 
1473 int btrfs_init_fs_root(struct btrfs_root *root)
1474 {
1475         int ret;
1476         struct btrfs_subvolume_writers *writers;
1477 
1478         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1479         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1480                                         GFP_NOFS);
1481         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1482                 ret = -ENOMEM;
1483                 goto fail;
1484         }
1485 
1486         writers = btrfs_alloc_subvolume_writers();
1487         if (IS_ERR(writers)) {
1488                 ret = PTR_ERR(writers);
1489                 goto fail;
1490         }
1491         root->subv_writers = writers;
1492 
1493         btrfs_init_free_ino_ctl(root);
1494         spin_lock_init(&root->ino_cache_lock);
1495         init_waitqueue_head(&root->ino_cache_wait);
1496 
1497         ret = get_anon_bdev(&root->anon_dev);
1498         if (ret)
1499                 goto fail;
1500 
1501         mutex_lock(&root->objectid_mutex);
1502         ret = btrfs_find_highest_objectid(root,
1503                                         &root->highest_objectid);
1504         if (ret) {
1505                 mutex_unlock(&root->objectid_mutex);
1506                 goto fail;
1507         }
1508 
1509         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1510 
1511         mutex_unlock(&root->objectid_mutex);
1512 
1513         return 0;
1514 fail:
1515         /* the caller is responsible to call free_fs_root */
1516         return ret;
1517 }
1518 
1519 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1520                                         u64 root_id)
1521 {
1522         struct btrfs_root *root;
1523 
1524         spin_lock(&fs_info->fs_roots_radix_lock);
1525         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1526                                  (unsigned long)root_id);
1527         spin_unlock(&fs_info->fs_roots_radix_lock);
1528         return root;
1529 }
1530 
1531 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1532                          struct btrfs_root *root)
1533 {
1534         int ret;
1535 
1536         ret = radix_tree_preload(GFP_NOFS);
1537         if (ret)
1538                 return ret;
1539 
1540         spin_lock(&fs_info->fs_roots_radix_lock);
1541         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1542                                 (unsigned long)root->root_key.objectid,
1543                                 root);
1544         if (ret == 0)
1545                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1546         spin_unlock(&fs_info->fs_roots_radix_lock);
1547         radix_tree_preload_end();
1548 
1549         return ret;
1550 }
1551 
1552 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1553                                      struct btrfs_key *location,
1554                                      bool check_ref)
1555 {
1556         struct btrfs_root *root;
1557         struct btrfs_path *path;
1558         struct btrfs_key key;
1559         int ret;
1560 
1561         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1562                 return fs_info->tree_root;
1563         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1564                 return fs_info->extent_root;
1565         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1566                 return fs_info->chunk_root;
1567         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1568                 return fs_info->dev_root;
1569         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1570                 return fs_info->csum_root;
1571         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1572                 return fs_info->quota_root ? fs_info->quota_root :
1573                                              ERR_PTR(-ENOENT);
1574         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1575                 return fs_info->uuid_root ? fs_info->uuid_root :
1576                                             ERR_PTR(-ENOENT);
1577         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1578                 return fs_info->free_space_root ? fs_info->free_space_root :
1579                                                   ERR_PTR(-ENOENT);
1580 again:
1581         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1582         if (root) {
1583                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1584                         return ERR_PTR(-ENOENT);
1585                 return root;
1586         }
1587 
1588         root = btrfs_read_fs_root(fs_info->tree_root, location);
1589         if (IS_ERR(root))
1590                 return root;
1591 
1592         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1593                 ret = -ENOENT;
1594                 goto fail;
1595         }
1596 
1597         ret = btrfs_init_fs_root(root);
1598         if (ret)
1599                 goto fail;
1600 
1601         path = btrfs_alloc_path();
1602         if (!path) {
1603                 ret = -ENOMEM;
1604                 goto fail;
1605         }
1606         key.objectid = BTRFS_ORPHAN_OBJECTID;
1607         key.type = BTRFS_ORPHAN_ITEM_KEY;
1608         key.offset = location->objectid;
1609 
1610         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1611         btrfs_free_path(path);
1612         if (ret < 0)
1613                 goto fail;
1614         if (ret == 0)
1615                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1616 
1617         ret = btrfs_insert_fs_root(fs_info, root);
1618         if (ret) {
1619                 if (ret == -EEXIST) {
1620                         free_fs_root(root);
1621                         goto again;
1622                 }
1623                 goto fail;
1624         }
1625         return root;
1626 fail:
1627         free_fs_root(root);
1628         return ERR_PTR(ret);
1629 }
1630 
1631 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1632 {
1633         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1634         int ret = 0;
1635         struct btrfs_device *device;
1636         struct backing_dev_info *bdi;
1637 
1638         rcu_read_lock();
1639         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1640                 if (!device->bdev)
1641                         continue;
1642                 bdi = device->bdev->bd_bdi;
1643                 if (bdi_congested(bdi, bdi_bits)) {
1644                         ret = 1;
1645                         break;
1646                 }
1647         }
1648         rcu_read_unlock();
1649         return ret;
1650 }
1651 
1652 /*
1653  * called by the kthread helper functions to finally call the bio end_io
1654  * functions.  This is where read checksum verification actually happens
1655  */
1656 static void end_workqueue_fn(struct btrfs_work *work)
1657 {
1658         struct bio *bio;
1659         struct btrfs_end_io_wq *end_io_wq;
1660 
1661         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1662         bio = end_io_wq->bio;
1663 
1664         bio->bi_status = end_io_wq->status;
1665         bio->bi_private = end_io_wq->private;
1666         bio->bi_end_io = end_io_wq->end_io;
1667         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1668         bio_endio(bio);
1669 }
1670 
1671 static int cleaner_kthread(void *arg)
1672 {
1673         struct btrfs_root *root = arg;
1674         struct btrfs_fs_info *fs_info = root->fs_info;
1675         int again;
1676         struct btrfs_trans_handle *trans;
1677 
1678         do {
1679                 again = 0;
1680 
1681                 /* Make the cleaner go to sleep early. */
1682                 if (btrfs_need_cleaner_sleep(fs_info))
1683                         goto sleep;
1684 
1685                 /*
1686                  * Do not do anything if we might cause open_ctree() to block
1687                  * before we have finished mounting the filesystem.
1688                  */
1689                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1690                         goto sleep;
1691 
1692                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1693                         goto sleep;
1694 
1695                 /*
1696                  * Avoid the problem that we change the status of the fs
1697                  * during the above check and trylock.
1698                  */
1699                 if (btrfs_need_cleaner_sleep(fs_info)) {
1700                         mutex_unlock(&fs_info->cleaner_mutex);
1701                         goto sleep;
1702                 }
1703 
1704                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1705                 btrfs_run_delayed_iputs(fs_info);
1706                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1707 
1708                 again = btrfs_clean_one_deleted_snapshot(root);
1709                 mutex_unlock(&fs_info->cleaner_mutex);
1710 
1711                 /*
1712                  * The defragger has dealt with the R/O remount and umount,
1713                  * needn't do anything special here.
1714                  */
1715                 btrfs_run_defrag_inodes(fs_info);
1716 
1717                 /*
1718                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1719                  * with relocation (btrfs_relocate_chunk) and relocation
1720                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1721                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1722                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1723                  * unused block groups.
1724                  */
1725                 btrfs_delete_unused_bgs(fs_info);
1726 sleep:
1727                 if (!again) {
1728                         set_current_state(TASK_INTERRUPTIBLE);
1729                         if (!kthread_should_stop())
1730                                 schedule();
1731                         __set_current_state(TASK_RUNNING);
1732                 }
1733         } while (!kthread_should_stop());
1734 
1735         /*
1736          * Transaction kthread is stopped before us and wakes us up.
1737          * However we might have started a new transaction and COWed some
1738          * tree blocks when deleting unused block groups for example. So
1739          * make sure we commit the transaction we started to have a clean
1740          * shutdown when evicting the btree inode - if it has dirty pages
1741          * when we do the final iput() on it, eviction will trigger a
1742          * writeback for it which will fail with null pointer dereferences
1743          * since work queues and other resources were already released and
1744          * destroyed by the time the iput/eviction/writeback is made.
1745          */
1746         trans = btrfs_attach_transaction(root);
1747         if (IS_ERR(trans)) {
1748                 if (PTR_ERR(trans) != -ENOENT)
1749                         btrfs_err(fs_info,
1750                                   "cleaner transaction attach returned %ld",
1751                                   PTR_ERR(trans));
1752         } else {
1753                 int ret;
1754 
1755                 ret = btrfs_commit_transaction(trans);
1756                 if (ret)
1757                         btrfs_err(fs_info,
1758                                   "cleaner open transaction commit returned %d",
1759                                   ret);
1760         }
1761 
1762         return 0;
1763 }
1764 
1765 static int transaction_kthread(void *arg)
1766 {
1767         struct btrfs_root *root = arg;
1768         struct btrfs_fs_info *fs_info = root->fs_info;
1769         struct btrfs_trans_handle *trans;
1770         struct btrfs_transaction *cur;
1771         u64 transid;
1772         unsigned long now;
1773         unsigned long delay;
1774         bool cannot_commit;
1775 
1776         do {
1777                 cannot_commit = false;
1778                 delay = HZ * fs_info->commit_interval;
1779                 mutex_lock(&fs_info->transaction_kthread_mutex);
1780 
1781                 spin_lock(&fs_info->trans_lock);
1782                 cur = fs_info->running_transaction;
1783                 if (!cur) {
1784                         spin_unlock(&fs_info->trans_lock);
1785                         goto sleep;
1786                 }
1787 
1788                 now = get_seconds();
1789                 if (cur->state < TRANS_STATE_BLOCKED &&
1790                     (now < cur->start_time ||
1791                      now - cur->start_time < fs_info->commit_interval)) {
1792                         spin_unlock(&fs_info->trans_lock);
1793                         delay = HZ * 5;
1794                         goto sleep;
1795                 }
1796                 transid = cur->transid;
1797                 spin_unlock(&fs_info->trans_lock);
1798 
1799                 /* If the file system is aborted, this will always fail. */
1800                 trans = btrfs_attach_transaction(root);
1801                 if (IS_ERR(trans)) {
1802                         if (PTR_ERR(trans) != -ENOENT)
1803                                 cannot_commit = true;
1804                         goto sleep;
1805                 }
1806                 if (transid == trans->transid) {
1807                         btrfs_commit_transaction(trans);
1808                 } else {
1809                         btrfs_end_transaction(trans);
1810                 }
1811 sleep:
1812                 wake_up_process(fs_info->cleaner_kthread);
1813                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1814 
1815                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1816                                       &fs_info->fs_state)))
1817                         btrfs_cleanup_transaction(fs_info);
1818                 set_current_state(TASK_INTERRUPTIBLE);
1819                 if (!kthread_should_stop() &&
1820                                 (!btrfs_transaction_blocked(fs_info) ||
1821                                  cannot_commit))
1822                         schedule_timeout(delay);
1823                 __set_current_state(TASK_RUNNING);
1824         } while (!kthread_should_stop());
1825         return 0;
1826 }
1827 
1828 /*
1829  * this will find the highest generation in the array of
1830  * root backups.  The index of the highest array is returned,
1831  * or -1 if we can't find anything.
1832  *
1833  * We check to make sure the array is valid by comparing the
1834  * generation of the latest  root in the array with the generation
1835  * in the super block.  If they don't match we pitch it.
