Btrfs: move the R/O check out of btrfs_clean_one_deleted_snapshot()
[linux-2.6-block.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/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <linux/uuid.h>
34 #include <asm/unaligned.h>
35 #include "compat.h"
36 #include "ctree.h"
37 #include "disk-io.h"
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "volumes.h"
41 #include "print-tree.h"
42 #include "async-thread.h"
43 #include "locking.h"
44 #include "tree-log.h"
45 #include "free-space-cache.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
50 #include "raid56.h"
51
52 #ifdef CONFIG_X86
53 #include <asm/cpufeature.h>
54 #endif
55
56 static struct extent_io_ops btree_extent_io_ops;
57 static void end_workqueue_fn(struct btrfs_work *work);
58 static void free_fs_root(struct btrfs_root *root);
59 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
60                                     int read_only);
61 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
62                                              struct btrfs_root *root);
63 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
64 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65                                       struct btrfs_root *root);
66 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
67 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
68 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69                                         struct extent_io_tree *dirty_pages,
70                                         int mark);
71 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72                                        struct extent_io_tree *pinned_extents);
73 static int btrfs_cleanup_transaction(struct btrfs_root *root);
74 static void btrfs_error_commit_super(struct btrfs_root *root);
75
76 /*
77  * end_io_wq structs are used to do processing in task context when an IO is
78  * complete.  This is used during reads to verify checksums, and it is used
79  * by writes to insert metadata for new file extents after IO is complete.
80  */
81 struct end_io_wq {
82         struct bio *bio;
83         bio_end_io_t *end_io;
84         void *private;
85         struct btrfs_fs_info *info;
86         int error;
87         int metadata;
88         struct list_head list;
89         struct btrfs_work work;
90 };
91
92 /*
93  * async submit bios are used to offload expensive checksumming
94  * onto the worker threads.  They checksum file and metadata bios
95  * just before they are sent down the IO stack.
96  */
97 struct async_submit_bio {
98         struct inode *inode;
99         struct bio *bio;
100         struct list_head list;
101         extent_submit_bio_hook_t *submit_bio_start;
102         extent_submit_bio_hook_t *submit_bio_done;
103         int rw;
104         int mirror_num;
105         unsigned long bio_flags;
106         /*
107          * bio_offset is optional, can be used if the pages in the bio
108          * can't tell us where in the file the bio should go
109          */
110         u64 bio_offset;
111         struct btrfs_work work;
112         int error;
113 };
114
115 /*
116  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
117  * eb, the lockdep key is determined by the btrfs_root it belongs to and
118  * the level the eb occupies in the tree.
119  *
120  * Different roots are used for different purposes and may nest inside each
121  * other and they require separate keysets.  As lockdep keys should be
122  * static, assign keysets according to the purpose of the root as indicated
123  * by btrfs_root->objectid.  This ensures that all special purpose roots
124  * have separate keysets.
125  *
126  * Lock-nesting across peer nodes is always done with the immediate parent
127  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
128  * subclass to avoid triggering lockdep warning in such cases.
129  *
130  * The key is set by the readpage_end_io_hook after the buffer has passed
131  * csum validation but before the pages are unlocked.  It is also set by
132  * btrfs_init_new_buffer on freshly allocated blocks.
133  *
134  * We also add a check to make sure the highest level of the tree is the
135  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
136  * needs update as well.
137  */
138 #ifdef CONFIG_DEBUG_LOCK_ALLOC
139 # if BTRFS_MAX_LEVEL != 8
140 #  error
141 # endif
142
143 static struct btrfs_lockdep_keyset {
144         u64                     id;             /* root objectid */
145         const char              *name_stem;     /* lock name stem */
146         char                    names[BTRFS_MAX_LEVEL + 1][20];
147         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
148 } btrfs_lockdep_keysets[] = {
149         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
150         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
151         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
152         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
153         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
154         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
155         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
156         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
157         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
158         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
159         { .id = 0,                              .name_stem = "tree"     },
160 };
161
162 void __init btrfs_init_lockdep(void)
163 {
164         int i, j;
165
166         /* initialize lockdep class names */
167         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
168                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
169
170                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
171                         snprintf(ks->names[j], sizeof(ks->names[j]),
172                                  "btrfs-%s-%02d", ks->name_stem, j);
173         }
174 }
175
176 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
177                                     int level)
178 {
179         struct btrfs_lockdep_keyset *ks;
180
181         BUG_ON(level >= ARRAY_SIZE(ks->keys));
182
183         /* find the matching keyset, id 0 is the default entry */
184         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
185                 if (ks->id == objectid)
186                         break;
187
188         lockdep_set_class_and_name(&eb->lock,
189                                    &ks->keys[level], ks->names[level]);
190 }
191
192 #endif
193
194 /*
195  * extents on the btree inode are pretty simple, there's one extent
196  * that covers the entire device
197  */
198 static struct extent_map *btree_get_extent(struct inode *inode,
199                 struct page *page, size_t pg_offset, u64 start, u64 len,
200                 int create)
201 {
202         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
203         struct extent_map *em;
204         int ret;
205
206         read_lock(&em_tree->lock);
207         em = lookup_extent_mapping(em_tree, start, len);
208         if (em) {
209                 em->bdev =
210                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
211                 read_unlock(&em_tree->lock);
212                 goto out;
213         }
214         read_unlock(&em_tree->lock);
215
216         em = alloc_extent_map();
217         if (!em) {
218                 em = ERR_PTR(-ENOMEM);
219                 goto out;
220         }
221         em->start = 0;
222         em->len = (u64)-1;
223         em->block_len = (u64)-1;
224         em->block_start = 0;
225         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
226
227         write_lock(&em_tree->lock);
228         ret = add_extent_mapping(em_tree, em, 0);
229         if (ret == -EEXIST) {
230                 free_extent_map(em);
231                 em = lookup_extent_mapping(em_tree, start, len);
232                 if (!em)
233                         em = ERR_PTR(-EIO);
234         } else if (ret) {
235                 free_extent_map(em);
236                 em = ERR_PTR(ret);
237         }
238         write_unlock(&em_tree->lock);
239
240 out:
241         return em;
242 }
243
244 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
245 {
246         return crc32c(seed, data, len);
247 }
248
249 void btrfs_csum_final(u32 crc, char *result)
250 {
251         put_unaligned_le32(~crc, result);
252 }
253
254 /*
255  * compute the csum for a btree block, and either verify it or write it
256  * into the csum field of the block.
257  */
258 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
259                            int verify)
260 {
261         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
262         char *result = NULL;
263         unsigned long len;
264         unsigned long cur_len;
265         unsigned long offset = BTRFS_CSUM_SIZE;
266         char *kaddr;
267         unsigned long map_start;
268         unsigned long map_len;
269         int err;
270         u32 crc = ~(u32)0;
271         unsigned long inline_result;
272
273         len = buf->len - offset;
274         while (len > 0) {
275                 err = map_private_extent_buffer(buf, offset, 32,
276                                         &kaddr, &map_start, &map_len);
277                 if (err)
278                         return 1;
279                 cur_len = min(len, map_len - (offset - map_start));
280                 crc = btrfs_csum_data(kaddr + offset - map_start,
281                                       crc, cur_len);
282                 len -= cur_len;
283                 offset += cur_len;
284         }
285         if (csum_size > sizeof(inline_result)) {
286                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
287                 if (!result)
288                         return 1;
289         } else {
290                 result = (char *)&inline_result;
291         }
292
293         btrfs_csum_final(crc, result);
294
295         if (verify) {
296                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
297                         u32 val;
298                         u32 found = 0;
299                         memcpy(&found, result, csum_size);
300
301                         read_extent_buffer(buf, &val, 0, csum_size);
302                         printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
303                                        "failed on %llu wanted %X found %X "
304                                        "level %d\n",
305                                        root->fs_info->sb->s_id,
306                                        (unsigned long long)buf->start, val, found,
307                                        btrfs_header_level(buf));
308                         if (result != (char *)&inline_result)
309                                 kfree(result);
310                         return 1;
311                 }
312         } else {
313                 write_extent_buffer(buf, result, 0, csum_size);
314         }
315         if (result != (char *)&inline_result)
316                 kfree(result);
317         return 0;
318 }
319
320 /*
321  * we can't consider a given block up to date unless the transid of the
322  * block matches the transid in the parent node's pointer.  This is how we
323  * detect blocks that either didn't get written at all or got written
324  * in the wrong place.
325  */
326 static int verify_parent_transid(struct extent_io_tree *io_tree,
327                                  struct extent_buffer *eb, u64 parent_transid,
328                                  int atomic)
329 {
330         struct extent_state *cached_state = NULL;
331         int ret;
332
333         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334                 return 0;
335
336         if (atomic)
337                 return -EAGAIN;
338
339         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
340                          0, &cached_state);
341         if (extent_buffer_uptodate(eb) &&
342             btrfs_header_generation(eb) == parent_transid) {
343                 ret = 0;
344                 goto out;
345         }
346         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
347                        "found %llu\n",
348                        (unsigned long long)eb->start,
349                        (unsigned long long)parent_transid,
350                        (unsigned long long)btrfs_header_generation(eb));
351         ret = 1;
352         clear_extent_buffer_uptodate(eb);
353 out:
354         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355                              &cached_state, GFP_NOFS);
356         return ret;
357 }
358
359 /*
360  * Return 0 if the superblock checksum type matches the checksum value of that
361  * algorithm. Pass the raw disk superblock data.
362  */
363 static int btrfs_check_super_csum(char *raw_disk_sb)
364 {
365         struct btrfs_super_block *disk_sb =
366                 (struct btrfs_super_block *)raw_disk_sb;
367         u16 csum_type = btrfs_super_csum_type(disk_sb);
368         int ret = 0;
369
370         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
371                 u32 crc = ~(u32)0;
372                 const int csum_size = sizeof(crc);
373                 char result[csum_size];
374
375                 /*
376                  * The super_block structure does not span the whole
377                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
378                  * is filled with zeros and is included in the checkum.
379                  */
380                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
381                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
382                 btrfs_csum_final(crc, result);
383
384                 if (memcmp(raw_disk_sb, result, csum_size))
385                         ret = 1;
386
387                 if (ret && btrfs_super_generation(disk_sb) < 10) {
388                         printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
389                         ret = 0;
390                 }
391         }
392
393         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
394                 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
395                                 csum_type);
396                 ret = 1;
397         }
398
399         return ret;
400 }
401
402 /*
403  * helper to read a given tree block, doing retries as required when
404  * the checksums don't match and we have alternate mirrors to try.
405  */
406 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
407                                           struct extent_buffer *eb,
408                                           u64 start, u64 parent_transid)
409 {
410         struct extent_io_tree *io_tree;
411         int failed = 0;
412         int ret;
413         int num_copies = 0;
414         int mirror_num = 0;
415         int failed_mirror = 0;
416
417         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
418         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
419         while (1) {
420                 ret = read_extent_buffer_pages(io_tree, eb, start,
421                                                WAIT_COMPLETE,
422                                                btree_get_extent, mirror_num);
423                 if (!ret) {
424                         if (!verify_parent_transid(io_tree, eb,
425                                                    parent_transid, 0))
426                                 break;
427                         else
428                                 ret = -EIO;
429                 }
430
431                 /*
432                  * This buffer's crc is fine, but its contents are corrupted, so
433                  * there is no reason to read the other copies, they won't be
434                  * any less wrong.
435                  */
436                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
437                         break;
438
439                 num_copies = btrfs_num_copies(root->fs_info,
440                                               eb->start, eb->len);
441                 if (num_copies == 1)
442                         break;
443
444                 if (!failed_mirror) {
445                         failed = 1;
446                         failed_mirror = eb->read_mirror;
447                 }
448
449                 mirror_num++;
450                 if (mirror_num == failed_mirror)
451                         mirror_num++;
452
453                 if (mirror_num > num_copies)
454                         break;
455         }
456
457         if (failed && !ret && failed_mirror)
458                 repair_eb_io_failure(root, eb, failed_mirror);
459
460         return ret;
461 }
462
463 /*
464  * checksum a dirty tree block before IO.  This has extra checks to make sure
465  * we only fill in the checksum field in the first page of a multi-page block
466  */
467
468 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
469 {
470         struct extent_io_tree *tree;
471         u64 start = page_offset(page);
472         u64 found_start;
473         struct extent_buffer *eb;
474
475         tree = &BTRFS_I(page->mapping->host)->io_tree;
476
477         eb = (struct extent_buffer *)page->private;
478         if (page != eb->pages[0])
479                 return 0;
480         found_start = btrfs_header_bytenr(eb);
481         if (found_start != start) {
482                 WARN_ON(1);
483                 return 0;
484         }
485         if (!PageUptodate(page)) {
486                 WARN_ON(1);
487                 return 0;
488         }
489         csum_tree_block(root, eb, 0);
490         return 0;
491 }
492
493 static int check_tree_block_fsid(struct btrfs_root *root,
494                                  struct extent_buffer *eb)
495 {
496         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
497         u8 fsid[BTRFS_UUID_SIZE];
498         int ret = 1;
499
500         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
501                            BTRFS_FSID_SIZE);
502         while (fs_devices) {
503                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
504                         ret = 0;
505                         break;
506                 }
507                 fs_devices = fs_devices->seed;
508         }
509         return ret;
510 }
511
512 #define CORRUPT(reason, eb, root, slot)                         \
513         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
514                "root=%llu, slot=%d\n", reason,                  \
515                (unsigned long long)btrfs_header_bytenr(eb),     \
516                (unsigned long long)root->objectid, slot)
517
518 static noinline int check_leaf(struct btrfs_root *root,
519                                struct extent_buffer *leaf)
520 {
521         struct btrfs_key key;
522         struct btrfs_key leaf_key;
523         u32 nritems = btrfs_header_nritems(leaf);
524         int slot;
525
526         if (nritems == 0)
527                 return 0;
528
529         /* Check the 0 item */
530         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
531             BTRFS_LEAF_DATA_SIZE(root)) {
532                 CORRUPT("invalid item offset size pair", leaf, root, 0);
533                 return -EIO;
534         }
535
536         /*
537          * Check to make sure each items keys are in the correct order and their
538          * offsets make sense.  We only have to loop through nritems-1 because
539          * we check the current slot against the next slot, which verifies the
540          * next slot's offset+size makes sense and that the current's slot
541          * offset is correct.
542          */
543         for (slot = 0; slot < nritems - 1; slot++) {
544                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
545                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
546
547                 /* Make sure the keys are in the right order */
548                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
549                         CORRUPT("bad key order", leaf, root, slot);
550                         return -EIO;
551                 }
552
553                 /*
554                  * Make sure the offset and ends are right, remember that the
555                  * item data starts at the end of the leaf and grows towards the
556                  * front.
557                  */
558                 if (btrfs_item_offset_nr(leaf, slot) !=
559                         btrfs_item_end_nr(leaf, slot + 1)) {
560                         CORRUPT("slot offset bad", leaf, root, slot);
561                         return -EIO;
562                 }
563
564                 /*
565                  * Check to make sure that we don't point outside of the leaf,
566                  * just incase all the items are consistent to eachother, but
567                  * all point outside of the leaf.
