Btrfs: replace tree->mapping with tree->private_data
[sfrench/cifs-2.6.git] / fs / btrfs / disk-io.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <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
54 #ifdef CONFIG_X86
55 #include <asm/cpufeature.h>
56 #endif
57
58 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
59                                  BTRFS_HEADER_FLAG_RELOC |\
60                                  BTRFS_SUPER_FLAG_ERROR |\
61                                  BTRFS_SUPER_FLAG_SEEDING |\
62                                  BTRFS_SUPER_FLAG_METADUMP)
63
64 static const struct extent_io_ops btree_extent_io_ops;
65 static void end_workqueue_fn(struct btrfs_work *work);
66 static void free_fs_root(struct btrfs_root *root);
67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info);
68 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
69 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
70                                       struct btrfs_fs_info *fs_info);
71 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
72 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
73                                         struct extent_io_tree *dirty_pages,
74                                         int mark);
75 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
76                                        struct extent_io_tree *pinned_extents);
77 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
78 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
79
80 /*
81  * btrfs_end_io_wq structs are used to do processing in task context when an IO
82  * is complete.  This is used during reads to verify checksums, and it is used
83  * by writes to insert metadata for new file extents after IO is complete.
84  */
85 struct btrfs_end_io_wq {
86         struct bio *bio;
87         bio_end_io_t *end_io;
88         void *private;
89         struct btrfs_fs_info *info;
90         int error;
91         enum btrfs_wq_endio_type metadata;
92         struct list_head list;
93         struct btrfs_work work;
94 };
95
96 static struct kmem_cache *btrfs_end_io_wq_cache;
97
98 int __init btrfs_end_io_wq_init(void)
99 {
100         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
101                                         sizeof(struct btrfs_end_io_wq),
102                                         0,
103                                         SLAB_MEM_SPREAD,
104                                         NULL);
105         if (!btrfs_end_io_wq_cache)
106                 return -ENOMEM;
107         return 0;
108 }
109
110 void btrfs_end_io_wq_exit(void)
111 {
112         kmem_cache_destroy(btrfs_end_io_wq_cache);
113 }
114
115 /*
116  * async submit bios are used to offload expensive checksumming
117  * onto the worker threads.  They checksum file and metadata bios
118  * just before they are sent down the IO stack.
119  */
120 struct async_submit_bio {
121         void *private_data;
122         struct btrfs_fs_info *fs_info;
123         struct bio *bio;
124         struct list_head list;
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         int error;
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_UUID_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 #define CORRUPT(reason, eb, root, slot)                                 \
549         btrfs_crit(root->fs_info,                                       \
550                    "corrupt %s, %s: block=%llu, root=%llu, slot=%d",    \
551                    btrfs_header_level(eb) == 0 ? "leaf" : "node",       \
552                    reason, btrfs_header_bytenr(eb), root->objectid, slot)
553
554 static noinline int check_leaf(struct btrfs_root *root,
555                                struct extent_buffer *leaf)
556 {
557         struct btrfs_fs_info *fs_info = root->fs_info;
558         struct btrfs_key key;
559         struct btrfs_key leaf_key;
560         u32 nritems = btrfs_header_nritems(leaf);
561         int slot;
562
563         /*
564          * Extent buffers from a relocation tree have a owner field that
565          * corresponds to the subvolume tree they are based on. So just from an
566          * extent buffer alone we can not find out what is the id of the
567          * corresponding subvolume tree, so we can not figure out if the extent
568          * buffer corresponds to the root of the relocation tree or not. So skip
569          * this check for relocation trees.
570          */
571         if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
572                 struct btrfs_root *check_root;
573
574                 key.objectid = btrfs_header_owner(leaf);
575                 key.type = BTRFS_ROOT_ITEM_KEY;
576                 key.offset = (u64)-1;
577
578                 check_root = btrfs_get_fs_root(fs_info, &key, false);
579                 /*
580                  * The only reason we also check NULL here is that during
581                  * open_ctree() some roots has not yet been set up.
582                  */
583                 if (!IS_ERR_OR_NULL(check_root)) {
584                         struct extent_buffer *eb;
585
586                         eb = btrfs_root_node(check_root);
587                         /* if leaf is the root, then it's fine */
588                         if (leaf != eb) {
589                                 CORRUPT("non-root leaf's nritems is 0",
590                                         leaf, check_root, 0);
591                                 free_extent_buffer(eb);
592                                 return -EIO;
593                         }
594                         free_extent_buffer(eb);
595                 }
596                 return 0;
597         }
598
599         if (nritems == 0)
600                 return 0;
601
602         /* Check the 0 item */
603         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
604             BTRFS_LEAF_DATA_SIZE(fs_info)) {
605                 CORRUPT("invalid item offset size pair", leaf, root, 0);
606                 return -EIO;
607         }
608
609         /*
610          * Check to make sure each items keys are in the correct order and their
611          * offsets make sense.  We only have to loop through nritems-1 because
612          * we check the current slot against the next slot, which verifies the
613          * next slot's offset+size makes sense and that the current's slot
614          * offset is correct.
615          */
616         for (slot = 0; slot < nritems - 1; slot++) {
617                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
618                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
619
620                 /* Make sure the keys are in the right order */
621                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
622                         CORRUPT("bad key order", leaf, root, slot);
623                         return -EIO;
624                 }
625
626                 /*
627                  * Make sure the offset and ends are right, remember that the
628                  * item data starts at the end of the leaf and grows towards the
629                  * front.
630                  */
631                 if (btrfs_item_offset_nr(leaf, slot) !=
632                         btrfs_item_end_nr(leaf, slot + 1)) {
633                         CORRUPT("slot offset bad", leaf, root, slot);
634                         return -EIO;
635                 }
636
637                 /*
638                  * Check to make sure that we don't point outside of the leaf,
639                  * just in case all the items are consistent to each other, but
640                  * all point outside of the leaf.
641                  */
642                 if (btrfs_item_end_nr(leaf, slot) >
643                     BTRFS_LEAF_DATA_SIZE(fs_info)) {
644                         CORRUPT("slot end outside of leaf", leaf, root, slot);
645                         return -EIO;
646                 }
647         }
648
649         return 0;
650 }
651
652 static int check_node(struct btrfs_root *root, struct extent_buffer *node)
653 {
654         unsigned long nr = btrfs_header_nritems(node);
655         struct btrfs_key key, next_key;
656         int slot;
657         u64 bytenr;
658         int ret = 0;
659
660         if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(root->fs_info)) {
661                 btrfs_crit(root->fs_info,
662                            "corrupt node: block %llu root %llu nritems %lu",
663                            node->start, root->objectid, nr);
664                 return -EIO;
665         }
666
667         for (slot = 0; slot < nr - 1; slot++) {
668                 bytenr = btrfs_node_blockptr(node, slot);
669                 btrfs_node_key_to_cpu(node, &key, slot);
670                 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
671
672                 if (!bytenr) {
673                         CORRUPT("invalid item slot", node, root, slot);
674                         ret = -EIO;
675                         goto out;
676                 }
677
678                 if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
679                         CORRUPT("bad key order", node, root, slot);
680                         ret = -EIO;
681                         goto out;
682                 }
683         }
684 out:
685         return ret;
686 }
687
688 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
689                                       u64 phy_offset, struct page *page,
690                                       u64 start, u64 end, int mirror)
691 {
692         u64 found_start;
693         int found_level;
694         struct extent_buffer *eb;
695         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
696         struct btrfs_fs_info *fs_info = root->fs_info;
697         int ret = 0;
698         int reads_done;
699
700         if (!page->private)
701                 goto out;
702
703         eb = (struct extent_buffer *)page->private;
704
705         /* the pending IO might have been the only thing that kept this buffer
706          * in memory.  Make sure we have a ref for all this other checks
707          */
708         extent_buffer_get(eb);
709
710         reads_done = atomic_dec_and_test(&eb->io_pages);
711         if (!reads_done)
712                 goto err;
713
714         eb->read_mirror = mirror;
715         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
716                 ret = -EIO;
717                 goto err;
718         }
719
720         found_start = btrfs_header_bytenr(eb);
721         if (found_start != eb->start) {
722                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
723                              found_start, eb->start);
724                 ret = -EIO;
725                 goto err;
726         }
727         if (check_tree_block_fsid(fs_info, eb)) {
728                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
729                              eb->start);
730                 ret = -EIO;
731                 goto err;
732         }
733         found_level = btrfs_header_level(eb);
734         if (found_level >= BTRFS_MAX_LEVEL) {
735                 btrfs_err(fs_info, "bad tree block level %d",
736                           (int)btrfs_header_level(eb));
737                 ret = -EIO;
738                 goto err;
739         }
740
741         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
742                                        eb, found_level);
743
744         ret = csum_tree_block(fs_info, eb, 1);
745         if (ret)
746                 goto err;
747
748         /*
749          * If this is a leaf block and it is corrupt, set the corrupt bit so
750          * that we don't try and read the other copies of this block, just
751          * return -EIO.