1836  */
1837 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1838 {
1839         u64 cur;
1840         int newest_index = -1;
1841         struct btrfs_root_backup *root_backup;
1842         int i;
1843 
1844         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1845                 root_backup = info->super_copy->super_roots + i;
1846                 cur = btrfs_backup_tree_root_gen(root_backup);
1847                 if (cur == newest_gen)
1848                         newest_index = i;
1849         }
1850 
1851         /* check to see if we actually wrapped around */
1852         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1853                 root_backup = info->super_copy->super_roots;
1854                 cur = btrfs_backup_tree_root_gen(root_backup);
1855                 if (cur == newest_gen)
1856                         newest_index = 0;
1857         }
1858         return newest_index;
1859 }
1860 
1861 
1862 /*
1863  * find the oldest backup so we know where to store new entries
1864  * in the backup array.  This will set the backup_root_index
1865  * field in the fs_info struct
1866  */
1867 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1868                                      u64 newest_gen)
1869 {
1870         int newest_index = -1;
1871 
1872         newest_index = find_newest_super_backup(info, newest_gen);
1873         /* if there was garbage in there, just move along */
1874         if (newest_index == -1) {
1875                 info->backup_root_index = 0;
1876         } else {
1877                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1878         }
1879 }
1880 
1881 /*
1882  * copy all the root pointers into the super backup array.
1883  * this will bump the backup pointer by one when it is
1884  * done
1885  */
1886 static void backup_super_roots(struct btrfs_fs_info *info)
1887 {
1888         int next_backup;
1889         struct btrfs_root_backup *root_backup;
1890         int last_backup;
1891 
1892         next_backup = info->backup_root_index;
1893         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1894                 BTRFS_NUM_BACKUP_ROOTS;
1895 
1896         /*
1897          * just overwrite the last backup if we're at the same generation
1898          * this happens only at umount
1899          */
1900         root_backup = info->super_for_commit->super_roots + last_backup;
1901         if (btrfs_backup_tree_root_gen(root_backup) ==
1902             btrfs_header_generation(info->tree_root->node))
1903                 next_backup = last_backup;
1904 
1905         root_backup = info->super_for_commit->super_roots + next_backup;
1906 
1907         /*
1908          * make sure all of our padding and empty slots get zero filled
1909          * regardless of which ones we use today
1910          */
1911         memset(root_backup, 0, sizeof(*root_backup));
1912 
1913         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1914 
1915         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1916         btrfs_set_backup_tree_root_gen(root_backup,
1917                                btrfs_header_generation(info->tree_root->node));
1918 
1919         btrfs_set_backup_tree_root_level(root_backup,
1920                                btrfs_header_level(info->tree_root->node));
1921 
1922         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1923         btrfs_set_backup_chunk_root_gen(root_backup,
1924                                btrfs_header_generation(info->chunk_root->node));
1925         btrfs_set_backup_chunk_root_level(root_backup,
1926                                btrfs_header_level(info->chunk_root->node));
1927 
1928         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1929         btrfs_set_backup_extent_root_gen(root_backup,
1930                                btrfs_header_generation(info->extent_root->node));
1931         btrfs_set_backup_extent_root_level(root_backup,
1932                                btrfs_header_level(info->extent_root->node));
1933 
1934         /*
1935          * we might commit during log recovery, which happens before we set
1936          * the fs_root.  Make sure it is valid before we fill it in.
1937          */
1938         if (info->fs_root && info->fs_root->node) {
1939                 btrfs_set_backup_fs_root(root_backup,
1940                                          info->fs_root->node->start);
1941                 btrfs_set_backup_fs_root_gen(root_backup,
1942                                btrfs_header_generation(info->fs_root->node));
1943                 btrfs_set_backup_fs_root_level(root_backup,
1944                                btrfs_header_level(info->fs_root->node));
1945         }
1946 
1947         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1948         btrfs_set_backup_dev_root_gen(root_backup,
1949                                btrfs_header_generation(info->dev_root->node));
1950         btrfs_set_backup_dev_root_level(root_backup,
1951                                        btrfs_header_level(info->dev_root->node));
1952 
1953         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1954         btrfs_set_backup_csum_root_gen(root_backup,
1955                                btrfs_header_generation(info->csum_root->node));
1956         btrfs_set_backup_csum_root_level(root_backup,
1957                                btrfs_header_level(info->csum_root->node));
1958 
1959         btrfs_set_backup_total_bytes(root_backup,
1960                              btrfs_super_total_bytes(info->super_copy));
1961         btrfs_set_backup_bytes_used(root_backup,
1962                              btrfs_super_bytes_used(info->super_copy));
1963         btrfs_set_backup_num_devices(root_backup,
1964                              btrfs_super_num_devices(info->super_copy));
1965 
1966         /*
1967          * if we don't copy this out to the super_copy, it won't get remembered
1968          * for the next commit
1969          */
1970         memcpy(&info->super_copy->super_roots,
1971                &info->super_for_commit->super_roots,
1972                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1973 }
1974 
1975 /*
1976  * this copies info out of the root backup array and back into
1977  * the in-memory super block.  It is meant to help iterate through
1978  * the array, so you send it the number of backups you've already
1979  * tried and the last backup index you used.
1980  *
1981  * this returns -1 when it has tried all the backups
1982  */
1983 static noinline int next_root_backup(struct btrfs_fs_info *info,
1984                                      struct btrfs_super_block *super,
1985                                      int *num_backups_tried, int *backup_index)
1986 {
1987         struct btrfs_root_backup *root_backup;
1988         int newest = *backup_index;
1989 
1990         if (*num_backups_tried == 0) {
1991                 u64 gen = btrfs_super_generation(super);
1992 
1993                 newest = find_newest_super_backup(info, gen);
1994                 if (newest == -1)
1995                         return -1;
1996 
1997                 *backup_index = newest;
1998                 *num_backups_tried = 1;
1999         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2000                 /* we've tried all the backups, all done */
2001                 return -1;
2002         } else {
2003                 /* jump to the next oldest backup */
2004                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2005                         BTRFS_NUM_BACKUP_ROOTS;
2006                 *backup_index = newest;
2007                 *num_backups_tried += 1;
2008         }
2009         root_backup = super->super_roots + newest;
2010 
2011         btrfs_set_super_generation(super,
2012                                    btrfs_backup_tree_root_gen(root_backup));
2013         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2014         btrfs_set_super_root_level(super,
2015                                    btrfs_backup_tree_root_level(root_backup));
2016         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2017 
2018         /*
2019          * fixme: the total bytes and num_devices need to match or we should
2020          * need a fsck
2021          */
2022         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2023         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2024         return 0;
2025 }
2026 
2027 /* helper to cleanup workers */
2028 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2029 {
2030         btrfs_destroy_workqueue(fs_info->fixup_workers);
2031         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2032         btrfs_destroy_workqueue(fs_info->workers);
2033         btrfs_destroy_workqueue(fs_info->endio_workers);
2034         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2035         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2036         btrfs_destroy_workqueue(fs_info->rmw_workers);
2037         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2038         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2039         btrfs_destroy_workqueue(fs_info->submit_workers);
2040         btrfs_destroy_workqueue(fs_info->delayed_workers);
2041         btrfs_destroy_workqueue(fs_info->caching_workers);
2042         btrfs_destroy_workqueue(fs_info->readahead_workers);
2043         btrfs_destroy_workqueue(fs_info->flush_workers);
2044         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2045         btrfs_destroy_workqueue(fs_info->extent_workers);
2046         /*
2047          * Now that all other work queues are destroyed, we can safely destroy
2048          * the queues used for metadata I/O, since tasks from those other work
2049          * queues can do metadata I/O operations.
2050          */
2051         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2052         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2053 }
2054 
2055 static void free_root_extent_buffers(struct btrfs_root *root)
2056 {
2057         if (root) {
2058                 free_extent_buffer(root->node);
2059                 free_extent_buffer(root->commit_root);
2060                 root->node = NULL;
2061                 root->commit_root = NULL;
2062         }
2063 }
2064 
2065 /* helper to cleanup tree roots */
2066 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2067 {
2068         free_root_extent_buffers(info->tree_root);
2069 
2070         free_root_extent_buffers(info->dev_root);
2071         free_root_extent_buffers(info->extent_root);
2072         free_root_extent_buffers(info->csum_root);
2073         free_root_extent_buffers(info->quota_root);
2074         free_root_extent_buffers(info->uuid_root);
2075         if (chunk_root)
2076                 free_root_extent_buffers(info->chunk_root);
2077         free_root_extent_buffers(info->free_space_root);
2078 }
2079 
2080 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2081 {
2082         int ret;
2083         struct btrfs_root *gang[8];
2084         int i;
2085 
2086         while (!list_empty(&fs_info->dead_roots)) {
2087                 gang[0] = list_entry(fs_info->dead_roots.next,
2088                                      struct btrfs_root, root_list);
2089                 list_del(&gang[0]->root_list);
2090 
2091                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2092                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2093                 } else {
2094                         free_extent_buffer(gang[0]->node);
2095                         free_extent_buffer(gang[0]->commit_root);
2096                         btrfs_put_fs_root(gang[0]);
2097                 }
2098         }
2099 
2100         while (1) {
2101                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2102                                              (void **)gang, 0,
2103                                              ARRAY_SIZE(gang));
2104                 if (!ret)
2105                         break;
2106                 for (i = 0; i < ret; i++)
2107                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2108         }
2109 
2110         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2111                 btrfs_free_log_root_tree(NULL, fs_info);
2112                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2113         }
2114 }
2115 
2116 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2117 {
2118         mutex_init(&fs_info->scrub_lock);
2119         atomic_set(&fs_info->scrubs_running, 0);
2120         atomic_set(&fs_info->scrub_pause_req, 0);
2121         atomic_set(&fs_info->scrubs_paused, 0);
2122         atomic_set(&fs_info->scrub_cancel_req, 0);
2123         init_waitqueue_head(&fs_info->scrub_pause_wait);
2124         fs_info->scrub_workers_refcnt = 0;
2125 }
2126 
2127 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2128 {
2129         spin_lock_init(&fs_info->balance_lock);
2130         mutex_init(&fs_info->balance_mutex);
2131         atomic_set(&fs_info->balance_running, 0);
2132         atomic_set(&fs_info->balance_pause_req, 0);
2133         atomic_set(&fs_info->balance_cancel_req, 0);
2134         fs_info->balance_ctl = NULL;
2135         init_waitqueue_head(&fs_info->balance_wait_q);
2136 }
2137 
2138 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2139 {
2140         struct inode *inode = fs_info->btree_inode;
2141 
2142         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2143         set_nlink(inode, 1);
2144         /*
2145          * we set the i_size on the btree inode to the max possible int.