568                  */
569                 if (btrfs_item_end_nr(leaf, slot) >
570                     BTRFS_LEAF_DATA_SIZE(root)) {
571                         CORRUPT("slot end outside of leaf", leaf, root, slot);
572                         return -EIO;
573                 }
574         }
575
576         return 0;
577 }
578
579 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
580                                struct extent_state *state, int mirror)
581 {
582         struct extent_io_tree *tree;
583         u64 found_start;
584         int found_level;
585         struct extent_buffer *eb;
586         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
587         int ret = 0;
588         int reads_done;
589
590         if (!page->private)
591                 goto out;
592
593         tree = &BTRFS_I(page->mapping->host)->io_tree;
594         eb = (struct extent_buffer *)page->private;
595
596         /* the pending IO might have been the only thing that kept this buffer
597          * in memory.  Make sure we have a ref for all this other checks
598          */
599         extent_buffer_get(eb);
600
601         reads_done = atomic_dec_and_test(&eb->io_pages);
602         if (!reads_done)
603                 goto err;
604
605         eb->read_mirror = mirror;
606         if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
607                 ret = -EIO;
608                 goto err;
609         }
610
611         found_start = btrfs_header_bytenr(eb);
612         if (found_start != eb->start) {
613                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
614                                "%llu %llu\n",
615                                (unsigned long long)found_start,
616                                (unsigned long long)eb->start);
617                 ret = -EIO;
618                 goto err;
619         }
620         if (check_tree_block_fsid(root, eb)) {
621                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
622                                (unsigned long long)eb->start);
623                 ret = -EIO;
624                 goto err;
625         }
626         found_level = btrfs_header_level(eb);
627         if (found_level >= BTRFS_MAX_LEVEL) {
628                 btrfs_info(root->fs_info, "bad tree block level %d\n",
629                            (int)btrfs_header_level(eb));
630                 ret = -EIO;
631                 goto err;
632         }
633
634         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
635                                        eb, found_level);
636
637         ret = csum_tree_block(root, eb, 1);
638         if (ret) {
639                 ret = -EIO;
640                 goto err;
641         }
642
643         /*
644          * If this is a leaf block and it is corrupt, set the corrupt bit so
645          * that we don't try and read the other copies of this block, just
646          * return -EIO.
647          */
648         if (found_level == 0 && check_leaf(root, eb)) {
649                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
650                 ret = -EIO;
651         }
652
653         if (!ret)
654                 set_extent_buffer_uptodate(eb);
655 err:
656         if (reads_done &&
657             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
658                 btree_readahead_hook(root, eb, eb->start, ret);
659
660         if (ret) {
661                 /*
662                  * our io error hook is going to dec the io pages
663                  * again, we have to make sure it has something
664                  * to decrement
665                  */
666                 atomic_inc(&eb->io_pages);
667                 clear_extent_buffer_uptodate(eb);
668         }
669         free_extent_buffer(eb);
670 out:
671         return ret;
672 }
673
674 static int btree_io_failed_hook(struct page *page, int failed_mirror)
675 {
676         struct extent_buffer *eb;
677         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
678
679         eb = (struct extent_buffer *)page->private;
680         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
681         eb->read_mirror = failed_mirror;
682         atomic_dec(&eb->io_pages);
683         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
684                 btree_readahead_hook(root, eb, eb->start, -EIO);
685         return -EIO;    /* we fixed nothing */
686 }
687
688 static void end_workqueue_bio(struct bio *bio, int err)
689 {
690         struct end_io_wq *end_io_wq = bio->bi_private;
691         struct btrfs_fs_info *fs_info;
692
693         fs_info = end_io_wq->info;
694         end_io_wq->error = err;
695         end_io_wq->work.func = end_workqueue_fn;
696         end_io_wq->work.flags = 0;
697
698         if (bio->bi_rw & REQ_WRITE) {
699                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
700                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
701                                            &end_io_wq->work);
702                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
703                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
704                                            &end_io_wq->work);
705                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
706                         btrfs_queue_worker(&fs_info->endio_raid56_workers,
707                                            &end_io_wq->work);
708                 else
709                         btrfs_queue_worker(&fs_info->endio_write_workers,
710                                            &end_io_wq->work);
711         } else {
712                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
713                         btrfs_queue_worker(&fs_info->endio_raid56_workers,
714                                            &end_io_wq->work);
715                 else if (end_io_wq->metadata)
716                         btrfs_queue_worker(&fs_info->endio_meta_workers,
717                                            &end_io_wq->work);
718                 else
719                         btrfs_queue_worker(&fs_info->endio_workers,
720                                            &end_io_wq->work);
721         }
722 }
723
724 /*
725  * For the metadata arg you want
726  *
727  * 0 - if data
728  * 1 - if normal metadta
729  * 2 - if writing to the free space cache area
730  * 3 - raid parity work
731  */
732 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
733                         int metadata)
734 {
735         struct end_io_wq *end_io_wq;
736         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
737         if (!end_io_wq)
738                 return -ENOMEM;
739
740         end_io_wq->private = bio->bi_private;
741         end_io_wq->end_io = bio->bi_end_io;
742         end_io_wq->info = info;
743         end_io_wq->error = 0;
744         end_io_wq->bio = bio;
745         end_io_wq->metadata = metadata;
746
747         bio->bi_private = end_io_wq;
748         bio->bi_end_io = end_workqueue_bio;
749         return 0;
750 }
751
752 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
753 {
754         unsigned long limit = min_t(unsigned long,
755                                     info->workers.max_workers,
756                                     info->fs_devices->open_devices);
757         return 256 * limit;
758 }
759
760 static void run_one_async_start(struct btrfs_work *work)
761 {
762         struct async_submit_bio *async;
763         int ret;
764
765         async = container_of(work, struct  async_submit_bio, work);
766         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
767                                       async->mirror_num, async->bio_flags,
768                                       async->bio_offset);
769         if (ret)
770                 async->error = ret;
771 }
772
773 static void run_one_async_done(struct btrfs_work *work)
774 {
775         struct btrfs_fs_info *fs_info;
776         struct async_submit_bio *async;
777         int limit;
778
779         async = container_of(work, struct  async_submit_bio, work);
780         fs_info = BTRFS_I(async->inode)->root->fs_info;
781
782         limit = btrfs_async_submit_limit(fs_info);
783         limit = limit * 2 / 3;
784
785         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
786             waitqueue_active(&fs_info->async_submit_wait))
787                 wake_up(&fs_info->async_submit_wait);
788
789         /* If an error occured we just want to clean up the bio and move on */
790         if (async->error) {
791                 bio_endio(async->bio, async->error);
792                 return;
793         }
794
795         async->submit_bio_done(async->inode, async->rw, async->bio,
796                                async->mirror_num, async->bio_flags,
797                                async->bio_offset);
798 }
799
800 static void run_one_async_free(struct btrfs_work *work)
801 {
802         struct async_submit_bio *async;
803
804         async = container_of(work, struct  async_submit_bio, work);
805         kfree(async);
806 }
807
808 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
809                         int rw, struct bio *bio, int mirror_num,
810                         unsigned long bio_flags,
811                         u64 bio_offset,
812                         extent_submit_bio_hook_t *submit_bio_start,
813                         extent_submit_bio_hook_t *submit_bio_done)
814 {
815         struct async_submit_bio *async;
816
817         async = kmalloc(sizeof(*async), GFP_NOFS);
818         if (!async)
819                 return -ENOMEM;
820
821         async->inode = inode;
822         async->rw = rw;
823         async->bio = bio;
824         async->mirror_num = mirror_num;
825         async->submit_bio_start = submit_bio_start;
826         async->submit_bio_done = submit_bio_done;
827
828         async->work.func = run_one_async_start;
829         async->work.ordered_func = run_one_async_done;
830         async->work.ordered_free = run_one_async_free;
831
832         async->work.flags = 0;
833         async->bio_flags = bio_flags;
834         async->bio_offset = bio_offset;
835
836         async->error = 0;
837
838         atomic_inc(&fs_info->nr_async_submits);
839
840         if (rw & REQ_SYNC)
841                 btrfs_set_work_high_prio(&async->work);
842
843         btrfs_queue_worker(&fs_info->workers, &async->work);
844
845         while (atomic_read(&fs_info->async_submit_draining) &&
846               atomic_read(&fs_info->nr_async_submits)) {
847                 wait_event(fs_info->async_submit_wait,
848                            (atomic_read(&fs_info->nr_async_submits) == 0));
849         }
850
851         return 0;
852 }
853
854 static int btree_csum_one_bio(struct bio *bio)
855 {
856         struct bio_vec *bvec = bio->bi_io_vec;
857         int bio_index = 0;
858         struct btrfs_root *root;
859         int ret = 0;
860
861         WARN_ON(bio->bi_vcnt <= 0);
862         while (bio_index < bio->bi_vcnt) {
863                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
864                 ret = csum_dirty_buffer(root, bvec->bv_page);
865                 if (ret)
866                         break;
867                 bio_index++;
868                 bvec++;
869         }
870         return ret;
871 }
872
873 static int __btree_submit_bio_start(struct inode *inode, int rw,
874                                     struct bio *bio, int mirror_num,
875                                     unsigned long bio_flags,
876                                     u64 bio_offset)
877 {
878         /*
879          * when we're called for a write, we're already in the async
880          * submission context.  Just jump into btrfs_map_bio
881          */
882         return btree_csum_one_bio(bio);
883 }
884
885 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
886                                  int mirror_num, unsigned long bio_flags,
887                                  u64 bio_offset)
888 {
889         int ret;
890
891         /*
892          * when we're called for a write, we're already in the async
893          * submission context.  Just jump into btrfs_map_bio
894          */
895         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
896         if (ret)
897                 bio_endio(bio, ret);
898         return ret;
899 }
900
901 static int check_async_write(struct inode *inode, unsigned long bio_flags)
902 {
903         if (bio_flags & EXTENT_BIO_TREE_LOG)
904                 return 0;
905 #ifdef CONFIG_X86
906         if (cpu_has_xmm4_2)
907                 return 0;
908 #endif
909         return 1;
910 }
911
912 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
913                                  int mirror_num, unsigned long bio_flags,
914                                  u64 bio_offset)
915 {
916         int async = check_async_write(inode, bio_flags);
917         int ret;
918
919         if (!(rw & REQ_WRITE)) {
920                 /*
921                  * called for a read, do the setup so that checksum validation
922                  * can happen in the async kernel threads
923                  */
924                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
925                                           bio, 1);
926                 if (ret)
927                         goto out_w_error;
928                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
929                                     mirror_num, 0);
930         } else if (!async) {
931                 ret = btree_csum_one_bio(bio);
932                 if (ret)
933                         goto out_w_error;
934                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
935                                     mirror_num, 0);
936         } else {
937                 /*
938                  * kthread helpers are used to submit writes so that
939                  * checksumming can happen in parallel across all CPUs
940                  */
941                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
942                                           inode, rw, bio, mirror_num, 0,
943                                           bio_offset,
944                                           __btree_submit_bio_start,
945                                           __btree_submit_bio_done);
946         }
947
948         if (ret) {
949 out_w_error:
950                 bio_endio(bio, ret);
951         }
952         return ret;
953 }
954
955 #ifdef CONFIG_MIGRATION
956 static int btree_migratepage(struct address_space *mapping,
957                         struct page *newpage, struct page *page,
958                         enum migrate_mode mode)
959 {
960         /*
961          * we can't safely write a btree page from here,
962          * we haven't done the locking hook
963          */
964         if (PageDirty(page))
965                 return -EAGAIN;
966         /*
967          * Buffers may be managed in a filesystem specific way.
968          * We must have no buffers or drop them.