752          */
753         if (found_level == 0 && check_leaf(root, eb)) {
754                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
755                 ret = -EIO;
756         }
757
758         if (found_level > 0 && check_node(root, eb))
759                 ret = -EIO;
760
761         if (!ret)
762                 set_extent_buffer_uptodate(eb);
763 err:
764         if (reads_done &&
765             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
766                 btree_readahead_hook(eb, ret);
767
768         if (ret) {
769                 /*
770                  * our io error hook is going to dec the io pages
771                  * again, we have to make sure it has something
772                  * to decrement
773                  */
774                 atomic_inc(&eb->io_pages);
775                 clear_extent_buffer_uptodate(eb);
776         }
777         free_extent_buffer(eb);
778 out:
779         return ret;
780 }
781
782 static int btree_io_failed_hook(struct page *page, int failed_mirror)
783 {
784         struct extent_buffer *eb;
785
786         eb = (struct extent_buffer *)page->private;
787         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
788         eb->read_mirror = failed_mirror;
789         atomic_dec(&eb->io_pages);
790         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
791                 btree_readahead_hook(eb, -EIO);
792         return -EIO;    /* we fixed nothing */
793 }
794
795 static void end_workqueue_bio(struct bio *bio)
796 {
797         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
798         struct btrfs_fs_info *fs_info;
799         struct btrfs_workqueue *wq;
800         btrfs_work_func_t func;
801
802         fs_info = end_io_wq->info;
803         end_io_wq->error = bio->bi_error;
804
805         if (bio_op(bio) == REQ_OP_WRITE) {
806                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
807                         wq = fs_info->endio_meta_write_workers;
808                         func = btrfs_endio_meta_write_helper;
809                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
810                         wq = fs_info->endio_freespace_worker;
811                         func = btrfs_freespace_write_helper;
812                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
813                         wq = fs_info->endio_raid56_workers;
814                         func = btrfs_endio_raid56_helper;
815                 } else {
816                         wq = fs_info->endio_write_workers;
817                         func = btrfs_endio_write_helper;
818                 }
819         } else {
820                 if (unlikely(end_io_wq->metadata ==
821                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
822                         wq = fs_info->endio_repair_workers;
823                         func = btrfs_endio_repair_helper;
824                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
825                         wq = fs_info->endio_raid56_workers;
826                         func = btrfs_endio_raid56_helper;
827                 } else if (end_io_wq->metadata) {
828                         wq = fs_info->endio_meta_workers;
829                         func = btrfs_endio_meta_helper;
830                 } else {
831                         wq = fs_info->endio_workers;
832                         func = btrfs_endio_helper;
833                 }
834         }
835
836         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
837         btrfs_queue_work(wq, &end_io_wq->work);
838 }
839
840 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
841                         enum btrfs_wq_endio_type metadata)
842 {
843         struct btrfs_end_io_wq *end_io_wq;
844
845         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
846         if (!end_io_wq)
847                 return -ENOMEM;
848
849         end_io_wq->private = bio->bi_private;
850         end_io_wq->end_io = bio->bi_end_io;
851         end_io_wq->info = info;
852         end_io_wq->error = 0;
853         end_io_wq->bio = bio;
854         end_io_wq->metadata = metadata;
855
856         bio->bi_private = end_io_wq;
857         bio->bi_end_io = end_workqueue_bio;
858         return 0;
859 }
860
861 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
862 {
863         unsigned long limit = min_t(unsigned long,
864                                     info->thread_pool_size,
865                                     info->fs_devices->open_devices);
866         return 256 * limit;
867 }
868
869 static void run_one_async_start(struct btrfs_work *work)
870 {
871         struct async_submit_bio *async;
872         int ret;
873
874         async = container_of(work, struct  async_submit_bio, work);
875         ret = async->submit_bio_start(async->private_data, async->bio,
876                                       async->mirror_num, async->bio_flags,
877                                       async->bio_offset);
878         if (ret)
879                 async->error = ret;
880 }
881
882 static void run_one_async_done(struct btrfs_work *work)
883 {
884         struct btrfs_fs_info *fs_info;
885         struct async_submit_bio *async;
886         int limit;
887
888         async = container_of(work, struct  async_submit_bio, work);
889         fs_info = async->fs_info;
890
891         limit = btrfs_async_submit_limit(fs_info);
892         limit = limit * 2 / 3;
893
894         /*
895          * atomic_dec_return implies a barrier for waitqueue_active
896          */
897         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
898             waitqueue_active(&fs_info->async_submit_wait))
899                 wake_up(&fs_info->async_submit_wait);
900
901         /* If an error occurred we just want to clean up the bio and move on */
902         if (async->error) {
903                 async->bio->bi_error = async->error;
904                 bio_endio(async->bio);
905                 return;
906         }
907
908         async->submit_bio_done(async->private_data, async->bio, async->mirror_num,
909                                async->bio_flags, async->bio_offset);
910 }
911
912 static void run_one_async_free(struct btrfs_work *work)
913 {
914         struct async_submit_bio *async;
915
916         async = container_of(work, struct  async_submit_bio, work);
917         kfree(async);
918 }
919
920 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
921                         int mirror_num, unsigned long bio_flags,
922                         u64 bio_offset, void *private_data,
923                         extent_submit_bio_hook_t *submit_bio_start,
924                         extent_submit_bio_hook_t *submit_bio_done)
925 {
926         struct async_submit_bio *async;
927
928         async = kmalloc(sizeof(*async), GFP_NOFS);
929         if (!async)
930                 return -ENOMEM;
931
932         async->private_data = private_data;
933         async->fs_info = fs_info;
934         async->bio = bio;
935         async->mirror_num = mirror_num;
936         async->submit_bio_start = submit_bio_start;
937         async->submit_bio_done = submit_bio_done;
938
939         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
940                         run_one_async_done, run_one_async_free);
941
942         async->bio_flags = bio_flags;
943         async->bio_offset = bio_offset;
944
945         async->error = 0;
946
947         atomic_inc(&fs_info->nr_async_submits);
948
949         if (op_is_sync(bio->bi_opf))
950                 btrfs_set_work_high_priority(&async->work);
951
952         btrfs_queue_work(fs_info->workers, &async->work);
953
954         while (atomic_read(&fs_info->async_submit_draining) &&
955               atomic_read(&fs_info->nr_async_submits)) {
956                 wait_event(fs_info->async_submit_wait,
957                            (atomic_read(&fs_info->nr_async_submits) == 0));
958         }
959
960         return 0;
961 }
962
963 static int btree_csum_one_bio(struct bio *bio)
964 {
965         struct bio_vec *bvec;
966         struct btrfs_root *root;
967         int i, ret = 0;
968
969         bio_for_each_segment_all(bvec, bio, i) {
970                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
971                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
972                 if (ret)
973                         break;
974         }
975
976         return ret;
977 }
978
979 static int __btree_submit_bio_start(void *private_data, struct bio *bio,
980                                     int mirror_num, unsigned long bio_flags,
981                                     u64 bio_offset)
982 {
983         /*
984          * when we're called for a write, we're already in the async
985          * submission context.  Just jump into btrfs_map_bio
986          */
987         return btree_csum_one_bio(bio);
988 }
989
990 static int __btree_submit_bio_done(void *private_data, struct bio *bio,
991                                  int mirror_num, unsigned long bio_flags,
992                                  u64 bio_offset)
993 {
994         struct inode *inode = private_data;
995         int ret;
996
997         /*
998          * when we're called for a write, we're already in the async
999          * submission context.  Just jump into btrfs_map_bio
1000          */
1001         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
1002         if (ret) {
1003                 bio->bi_error = ret;
1004                 bio_endio(bio);
1005         }
1006         return ret;
1007 }
1008
1009 static int check_async_write(unsigned long bio_flags)
1010 {
1011         if (bio_flags & EXTENT_BIO_TREE_LOG)
1012                 return 0;
1013 #ifdef CONFIG_X86
1014         if (static_cpu_has(X86_FEATURE_XMM4_2))
1015                 return 0;
1016 #endif
1017         return 1;
1018 }
1019
1020 static int btree_submit_bio_hook(void *private_data, struct bio *bio,
1021                                  int mirror_num, unsigned long bio_flags,
1022                                  u64 bio_offset)
1023 {
1024         struct inode *inode = private_data;
1025         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1026         int async = check_async_write(bio_flags);
1027         int ret;
1028
1029         if (bio_op(bio) != REQ_OP_WRITE) {
1030                 /*
1031                  * called for a read, do the setup so that checksum validation
1032                  * can happen in the async kernel threads
1033                  */
1034                 ret = btrfs_bio_wq_end_io(fs_info, bio,
1035                                           BTRFS_WQ_ENDIO_METADATA);
1036                 if (ret)
1037                         goto out_w_error;
1038                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1039         } else if (!async) {
1040                 ret = btree_csum_one_bio(bio);
1041                 if (ret)
1042                         goto out_w_error;
1043                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1044         } else {
1045                 /*
1046                  * kthread helpers are used to submit writes so that
1047                  * checksumming can happen in parallel across all CPUs
1048                  */
1049                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
1050                                           bio_offset, private_data,
1051                                           __btree_submit_bio_start,
1052                                           __btree_submit_bio_done);
1053         }
1054
1055         if (ret)
1056                 goto out_w_error;
1057         return 0;
1058
1059 out_w_error:
1060         bio->bi_error = ret;
1061         bio_endio(bio);
1062         return ret;
1063 }
1064
1065 #ifdef CONFIG_MIGRATION
1066 static int btree_migratepage(struct address_space *mapping,
1067                         struct page *newpage, struct page *page,
1068                         enum migrate_mode mode)
1069 {
1070         /*
1071          * we can't safely write a btree page from here,
1072          * we haven't done the locking hook
1073          */
1074         if (PageDirty(page))
1075                 return -EAGAIN;
1076         /*
1077          * Buffers may be managed in a filesystem specific way.
1078          * We must have no buffers or drop them.
1079          */
1080         if (page_has_private(page) &&
1081             !try_to_release_page(page, GFP_KERNEL))
1082                 return -EAGAIN;
1083         return migrate_page(mapping, newpage, page, mode);
1084 }
1085 #endif
1086
1087
1088 static int btree_writepages(struct address_space *mapping,
1089                             struct writeback_control *wbc)
1090 {
1091         struct btrfs_fs_info *fs_info;
1092         int ret;
1093
1094         if (wbc->sync_mode == WB_SYNC_NONE) {
1095
1096                 if (wbc->for_kupdate)
1097                         return 0;
1098
1099                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1100                 /* this is a bit racy, but that's ok */
1101                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1102                                              BTRFS_DIRTY_METADATA_THRESH);
1103                 if (ret < 0)
1104                         return 0;
1105         }
1106         return btree_write_cache_pages(mapping, wbc);
1107 }
1108
1109 static int btree_readpage(struct file *file, struct page *page)
1110 {
1111         struct extent_io_tree *tree;
1112         tree = &BTRFS_I(page->mapping->host)->io_tree;
1113         return extent_read_full_page(tree, page, btree_get_extent, 0);
1114 }
1115
1116 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1117 {
1118         if (PageWriteback(page) || PageDirty(page))
1119                 return 0;
1120
1121         return try_release_extent_buffer(page);
1122 }
1123
1124 static void btree_invalidatepage(struct page *page, unsigned int offset,
1125                                  unsigned int length)
1126 {
1127         struct extent_io_tree *tree;
1128         tree = &BTRFS_I(page->mapping->host)->io_tree;
1129         extent_invalidatepage(tree, page, offset);
1130         btree_releasepage(page, GFP_NOFS);
1131         if (PagePrivate(page)) {
1132                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1133                            "page private not zero on page %llu",
1134                            (unsigned long long)page_offset(page));
1135                 ClearPagePrivate(page);
1136                 set_page_private(page, 0);
1137                 put_page(page);
1138         }
1139 }
1140
1141 static int btree_set_page_dirty(struct page *page)
1142 {
1143 #ifdef DEBUG
1144         struct extent_buffer *eb;
1145
1146         BUG_ON(!PagePrivate(page));
1147         eb = (struct extent_buffer *)page->private;
1148         BUG_ON(!eb);
1149         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1150         BUG_ON(!atomic_read(&eb->refs));
1151         btrfs_assert_tree_locked(eb);
1152 #endif
1153         return __set_page_dirty_nobuffers(page);
1154 }
1155
1156 static const struct address_space_operations btree_aops = {
1157         .readpage       = btree_readpage,
1158         .writepages     = btree_writepages,
1159         .releasepage    = btree_releasepage,
1160         .invalidatepage = btree_invalidatepage,
1161 #ifdef CONFIG_MIGRATION
1162         .migratepage    = btree_migratepage,
1163 #endif
1164         .set_page_dirty = btree_set_page_dirty,
1165 };
1166
1167 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1168 {
1169         struct extent_buffer *buf = NULL;
1170         struct inode *btree_inode = fs_info->btree_inode;
1171
1172         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1173         if (IS_ERR(buf))
1174                 return;
1175         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1176                                  buf, WAIT_NONE, btree_get_extent, 0);
1177         free_extent_buffer(buf);
1178 }
1179
1180 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1181                          int mirror_num, struct extent_buffer **eb)
1182 {
1183         struct extent_buffer *buf = NULL;
1184         struct inode *btree_inode = fs_info->btree_inode;
1185         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1186         int ret;
1187
1188         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1189         if (IS_ERR(buf))