2146          * the real end of the address space is determined by all of
2147          * the devices in the system
2148          */
2149         inode->i_size = OFFSET_MAX;
2150         inode->i_mapping->a_ops = &btree_aops;
2151 
2152         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2153         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2154         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2155         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2156 
2157         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2158 
2159         BTRFS_I(inode)->root = fs_info->tree_root;
2160         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2161         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2162         btrfs_insert_inode_hash(inode);
2163 }
2164 
2165 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2166 {
2167         fs_info->dev_replace.lock_owner = 0;
2168         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2169         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2170         rwlock_init(&fs_info->dev_replace.lock);
2171         atomic_set(&fs_info->dev_replace.read_locks, 0);
2172         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2173         init_waitqueue_head(&fs_info->replace_wait);
2174         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2175 }
2176 
2177 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2178 {
2179         spin_lock_init(&fs_info->qgroup_lock);
2180         mutex_init(&fs_info->qgroup_ioctl_lock);
2181         fs_info->qgroup_tree = RB_ROOT;
2182         fs_info->qgroup_op_tree = RB_ROOT;
2183         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2184         fs_info->qgroup_seq = 1;
2185         fs_info->qgroup_ulist = NULL;
2186         fs_info->qgroup_rescan_running = false;
2187         mutex_init(&fs_info->qgroup_rescan_lock);
2188 }
2189 
2190 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2191                 struct btrfs_fs_devices *fs_devices)
2192 {
2193         int max_active = fs_info->thread_pool_size;
2194         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2195 
2196         fs_info->workers =
2197                 btrfs_alloc_workqueue(fs_info, "worker",
2198                                       flags | WQ_HIGHPRI, max_active, 16);
2199 
2200         fs_info->delalloc_workers =
2201                 btrfs_alloc_workqueue(fs_info, "delalloc",
2202                                       flags, max_active, 2);
2203 
2204         fs_info->flush_workers =
2205                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2206                                       flags, max_active, 0);
2207 
2208         fs_info->caching_workers =
2209                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2210 
2211         /*
2212          * a higher idle thresh on the submit workers makes it much more
2213          * likely that bios will be send down in a sane order to the
2214          * devices
2215          */
2216         fs_info->submit_workers =
2217                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2218                                       min_t(u64, fs_devices->num_devices,
2219                                             max_active), 64);
2220 
2221         fs_info->fixup_workers =
2222                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2223 
2224         /*
2225          * endios are largely parallel and should have a very
2226          * low idle thresh
2227          */
2228         fs_info->endio_workers =
2229                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2230         fs_info->endio_meta_workers =
2231                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2232                                       max_active, 4);
2233         fs_info->endio_meta_write_workers =
2234                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2235                                       max_active, 2);
2236         fs_info->endio_raid56_workers =
2237                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2238                                       max_active, 4);
2239         fs_info->endio_repair_workers =
2240                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2241         fs_info->rmw_workers =
2242                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2243         fs_info->endio_write_workers =
2244                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2245                                       max_active, 2);
2246         fs_info->endio_freespace_worker =
2247                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2248                                       max_active, 0);
2249         fs_info->delayed_workers =
2250                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2251                                       max_active, 0);
2252         fs_info->readahead_workers =
2253                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2254                                       max_active, 2);
2255         fs_info->qgroup_rescan_workers =
2256                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2257         fs_info->extent_workers =
2258                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2259                                       min_t(u64, fs_devices->num_devices,
2260                                             max_active), 8);
2261 
2262         if (!(fs_info->workers && fs_info->delalloc_workers &&
2263               fs_info->submit_workers && fs_info->flush_workers &&
2264               fs_info->endio_workers && fs_info->endio_meta_workers &&
2265               fs_info->endio_meta_write_workers &&
2266               fs_info->endio_repair_workers &&
2267               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2268               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2269               fs_info->caching_workers && fs_info->readahead_workers &&
2270               fs_info->fixup_workers && fs_info->delayed_workers &&
2271               fs_info->extent_workers &&
2272               fs_info->qgroup_rescan_workers)) {
2273                 return -ENOMEM;
2274         }
2275 
2276         return 0;
2277 }
2278 
2279 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2280                             struct btrfs_fs_devices *fs_devices)
2281 {
2282         int ret;
2283         struct btrfs_root *log_tree_root;
2284         struct btrfs_super_block *disk_super = fs_info->super_copy;
2285         u64 bytenr = btrfs_super_log_root(disk_super);
2286 
2287         if (fs_devices->rw_devices == 0) {
2288                 btrfs_warn(fs_info, "log replay required on RO media");
2289                 return -EIO;
2290         }
2291 
2292         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2293         if (!log_tree_root)
2294                 return -ENOMEM;
2295 
2296         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2297 
2298         log_tree_root->node = read_tree_block(fs_info, bytenr,
2299                                               fs_info->generation + 1);
2300         if (IS_ERR(log_tree_root->node)) {
2301                 btrfs_warn(fs_info, "failed to read log tree");
2302                 ret = PTR_ERR(log_tree_root->node);
2303                 kfree(log_tree_root);
2304                 return ret;
2305         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2306                 btrfs_err(fs_info, "failed to read log tree");
2307                 free_extent_buffer(log_tree_root->node);
2308                 kfree(log_tree_root);
2309                 return -EIO;
2310         }
2311         /* returns with log_tree_root freed on success */
2312         ret = btrfs_recover_log_trees(log_tree_root);
2313         if (ret) {
2314                 btrfs_handle_fs_error(fs_info, ret,
2315                                       "Failed to recover log tree");
2316                 free_extent_buffer(log_tree_root->node);
2317                 kfree(log_tree_root);
2318                 return ret;
2319         }
2320 
2321         if (sb_rdonly(fs_info->sb)) {
2322                 ret = btrfs_commit_super(fs_info);
2323                 if (ret)
2324                         return ret;
2325         }
2326 
2327         return 0;
2328 }
2329 
2330 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2331 {
2332         struct btrfs_root *tree_root = fs_info->tree_root;
2333         struct btrfs_root *root;
2334         struct btrfs_key location;
2335         int ret;
2336 
2337         BUG_ON(!fs_info->tree_root);
2338 
2339         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2340         location.type = BTRFS_ROOT_ITEM_KEY;
2341         location.offset = 0;
2342 
2343         root = btrfs_read_tree_root(tree_root, &location);
2344         if (IS_ERR(root))
2345                 return PTR_ERR(root);
2346         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2347         fs_info->extent_root = root;
2348 
2349         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2350         root = btrfs_read_tree_root(tree_root, &location);
2351         if (IS_ERR(root))
2352                 return PTR_ERR(root);
2353         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2354         fs_info->dev_root = root;
2355         btrfs_init_devices_late(fs_info);
2356 
2357         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2358         root = btrfs_read_tree_root(tree_root, &location);
2359         if (IS_ERR(root))
2360                 return PTR_ERR(root);
2361         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2362         fs_info->csum_root = root;
2363 
2364         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2365         root = btrfs_read_tree_root(tree_root, &location);
2366         if (!IS_ERR(root)) {
2367                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2368                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2369                 fs_info->quota_root = root;
2370         }
2371 
2372         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2373         root = btrfs_read_tree_root(tree_root, &location);
2374         if (IS_ERR(root)) {
2375                 ret = PTR_ERR(root);
2376                 if (ret != -ENOENT)
2377                         return ret;
2378         } else {
2379                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2380                 fs_info->uuid_root = root;
2381         }
2382 
2383         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2384                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2385                 root = btrfs_read_tree_root(tree_root, &location);
2386                 if (IS_ERR(root))
2387                         return PTR_ERR(root);
2388                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2389                 fs_info->free_space_root = root;
2390         }
2391 
2392         return 0;
2393 }
2394 
2395 int open_ctree(struct super_block *sb,
2396                struct btrfs_fs_devices *fs_devices,
2397                char *options)
2398 {
2399         u32 sectorsize;
2400         u32 nodesize;
2401         u32 stripesize;
2402         u64 generation;
2403         u64 features;
2404         struct btrfs_key location;
2405         struct buffer_head *bh;
2406         struct btrfs_super_block *disk_super;
2407         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2408         struct btrfs_root *tree_root;
2409         struct btrfs_root *chunk_root;
2410         int ret;
2411         int err = -EINVAL;
2412         int num_backups_tried = 0;
2413         int backup_index = 0;
2414         int max_active;
2415         int clear_free_space_tree = 0;
2416 
2417         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2418         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2419         if (!tree_root || !chunk_root) {
2420                 err = -ENOMEM;
2421                 goto fail;
2422         }
2423 
2424         ret = init_srcu_struct(&fs_info->subvol_srcu);
2425         if (ret) {
2426                 err = ret;
2427                 goto fail;
2428         }
2429 
2430         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2431         if (ret) {
2432                 err = ret;
2433                 goto fail_srcu;
2434         }
2435         fs_info->dirty_metadata_batch = PAGE_SIZE *
2436                                         (1 + ilog2(nr_cpu_ids));
2437 
2438         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2439         if (ret) {
2440                 err = ret;
2441                 goto fail_dirty_metadata_bytes;
2442         }
2443 
2444         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2445         if (ret) {
2446                 err = ret;
2447                 goto fail_delalloc_bytes;
2448         }
2449 
2450         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2451         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2452         INIT_LIST_HEAD(&fs_info->trans_list);
2453         INIT_LIST_HEAD(&fs_info->dead_roots);
2454         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2455         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2456         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2457         spin_lock_init(&fs_info->delalloc_root_lock);
2458         spin_lock_init(&fs_info->trans_lock);
2459         spin_lock_init(&fs_info->fs_roots_radix_lock);
2460         spin_lock_init(&fs_info->delayed_iput_lock);
2461         spin_lock_init(&fs_info->defrag_inodes_lock);
2462         spin_lock_init(&fs_info->tree_mod_seq_lock);
2463         spin_lock_init(&fs_info->super_lock);
2464         spin_lock_init(&fs_info->qgroup_op_lock);
2465         spin_lock_init(&fs_info->buffer_lock);
2466         spin_lock_init(&fs_info->unused_bgs_lock);
2467         rwlock_init(&fs_info->tree_mod_log_lock);
2468         mutex_init(&fs_info->unused_bg_unpin_mutex);
2469         mutex_init(&fs_info->delete_unused_bgs_mutex);
2470         mutex_init(&fs_info->reloc_mutex);
2471         mutex_init(&fs_info->delalloc_root_mutex);
2472         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2473         seqlock_init(&fs_info->profiles_lock);
2474 
2475         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2476         INIT_LIST_HEAD(&fs_info->space_info);
2477         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2478         INIT_LIST_HEAD(&fs_info->unused_bgs);
2479         btrfs_mapping_init(&fs_info->mapping_tree);
2480         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2481                              BTRFS_BLOCK_RSV_GLOBAL);
2482         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2483         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2484         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2485         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2486                              BTRFS_BLOCK_RSV_DELOPS);
2487         atomic_set(&fs_info->async_delalloc_pages, 0);
2488         atomic_set(&fs_info->defrag_running, 0);
2489         atomic_set(&fs_info->qgroup_op_seq, 0);
2490         atomic_set(&fs_info->reada_works_cnt, 0);
2491         atomic64_set(&fs_info->tree_mod_seq, 0);
2492         fs_info->sb = sb;
2493         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2494         fs_info->metadata_ratio = 0;
2495         fs_info->defrag_inodes = RB_ROOT;
2496         atomic64_set(&fs_info->free_chunk_space, 0);
2497         fs_info->tree_mod_log = RB_ROOT;
2498         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2499         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2500         /* readahead state */
2501         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2502         spin_lock_init(&fs_info->reada_lock);
2503         btrfs_init_ref_verify(fs_info);
2504 
2505         fs_info->thread_pool_size = min_t(unsigned long,
2506                                           num_online_cpus() + 2, 8);
2507 
2508         INIT_LIST_HEAD(&fs_info->ordered_roots);
2509         spin_lock_init(&fs_info->ordered_root_lock);
2510 
2511         fs_info->btree_inode = new_inode(sb);
2512         if (!fs_info->btree_inode) {
2513                 err = -ENOMEM;
2514                 goto fail_bio_counter;
2515         }
2516         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2517 
2518         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2519                                         GFP_KERNEL);
2520         if (!fs_info->delayed_root) {
2521                 err = -ENOMEM;
2522                 goto fail_iput;
2523         }
2524         btrfs_init_delayed_root(fs_info->delayed_root);
2525 
2526         btrfs_init_scrub(fs_info);
2527 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2528         fs_info->check_integrity_print_mask = 0;
2529 #endif
2530         btrfs_init_balance(fs_info);
2531         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2532 
2533         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2534         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2535 
2536         btrfs_init_btree_inode(fs_info);
2537 
2538         spin_lock_init(&fs_info->block_group_cache_lock);
2539         fs_info->block_group_cache_tree = RB_ROOT;
2540         fs_info->first_logical_byte = (u64)-1;
2541 
2542         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2543         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2544         fs_info->pinned_extents = &fs_info->freed_extents[0];
2545         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2546 
2547         mutex_init(&fs_info->ordered_operations_mutex);
2548         mutex_init(&fs_info->tree_log_mutex);
2549         mutex_init(&fs_info->chunk_mutex);
2550         mutex_init(&fs_info->transaction_kthread_mutex);
2551         mutex_init(&fs_info->cleaner_mutex);
2552         mutex_init(&fs_info->volume_mutex);
2553         mutex_init(&fs_info->ro_block_group_mutex);
2554         init_rwsem(&fs_info->commit_root_sem);
2555         init_rwsem(&fs_info->cleanup_work_sem);
2556         init_rwsem(&fs_info->subvol_sem);
2557         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2558 
2559         btrfs_init_dev_replace_locks(fs_info);
2560         btrfs_init_qgroup(fs_info);
2561 
2562         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2563         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2564 
2565         init_waitqueue_head(&fs_info->transaction_throttle);
2566         init_waitqueue_head(&fs_info->transaction_wait);
2567         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2568         init_waitqueue_head(&fs_info->async_submit_wait);
2569 
2570         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2571 
2572         /* Usable values until the real ones are cached from the superblock */
2573         fs_info->nodesize = 4096;
2574         fs_info->sectorsize = 4096;
2575         fs_info->stripesize = 4096;
2576 
2577         ret = btrfs_alloc_stripe_hash_table(fs_info);
2578         if (ret) {
2579                 err = ret;
2580                 goto fail_alloc;
2581         }
2582 
2583         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2584 
2585         invalidate_bdev(fs_devices->latest_bdev);
2586 
2587         /*
2588          * Read super block and check the signature bytes only
2589          */
2590         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2591         if (IS_ERR(bh)) {
2592                 err = PTR_ERR(bh);
2593                 goto fail_alloc;
2594         }
2595 
2596         /*
2597          * We want to check superblock checksum, the type is stored inside.