969          */
970         if (page_has_private(page) &&
971             !try_to_release_page(page, GFP_KERNEL))
972                 return -EAGAIN;
973         return migrate_page(mapping, newpage, page, mode);
974 }
975 #endif
976
977
978 static int btree_writepages(struct address_space *mapping,
979                             struct writeback_control *wbc)
980 {
981         struct extent_io_tree *tree;
982         struct btrfs_fs_info *fs_info;
983         int ret;
984
985         tree = &BTRFS_I(mapping->host)->io_tree;
986         if (wbc->sync_mode == WB_SYNC_NONE) {
987
988                 if (wbc->for_kupdate)
989                         return 0;
990
991                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
992                 /* this is a bit racy, but that's ok */
993                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
994                                              BTRFS_DIRTY_METADATA_THRESH);
995                 if (ret < 0)
996                         return 0;
997         }
998         return btree_write_cache_pages(mapping, wbc);
999 }
1000
1001 static int btree_readpage(struct file *file, struct page *page)
1002 {
1003         struct extent_io_tree *tree;
1004         tree = &BTRFS_I(page->mapping->host)->io_tree;
1005         return extent_read_full_page(tree, page, btree_get_extent, 0);
1006 }
1007
1008 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1009 {
1010         if (PageWriteback(page) || PageDirty(page))
1011                 return 0;
1012
1013         return try_release_extent_buffer(page);
1014 }
1015
1016 static void btree_invalidatepage(struct page *page, unsigned long offset)
1017 {
1018         struct extent_io_tree *tree;
1019         tree = &BTRFS_I(page->mapping->host)->io_tree;
1020         extent_invalidatepage(tree, page, offset);
1021         btree_releasepage(page, GFP_NOFS);
1022         if (PagePrivate(page)) {
1023                 printk(KERN_WARNING "btrfs warning page private not zero "
1024                        "on page %llu\n", (unsigned long long)page_offset(page));
1025                 ClearPagePrivate(page);
1026                 set_page_private(page, 0);
1027                 page_cache_release(page);
1028         }
1029 }
1030
1031 static int btree_set_page_dirty(struct page *page)
1032 {
1033 #ifdef DEBUG
1034         struct extent_buffer *eb;
1035
1036         BUG_ON(!PagePrivate(page));
1037         eb = (struct extent_buffer *)page->private;
1038         BUG_ON(!eb);
1039         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1040         BUG_ON(!atomic_read(&eb->refs));
1041         btrfs_assert_tree_locked(eb);
1042 #endif
1043         return __set_page_dirty_nobuffers(page);
1044 }
1045
1046 static const struct address_space_operations btree_aops = {
1047         .readpage       = btree_readpage,
1048         .writepages     = btree_writepages,
1049         .releasepage    = btree_releasepage,
1050         .invalidatepage = btree_invalidatepage,
1051 #ifdef CONFIG_MIGRATION
1052         .migratepage    = btree_migratepage,
1053 #endif
1054         .set_page_dirty = btree_set_page_dirty,
1055 };
1056
1057 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1058                          u64 parent_transid)
1059 {
1060         struct extent_buffer *buf = NULL;
1061         struct inode *btree_inode = root->fs_info->btree_inode;
1062         int ret = 0;
1063
1064         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1065         if (!buf)
1066                 return 0;
1067         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1068                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1069         free_extent_buffer(buf);
1070         return ret;
1071 }
1072
1073 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1074                          int mirror_num, struct extent_buffer **eb)
1075 {
1076         struct extent_buffer *buf = NULL;
1077         struct inode *btree_inode = root->fs_info->btree_inode;
1078         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1079         int ret;
1080
1081         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1082         if (!buf)
1083                 return 0;
1084
1085         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1086
1087         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1088                                        btree_get_extent, mirror_num);
1089         if (ret) {
1090                 free_extent_buffer(buf);
1091                 return ret;
1092         }
1093
1094         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1095                 free_extent_buffer(buf);
1096                 return -EIO;
1097         } else if (extent_buffer_uptodate(buf)) {
1098                 *eb = buf;
1099         } else {
1100                 free_extent_buffer(buf);
1101         }
1102         return 0;
1103 }
1104
1105 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1106                                             u64 bytenr, u32 blocksize)
1107 {
1108         struct inode *btree_inode = root->fs_info->btree_inode;
1109         struct extent_buffer *eb;
1110         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1111                                 bytenr, blocksize);
1112         return eb;
1113 }
1114
1115 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1116                                                  u64 bytenr, u32 blocksize)
1117 {
1118         struct inode *btree_inode = root->fs_info->btree_inode;
1119         struct extent_buffer *eb;
1120
1121         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1122                                  bytenr, blocksize);
1123         return eb;
1124 }
1125
1126
1127 int btrfs_write_tree_block(struct extent_buffer *buf)
1128 {
1129         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1130                                         buf->start + buf->len - 1);
1131 }
1132
1133 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1134 {
1135         return filemap_fdatawait_range(buf->pages[0]->mapping,
1136                                        buf->start, buf->start + buf->len - 1);
1137 }
1138
1139 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1140                                       u32 blocksize, u64 parent_transid)
1141 {
1142         struct extent_buffer *buf = NULL;
1143         int ret;
1144
1145         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1146         if (!buf)
1147                 return NULL;
1148
1149         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1150         return buf;
1151
1152 }
1153
1154 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1155                       struct extent_buffer *buf)
1156 {
1157         struct btrfs_fs_info *fs_info = root->fs_info;
1158
1159         if (btrfs_header_generation(buf) ==
1160             fs_info->running_transaction->transid) {
1161                 btrfs_assert_tree_locked(buf);
1162
1163                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1164                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1165                                              -buf->len,
1166                                              fs_info->dirty_metadata_batch);
1167                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1168                         btrfs_set_lock_blocking(buf);
1169                         clear_extent_buffer_dirty(buf);
1170                 }
1171         }
1172 }
1173
1174 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1175                          u32 stripesize, struct btrfs_root *root,
1176                          struct btrfs_fs_info *fs_info,
1177                          u64 objectid)
1178 {
1179         root->node = NULL;
1180         root->commit_root = NULL;
1181         root->sectorsize = sectorsize;
1182         root->nodesize = nodesize;
1183         root->leafsize = leafsize;
1184         root->stripesize = stripesize;
1185         root->ref_cows = 0;
1186         root->track_dirty = 0;
1187         root->in_radix = 0;
1188         root->orphan_item_inserted = 0;
1189         root->orphan_cleanup_state = 0;
1190
1191         root->objectid = objectid;
1192         root->last_trans = 0;
1193         root->highest_objectid = 0;
1194         root->name = NULL;
1195         root->inode_tree = RB_ROOT;
1196         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1197         root->block_rsv = NULL;
1198         root->orphan_block_rsv = NULL;
1199
1200         INIT_LIST_HEAD(&root->dirty_list);
1201         INIT_LIST_HEAD(&root->root_list);
1202         INIT_LIST_HEAD(&root->logged_list[0]);
1203         INIT_LIST_HEAD(&root->logged_list[1]);
1204         spin_lock_init(&root->orphan_lock);
1205         spin_lock_init(&root->inode_lock);
1206         spin_lock_init(&root->accounting_lock);
1207         spin_lock_init(&root->log_extents_lock[0]);
1208         spin_lock_init(&root->log_extents_lock[1]);
1209         mutex_init(&root->objectid_mutex);
1210         mutex_init(&root->log_mutex);
1211         init_waitqueue_head(&root->log_writer_wait);
1212         init_waitqueue_head(&root->log_commit_wait[0]);
1213         init_waitqueue_head(&root->log_commit_wait[1]);
1214         atomic_set(&root->log_commit[0], 0);
1215         atomic_set(&root->log_commit[1], 0);
1216         atomic_set(&root->log_writers, 0);
1217         atomic_set(&root->log_batch, 0);
1218         atomic_set(&root->orphan_inodes, 0);
1219         root->log_transid = 0;
1220         root->last_log_commit = 0;
1221         extent_io_tree_init(&root->dirty_log_pages,
1222                              fs_info->btree_inode->i_mapping);
1223
1224         memset(&root->root_key, 0, sizeof(root->root_key));
1225         memset(&root->root_item, 0, sizeof(root->root_item));
1226         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1227         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1228         root->defrag_trans_start = fs_info->generation;
1229         init_completion(&root->kobj_unregister);
1230         root->defrag_running = 0;
1231         root->root_key.objectid = objectid;
1232         root->anon_dev = 0;
1233
1234         spin_lock_init(&root->root_item_lock);
1235 }
1236
1237 static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1238                                             struct btrfs_fs_info *fs_info,
1239                                             u64 objectid,
1240                                             struct btrfs_root *root)
1241 {
1242         int ret;
1243         u32 blocksize;
1244         u64 generation;
1245
1246         __setup_root(tree_root->nodesize, tree_root->leafsize,
1247                      tree_root->sectorsize, tree_root->stripesize,
1248                      root, fs_info, objectid);
1249         ret = btrfs_find_last_root(tree_root, objectid,
1250                                    &root->root_item, &root->root_key);
1251         if (ret > 0)
1252                 return -ENOENT;
1253         else if (ret < 0)
1254                 return ret;
1255
1256         generation = btrfs_root_generation(&root->root_item);
1257         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1258         root->commit_root = NULL;
1259         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1260                                      blocksize, generation);
1261         if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1262                 free_extent_buffer(root->node);
1263                 root->node = NULL;
1264                 return -EIO;
1265         }
1266         root->commit_root = btrfs_root_node(root);
1267         return 0;
1268 }
1269
1270 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1271 {
1272         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1273         if (root)
1274                 root->fs_info = fs_info;
1275         return root;
1276 }
1277
1278 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1279                                      struct btrfs_fs_info *fs_info,
1280                                      u64 objectid)
1281 {
1282         struct extent_buffer *leaf;
1283         struct btrfs_root *tree_root = fs_info->tree_root;
1284         struct btrfs_root *root;
1285         struct btrfs_key key;
1286         int ret = 0;
1287         u64 bytenr;
1288         uuid_le uuid;
1289
1290         root = btrfs_alloc_root(fs_info);
1291         if (!root)
1292                 return ERR_PTR(-ENOMEM);
1293
1294         __setup_root(tree_root->nodesize, tree_root->leafsize,
1295                      tree_root->sectorsize, tree_root->stripesize,
1296                      root, fs_info, objectid);
1297         root->root_key.objectid = objectid;
1298         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1299         root->root_key.offset = 0;
1300
1301         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1302                                       0, objectid, NULL, 0, 0, 0);
1303         if (IS_ERR(leaf)) {
1304                 ret = PTR_ERR(leaf);
1305                 leaf = NULL;
1306                 goto fail;
1307         }
1308
1309         bytenr = leaf->start;
1310         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1311         btrfs_set_header_bytenr(leaf, leaf->start);
1312         btrfs_set_header_generation(leaf, trans->transid);
1313         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1314         btrfs_set_header_owner(leaf, objectid);
1315         root->node = leaf;
1316
1317         write_extent_buffer(leaf, fs_info->fsid,
1318                             (unsigned long)btrfs_header_fsid(leaf),
1319                             BTRFS_FSID_SIZE);
1320         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1321                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1322                             BTRFS_UUID_SIZE);
1323         btrfs_mark_buffer_dirty(leaf);
1324
1325         root->commit_root = btrfs_root_node(root);
1326         root->track_dirty = 1;
1327
1328
1329         root->root_item.flags = 0;
1330         root->root_item.byte_limit = 0;
1331         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1332         btrfs_set_root_generation(&root->root_item, trans->transid);
1333         btrfs_set_root_level(&root->root_item, 0);
1334         btrfs_set_root_refs(&root->root_item, 1);
1335         btrfs_set_root_used(&root->root_item, leaf->len);
1336         btrfs_set_root_last_snapshot(&root->root_item, 0);
1337         btrfs_set_root_dirid(&root->root_item, 0);
1338         uuid_le_gen(&uuid);
1339         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1340         root->root_item.drop_level = 0;
1341
1342         key.objectid = objectid;
1343         key.type = BTRFS_ROOT_ITEM_KEY;
1344         key.offset = 0;
1345         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1346         if (ret)
1347                 goto fail;
1348
1349         btrfs_tree_unlock(leaf);
1350
1351         return root;
1352
1353 fail:
1354         if (leaf) {
1355                 btrfs_tree_unlock(leaf);
1356                 free_extent_buffer(leaf);
1357         }
1358         kfree(root);
1359
1360         return ERR_PTR(ret);
1361 }
1362
1363 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1364                                          struct btrfs_fs_info *fs_info)
1365 {
1366         struct btrfs_root *root;
1367         struct btrfs_root *tree_root = fs_info->tree_root;
1368         struct extent_buffer *leaf;
1369
1370         root = btrfs_alloc_root(fs_info);
1371         if (!root)
1372                 return ERR_PTR(-ENOMEM);
1373
1374         __setup_root(tree_root->nodesize, tree_root->leafsize,
1375                      tree_root->sectorsize, tree_root->stripesize,
1376                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1377
1378         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1379         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1380         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1381         /*
1382          * log trees do not get reference counted because they go away
1383          * before a real commit is actually done.  They do store pointers
1384          * to file data extents, and those reference counts still get
1385          * updated (along with back refs to the log tree).
1386          */
1387         root->ref_cows = 0;
1388
1389         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1390                                       BTRFS_TREE_LOG_OBJECTID, NULL,
1391                                       0, 0, 0);
1392         if (IS_ERR(leaf)) {
1393                 kfree(root);
1394                 return ERR_CAST(leaf);
1395         }
1396
1397         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1398         btrfs_set_header_bytenr(leaf, leaf->start);
1399         btrfs_set_header_generation(leaf, trans->transid);
1400         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1401         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1402         root->node = leaf;
1403
1404         write_extent_buffer(root->node, root->fs_info->fsid,
1405                             (unsigned long)btrfs_header_fsid(root->node),
1406                             BTRFS_FSID_SIZE);
1407         btrfs_mark_buffer_dirty(root->node);
1408         btrfs_tree_unlock(root->node);
1409         return root;
1410 }
1411
1412 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1413                              struct btrfs_fs_info *fs_info)
1414 {
1415         struct btrfs_root *log_root;
1416
1417         log_root = alloc_log_tree(trans, fs_info);
1418         if (IS_ERR(log_root))
1419                 return PTR_ERR(log_root);
1420         WARN_ON(fs_info->log_root_tree);
1421         fs_info->log_root_tree = log_root;
1422         return 0;
1423 }
1424
1425 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1426                        struct btrfs_root *root)
1427 {
1428         struct btrfs_root *log_root;
1429         struct btrfs_inode_item *inode_item;
1430
1431         log_root = alloc_log_tree(trans, root->fs_info);
1432         if (IS_ERR(log_root))
1433                 return PTR_ERR(log_root);
1434
1435         log_root->last_trans = trans->transid;
1436         log_root->root_key.offset = root->root_key.objectid;
1437
1438         inode_item = &log_root->root_item.inode;
1439         inode_item->generation = cpu_to_le64(1);
1440         inode_item->size = cpu_to_le64(3);
1441         inode_item->nlink = cpu_to_le32(1);
1442         inode_item->nbytes = cpu_to_le64(root->leafsize);
1443         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1444
1445         btrfs_set_root_node(&log_root->root_item, log_root->node);
1446
1447         WARN_ON(root->log_root);
1448         root->log_root = log_root;
1449         root->log_transid = 0;
1450         root->last_log_commit = 0;
1451         return 0;
1452 }
1453
1454 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1455                                                struct btrfs_key *location)
1456 {
1457         struct btrfs_root *root;
1458         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1459         struct btrfs_path *path;
1460         struct extent_buffer *l;
1461         u64 generation;
1462         u32 blocksize;
1463         int ret = 0;
1464         int slot;
1465
1466         root = btrfs_alloc_root(fs_info);
1467         if (!root)
1468                 return ERR_PTR(-ENOMEM);
1469         if (location->offset == (u64)-1) {
1470                 ret = find_and_setup_root(tree_root, fs_info,
1471                                           location->objectid, root);
1472                 if (ret) {
1473                         kfree(root);
1474                         return ERR_PTR(ret);
1475                 }
1476                 goto out;
1477         }
1478
1479         __setup_root(tree_root->nodesize, tree_root->leafsize,
1480                      tree_root->sectorsize, tree_root->stripesize,
1481                      root, fs_info, location->objectid);
1482
1483         path = btrfs_alloc_path();
1484         if (!path) {
1485                 kfree(root);
1486                 return ERR_PTR(-ENOMEM);
1487         }
1488         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1489         if (ret == 0) {
1490                 l = path->nodes[0];
1491                 slot = path->slots[0];
1492                 btrfs_read_root_item(l, slot, &root->root_item);
1493                 memcpy(&root->root_key, location, sizeof(*location));
1494         }
1495         btrfs_free_path(path);
1496         if (ret) {
1497                 kfree(root);
1498                 if (ret > 0)
1499                         ret = -ENOENT;
1500                 return ERR_PTR(ret);
1501         }
1502
1503         generation = btrfs_root_generation(&root->root_item);
1504         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1505         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1506                                      blocksize, generation);
1507         if (!root->node || !extent_buffer_uptodate(root->node)) {
1508                 ret = (!root->node) ? -ENOMEM : -EIO;
1509
1510                 free_extent_buffer(root->node);
1511                 kfree(root);
1512                 return ERR_PTR(ret);
1513         }
1514
1515         root->commit_root = btrfs_root_node(root);
1516 out:
1517         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1518                 root->ref_cows = 1;
1519                 btrfs_check_and_init_root_item(&root->root_item);
1520         }
1521
1522         return root;
1523 }
1524
1525 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1526                                               struct btrfs_key *location)
1527 {
1528         struct btrfs_root *root;
1529         int ret;
1530
1531         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1532                 return fs_info->tree_root;
1533         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1534                 return fs_info->extent_root;
1535         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1536                 return fs_info->chunk_root;
1537         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1538                 return fs_info->dev_root;
1539         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1540                 return fs_info->csum_root;
1541         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1542                 return fs_info->quota_root ? fs_info->quota_root :
1543                                              ERR_PTR(-ENOENT);
1544 again:
1545         spin_lock(&fs_info->fs_roots_radix_lock);
1546         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1547                                  (unsigned long)location->objectid);
1548         spin_unlock(&fs_info->fs_roots_radix_lock);
1549         if (root)
1550                 return root;
1551
1552         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1553         if (IS_ERR(root))
1554                 return root;
1555
1556         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1557         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1558                                         GFP_NOFS);
1559         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1560                 ret = -ENOMEM;
1561                 goto fail;
1562         }
1563
1564         btrfs_init_free_ino_ctl(root);
1565         mutex_init(&root->fs_commit_mutex);
1566         spin_lock_init(&root->cache_lock);
1567         init_waitqueue_head(&root->cache_wait);
1568
1569         ret = get_anon_bdev(&root->anon_dev);
1570         if (ret)
1571                 goto fail;
1572
1573         if (btrfs_root_refs(&root->root_item) == 0) {
1574                 ret = -ENOENT;
1575                 goto fail;
1576         }
1577
1578         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1579         if (ret < 0)
1580                 goto fail;
1581         if (ret == 0)
1582                 root->orphan_item_inserted = 1;
1583
1584         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1585         if (ret)
1586                 goto fail;
1587
1588         spin_lock(&fs_info->fs_roots_radix_lock);
1589         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1590                                 (unsigned long)root->root_key.objectid,
1591                                 root);
1592         if (ret == 0)
1593                 root->in_radix = 1;
1594
1595         spin_unlock(&fs_info->fs_roots_radix_lock);
1596         radix_tree_preload_end();
1597         if (ret) {
1598                 if (ret == -EEXIST) {
1599                         free_fs_root(root);
1600                         goto again;
1601                 }
1602                 goto fail;
1603         }
1604
1605         ret = btrfs_find_dead_roots(fs_info->tree_root,
1606                                     root->root_key.objectid);
1607         WARN_ON(ret);
1608         return root;
1609 fail:
1610         free_fs_root(root);
1611         return ERR_PTR(ret);
1612 }
1613
1614 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1615 {
1616         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1617         int ret = 0;
1618         struct btrfs_device *device;
1619         struct backing_dev_info *bdi;
1620
1621         rcu_read_lock();
1622         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1623                 if (!device->bdev)
1624                         continue;
1625                 bdi = blk_get_backing_dev_info(device->bdev);
1626                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1627                         ret = 1;
1628                         break;
1629                 }
1630         }
1631         rcu_read_unlock();
1632         return ret;
1633 }
1634
1635 /*
1636  * If this fails, caller must call bdi_destroy() to get rid of the
1637  * bdi again.