1190                 return 0;
1191
1192         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1193
1194         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1195                                        btree_get_extent, mirror_num);
1196         if (ret) {
1197                 free_extent_buffer(buf);
1198                 return ret;
1199         }
1200
1201         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1202                 free_extent_buffer(buf);
1203                 return -EIO;
1204         } else if (extent_buffer_uptodate(buf)) {
1205                 *eb = buf;
1206         } else {
1207                 free_extent_buffer(buf);
1208         }
1209         return 0;
1210 }
1211
1212 struct extent_buffer *btrfs_find_create_tree_block(
1213                                                 struct btrfs_fs_info *fs_info,
1214                                                 u64 bytenr)
1215 {
1216         if (btrfs_is_testing(fs_info))
1217                 return alloc_test_extent_buffer(fs_info, bytenr);
1218         return alloc_extent_buffer(fs_info, bytenr);
1219 }
1220
1221
1222 int btrfs_write_tree_block(struct extent_buffer *buf)
1223 {
1224         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1225                                         buf->start + buf->len - 1);
1226 }
1227
1228 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1229 {
1230         return filemap_fdatawait_range(buf->pages[0]->mapping,
1231                                        buf->start, buf->start + buf->len - 1);
1232 }
1233
1234 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1235                                       u64 parent_transid)
1236 {
1237         struct extent_buffer *buf = NULL;
1238         int ret;
1239
1240         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1241         if (IS_ERR(buf))
1242                 return buf;
1243
1244         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1245         if (ret) {
1246                 free_extent_buffer(buf);
1247                 return ERR_PTR(ret);
1248         }
1249         return buf;
1250
1251 }
1252
1253 void clean_tree_block(struct btrfs_fs_info *fs_info,
1254                       struct extent_buffer *buf)
1255 {
1256         if (btrfs_header_generation(buf) ==
1257             fs_info->running_transaction->transid) {
1258                 btrfs_assert_tree_locked(buf);
1259
1260                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1261                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1262                                              -buf->len,
1263                                              fs_info->dirty_metadata_batch);
1264                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1265                         btrfs_set_lock_blocking(buf);
1266                         clear_extent_buffer_dirty(buf);
1267                 }
1268         }
1269 }
1270
1271 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1272 {
1273         struct btrfs_subvolume_writers *writers;
1274         int ret;
1275
1276         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1277         if (!writers)
1278                 return ERR_PTR(-ENOMEM);
1279
1280         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1281         if (ret < 0) {
1282                 kfree(writers);
1283                 return ERR_PTR(ret);
1284         }
1285
1286         init_waitqueue_head(&writers->wait);
1287         return writers;
1288 }
1289
1290 static void
1291 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1292 {
1293         percpu_counter_destroy(&writers->counter);
1294         kfree(writers);
1295 }
1296
1297 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1298                          u64 objectid)
1299 {
1300         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1301         root->node = NULL;
1302         root->commit_root = NULL;
1303         root->state = 0;
1304         root->orphan_cleanup_state = 0;
1305
1306         root->objectid = objectid;
1307         root->last_trans = 0;
1308         root->highest_objectid = 0;
1309         root->nr_delalloc_inodes = 0;
1310         root->nr_ordered_extents = 0;
1311         root->name = NULL;
1312         root->inode_tree = RB_ROOT;
1313         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1314         root->block_rsv = NULL;
1315         root->orphan_block_rsv = NULL;
1316
1317         INIT_LIST_HEAD(&root->dirty_list);
1318         INIT_LIST_HEAD(&root->root_list);
1319         INIT_LIST_HEAD(&root->delalloc_inodes);
1320         INIT_LIST_HEAD(&root->delalloc_root);
1321         INIT_LIST_HEAD(&root->ordered_extents);
1322         INIT_LIST_HEAD(&root->ordered_root);
1323         INIT_LIST_HEAD(&root->logged_list[0]);
1324         INIT_LIST_HEAD(&root->logged_list[1]);
1325         spin_lock_init(&root->orphan_lock);
1326         spin_lock_init(&root->inode_lock);
1327         spin_lock_init(&root->delalloc_lock);
1328         spin_lock_init(&root->ordered_extent_lock);
1329         spin_lock_init(&root->accounting_lock);
1330         spin_lock_init(&root->log_extents_lock[0]);
1331         spin_lock_init(&root->log_extents_lock[1]);
1332         mutex_init(&root->objectid_mutex);
1333         mutex_init(&root->log_mutex);
1334         mutex_init(&root->ordered_extent_mutex);
1335         mutex_init(&root->delalloc_mutex);
1336         init_waitqueue_head(&root->log_writer_wait);
1337         init_waitqueue_head(&root->log_commit_wait[0]);
1338         init_waitqueue_head(&root->log_commit_wait[1]);
1339         INIT_LIST_HEAD(&root->log_ctxs[0]);
1340         INIT_LIST_HEAD(&root->log_ctxs[1]);
1341         atomic_set(&root->log_commit[0], 0);
1342         atomic_set(&root->log_commit[1], 0);
1343         atomic_set(&root->log_writers, 0);
1344         atomic_set(&root->log_batch, 0);
1345         atomic_set(&root->orphan_inodes, 0);
1346         refcount_set(&root->refs, 1);
1347         atomic_set(&root->will_be_snapshoted, 0);
1348         atomic64_set(&root->qgroup_meta_rsv, 0);
1349         root->log_transid = 0;
1350         root->log_transid_committed = -1;
1351         root->last_log_commit = 0;
1352         if (!dummy)
1353                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1354
1355         memset(&root->root_key, 0, sizeof(root->root_key));
1356         memset(&root->root_item, 0, sizeof(root->root_item));
1357         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1358         if (!dummy)
1359                 root->defrag_trans_start = fs_info->generation;
1360         else
1361                 root->defrag_trans_start = 0;
1362         root->root_key.objectid = objectid;
1363         root->anon_dev = 0;
1364
1365         spin_lock_init(&root->root_item_lock);
1366 }
1367
1368 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1369                 gfp_t flags)
1370 {
1371         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1372         if (root)
1373                 root->fs_info = fs_info;
1374         return root;
1375 }
1376
1377 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1378 /* Should only be used by the testing infrastructure */
1379 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1380 {
1381         struct btrfs_root *root;
1382
1383         if (!fs_info)
1384                 return ERR_PTR(-EINVAL);
1385
1386         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1387         if (!root)
1388                 return ERR_PTR(-ENOMEM);
1389
1390         /* We don't use the stripesize in selftest, set it as sectorsize */
1391         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1392         root->alloc_bytenr = 0;
1393
1394         return root;
1395 }
1396 #endif
1397
1398 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1399                                      struct btrfs_fs_info *fs_info,
1400                                      u64 objectid)
1401 {
1402         struct extent_buffer *leaf;
1403         struct btrfs_root *tree_root = fs_info->tree_root;
1404         struct btrfs_root *root;
1405         struct btrfs_key key;
1406         int ret = 0;
1407         uuid_le uuid;
1408
1409         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1410         if (!root)
1411                 return ERR_PTR(-ENOMEM);
1412
1413         __setup_root(root, fs_info, objectid);
1414         root->root_key.objectid = objectid;
1415         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1416         root->root_key.offset = 0;
1417
1418         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1419         if (IS_ERR(leaf)) {
1420                 ret = PTR_ERR(leaf);
1421                 leaf = NULL;
1422                 goto fail;
1423         }
1424
1425         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1426         btrfs_set_header_bytenr(leaf, leaf->start);
1427         btrfs_set_header_generation(leaf, trans->transid);
1428         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1429         btrfs_set_header_owner(leaf, objectid);
1430         root->node = leaf;
1431
1432         write_extent_buffer_fsid(leaf, fs_info->fsid);
1433         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1434         btrfs_mark_buffer_dirty(leaf);
1435
1436         root->commit_root = btrfs_root_node(root);
1437         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1438
1439         root->root_item.flags = 0;
1440         root->root_item.byte_limit = 0;
1441         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1442         btrfs_set_root_generation(&root->root_item, trans->transid);
1443         btrfs_set_root_level(&root->root_item, 0);
1444         btrfs_set_root_refs(&root->root_item, 1);
1445         btrfs_set_root_used(&root->root_item, leaf->len);
1446         btrfs_set_root_last_snapshot(&root->root_item, 0);
1447         btrfs_set_root_dirid(&root->root_item, 0);
1448         uuid_le_gen(&uuid);
1449         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1450         root->root_item.drop_level = 0;
1451
1452         key.objectid = objectid;
1453         key.type = BTRFS_ROOT_ITEM_KEY;
1454         key.offset = 0;
1455         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1456         if (ret)
1457                 goto fail;
1458
1459         btrfs_tree_unlock(leaf);
1460
1461         return root;
1462
1463 fail:
1464         if (leaf) {
1465                 btrfs_tree_unlock(leaf);
1466                 free_extent_buffer(root->commit_root);
1467                 free_extent_buffer(leaf);
1468         }
1469         kfree(root);
1470
1471         return ERR_PTR(ret);
1472 }
1473
1474 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1475                                          struct btrfs_fs_info *fs_info)
1476 {
1477         struct btrfs_root *root;
1478         struct extent_buffer *leaf;
1479
1480         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1481         if (!root)
1482                 return ERR_PTR(-ENOMEM);
1483
1484         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1485
1486         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1487         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1488         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1489
1490         /*
1491          * DON'T set REF_COWS for log trees
1492          *
1493          * log trees do not get reference counted because they go away
1494          * before a real commit is actually done.  They do store pointers
1495          * to file data extents, and those reference counts still get
1496          * updated (along with back refs to the log tree).
1497          */
1498
1499         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1500                         NULL, 0, 0, 0);
1501         if (IS_ERR(leaf)) {
1502                 kfree(root);
1503                 return ERR_CAST(leaf);
1504         }
1505
1506         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1507         btrfs_set_header_bytenr(leaf, leaf->start);
1508         btrfs_set_header_generation(leaf, trans->transid);
1509         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1510         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1511         root->node = leaf;
1512
1513         write_extent_buffer_fsid(root->node, fs_info->fsid);
1514         btrfs_mark_buffer_dirty(root->node);
1515         btrfs_tree_unlock(root->node);
1516         return root;
1517 }
1518
1519 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1520                              struct btrfs_fs_info *fs_info)
1521 {
1522         struct btrfs_root *log_root;
1523
1524         log_root = alloc_log_tree(trans, fs_info);
1525         if (IS_ERR(log_root))
1526                 return PTR_ERR(log_root);
1527         WARN_ON(fs_info->log_root_tree);
1528         fs_info->log_root_tree = log_root;
1529         return 0;
1530 }
1531
1532 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1533                        struct btrfs_root *root)
1534 {
1535         struct btrfs_fs_info *fs_info = root->fs_info;
1536         struct btrfs_root *log_root;
1537         struct btrfs_inode_item *inode_item;
1538
1539         log_root = alloc_log_tree(trans, fs_info);
1540         if (IS_ERR(log_root))
1541                 return PTR_ERR(log_root);
1542
1543         log_root->last_trans = trans->transid;
1544         log_root->root_key.offset = root->root_key.objectid;
1545
1546         inode_item = &log_root->root_item.inode;
1547         btrfs_set_stack_inode_generation(inode_item, 1);
1548         btrfs_set_stack_inode_size(inode_item, 3);
1549         btrfs_set_stack_inode_nlink(inode_item, 1);
1550         btrfs_set_stack_inode_nbytes(inode_item,
1551                                      fs_info->nodesize);
1552         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1553
1554         btrfs_set_root_node(&log_root->root_item, log_root->node);
1555
1556         WARN_ON(root->log_root);
1557         root->log_root = log_root;
1558         root->log_transid = 0;
1559         root->log_transid_committed = -1;
1560         root->last_log_commit = 0;
1561         return 0;
1562 }
1563
1564 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1565                                                struct btrfs_key *key)
1566 {
1567         struct btrfs_root *root;
1568         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1569         struct btrfs_path *path;
1570         u64 generation;
1571         int ret;
1572
1573         path = btrfs_alloc_path();
1574         if (!path)
1575                 return ERR_PTR(-ENOMEM);
1576
1577         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1578         if (!root) {
1579                 ret = -ENOMEM;
1580                 goto alloc_fail;
1581         }
1582
1583         __setup_root(root, fs_info, key->objectid);
1584
1585         ret = btrfs_find_root(tree_root, key, path,
1586                               &root->root_item, &root->root_key);
1587         if (ret) {
1588                 if (ret > 0)
1589                         ret = -ENOENT;
1590                 goto find_fail;
1591         }
1592
1593         generation = btrfs_root_generation(&root->root_item);
1594         root->node = read_tree_block(fs_info,
1595                                      btrfs_root_bytenr(&root->root_item),
1596                                      generation);
1597         if (IS_ERR(root->node)) {
1598                 ret = PTR_ERR(root->node);
1599                 goto find_fail;
1600         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1601                 ret = -EIO;
1602                 free_extent_buffer(root->node);
1603                 goto find_fail;
1604         }
1605         root->commit_root = btrfs_root_node(root);
1606 out:
1607         btrfs_free_path(path);
1608         return root;
1609
1610 find_fail:
1611         kfree(root);
1612 alloc_fail:
1613         root = ERR_PTR(ret);
1614         goto out;
1615 }
1616
1617 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1618                                       struct btrfs_key *location)
1619 {
1620         struct btrfs_root *root;
1621
1622         root = btrfs_read_tree_root(tree_root, location);
1623         if (IS_ERR(root))
1624                 return root;
1625