2598          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2599          */
2600         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2601                 btrfs_err(fs_info, "superblock checksum mismatch");
2602                 err = -EINVAL;
2603                 brelse(bh);
2604                 goto fail_alloc;
2605         }
2606 
2607         /*
2608          * super_copy is zeroed at allocation time and we never touch the
2609          * following bytes up to INFO_SIZE, the checksum is calculated from
2610          * the whole block of INFO_SIZE
2611          */
2612         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2613         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2614                sizeof(*fs_info->super_for_commit));
2615         brelse(bh);
2616 
2617         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2618 
2619         ret = btrfs_check_super_valid(fs_info);
2620         if (ret) {
2621                 btrfs_err(fs_info, "superblock contains fatal errors");
2622                 err = -EINVAL;
2623                 goto fail_alloc;
2624         }
2625 
2626         disk_super = fs_info->super_copy;
2627         if (!btrfs_super_root(disk_super))
2628                 goto fail_alloc;
2629 
2630         /* check FS state, whether FS is broken. */
2631         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2632                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2633 
2634         /*
2635          * run through our array of backup supers and setup
2636          * our ring pointer to the oldest one
2637          */
2638         generation = btrfs_super_generation(disk_super);
2639         find_oldest_super_backup(fs_info, generation);
2640 
2641         /*
2642          * In the long term, we'll store the compression type in the super
2643          * block, and it'll be used for per file compression control.
2644          */
2645         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2646 
2647         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2648         if (ret) {
2649                 err = ret;
2650                 goto fail_alloc;
2651         }
2652 
2653         features = btrfs_super_incompat_flags(disk_super) &
2654                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2655         if (features) {
2656                 btrfs_err(fs_info,
2657                     "cannot mount because of unsupported optional features (%llx)",
2658                     features);
2659                 err = -EINVAL;
2660                 goto fail_alloc;
2661         }
2662 
2663         features = btrfs_super_incompat_flags(disk_super);
2664         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2665         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2666                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2667         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2668                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2669 
2670         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2671                 btrfs_info(fs_info, "has skinny extents");
2672 
2673         /*
2674          * flag our filesystem as having big metadata blocks if
2675          * they are bigger than the page size
2676          */
2677         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2678                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2679                         btrfs_info(fs_info,
2680                                 "flagging fs with big metadata feature");
2681                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2682         }
2683 
2684         nodesize = btrfs_super_nodesize(disk_super);
2685         sectorsize = btrfs_super_sectorsize(disk_super);
2686         stripesize = sectorsize;
2687         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2688         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2689 
2690         /* Cache block sizes */
2691         fs_info->nodesize = nodesize;
2692         fs_info->sectorsize = sectorsize;
2693         fs_info->stripesize = stripesize;
2694 
2695         /*
2696          * mixed block groups end up with duplicate but slightly offset
2697          * extent buffers for the same range.  It leads to corruptions
2698          */
2699         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2700             (sectorsize != nodesize)) {
2701                 btrfs_err(fs_info,
2702 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2703                         nodesize, sectorsize);
2704                 goto fail_alloc;
2705         }
2706 
2707         /*
2708          * Needn't use the lock because there is no other task which will
2709          * update the flag.
2710          */
2711         btrfs_set_super_incompat_flags(disk_super, features);
2712 
2713         features = btrfs_super_compat_ro_flags(disk_super) &
2714                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2715         if (!sb_rdonly(sb) && features) {
2716                 btrfs_err(fs_info,
2717         "cannot mount read-write because of unsupported optional features (%llx)",
2718                        features);
2719                 err = -EINVAL;
2720                 goto fail_alloc;
2721         }
2722 
2723         max_active = fs_info->thread_pool_size;
2724 
2725         ret = btrfs_init_workqueues(fs_info, fs_devices);
2726         if (ret) {
2727                 err = ret;
2728                 goto fail_sb_buffer;
2729         }
2730 
2731         sb->s_bdi->congested_fn = btrfs_congested_fn;
2732         sb->s_bdi->congested_data = fs_info;
2733         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2734         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2735         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2736         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2737 
2738         sb->s_blocksize = sectorsize;
2739         sb->s_blocksize_bits = blksize_bits(sectorsize);
2740         memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2741 
2742         mutex_lock(&fs_info->chunk_mutex);
2743         ret = btrfs_read_sys_array(fs_info);
2744         mutex_unlock(&fs_info->chunk_mutex);
2745         if (ret) {
2746                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2747                 goto fail_sb_buffer;
2748         }
2749 
2750         generation = btrfs_super_chunk_root_generation(disk_super);
2751 
2752         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2753 
2754         chunk_root->node = read_tree_block(fs_info,
2755                                            btrfs_super_chunk_root(disk_super),
2756                                            generation);
2757         if (IS_ERR(chunk_root->node) ||
2758             !extent_buffer_uptodate(chunk_root->node)) {
2759                 btrfs_err(fs_info, "failed to read chunk root");
2760                 if (!IS_ERR(chunk_root->node))
2761                         free_extent_buffer(chunk_root->node);
2762                 chunk_root->node = NULL;
2763                 goto fail_tree_roots;
2764         }
2765         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2766         chunk_root->commit_root = btrfs_root_node(chunk_root);
2767 
2768         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2769            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2770 
2771         ret = btrfs_read_chunk_tree(fs_info);
2772         if (ret) {
2773                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2774                 goto fail_tree_roots;
2775         }
2776 
2777         /*
2778          * keep the device that is marked to be the target device for the
2779          * dev_replace procedure
2780          */
2781         btrfs_close_extra_devices(fs_devices, 0);
2782 
2783         if (!fs_devices->latest_bdev) {
2784                 btrfs_err(fs_info, "failed to read devices");
2785                 goto fail_tree_roots;
2786         }
2787 
2788 retry_root_backup:
2789         generation = btrfs_super_generation(disk_super);
2790 
2791         tree_root->node = read_tree_block(fs_info,
2792                                           btrfs_super_root(disk_super),
2793                                           generation);
2794         if (IS_ERR(tree_root->node) ||
2795             !extent_buffer_uptodate(tree_root->node)) {
2796                 btrfs_warn(fs_info, "failed to read tree root");
2797                 if (!IS_ERR(tree_root->node))
2798                         free_extent_buffer(tree_root->node);
2799                 tree_root->node = NULL;
2800                 goto recovery_tree_root;
2801         }
2802 
2803         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2804         tree_root->commit_root = btrfs_root_node(tree_root);
2805         btrfs_set_root_refs(&tree_root->root_item, 1);
2806 
2807         mutex_lock(&tree_root->objectid_mutex);
2808         ret = btrfs_find_highest_objectid(tree_root,
2809                                         &tree_root->highest_objectid);
2810         if (ret) {
2811                 mutex_unlock(&tree_root->objectid_mutex);
2812                 goto recovery_tree_root;
2813         }
2814 
2815         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2816 
2817         mutex_unlock(&tree_root->objectid_mutex);
2818 
2819         ret = btrfs_read_roots(fs_info);
2820         if (ret)
2821                 goto recovery_tree_root;
2822 
2823         fs_info->generation = generation;
2824         fs_info->last_trans_committed = generation;
2825 
2826         ret = btrfs_recover_balance(fs_info);
2827         if (ret) {
2828                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2829                 goto fail_block_groups;
2830         }
2831 
2832         ret = btrfs_init_dev_stats(fs_info);
2833         if (ret) {
2834                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2835                 goto fail_block_groups;
2836         }
2837 
2838         ret = btrfs_init_dev_replace(fs_info);
2839         if (ret) {
2840                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2841                 goto fail_block_groups;
2842         }
2843 
2844         btrfs_close_extra_devices(fs_devices, 1);
2845 
2846         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2847         if (ret) {
2848                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2849                                 ret);
2850                 goto fail_block_groups;
2851         }
2852 
2853         ret = btrfs_sysfs_add_device(fs_devices);
2854         if (ret) {
2855                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2856                                 ret);
2857                 goto fail_fsdev_sysfs;
2858         }
2859 
2860         ret = btrfs_sysfs_add_mounted(fs_info);
2861         if (ret) {
2862                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2863                 goto fail_fsdev_sysfs;
2864         }
2865 
2866         ret = btrfs_init_space_info(fs_info);
2867         if (ret) {
2868                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2869                 goto fail_sysfs;
2870         }
2871 
2872         ret = btrfs_read_block_groups(fs_info);
2873         if (ret) {
2874                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2875                 goto fail_sysfs;
2876         }
2877 
2878         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info)) {
2879                 btrfs_warn(fs_info,
2880                 "writeable mount is not allowed due to too many missing devices");
2881                 goto fail_sysfs;
2882         }
2883 
2884         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2885                                                "btrfs-cleaner");
2886         if (IS_ERR(fs_info->cleaner_kthread))
2887                 goto fail_sysfs;
2888 
2889         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2890                                                    tree_root,
2891                                                    "btrfs-transaction");
2892         if (IS_ERR(fs_info->transaction_kthread))
2893                 goto fail_cleaner;
2894 
2895         if (!btrfs_test_opt(fs_info, NOSSD) &&
2896             !fs_info->fs_devices->rotating) {
2897                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2898         }
2899 
2900         /*
2901          * Mount does not set all options immediately, we can do it now and do
2902          * not have to wait for transaction commit
2903          */
2904         btrfs_apply_pending_changes(fs_info);
2905 
2906 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2907         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2908                 ret = btrfsic_mount(fs_info, fs_devices,
2909                                     btrfs_test_opt(fs_info,
2910                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2911                                     1 : 0,
2912                                     fs_info->check_integrity_print_mask);
2913                 if (ret)
2914                         btrfs_warn(fs_info,
2915                                 "failed to initialize integrity check module: %d",
2916                                 ret);
2917         }
2918 #endif
2919         ret = btrfs_read_qgroup_config(fs_info);
2920         if (ret)
2921                 goto fail_trans_kthread;
2922 
2923         if (btrfs_build_ref_tree(fs_info))
2924                 btrfs_err(fs_info, "couldn't build ref tree");
2925 
2926         /* do not make disk changes in broken FS or nologreplay is given */
2927         if (btrfs_super_log_root(disk_super) != 0 &&
2928             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2929                 ret = btrfs_replay_log(fs_info, fs_devices);
2930                 if (ret) {
2931                         err = ret;
2932                         goto fail_qgroup;
2933                 }
2934         }
2935 
2936         ret = btrfs_find_orphan_roots(fs_info);
2937         if (ret)
2938                 goto fail_qgroup;
2939 
2940         if (!sb_rdonly(sb)) {
2941                 ret = btrfs_cleanup_fs_roots(fs_info);
2942                 if (ret)
2943                         goto fail_qgroup;
2944 
2945                 mutex_lock(&fs_info->cleaner_mutex);
2946                 ret = btrfs_recover_relocation(tree_root);
2947                 mutex_unlock(&fs_info->cleaner_mutex);
2948                 if (ret < 0) {
2949                         btrfs_warn(fs_info, "failed to recover relocation: %d",
2950                                         ret);
2951                         err = -EINVAL;
2952                         goto fail_qgroup;