1638  */
1639 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1640 {
1641         int err;
1642
1643         bdi->capabilities = BDI_CAP_MAP_COPY;
1644         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1645         if (err)
1646                 return err;
1647
1648         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1649         bdi->congested_fn       = btrfs_congested_fn;
1650         bdi->congested_data     = info;
1651         return 0;
1652 }
1653
1654 /*
1655  * called by the kthread helper functions to finally call the bio end_io
1656  * functions.  This is where read checksum verification actually happens
1657  */
1658 static void end_workqueue_fn(struct btrfs_work *work)
1659 {
1660         struct bio *bio;
1661         struct end_io_wq *end_io_wq;
1662         struct btrfs_fs_info *fs_info;
1663         int error;
1664
1665         end_io_wq = container_of(work, struct end_io_wq, work);
1666         bio = end_io_wq->bio;
1667         fs_info = end_io_wq->info;
1668
1669         error = end_io_wq->error;
1670         bio->bi_private = end_io_wq->private;
1671         bio->bi_end_io = end_io_wq->end_io;
1672         kfree(end_io_wq);
1673         bio_endio(bio, error);
1674 }
1675
1676 /*
1677  * If we remount the fs to be R/O or umount the fs, the cleaner needn't do
1678  * anything except sleeping. This function is used to check the status of
1679  * the fs.
1680  */
1681 static inline int need_cleaner_sleep(struct btrfs_root *root)
1682 {
1683         return (root->fs_info->sb->s_flags & MS_RDONLY ||
1684                 btrfs_fs_closing(root->fs_info));
1685 }
1686
1687 static int cleaner_kthread(void *arg)
1688 {
1689         struct btrfs_root *root = arg;
1690         int again;
1691
1692         do {
1693                 again = 0;
1694
1695                 /* Make the cleaner go to sleep early. */
1696                 if (need_cleaner_sleep(root))
1697                         goto sleep;
1698
1699                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1700                         goto sleep;
1701
1702                 /*
1703                  * Avoid the problem that we change the status of the fs
1704                  * during the above check and trylock.
1705                  */
1706                 if (need_cleaner_sleep(root)) {
1707                         mutex_unlock(&root->fs_info->cleaner_mutex);
1708                         goto sleep;
1709                 }
1710
1711                 btrfs_run_delayed_iputs(root);
1712                 again = btrfs_clean_one_deleted_snapshot(root);
1713                 mutex_unlock(&root->fs_info->cleaner_mutex);
1714
1715                 /*
1716                  * The defragger has dealt with the R/O remount and umount,
1717                  * needn't do anything special here.
1718                  */
1719                 btrfs_run_defrag_inodes(root->fs_info);
1720 sleep:
1721                 if (!try_to_freeze() && !again) {
1722                         set_current_state(TASK_INTERRUPTIBLE);
1723                         if (!kthread_should_stop())
1724                                 schedule();
1725                         __set_current_state(TASK_RUNNING);
1726                 }
1727         } while (!kthread_should_stop());
1728         return 0;
1729 }
1730
1731 static int transaction_kthread(void *arg)
1732 {
1733         struct btrfs_root *root = arg;
1734         struct btrfs_trans_handle *trans;
1735         struct btrfs_transaction *cur;
1736         u64 transid;
1737         unsigned long now;
1738         unsigned long delay;
1739         bool cannot_commit;
1740
1741         do {
1742                 cannot_commit = false;
1743                 delay = HZ * 30;
1744                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1745
1746                 spin_lock(&root->fs_info->trans_lock);
1747                 cur = root->fs_info->running_transaction;
1748                 if (!cur) {
1749                         spin_unlock(&root->fs_info->trans_lock);
1750                         goto sleep;
1751                 }
1752
1753                 now = get_seconds();
1754                 if (!cur->blocked &&
1755                     (now < cur->start_time || now - cur->start_time < 30)) {
1756                         spin_unlock(&root->fs_info->trans_lock);
1757                         delay = HZ * 5;
1758                         goto sleep;
1759                 }
1760                 transid = cur->transid;
1761                 spin_unlock(&root->fs_info->trans_lock);
1762
1763                 /* If the file system is aborted, this will always fail. */
1764                 trans = btrfs_attach_transaction(root);
1765                 if (IS_ERR(trans)) {
1766                         if (PTR_ERR(trans) != -ENOENT)
1767                                 cannot_commit = true;
1768                         goto sleep;
1769                 }
1770                 if (transid == trans->transid) {
1771                         btrfs_commit_transaction(trans, root);
1772                 } else {
1773                         btrfs_end_transaction(trans, root);
1774                 }
1775 sleep:
1776                 wake_up_process(root->fs_info->cleaner_kthread);
1777                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1778
1779                 if (!try_to_freeze()) {
1780                         set_current_state(TASK_INTERRUPTIBLE);
1781                         if (!kthread_should_stop() &&
1782                             (!btrfs_transaction_blocked(root->fs_info) ||
1783                              cannot_commit))
1784                                 schedule_timeout(delay);
1785                         __set_current_state(TASK_RUNNING);
1786                 }
1787         } while (!kthread_should_stop());
1788         return 0;
1789 }
1790
1791 /*
1792  * this will find the highest generation in the array of
1793  * root backups.  The index of the highest array is returned,
1794  * or -1 if we can't find anything.
1795  *
1796  * We check to make sure the array is valid by comparing the
1797  * generation of the latest  root in the array with the generation
1798  * in the super block.  If they don't match we pitch it.
1799  */
1800 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1801 {
1802         u64 cur;
1803         int newest_index = -1;
1804         struct btrfs_root_backup *root_backup;
1805         int i;
1806
1807         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1808                 root_backup = info->super_copy->super_roots + i;
1809                 cur = btrfs_backup_tree_root_gen(root_backup);
1810                 if (cur == newest_gen)
1811                         newest_index = i;
1812         }
1813
1814         /* check to see if we actually wrapped around */
1815         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1816                 root_backup = info->super_copy->super_roots;
1817                 cur = btrfs_backup_tree_root_gen(root_backup);
1818                 if (cur == newest_gen)
1819                         newest_index = 0;
1820         }
1821         return newest_index;
1822 }
1823
1824
1825 /*
1826  * find the oldest backup so we know where to store new entries
1827  * in the backup array.  This will set the backup_root_index
1828  * field in the fs_info struct
1829  */
1830 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1831                                      u64 newest_gen)
1832 {
1833         int newest_index = -1;
1834
1835         newest_index = find_newest_super_backup(info, newest_gen);
1836         /* if there was garbage in there, just move along */
1837         if (newest_index == -1) {
1838                 info->backup_root_index = 0;
1839         } else {
1840                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1841         }
1842 }
1843
1844 /*
1845  * copy all the root pointers into the super backup array.
1846  * this will bump the backup pointer by one when it is
1847  * done
1848  */
1849 static void backup_super_roots(struct btrfs_fs_info *info)
1850 {
1851         int next_backup;
1852         struct btrfs_root_backup *root_backup;
1853         int last_backup;
1854
1855         next_backup = info->backup_root_index;
1856         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1857                 BTRFS_NUM_BACKUP_ROOTS;
1858
1859         /*
1860          * just overwrite the last backup if we're at the same generation
1861          * this happens only at umount
1862          */
1863         root_backup = info->super_for_commit->super_roots + last_backup;
1864         if (btrfs_backup_tree_root_gen(root_backup) ==
1865             btrfs_header_generation(info->tree_root->node))
1866                 next_backup = last_backup;
1867
1868         root_backup = info->super_for_commit->super_roots + next_backup;
1869
1870         /*
1871          * make sure all of our padding and empty slots get zero filled
1872          * regardless of which ones we use today
1873          */
1874         memset(root_backup, 0, sizeof(*root_backup));
1875
1876         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1877
1878         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1879         btrfs_set_backup_tree_root_gen(root_backup,
1880                                btrfs_header_generation(info->tree_root->node));
1881
1882         btrfs_set_backup_tree_root_level(root_backup,
1883                                btrfs_header_level(info->tree_root->node));
1884
1885         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1886         btrfs_set_backup_chunk_root_gen(root_backup,
1887                                btrfs_header_generation(info->chunk_root->node));
1888         btrfs_set_backup_chunk_root_level(root_backup,
1889                                btrfs_header_level(info->chunk_root->node));
1890
1891         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1892         btrfs_set_backup_extent_root_gen(root_backup,
1893                                btrfs_header_generation(info->extent_root->node));
1894         btrfs_set_backup_extent_root_level(root_backup,
1895                                btrfs_header_level(info->extent_root->node));
1896
1897         /*
1898          * we might commit during log recovery, which happens before we set
1899          * the fs_root.  Make sure it is valid before we fill it in.
1900          */
1901         if (info->fs_root && info->fs_root->node) {
1902                 btrfs_set_backup_fs_root(root_backup,
1903                                          info->fs_root->node->start);
1904                 btrfs_set_backup_fs_root_gen(root_backup,
1905                                btrfs_header_generation(info->fs_root->node));
1906                 btrfs_set_backup_fs_root_level(root_backup,
1907                                btrfs_header_level(info->fs_root->node));
1908         }
1909
1910         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1911         btrfs_set_backup_dev_root_gen(root_backup,
1912                                btrfs_header_generation(info->dev_root->node));
1913         btrfs_set_backup_dev_root_level(root_backup,
1914                                        btrfs_header_level(info->dev_root->node));
1915
1916         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1917         btrfs_set_backup_csum_root_gen(root_backup,
1918                                btrfs_header_generation(info->csum_root->node));
1919         btrfs_set_backup_csum_root_level(root_backup,
1920                                btrfs_header_level(info->csum_root->node));
1921
1922         btrfs_set_backup_total_bytes(root_backup,
1923                              btrfs_super_total_bytes(info->super_copy));
1924         btrfs_set_backup_bytes_used(root_backup,
1925                              btrfs_super_bytes_used(info->super_copy));
1926         btrfs_set_backup_num_devices(root_backup,
1927                              btrfs_super_num_devices(info->super_copy));
1928
1929         /*
1930          * if we don't copy this out to the super_copy, it won't get remembered
1931          * for the next commit
1932          */
1933         memcpy(&info->super_copy->super_roots,
1934                &info->super_for_commit->super_roots,
1935                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1936 }
1937
1938 /*
1939  * this copies info out of the root backup array and back into
1940  * the in-memory super block.  It is meant to help iterate through
1941  * the array, so you send it the number of backups you've already
1942  * tried and the last backup index you used.
1943  *
1944  * this returns -1 when it has tried all the backups
1945  */
1946 static noinline int next_root_backup(struct btrfs_fs_info *info,
1947                                      struct btrfs_super_block *super,
1948                                      int *num_backups_tried, int *backup_index)
1949 {
1950         struct btrfs_root_backup *root_backup;
1951         int newest = *backup_index;
1952
1953         if (*num_backups_tried == 0) {
1954                 u64 gen = btrfs_super_generation(super);
1955
1956                 newest = find_newest_super_backup(info, gen);
1957                 if (newest == -1)
1958                         return -1;
1959
1960                 *backup_index = newest;
1961                 *num_backups_tried = 1;
1962         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1963                 /* we've tried all the backups, all done */
1964                 return -1;
1965         } else {
1966                 /* jump to the next oldest backup */
1967                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1968                         BTRFS_NUM_BACKUP_ROOTS;
1969                 *backup_index = newest;
1970                 *num_backups_tried += 1;
1971         }
1972         root_backup = super->super_roots + newest;
1973
1974         btrfs_set_super_generation(super,
1975                                    btrfs_backup_tree_root_gen(root_backup));
1976         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1977         btrfs_set_super_root_level(super,
1978                                    btrfs_backup_tree_root_level(root_backup));
1979         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1980
1981         /*
1982          * fixme: the total bytes and num_devices need to match or we should
1983          * need a fsck
1984          */
1985         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1986         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1987         return 0;
1988 }
1989
1990 /* helper to cleanup workers */
1991 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1992 {
1993         btrfs_stop_workers(&fs_info->generic_worker);
1994         btrfs_stop_workers(&fs_info->fixup_workers);
1995         btrfs_stop_workers(&fs_info->delalloc_workers);
1996         btrfs_stop_workers(&fs_info->workers);
1997         btrfs_stop_workers(&fs_info->endio_workers);
1998         btrfs_stop_workers(&fs_info->endio_meta_workers);
1999         btrfs_stop_workers(&fs_info->endio_raid56_workers);
2000         btrfs_stop_workers(&fs_info->rmw_workers);
2001         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2002         btrfs_stop_workers(&fs_info->endio_write_workers);
2003         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2004         btrfs_stop_workers(&fs_info->submit_workers);
2005         btrfs_stop_workers(&fs_info->delayed_workers);
2006         btrfs_stop_workers(&fs_info->caching_workers);
2007         btrfs_stop_workers(&fs_info->readahead_workers);
2008         btrfs_stop_workers(&fs_info->flush_workers);
2009         btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2010 }
2011
2012 /* helper to cleanup tree roots */
2013 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2014 {
2015         free_extent_buffer(info->tree_root->node);
2016         free_extent_buffer(info->tree_root->commit_root);
2017         info->tree_root->node = NULL;
2018         info->tree_root->commit_root = NULL;
2019
2020         if (info->dev_root) {
2021                 free_extent_buffer(info->dev_root->node);
2022                 free_extent_buffer(info->dev_root->commit_root);
2023                 info->dev_root->node = NULL;
2024                 info->dev_root->commit_root = NULL;
2025         }
2026         if (info->extent_root) {
2027                 free_extent_buffer(info->extent_root->node);
2028                 free_extent_buffer(info->extent_root->commit_root);
2029                 info->extent_root->node = NULL;
2030                 info->extent_root->commit_root = NULL;
2031         }
2032         if (info->csum_root) {
2033                 free_extent_buffer(info->csum_root->node);
2034                 free_extent_buffer(info->csum_root->commit_root);
2035                 info->csum_root->node = NULL;
2036                 info->csum_root->commit_root = NULL;
2037         }
2038         if (info->quota_root) {
2039                 free_extent_buffer(info->quota_root->node);
2040                 free_extent_buffer(info->quota_root->commit_root);
2041                 info->quota_root->node = NULL;
2042                 info->quota_root->commit_root = NULL;
2043         }
2044         if (chunk_root) {
2045                 free_extent_buffer(info->chunk_root->node);
2046                 free_extent_buffer(info->chunk_root->commit_root);
2047                 info->chunk_root->node = NULL;
2048                 info->chunk_root->commit_root = NULL;
2049         }
2050 }
2051
2052 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2053 {
2054         int ret;
2055         struct btrfs_root *gang[8];
2056         int i;
2057
2058         while (!list_empty(&fs_info->dead_roots)) {
2059                 gang[0] = list_entry(fs_info->dead_roots.next,
2060                                      struct btrfs_root, root_list);
2061                 list_del(&gang[0]->root_list);
2062
2063                 if (gang[0]->in_radix) {
2064                         btrfs_free_fs_root(fs_info, gang[0]);
2065                 } else {
2066                         free_extent_buffer(gang[0]->node);
2067                         free_extent_buffer(gang[0]->commit_root);
2068                         kfree(gang[0]);
2069                 }
2070         }
2071
2072         while (1) {
2073                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2074                                              (void **)gang, 0,
2075                                              ARRAY_SIZE(gang));
2076                 if (!ret)
2077                         break;
2078                 for (i = 0; i < ret; i++)
2079                         btrfs_free_fs_root(fs_info, gang[i]);
2080         }
2081 }
2082
2083 int open_ctree(struct super_block *sb,
2084                struct btrfs_fs_devices *fs_devices,
2085                char *options)
2086 {
2087         u32 sectorsize;
2088         u32 nodesize;
2089         u32 leafsize;
2090         u32 blocksize;
2091         u32 stripesize;
2092         u64 generation;
2093         u64 features;
2094         struct btrfs_key location;
2095         struct buffer_head *bh;
2096         struct btrfs_super_block *disk_super;
2097         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2098         struct btrfs_root *tree_root;
2099         struct btrfs_root *extent_root;
2100         struct btrfs_root *csum_root;
2101         struct btrfs_root *chunk_root;
2102         struct btrfs_root *dev_root;
2103         struct btrfs_root *quota_root;
2104         struct btrfs_root *log_tree_root;
2105         int ret;
2106         int err = -EINVAL;
2107         int num_backups_tried = 0;
2108         int backup_index = 0;
2109
2110         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2111         extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
2112         csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