1626         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1627                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1628                 btrfs_check_and_init_root_item(&root->root_item);
1629         }
1630
1631         return root;
1632 }
1633
1634 int btrfs_init_fs_root(struct btrfs_root *root)
1635 {
1636         int ret;
1637         struct btrfs_subvolume_writers *writers;
1638
1639         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1640         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1641                                         GFP_NOFS);
1642         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1643                 ret = -ENOMEM;
1644                 goto fail;
1645         }
1646
1647         writers = btrfs_alloc_subvolume_writers();
1648         if (IS_ERR(writers)) {
1649                 ret = PTR_ERR(writers);
1650                 goto fail;
1651         }
1652         root->subv_writers = writers;
1653
1654         btrfs_init_free_ino_ctl(root);
1655         spin_lock_init(&root->ino_cache_lock);
1656         init_waitqueue_head(&root->ino_cache_wait);
1657
1658         ret = get_anon_bdev(&root->anon_dev);
1659         if (ret)
1660                 goto fail;
1661
1662         mutex_lock(&root->objectid_mutex);
1663         ret = btrfs_find_highest_objectid(root,
1664                                         &root->highest_objectid);
1665         if (ret) {
1666                 mutex_unlock(&root->objectid_mutex);
1667                 goto fail;
1668         }
1669
1670         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1671
1672         mutex_unlock(&root->objectid_mutex);
1673
1674         return 0;
1675 fail:
1676         /* the caller is responsible to call free_fs_root */
1677         return ret;
1678 }
1679
1680 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1681                                         u64 root_id)
1682 {
1683         struct btrfs_root *root;
1684
1685         spin_lock(&fs_info->fs_roots_radix_lock);
1686         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1687                                  (unsigned long)root_id);
1688         spin_unlock(&fs_info->fs_roots_radix_lock);
1689         return root;
1690 }
1691
1692 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1693                          struct btrfs_root *root)
1694 {
1695         int ret;
1696
1697         ret = radix_tree_preload(GFP_NOFS);
1698         if (ret)
1699                 return ret;
1700
1701         spin_lock(&fs_info->fs_roots_radix_lock);
1702         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1703                                 (unsigned long)root->root_key.objectid,
1704                                 root);
1705         if (ret == 0)
1706                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1707         spin_unlock(&fs_info->fs_roots_radix_lock);
1708         radix_tree_preload_end();
1709
1710         return ret;
1711 }
1712
1713 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1714                                      struct btrfs_key *location,
1715                                      bool check_ref)
1716 {
1717         struct btrfs_root *root;
1718         struct btrfs_path *path;
1719         struct btrfs_key key;
1720         int ret;
1721
1722         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1723                 return fs_info->tree_root;
1724         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1725                 return fs_info->extent_root;
1726         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1727                 return fs_info->chunk_root;
1728         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1729                 return fs_info->dev_root;
1730         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1731                 return fs_info->csum_root;
1732         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1733                 return fs_info->quota_root ? fs_info->quota_root :
1734                                              ERR_PTR(-ENOENT);
1735         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1736                 return fs_info->uuid_root ? fs_info->uuid_root :
1737                                             ERR_PTR(-ENOENT);
1738         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1739                 return fs_info->free_space_root ? fs_info->free_space_root :
1740                                                   ERR_PTR(-ENOENT);
1741 again:
1742         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1743         if (root) {
1744                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1745                         return ERR_PTR(-ENOENT);
1746                 return root;
1747         }
1748
1749         root = btrfs_read_fs_root(fs_info->tree_root, location);
1750         if (IS_ERR(root))
1751                 return root;
1752
1753         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1754                 ret = -ENOENT;
1755                 goto fail;
1756         }
1757
1758         ret = btrfs_init_fs_root(root);
1759         if (ret)
1760                 goto fail;
1761
1762         path = btrfs_alloc_path();
1763         if (!path) {
1764                 ret = -ENOMEM;
1765                 goto fail;
1766         }
1767         key.objectid = BTRFS_ORPHAN_OBJECTID;
1768         key.type = BTRFS_ORPHAN_ITEM_KEY;
1769         key.offset = location->objectid;
1770
1771         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1772         btrfs_free_path(path);
1773         if (ret < 0)
1774                 goto fail;
1775         if (ret == 0)
1776                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1777
1778         ret = btrfs_insert_fs_root(fs_info, root);
1779         if (ret) {
1780                 if (ret == -EEXIST) {
1781                         free_fs_root(root);
1782                         goto again;
1783                 }
1784                 goto fail;
1785         }
1786         return root;
1787 fail:
1788         free_fs_root(root);
1789         return ERR_PTR(ret);
1790 }
1791
1792 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1793 {
1794         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1795         int ret = 0;
1796         struct btrfs_device *device;
1797         struct backing_dev_info *bdi;
1798
1799         rcu_read_lock();
1800         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1801                 if (!device->bdev)
1802                         continue;
1803                 bdi = device->bdev->bd_bdi;
1804                 if (bdi_congested(bdi, bdi_bits)) {
1805                         ret = 1;
1806                         break;
1807                 }
1808         }
1809         rcu_read_unlock();
1810         return ret;
1811 }
1812
1813 /*
1814  * called by the kthread helper functions to finally call the bio end_io
1815  * functions.  This is where read checksum verification actually happens
1816  */
1817 static void end_workqueue_fn(struct btrfs_work *work)
1818 {
1819         struct bio *bio;
1820         struct btrfs_end_io_wq *end_io_wq;
1821
1822         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1823         bio = end_io_wq->bio;
1824
1825         bio->bi_error = end_io_wq->error;
1826         bio->bi_private = end_io_wq->private;
1827         bio->bi_end_io = end_io_wq->end_io;
1828         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1829         bio_endio(bio);
1830 }
1831
1832 static int cleaner_kthread(void *arg)
1833 {
1834         struct btrfs_root *root = arg;
1835         struct btrfs_fs_info *fs_info = root->fs_info;
1836         int again;
1837         struct btrfs_trans_handle *trans;
1838
1839         do {
1840                 again = 0;
1841
1842                 /* Make the cleaner go to sleep early. */
1843                 if (btrfs_need_cleaner_sleep(fs_info))
1844                         goto sleep;
1845
1846                 /*
1847                  * Do not do anything if we might cause open_ctree() to block
1848                  * before we have finished mounting the filesystem.
1849                  */
1850                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1851                         goto sleep;
1852
1853                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1854                         goto sleep;
1855
1856                 /*
1857                  * Avoid the problem that we change the status of the fs
1858                  * during the above check and trylock.
1859                  */
1860                 if (btrfs_need_cleaner_sleep(fs_info)) {
1861                         mutex_unlock(&fs_info->cleaner_mutex);
1862                         goto sleep;
1863                 }
1864
1865                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1866                 btrfs_run_delayed_iputs(fs_info);
1867                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1868
1869                 again = btrfs_clean_one_deleted_snapshot(root);
1870                 mutex_unlock(&fs_info->cleaner_mutex);
1871
1872                 /*
1873                  * The defragger has dealt with the R/O remount and umount,
1874                  * needn't do anything special here.
1875                  */
1876                 btrfs_run_defrag_inodes(fs_info);
1877
1878                 /*
1879                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1880                  * with relocation (btrfs_relocate_chunk) and relocation
1881                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1882                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1883                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1884                  * unused block groups.
1885                  */
1886                 btrfs_delete_unused_bgs(fs_info);
1887 sleep:
1888                 if (!again) {
1889                         set_current_state(TASK_INTERRUPTIBLE);
1890                         if (!kthread_should_stop())
1891                                 schedule();
1892                         __set_current_state(TASK_RUNNING);
1893                 }
1894         } while (!kthread_should_stop());
1895
1896         /*
1897          * Transaction kthread is stopped before us and wakes us up.
1898          * However we might have started a new transaction and COWed some
1899          * tree blocks when deleting unused block groups for example. So
1900          * make sure we commit the transaction we started to have a clean
1901          * shutdown when evicting the btree inode - if it has dirty pages
1902          * when we do the final iput() on it, eviction will trigger a
1903          * writeback for it which will fail with null pointer dereferences
1904          * since work queues and other resources were already released and
1905          * destroyed by the time the iput/eviction/writeback is made.
1906          */
1907         trans = btrfs_attach_transaction(root);
1908         if (IS_ERR(trans)) {
1909                 if (PTR_ERR(trans) != -ENOENT)
1910                         btrfs_err(fs_info,
1911                                   "cleaner transaction attach returned %ld",
1912                                   PTR_ERR(trans));
1913         } else {
1914                 int ret;
1915
1916                 ret = btrfs_commit_transaction(trans);
1917                 if (ret)
1918                         btrfs_err(fs_info,
1919                                   "cleaner open transaction commit returned %d",
1920                                   ret);
1921         }
1922
1923         return 0;
1924 }
1925
1926 static int transaction_kthread(void *arg)
1927 {
1928         struct btrfs_root *root = arg;
1929         struct btrfs_fs_info *fs_info = root->fs_info;
1930         struct btrfs_trans_handle *trans;
1931         struct btrfs_transaction *cur;
1932         u64 transid;
1933         unsigned long now;
1934         unsigned long delay;
1935         bool cannot_commit;
1936
1937         do {
1938                 cannot_commit = false;
1939                 delay = HZ * fs_info->commit_interval;
1940                 mutex_lock(&fs_info->transaction_kthread_mutex);
1941
1942                 spin_lock(&fs_info->trans_lock);
1943                 cur = fs_info->running_transaction;
1944                 if (!cur) {
1945                         spin_unlock(&fs_info->trans_lock);
1946                         goto sleep;
1947                 }
1948
1949                 now = get_seconds();
1950                 if (cur->state < TRANS_STATE_BLOCKED &&
1951                     (now < cur->start_time ||
1952                      now - cur->start_time < fs_info->commit_interval)) {
1953                         spin_unlock(&fs_info->trans_lock);
1954                         delay = HZ * 5;
1955                         goto sleep;
1956                 }
1957                 transid = cur->transid;
1958                 spin_unlock(&fs_info->trans_lock);
1959
1960                 /* If the file system is aborted, this will always fail. */
1961                 trans = btrfs_attach_transaction(root);
1962                 if (IS_ERR(trans)) {
1963                         if (PTR_ERR(trans) != -ENOENT)
1964                                 cannot_commit = true;
1965                         goto sleep;
1966                 }
1967                 if (transid == trans->transid) {
1968                         btrfs_commit_transaction(trans);
1969                 } else {
1970                         btrfs_end_transaction(trans);
1971                 }
1972 sleep:
1973                 wake_up_process(fs_info->cleaner_kthread);
1974                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1975
1976                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1977                                       &fs_info->fs_state)))
1978                         btrfs_cleanup_transaction(fs_info);
1979                 set_current_state(TASK_INTERRUPTIBLE);
1980                 if (!kthread_should_stop() &&
1981                                 (!btrfs_transaction_blocked(fs_info) ||
1982                                  cannot_commit))
1983                         schedule_timeout(delay);
1984                 __set_current_state(TASK_RUNNING);
1985         } while (!kthread_should_stop());
1986         return 0;
1987 }
1988
1989 /*
1990  * this will find the highest generation in the array of
1991  * root backups.  The index of the highest array is returned,
1992  * or -1 if we can't find anything.
1993  *
1994  * We check to make sure the array is valid by comparing the
1995  * generation of the latest  root in the array with the generation
1996  * in the super block.  If they don't match we pitch it.
1997  */
1998 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1999 {
2000         u64 cur;
2001         int newest_index = -1;
2002         struct btrfs_root_backup *root_backup;
2003         int i;
2004
2005         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2006                 root_backup = info->super_copy->super_roots + i;
2007                 cur = btrfs_backup_tree_root_gen(root_backup);
2008                 if (cur == newest_gen)
2009                         newest_index = i;
2010         }
2011
2012         /* check to see if we actually wrapped around */
2013         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
2014                 root_backup = info->super_copy->super_roots;
2015                 cur = btrfs_backup_tree_root_gen(root_backup);
2016                 if (cur == newest_gen)
2017                         newest_index = 0;
2018         }
2019         return newest_index;
2020 }
2021
2022
2023 /*
2024  * find the oldest backup so we know where to store new entries
2025  * in the backup array.  This will set the backup_root_index
2026  * field in the fs_info struct
2027  */
2028 static void find_oldest_super_backup(struct btrfs_fs_info *info,
2029                                      u64 newest_gen)
2030 {
2031         int newest_index = -1;
2032
2033         newest_index = find_newest_super_backup(info, newest_gen);
2034         /* if there was garbage in there, just move along */
2035         if (newest_index == -1) {
2036                 info->backup_root_index = 0;
2037         } else {
2038                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2039         }
2040 }
2041
2042 /*
2043  * copy all the root pointers into the super backup array.
2044  * this will bump the backup pointer by one when it is
2045  * done
2046  */
2047 static void backup_super_roots(struct btrfs_fs_info *info)
2048 {
2049         int next_backup;
2050         struct btrfs_root_backup *root_backup;
2051         int last_backup;
2052
2053         next_backup = info->backup_root_index;
2054         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2055                 BTRFS_NUM_BACKUP_ROOTS;
2056
2057         /*
2058          * just overwrite the last backup if we're at the same generation
2059          * this happens only at umount
2060          */
2061         root_backup = info->super_for_commit->super_roots + last_backup;
2062         if (btrfs_backup_tree_root_gen(root_backup) ==
2063             btrfs_header_generation(info->tree_root->node))
2064                 next_backup = last_backup;
2065
2066         root_backup = info->super_for_commit->super_roots + next_backup;
2067
2068         /*
2069          * make sure all of our padding and empty slots get zero filled
2070          * regardless of which ones we use today
2071          */
2072         memset(root_backup, 0, sizeof(*root_backup));
2073
2074         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2075
2076         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2077         btrfs_set_backup_tree_root_gen(root_backup,
2078                                btrfs_header_generation(info->tree_root->node));
2079
2080         btrfs_set_backup_tree_root_level(root_backup,
2081                                btrfs_header_level(info->tree_root->node));
2082
2083         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2084         btrfs_set_backup_chunk_root_gen(root_backup,
2085                                btrfs_header_generation(info->chunk_root->node));
2086         btrfs_set_backup_chunk_root_level(root_backup,
2087                                btrfs_header_level(info->chunk_root->node));
2088
2089         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2090         btrfs_set_backup_extent_root_gen(root_backup,
2091                                btrfs_header_generation(info->extent_root->node));
2092         btrfs_set_backup_extent_root_level(root_backup,
2093                                btrfs_header_level(info->extent_root->node));
2094
2095         /*
2096          * we might commit during log recovery, which happens before we set
2097          * the fs_root.  Make sure it is valid before we fill it in.
2098          */
2099         if (info->fs_root && info->fs_root->node) {
2100                 btrfs_set_backup_fs_root(root_backup,
2101                                          info->fs_root->node->start);
2102                 btrfs_set_backup_fs_root_gen(root_backup,
2103                                btrfs_header_generation(info->fs_root->node));
2104                 btrfs_set_backup_fs_root_level(root_backup,
2105                                btrfs_header_level(info->fs_root->node));
2106         }
2107
2108         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2109         btrfs_set_backup_dev_root_gen(root_backup,
2110                                btrfs_header_generation(info->dev_root->node));
2111         btrfs_set_backup_dev_root_level(root_backup,
2112                                        btrfs_header_level(info->dev_root->node));
2113
2114         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2115         btrfs_set_backup_csum_root_gen(root_backup,
2116                                btrfs_header_generation(info->csum_root->node));
2117         btrfs_set_backup_csum_root_level(root_backup,
2118                                btrfs_header_level(info->csum_root->node));
2119
2120         btrfs_set_backup_total_bytes(root_backup,
2121                              btrfs_super_total_bytes(info->super_copy));
2122         btrfs_set_backup_bytes_used(root_backup,
2123                              btrfs_super_bytes_used(info->super_copy));
2124         btrfs_set_backup_num_devices(root_backup,
2125                              btrfs_super_num_devices(info->super_copy));
2126
2127         /*
2128          * if we don't copy this out to the super_copy, it won't get remembered
2129          * for the next commit
2130          */
2131         memcpy(&info->super_copy->super_roots,
2132                &info->super_for_commit->super_roots,
2133                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2134 }
2135
2136 /*
2137  * this copies info out of the root backup array and back into
2138  * the in-memory super block.  It is meant to help iterate through
2139  * the array, so you send it the number of backups you've already
2140  * tried and the last backup index you used.
2141  *
2142  * this returns -1 when it has tried all the backups
2143  */
2144 static noinline int next_root_backup(struct btrfs_fs_info *info,
2145                                      struct btrfs_super_block *super,
2146                                      int *num_backups_tried, int *backup_index)
2147 {
2148         struct btrfs_root_backup *root_backup;
2149         int newest = *backup_index;
2150
2151         if (*num_backups_tried == 0) {
2152                 u64 gen = btrfs_super_generation(super);
2153
2154                 newest = find_newest_super_backup(info, gen);
2155                 if (newest == -1)
2156                         return -1;
2157
2158                 *backup_index = newest;
2159                 *num_backups_tried = 1;
2160         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2161                 /* we've tried all the backups, all done */
2162                 return -1;
2163         } else {
2164                 /* jump to the next oldest backup */
2165                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2166                         BTRFS_NUM_BACKUP_ROOTS;
2167                 *backup_index = newest;
2168                 *num_backups_tried += 1;
2169         }
2170         root_backup = super->super_roots + newest;
2171
2172         btrfs_set_super_generation(super,
2173                                    btrfs_backup_tree_root_gen(root_backup));
2174         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2175         btrfs_set_super_root_level(super,
2176                                    btrfs_backup_tree_root_level(root_backup));
2177         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2178
2179         /*
2180          * fixme: the total bytes and num_devices need to match or we should
2181          * need a fsck
2182          */
2183         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2184         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2185         return 0;
2186 }
2187
2188 /* helper to cleanup workers */
2189 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2190 {
2191         btrfs_destroy_workqueue(fs_info->fixup_workers);
2192         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2193         btrfs_destroy_workqueue(fs_info->workers);
2194         btrfs_destroy_workqueue(fs_info->endio_workers);
2195         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2196         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2197         btrfs_destroy_workqueue(fs_info->rmw_workers);
2198         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2199         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2200         btrfs_destroy_workqueue(fs_info->submit_workers);
2201         btrfs_destroy_workqueue(fs_info->delayed_workers);
2202         btrfs_destroy_workqueue(fs_info->caching_workers);
2203         btrfs_destroy_workqueue(fs_info->readahead_workers);
2204         btrfs_destroy_workqueue(fs_info->flush_workers);
2205         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2206         btrfs_destroy_workqueue(fs_info->extent_workers);
2207         /*
2208          * Now that all other work queues are destroyed, we can safely destroy
2209          * the queues used for metadata I/O, since tasks from those other work
2210          * queues can do metadata I/O operations.
2211          */
2212         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2213         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2214 }
2215
2216 static void free_root_extent_buffers(struct btrfs_root *root)
2217 {
2218         if (root) {
2219                 free_extent_buffer(root->node);
2220                 free_extent_buffer(root->commit_root);
2221                 root->node = NULL;
2222                 root->commit_root = NULL;
2223         }
2224 }
2225
2226 /* helper to cleanup tree roots */
2227 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2228 {
2229         free_root_extent_buffers(info->tree_root);
2230
2231         free_root_extent_buffers(info->dev_root);
2232         free_root_extent_buffers(info->extent_root);
2233         free_root_extent_buffers(info->csum_root);
2234         free_root_extent_buffers(info->quota_root);
2235         free_root_extent_buffers(info->uuid_root);
2236         if (chunk_root)
2237                 free_root_extent_buffers(info->chunk_root);
2238         free_root_extent_buffers(info->free_space_root);
2239 }
2240
2241 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2242 {
2243         int ret;
2244         struct btrfs_root *gang[8];
2245         int i;
2246
2247         while (!list_empty(&fs_info->dead_roots)) {
2248                 gang[0] = list_entry(fs_info->dead_roots.next,
2249                                      struct btrfs_root, root_list);
2250                 list_del(&gang[0]->root_list);
2251
2252                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2253                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2254                 } else {
2255                         free_extent_buffer(gang[0]->node);
2256                         free_extent_buffer(gang[0]->commit_root);
2257                         btrfs_put_fs_root(gang[0]);
2258                 }
2259         }
2260
2261         while (1) {
2262                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2263                                              (void **)gang, 0,
2264                                              ARRAY_SIZE(gang));
2265                 if (!ret)
2266                         break;
2267                 for (i = 0; i < ret; i++)
2268                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2269         }
2270
2271         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2272                 btrfs_free_log_root_tree(NULL, fs_info);
2273                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2274         }
2275 }
2276
2277 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2278 {
2279         mutex_init(&fs_info->scrub_lock);
2280         atomic_set(&fs_info->scrubs_running, 0);
2281         atomic_set(&fs_info->scrub_pause_req, 0);
2282         atomic_set(&fs_info->scrubs_paused, 0);
2283         atomic_set(&fs_info->scrub_cancel_req, 0);
2284         init_waitqueue_head(&fs_info->scrub_pause_wait);
2285         fs_info->scrub_workers_refcnt = 0;
2286 }
2287
2288 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2289 {
2290         spin_lock_init(&fs_info->balance_lock);
2291         mutex_init(&fs_info->balance_mutex);
2292         atomic_set(&fs_info->balance_running, 0);
2293         atomic_set(&fs_info->balance_pause_req, 0);
2294         atomic_set(&fs_info->balance_cancel_req, 0);
2295         fs_info->balance_ctl = NULL;
2296         init_waitqueue_head(&fs_info->balance_wait_q);
2297 }
2298
2299 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2300 {
2301         struct inode *inode = fs_info->btree_inode;
2302
2303         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2304         set_nlink(inode, 1);
2305         /*
2306          * we set the i_size on the btree inode to the max possible int.
2307          * the real end of the address space is determined by all of
2308          * the devices in the system
2309          */
2310         inode->i_size = OFFSET_MAX;
2311         inode->i_mapping->a_ops = &btree_aops;
2312
2313         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2314         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2315         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2316         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2317
2318         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2319
2320         BTRFS_I(inode)->root = fs_info->tree_root;
2321         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2322         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2323         btrfs_insert_inode_hash(inode);
2324 }
2325
2326 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2327 {
2328         fs_info->dev_replace.lock_owner = 0;
2329         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2330         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2331         rwlock_init(&fs_info->dev_replace.lock);
2332         atomic_set(&fs_info->dev_replace.read_locks, 0);
2333         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2334         init_waitqueue_head(&fs_info->replace_wait);
2335         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2336 }
2337
2338 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2339 {
2340         spin_lock_init(&fs_info->qgroup_lock);
2341         mutex_init(&fs_info->qgroup_ioctl_lock);
2342         fs_info->qgroup_tree = RB_ROOT;
2343         fs_info->qgroup_op_tree = RB_ROOT;
2344         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2345         fs_info->qgroup_seq = 1;
2346         fs_info->qgroup_ulist = NULL;
2347         fs_info->qgroup_rescan_running = false;
2348         mutex_init(&fs_info->qgroup_rescan_lock);
2349 }
2350
2351 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2352                 struct btrfs_fs_devices *fs_devices)
2353 {
2354         int max_active = fs_info->thread_pool_size;
2355         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2356
2357         fs_info->workers =
2358                 btrfs_alloc_workqueue(fs_info, "worker",
2359                                       flags | WQ_HIGHPRI, max_active, 16);
2360
2361         fs_info->delalloc_workers =
2362                 btrfs_alloc_workqueue(fs_info, "delalloc",
2363                                       flags, max_active, 2);
2364
2365         fs_info->flush_workers =
2366                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2367                                       flags, max_active, 0);
2368
2369         fs_info->caching_workers =
2370                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2371
2372         /*
2373          * a higher idle thresh on the submit workers makes it much more
2374          * likely that bios will be send down in a sane order to the
2375          * devices
2376          */
2377         fs_info->submit_workers =
2378                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2379                                       min_t(u64, fs_devices->num_devices,
2380                                             max_active), 64);
2381
2382         fs_info->fixup_workers =
2383                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2384
2385         /*
2386          * endios are largely parallel and should have a very
2387          * low idle thresh
2388          */
2389         fs_info->endio_workers =
2390                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2391         fs_info->endio_meta_workers =
2392                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2393                                       max_active, 4);
2394         fs_info->endio_meta_write_workers =
2395                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2396                                       max_active, 2);
2397         fs_info->endio_raid56_workers =
2398                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2399                                       max_active, 4);
2400         fs_info->endio_repair_workers =
2401                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2402         fs_info->rmw_workers =
2403                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2404         fs_info->endio_write_workers =
2405                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2406                                       max_active, 2);
2407         fs_info->endio_freespace_worker =
2408                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2409                                       max_active, 0);
2410         fs_info->delayed_workers =
2411                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2412                                       max_active, 0);
2413         fs_info->readahead_workers =
2414                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2415                                       max_active, 2);
2416         fs_info->qgroup_rescan_workers =
2417                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2418         fs_info->extent_workers =
2419                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2420                                       min_t(u64, fs_devices->num_devices,
2421                                             max_active), 8);
2422
2423         if (!(fs_info->workers && fs_info->delalloc_workers &&
2424               fs_info->submit_workers && fs_info->flush_workers &&
2425               fs_info->endio_workers && fs_info->endio_meta_workers &&
2426               fs_info->endio_meta_write_workers &&
2427               fs_info->endio_repair_workers &&
2428               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2429               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2430               fs_info->caching_workers && fs_info->readahead_workers &&
2431               fs_info->fixup_workers && fs_info->delayed_workers &&
2432               fs_info->extent_workers &&
2433               fs_info->qgroup_rescan_workers)) {
2434                 return -ENOMEM;
2435         }
2436
2437         return 0;
2438 }
2439
2440 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2441                             struct btrfs_fs_devices *fs_devices)
2442 {
2443         int ret;
2444         struct btrfs_root *log_tree_root;
2445         struct btrfs_super_block *disk_super = fs_info->super_copy;
2446         u64 bytenr = btrfs_super_log_root(disk_super);
2447
2448         if (fs_devices->rw_devices == 0) {
2449                 btrfs_warn(fs_info, "log replay required on RO media");
2450                 return -EIO;
2451         }
2452
2453         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2454         if (!log_tree_root)
2455                 return -ENOMEM;
2456
2457         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2458
2459         log_tree_root->node = read_tree_block(fs_info, bytenr,
2460                                               fs_info->generation + 1);
2461         if (IS_ERR(log_tree_root->node)) {
2462                 btrfs_warn(fs_info, "failed to read log tree");
2463                 ret = PTR_ERR(log_tree_root->node);
2464                 kfree(log_tree_root);
2465                 return ret;
2466         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2467                 btrfs_err(fs_info, "failed to read log tree");
2468                 free_extent_buffer(log_tree_root->node);
2469                 kfree(log_tree_root);
2470                 return -EIO;
2471         }
2472         /* returns with log_tree_root freed on success */
2473         ret = btrfs_recover_log_trees(log_tree_root);
2474         if (ret) {
2475                 btrfs_handle_fs_error(fs_info, ret,
2476                                       "Failed to recover log tree");
2477                 free_extent_buffer(log_tree_root->node);
2478                 kfree(log_tree_root);
2479                 return ret;
2480         }
2481
2482         if (fs_info->sb->s_flags & MS_RDONLY) {
2483                 ret = btrfs_commit_super(fs_info);
2484                 if (ret)
2485                         return ret;
2486         }
2487
2488         return 0;
2489 }
2490
2491 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2492 {
2493         struct btrfs_root *tree_root = fs_info->tree_root;
2494         struct btrfs_root *root;
2495         struct btrfs_key location;
2496         int ret;
2497
2498         BUG_ON(!fs_info->tree_root);
2499
2500         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2501         location.type = BTRFS_ROOT_ITEM_KEY;
2502         location.offset = 0;
2503
2504         root = btrfs_read_tree_root(tree_root, &location);
2505         if (IS_ERR(root))
2506                 return PTR_ERR(root);
2507         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2508         fs_info->extent_root = root;
2509
2510         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2511         root = btrfs_read_tree_root(tree_root, &location);
2512         if (IS_ERR(root))
2513                 return PTR_ERR(root);
2514         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2515         fs_info->dev_root = root;
2516         btrfs_init_devices_late(fs_info);
2517
2518         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2519         root = btrfs_read_tree_root(tree_root, &location);
2520         if (IS_ERR(root))
2521                 return PTR_ERR(root);
2522         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2523         fs_info->csum_root = root;
2524
2525         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2526         root = btrfs_read_tree_root(tree_root, &location);
2527         if (!IS_ERR(root)) {
2528                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2529                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2530                 fs_info->quota_root = root;
2531         }
2532
2533         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2534         root = btrfs_read_tree_root(tree_root, &location);
2535         if (IS_ERR(root)) {
2536                 ret = PTR_ERR(root);
2537                 if (ret != -ENOENT)
2538                         return ret;
2539         } else {
2540                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2541                 fs_info->uuid_root = root;
2542         }
2543
2544         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2545                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2546                 root = btrfs_read_tree_root(tree_root, &location);
2547                 if (IS_ERR(root))
2548                         return PTR_ERR(root);
2549                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2550                 fs_info->free_space_root = root;
2551         }
2552
2553         return 0;
2554 }
2555
2556 int open_ctree(struct super_block *sb,
2557                struct btrfs_fs_devices *fs_devices,
2558                char *options)
2559 {
2560         u32 sectorsize;
2561         u32 nodesize;
2562         u32 stripesize;
2563         u64 generation;
2564         u64 features;
2565         struct btrfs_key location;
2566         struct buffer_head *bh;
2567         struct btrfs_super_block *disk_super;
2568         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2569         struct btrfs_root *tree_root;
2570         struct btrfs_root *chunk_root;
2571         int ret;
2572         int err = -EINVAL;
2573         int num_backups_tried = 0;
2574         int backup_index = 0;
2575         int max_active;
2576         int clear_free_space_tree = 0;
2577
2578         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2579         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2580         if (!tree_root || !chunk_root) {
2581                 err = -ENOMEM;
2582                 goto fail;
2583         }
2584
2585         ret = init_srcu_struct(&fs_info->subvol_srcu);
2586         if (ret) {
2587                 err = ret;
2588                 goto fail;
2589         }
2590
2591         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2592         if (ret) {
2593                 err = ret;
2594                 goto fail_srcu;
2595         }
2596         fs_info->dirty_metadata_batch = PAGE_SIZE *
2597                                         (1 + ilog2(nr_cpu_ids));
2598
2599         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2600         if (ret) {
2601                 err = ret;
2602                 goto fail_dirty_metadata_bytes;
2603         }
2604
2605         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2606         if (ret) {
2607                 err = ret;
2608                 goto fail_delalloc_bytes;
2609         }
2610
2611         fs_info->btree_inode = new_inode(sb);
2612         if (!fs_info->btree_inode) {
2613                 err = -ENOMEM;
2614                 goto fail_bio_counter;
2615         }
2616
2617         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2618
2619         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2620         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2621         INIT_LIST_HEAD(&fs_info->trans_list);
2622         INIT_LIST_HEAD(&fs_info->dead_roots);
2623         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2624         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2625         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2626         spin_lock_init(&fs_info->delalloc_root_lock);
2627         spin_lock_init(&fs_info->trans_lock);
2628         spin_lock_init(&fs_info->fs_roots_radix_lock);
2629         spin_lock_init(&fs_info->delayed_iput_lock);
2630         spin_lock_init(&fs_info->defrag_inodes_lock);
2631         spin_lock_init(&fs_info->tree_mod_seq_lock);
2632         spin_lock_init(&fs_info->super_lock);
2633         spin_lock_init(&fs_info->qgroup_op_lock);
2634         spin_lock_init(&fs_info->buffer_lock);
2635         spin_lock_init(&fs_info->unused_bgs_lock);
2636         rwlock_init(&fs_info->tree_mod_log_lock);
2637         mutex_init(&fs_info->unused_bg_unpin_mutex);
2638         mutex_init(&fs_info->delete_unused_bgs_mutex);
2639         mutex_init(&fs_info->reloc_mutex);
2640         mutex_init(&fs_info->delalloc_root_mutex);
2641         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2642         seqlock_init(&fs_info->profiles_lock);
2643
2644         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2645         INIT_LIST_HEAD(&fs_info->space_info);
2646         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2647         INIT_LIST_HEAD(&fs_info->unused_bgs);
2648         btrfs_mapping_init(&fs_info->mapping_tree);
2649         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2650                              BTRFS_BLOCK_RSV_GLOBAL);
2651         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2652                              BTRFS_BLOCK_RSV_DELALLOC);
2653         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2654         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2655         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2656         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2657                              BTRFS_BLOCK_RSV_DELOPS);
2658         atomic_set(&fs_info->nr_async_submits, 0);
2659         atomic_set(&fs_info->async_delalloc_pages, 0);
2660         atomic_set(&fs_info->async_submit_draining, 0);
2661         atomic_set(&fs_info->nr_async_bios, 0);
2662         atomic_set(&fs_info->defrag_running, 0);
2663         atomic_set(&fs_info->qgroup_op_seq, 0);
2664         atomic_set(&fs_info->reada_works_cnt, 0);
2665         atomic64_set(&fs_info->tree_mod_seq, 0);
2666         fs_info->fs_frozen = 0;
2667         fs_info->sb = sb;
2668         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2669         fs_info->metadata_ratio = 0;
2670         fs_info->defrag_inodes = RB_ROOT;
2671         atomic64_set(&fs_info->free_chunk_space, 0);
2672         fs_info->tree_mod_log = RB_ROOT;
2673         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2674         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2675         /* readahead state */
2676         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2677         spin_lock_init(&fs_info->reada_lock);
2678
2679         fs_info->thread_pool_size = min_t(unsigned long,
2680                                           num_online_cpus() + 2, 8);
2681
2682         INIT_LIST_HEAD(&fs_info->ordered_roots);
2683         spin_lock_init(&fs_info->ordered_root_lock);
2684         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2685                                         GFP_KERNEL);
2686         if (!fs_info->delayed_root) {
2687                 err = -ENOMEM;
2688                 goto fail_iput;
2689         }
2690         btrfs_init_delayed_root(fs_info->delayed_root);
2691
2692         btrfs_init_scrub(fs_info);
2693 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2694         fs_info->check_integrity_print_mask = 0;
2695 #endif
2696         btrfs_init_balance(fs_info);
2697         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2698
2699         sb->s_blocksize = 4096;
2700         sb->s_blocksize_bits = blksize_bits(4096);
2701
2702         btrfs_init_btree_inode(fs_info);
2703
2704         spin_lock_init(&fs_info->block_group_cache_lock);
2705         fs_info->block_group_cache_tree = RB_ROOT;
2706         fs_info->first_logical_byte = (u64)-1;
2707
2708         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2709         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2710         fs_info->pinned_extents = &fs_info->freed_extents[0];
2711         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2712
2713         mutex_init(&fs_info->ordered_operations_mutex);
2714         mutex_init(&fs_info->tree_log_mutex);
2715         mutex_init(&fs_info->chunk_mutex);
2716         mutex_init(&fs_info->transaction_kthread_mutex);
2717         mutex_init(&fs_info->cleaner_mutex);
2718         mutex_init(&fs_info->volume_mutex);
2719         mutex_init(&fs_info->ro_block_group_mutex);
2720         init_rwsem(&fs_info->commit_root_sem);
2721         init_rwsem(&fs_info->cleanup_work_sem);
2722         init_rwsem(&fs_info->subvol_sem);
2723         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2724
2725         btrfs_init_dev_replace_locks(fs_info);
2726         btrfs_init_qgroup(fs_info);
2727
2728         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2729         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2730
2731         init_waitqueue_head(&fs_info->transaction_throttle);
2732         init_waitqueue_head(&fs_info->transaction_wait);
2733         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2734         init_waitqueue_head(&fs_info->async_submit_wait);
2735
2736         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2737
2738         /* Usable values until the real ones are cached from the superblock */
2739         fs_info->nodesize = 4096;
2740         fs_info->sectorsize = 4096;
2741         fs_info->stripesize = 4096;
2742
2743         ret = btrfs_alloc_stripe_hash_table(fs_info);
2744         if (ret) {
2745                 err = ret;
2746                 goto fail_alloc;
2747         }
2748
2749         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2750
2751         invalidate_bdev(fs_devices->latest_bdev);
2752
2753         /*
2754          * Read super block and check the signature bytes only
2755          */
2756         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2757         if (IS_ERR(bh)) {
2758                 err = PTR_ERR(bh);
2759                 goto fail_alloc;
2760         }
2761
2762         /*
2763          * We want to check superblock checksum, the type is stored inside.
2764          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2765          */
2766         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2767                 btrfs_err(fs_info, "superblock checksum mismatch");
2768                 err = -EINVAL;
2769                 brelse(bh);
2770                 goto fail_alloc;
2771         }
2772
2773         /*
2774          * super_copy is zeroed at allocation time and we never touch the
2775          * following bytes up to INFO_SIZE, the checksum is calculated from
2776          * the whole block of INFO_SIZE
2777          */
2778         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2779         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2780                sizeof(*fs_info->super_for_commit));
2781         brelse(bh);
2782
2783         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2784
2785         ret = btrfs_check_super_valid(fs_info);
2786         if (ret) {
2787                 btrfs_err(fs_info, "superblock contains fatal errors");
2788                 err = -EINVAL;
2789                 goto fail_alloc;
2790         }
2791
2792         disk_super = fs_info->super_copy;
2793         if (!btrfs_super_root(disk_super))
2794                 goto fail_alloc;
2795
2796         /* check FS state, whether FS is broken. */
2797         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2798                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2799
2800         /*
2801          * run through our array of backup supers and setup
2802          * our ring pointer to the oldest one
2803          */
2804         generation = btrfs_super_generation(disk_super);
2805         find_oldest_super_backup(fs_info, generation);
2806
2807         /*
2808          * In the long term, we'll store the compression type in the super
2809          * block, and it'll be used for per file compression control.
2810          */
2811         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2812
2813         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2814         if (ret) {
2815                 err = ret;
2816                 goto fail_alloc;
2817         }
2818
2819         features = btrfs_super_incompat_flags(disk_super) &
2820                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2821         if (features) {
2822                 btrfs_err(fs_info,
2823                     "cannot mount because of unsupported optional features (%llx)",
2824                     features);
2825                 err = -EINVAL;
2826                 goto fail_alloc;
2827         }
2828
2829         features = btrfs_super_incompat_flags(disk_super);
2830         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2831         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2832                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2833
2834         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2835                 btrfs_info(fs_info, "has skinny extents");
2836
2837         /*
2838          * flag our filesystem as having big metadata blocks if
2839          * they are bigger than the page size
2840          */
2841         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2842                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2843                         btrfs_info(fs_info,
2844                                 "flagging fs with big metadata feature");
2845                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2846         }
2847
2848         nodesize = btrfs_super_nodesize(disk_super);
2849         sectorsize = btrfs_super_sectorsize(disk_super);
2850         stripesize = sectorsize;
2851         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2852         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2853
2854         /* Cache block sizes */
2855         fs_info->nodesize = nodesize;
2856         fs_info->sectorsize = sectorsize;
2857         fs_info->stripesize = stripesize;
2858
2859         /*
2860          * mixed block groups end up with duplicate but slightly offset
2861          * extent buffers for the same range.  It leads to corruptions
2862          */
2863         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2864             (sectorsize != nodesize)) {
2865                 btrfs_err(fs_info,
2866 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2867                         nodesize, sectorsize);
2868                 goto fail_alloc;
2869         }
2870
2871         /*
2872          * Needn't use the lock because there is no other task which will
2873          * update the flag.
2874          */
2875         btrfs_set_super_incompat_flags(disk_super, features);
2876
2877         features = btrfs_super_compat_ro_flags(disk_super) &
2878                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2879         if (!(sb->s_flags & MS_RDONLY) && features) {
2880                 btrfs_err(fs_info,
2881         "cannot mount read-write because of unsupported optional features (%llx)",
2882                        features);
2883                 err = -EINVAL;
2884                 goto fail_alloc;
2885         }
2886
2887         max_active = fs_info->thread_pool_size;
2888
2889         ret = btrfs_init_workqueues(fs_info, fs_devices);
2890         if (ret) {
2891                 err = ret;
2892                 goto fail_sb_buffer;
2893         }
2894
2895         sb->s_bdi->congested_fn = btrfs_congested_fn;
2896         sb->s_bdi->congested_data = fs_info;
2897         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2898         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
2899         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2900         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2901
2902         sb->s_blocksize = sectorsize;
2903         sb->s_blocksize_bits = blksize_bits(sectorsize);
2904
2905         mutex_lock(&fs_info->chunk_mutex);
2906         ret = btrfs_read_sys_array(fs_info);
2907         mutex_unlock(&fs_info->chunk_mutex);
2908         if (ret) {
2909                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2910                 goto fail_sb_buffer;
2911         }
2912
2913         generation = btrfs_super_chunk_root_generation(disk_super);
2914
2915         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2916
2917         chunk_root->node = read_tree_block(fs_info,
2918                                            btrfs_super_chunk_root(disk_super),
2919                                            generation);
2920         if (IS_ERR(chunk_root->node) ||
2921             !extent_buffer_uptodate(chunk_root->node)) {
2922                 btrfs_err(fs_info, "failed to read chunk root");
2923                 if (!IS_ERR(chunk_root->node))
2924                         free_extent_buffer(chunk_root->node);
2925                 chunk_root->node = NULL;
2926                 goto fail_tree_roots;
2927         }
2928         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2929         chunk_root->commit_root = btrfs_root_node(chunk_root);
2930
2931         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2932            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2933
2934         ret = btrfs_read_chunk_tree(fs_info);
2935         if (ret) {
2936                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2937                 goto fail_tree_roots;
2938         }
2939
2940         /*
2941          * keep the device that is marked to be the target device for the
2942          * dev_replace procedure
2943          */
2944         btrfs_close_extra_devices(fs_devices, 0);
2945
2946         if (!fs_devices->latest_bdev) {
2947                 btrfs_err(fs_info, "failed to read devices");
2948                 goto fail_tree_roots;
2949         }
2950
2951 retry_root_backup:
2952         generation = btrfs_super_generation(disk_super);
2953
2954         tree_root->node = read_tree_block(fs_info,
2955                                           btrfs_super_root(disk_super),
2956                                           generation);
2957         if (IS_ERR(tree_root->node) ||
2958             !extent_buffer_uptodate(tree_root->node)) {
2959                 btrfs_warn(fs_info, "failed to read tree root");
2960                 if (!IS_ERR(tree_root->node))
2961                         free_extent_buffer(tree_root->node);
2962                 tree_root->node = NULL;
2963                 goto recovery_tree_root;
2964         }
2965
2966         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2967         tree_root->commit_root = btrfs_root_node(tree_root);
2968         btrfs_set_root_refs(&tree_root->root_item, 1);
2969
2970         mutex_lock(&tree_root->objectid_mutex);
2971         ret = btrfs_find_highest_objectid(tree_root,
2972                                         &tree_root->highest_objectid);
2973         if (ret) {
2974                 mutex_unlock(&tree_root->objectid_mutex);
2975                 goto recovery_tree_root;
2976         }
2977
2978         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2979
2980         mutex_unlock(&tree_root->objectid_mutex);
2981
2982         ret = btrfs_read_roots(fs_info);
2983         if (ret)
2984                 goto recovery_tree_root;
2985
2986         fs_info->generation = generation;
2987         fs_info->last_trans_committed = generation;
2988
2989         ret = btrfs_recover_balance(fs_info);
2990         if (ret) {
2991                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2992                 goto fail_block_groups;
2993         }
2994
2995         ret = btrfs_init_dev_stats(fs_info);
2996         if (ret) {
2997                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2998                 goto fail_block_groups;
2999         }
3000
3001         ret = btrfs_init_dev_replace(fs_info);
3002         if (ret) {
3003                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3004                 goto fail_block_groups;
3005         }
3006
3007         btrfs_close_extra_devices(fs_devices, 1);
3008
3009         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3010         if (ret) {
3011                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3012                                 ret);
3013                 goto fail_block_groups;
3014         }
3015
3016         ret = btrfs_sysfs_add_device(fs_devices);
3017         if (ret) {
3018                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3019                                 ret);
3020                 goto fail_fsdev_sysfs;
3021         }
3022
3023         ret = btrfs_sysfs_add_mounted(fs_info);
3024         if (ret) {
3025                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3026                 goto fail_fsdev_sysfs;
3027         }
3028
3029         ret = btrfs_init_space_info(fs_info);
3030         if (ret) {
3031                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3032                 goto fail_sysfs;
3033         }
3034
3035         ret = btrfs_read_block_groups(fs_info);
3036         if (ret) {
3037                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3038                 goto fail_sysfs;
3039         }
3040         fs_info->num_tolerated_disk_barrier_failures =
3041                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3042         if (fs_info->fs_devices->missing_devices >
3043              fs_info->num_tolerated_disk_barrier_failures &&
3044             !(sb->s_flags & MS_RDONLY)) {
3045                 btrfs_warn(fs_info,
3046 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3047                         fs_info->fs_devices->missing_devices,
3048                         fs_info->num_tolerated_disk_barrier_failures);
3049                 goto fail_sysfs;
3050         }
3051
3052         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3053                                                "btrfs-cleaner");
3054         if (IS_ERR(fs_info->cleaner_kthread))
3055                 goto fail_sysfs;
3056
3057         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3058                                                    tree_root,
3059                                                    "btrfs-transaction");
3060         if (IS_ERR(fs_info->transaction_kthread))
3061                 goto fail_cleaner;
3062
3063         if (!btrfs_test_opt(fs_info, SSD) &&
3064             !btrfs_test_opt(fs_info, NOSSD) &&
3065             !fs_info->fs_devices->rotating) {
3066                 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3067                 btrfs_set_opt(fs_info->mount_opt, SSD);
3068         }
3069
3070         /*
3071          * Mount does not set all options immediately, we can do it now and do
3072          * not have to wait for transaction commit
3073          */
3074         btrfs_apply_pending_changes(fs_info);
3075
3076 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3077         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3078                 ret = btrfsic_mount(fs_info, fs_devices,
3079                                     btrfs_test_opt(fs_info,
3080                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3081                                     1 : 0,
3082                                     fs_info->check_integrity_print_mask);
3083                 if (ret)
3084                         btrfs_warn(fs_info,
3085                                 "failed to initialize integrity check module: %d",
3086                                 ret);
3087         }
3088 #endif
3089         ret = btrfs_read_qgroup_config(fs_info);
3090         if (ret)
3091                 goto fail_trans_kthread;
3092
3093         /* do not make disk changes in broken FS or nologreplay is given */
3094         if (btrfs_super_log_root(disk_super) != 0 &&
3095             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3096                 ret = btrfs_replay_log(fs_info, fs_devices);
3097                 if (ret) {
3098                         err = ret;
3099                         goto fail_qgroup;
3100                 }
3101         }
3102
3103         ret = btrfs_find_orphan_roots(fs_info);
3104         if (ret)
3105                 goto fail_qgroup;
3106
3107         if (!(sb->s_flags & MS_RDONLY)) {
3108                 ret = btrfs_cleanup_fs_roots(fs_info);
3109                 if (ret)
3110                         goto fail_qgroup;
3111
3112                 mutex_lock(&fs_info->cleaner_mutex);
3113                 ret = btrfs_recover_relocation(tree_root);
3114                 mutex_unlock(&fs_info->cleaner_mutex);
3115                 if (ret < 0) {
3116                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3117                                         ret);
3118                         err = -EINVAL;
3119                         goto fail_qgroup;
3120                 }
3121         }
3122
3123         location.objectid = BTRFS_FS_TREE_OBJECTID;
3124         location.type = BTRFS_ROOT_ITEM_KEY;
3125         location.offset = 0;
3126
3127         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3128         if (IS_ERR(fs_info->fs_root)) {
3129                 err = PTR_ERR(fs_info->fs_root);
3130                 goto fail_qgroup;
3131         }
3132
3133         if (sb->s_flags & MS_RDONLY)
3134                 return 0;
3135
3136         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3137             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3138                 clear_free_space_tree = 1;
3139         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3140                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3141                 btrfs_warn(fs_info, "free space tree is invalid");
3142                 clear_free_space_tree = 1;
3143         }
3144
3145         if (clear_free_space_tree) {
3146                 btrfs_info(fs_info, "clearing free space tree");
3147                 ret = btrfs_clear_free_space_tree(fs_info);
3148                 if (ret) {
3149                         btrfs_warn(fs_info,
3150                                    "failed to clear free space tree: %d", ret);
3151                         close_ctree(fs_info);
3152                         return ret;
3153                 }
3154         }
3155
3156         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3157             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3158                 btrfs_info(fs_info, "creating free space tree");
3159                 ret = btrfs_create_free_space_tree(fs_info);
3160                 if (ret) {
3161                         btrfs_warn(fs_info,
3162                                 "failed to create free space tree: %d", ret);
3163                         close_ctree(fs_info);
3164                         return ret;
3165                 }
3166         }
3167
3168         down_read(&fs_info->cleanup_work_sem);
3169         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3170             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3171                 up_read(&fs_info->cleanup_work_sem);
3172                 close_ctree(fs_info);
3173                 return ret;
3174         }
3175         up_read(&fs_info->cleanup_work_sem);
3176
3177         ret = btrfs_resume_balance_async(fs_info);
3178         if (ret) {
3179                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3180                 close_ctree(fs_info);
3181                 return ret;
3182         }
3183
3184         ret = btrfs_resume_dev_replace_async(fs_info);
3185         if (ret) {
3186                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3187                 close_ctree(fs_info);
3188                 return ret;
3189         }
3190
3191         btrfs_qgroup_rescan_resume(fs_info);
3192
3193         if (!fs_info->uuid_root) {
3194                 btrfs_info(fs_info, "creating UUID tree");
3195                 ret = btrfs_create_uuid_tree(fs_info);
3196                 if (ret) {
3197                         btrfs_warn(fs_info,
3198                                 "failed to create the UUID tree: %d", ret);
3199                         close_ctree(fs_info);
3200                         return ret;
3201                 }
3202         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3203                    fs_info->generation !=
3204                                 btrfs_super_uuid_tree_generation(disk_super)) {
3205                 btrfs_info(fs_info, "checking UUID tree");
3206                 ret = btrfs_check_uuid_tree(fs_info);
3207                 if (ret) {
3208                         btrfs_warn(fs_info,
3209                                 "failed to check the UUID tree: %d", ret);
3210                         close_ctree(fs_info);
3211                         return ret;
3212                 }
3213         } else {
3214                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3215         }
3216         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3217
3218         /*
3219          * backuproot only affect mount behavior, and if open_ctree succeeded,
3220          * no need to keep the flag
3221          */
3222         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3223
3224         return 0;
3225
3226 fail_qgroup:
3227         btrfs_free_qgroup_config(fs_info);
3228 fail_trans_kthread:
3229         kthread_stop(fs_info->transaction_kthread);
3230         btrfs_cleanup_transaction(fs_info);
3231         btrfs_free_fs_roots(fs_info);
3232 fail_cleaner:
3233         kthread_stop(fs_info->cleaner_kthread);
3234
3235         /*
3236          * make sure we're done with the btree inode before we stop our
3237          * kthreads
3238          */
3239         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3240
3241 fail_sysfs:
3242         btrfs_sysfs_remove_mounted(fs_info);
3243
3244 fail_fsdev_sysfs:
3245         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3246
3247 fail_block_groups:
3248         btrfs_put_block_group_cache(fs_info);
3249
3250 fail_tree_roots:
3251         free_root_pointers(fs_info, 1);
3252         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3253
3254 fail_sb_buffer:
3255         btrfs_stop_all_workers(fs_info);
3256         btrfs_free_block_groups(fs_info);
3257 fail_alloc:
3258 fail_iput:
3259         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3260
3261         iput(fs_info->btree_inode);
3262 fail_bio_counter:
3263         percpu_counter_destroy(&fs_info->bio_counter);
3264 fail_delalloc_bytes:
3265         percpu_counter_destroy(&fs_info->delalloc_bytes);
3266 fail_dirty_metadata_bytes:
3267         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3268 fail_srcu:
3269         cleanup_srcu_struct(&fs_info->subvol_srcu);
3270 fail:
3271         btrfs_free_stripe_hash_table(fs_info);
3272         btrfs_close_devices(fs_info->fs_devices);
3273         return err;
3274
3275 recovery_tree_root:
3276         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3277                 goto fail_tree_roots;
3278
3279         free_root_pointers(fs_info, 0);
3280
3281         /* don't use the log in recovery mode, it won't be valid */
3282         btrfs_set_super_log_root(disk_super, 0);
3283
3284         /* we can't trust the free space cache either */
3285         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3286
3287         ret = next_root_backup(fs_info, fs_info->super_copy,
3288                                &num_backups_tried, &backup_index);
3289         if (ret == -1)
3290                 goto fail_block_groups;
3291         goto retry_root_backup;
3292 }
3293
3294 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3295 {
3296         if (uptodate) {
3297                 set_buffer_uptodate(bh);
3298         } else {
3299                 struct btrfs_device *device = (struct btrfs_device *)
3300                         bh->b_private;
3301
3302                 btrfs_warn_rl_in_rcu(device->fs_info,
3303                                 "lost page write due to IO error on %s",
3304                                           rcu_str_deref(device->name));
3305                 /* note, we don't set_buffer_write_io_error because we have
3306                  * our own ways of dealing with the IO errors
3307                  */
3308                 clear_buffer_uptodate(bh);
3309                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3310         }
3311         unlock_buffer(bh);
3312         put_bh(bh);
3313 }
3314
3315 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3316                         struct buffer_head **bh_ret)
3317 {
3318         struct buffer_head *bh;
3319         struct btrfs_super_block *super;
3320         u64 bytenr;
3321
3322         bytenr = btrfs_sb_offset(copy_num);
3323         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3324                 return -EINVAL;
3325
3326         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3327         /*
3328          * If we fail to read from the underlying devices, as of now
3329          * the best option we have is to mark it EIO.
3330          */
3331         if (!bh)
3332                 return -EIO;
3333
3334         super = (struct btrfs_super_block *)bh->b_data;
3335         if (btrfs_super_bytenr(super) != bytenr ||
3336                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3337                 brelse(bh);
3338                 return -EINVAL;
3339         }
3340
3341         *bh_ret = bh;
3342         return 0;
3343 }
3344
3345
3346 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3347 {
3348         struct buffer_head *bh;
3349         struct buffer_head *latest = NULL;
3350         struct btrfs_super_block *super;
3351         int i;
3352         u64 transid = 0;
3353         int ret = -EINVAL;
3354
3355         /* we would like to check all the supers, but that would make
3356          * a btrfs mount succeed after a mkfs from a different FS.
3357          * So, we need to add a special mount option to scan for
3358          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3359          */
3360         for (i = 0; i < 1; i++) {
3361                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3362                 if (ret)
3363                         continue;
3364
3365                 super = (struct btrfs_super_block *)bh->b_data;
3366
3367                 if (!latest || btrfs_super_generation(super) > transid) {
3368                         brelse(latest);
3369                         latest = bh;
3370                         transid = btrfs_super_generation(super);
3371                 } else {
3372                         brelse(bh);
3373                 }
3374         }
3375
3376         if (!latest)
3377                 return ERR_PTR(ret);
3378
3379         return latest;
3380 }
3381
3382 /*
3383  * this should be called twice, once with wait == 0 and
3384  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3385  * we write are pinned.
3386  *
3387  * They are released when wait == 1 is done.
3388  * max_mirrors must be the same for both runs, and it indicates how
3389  * many supers on this one device should be written.
3390  *
3391  * max_mirrors == 0 means to write them all.
3392  */
3393 static int write_dev_supers(struct btrfs_device *device,
3394                             struct btrfs_super_block *sb,
3395                             int wait, int max_mirrors)
3396 {
3397         struct buffer_head *bh;
3398         int i;
3399         int ret;
3400         int errors = 0;
3401         u32 crc;
3402         u64 bytenr;
3403
3404         if (max_mirrors == 0)
3405                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3406
3407         for (i = 0; i < max_mirrors; i++) {
3408                 bytenr = btrfs_sb_offset(i);
3409                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3410                     device->commit_total_bytes)
3411                         break;
3412
3413                 if (wait) {
3414                         bh = __find_get_block(device->bdev, bytenr / 4096,
3415                                               BTRFS_SUPER_INFO_SIZE);
3416                         if (!bh) {
3417                                 errors++;
3418                                 continue;
3419                         }
3420                         wait_on_buffer(bh);
3421                         if (!buffer_uptodate(bh))
3422                                 errors++;
3423
3424                         /* drop our reference */
3425                         brelse(bh);
3426
3427                         /* drop the reference from the wait == 0 run */
3428                         brelse(bh);
3429                         continue;
3430                 } else {
3431                         btrfs_set_super_bytenr(sb, bytenr);
3432
3433                         crc = ~(u32)0;
3434                         crc = btrfs_csum_data((const char *)sb +
3435                                               BTRFS_CSUM_SIZE, crc,
3436                                               BTRFS_SUPER_INFO_SIZE -
3437                                               BTRFS_CSUM_SIZE);
3438                         btrfs_csum_final(crc, sb->csum);
3439
3440                         /*
3441                          * one reference for us, and we leave it for the
3442                          * caller
3443                          */
3444                         bh = __getblk(device->bdev, bytenr / 4096,
3445                                       BTRFS_SUPER_INFO_SIZE);
3446                         if (!bh) {
3447                                 btrfs_err(device->fs_info,
3448                                     "couldn't get super buffer head for bytenr %llu",
3449                                     bytenr);
3450                                 errors++;
3451                                 continue;
3452                         }
3453
3454                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3455
3456                         /* one reference for submit_bh */
3457                         get_bh(bh);
3458
3459                         set_buffer_uptodate(bh);
3460                         lock_buffer(bh);
3461                         bh->b_end_io = btrfs_end_buffer_write_sync;
3462                         bh->b_private = device;
3463                 }
3464
3465                 /*
3466                  * we fua the first super.  The others we allow
3467                  * to go down lazy.
3468                  */
3469                 if (i == 0) {
3470                         ret = btrfsic_submit_bh(REQ_OP_WRITE,
3471                                                 REQ_SYNC | REQ_FUA, bh);
3472                 } else {
3473                         ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
3474                 }
3475                 if (ret)
3476                         errors++;
3477         }
3478         return errors < i ? 0 : -1;
3479 }
3480
3481 /*
3482  * endio for the write_dev_flush, this will wake anyone waiting
3483  * for the barrier when it is done
3484  */
3485 static void btrfs_end_empty_barrier(struct bio *bio)
3486 {
3487         if (bio->bi_private)
3488                 complete(bio->bi_private);
3489         bio_put(bio);
3490 }
3491
3492 /*
3493  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3494  * sent down.  With wait == 1, it waits for the previous flush.
3495  *
3496  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3497  * capable
3498  */
3499 static int write_dev_flush(struct btrfs_device *device, int wait)
3500 {
3501         struct request_queue *q = bdev_get_queue(device->bdev);
3502         struct bio *bio;
3503         int ret = 0;
3504
3505         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3506                 return 0;
3507
3508         if (wait) {
3509                 bio = device->flush_bio;
3510                 if (!bio)
3511                         /*
3512                          * This means the alloc has failed with ENOMEM, however
3513                          * here we return 0, as its not a device error.
3514                          */
3515                         return 0;
3516
3517                 wait_for_completion(&device->flush_wait);
3518
3519                 if (bio->bi_error) {
3520                         ret = bio->bi_error;
3521                         btrfs_dev_stat_inc_and_print(device,
3522                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3523                 }
3524
3525                 /* drop the reference from the wait == 0 run */
3526                 bio_put(bio);
3527                 device->flush_bio = NULL;
3528
3529                 return ret;
3530         }
3531
3532         /*
3533          * one reference for us, and we leave it for the
3534          * caller
3535          */
3536         device->flush_bio = NULL;
3537         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3538         if (!bio)
3539                 return -ENOMEM;
3540
3541         bio->bi_end_io = btrfs_end_empty_barrier;
3542         bio->bi_bdev = device->bdev;
3543         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3544         init_completion(&device->flush_wait);
3545         bio->bi_private = &device->flush_wait;
3546         device->flush_bio = bio;
3547
3548         bio_get(bio);
3549         btrfsic_submit_bio(bio);
3550
3551         return 0;
3552 }
3553
3554 static int check_barrier_error(struct btrfs_fs_devices *fsdevs)
3555 {
3556         int submit_flush_error = 0;
3557         int dev_flush_error = 0;
3558         struct btrfs_device *dev;
3559         int tolerance;
3560