2953                 }
2954         }
2955 
2956         location.objectid = BTRFS_FS_TREE_OBJECTID;
2957         location.type = BTRFS_ROOT_ITEM_KEY;
2958         location.offset = 0;
2959 
2960         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2961         if (IS_ERR(fs_info->fs_root)) {
2962                 err = PTR_ERR(fs_info->fs_root);
2963                 goto fail_qgroup;
2964         }
2965 
2966         if (sb_rdonly(sb))
2967                 return 0;
2968 
2969         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2970             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2971                 clear_free_space_tree = 1;
2972         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2973                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2974                 btrfs_warn(fs_info, "free space tree is invalid");
2975                 clear_free_space_tree = 1;
2976         }
2977 
2978         if (clear_free_space_tree) {
2979                 btrfs_info(fs_info, "clearing free space tree");
2980                 ret = btrfs_clear_free_space_tree(fs_info);
2981                 if (ret) {
2982                         btrfs_warn(fs_info,
2983                                    "failed to clear free space tree: %d", ret);
2984                         close_ctree(fs_info);
2985                         return ret;
2986                 }
2987         }
2988 
2989         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
2990             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2991                 btrfs_info(fs_info, "creating free space tree");
2992                 ret = btrfs_create_free_space_tree(fs_info);
2993                 if (ret) {
2994                         btrfs_warn(fs_info,
2995                                 "failed to create free space tree: %d", ret);
2996                         close_ctree(fs_info);
2997                         return ret;
2998                 }
2999         }
3000 
3001         down_read(&fs_info->cleanup_work_sem);
3002         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3003             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3004                 up_read(&fs_info->cleanup_work_sem);
3005                 close_ctree(fs_info);
3006                 return ret;
3007         }
3008         up_read(&fs_info->cleanup_work_sem);
3009 
3010         ret = btrfs_resume_balance_async(fs_info);
3011         if (ret) {
3012                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3013                 close_ctree(fs_info);
3014                 return ret;
3015         }
3016 
3017         ret = btrfs_resume_dev_replace_async(fs_info);
3018         if (ret) {
3019                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3020                 close_ctree(fs_info);
3021                 return ret;
3022         }
3023 
3024         btrfs_qgroup_rescan_resume(fs_info);
3025 
3026         if (!fs_info->uuid_root) {
3027                 btrfs_info(fs_info, "creating UUID tree");
3028                 ret = btrfs_create_uuid_tree(fs_info);
3029                 if (ret) {
3030                         btrfs_warn(fs_info,
3031                                 "failed to create the UUID tree: %d", ret);
3032                         close_ctree(fs_info);
3033                         return ret;
3034                 }
3035         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3036                    fs_info->generation !=
3037                                 btrfs_super_uuid_tree_generation(disk_super)) {
3038                 btrfs_info(fs_info, "checking UUID tree");
3039                 ret = btrfs_check_uuid_tree(fs_info);
3040                 if (ret) {
3041                         btrfs_warn(fs_info,
3042                                 "failed to check the UUID tree: %d", ret);
3043                         close_ctree(fs_info);
3044                         return ret;
3045                 }
3046         } else {
3047                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3048         }
3049         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3050 
3051         /*
3052          * backuproot only affect mount behavior, and if open_ctree succeeded,
3053          * no need to keep the flag
3054          */
3055         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3056 
3057         return 0;
3058 
3059 fail_qgroup:
3060         btrfs_free_qgroup_config(fs_info);
3061 fail_trans_kthread:
3062         kthread_stop(fs_info->transaction_kthread);
3063         btrfs_cleanup_transaction(fs_info);
3064         btrfs_free_fs_roots(fs_info);
3065 fail_cleaner:
3066         kthread_stop(fs_info->cleaner_kthread);
3067 
3068         /*
3069          * make sure we're done with the btree inode before we stop our
3070          * kthreads
3071          */
3072         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3073 
3074 fail_sysfs:
3075         btrfs_sysfs_remove_mounted(fs_info);
3076 
3077 fail_fsdev_sysfs:
3078         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3079 
3080 fail_block_groups:
3081         btrfs_put_block_group_cache(fs_info);
3082 
3083 fail_tree_roots:
3084         free_root_pointers(fs_info, 1);
3085         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3086 
3087 fail_sb_buffer:
3088         btrfs_stop_all_workers(fs_info);
3089         btrfs_free_block_groups(fs_info);
3090 fail_alloc:
3091 fail_iput:
3092         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3093 
3094         iput(fs_info->btree_inode);
3095 fail_bio_counter:
3096         percpu_counter_destroy(&fs_info->bio_counter);
3097 fail_delalloc_bytes:
3098         percpu_counter_destroy(&fs_info->delalloc_bytes);
3099 fail_dirty_metadata_bytes:
3100         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3101 fail_srcu:
3102         cleanup_srcu_struct(&fs_info->subvol_srcu);
3103 fail:
3104         btrfs_free_stripe_hash_table(fs_info);
3105         btrfs_close_devices(fs_info->fs_devices);
3106         return err;
3107 
3108 recovery_tree_root:
3109         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3110                 goto fail_tree_roots;
3111 
3112         free_root_pointers(fs_info, 0);
3113 
3114         /* don't use the log in recovery mode, it won't be valid */
3115         btrfs_set_super_log_root(disk_super, 0);
3116 
3117         /* we can't trust the free space cache either */
3118         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3119 
3120         ret = next_root_backup(fs_info, fs_info->super_copy,
3121                                &num_backups_tried, &backup_index);
3122         if (ret == -1)
3123                 goto fail_block_groups;
3124         goto retry_root_backup;
3125 }
3126 
3127 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3128 {
3129         if (uptodate) {
3130                 set_buffer_uptodate(bh);
3131         } else {
3132                 struct btrfs_device *device = (struct btrfs_device *)
3133                         bh->b_private;
3134 
3135                 btrfs_warn_rl_in_rcu(device->fs_info,
3136                                 "lost page write due to IO error on %s",
3137                                           rcu_str_deref(device->name));
3138                 /* note, we don't set_buffer_write_io_error because we have
3139                  * our own ways of dealing with the IO errors
3140                  */
3141                 clear_buffer_uptodate(bh);
3142                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3143         }
3144         unlock_buffer(bh);
3145         put_bh(bh);
3146 }
3147 
3148 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3149                         struct buffer_head **bh_ret)
3150 {
3151         struct buffer_head *bh;
3152         struct btrfs_super_block *super;
3153         u64 bytenr;
3154 
3155         bytenr = btrfs_sb_offset(copy_num);
3156         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3157                 return -EINVAL;
3158 
3159         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3160         /*
3161          * If we fail to read from the underlying devices, as of now
3162          * the best option we have is to mark it EIO.
3163          */
3164         if (!bh)
3165                 return -EIO;
3166 
3167         super = (struct btrfs_super_block *)bh->b_data;
3168         if (btrfs_super_bytenr(super) != bytenr ||
3169                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3170                 brelse(bh);
3171                 return -EINVAL;
3172         }
3173 
3174         *bh_ret = bh;
3175         return 0;
3176 }
3177 
3178 
3179 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3180 {
3181         struct buffer_head *bh;
3182         struct buffer_head *latest = NULL;
3183         struct btrfs_super_block *super;
3184         int i;
3185         u64 transid = 0;
3186         int ret = -EINVAL;
3187 
3188         /* we would like to check all the supers, but that would make
3189          * a btrfs mount succeed after a mkfs from a different FS.
3190          * So, we need to add a special mount option to scan for
3191          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3192          */
3193         for (i = 0; i < 1; i++) {
3194                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3195                 if (ret)
3196                         continue;
3197 
3198                 super = (struct btrfs_super_block *)bh->b_data;
3199 
3200                 if (!latest || btrfs_super_generation(super) > transid) {
3201                         brelse(latest);
3202                         latest = bh;
3203                         transid = btrfs_super_generation(super);
3204                 } else {
3205                         brelse(bh);
3206                 }
3207         }
3208 
3209         if (!latest)
3210                 return ERR_PTR(ret);
3211 
3212         return latest;
3213 }
3214 
3215 /*
3216  * Write superblock @sb to the @device. Do not wait for completion, all the
3217  * buffer heads we write are pinned.
3218  *
3219  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3220  * the expected device size at commit time. Note that max_mirrors must be
3221  * same for write and wait phases.
3222  *
3223  * Return number of errors when buffer head is not found or submission fails.
3224  */
3225 static int write_dev_supers(struct btrfs_device *device,
3226                             struct btrfs_super_block *sb, int max_mirrors)
3227 {
3228         struct buffer_head *bh;
3229         int i;
3230         int ret;
3231         int errors = 0;
3232         u32 crc;
3233         u64 bytenr;
3234         int op_flags;
3235 
3236         if (max_mirrors == 0)
3237                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3238 
3239         for (i = 0; i < max_mirrors; i++) {
3240                 bytenr = btrfs_sb_offset(i);
3241                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3242                     device->commit_total_bytes)
3243                         break;
3244 
3245                 btrfs_set_super_bytenr(sb, bytenr);
3246 
3247                 crc = ~(u32)0;
3248                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3249                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3250                 btrfs_csum_final(crc, sb->csum);
3251 
3252                 /* One reference for us, and we leave it for the caller */
3253                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3254                               BTRFS_SUPER_INFO_SIZE);
3255                 if (!bh) {
3256                         btrfs_err(device->fs_info,
3257                             "couldn't get super buffer head for bytenr %llu",
3258                             bytenr);
3259                         errors++;
3260                         continue;
3261                 }
3262 
3263                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3264 
3265                 /* one reference for submit_bh */
3266                 get_bh(bh);
3267 
3268                 set_buffer_uptodate(bh);
3269                 lock_buffer(bh);
3270                 bh->b_end_io = btrfs_end_buffer_write_sync;
3271                 bh->b_private = device;
3272 
3273                 /*
3274                  * we fua the first super.  The others we allow
3275                  * to go down lazy.
3276                  */
3277                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3278                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3279                         op_flags |= REQ_FUA;
3280                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3281                 if (ret)
3282                         errors++;
3283         }
3284         return errors < i ? 0 : -1;
3285 }
3286 
3287 /*
3288  * Wait for write completion of superblocks done by write_dev_supers,
3289  * @max_mirrors same for write and wait phases.
3290  *
3291  * Return number of errors when buffer head is not found or not marked up to
3292  * date.
3293  */
3294 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3295 {
3296         struct buffer_head *bh;
3297         int i;
3298         int errors = 0;
3299         u64 bytenr;
3300 
3301         if (max_mirrors == 0)
3302                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3303 
3304         for (i = 0; i < max_mirrors; i++) {
3305                 bytenr = btrfs_sb_offset(i);
3306                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3307                     device->commit_total_bytes)
3308                         break;
3309 
3310                 bh = __find_get_block(device->bdev,
3311                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3312                                       BTRFS_SUPER_INFO_SIZE);
3313                 if (!bh) {
3314                         errors++;
3315                         continue;
3316                 }
3317                 wait_on_buffer(bh);
3318                 if (!buffer_uptodate(bh))
3319                         errors++;
3320 
3321                 /* drop our reference */
3322                 brelse(bh);
3323 
3324                 /* drop the reference from the writing run */
3325                 brelse(bh);
3326         }
3327 
3328         return errors < i ? 0 : -1;
3329 }
3330 
3331 /*
3332  * endio for the write_dev_flush, this will wake anyone waiting
3333  * for the barrier when it is done
3334  */
3335 static void btrfs_end_empty_barrier(struct bio *bio)
3336 {
3337         complete(bio->bi_private);
3338 }
3339 
3340 /*
3341  * Submit a flush request to the device if it supports it. Error handling is
3342  * done in the waiting counterpart.