2113         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2114         dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
2115         quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
2116
2117         if (!tree_root || !extent_root || !csum_root ||
2118             !chunk_root || !dev_root || !quota_root) {
2119                 err = -ENOMEM;
2120                 goto fail;
2121         }
2122
2123         ret = init_srcu_struct(&fs_info->subvol_srcu);
2124         if (ret) {
2125                 err = ret;
2126                 goto fail;
2127         }
2128
2129         ret = setup_bdi(fs_info, &fs_info->bdi);
2130         if (ret) {
2131                 err = ret;
2132                 goto fail_srcu;
2133         }
2134
2135         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2136         if (ret) {
2137                 err = ret;
2138                 goto fail_bdi;
2139         }
2140         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2141                                         (1 + ilog2(nr_cpu_ids));
2142
2143         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2144         if (ret) {
2145                 err = ret;
2146                 goto fail_dirty_metadata_bytes;
2147         }
2148
2149         fs_info->btree_inode = new_inode(sb);
2150         if (!fs_info->btree_inode) {
2151                 err = -ENOMEM;
2152                 goto fail_delalloc_bytes;
2153         }
2154
2155         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2156
2157         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2158         INIT_LIST_HEAD(&fs_info->trans_list);
2159         INIT_LIST_HEAD(&fs_info->dead_roots);
2160         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2161         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
2162         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2163         spin_lock_init(&fs_info->delalloc_lock);
2164         spin_lock_init(&fs_info->trans_lock);
2165         spin_lock_init(&fs_info->fs_roots_radix_lock);
2166         spin_lock_init(&fs_info->delayed_iput_lock);
2167         spin_lock_init(&fs_info->defrag_inodes_lock);
2168         spin_lock_init(&fs_info->free_chunk_lock);
2169         spin_lock_init(&fs_info->tree_mod_seq_lock);
2170         spin_lock_init(&fs_info->super_lock);
2171         rwlock_init(&fs_info->tree_mod_log_lock);
2172         mutex_init(&fs_info->reloc_mutex);
2173         seqlock_init(&fs_info->profiles_lock);
2174
2175         init_completion(&fs_info->kobj_unregister);
2176         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2177         INIT_LIST_HEAD(&fs_info->space_info);
2178         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2179         btrfs_mapping_init(&fs_info->mapping_tree);
2180         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2181                              BTRFS_BLOCK_RSV_GLOBAL);
2182         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2183                              BTRFS_BLOCK_RSV_DELALLOC);
2184         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2185         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2186         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2187         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2188                              BTRFS_BLOCK_RSV_DELOPS);
2189         atomic_set(&fs_info->nr_async_submits, 0);
2190         atomic_set(&fs_info->async_delalloc_pages, 0);
2191         atomic_set(&fs_info->async_submit_draining, 0);
2192         atomic_set(&fs_info->nr_async_bios, 0);
2193         atomic_set(&fs_info->defrag_running, 0);
2194         atomic64_set(&fs_info->tree_mod_seq, 0);
2195         fs_info->sb = sb;
2196         fs_info->max_inline = 8192 * 1024;
2197         fs_info->metadata_ratio = 0;
2198         fs_info->defrag_inodes = RB_ROOT;
2199         fs_info->trans_no_join = 0;
2200         fs_info->free_chunk_space = 0;
2201         fs_info->tree_mod_log = RB_ROOT;
2202
2203         /* readahead state */
2204         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2205         spin_lock_init(&fs_info->reada_lock);
2206
2207         fs_info->thread_pool_size = min_t(unsigned long,
2208                                           num_online_cpus() + 2, 8);
2209
2210         INIT_LIST_HEAD(&fs_info->ordered_extents);
2211         spin_lock_init(&fs_info->ordered_extent_lock);
2212         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2213                                         GFP_NOFS);
2214         if (!fs_info->delayed_root) {
2215                 err = -ENOMEM;
2216                 goto fail_iput;
2217         }
2218         btrfs_init_delayed_root(fs_info->delayed_root);
2219
2220         mutex_init(&fs_info->scrub_lock);
2221         atomic_set(&fs_info->scrubs_running, 0);
2222         atomic_set(&fs_info->scrub_pause_req, 0);
2223         atomic_set(&fs_info->scrubs_paused, 0);
2224         atomic_set(&fs_info->scrub_cancel_req, 0);
2225         init_waitqueue_head(&fs_info->scrub_pause_wait);
2226         init_rwsem(&fs_info->scrub_super_lock);
2227         fs_info->scrub_workers_refcnt = 0;
2228 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2229         fs_info->check_integrity_print_mask = 0;
2230 #endif
2231
2232         spin_lock_init(&fs_info->balance_lock);
2233         mutex_init(&fs_info->balance_mutex);
2234         atomic_set(&fs_info->balance_running, 0);
2235         atomic_set(&fs_info->balance_pause_req, 0);
2236         atomic_set(&fs_info->balance_cancel_req, 0);
2237         fs_info->balance_ctl = NULL;
2238         init_waitqueue_head(&fs_info->balance_wait_q);
2239
2240         sb->s_blocksize = 4096;
2241         sb->s_blocksize_bits = blksize_bits(4096);
2242         sb->s_bdi = &fs_info->bdi;
2243
2244         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2245         set_nlink(fs_info->btree_inode, 1);
2246         /*
2247          * we set the i_size on the btree inode to the max possible int.
2248          * the real end of the address space is determined by all of
2249          * the devices in the system
2250          */
2251         fs_info->btree_inode->i_size = OFFSET_MAX;
2252         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2253         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2254
2255         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2256         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2257                              fs_info->btree_inode->i_mapping);
2258         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2259         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2260
2261         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2262
2263         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2264         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2265                sizeof(struct btrfs_key));
2266         set_bit(BTRFS_INODE_DUMMY,
2267                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2268         insert_inode_hash(fs_info->btree_inode);
2269
2270         spin_lock_init(&fs_info->block_group_cache_lock);
2271         fs_info->block_group_cache_tree = RB_ROOT;
2272         fs_info->first_logical_byte = (u64)-1;
2273
2274         extent_io_tree_init(&fs_info->freed_extents[0],
2275                              fs_info->btree_inode->i_mapping);
2276         extent_io_tree_init(&fs_info->freed_extents[1],
2277                              fs_info->btree_inode->i_mapping);
2278         fs_info->pinned_extents = &fs_info->freed_extents[0];
2279         fs_info->do_barriers = 1;
2280
2281
2282         mutex_init(&fs_info->ordered_operations_mutex);
2283         mutex_init(&fs_info->tree_log_mutex);
2284         mutex_init(&fs_info->chunk_mutex);
2285         mutex_init(&fs_info->transaction_kthread_mutex);
2286         mutex_init(&fs_info->cleaner_mutex);
2287         mutex_init(&fs_info->volume_mutex);
2288         init_rwsem(&fs_info->extent_commit_sem);
2289         init_rwsem(&fs_info->cleanup_work_sem);
2290         init_rwsem(&fs_info->subvol_sem);
2291         fs_info->dev_replace.lock_owner = 0;
2292         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2293         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2294         mutex_init(&fs_info->dev_replace.lock_management_lock);
2295         mutex_init(&fs_info->dev_replace.lock);
2296
2297         spin_lock_init(&fs_info->qgroup_lock);
2298         mutex_init(&fs_info->qgroup_ioctl_lock);
2299         fs_info->qgroup_tree = RB_ROOT;
2300         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2301         fs_info->qgroup_seq = 1;
2302         fs_info->quota_enabled = 0;
2303         fs_info->pending_quota_state = 0;
2304         fs_info->qgroup_ulist = NULL;
2305         mutex_init(&fs_info->qgroup_rescan_lock);
2306
2307         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2308         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2309
2310         init_waitqueue_head(&fs_info->transaction_throttle);
2311         init_waitqueue_head(&fs_info->transaction_wait);
2312         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2313         init_waitqueue_head(&fs_info->async_submit_wait);
2314
2315         ret = btrfs_alloc_stripe_hash_table(fs_info);
2316         if (ret) {
2317                 err = ret;
2318                 goto fail_alloc;
2319         }
2320
2321         __setup_root(4096, 4096, 4096, 4096, tree_root,
2322                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2323
2324         invalidate_bdev(fs_devices->latest_bdev);
2325
2326         /*
2327          * Read super block and check the signature bytes only
2328          */
2329         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2330         if (!bh) {
2331                 err = -EINVAL;
2332                 goto fail_alloc;
2333         }
2334
2335         /*
2336          * We want to check superblock checksum, the type is stored inside.
2337          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2338          */
2339         if (btrfs_check_super_csum(bh->b_data)) {
2340                 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2341                 err = -EINVAL;
2342                 goto fail_alloc;
2343         }
2344
2345         /*
2346          * super_copy is zeroed at allocation time and we never touch the
2347          * following bytes up to INFO_SIZE, the checksum is calculated from
2348          * the whole block of INFO_SIZE
2349          */
2350         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2351         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2352                sizeof(*fs_info->super_for_commit));
2353         brelse(bh);
2354
2355         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2356
2357         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2358         if (ret) {
2359                 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2360                 err = -EINVAL;
2361                 goto fail_alloc;
2362         }
2363
2364         disk_super = fs_info->super_copy;
2365         if (!btrfs_super_root(disk_super))
2366                 goto fail_alloc;
2367
2368         /* check FS state, whether FS is broken. */
2369         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2370                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2371
2372         /*
2373          * run through our array of backup supers and setup
2374          * our ring pointer to the oldest one
2375          */
2376         generation = btrfs_super_generation(disk_super);
2377         find_oldest_super_backup(fs_info, generation);
2378
2379         /*
2380          * In the long term, we'll store the compression type in the super
2381          * block, and it'll be used for per file compression control.
2382          */
2383         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2384
2385         ret = btrfs_parse_options(tree_root, options);
2386         if (ret) {
2387                 err = ret;
2388                 goto fail_alloc;
2389         }
2390
2391         features = btrfs_super_incompat_flags(disk_super) &
2392                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2393         if (features) {
2394                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2395                        "unsupported optional features (%Lx).\n",
2396                        (unsigned long long)features);
2397                 err = -EINVAL;
2398                 goto fail_alloc;
2399         }
2400
2401         if (btrfs_super_leafsize(disk_super) !=
2402             btrfs_super_nodesize(disk_super)) {
2403                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2404                        "blocksizes don't match.  node %d leaf %d\n",
2405                        btrfs_super_nodesize(disk_super),
2406                        btrfs_super_leafsize(disk_super));
2407                 err = -EINVAL;
2408                 goto fail_alloc;
2409         }
2410         if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2411                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2412                        "blocksize (%d) was too large\n",
2413                        btrfs_super_leafsize(disk_super));
2414                 err = -EINVAL;
2415                 goto fail_alloc;
2416         }
2417
2418         features = btrfs_super_incompat_flags(disk_super);
2419         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2420         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2421                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2422
2423         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2424                 printk(KERN_ERR "btrfs: has skinny extents\n");
2425
2426         /*
2427          * flag our filesystem as having big metadata blocks if
2428          * they are bigger than the page size
2429          */
2430         if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2431                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2432                         printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2433                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2434         }
2435
2436         nodesize = btrfs_super_nodesize(disk_super);
2437         leafsize = btrfs_super_leafsize(disk_super);
2438         sectorsize = btrfs_super_sectorsize(disk_super);
2439         stripesize = btrfs_super_stripesize(disk_super);
2440         fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2441         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2442
2443         /*
2444          * mixed block groups end up with duplicate but slightly offset
2445          * extent buffers for the same range.  It leads to corruptions
2446          */
2447         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2448             (sectorsize != leafsize)) {
2449                 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2450                                 "are not allowed for mixed block groups on %s\n",
2451                                 sb->s_id);
2452                 goto fail_alloc;
2453         }
2454
2455         /*
2456          * Needn't use the lock because there is no other task which will
2457          * update the flag.
2458          */
2459         btrfs_set_super_incompat_flags(disk_super, features);
2460
2461         features = btrfs_super_compat_ro_flags(disk_super) &
2462                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2463         if (!(sb->s_flags & MS_RDONLY) && features) {
2464                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2465                        "unsupported option features (%Lx).\n",
2466                        (unsigned long long)features);
2467                 err = -EINVAL;
2468                 goto fail_alloc;
2469         }
2470
2471         btrfs_init_workers(&fs_info->generic_worker,
2472                            "genwork", 1, NULL);
2473
2474         btrfs_init_workers(&fs_info->workers, "worker",
2475                            fs_info->thread_pool_size,
2476                            &fs_info->generic_worker);
2477
2478         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2479                            fs_info->thread_pool_size,
2480                            &fs_info->generic_worker);
2481
2482         btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2483                            fs_info->thread_pool_size,
2484                            &fs_info->generic_worker);
2485
2486         btrfs_init_workers(&fs_info->submit_workers, "submit",
2487                            min_t(u64, fs_devices->num_devices,
2488                            fs_info->thread_pool_size),
2489                            &fs_info->generic_worker);
2490
2491         btrfs_init_workers(&fs_info->caching_workers, "cache",
2492                            2, &fs_info->generic_worker);
2493
2494         /* a higher idle thresh on the submit workers makes it much more
2495          * likely that bios will be send down in a sane order to the
2496          * devices
2497          */
2498         fs_info->submit_workers.idle_thresh = 64;
2499
2500         fs_info->workers.idle_thresh = 16;
2501         fs_info->workers.ordered = 1;
2502
2503         fs_info->delalloc_workers.idle_thresh = 2;
2504         fs_info->delalloc_workers.ordered = 1;
2505
2506         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2507                            &fs_info->generic_worker);
2508         btrfs_init_workers(&fs_info->endio_workers, "endio",
2509                            fs_info->thread_pool_size,
2510                            &fs_info->generic_worker);
2511         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2512                            fs_info->thread_pool_size,
2513                            &fs_info->generic_worker);
2514         btrfs_init_workers(&fs_info->endio_meta_write_workers,
2515                            "endio-meta-write", fs_info->thread_pool_size,
2516                            &fs_info->generic_worker);
2517         btrfs_init_workers(&fs_info->endio_raid56_workers,
2518                            "endio-raid56", fs_info->thread_pool_size,
2519                            &fs_info->generic_worker);
2520         btrfs_init_workers(&fs_info->rmw_workers,
2521                            "rmw", fs_info->thread_pool_size,
2522                            &fs_info->generic_worker);
2523         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2524                            fs_info->thread_pool_size,
2525                            &fs_info->generic_worker);
2526         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2527                            1, &fs_info->generic_worker);
2528         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2529                            fs_info->thread_pool_size,
2530                            &fs_info->generic_worker);
2531         btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2532                            fs_info->thread_pool_size,
2533                            &fs_info->generic_worker);
2534         btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2535                            &fs_info->generic_worker);
2536
2537         /*
2538          * endios are largely parallel and should have a very
2539          * low idle thresh
2540          */
2541         fs_info->endio_workers.idle_thresh = 4;
2542         fs_info->endio_meta_workers.idle_thresh = 4;
2543         fs_info->endio_raid56_workers.idle_thresh = 4;
2544         fs_info->rmw_workers.idle_thresh = 2;
2545
2546         fs_info->endio_write_workers.idle_thresh = 2;
2547         fs_info->endio_meta_write_workers.idle_thresh = 2;
2548         fs_info->readahead_workers.idle_thresh = 2;
2549
2550         /*
2551          * btrfs_start_workers can really only fail because of ENOMEM so just
2552          * return -ENOMEM if any of these fail.
2553          */
2554         ret = btrfs_start_workers(&fs_info->workers);
2555         ret |= btrfs_start_workers(&fs_info->generic_worker);
2556         ret |= btrfs_start_workers(&fs_info->submit_workers);
2557         ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2558         ret |= btrfs_start_workers(&fs_info->fixup_workers);
2559         ret |= btrfs_start_workers(&fs_info->endio_workers);
2560         ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2561         ret |= btrfs_start_workers(&fs_info->rmw_workers);
2562         ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2563         ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2564         ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2565         ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2566         ret |= btrfs_start_workers(&fs_info->delayed_workers);
2567         ret |= btrfs_start_workers(&fs_info->caching_workers);
2568         ret |= btrfs_start_workers(&fs_info->readahead_workers);
2569         ret |= btrfs_start_workers(&fs_info->flush_workers);
2570         ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2571         if (ret) {
2572                 err = -ENOMEM;
2573                 goto fail_sb_buffer;
2574         }
2575
2576         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2577         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2578                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2579
2580         tree_root->nodesize = nodesize;
2581         tree_root->leafsize = leafsize;
2582         tree_root->sectorsize = sectorsize;
2583         tree_root->stripesize = stripesize;
2584
2585         sb->s_blocksize = sectorsize;
2586         sb->s_blocksize_bits = blksize_bits(sectorsize);
2587
2588         if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2589                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2590                 goto fail_sb_buffer;
2591         }
2592
2593         if (sectorsize != PAGE_SIZE) {
2594                 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2595                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2596                 goto fail_sb_buffer;
2597         }
2598
2599         mutex_lock(&fs_info->chunk_mutex);
2600         ret = btrfs_read_sys_array(tree_root);
2601         mutex_unlock(&fs_info->chunk_mutex);
2602         if (ret) {
2603                 printk(KERN_WARNING "btrfs: failed to read the system "
2604                        "array on %s\n", sb->s_id);
2605                 goto fail_sb_buffer;
2606         }
2607
2608         blocksize = btrfs_level_size(tree_root,
2609                                      btrfs_super_chunk_root_level(disk_super));
2610         generation = btrfs_super_chunk_root_generation(disk_super);
2611
2612         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2613                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2614
2615         chunk_root->node = read_tree_block(chunk_root,
2616                                            btrfs_super_chunk_root(disk_super),
2617                                            blocksize, generation);
2618         if (!chunk_root->node ||
2619             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2620                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2621                        sb->s_id);
2622                 goto fail_tree_roots;
2623         }
2624         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2625         chunk_root->commit_root = btrfs_root_node(chunk_root);
2626
2627         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2628            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2629            BTRFS_UUID_SIZE);
2630
2631         ret = btrfs_read_chunk_tree(chunk_root);
2632         if (ret) {
2633                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2634                        sb->s_id);
2635                 goto fail_tree_roots;
2636         }
2637
2638         /*
2639          * keep the device that is marked to be the target device for the
2640          * dev_replace procedure
2641          */
2642         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2643
2644         if (!fs_devices->latest_bdev) {
2645                 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2646                        sb->s_id);
2647                 goto fail_tree_roots;
2648         }
2649
2650 retry_root_backup:
2651         blocksize = btrfs_level_size(tree_root,
2652                                      btrfs_super_root_level(disk_super));
2653         generation = btrfs_super_generation(disk_super);
2654
2655         tree_root->node = read_tree_block(tree_root,
2656                                           btrfs_super_root(disk_super),
2657                                           blocksize, generation);
2658         if (!tree_root->node ||
2659             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2660                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2661                        sb->s_id);
2662
2663                 goto recovery_tree_root;
2664         }
2665
2666         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2667         tree_root->commit_root = btrfs_root_node(tree_root);
2668
2669         ret = find_and_setup_root(tree_root, fs_info,
2670                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2671         if (ret)
2672                 goto recovery_tree_root;
2673         extent_root->track_dirty = 1;
2674
2675         ret = find_and_setup_root(tree_root, fs_info,
2676                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2677         if (ret)
2678                 goto recovery_tree_root;
2679         dev_root->track_dirty = 1;
2680
2681         ret = find_and_setup_root(tree_root, fs_info,
2682                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2683         if (ret)
2684                 goto recovery_tree_root;
2685         csum_root->track_dirty = 1;
2686
2687         ret = find_and_setup_root(tree_root, fs_info,
2688                                   BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2689         if (ret) {
2690                 kfree(quota_root);
2691                 quota_root = fs_info->quota_root = NULL;
2692         } else {
2693                 quota_root->track_dirty = 1;
2694                 fs_info->quota_enabled = 1;
2695                 fs_info->pending_quota_state = 1;
2696         }
2697
2698         fs_info->generation = generation;
2699         fs_info->last_trans_committed = generation;
2700
2701         ret = btrfs_recover_balance(fs_info);
2702         if (ret) {
2703                 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2704                 goto fail_block_groups;
2705         }
2706
2707         ret = btrfs_init_dev_stats(fs_info);
2708         if (ret) {
2709                 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2710                        ret);
2711                 goto fail_block_groups;
2712         }
2713
2714         ret = btrfs_init_dev_replace(fs_info);
2715         if (ret) {
2716                 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2717                 goto fail_block_groups;
2718         }
2719
2720         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2721
2722         ret = btrfs_init_space_info(fs_info);
2723         if (ret) {
2724                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2725                 goto fail_block_groups;
2726         }
2727
2728         ret = btrfs_read_block_groups(extent_root);
2729         if (ret) {
2730                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2731                 goto fail_block_groups;
2732         }
2733         fs_info->num_tolerated_disk_barrier_failures =
2734                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2735         if (fs_info->fs_devices->missing_devices >
2736              fs_info->num_tolerated_disk_barrier_failures &&
2737             !(sb->s_flags & MS_RDONLY)) {
2738                 printk(KERN_WARNING
2739                        "Btrfs: too many missing devices, writeable mount is not allowed\n");
2740                 goto fail_block_groups;
2741         }
2742
2743         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2744                                                "btrfs-cleaner");
2745         if (IS_ERR(fs_info->cleaner_kthread))
2746                 goto fail_block_groups;
2747
2748         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2749                                                    tree_root,
2750                                                    "btrfs-transaction");
2751         if (IS_ERR(fs_info->transaction_kthread))
2752                 goto fail_cleaner;
2753
2754         if (!btrfs_test_opt(tree_root, SSD) &&
2755             !btrfs_test_opt(tree_root, NOSSD) &&
2756             !fs_info->fs_devices->rotating) {
2757                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2758                        "mode\n");
2759                 btrfs_set_opt(fs_info->mount_opt, SSD);
2760         }
2761
2762 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2763         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2764                 ret = btrfsic_mount(tree_root, fs_devices,
2765                                     btrfs_test_opt(tree_root,
2766                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2767                                     1 : 0,
2768                                     fs_info->check_integrity_print_mask);
2769                 if (ret)
2770                         printk(KERN_WARNING "btrfs: failed to initialize"
2771                                " integrity check module %s\n", sb->s_id);
2772         }
2773 #endif
2774         ret = btrfs_read_qgroup_config(fs_info);
2775         if (ret)
2776                 goto fail_trans_kthread;
2777
2778         /* do not make disk changes in broken FS */
2779         if (btrfs_super_log_root(disk_super) != 0) {
2780                 u64 bytenr = btrfs_super_log_root(disk_super);
2781
2782                 if (fs_devices->rw_devices == 0) {
2783                         printk(KERN_WARNING "Btrfs log replay required "
2784                                "on RO media\n");
2785                         err = -EIO;
2786                         goto fail_qgroup;
2787                 }
2788                 blocksize =
2789                      btrfs_level_size(tree_root,
2790                                       btrfs_super_log_root_level(disk_super));
2791
2792                 log_tree_root = btrfs_alloc_root(fs_info);
2793                 if (!log_tree_root) {
2794                         err = -ENOMEM;
2795                         goto fail_qgroup;
2796                 }
2797
2798                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2799                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2800
2801                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2802                                                       blocksize,
2803                                                       generation + 1);
2804                 if (!log_tree_root->node ||
2805                     !extent_buffer_uptodate(log_tree_root->node)) {
2806                         printk(KERN_ERR "btrfs: failed to read log tree\n");
2807                         free_extent_buffer(log_tree_root->node);
2808                         kfree(log_tree_root);
2809                         goto fail_trans_kthread;
2810                 }
2811                 /* returns with log_tree_root freed on success */
2812                 ret = btrfs_recover_log_trees(log_tree_root);
2813                 if (ret) {
2814                         btrfs_error(tree_root->fs_info, ret,
2815                                     "Failed to recover log tree");
2816                         free_extent_buffer(log_tree_root->node);
2817                         kfree(log_tree_root);
2818                         goto fail_trans_kthread;
2819                 }
2820
2821                 if (sb->s_flags & MS_RDONLY) {
2822                         ret = btrfs_commit_super(tree_root);
2823                         if (ret)
2824                                 goto fail_trans_kthread;
2825                 }
2826         }
2827
2828         ret = btrfs_find_orphan_roots(tree_root);
2829         if (ret)
2830                 goto fail_trans_kthread;
2831
2832         if (!(sb->s_flags & MS_RDONLY)) {
2833                 ret = btrfs_cleanup_fs_roots(fs_info);
2834                 if (ret)
2835                         goto fail_trans_kthread;
2836
2837                 ret = btrfs_recover_relocation(tree_root);
2838                 if (ret < 0) {
2839                         printk(KERN_WARNING
2840                                "btrfs: failed to recover relocation\n");
2841                         err = -EINVAL;
2842                         goto fail_qgroup;
2843                 }
2844         }
2845
2846         location.objectid = BTRFS_FS_TREE_OBJECTID;
2847         location.type = BTRFS_ROOT_ITEM_KEY;
2848         location.offset = (u64)-1;
2849
2850         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2851         if (IS_ERR(fs_info->fs_root)) {
2852                 err = PTR_ERR(fs_info->fs_root);
2853                 goto fail_qgroup;
2854         }
2855
2856         if (sb->s_flags & MS_RDONLY)
2857                 return 0;
2858
2859         down_read(&fs_info->cleanup_work_sem);
2860         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2861             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2862                 up_read(&fs_info->cleanup_work_sem);
2863                 close_ctree(tree_root);
2864                 return ret;
2865         }
2866         up_read(&fs_info->cleanup_work_sem);
2867
2868         ret = btrfs_resume_balance_async(fs_info);
2869         if (ret) {
2870                 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2871                 close_ctree(tree_root);
2872                 return ret;
2873         }
2874
2875         ret = btrfs_resume_dev_replace_async(fs_info);
2876         if (ret) {
2877                 pr_warn("btrfs: failed to resume dev_replace\n");
2878                 close_ctree(tree_root);
2879                 return ret;
2880         }
2881
2882         return 0;
2883
2884 fail_qgroup:
2885         btrfs_free_qgroup_config(fs_info);
2886 fail_trans_kthread:
2887         kthread_stop(fs_info->transaction_kthread);
2888         btrfs_cleanup_transaction(fs_info->tree_root);
2889         del_fs_roots(fs_info);
2890 fail_cleaner:
2891         kthread_stop(fs_info->cleaner_kthread);
2892
2893         /*
2894          * make sure we're done with the btree inode before we stop our
2895          * kthreads
2896          */
2897         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2898
2899 fail_block_groups:
2900         btrfs_put_block_group_cache(fs_info);
2901         btrfs_free_block_groups(fs_info);
2902
2903 fail_tree_roots:
2904         free_root_pointers(fs_info, 1);
2905         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2906
2907 fail_sb_buffer:
2908         btrfs_stop_all_workers(fs_info);
2909 fail_alloc:
2910 fail_iput:
2911         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2912
2913         iput(fs_info->btree_inode);
2914 fail_delalloc_bytes:
2915         percpu_counter_destroy(&fs_info->delalloc_bytes);
2916 fail_dirty_metadata_bytes:
2917         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2918 fail_bdi:
2919         bdi_destroy(&fs_info->bdi);
2920 fail_srcu:
2921         cleanup_srcu_struct(&fs_info->subvol_srcu);
2922 fail:
2923         btrfs_free_stripe_hash_table(fs_info);
2924         btrfs_close_devices(fs_info->fs_devices);
2925         return err;
2926
2927 recovery_tree_root:
2928         if (!btrfs_test_opt(tree_root, RECOVERY))
2929                 goto fail_tree_roots;
2930
2931         free_root_pointers(fs_info, 0);
2932
2933         /* don't use the log in recovery mode, it won't be valid */
2934         btrfs_set_super_log_root(disk_super, 0);
2935
2936         /* we can't trust the free space cache either */
2937         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2938
2939         ret = next_root_backup(fs_info, fs_info->super_copy,
2940                                &num_backups_tried, &backup_index);
2941         if (ret == -1)
2942                 goto fail_block_groups;
2943         goto retry_root_backup;
2944 }
2945
2946 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2947 {
2948         if (uptodate) {
2949                 set_buffer_uptodate(bh);
2950         } else {
2951                 struct btrfs_device *device = (struct btrfs_device *)
2952                         bh->b_private;
2953
2954                 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2955                                           "I/O error on %s\n",
2956                                           rcu_str_deref(device->name));
2957                 /* note, we dont' set_buffer_write_io_error because we have
2958                  * our own ways of dealing with the IO errors
2959                  */
2960                 clear_buffer_uptodate(bh);
2961                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2962         }
2963         unlock_buffer(bh);
2964         put_bh(bh);
2965 }
2966
2967 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2968 {
2969         struct buffer_head *bh;
2970         struct buffer_head *latest = NULL;
2971         struct btrfs_super_block *super;
2972         int i;
2973         u64 transid = 0;
2974         u64 bytenr;
2975
2976         /* we would like to check all the supers, but that would make
2977          * a btrfs mount succeed after a mkfs from a different FS.
2978          * So, we need to add a special mount option to scan for
2979          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2980          */
2981         for (i = 0; i < 1; i++) {
2982                 bytenr = btrfs_sb_offset(i);
2983                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2984                         break;
2985                 bh = __bread(bdev, bytenr / 4096, 4096);
2986                 if (!bh)
2987                         continue;
2988
2989                 super = (struct btrfs_super_block *)bh->b_data;
2990                 if (btrfs_super_bytenr(super) != bytenr ||
2991                     super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2992                         brelse(bh);
2993                         continue;
2994                 }
2995
2996                 if (!latest || btrfs_super_generation(super) > transid) {
2997                         brelse(latest);
2998                         latest = bh;
2999                         transid = btrfs_super_generation(super);
3000                 } else {
3001                         brelse(bh);
3002                 }
3003         }
3004         return latest;
3005 }
3006
3007 /*
3008  * this should be called twice, once with wait == 0 and
3009  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3010  * we write are pinned.
3011  *
3012  * They are released when wait == 1 is done.
3013  * max_mirrors must be the same for both runs, and it indicates how
3014  * many supers on this one device should be written.
3015  *
3016  * max_mirrors == 0 means to write them all.
3017  */
3018 static int write_dev_supers(struct btrfs_device *device,
3019                             struct btrfs_super_block *sb,
3020                             int do_barriers, int wait, int max_mirrors)
3021 {
3022         struct buffer_head *bh;
3023         int i;
3024         int ret;
3025         int errors = 0;
3026         u32 crc;
3027         u64 bytenr;
3028
3029         if (max_mirrors == 0)
3030                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3031
3032         for (i = 0; i < max_mirrors; i++) {
3033                 bytenr = btrfs_sb_offset(i);
3034                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3035                         break;
3036
3037                 if (wait) {
3038                         bh = __find_get_block(device->bdev, bytenr / 4096,
3039                                               BTRFS_SUPER_INFO_SIZE);
3040                         if (!bh) {
3041                                 errors++;
3042                                 continue;
3043                         }
3044                         wait_on_buffer(bh);
3045                         if (!buffer_uptodate(bh))
3046                                 errors++;
3047
3048                         /* drop our reference */
3049                         brelse(bh);
3050
3051                         /* drop the reference from the wait == 0 run */
3052                         brelse(bh);
3053                         continue;
3054                 } else {
3055                         btrfs_set_super_bytenr(sb, bytenr);
3056
3057                         crc = ~(u32)0;
3058                         crc = btrfs_csum_data((char *)sb +
3059                                               BTRFS_CSUM_SIZE, crc,
3060                                               BTRFS_SUPER_INFO_SIZE -
3061                                               BTRFS_CSUM_SIZE);
3062                         btrfs_csum_final(crc, sb->csum);
3063
3064                         /*
3065                          * one reference for us, and we leave it for the
3066                          * caller
3067                          */
3068                         bh = __getblk(device->bdev, bytenr / 4096,
3069                                       BTRFS_SUPER_INFO_SIZE);
3070                         if (!bh) {
3071                                 printk(KERN_ERR "btrfs: couldn't get super "
3072                                        "buffer head for bytenr %Lu\n", bytenr);
3073                                 errors++;
3074                                 continue;
3075                         }
3076
3077                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3078
3079                         /* one reference for submit_bh */
3080                         get_bh(bh);
3081
3082                         set_buffer_uptodate(bh);
3083                         lock_buffer(bh);
3084                         bh->b_end_io = btrfs_end_buffer_write_sync;
3085                         bh->b_private = device;
3086                 }
3087
3088                 /*
3089                  * we fua the first super.  The others we allow
3090                  * to go down lazy.
3091                  */
3092                 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3093                 if (ret)
3094                         errors++;
3095         }
3096         return errors < i ? 0 : -1;
3097 }
3098
3099 /*
3100  * endio for the write_dev_flush, this will wake anyone waiting
3101  * for the barrier when it is done
3102  */
3103 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3104 {
3105         if (err) {
3106                 if (err == -EOPNOTSUPP)
3107                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3108                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3109         }
3110         if (bio->bi_private)
3111                 complete(bio->bi_private);
3112         bio_put(bio);
3113 }
3114
3115 /*
3116  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3117  * sent down.  With wait == 1, it waits for the previous flush.
3118  *
3119  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3120  * capable
3121  */
3122 static int write_dev_flush(struct btrfs_device *device, int wait)
3123 {
3124         struct bio *bio;
3125         int ret = 0;
3126
3127         if (device->nobarriers)
3128                 return 0;
3129
3130         if (wait) {
3131                 bio = device->flush_bio;
3132                 if (!bio)
3133                         return 0;
3134
3135                 wait_for_completion(&device->flush_wait);
3136
3137                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3138                         printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3139                                       rcu_str_deref(device->name));
3140                         device->nobarriers = 1;
3141                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3142                         ret = -EIO;
3143                         btrfs_dev_stat_inc_and_print(device,
3144                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3145                 }
3146
3147                 /* drop the reference from the wait == 0 run */
3148                 bio_put(bio);
3149                 device->flush_bio = NULL;
3150
3151                 return ret;
3152         }
3153
3154         /*
3155          * one reference for us, and we leave it for the
3156          * caller
3157          */
3158         device->flush_bio = NULL;
3159         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3160         if (!bio)
3161                 return -ENOMEM;
3162
3163         bio->bi_end_io = btrfs_end_empty_barrier;
3164         bio->bi_bdev = device->bdev;
3165         init_completion(&device->flush_wait);
3166         bio->bi_private = &device->flush_wait;
3167         device->flush_bio = bio;
3168
3169         bio_get(bio);
3170         btrfsic_submit_bio(WRITE_FLUSH, bio);
3171
3172         return 0;
3173 }
3174
3175 /*
3176  * send an empty flush down to each device in parallel,
3177  * then wait for them
3178  */
3179 static int barrier_all_devices(struct btrfs_fs_info *info)
3180 {
3181         struct list_head *head;
3182         struct btrfs_device *dev;
3183         int errors_send = 0;
3184         int errors_wait = 0;
3185         int ret;
3186
3187         /* send down all the barriers */
3188         head = &info->fs_devices->devices;
3189         list_for_each_entry_rcu(dev, head, dev_list) {
3190                 if (!dev->bdev) {
3191                         errors_send++;
3192                         continue;
3193                 }
3194                 if (!dev->in_fs_metadata || !dev->writeable)
3195                         continue;
3196
3197                 ret = write_dev_flush(dev, 0);
3198                 if (ret)
3199                         errors_send++;
3200         }
3201
3202         /* wait for all the barriers */
3203         list_for_each_entry_rcu(dev, head, dev_list) {
3204                 if (!dev->bdev) {
3205                         errors_wait++;
3206                         continue;
3207                 }
3208                 if (!dev->in_fs_metadata || !dev->writeable)
3209                         continue;
3210
3211                 ret = write_dev_flush(dev, 1);
3212                 if (ret)
3213                         errors_wait++;
3214         }
3215         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3216             errors_wait > info->num_tolerated_disk_barrier_failures)
3217                 return -EIO;
3218         return 0;
3219 }
3220
3221 int btrfs_calc_num_tolerated_disk_barrier_failures(
3222         struct btrfs_fs_info *fs_info)
3223 {
3224         struct btrfs_ioctl_space_info space;
3225         struct btrfs_space_info *sinfo;
3226         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3227                        BTRFS_BLOCK_GROUP_SYSTEM,
3228                        BTRFS_BLOCK_GROUP_METADATA,
3229                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3230         int num_types = 4;
3231         int i;
3232         int c;
3233         int num_tolerated_disk_barrier_failures =
3234                 (int)fs_info->fs_devices->num_devices;
3235
3236         for (i = 0; i < num_types; i++) {
3237                 struct btrfs_space_info *tmp;
3238
3239                 sinfo = NULL;
3240                 rcu_read_lock();
3241                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3242                         if (tmp->flags == types[i]) {
3243                                 sinfo = tmp;
3244                                 break;
3245                         }
3246                 }
3247                 rcu_read_unlock();
3248
3249                 if (!sinfo)
3250                         continue;
3251
3252                 down_read(&sinfo->groups_sem);
3253                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3254                         if (!list_empty(&sinfo->block_groups[c])) {
3255                                 u64 flags;
3256
3257                                 btrfs_get_block_group_info(
3258                                         &sinfo->block_groups[c], &space);
3259                                 if (space.total_bytes == 0 ||
3260                                     space.used_bytes == 0)
3261                                         continue;
3262                                 flags = space.flags;
3263                                 /*
3264                                  * return
3265                                  * 0: if dup, single or RAID0 is configured for
3266                                  *    any of metadata, system or data, else
3267                                  * 1: if RAID5 is configured, or if RAID1 or
3268                                  *    RAID10 is configured and only two mirrors
3269                                  *    are used, else
3270                                  * 2: if RAID6 is configured, else
3271                                  * num_mirrors - 1: if RAID1 or RAID10 is
3272                                  *                  configured and more than
3273                                  *                  2 mirrors are used.
3274                                  */
3275                                 if (num_tolerated_disk_barrier_failures > 0 &&
3276                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3277                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3278                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3279                                       == 0)))
3280                                         num_tolerated_disk_barrier_failures = 0;
3281                                 else if (num_tolerated_disk_barrier_failures > 1) {
3282                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3283                                             BTRFS_BLOCK_GROUP_RAID5 |
3284                                             BTRFS_BLOCK_GROUP_RAID10)) {
3285                                                 num_tolerated_disk_barrier_failures = 1;
3286                                         } else if (flags &
3287                                                    BTRFS_BLOCK_GROUP_RAID6) {
3288                                                 num_tolerated_disk_barrier_failures = 2;
3289                                         }
3290                                 }
3291                         }
3292                 }
3293                 up_read(&sinfo->groups_sem);
3294         }
3295
3296         return num_tolerated_disk_barrier_failures;
3297 }
3298
3299 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3300 {
3301         struct list_head *head;
3302         struct btrfs_device *dev;
3303         struct btrfs_super_block *sb;
3304         struct btrfs_dev_item *dev_item;
3305         int ret;
3306         int do_barriers;
3307         int max_errors;
3308         int total_errors = 0;
3309         u64 flags;
3310
3311         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3312         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3313         backup_super_roots(root->fs_info);
3314
3315         sb = root->fs_info->super_for_commit;
3316         dev_item = &sb->dev_item;
3317
3318         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3319         head = &root->fs_info->fs_devices->devices;
3320
3321         if (do_barriers) {
3322                 ret = barrier_all_devices(root->fs_info);
3323                 if (ret) {
3324                         mutex_unlock(
3325                                 &root->fs_info->fs_devices->device_list_mutex);
3326                         btrfs_error(root->fs_info, ret,
3327                                     "errors while submitting device barriers.");
3328                         return ret;
3329                 }
3330         }
3331
3332         list_for_each_entry_rcu(dev, head, dev_list) {
3333                 if (!dev->bdev) {
3334                         total_errors++;
3335                         continue;
3336                 }
3337                 if (!dev->in_fs_metadata || !dev->writeable)
3338                         continue;
3339
3340                 btrfs_set_stack_device_generation(dev_item, 0);
3341                 btrfs_set_stack_device_type(dev_item, dev->type);
3342                 btrfs_set_stack_device_id(dev_item, dev->devid);
3343                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3344                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3345                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3346                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3347                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3348                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3349                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3350
3351                 flags = btrfs_super_flags(sb);
3352                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3353
3354                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3355                 if (ret)
3356                         total_errors++;
3357         }
3358         if (total_errors > max_errors) {
3359                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3360                        total_errors);
3361
3362                 /* This shouldn't happen. FUA is masked off if unsupported */
3363                 BUG();
3364         }
3365
3366         total_errors = 0;
3367         list_for_each_entry_rcu(dev, head, dev_list) {
3368                 if (!dev->bdev)
3369                         continue;
3370                 if (!dev->in_fs_metadata || !dev->writeable)
3371                         continue;
3372
3373                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3374                 if (ret)
3375                         total_errors++;
3376         }
3377         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3378         if (total_errors > max_errors) {
3379                 btrfs_error(root->fs_info, -EIO,
3380                             "%d errors while writing supers", total_errors);
3381                 return -EIO;
3382         }
3383         return 0;
3384 }
3385
3386 int write_ctree_super(struct btrfs_trans_handle *trans,
3387                       struct btrfs_root *root, int max_mirrors)
3388 {
3389         int ret;
3390
3391         ret = write_all_supers(root, max_mirrors);
3392         return ret;
3393 }
3394
3395 void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3396 {
3397         spin_lock(&fs_info->fs_roots_radix_lock);
3398         radix_tree_delete(&fs_info->fs_roots_radix,
3399                           (unsigned long)root->root_key.objectid);
3400         spin_unlock(&fs_info->fs_roots_radix_lock);
3401
3402         if (btrfs_root_refs(&root->root_item) == 0)
3403                 synchronize_srcu(&fs_info->subvol_srcu);
3404
3405         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3406                 btrfs_free_log(NULL, root);
3407                 btrfs_free_log_root_tree(NULL, fs_info);
3408         }
3409
3410         __btrfs_remove_free_space_cache(root->free_ino_pinned);
3411         __btrfs_remove_free_space_cache(root->free_ino_ctl);
3412         free_fs_root(root);
3413 }
3414
3415 static void free_fs_root(struct btrfs_root *root)
3416 {
3417         iput(root->cache_inode);
3418         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3419         if (root->anon_dev)
3420                 free_anon_bdev(root->anon_dev);
3421         free_extent_buffer(root->node);
3422         free_extent_buffer(root->commit_root);
3423         kfree(root->free_ino_ctl);
3424         kfree(root->free_ino_pinned);
3425         kfree(root->name);
3426         kfree(root);
3427 }
3428
3429 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3430 {
3431         u64 root_objectid = 0;
3432         struct btrfs_root *gang[8];
3433         int i;
3434         int ret;
3435
3436         while (1) {
3437                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3438                                              (void **)gang, root_objectid,
3439                                              ARRAY_SIZE(gang));
3440                 if (!ret)
3441                         break;
3442
3443                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3444                 for (i = 0; i < ret; i++) {
3445                         int err;
3446
3447                         root_objectid = gang[i]->root_key.objectid;
3448                         err = btrfs_orphan_cleanup(gang[i]);
3449                         if (err)
3450                                 return err;
3451                 }
3452                 root_objectid++;
3453         }
3454         return 0;
3455 }
3456
3457 int btrfs_commit_super(struct btrfs_root *root)
3458 {
3459         struct btrfs_trans_handle *trans;
3460         int ret;
3461
3462         mutex_lock(&root->fs_info->cleaner_mutex);
3463         btrfs_run_delayed_iputs(root);
3464         mutex_unlock(&root->fs_info->cleaner_mutex);
3465         wake_up_process(root->fs_info->cleaner_kthread);
3466
3467         /* wait until ongoing cleanup work done */
3468         down_write(&root->fs_info->cleanup_work_sem);
3469         up_write(&root->fs_info->cleanup_work_sem);
3470
3471         trans = btrfs_join_transaction(root);
3472         if (IS_ERR(trans))
3473                 return PTR_ERR(trans);
3474         ret = btrfs_commit_transaction(trans, root);
3475         if (ret)
3476                 return ret;
3477         /* run commit again to drop the original snapshot */
3478         trans = btrfs_join_transaction(root);
3479         if (IS_ERR(trans))
3480                 return PTR_ERR(trans);
3481         ret = btrfs_commit_transaction(trans, root);
3482         if (ret)
3483                 return ret;
3484         ret = btrfs_write_and_wait_transaction(NULL, root);
3485         if (ret) {
3486                 btrfs_error(root->fs_info, ret,
3487                             "Failed to sync btree inode to disk.");
3488                 return ret;
3489         }
3490
3491         ret = write_ctree_super(NULL, root, 0);
3492         return ret;
3493 }
3494
3495 int close_ctree(struct btrfs_root *root)
3496 {
3497         struct btrfs_fs_info *fs_info = root->fs_info;
3498         int ret;
3499
3500         fs_info->closing = 1;
3501         smp_mb();
3502
3503         /* pause restriper - we want to resume on mount */
3504         btrfs_pause_balance(fs_info);
3505
3506         btrfs_dev_replace_suspend_for_unmount(fs_info);
3507
3508         btrfs_scrub_cancel(fs_info);
3509
3510         /* wait for any defraggers to finish */
3511         wait_event(fs_info->transaction_wait,
3512                    (atomic_read(&fs_info->defrag_running) == 0));
3513
3514         /* clear out the rbtree of defraggable inodes */
3515         btrfs_cleanup_defrag_inodes(fs_info);
3516
3517         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3518                 ret = btrfs_commit_super(root);
3519                 if (ret)
3520                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3521         }
3522
3523         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3524                 btrfs_error_commit_super(root);
3525
3526         btrfs_put_block_group_cache(fs_info);
3527
3528         kthread_stop(fs_info->transaction_kthread);
3529         kthread_stop(fs_info->cleaner_kthread);
3530
3531         fs_info->closing = 2;
3532         smp_mb();
3533
3534         btrfs_free_qgroup_config(root->fs_info);
3535
3536         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3537                 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3538                        percpu_counter_sum(&fs_info->delalloc_bytes));
3539         }
3540
3541         btrfs_free_block_groups(fs_info);
3542
3543         btrfs_stop_all_workers(fs_info);
3544
3545         del_fs_roots(fs_info);
3546
3547         free_root_pointers(fs_info, 1);
3548
3549         iput(fs_info->btree_inode);
3550
3551 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3552         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3553                 btrfsic_unmount(root, fs_info->fs_devices);
3554 #endif
3555
3556         btrfs_close_devices(fs_info->fs_devices);
3557         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3558
3559         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3560         percpu_counter_destroy(&fs_info->delalloc_bytes);
3561         bdi_destroy(&fs_info->bdi);
3562         cleanup_srcu_struct(&fs_info->subvol_srcu);
3563
3564         btrfs_free_stripe_hash_table(fs_info);
3565
3566         return 0;
3567 }
3568
3569 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3570                           int atomic)
3571 {
3572         int ret;
3573         struct inode *btree_inode = buf->pages[0]->mapping->host;
3574
3575         ret = extent_buffer_uptodate(buf);
3576         if (!ret)
3577                 return ret;
3578
3579         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3580                                     parent_transid, atomic);
3581         if (ret == -EAGAIN)
3582                 return ret;
3583         return !ret;
3584 }
3585
3586 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3587 {
3588         return set_extent_buffer_uptodate(buf);
3589 }
3590
3591 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3592 {
3593         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3594         u64 transid = btrfs_header_generation(buf);
3595         int was_dirty;
3596
3597         btrfs_assert_tree_locked(buf);
3598         if (transid != root->fs_info->generation)
3599                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3600                        "found %llu running %llu\n",
3601                         (unsigned long long)buf->start,
3602                         (unsigned long long)transid,
3603                         (unsigned long long)root->fs_info->generation);
3604         was_dirty = set_extent_buffer_dirty(buf);
3605         if (!was_dirty)
3606                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3607                                      buf->len,
3608                                      root->fs_info->dirty_metadata_batch);
3609 }
3610
3611 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3612                                         int flush_delayed)
3613 {
3614         /*
3615          * looks as though older kernels can get into trouble with
3616          * this code, they end up stuck in balance_dirty_pages forever
3617          */
3618         int ret;
3619
3620         if (current->flags & PF_MEMALLOC)
3621                 return;
3622
3623         if (flush_delayed)
3624                 btrfs_balance_delayed_items(root);
3625
3626         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3627                                      BTRFS_DIRTY_METADATA_THRESH);
3628         if (ret > 0) {
3629                 balance_dirty_pages_ratelimited(
3630                                    root->fs_info->btree_inode->i_mapping);
3631         }
3632         return;
3633 }
3634
3635 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3636 {
3637         __btrfs_btree_balance_dirty(root, 1);
3638 }
3639
3640 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3641 {
3642         __btrfs_btree_balance_dirty(root, 0);
3643 }
3644
3645 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3646 {
3647         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3648         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3649 }
3650
3651 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3652                               int read_only)
3653 {
3654         /*
3655          * Placeholder for checks
3656          */
3657         return 0;
3658 }
3659
3660 static void btrfs_error_commit_super(struct btrfs_root *root)
3661 {
3662         mutex_lock(&root->fs_info->cleaner_mutex);
3663         btrfs_run_delayed_iputs(root);
3664         mutex_unlock(&root->fs_info->cleaner_mutex);
3665
3666         down_write(&root->fs_info->cleanup_work_sem);
3667         up_write(&root->fs_info->cleanup_work_sem);
3668
3669         /* cleanup FS via transaction */
3670         btrfs_cleanup_transaction(root);
3671 }
3672
3673 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3674                                              struct btrfs_root *root)
3675 {
3676         struct btrfs_inode *btrfs_inode;
3677         struct list_head splice;
3678
3679         INIT_LIST_HEAD(&splice);
3680
3681         mutex_lock(&root->fs_info->ordered_operations_mutex);
3682         spin_lock(&root->fs_info->ordered_extent_lock);
3683
3684         list_splice_init(&t->ordered_operations, &splice);
3685         while (!list_empty(&splice)) {
3686                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3687                                          ordered_operations);
3688
3689                 list_del_init(&btrfs_inode->ordered_operations);
3690                 spin_unlock(&root->fs_info->ordered_extent_lock);
3691
3692                 btrfs_invalidate_inodes(btrfs_inode->root);
3693
3694                 spin_lock(&root->fs_info->ordered_extent_lock);
3695         }
3696
3697         spin_unlock(&root->fs_info->ordered_extent_lock);
3698         mutex_unlock(&root->fs_info->ordered_operations_mutex);
3699 }
3700
3701 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3702 {
3703         struct btrfs_ordered_extent *ordered;
3704
3705         spin_lock(&root->fs_info->ordered_extent_lock);
3706         /*
3707          * This will just short circuit the ordered completion stuff which will
3708          * make sure the ordered extent gets properly cleaned up.
3709          */
3710         list_for_each_entry(ordered, &root->fs_info->ordered_extents,
3711                             root_extent_list)
3712                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3713         spin_unlock(&root->fs_info->ordered_extent_lock);
3714 }
3715
3716 int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3717                                struct btrfs_root *root)
3718 {
3719         struct rb_node *node;
3720         struct btrfs_delayed_ref_root *delayed_refs;
3721         struct btrfs_delayed_ref_node *ref;
3722         int ret = 0;
3723
3724         delayed_refs = &trans->delayed_refs;
3725
3726         spin_lock(&delayed_refs->lock);
3727         if (delayed_refs->num_entries == 0) {
3728                 spin_unlock(&delayed_refs->lock);
3729                 printk(KERN_INFO "delayed_refs has NO entry\n");
3730                 return ret;
3731         }
3732
3733         while ((node = rb_first(&delayed_refs->root)) != NULL) {
3734                 struct btrfs_delayed_ref_head *head = NULL;
3735
3736                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3737                 atomic_set(&ref->refs, 1);
3738                 if (btrfs_delayed_ref_is_head(ref)) {
3739
3740                         head = btrfs_delayed_node_to_head(ref);
3741                         if (!mutex_trylock(&head->mutex)) {
3742                                 atomic_inc(&ref->refs);
3743                                 spin_unlock(&delayed_refs->lock);
3744
3745                                 /* Need to wait for the delayed ref to run */
3746                                 mutex_lock(&head->mutex);
3747                                 mutex_unlock(&head->mutex);
3748                                 btrfs_put_delayed_ref(ref);
3749
3750                                 spin_lock(&delayed_refs->lock);
3751                                 continue;
3752                         }
3753
3754                         if (head->must_insert_reserved)
3755                                 btrfs_pin_extent(root, ref->bytenr,
3756                                                  ref->num_bytes, 1);
3757                         btrfs_free_delayed_extent_op(head->extent_op);
3758                         delayed_refs->num_heads--;
3759                         if (list_empty(&head->cluster))
3760                                 delayed_refs->num_heads_ready--;
3761                         list_del_init(&head->cluster);
3762                 }
3763
3764                 ref->in_tree = 0;
3765                 rb_erase(&ref->rb_node, &delayed_refs->root);
3766                 delayed_refs->num_entries--;
3767                 if (head)
3768                         mutex_unlock(&head->mutex);
3769                 spin_unlock(&delayed_refs->lock);
3770                 btrfs_put_delayed_ref(ref);
3771
3772                 cond_resched();
3773                 spin_lock(&delayed_refs->lock);
3774         }
3775
3776         spin_unlock(&delayed_refs->lock);
3777
3778         return ret;
3779 }
3780
3781 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3782 {
3783         struct btrfs_pending_snapshot *snapshot;
3784         struct list_head splice;
3785
3786         INIT_LIST_HEAD(&splice);
3787
3788         list_splice_init(&t->pending_snapshots, &splice);
3789
3790         while (!list_empty(&splice)) {
3791                 snapshot = list_entry(splice.next,
3792                                       struct btrfs_pending_snapshot,
3793                                       list);
3794                 snapshot->error = -ECANCELED;
3795                 list_del_init(&snapshot->list);
3796         }
3797 }
3798
3799 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3800 {
3801         struct btrfs_inode *btrfs_inode;
3802         struct list_head splice;
3803
3804         INIT_LIST_HEAD(&splice);
3805
3806         spin_lock(&root->fs_info->delalloc_lock);
3807         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3808
3809         while (!list_empty(&splice)) {
3810                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3811                                     delalloc_inodes);
3812
3813                 list_del_init(&btrfs_inode->delalloc_inodes);
3814                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3815                           &btrfs_inode->runtime_flags);
3816                 spin_unlock(&root->fs_info->delalloc_lock);
3817
3818                 btrfs_invalidate_inodes(btrfs_inode->root);
3819
3820                 spin_lock(&root->fs_info->delalloc_lock);
3821         }
3822
3823         spin_unlock(&root->fs_info->delalloc_lock);
3824 }
3825
3826 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3827                                         struct extent_io_tree *dirty_pages,
3828                                         int mark)
3829 {
3830         int ret;
3831         struct extent_buffer *eb;
3832         u64 start = 0;
3833         u64 end;
3834
3835         while (1) {
3836                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3837                                             mark, NULL);
3838                 if (ret)
3839                         break;
3840
3841                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3842                 while (start <= end) {
3843                         eb = btrfs_find_tree_block(root, start,
3844                                                    root->leafsize);
3845                         start += root->leafsize;
3846                         if (!eb)
3847                                 continue;
3848                         wait_on_extent_buffer_writeback(eb);
3849
3850                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3851                                                &eb->bflags))
3852                                 clear_extent_buffer_dirty(eb);
3853                         free_extent_buffer_stale(eb);
3854                 }
3855         }
3856
3857         return ret;
3858 }
3859
3860 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3861                                        struct extent_io_tree *pinned_extents)
3862 {
3863         struct extent_io_tree *unpin;
3864         u64 start;
3865         u64 end;
3866         int ret;
3867         bool loop = true;
3868
3869         unpin = pinned_extents;
3870 again:
3871         while (1) {
3872                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3873                                             EXTENT_DIRTY, NULL);
3874                 if (ret)
3875                         break;
3876
3877                 /* opt_discard */
3878                 if (btrfs_test_opt(root, DISCARD))
3879                         ret = btrfs_error_discard_extent(root, start,
3880                                                          end + 1 - start,
3881                                                          NULL);
3882
3883                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3884                 btrfs_error_unpin_extent_range(root, start, end);
3885                 cond_resched();
3886         }
3887
3888         if (loop) {
3889                 if (unpin == &root->fs_info->freed_extents[0])
3890                         unpin = &root->fs_info->freed_extents[1];
3891                 else
3892                         unpin = &root->fs_info->freed_extents[0];
3893                 loop = false;
3894                 goto again;
3895         }
3896
3897         return 0;
3898 }
3899
3900 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3901                                    struct btrfs_root *root)
3902 {
3903         btrfs_destroy_delayed_refs(cur_trans, root);
3904         btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3905                                 cur_trans->dirty_pages.dirty_bytes);
3906
3907         /* FIXME: cleanup wait for commit */
3908         cur_trans->in_commit = 1;
3909         cur_trans->blocked = 1;
3910         wake_up(&root->fs_info->transaction_blocked_wait);
3911
3912         btrfs_evict_pending_snapshots(cur_trans);
3913
3914         cur_trans->blocked = 0;
3915         wake_up(&root->fs_info->transaction_wait);
3916
3917         cur_trans->commit_done = 1;
3918         wake_up(&cur_trans->commit_wait);
3919
3920         btrfs_destroy_delayed_inodes(root);
3921         btrfs_assert_delayed_root_empty(root);
3922
3923         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3924                                      EXTENT_DIRTY);
3925         btrfs_destroy_pinned_extent(root,
3926                                     root->fs_info->pinned_extents);
3927
3928         /*
3929         memset(cur_trans, 0, sizeof(*cur_trans));
3930         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3931         */
3932 }
3933
3934 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3935 {
3936         struct btrfs_transaction *t;
3937         LIST_HEAD(list);
3938
3939         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3940
3941         spin_lock(&root->fs_info->trans_lock);
3942         list_splice_init(&root->fs_info->trans_list, &list);
3943         root->fs_info->trans_no_join = 1;
3944         spin_unlock(&root->fs_info->trans_lock);
3945
3946         while (!list_empty(&list)) {
3947                 t = list_entry(list.next, struct btrfs_transaction, list);
3948
3949                 btrfs_destroy_ordered_operations(t, root);
3950
3951                 btrfs_destroy_ordered_extents(root);
3952
3953                 btrfs_destroy_delayed_refs(t, root);
3954
3955                 /* FIXME: cleanup wait for commit */
3956                 t->in_commit = 1;
3957                 t->blocked = 1;
3958                 smp_mb();
3959                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3960                         wake_up(&root->fs_info->transaction_blocked_wait);
3961
3962                 btrfs_evict_pending_snapshots(t);
3963
3964                 t->blocked = 0;
3965                 smp_mb();
3966                 if (waitqueue_active(&root->fs_info->transaction_wait))
3967                         wake_up(&root->fs_info->transaction_wait);
3968
3969                 t->commit_done = 1;
3970                 smp_mb();
3971                 if (waitqueue_active(&t->commit_wait))
3972                         wake_up(&t->commit_wait);
3973
3974                 btrfs_destroy_delayed_inodes(root);
3975                 btrfs_assert_delayed_root_empty(root);
3976
3977                 btrfs_destroy_delalloc_inodes(root);
3978
3979                 spin_lock(&root->fs_info->trans_lock);
3980                 root->fs_info->running_transaction = NULL;
3981                 spin_unlock(&root->fs_info->trans_lock);
3982
3983                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3984                                              EXTENT_DIRTY);
3985
3986                 btrfs_destroy_pinned_extent(root,
3987                                             root->fs_info->pinned_extents);
3988
3989                 atomic_set(&t->use_count, 0);
3990                 list_del_init(&t->list);
3991                 memset(t, 0, sizeof(*t));
3992                 kmem_cache_free(btrfs_transaction_cachep, t);
3993         }
3994
3995         spin_lock(&root->fs_info->trans_lock);
3996         root->fs_info->trans_no_join = 0;
3997         spin_unlock(&root->fs_info->trans_lock);
3998         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3999
4000         return 0;
4001 }
4002
4003 static struct extent_io_ops btree_extent_io_ops = {
4004         .readpage_end_io_hook = btree_readpage_end_io_hook,
4005         .readpage_io_failed_hook = btree_io_failed_hook,
4006         .submit_bio_hook = btree_submit_bio_hook,
4007         /* note we're sharing with inode.c for the merge bio hook */
4008         .merge_bio_hook = btrfs_merge_bio_hook,
4009 };