3343  */
3344 static void write_dev_flush(struct btrfs_device *device)
3345 {
3346         struct request_queue *q = bdev_get_queue(device->bdev);
3347         struct bio *bio = device->flush_bio;
3348 
3349         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3350                 return;
3351 
3352         bio_reset(bio);
3353         bio->bi_end_io = btrfs_end_empty_barrier;
3354         bio_set_dev(bio, device->bdev);
3355         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3356         init_completion(&device->flush_wait);
3357         bio->bi_private = &device->flush_wait;
3358 
3359         btrfsic_submit_bio(bio);
3360         device->flush_bio_sent = 1;
3361 }
3362 
3363 /*
3364  * If the flush bio has been submitted by write_dev_flush, wait for it.
3365  */
3366 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3367 {
3368         struct bio *bio = device->flush_bio;
3369 
3370         if (!device->flush_bio_sent)
3371                 return BLK_STS_OK;
3372 
3373         device->flush_bio_sent = 0;
3374         wait_for_completion_io(&device->flush_wait);
3375 
3376         return bio->bi_status;
3377 }
3378 
3379 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3380 {
3381         if (!btrfs_check_rw_degradable(fs_info))
3382                 return -EIO;
3383         return 0;
3384 }
3385 
3386 /*
3387  * send an empty flush down to each device in parallel,
3388  * then wait for them
3389  */
3390 static int barrier_all_devices(struct btrfs_fs_info *info)
3391 {
3392         struct list_head *head;
3393         struct btrfs_device *dev;
3394         int errors_wait = 0;
3395         blk_status_t ret;
3396 
3397         /* send down all the barriers */
3398         head = &info->fs_devices->devices;
3399         list_for_each_entry_rcu(dev, head, dev_list) {
3400                 if (dev->missing)
3401                         continue;
3402                 if (!dev->bdev)
3403                         continue;
3404                 if (!dev->in_fs_metadata || !dev->writeable)
3405                         continue;
3406 
3407                 write_dev_flush(dev);
3408                 dev->last_flush_error = BLK_STS_OK;
3409         }
3410 
3411         /* wait for all the barriers */
3412         list_for_each_entry_rcu(dev, head, dev_list) {
3413                 if (dev->missing)
3414                         continue;
3415                 if (!dev->bdev) {
3416                         errors_wait++;
3417                         continue;
3418                 }
3419                 if (!dev->in_fs_metadata || !dev->writeable)
3420                         continue;
3421 
3422                 ret = wait_dev_flush(dev);
3423                 if (ret) {
3424                         dev->last_flush_error = ret;
3425                         btrfs_dev_stat_inc_and_print(dev,
3426                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3427                         errors_wait++;
3428                 }
3429         }
3430 
3431         if (errors_wait) {
3432                 /*
3433                  * At some point we need the status of all disks
3434                  * to arrive at the volume status. So error checking
3435                  * is being pushed to a separate loop.
3436                  */
3437                 return check_barrier_error(info);
3438         }
3439         return 0;
3440 }
3441 
3442 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3443 {
3444         int raid_type;
3445         int min_tolerated = INT_MAX;
3446 
3447         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3448             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3449                 min_tolerated = min(min_tolerated,
3450                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3451                                     tolerated_failures);
3452 
3453         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3454                 if (raid_type == BTRFS_RAID_SINGLE)
3455                         continue;
3456                 if (!(flags & btrfs_raid_group[raid_type]))
3457                         continue;
3458                 min_tolerated = min(min_tolerated,
3459                                     btrfs_raid_array[raid_type].
3460                                     tolerated_failures);
3461         }
3462 
3463         if (min_tolerated == INT_MAX) {
3464                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3465                 min_tolerated = 0;
3466         }
3467 
3468         return min_tolerated;
3469 }
3470 
3471 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3472 {
3473         struct list_head *head;
3474         struct btrfs_device *dev;
3475         struct btrfs_super_block *sb;
3476         struct btrfs_dev_item *dev_item;
3477         int ret;
3478         int do_barriers;
3479         int max_errors;
3480         int total_errors = 0;
3481         u64 flags;
3482 
3483         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3484 
3485         /*
3486          * max_mirrors == 0 indicates we're from commit_transaction,
3487          * not from fsync where the tree roots in fs_info have not
3488          * been consistent on disk.
3489          */
3490         if (max_mirrors == 0)
3491                 backup_super_roots(fs_info);
3492 
3493         sb = fs_info->super_for_commit;
3494         dev_item = &sb->dev_item;
3495 
3496         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3497         head = &fs_info->fs_devices->devices;
3498         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3499 
3500         if (do_barriers) {
3501                 ret = barrier_all_devices(fs_info);
3502                 if (ret) {
3503                         mutex_unlock(
3504                                 &fs_info->fs_devices->device_list_mutex);
3505                         btrfs_handle_fs_error(fs_info, ret,
3506                                               "errors while submitting device barriers.");
3507                         return ret;
3508                 }
3509         }
3510 
3511         list_for_each_entry_rcu(dev, head, dev_list) {
3512                 if (!dev->bdev) {
3513                         total_errors++;
3514                         continue;
3515                 }
3516                 if (!dev->in_fs_metadata || !dev->writeable)
3517                         continue;
3518 
3519                 btrfs_set_stack_device_generation(dev_item, 0);
3520                 btrfs_set_stack_device_type(dev_item, dev->type);
3521                 btrfs_set_stack_device_id(dev_item, dev->devid);
3522                 btrfs_set_stack_device_total_bytes(dev_item,
3523                                                    dev->commit_total_bytes);
3524                 btrfs_set_stack_device_bytes_used(dev_item,
3525                                                   dev->commit_bytes_used);
3526                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3527                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3528                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3529                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3530                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3531 
3532                 flags = btrfs_super_flags(sb);
3533                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3534 
3535                 ret = write_dev_supers(dev, sb, max_mirrors);
3536                 if (ret)
3537                         total_errors++;
3538         }
3539         if (total_errors > max_errors) {
3540                 btrfs_err(fs_info, "%d errors while writing supers",
3541                           total_errors);
3542                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3543 
3544                 /* FUA is masked off if unsupported and can't be the reason */
3545                 btrfs_handle_fs_error(fs_info, -EIO,
3546                                       "%d errors while writing supers",
3547                                       total_errors);
3548                 return -EIO;
3549         }
3550 
3551         total_errors = 0;
3552         list_for_each_entry_rcu(dev, head, dev_list) {
3553                 if (!dev->bdev)
3554                         continue;
3555                 if (!dev->in_fs_metadata || !dev->writeable)
3556                         continue;
3557 
3558                 ret = wait_dev_supers(dev, max_mirrors);
3559                 if (ret)
3560                         total_errors++;
3561         }
3562         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3563         if (total_errors > max_errors) {
3564                 btrfs_handle_fs_error(fs_info, -EIO,
3565                                       "%d errors while writing supers",
3566                                       total_errors);
3567                 return -EIO;
3568         }
3569         return 0;
3570 }
3571 
3572 /* Drop a fs root from the radix tree and free it. */
3573 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3574                                   struct btrfs_root *root)
3575 {
3576         spin_lock(&fs_info->fs_roots_radix_lock);
3577         radix_tree_delete(&fs_info->fs_roots_radix,
3578                           (unsigned long)root->root_key.objectid);
3579         spin_unlock(&fs_info->fs_roots_radix_lock);
3580 
3581         if (btrfs_root_refs(&root->root_item) == 0)
3582                 synchronize_srcu(&fs_info->subvol_srcu);
3583 
3584         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3585                 btrfs_free_log(NULL, root);
3586                 if (root->reloc_root) {
3587                         free_extent_buffer(root->reloc_root->node);
3588                         free_extent_buffer(root->reloc_root->commit_root);
3589                         btrfs_put_fs_root(root->reloc_root);
3590                         root->reloc_root = NULL;
3591                 }
3592         }
3593 
3594         if (root->free_ino_pinned)
3595                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3596         if (root->free_ino_ctl)
3597                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3598         free_fs_root(root);
3599 }
3600 
3601 static void free_fs_root(struct btrfs_root *root)
3602 {
3603         iput(root->ino_cache_inode);
3604         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3605         btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3606         root->orphan_block_rsv = NULL;
3607         if (root->anon_dev)
3608                 free_anon_bdev(root->anon_dev);
3609         if (root->subv_writers)
3610                 btrfs_free_subvolume_writers(root->subv_writers);
3611         free_extent_buffer(root->node);
3612         free_extent_buffer(root->commit_root);
3613         kfree(root->free_ino_ctl);
3614         kfree(root->free_ino_pinned);
3615         kfree(root->name);
3616         btrfs_put_fs_root(root);
3617 }
3618 
3619 void btrfs_free_fs_root(struct btrfs_root *root)
3620 {
3621         free_fs_root(root);
3622 }
3623 
3624 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3625 {
3626         u64 root_objectid = 0;
3627         struct btrfs_root *gang[8];
3628         int i = 0;
3629         int err = 0;
3630         unsigned int ret = 0;
3631         int index;
3632 
3633         while (1) {
3634                 index = srcu_read_lock(&fs_info->subvol_srcu);
3635                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3636                                              (void **)gang, root_objectid,
3637                                              ARRAY_SIZE(gang));
3638                 if (!ret) {
3639                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3640                         break;
3641                 }
3642                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3643 
3644                 for (i = 0; i < ret; i++) {
3645                         /* Avoid to grab roots in dead_roots */
3646                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3647                                 gang[i] = NULL;
3648                                 continue;
3649                         }
3650                         /* grab all the search result for later use */
3651                         gang[i] = btrfs_grab_fs_root(gang[i]);
3652                 }
3653                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3654 
3655                 for (i = 0; i < ret; i++) {
3656                         if (!gang[i])
3657                                 continue;
3658                         root_objectid = gang[i]->root_key.objectid;
3659                         err = btrfs_orphan_cleanup(gang[i]);
3660                         if (err)
3661                                 break;
3662                         btrfs_put_fs_root(gang[i]);
3663                 }
3664                 root_objectid++;
3665         }
3666 
3667         /* release the uncleaned roots due to error */
3668         for (; i < ret; i++) {
3669                 if (gang[i])
3670                         btrfs_put_fs_root(gang[i]);
3671         }
3672         return err;
3673 }
3674 
3675 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3676 {
3677         struct btrfs_root *root = fs_info->tree_root;
3678         struct btrfs_trans_handle *trans;
3679 
3680         mutex_lock(&fs_info->cleaner_mutex);
3681         btrfs_run_delayed_iputs(fs_info);
3682         mutex_unlock(&fs_info->cleaner_mutex);
3683         wake_up_process(fs_info->cleaner_kthread);
3684 
3685         /* wait until ongoing cleanup work done */
3686         down_write(&fs_info->cleanup_work_sem);
3687         up_write(&fs_info->cleanup_work_sem);
3688 
3689         trans = btrfs_join_transaction(root);
3690         if (IS_ERR(trans))
3691                 return PTR_ERR(trans);
3692         return btrfs_commit_transaction(trans);
3693 }
3694 
3695 void close_ctree(struct btrfs_fs_info *fs_info)
3696 {
3697         struct btrfs_root *root = fs_info->tree_root;
3698         int ret;
3699 
3700         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3701 
3702         /* wait for the qgroup rescan worker to stop */
3703         btrfs_qgroup_wait_for_completion(fs_info, false);
3704 
3705         /* wait for the uuid_scan task to finish */
3706         down(&fs_info->uuid_tree_rescan_sem);
3707         /* avoid complains from lockdep et al., set sem back to initial state */
3708         up(&fs_info->uuid_tree_rescan_sem);
3709 
3710         /* pause restriper - we want to resume on mount */
3711         btrfs_pause_balance(fs_info);
3712 
3713         btrfs_dev_replace_suspend_for_unmount(fs_info);
3714 
3715         btrfs_scrub_cancel(fs_info);
3716 
3717         /* wait for any defraggers to finish */
3718         wait_event(fs_info->transaction_wait,
3719                    (atomic_read(&fs_info->defrag_running) == 0));
3720 
3721         /* clear out the rbtree of defraggable inodes */
3722         btrfs_cleanup_defrag_inodes(fs_info);
3723 
3724         cancel_work_sync(&fs_info->async_reclaim_work);
3725 
3726         if (!sb_rdonly(fs_info->sb)) {
3727                 /*
3728                  * If the cleaner thread is stopped and there are
3729                  * block groups queued for removal, the deletion will be
3730                  * skipped when we quit the cleaner thread.
3731                  */
3732                 btrfs_delete_unused_bgs(fs_info);
3733 
3734                 ret = btrfs_commit_super(fs_info);
3735                 if (ret)
3736                         btrfs_err(fs_info, "commit super ret %d", ret);
3737         }
3738 
3739         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3740                 btrfs_error_commit_super(fs_info);
3741 
3742         kthread_stop(fs_info->transaction_kthread);
3743         kthread_stop(fs_info->cleaner_kthread);
3744 
3745         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3746 
3747         btrfs_free_qgroup_config(fs_info);
3748 
3749         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3750                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3751                        percpu_counter_sum(&fs_info->delalloc_bytes));
3752         }
3753 
3754         btrfs_sysfs_remove_mounted(fs_info);
3755         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3756 
3757         btrfs_free_fs_roots(fs_info);
3758 
3759         btrfs_put_block_group_cache(fs_info);
3760 
3761         /*
3762          * we must make sure there is not any read request to
3763          * submit after we stopping all workers.
3764          */
3765         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3766         btrfs_stop_all_workers(fs_info);
3767 
3768         btrfs_free_block_groups(fs_info);
3769 
3770         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3771         free_root_pointers(fs_info, 1);
3772 
3773         iput(fs_info->btree_inode);
3774 
3775 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3776         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3777                 btrfsic_unmount(fs_info->fs_devices);
3778 #endif
3779 
3780         btrfs_close_devices(fs_info->fs_devices);
3781         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3782 
3783         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3784         percpu_counter_destroy(&fs_info->delalloc_bytes);
3785         percpu_counter_destroy(&fs_info->bio_counter);
3786         cleanup_srcu_struct(&fs_info->subvol_srcu);
3787 
3788         btrfs_free_stripe_hash_table(fs_info);
3789         btrfs_free_ref_cache(fs_info);
3790 
3791         __btrfs_free_block_rsv(root->orphan_block_rsv);
3792         root->orphan_block_rsv = NULL;
3793 
3794         while (!list_empty(&fs_info->pinned_chunks)) {
3795                 struct extent_map *em;
3796 
3797                 em = list_first_entry(&fs_info->pinned_chunks,
3798                                       struct extent_map, list);
3799                 list_del_init(&em->list);
3800                 free_extent_map(em);
3801         }
3802 }
3803 
3804 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3805                           int atomic)
3806 {
3807         int ret;
3808         struct inode *btree_inode = buf->pages[0]->mapping->host;
3809 
3810         ret = extent_buffer_uptodate(buf);
3811         if (!ret)
3812                 return ret;
3813 
3814         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3815                                     parent_transid, atomic);
3816         if (ret == -EAGAIN)
3817                 return ret;
3818         return !ret;
3819 }
3820 
3821 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3822 {
3823         struct btrfs_fs_info *fs_info;
3824         struct btrfs_root *root;
3825         u64 transid = btrfs_header_generation(buf);
3826         int was_dirty;
3827 
3828 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3829         /*
3830          * This is a fast path so only do this check if we have sanity tests
3831          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3832          * outside of the sanity tests.
3833          */
3834         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3835                 return;
3836 #endif
3837         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3838         fs_info = root->fs_info;
3839         btrfs_assert_tree_locked(buf);
3840         if (transid != fs_info->generation)
3841                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
3842                         buf->start, transid, fs_info->generation);
3843         was_dirty = set_extent_buffer_dirty(buf);
3844         if (!was_dirty)
3845                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3846                                          buf->len,
3847                                          fs_info->dirty_metadata_batch);
3848 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3849         /*
3850          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
3851          * but item data not updated.
3852          * So here we should only check item pointers, not item data.
3853          */
3854         if (btrfs_header_level(buf) == 0 &&
3855             btrfs_check_leaf_relaxed(root, buf)) {
3856                 btrfs_print_leaf(buf);
3857                 ASSERT(0);
3858         }
3859 #endif
3860 }
3861 
3862 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
3863                                         int flush_delayed)
3864 {
3865         /*
3866          * looks as though older kernels can get into trouble with
3867          * this code, they end up stuck in balance_dirty_pages forever
3868          */
3869         int ret;
3870 
3871         if (current->flags & PF_MEMALLOC)
3872                 return;
3873 
3874         if (flush_delayed)
3875                 btrfs_balance_delayed_items(fs_info);
3876 
3877         ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
3878                                      BTRFS_DIRTY_METADATA_THRESH);
3879         if (ret > 0) {
3880                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
3881         }
3882 }
3883 
3884 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
3885 {
3886         __btrfs_btree_balance_dirty(fs_info, 1);
3887 }
3888 
3889 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
3890 {
3891         __btrfs_btree_balance_dirty(fs_info, 0);
3892 }
3893 
3894 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3895 {
3896         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3897         struct btrfs_fs_info *fs_info = root->fs_info;
3898 
3899         return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
3900 }
3901 
3902 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
3903 {
3904         struct btrfs_super_block *sb = fs_info->super_copy;
3905         u64 nodesize = btrfs_super_nodesize(sb);
3906         u64 sectorsize = btrfs_super_sectorsize(sb);
3907         int ret = 0;
3908 
3909         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
3910                 btrfs_err(fs_info, "no valid FS found");
3911                 ret = -EINVAL;
3912         }
3913         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
3914                 btrfs_warn(fs_info, "unrecognized super flag: %llu",
3915                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
3916         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3917                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
3918                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3919                 ret = -EINVAL;
3920         }
3921         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3922                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
3923                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3924                 ret = -EINVAL;
3925         }
3926         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3927                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
3928                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3929                 ret = -EINVAL;
3930         }
3931 
3932         /*
3933          * Check sectorsize and nodesize first, other check will need it.
3934          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
3935          */
3936         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
3937             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3938                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
3939                 ret = -EINVAL;
3940         }
3941         /* Only PAGE SIZE is supported yet */
3942         if (sectorsize != PAGE_SIZE) {
3943                 btrfs_err(fs_info,
3944                         "sectorsize %llu not supported yet, only support %lu",
3945                         sectorsize, PAGE_SIZE);
3946                 ret = -EINVAL;
3947         }
3948         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
3949             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
3950                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
3951                 ret = -EINVAL;
3952         }
3953         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
3954                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
3955                           le32_to_cpu(sb->__unused_leafsize), nodesize);
3956                 ret = -EINVAL;
3957         }
3958 
3959         /* Root alignment check */
3960         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
3961                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
3962                            btrfs_super_root(sb));
3963                 ret = -EINVAL;
3964         }
3965         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
3966                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
3967                            btrfs_super_chunk_root(sb));
3968                 ret = -EINVAL;
3969         }
3970         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
3971                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
3972                            btrfs_super_log_root(sb));
3973                 ret = -EINVAL;
3974         }
3975 
3976         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
3977                 btrfs_err(fs_info,
3978                            "dev_item UUID does not match fsid: %pU != %pU",
3979                            fs_info->fsid, sb->dev_item.fsid);
3980                 ret = -EINVAL;
3981         }
3982 
3983         /*
3984          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3985          * done later
3986          */
3987         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
3988                 btrfs_err(fs_info, "bytes_used is too small %llu",
3989                           btrfs_super_bytes_used(sb));
3990                 ret = -EINVAL;
3991         }
3992         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
3993                 btrfs_err(fs_info, "invalid stripesize %u",
3994                           btrfs_super_stripesize(sb));
3995                 ret = -EINVAL;
3996         }
3997         if (btrfs_super_num_devices(sb) > (1UL << 31))
3998                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
3999                            btrfs_super_num_devices(sb));
4000         if (btrfs_super_num_devices(sb) == 0) {
4001                 btrfs_err(fs_info, "number of devices is 0");
4002                 ret = -EINVAL;
4003         }
4004 
4005         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4006                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4007                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4008                 ret = -EINVAL;
4009         }
4010 
4011         /*
4012          * Obvious sys_chunk_array corruptions, it must hold at least one key
4013          * and one chunk
4014          */
4015         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4016                 btrfs_err(fs_info, "system chunk array too big %u > %u",
4017                           btrfs_super_sys_array_size(sb),
4018                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4019                 ret = -EINVAL;
4020         }
4021         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4022                         + sizeof(struct btrfs_chunk)) {
4023                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4024                           btrfs_super_sys_array_size(sb),
4025                           sizeof(struct btrfs_disk_key)
4026                           + sizeof(struct btrfs_chunk));
4027                 ret = -EINVAL;
4028         }
4029 
4030         /*
4031          * The generation is a global counter, we'll trust it more than the others
4032          * but it's still possible that it's the one that's wrong.
4033          */
4034         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4035                 btrfs_warn(fs_info,
4036                         "suspicious: generation < chunk_root_generation: %llu < %llu",
4037                         btrfs_super_generation(sb),
4038                         btrfs_super_chunk_root_generation(sb));
4039         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4040             && btrfs_super_cache_generation(sb) != (u64)-1)
4041                 btrfs_warn(fs_info,
4042                         "suspicious: generation < cache_generation: %llu < %llu",
4043                         btrfs_super_generation(sb),
4044                         btrfs_super_cache_generation(sb));
4045 
4046         return ret;
4047 }
4048 
4049 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4050 {
4051         mutex_lock(&fs_info->cleaner_mutex);
4052         btrfs_run_delayed_iputs(fs_info);
4053         mutex_unlock(&fs_info->cleaner_mutex);
4054 
4055         down_write(&fs_info->cleanup_work_sem);
4056         up_write(&fs_info->cleanup_work_sem);
4057 
4058         /* cleanup FS via transaction */
4059         btrfs_cleanup_transaction(fs_info);
4060 }
4061 
4062 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4063 {
4064         struct btrfs_ordered_extent *ordered;
4065 
4066         spin_lock(&root->ordered_extent_lock);
4067         /*
4068          * This will just short circuit the ordered completion stuff which will
4069          * make sure the ordered extent gets properly cleaned up.
4070          */
4071         list_for_each_entry(ordered, &root->ordered_extents,
4072                             root_extent_list)
4073                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4074         spin_unlock(&root->ordered_extent_lock);
4075 }
4076 
4077 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4078 {
4079         struct btrfs_root *root;
4080         struct list_head splice;
4081 
4082         INIT_LIST_HEAD(&splice);
4083 
4084         spin_lock(&fs_info->ordered_root_lock);
4085         list_splice_init(&fs_info->ordered_roots, &splice);
4086         while (!list_empty(&splice)) {
4087                 root = list_first_entry(&splice, struct btrfs_root,
4088                                         ordered_root);
4089                 list_move_tail(&root->ordered_root,
4090                                &fs_info->ordered_roots);
4091 
4092                 spin_unlock(&fs_info->ordered_root_lock);
4093                 btrfs_destroy_ordered_extents(root);
4094 
4095                 cond_resched();
4096                 spin_lock(&fs_info->ordered_root_lock);
4097         }
4098         spin_unlock(&fs_info->ordered_root_lock);
4099 }
4100 
4101 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4102                                       struct btrfs_fs_info *fs_info)
4103 {
4104         struct rb_node *node;
4105         struct btrfs_delayed_ref_root *delayed_refs;
4106         struct btrfs_delayed_ref_node *ref;
4107         int ret = 0;
4108 
4109         delayed_refs = &trans->delayed_refs;
4110 
4111         spin_lock(&delayed_refs->lock);
4112         if (atomic_read(&delayed_refs->num_entries) == 0) {
4113                 spin_unlock(&delayed_refs->lock);
4114                 btrfs_info(fs_info, "delayed_refs has NO entry");
4115                 return ret;
4116         }
4117 
4118         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4119                 struct btrfs_delayed_ref_head *head;
4120                 struct rb_node *n;
4121                 bool pin_bytes = false;
4122 
4123                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4124                                 href_node);
4125                 if (!mutex_trylock(&head->mutex)) {
4126                         refcount_inc(&head->refs);
4127                         spin_unlock(&delayed_refs->lock);
4128 
4129                         mutex_lock(&head->mutex);
4130                         mutex_unlock(&head->mutex);
4131                         btrfs_put_delayed_ref_head(head);
4132                         spin_lock(&delayed_refs->lock);
4133                         continue;
4134                 }
4135                 spin_lock(&head->lock);
4136                 while ((n = rb_first(&head->ref_tree)) != NULL) {
4137                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4138                                        ref_node);
4139                         ref->in_tree = 0;
4140                         rb_erase(&ref->ref_node, &head->ref_tree);
4141                         RB_CLEAR_NODE(&ref->ref_node);
4142                         if (!list_empty(&ref->add_list))
4143                                 list_del(&ref->add_list);
4144                         atomic_dec(&delayed_refs->num_entries);
4145                         btrfs_put_delayed_ref(ref);
4146                 }
4147                 if (head->must_insert_reserved)
4148                         pin_bytes = true;
4149                 btrfs_free_delayed_extent_op(head->extent_op);
4150                 delayed_refs->num_heads--;
4151                 if (head->processing == 0)
4152                         delayed_refs->num_heads_ready--;
4153                 atomic_dec(&delayed_refs->num_entries);
4154                 rb_erase(&head->href_node, &delayed_refs->href_root);
4155                 RB_CLEAR_NODE(&head->href_node);
4156                 spin_unlock(&head->lock);
4157                 spin_unlock(&delayed_refs->lock);
4158                 mutex_unlock(&head->mutex);
4159 
4160                 if (pin_bytes)
4161                         btrfs_pin_extent(fs_info, head->bytenr,
4162                                          head->num_bytes, 1);
4163                 btrfs_put_delayed_ref_head(head);
4164                 cond_resched();
4165                 spin_lock(&delayed_refs->lock);
4166         }
4167 
4168         spin_unlock(&delayed_refs->lock);
4169 
4170         return ret;
4171 }
4172 
4173 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4174 {
4175         struct btrfs_inode *btrfs_inode;
4176         struct list_head splice;
4177 
4178         INIT_LIST_HEAD(&splice);
4179 
4180         spin_lock(&root->delalloc_lock);
4181         list_splice_init(&root->delalloc_inodes, &splice);
4182 
4183         while (!list_empty(&splice)) {
4184                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4185                                                delalloc_inodes);
4186 
4187                 list_del_init(&btrfs_inode->delalloc_inodes);
4188                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4189                           &btrfs_inode->runtime_flags);
4190                 spin_unlock(&root->delalloc_lock);
4191 
4192                 btrfs_invalidate_inodes(btrfs_inode->root);
4193 
4194                 spin_lock(&root->delalloc_lock);
4195         }
4196 
4197         spin_unlock(&root->delalloc_lock);
4198 }
4199 
4200 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4201 {
4202         struct btrfs_root *root;
4203         struct list_head splice;
4204 
4205         INIT_LIST_HEAD(&splice);
4206 
4207         spin_lock(&fs_info->delalloc_root_lock);
4208         list_splice_init(&fs_info->delalloc_roots, &splice);
4209         while (!list_empty(&splice)) {
4210                 root = list_first_entry(&splice, struct btrfs_root,
4211                                          delalloc_root);
4212                 list_del_init(&root->delalloc_root);
4213                 root = btrfs_grab_fs_root(root);
4214                 BUG_ON(!root);
4215                 spin_unlock(&fs_info->delalloc_root_lock);
4216 
4217                 btrfs_destroy_delalloc_inodes(root);
4218                 btrfs_put_fs_root(root);
4219 
4220                 spin_lock(&fs_info->delalloc_root_lock);
4221         }
4222         spin_unlock(&fs_info->delalloc_root_lock);
4223 }
4224 
4225 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4226                                         struct extent_io_tree *dirty_pages,
4227                                         int mark)
4228 {
4229         int ret;
4230         struct extent_buffer *eb;
4231         u64 start = 0;
4232         u64 end;
4233 
4234         while (1) {
4235                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4236                                             mark, NULL);
4237                 if (ret)
4238                         break;
4239 
4240                 clear_extent_bits(dirty_pages, start, end, mark);
4241                 while (start <= end) {
4242                         eb = find_extent_buffer(fs_info, start);
4243                         start += fs_info->nodesize;
4244                         if (!eb)
4245                                 continue;
4246                         wait_on_extent_buffer_writeback(eb);
4247 
4248                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4249                                                &eb->bflags))
4250                                 clear_extent_buffer_dirty(eb);
4251                         free_extent_buffer_stale(eb);
4252                 }
4253         }
4254 
4255         return ret;
4256 }
4257 
4258 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4259                                        struct extent_io_tree *pinned_extents)
4260 {
4261         struct extent_io_tree *unpin;
4262         u64 start;
4263         u64 end;
4264         int ret;
4265         bool loop = true;
4266 
4267         unpin = pinned_extents;
4268 again:
4269         while (1) {
4270                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4271                                             EXTENT_DIRTY, NULL);
4272                 if (ret)
4273                         break;
4274 
4275                 clear_extent_dirty(unpin, start, end);
4276                 btrfs_error_unpin_extent_range(fs_info, start, end);
4277                 cond_resched();
4278         }
4279 
4280         if (loop) {
4281                 if (unpin == &fs_info->freed_extents[0])
4282                         unpin = &fs_info->freed_extents[1];
4283                 else
4284                         unpin = &fs_info->freed_extents[0];
4285                 loop = false;
4286                 goto again;
4287         }
4288 
4289         return 0;
4290 }
4291 
4292 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4293 {
4294         struct inode *inode;
4295 
4296         inode = cache->io_ctl.inode;
4297         if (inode) {
4298                 invalidate_inode_pages2(inode->i_mapping);
4299                 BTRFS_I(inode)->generation = 0;
4300                 cache->io_ctl.inode = NULL;
4301                 iput(inode);
4302         }
4303         btrfs_put_block_group(cache);
4304 }
4305 
4306 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4307                              struct btrfs_fs_info *fs_info)
4308 {
4309         struct btrfs_block_group_cache *cache;
4310 
4311         spin_lock(&cur_trans->dirty_bgs_lock);
4312         while (!list_empty(&cur_trans->dirty_bgs)) {
4313                 cache = list_first_entry(&cur_trans->dirty_bgs,
4314                                          struct btrfs_block_group_cache,
4315                                          dirty_list);
4316                 if (!cache) {
4317                         btrfs_err(fs_info, "orphan block group dirty_bgs list");
4318                         spin_unlock(&cur_trans->dirty_bgs_lock);
4319                         return;
4320                 }
4321 
4322                 if (!list_empty(&cache->io_list)) {
4323                         spin_unlock(&cur_trans->dirty_bgs_lock);
4324                         list_del_init(&cache->io_list);
4325                         btrfs_cleanup_bg_io(cache);
4326                         spin_lock(&cur_trans->dirty_bgs_lock);
4327                 }
4328 
4329                 list_del_init(&cache->dirty_list);
4330                 spin_lock(&cache->lock);
4331                 cache->disk_cache_state = BTRFS_DC_ERROR;
4332                 spin_unlock(&cache->lock);
4333 
4334                 spin_unlock(&cur_trans->dirty_bgs_lock);
4335                 btrfs_put_block_group(cache);
4336                 spin_lock(&cur_trans->dirty_bgs_lock);
4337         }
4338         spin_unlock(&cur_trans->dirty_bgs_lock);
4339 
4340         while (!list_empty(&cur_trans->io_bgs)) {
4341                 cache = list_first_entry(&cur_trans->io_bgs,
4342                                          struct btrfs_block_group_cache,
4343                                          io_list);
4344                 if (!cache) {
4345                         btrfs_err(fs_info, "orphan block group on io_bgs list");
4346                         return;
4347                 }
4348 
4349                 list_del_init(&cache->io_list);
4350                 spin_lock(&cache->lock);
4351                 cache->disk_cache_state = BTRFS_DC_ERROR;
4352                 spin_unlock(&cache->lock);
4353                 btrfs_cleanup_bg_io(cache);
4354         }
4355 }
4356 
4357 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4358                                    struct btrfs_fs_info *fs_info)
4359 {
4360         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4361         ASSERT(list_empty(&cur_trans->dirty_bgs));
4362         ASSERT(list_empty(&cur_trans->io_bgs));
4363 
4364         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4365 
4366         cur_trans->state = TRANS_STATE_COMMIT_START;
4367         wake_up(&fs_info->transaction_blocked_wait);
4368 
4369         cur_trans->state = TRANS_STATE_UNBLOCKED;
4370         wake_up(&fs_info->transaction_wait);
4371 
4372         btrfs_destroy_delayed_inodes(fs_info);
4373         btrfs_assert_delayed_root_empty(fs_info);
4374 
4375         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4376                                      EXTENT_DIRTY);
4377         btrfs_destroy_pinned_extent(fs_info,
4378                                     fs_info->pinned_extents);
4379 
4380         cur_trans->state =TRANS_STATE_COMPLETED;
4381         wake_up(&cur_trans->commit_wait);
4382 }
4383 
4384 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4385 {
4386         struct btrfs_transaction *t;
4387 
4388         mutex_lock(&fs_info->transaction_kthread_mutex);
4389 
4390         spin_lock(&fs_info->trans_lock);
4391         while (!list_empty(&fs_info->trans_list)) {
4392                 t = list_first_entry(&fs_info->trans_list,
4393                                      struct btrfs_transaction, list);
4394                 if (t->state >= TRANS_STATE_COMMIT_START) {
4395                         refcount_inc(&t->use_count);
4396                         spin_unlock(&fs_info->trans_lock);
4397                         btrfs_wait_for_commit(fs_info, t->transid);
4398                         btrfs_put_transaction(t);
4399                         spin_lock(&fs_info->trans_lock);
4400                         continue;
4401                 }
4402                 if (t == fs_info->running_transaction) {
4403                         t->state = TRANS_STATE_COMMIT_DOING;
4404                         spin_unlock(&fs_info->trans_lock);
4405                         /*
4406                          * We wait for 0 num_writers since we don't hold a trans
4407                          * handle open currently for this transaction.
4408                          */
4409                         wait_event(t->writer_wait,
4410                                    atomic_read(&t->num_writers) == 0);
4411                 } else {
4412                         spin_unlock(&fs_info->trans_lock);
4413                 }
4414                 btrfs_cleanup_one_transaction(t, fs_info);
4415 
4416                 spin_lock(&fs_info->trans_lock);
4417                 if (t == fs_info->running_transaction)
4418                         fs_info->running_transaction = NULL;
4419                 list_del_init(&t->list);
4420                 spin_unlock(&fs_info->trans_lock);
4421 
4422                 btrfs_put_transaction(t);
4423                 trace_btrfs_transaction_commit(fs_info->tree_root);
4424                 spin_lock(&fs_info->trans_lock);
4425         }
4426         spin_unlock(&fs_info->trans_lock);
4427         btrfs_destroy_all_ordered_extents(fs_info);
4428         btrfs_destroy_delayed_inodes(fs_info);
4429         btrfs_assert_delayed_root_empty(fs_info);
4430         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4431         btrfs_destroy_all_delalloc_inodes(fs_info);
4432         mutex_unlock(&fs_info->transaction_kthread_mutex);
4433 
4434         return 0;
4435 }
4436 
4437 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4438 {
4439         struct inode *inode = private_data;
4440         return btrfs_sb(inode->i_sb);
4441 }
4442 
4443 static const struct extent_io_ops btree_extent_io_ops = {
4444         /* mandatory callbacks */
4445         .submit_bio_hook = btree_submit_bio_hook,
4446         .readpage_end_io_hook = btree_readpage_end_io_hook,
4447         /* note we're sharing with inode.c for the merge bio hook */
4448         .merge_bio_hook = btrfs_merge_bio_hook,
4449         .readpage_io_failed_hook = btree_io_failed_hook,
4450         .set_range_writeback = btrfs_set_range_writeback,
4451         .tree_fs_info = btree_fs_info,
4452 
4453         /* optional callbacks */
4454 };
4455 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp