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

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

Version: ~ [ linux-5.11-rc3 ] ~ [ linux-5.10.7 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.89 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.167 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.215 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.251 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.251 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (C) 2007 Oracle.  All rights reserved.
  3  *
  4  * This program is free software; you can redistribute it and/or
  5  * modify it under the terms of the GNU General Public
  6  * License v2 as published by the Free Software Foundation.
  7  *
  8  * This program is distributed in the hope that it will be useful,
  9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 11  * General Public License for more details.
 12  *
 13  * You should have received a copy of the GNU General Public
 14  * License along with this program; if not, write to the
 15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 16  * Boston, MA 021110-1307, USA.
 17  */
 18 
 19 #include <linux/fs.h>
 20 #include <linux/blkdev.h>
 21 #include <linux/scatterlist.h>
 22 #include <linux/swap.h>
 23 #include <linux/radix-tree.h>
 24 #include <linux/writeback.h>
 25 #include <linux/buffer_head.h>
 26 #include <linux/workqueue.h>
 27 #include <linux/kthread.h>
 28 #include <linux/freezer.h>
 29 #include <linux/slab.h>
 30 #include <linux/migrate.h>
 31 #include <linux/ratelimit.h>
 32 #include <linux/uuid.h>
 33 #include <linux/semaphore.h>
 34 #include <asm/unaligned.h>
 35 #include "ctree.h"
 36 #include "disk-io.h"
 37 #include "hash.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 #include "sysfs.h"
 52 
 53 #ifdef CONFIG_X86
 54 #include <asm/cpufeature.h>
 55 #endif
 56 
 57 static struct extent_io_ops btree_extent_io_ops;
 58 static void end_workqueue_fn(struct btrfs_work *work);
 59 static void free_fs_root(struct btrfs_root *root);
 60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
 61                                     int read_only);
 62 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
 63                                              struct btrfs_root *root);
 64 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
 65 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
 66                                       struct btrfs_root *root);
 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 = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
160         { .id = 0,                              .name_stem = "tree"     },
161 };
162 
163 void __init btrfs_init_lockdep(void)
164 {
165         int i, j;
166 
167         /* initialize lockdep class names */
168         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
170 
171                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172                         snprintf(ks->names[j], sizeof(ks->names[j]),
173                                  "btrfs-%s-%02d", ks->name_stem, j);
174         }
175 }
176 
177 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
178                                     int level)
179 {
180         struct btrfs_lockdep_keyset *ks;
181 
182         BUG_ON(level >= ARRAY_SIZE(ks->keys));
183 
184         /* find the matching keyset, id 0 is the default entry */
185         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186                 if (ks->id == objectid)
187                         break;
188 
189         lockdep_set_class_and_name(&eb->lock,
190                                    &ks->keys[level], ks->names[level]);
191 }
192 
193 #endif
194 
195 /*
196  * extents on the btree inode are pretty simple, there's one extent
197  * that covers the entire device
198  */
199 static struct extent_map *btree_get_extent(struct inode *inode,
200                 struct page *page, size_t pg_offset, u64 start, u64 len,
201                 int create)
202 {
203         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204         struct extent_map *em;
205         int ret;
206 
207         read_lock(&em_tree->lock);
208         em = lookup_extent_mapping(em_tree, start, len);
209         if (em) {
210                 em->bdev =
211                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
212                 read_unlock(&em_tree->lock);
213                 goto out;
214         }
215         read_unlock(&em_tree->lock);
216 
217         em = alloc_extent_map();
218         if (!em) {
219                 em = ERR_PTR(-ENOMEM);
220                 goto out;
221         }
222         em->start = 0;
223         em->len = (u64)-1;
224         em->block_len = (u64)-1;
225         em->block_start = 0;
226         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
227 
228         write_lock(&em_tree->lock);
229         ret = add_extent_mapping(em_tree, em, 0);
230         if (ret == -EEXIST) {
231                 free_extent_map(em);
232                 em = lookup_extent_mapping(em_tree, start, len);
233                 if (!em)
234                         em = ERR_PTR(-EIO);
235         } else if (ret) {
236                 free_extent_map(em);
237                 em = ERR_PTR(ret);
238         }
239         write_unlock(&em_tree->lock);
240 
241 out:
242         return em;
243 }
244 
245 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
246 {
247         return btrfs_crc32c(seed, data, len);
248 }
249 
250 void btrfs_csum_final(u32 crc, char *result)
251 {
252         put_unaligned_le32(~crc, result);
253 }
254 
255 /*
256  * compute the csum for a btree block, and either verify it or write it
257  * into the csum field of the block.
258  */
259 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260                            int verify)
261 {
262         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
263         char *result = NULL;
264         unsigned long len;
265         unsigned long cur_len;
266         unsigned long offset = BTRFS_CSUM_SIZE;
267         char *kaddr;
268         unsigned long map_start;
269         unsigned long map_len;
270         int err;
271         u32 crc = ~(u32)0;
272         unsigned long inline_result;
273 
274         len = buf->len - offset;
275         while (len > 0) {
276                 err = map_private_extent_buffer(buf, offset, 32,
277                                         &kaddr, &map_start, &map_len);
278                 if (err)
279                         return 1;
280                 cur_len = min(len, map_len - (offset - map_start));
281                 crc = btrfs_csum_data(kaddr + offset - map_start,
282                                       crc, cur_len);
283                 len -= cur_len;
284                 offset += cur_len;
285         }
286         if (csum_size > sizeof(inline_result)) {
287                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
288                 if (!result)
289                         return 1;
290         } else {
291                 result = (char *)&inline_result;
292         }
293 
294         btrfs_csum_final(crc, result);
295 
296         if (verify) {
297                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
298                         u32 val;
299                         u32 found = 0;
300                         memcpy(&found, result, csum_size);
301 
302                         read_extent_buffer(buf, &val, 0, csum_size);
303                         printk_ratelimited(KERN_INFO
304                                 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
305                                 "level %d\n",
306                                 root->fs_info->sb->s_id, buf->start,
307                                 val, found, 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                        eb->start, parent_transid, btrfs_header_generation(eb));
349         ret = 1;
350         clear_extent_buffer_uptodate(eb);
351 out:
352         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
353                              &cached_state, GFP_NOFS);
354         return ret;
355 }
356 
357 /*
358  * Return 0 if the superblock checksum type matches the checksum value of that
359  * algorithm. Pass the raw disk superblock data.
360  */
361 static int btrfs_check_super_csum(char *raw_disk_sb)
362 {
363         struct btrfs_super_block *disk_sb =
364                 (struct btrfs_super_block *)raw_disk_sb;
365         u16 csum_type = btrfs_super_csum_type(disk_sb);
366         int ret = 0;
367 
368         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
369                 u32 crc = ~(u32)0;
370                 const int csum_size = sizeof(crc);
371                 char result[csum_size];
372 
373                 /*
374                  * The super_block structure does not span the whole
375                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
376                  * is filled with zeros and is included in the checkum.
377                  */
378                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
379                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
380                 btrfs_csum_final(crc, result);
381 
382                 if (memcmp(raw_disk_sb, result, csum_size))
383                         ret = 1;
384 
385                 if (ret && btrfs_super_generation(disk_sb) < 10) {
386                         printk(KERN_WARNING
387                                 "BTRFS: super block crcs don't match, older mkfs detected\n");
388                         ret = 0;
389                 }
390         }
391 
392         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
393                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
394                                 csum_type);
395                 ret = 1;
396         }
397 
398         return ret;
399 }
400 
401 /*
402  * helper to read a given tree block, doing retries as required when
403  * the checksums don't match and we have alternate mirrors to try.
404  */
405 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
406                                           struct extent_buffer *eb,
407                                           u64 start, u64 parent_transid)
408 {
409         struct extent_io_tree *io_tree;
410         int failed = 0;
411         int ret;
412         int num_copies = 0;
413         int mirror_num = 0;
414         int failed_mirror = 0;
415 
416         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
417         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
418         while (1) {
419                 ret = read_extent_buffer_pages(io_tree, eb, start,
420                                                WAIT_COMPLETE,
421                                                btree_get_extent, mirror_num);
422                 if (!ret) {
423                         if (!verify_parent_transid(io_tree, eb,
424                                                    parent_transid, 0))
425                                 break;
426                         else
427                                 ret = -EIO;
428                 }
429 
430                 /*
431                  * This buffer's crc is fine, but its contents are corrupted, so
432                  * there is no reason to read the other copies, they won't be
433                  * any less wrong.
434                  */
435                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
436                         break;
437 
438                 num_copies = btrfs_num_copies(root->fs_info,
439                                               eb->start, eb->len);
440                 if (num_copies == 1)
441                         break;
442 
443                 if (!failed_mirror) {
444                         failed = 1;
445                         failed_mirror = eb->read_mirror;
446                 }
447 
448                 mirror_num++;
449                 if (mirror_num == failed_mirror)
450                         mirror_num++;
451 
452                 if (mirror_num > num_copies)
453                         break;
454         }
455 
456         if (failed && !ret && failed_mirror)
457                 repair_eb_io_failure(root, eb, failed_mirror);
458 
459         return ret;
460 }
461 
462 /*
463  * checksum a dirty tree block before IO.  This has extra checks to make sure
464  * we only fill in the checksum field in the first page of a multi-page block
465  */
466 
467 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
468 {
469         u64 start = page_offset(page);
470         u64 found_start;
471         struct extent_buffer *eb;
472 
473         eb = (struct extent_buffer *)page->private;
474         if (page != eb->pages[0])
475                 return 0;
476         found_start = btrfs_header_bytenr(eb);
477         if (WARN_ON(found_start != start || !PageUptodate(page)))
478                 return 0;
479         csum_tree_block(root, eb, 0);
480         return 0;
481 }
482 
483 static int check_tree_block_fsid(struct btrfs_root *root,
484                                  struct extent_buffer *eb)
485 {
486         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
487         u8 fsid[BTRFS_UUID_SIZE];
488         int ret = 1;
489 
490         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
491         while (fs_devices) {
492                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
493                         ret = 0;
494                         break;
495                 }
496                 fs_devices = fs_devices->seed;
497         }
498         return ret;
499 }
500 
501 #define CORRUPT(reason, eb, root, slot)                         \
502         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
503                    "root=%llu, slot=%d", reason,                        \
504                btrfs_header_bytenr(eb), root->objectid, slot)
505 
506 static noinline int check_leaf(struct btrfs_root *root,
507                                struct extent_buffer *leaf)
508 {
509         struct btrfs_key key;
510         struct btrfs_key leaf_key;
511         u32 nritems = btrfs_header_nritems(leaf);
512         int slot;
513 
514         if (nritems == 0)
515                 return 0;
516 
517         /* Check the 0 item */
518         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
519             BTRFS_LEAF_DATA_SIZE(root)) {
520                 CORRUPT("invalid item offset size pair", leaf, root, 0);
521                 return -EIO;
522         }
523 
524         /*
525          * Check to make sure each items keys are in the correct order and their
526          * offsets make sense.  We only have to loop through nritems-1 because
527          * we check the current slot against the next slot, which verifies the
528          * next slot's offset+size makes sense and that the current's slot
529          * offset is correct.
530          */
531         for (slot = 0; slot < nritems - 1; slot++) {
532                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
533                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
534 
535                 /* Make sure the keys are in the right order */
536                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
537                         CORRUPT("bad key order", leaf, root, slot);
538                         return -EIO;
539                 }
540 
541                 /*
542                  * Make sure the offset and ends are right, remember that the
543                  * item data starts at the end of the leaf and grows towards the
544                  * front.
545                  */
546                 if (btrfs_item_offset_nr(leaf, slot) !=
547                         btrfs_item_end_nr(leaf, slot + 1)) {
548                         CORRUPT("slot offset bad", leaf, root, slot);
549                         return -EIO;
550                 }
551 
552                 /*
553                  * Check to make sure that we don't point outside of the leaf,
554                  * just incase all the items are consistent to eachother, but
555                  * all point outside of the leaf.
556                  */
557                 if (btrfs_item_end_nr(leaf, slot) >
558                     BTRFS_LEAF_DATA_SIZE(root)) {
559                         CORRUPT("slot end outside of leaf", leaf, root, slot);
560                         return -EIO;
561                 }
562         }
563 
564         return 0;
565 }
566 
567 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
568                                       u64 phy_offset, struct page *page,
569                                       u64 start, u64 end, int mirror)
570 {
571         u64 found_start;
572         int found_level;
573         struct extent_buffer *eb;
574         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
575         int ret = 0;
576         int reads_done;
577 
578         if (!page->private)
579                 goto out;
580 
581         eb = (struct extent_buffer *)page->private;
582 
583         /* the pending IO might have been the only thing that kept this buffer
584          * in memory.  Make sure we have a ref for all this other checks
585          */
586         extent_buffer_get(eb);
587 
588         reads_done = atomic_dec_and_test(&eb->io_pages);
589         if (!reads_done)
590                 goto err;
591 
592         eb->read_mirror = mirror;
593         if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
594                 ret = -EIO;
595                 goto err;
596         }
597 
598         found_start = btrfs_header_bytenr(eb);
599         if (found_start != eb->start) {
600                 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
601                                "%llu %llu\n",
602                                found_start, eb->start);
603                 ret = -EIO;
604                 goto err;
605         }
606         if (check_tree_block_fsid(root, eb)) {
607                 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
608                                eb->start);
609                 ret = -EIO;
610                 goto err;
611         }
612         found_level = btrfs_header_level(eb);
613         if (found_level >= BTRFS_MAX_LEVEL) {
614                 btrfs_info(root->fs_info, "bad tree block level %d",
615                            (int)btrfs_header_level(eb));
616                 ret = -EIO;
617                 goto err;
618         }
619 
620         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
621                                        eb, found_level);
622 
623         ret = csum_tree_block(root, eb, 1);
624         if (ret) {
625                 ret = -EIO;
626                 goto err;
627         }
628 
629         /*
630          * If this is a leaf block and it is corrupt, set the corrupt bit so
631          * that we don't try and read the other copies of this block, just
632          * return -EIO.
633          */
634         if (found_level == 0 && check_leaf(root, eb)) {
635                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
636                 ret = -EIO;
637         }
638 
639         if (!ret)
640                 set_extent_buffer_uptodate(eb);
641 err:
642         if (reads_done &&
643             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
644                 btree_readahead_hook(root, eb, eb->start, ret);
645 
646         if (ret) {
647                 /*
648                  * our io error hook is going to dec the io pages
649                  * again, we have to make sure it has something
650                  * to decrement
651                  */
652                 atomic_inc(&eb->io_pages);
653                 clear_extent_buffer_uptodate(eb);
654         }
655         free_extent_buffer(eb);
656 out:
657         return ret;
658 }
659 
660 static int btree_io_failed_hook(struct page *page, int failed_mirror)
661 {
662         struct extent_buffer *eb;
663         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
664 
665         eb = (struct extent_buffer *)page->private;
666         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
667         eb->read_mirror = failed_mirror;
668         atomic_dec(&eb->io_pages);
669         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
670                 btree_readahead_hook(root, eb, eb->start, -EIO);
671         return -EIO;    /* we fixed nothing */
672 }
673 
674 static void end_workqueue_bio(struct bio *bio, int err)
675 {
676         struct end_io_wq *end_io_wq = bio->bi_private;
677         struct btrfs_fs_info *fs_info;
678 
679         fs_info = end_io_wq->info;
680         end_io_wq->error = err;
681         end_io_wq->work.func = end_workqueue_fn;
682         end_io_wq->work.flags = 0;
683 
684         if (bio->bi_rw & REQ_WRITE) {
685                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
686                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
687                                            &end_io_wq->work);
688                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
689                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
690                                            &end_io_wq->work);
691                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
692                         btrfs_queue_worker(&fs_info->endio_raid56_workers,
693                                            &end_io_wq->work);
694                 else
695                         btrfs_queue_worker(&fs_info->endio_write_workers,
696                                            &end_io_wq->work);
697         } else {
698                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
699                         btrfs_queue_worker(&fs_info->endio_raid56_workers,
700                                            &end_io_wq->work);
701                 else if (end_io_wq->metadata)
702                         btrfs_queue_worker(&fs_info->endio_meta_workers,
703                                            &end_io_wq->work);
704                 else
705                         btrfs_queue_worker(&fs_info->endio_workers,
706                                            &end_io_wq->work);
707         }
708 }
709 
710 /*
711  * For the metadata arg you want
712  *
713  * 0 - if data
714  * 1 - if normal metadta
715  * 2 - if writing to the free space cache area
716  * 3 - raid parity work
717  */
718 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
719                         int metadata)
720 {
721         struct end_io_wq *end_io_wq;
722         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
723         if (!end_io_wq)
724                 return -ENOMEM;
725 
726         end_io_wq->private = bio->bi_private;
727         end_io_wq->end_io = bio->bi_end_io;
728         end_io_wq->info = info;
729         end_io_wq->error = 0;
730         end_io_wq->bio = bio;
731         end_io_wq->metadata = metadata;
732 
733         bio->bi_private = end_io_wq;
734         bio->bi_end_io = end_workqueue_bio;
735         return 0;
736 }
737 
738 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
739 {
740         unsigned long limit = min_t(unsigned long,
741                                     info->workers.max_workers,
742                                     info->fs_devices->open_devices);
743         return 256 * limit;
744 }
745 
746 static void run_one_async_start(struct btrfs_work *work)
747 {
748         struct async_submit_bio *async;
749         int ret;
750 
751         async = container_of(work, struct  async_submit_bio, work);
752         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
753                                       async->mirror_num, async->bio_flags,
754                                       async->bio_offset);
755         if (ret)
756                 async->error = ret;
757 }
758 
759 static void run_one_async_done(struct btrfs_work *work)
760 {
761         struct btrfs_fs_info *fs_info;
762         struct async_submit_bio *async;
763         int limit;
764 
765         async = container_of(work, struct  async_submit_bio, work);
766         fs_info = BTRFS_I(async->inode)->root->fs_info;
767 
768         limit = btrfs_async_submit_limit(fs_info);
769         limit = limit * 2 / 3;
770 
771         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
772             waitqueue_active(&fs_info->async_submit_wait))
773                 wake_up(&fs_info->async_submit_wait);
774 
775         /* If an error occured we just want to clean up the bio and move on */
776         if (async->error) {
777                 bio_endio(async->bio, async->error);
778                 return;
779         }
780 
781         async->submit_bio_done(async->inode, async->rw, async->bio,
782                                async->mirror_num, async->bio_flags,
783                                async->bio_offset);
784 }
785 
786 static void run_one_async_free(struct btrfs_work *work)
787 {
788         struct async_submit_bio *async;
789 
790         async = container_of(work, struct  async_submit_bio, work);
791         kfree(async);
792 }
793 
794 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
795                         int rw, struct bio *bio, int mirror_num,
796                         unsigned long bio_flags,
797                         u64 bio_offset,
798                         extent_submit_bio_hook_t *submit_bio_start,
799                         extent_submit_bio_hook_t *submit_bio_done)
800 {
801         struct async_submit_bio *async;
802 
803         async = kmalloc(sizeof(*async), GFP_NOFS);
804         if (!async)
805                 return -ENOMEM;
806 
807         async->inode = inode;
808         async->rw = rw;
809         async->bio = bio;
810         async->mirror_num = mirror_num;
811         async->submit_bio_start = submit_bio_start;
812         async->submit_bio_done = submit_bio_done;
813 
814         async->work.func = run_one_async_start;
815         async->work.ordered_func = run_one_async_done;
816         async->work.ordered_free = run_one_async_free;
817 
818         async->work.flags = 0;
819         async->bio_flags = bio_flags;
820         async->bio_offset = bio_offset;
821 
822         async->error = 0;
823 
824         atomic_inc(&fs_info->nr_async_submits);
825 
826         if (rw & REQ_SYNC)
827                 btrfs_set_work_high_prio(&async->work);
828 
829         btrfs_queue_worker(&fs_info->workers, &async->work);
830 
831         while (atomic_read(&fs_info->async_submit_draining) &&
832               atomic_read(&fs_info->nr_async_submits)) {
833                 wait_event(fs_info->async_submit_wait,
834                            (atomic_read(&fs_info->nr_async_submits) == 0));
835         }
836 
837         return 0;
838 }
839 
840 static int btree_csum_one_bio(struct bio *bio)
841 {
842         struct bio_vec *bvec;
843         struct btrfs_root *root;
844         int i, ret = 0;
845 
846         bio_for_each_segment_all(bvec, bio, i) {
847                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
848                 ret = csum_dirty_buffer(root, bvec->bv_page);
849                 if (ret)
850                         break;
851         }
852 
853         return ret;
854 }
855 
856 static int __btree_submit_bio_start(struct inode *inode, int rw,
857                                     struct bio *bio, int mirror_num,
858                                     unsigned long bio_flags,
859                                     u64 bio_offset)
860 {
861         /*
862          * when we're called for a write, we're already in the async
863          * submission context.  Just jump into btrfs_map_bio
864          */
865         return btree_csum_one_bio(bio);
866 }
867 
868 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
869                                  int mirror_num, unsigned long bio_flags,
870                                  u64 bio_offset)
871 {
872         int ret;
873 
874         /*
875          * when we're called for a write, we're already in the async
876          * submission context.  Just jump into btrfs_map_bio
877          */
878         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
879         if (ret)
880                 bio_endio(bio, ret);
881         return ret;
882 }
883 
884 static int check_async_write(struct inode *inode, unsigned long bio_flags)
885 {
886         if (bio_flags & EXTENT_BIO_TREE_LOG)
887                 return 0;
888 #ifdef CONFIG_X86
889         if (cpu_has_xmm4_2)
890                 return 0;
891 #endif
892         return 1;
893 }
894 
895 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
896                                  int mirror_num, unsigned long bio_flags,
897                                  u64 bio_offset)
898 {
899         int async = check_async_write(inode, bio_flags);
900         int ret;
901 
902         if (!(rw & REQ_WRITE)) {
903                 /*
904                  * called for a read, do the setup so that checksum validation
905                  * can happen in the async kernel threads
906                  */
907                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
908                                           bio, 1);
909                 if (ret)
910                         goto out_w_error;
911                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
912                                     mirror_num, 0);
913         } else if (!async) {
914                 ret = btree_csum_one_bio(bio);
915                 if (ret)
916                         goto out_w_error;
917                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
918                                     mirror_num, 0);
919         } else {
920                 /*
921                  * kthread helpers are used to submit writes so that
922                  * checksumming can happen in parallel across all CPUs
923                  */
924                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
925                                           inode, rw, bio, mirror_num, 0,
926                                           bio_offset,
927                                           __btree_submit_bio_start,
928                                           __btree_submit_bio_done);
929         }
930 
931         if (ret) {
932 out_w_error:
933                 bio_endio(bio, ret);
934         }
935         return ret;
936 }
937 
938 #ifdef CONFIG_MIGRATION
939 static int btree_migratepage(struct address_space *mapping,
940                         struct page *newpage, struct page *page,
941                         enum migrate_mode mode)
942 {
943         /*
944          * we can't safely write a btree page from here,
945          * we haven't done the locking hook
946          */
947         if (PageDirty(page))
948                 return -EAGAIN;
949         /*
950          * Buffers may be managed in a filesystem specific way.
951          * We must have no buffers or drop them.
952          */
953         if (page_has_private(page) &&
954             !try_to_release_page(page, GFP_KERNEL))
955                 return -EAGAIN;
956         return migrate_page(mapping, newpage, page, mode);
957 }
958 #endif
959 
960 
961 static int btree_writepages(struct address_space *mapping,
962                             struct writeback_control *wbc)
963 {
964         struct btrfs_fs_info *fs_info;
965         int ret;
966 
967         if (wbc->sync_mode == WB_SYNC_NONE) {
968 
969                 if (wbc->for_kupdate)
970                         return 0;
971 
972                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
973                 /* this is a bit racy, but that's ok */
974                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
975                                              BTRFS_DIRTY_METADATA_THRESH);
976                 if (ret < 0)
977                         return 0;
978         }
979         return btree_write_cache_pages(mapping, wbc);
980 }
981 
982 static int btree_readpage(struct file *file, struct page *page)
983 {
984         struct extent_io_tree *tree;
985         tree = &BTRFS_I(page->mapping->host)->io_tree;
986         return extent_read_full_page(tree, page, btree_get_extent, 0);
987 }
988 
989 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
990 {
991         if (PageWriteback(page) || PageDirty(page))
992                 return 0;
993 
994         return try_release_extent_buffer(page);
995 }
996 
997 static void btree_invalidatepage(struct page *page, unsigned int offset,
998                                  unsigned int length)
999 {
1000         struct extent_io_tree *tree;
1001         tree = &BTRFS_I(page->mapping->host)->io_tree;
1002         extent_invalidatepage(tree, page, offset);
1003         btree_releasepage(page, GFP_NOFS);
1004         if (PagePrivate(page)) {
1005                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1006                            "page private not zero on page %llu",
1007                            (unsigned long long)page_offset(page));
1008                 ClearPagePrivate(page);
1009                 set_page_private(page, 0);
1010                 page_cache_release(page);
1011         }
1012 }
1013 
1014 static int btree_set_page_dirty(struct page *page)
1015 {
1016 #ifdef DEBUG
1017         struct extent_buffer *eb;
1018 
1019         BUG_ON(!PagePrivate(page));
1020         eb = (struct extent_buffer *)page->private;
1021         BUG_ON(!eb);
1022         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1023         BUG_ON(!atomic_read(&eb->refs));
1024         btrfs_assert_tree_locked(eb);
1025 #endif
1026         return __set_page_dirty_nobuffers(page);
1027 }
1028 
1029 static const struct address_space_operations btree_aops = {
1030         .readpage       = btree_readpage,
1031         .writepages     = btree_writepages,
1032         .releasepage    = btree_releasepage,
1033         .invalidatepage = btree_invalidatepage,
1034 #ifdef CONFIG_MIGRATION
1035         .migratepage    = btree_migratepage,
1036 #endif
1037         .set_page_dirty = btree_set_page_dirty,
1038 };
1039 
1040 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1041                          u64 parent_transid)
1042 {
1043         struct extent_buffer *buf = NULL;
1044         struct inode *btree_inode = root->fs_info->btree_inode;
1045         int ret = 0;
1046 
1047         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1048         if (!buf)
1049                 return 0;
1050         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1051                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1052         free_extent_buffer(buf);
1053         return ret;
1054 }
1055 
1056 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1057                          int mirror_num, struct extent_buffer **eb)
1058 {
1059         struct extent_buffer *buf = NULL;
1060         struct inode *btree_inode = root->fs_info->btree_inode;
1061         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1062         int ret;
1063 
1064         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1065         if (!buf)
1066                 return 0;
1067 
1068         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1069 
1070         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1071                                        btree_get_extent, mirror_num);
1072         if (ret) {
1073                 free_extent_buffer(buf);
1074                 return ret;
1075         }
1076 
1077         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1078                 free_extent_buffer(buf);
1079                 return -EIO;
1080         } else if (extent_buffer_uptodate(buf)) {
1081                 *eb = buf;
1082         } else {
1083                 free_extent_buffer(buf);
1084         }
1085         return 0;
1086 }
1087 
1088 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1089                                             u64 bytenr, u32 blocksize)
1090 {
1091         return find_extent_buffer(root->fs_info, bytenr);
1092 }
1093 
1094 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1095                                                  u64 bytenr, u32 blocksize)
1096 {
1097         return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1098 }
1099 
1100 
1101 int btrfs_write_tree_block(struct extent_buffer *buf)
1102 {
1103         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1104                                         buf->start + buf->len - 1);
1105 }
1106 
1107 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1108 {
1109         return filemap_fdatawait_range(buf->pages[0]->mapping,
1110                                        buf->start, buf->start + buf->len - 1);
1111 }
1112 
1113 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1114                                       u32 blocksize, u64 parent_transid)
1115 {
1116         struct extent_buffer *buf = NULL;
1117         int ret;
1118 
1119         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1120         if (!buf)
1121                 return NULL;
1122 
1123         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1124         if (ret) {
1125                 free_extent_buffer(buf);
1126                 return NULL;
1127         }
1128         return buf;
1129 
1130 }
1131 
1132 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1133                       struct extent_buffer *buf)
1134 {
1135         struct btrfs_fs_info *fs_info = root->fs_info;
1136 
1137         if (btrfs_header_generation(buf) ==
1138             fs_info->running_transaction->transid) {
1139                 btrfs_assert_tree_locked(buf);
1140 
1141                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1142                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1143                                              -buf->len,
1144                                              fs_info->dirty_metadata_batch);
1145                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1146                         btrfs_set_lock_blocking(buf);
1147                         clear_extent_buffer_dirty(buf);
1148                 }
1149         }
1150 }
1151 
1152 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1153                          u32 stripesize, struct btrfs_root *root,
1154                          struct btrfs_fs_info *fs_info,
1155                          u64 objectid)
1156 {
1157         root->node = NULL;
1158         root->commit_root = NULL;
1159         root->sectorsize = sectorsize;
1160         root->nodesize = nodesize;
1161         root->leafsize = leafsize;
1162         root->stripesize = stripesize;
1163         root->ref_cows = 0;
1164         root->track_dirty = 0;
1165         root->in_radix = 0;
1166         root->orphan_item_inserted = 0;
1167         root->orphan_cleanup_state = 0;
1168 
1169         root->objectid = objectid;
1170         root->last_trans = 0;
1171         root->highest_objectid = 0;
1172         root->nr_delalloc_inodes = 0;
1173         root->nr_ordered_extents = 0;
1174         root->name = NULL;
1175         root->inode_tree = RB_ROOT;
1176         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1177         root->block_rsv = NULL;
1178         root->orphan_block_rsv = NULL;
1179 
1180         INIT_LIST_HEAD(&root->dirty_list);
1181         INIT_LIST_HEAD(&root->root_list);
1182         INIT_LIST_HEAD(&root->delalloc_inodes);
1183         INIT_LIST_HEAD(&root->delalloc_root);
1184         INIT_LIST_HEAD(&root->ordered_extents);
1185         INIT_LIST_HEAD(&root->ordered_root);
1186         INIT_LIST_HEAD(&root->logged_list[0]);
1187         INIT_LIST_HEAD(&root->logged_list[1]);
1188         spin_lock_init(&root->orphan_lock);
1189         spin_lock_init(&root->inode_lock);
1190         spin_lock_init(&root->delalloc_lock);
1191         spin_lock_init(&root->ordered_extent_lock);
1192         spin_lock_init(&root->accounting_lock);
1193         spin_lock_init(&root->log_extents_lock[0]);
1194         spin_lock_init(&root->log_extents_lock[1]);
1195         mutex_init(&root->objectid_mutex);
1196         mutex_init(&root->log_mutex);
1197         init_waitqueue_head(&root->log_writer_wait);
1198         init_waitqueue_head(&root->log_commit_wait[0]);
1199         init_waitqueue_head(&root->log_commit_wait[1]);
1200         atomic_set(&root->log_commit[0], 0);
1201         atomic_set(&root->log_commit[1], 0);
1202         atomic_set(&root->log_writers, 0);
1203         atomic_set(&root->log_batch, 0);
1204         atomic_set(&root->orphan_inodes, 0);
1205         atomic_set(&root->refs, 1);
1206         root->log_transid = 0;
1207         root->last_log_commit = 0;
1208         if (fs_info)
1209                 extent_io_tree_init(&root->dirty_log_pages,
1210                                      fs_info->btree_inode->i_mapping);
1211 
1212         memset(&root->root_key, 0, sizeof(root->root_key));
1213         memset(&root->root_item, 0, sizeof(root->root_item));
1214         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1215         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1216         if (fs_info)
1217                 root->defrag_trans_start = fs_info->generation;
1218         else
1219                 root->defrag_trans_start = 0;
1220         init_completion(&root->kobj_unregister);
1221         root->defrag_running = 0;
1222         root->root_key.objectid = objectid;
1223         root->anon_dev = 0;
1224 
1225         spin_lock_init(&root->root_item_lock);
1226 }
1227 
1228 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1229 {
1230         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1231         if (root)
1232                 root->fs_info = fs_info;
1233         return root;
1234 }
1235 
1236 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1237 /* Should only be used by the testing infrastructure */
1238 struct btrfs_root *btrfs_alloc_dummy_root(void)
1239 {
1240         struct btrfs_root *root;
1241 
1242         root = btrfs_alloc_root(NULL);
1243         if (!root)
1244                 return ERR_PTR(-ENOMEM);
1245         __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1246         root->dummy_root = 1;
1247 
1248         return root;
1249 }
1250 #endif
1251 
1252 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1253                                      struct btrfs_fs_info *fs_info,
1254                                      u64 objectid)
1255 {
1256         struct extent_buffer *leaf;
1257         struct btrfs_root *tree_root = fs_info->tree_root;
1258         struct btrfs_root *root;
1259         struct btrfs_key key;
1260         int ret = 0;
1261         uuid_le uuid;
1262 
1263         root = btrfs_alloc_root(fs_info);
1264         if (!root)
1265                 return ERR_PTR(-ENOMEM);
1266 
1267         __setup_root(tree_root->nodesize, tree_root->leafsize,
1268                      tree_root->sectorsize, tree_root->stripesize,
1269                      root, fs_info, objectid);
1270         root->root_key.objectid = objectid;
1271         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1272         root->root_key.offset = 0;
1273 
1274         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1275                                       0, objectid, NULL, 0, 0, 0);
1276         if (IS_ERR(leaf)) {
1277                 ret = PTR_ERR(leaf);
1278                 leaf = NULL;
1279                 goto fail;
1280         }
1281 
1282         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1283         btrfs_set_header_bytenr(leaf, leaf->start);
1284         btrfs_set_header_generation(leaf, trans->transid);
1285         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1286         btrfs_set_header_owner(leaf, objectid);
1287         root->node = leaf;
1288 
1289         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1290                             BTRFS_FSID_SIZE);
1291         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1292                             btrfs_header_chunk_tree_uuid(leaf),
1293                             BTRFS_UUID_SIZE);
1294         btrfs_mark_buffer_dirty(leaf);
1295 
1296         root->commit_root = btrfs_root_node(root);
1297         root->track_dirty = 1;
1298 
1299 
1300         root->root_item.flags = 0;
1301         root->root_item.byte_limit = 0;
1302         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1303         btrfs_set_root_generation(&root->root_item, trans->transid);
1304         btrfs_set_root_level(&root->root_item, 0);
1305         btrfs_set_root_refs(&root->root_item, 1);
1306         btrfs_set_root_used(&root->root_item, leaf->len);
1307         btrfs_set_root_last_snapshot(&root->root_item, 0);
1308         btrfs_set_root_dirid(&root->root_item, 0);
1309         uuid_le_gen(&uuid);
1310         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1311         root->root_item.drop_level = 0;
1312 
1313         key.objectid = objectid;
1314         key.type = BTRFS_ROOT_ITEM_KEY;
1315         key.offset = 0;
1316         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1317         if (ret)
1318                 goto fail;
1319 
1320         btrfs_tree_unlock(leaf);
1321 
1322         return root;
1323 
1324 fail:
1325         if (leaf) {
1326                 btrfs_tree_unlock(leaf);
1327                 free_extent_buffer(leaf);
1328         }
1329         kfree(root);
1330 
1331         return ERR_PTR(ret);
1332 }
1333 
1334 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1335                                          struct btrfs_fs_info *fs_info)
1336 {
1337         struct btrfs_root *root;
1338         struct btrfs_root *tree_root = fs_info->tree_root;
1339         struct extent_buffer *leaf;
1340 
1341         root = btrfs_alloc_root(fs_info);
1342         if (!root)
1343                 return ERR_PTR(-ENOMEM);
1344 
1345         __setup_root(tree_root->nodesize, tree_root->leafsize,
1346                      tree_root->sectorsize, tree_root->stripesize,
1347                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1348 
1349         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1350         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1351         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1352         /*
1353          * log trees do not get reference counted because they go away
1354          * before a real commit is actually done.  They do store pointers
1355          * to file data extents, and those reference counts still get
1356          * updated (along with back refs to the log tree).
1357          */
1358         root->ref_cows = 0;
1359 
1360         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1361                                       BTRFS_TREE_LOG_OBJECTID, NULL,
1362                                       0, 0, 0);
1363         if (IS_ERR(leaf)) {
1364                 kfree(root);
1365                 return ERR_CAST(leaf);
1366         }
1367 
1368         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1369         btrfs_set_header_bytenr(leaf, leaf->start);
1370         btrfs_set_header_generation(leaf, trans->transid);
1371         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1372         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1373         root->node = leaf;
1374 
1375         write_extent_buffer(root->node, root->fs_info->fsid,
1376                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1377         btrfs_mark_buffer_dirty(root->node);
1378         btrfs_tree_unlock(root->node);
1379         return root;
1380 }
1381 
1382 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1383                              struct btrfs_fs_info *fs_info)
1384 {
1385         struct btrfs_root *log_root;
1386 
1387         log_root = alloc_log_tree(trans, fs_info);
1388         if (IS_ERR(log_root))
1389                 return PTR_ERR(log_root);
1390         WARN_ON(fs_info->log_root_tree);
1391         fs_info->log_root_tree = log_root;
1392         return 0;
1393 }
1394 
1395 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1396                        struct btrfs_root *root)
1397 {
1398         struct btrfs_root *log_root;
1399         struct btrfs_inode_item *inode_item;
1400 
1401         log_root = alloc_log_tree(trans, root->fs_info);
1402         if (IS_ERR(log_root))
1403                 return PTR_ERR(log_root);
1404 
1405         log_root->last_trans = trans->transid;
1406         log_root->root_key.offset = root->root_key.objectid;
1407 
1408         inode_item = &log_root->root_item.inode;
1409         btrfs_set_stack_inode_generation(inode_item, 1);
1410         btrfs_set_stack_inode_size(inode_item, 3);
1411         btrfs_set_stack_inode_nlink(inode_item, 1);
1412         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1413         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1414 
1415         btrfs_set_root_node(&log_root->root_item, log_root->node);
1416 
1417         WARN_ON(root->log_root);
1418         root->log_root = log_root;
1419         root->log_transid = 0;
1420         root->last_log_commit = 0;
1421         return 0;
1422 }
1423 
1424 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1425                                                struct btrfs_key *key)
1426 {
1427         struct btrfs_root *root;
1428         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1429         struct btrfs_path *path;
1430         u64 generation;
1431         u32 blocksize;
1432         int ret;
1433 
1434         path = btrfs_alloc_path();
1435         if (!path)
1436                 return ERR_PTR(-ENOMEM);
1437 
1438         root = btrfs_alloc_root(fs_info);
1439         if (!root) {
1440                 ret = -ENOMEM;
1441                 goto alloc_fail;
1442         }
1443 
1444         __setup_root(tree_root->nodesize, tree_root->leafsize,
1445                      tree_root->sectorsize, tree_root->stripesize,
1446                      root, fs_info, key->objectid);
1447 
1448         ret = btrfs_find_root(tree_root, key, path,
1449                               &root->root_item, &root->root_key);
1450         if (ret) {
1451                 if (ret > 0)
1452                         ret = -ENOENT;
1453                 goto find_fail;
1454         }
1455 
1456         generation = btrfs_root_generation(&root->root_item);
1457         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1458         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1459                                      blocksize, generation);
1460         if (!root->node) {
1461                 ret = -ENOMEM;
1462                 goto find_fail;
1463         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1464                 ret = -EIO;
1465                 goto read_fail;
1466         }
1467         root->commit_root = btrfs_root_node(root);
1468 out:
1469         btrfs_free_path(path);
1470         return root;
1471 
1472 read_fail:
1473         free_extent_buffer(root->node);
1474 find_fail:
1475         kfree(root);
1476 alloc_fail:
1477         root = ERR_PTR(ret);
1478         goto out;
1479 }
1480 
1481 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1482                                       struct btrfs_key *location)
1483 {
1484         struct btrfs_root *root;
1485 
1486         root = btrfs_read_tree_root(tree_root, location);
1487         if (IS_ERR(root))
1488                 return root;
1489 
1490         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1491                 root->ref_cows = 1;
1492                 btrfs_check_and_init_root_item(&root->root_item);
1493         }
1494 
1495         return root;
1496 }
1497 
1498 int btrfs_init_fs_root(struct btrfs_root *root)
1499 {
1500         int ret;
1501 
1502         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1503         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1504                                         GFP_NOFS);
1505         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1506                 ret = -ENOMEM;
1507                 goto fail;
1508         }
1509 
1510         btrfs_init_free_ino_ctl(root);
1511         mutex_init(&root->fs_commit_mutex);
1512         spin_lock_init(&root->cache_lock);
1513         init_waitqueue_head(&root->cache_wait);
1514 
1515         ret = get_anon_bdev(&root->anon_dev);
1516         if (ret)
1517                 goto fail;
1518         return 0;
1519 fail:
1520         kfree(root->free_ino_ctl);
1521         kfree(root->free_ino_pinned);
1522         return ret;
1523 }
1524 
1525 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1526                                                u64 root_id)
1527 {
1528         struct btrfs_root *root;
1529 
1530         spin_lock(&fs_info->fs_roots_radix_lock);
1531         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1532                                  (unsigned long)root_id);
1533         spin_unlock(&fs_info->fs_roots_radix_lock);
1534         return root;
1535 }
1536 
1537 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1538                          struct btrfs_root *root)
1539 {
1540         int ret;
1541 
1542         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1543         if (ret)
1544                 return ret;
1545 
1546         spin_lock(&fs_info->fs_roots_radix_lock);
1547         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1548                                 (unsigned long)root->root_key.objectid,
1549                                 root);
1550         if (ret == 0)
1551                 root->in_radix = 1;
1552         spin_unlock(&fs_info->fs_roots_radix_lock);
1553         radix_tree_preload_end();
1554 
1555         return ret;
1556 }
1557 
1558 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1559                                      struct btrfs_key *location,
1560                                      bool check_ref)
1561 {
1562         struct btrfs_root *root;
1563         struct btrfs_path *path;
1564         int ret;
1565 
1566         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1567                 return fs_info->tree_root;
1568         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1569                 return fs_info->extent_root;
1570         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1571                 return fs_info->chunk_root;
1572         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1573                 return fs_info->dev_root;
1574         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1575                 return fs_info->csum_root;
1576         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1577                 return fs_info->quota_root ? fs_info->quota_root :
1578                                              ERR_PTR(-ENOENT);
1579         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1580                 return fs_info->uuid_root ? fs_info->uuid_root :
1581                                             ERR_PTR(-ENOENT);
1582 again:
1583         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1584         if (root) {
1585                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1586                         return ERR_PTR(-ENOENT);
1587                 return root;
1588         }
1589 
1590         root = btrfs_read_fs_root(fs_info->tree_root, location);
1591         if (IS_ERR(root))
1592                 return root;
1593 
1594         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1595                 ret = -ENOENT;
1596                 goto fail;
1597         }
1598 
1599         ret = btrfs_init_fs_root(root);
1600         if (ret)
1601                 goto fail;
1602 
1603         path = btrfs_alloc_path();
1604         if (!path) {
1605                 ret = -ENOMEM;
1606                 goto fail;
1607         }
1608         ret = btrfs_find_item(fs_info->tree_root, path, BTRFS_ORPHAN_OBJECTID,
1609                         location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1610         btrfs_free_path(path);
1611         if (ret < 0)
1612                 goto fail;
1613         if (ret == 0)
1614                 root->orphan_item_inserted = 1;
1615 
1616         ret = btrfs_insert_fs_root(fs_info, root);
1617         if (ret) {
1618                 if (ret == -EEXIST) {
1619                         free_fs_root(root);
1620                         goto again;
1621                 }
1622                 goto fail;
1623         }
1624         return root;
1625 fail:
1626         free_fs_root(root);
1627         return ERR_PTR(ret);
1628 }
1629 
1630 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1631 {
1632         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1633         int ret = 0;
1634         struct btrfs_device *device;
1635         struct backing_dev_info *bdi;
1636 
1637         rcu_read_lock();
1638         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1639                 if (!device->bdev)
1640                         continue;
1641                 bdi = blk_get_backing_dev_info(device->bdev);
1642                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1643                         ret = 1;
1644                         break;
1645                 }
1646         }
1647         rcu_read_unlock();
1648         return ret;
1649 }
1650 
1651 /*
1652  * If this fails, caller must call bdi_destroy() to get rid of the
1653  * bdi again.
1654  */
1655 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1656 {
1657         int err;
1658 
1659         bdi->capabilities = BDI_CAP_MAP_COPY;
1660         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1661         if (err)
1662                 return err;
1663 
1664         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1665         bdi->congested_fn       = btrfs_congested_fn;
1666         bdi->congested_data     = info;
1667         return 0;
1668 }
1669 
1670 /*
1671  * called by the kthread helper functions to finally call the bio end_io
1672  * functions.  This is where read checksum verification actually happens
1673  */
1674 static void end_workqueue_fn(struct btrfs_work *work)
1675 {
1676         struct bio *bio;
1677         struct end_io_wq *end_io_wq;
1678         int error;
1679 
1680         end_io_wq = container_of(work, struct end_io_wq, work);
1681         bio = end_io_wq->bio;
1682 
1683         error = end_io_wq->error;
1684         bio->bi_private = end_io_wq->private;
1685         bio->bi_end_io = end_io_wq->end_io;
1686         kfree(end_io_wq);
1687         bio_endio_nodec(bio, error);
1688 }
1689 
1690 static int cleaner_kthread(void *arg)
1691 {
1692         struct btrfs_root *root = arg;
1693         int again;
1694 
1695         do {
1696                 again = 0;
1697 
1698                 /* Make the cleaner go to sleep early. */
1699                 if (btrfs_need_cleaner_sleep(root))
1700                         goto sleep;
1701 
1702                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1703                         goto sleep;
1704 
1705                 /*
1706                  * Avoid the problem that we change the status of the fs
1707                  * during the above check and trylock.
1708                  */
1709                 if (btrfs_need_cleaner_sleep(root)) {
1710                         mutex_unlock(&root->fs_info->cleaner_mutex);
1711                         goto sleep;
1712                 }
1713 
1714                 btrfs_run_delayed_iputs(root);
1715                 again = btrfs_clean_one_deleted_snapshot(root);
1716                 mutex_unlock(&root->fs_info->cleaner_mutex);
1717 
1718                 /*
1719                  * The defragger has dealt with the R/O remount and umount,
1720                  * needn't do anything special here.
1721                  */
1722                 btrfs_run_defrag_inodes(root->fs_info);
1723 sleep:
1724                 if (!try_to_freeze() && !again) {
1725                         set_current_state(TASK_INTERRUPTIBLE);
1726                         if (!kthread_should_stop())
1727                                 schedule();
1728                         __set_current_state(TASK_RUNNING);
1729                 }
1730         } while (!kthread_should_stop());
1731         return 0;
1732 }
1733 
1734 static int transaction_kthread(void *arg)
1735 {
1736         struct btrfs_root *root = arg;
1737         struct btrfs_trans_handle *trans;
1738         struct btrfs_transaction *cur;
1739         u64 transid;
1740         unsigned long now;
1741         unsigned long delay;
1742         bool cannot_commit;
1743 
1744         do {
1745                 cannot_commit = false;
1746                 delay = HZ * root->fs_info->commit_interval;
1747                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1748 
1749                 spin_lock(&root->fs_info->trans_lock);
1750                 cur = root->fs_info->running_transaction;
1751                 if (!cur) {
1752                         spin_unlock(&root->fs_info->trans_lock);
1753                         goto sleep;
1754                 }
1755 
1756                 now = get_seconds();
1757                 if (cur->state < TRANS_STATE_BLOCKED &&
1758                     (now < cur->start_time ||
1759                      now - cur->start_time < root->fs_info->commit_interval)) {
1760                         spin_unlock(&root->fs_info->trans_lock);
1761                         delay = HZ * 5;
1762                         goto sleep;
1763                 }
1764                 transid = cur->transid;
1765                 spin_unlock(&root->fs_info->trans_lock);
1766 
1767                 /* If the file system is aborted, this will always fail. */
1768                 trans = btrfs_attach_transaction(root);
1769                 if (IS_ERR(trans)) {
1770                         if (PTR_ERR(trans) != -ENOENT)
1771                                 cannot_commit = true;
1772                         goto sleep;
1773                 }
1774                 if (transid == trans->transid) {
1775                         btrfs_commit_transaction(trans, root);
1776                 } else {
1777                         btrfs_end_transaction(trans, root);
1778                 }
1779 sleep:
1780                 wake_up_process(root->fs_info->cleaner_kthread);
1781                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1782 
1783                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1784                                       &root->fs_info->fs_state)))
1785                         btrfs_cleanup_transaction(root);
1786                 if (!try_to_freeze()) {
1787                         set_current_state(TASK_INTERRUPTIBLE);
1788                         if (!kthread_should_stop() &&
1789                             (!btrfs_transaction_blocked(root->fs_info) ||
1790                              cannot_commit))
1791                                 schedule_timeout(delay);
1792                         __set_current_state(TASK_RUNNING);
1793                 }
1794         } while (!kthread_should_stop());
1795         return 0;
1796 }
1797 
1798 /*
1799  * this will find the highest generation in the array of
1800  * root backups.  The index of the highest array is returned,
1801  * or -1 if we can't find anything.
1802  *
1803  * We check to make sure the array is valid by comparing the
1804  * generation of the latest  root in the array with the generation
1805  * in the super block.  If they don't match we pitch it.
1806  */
1807 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1808 {
1809         u64 cur;
1810         int newest_index = -1;
1811         struct btrfs_root_backup *root_backup;
1812         int i;
1813 
1814         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1815                 root_backup = info->super_copy->super_roots + i;
1816                 cur = btrfs_backup_tree_root_gen(root_backup);
1817                 if (cur == newest_gen)
1818                         newest_index = i;
1819         }
1820 
1821         /* check to see if we actually wrapped around */
1822         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1823                 root_backup = info->super_copy->super_roots;
1824                 cur = btrfs_backup_tree_root_gen(root_backup);
1825                 if (cur == newest_gen)
1826                         newest_index = 0;
1827         }
1828         return newest_index;
1829 }
1830 
1831 
1832 /*
1833  * find the oldest backup so we know where to store new entries
1834  * in the backup array.  This will set the backup_root_index
1835  * field in the fs_info struct
1836  */
1837 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1838                                      u64 newest_gen)
1839 {
1840         int newest_index = -1;
1841 
1842         newest_index = find_newest_super_backup(info, newest_gen);
1843         /* if there was garbage in there, just move along */
1844         if (newest_index == -1) {
1845                 info->backup_root_index = 0;
1846         } else {
1847                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1848         }
1849 }
1850 
1851 /*
1852  * copy all the root pointers into the super backup array.
1853  * this will bump the backup pointer by one when it is
1854  * done
1855  */
1856 static void backup_super_roots(struct btrfs_fs_info *info)
1857 {
1858         int next_backup;
1859         struct btrfs_root_backup *root_backup;
1860         int last_backup;
1861 
1862         next_backup = info->backup_root_index;
1863         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1864                 BTRFS_NUM_BACKUP_ROOTS;
1865 
1866         /*
1867          * just overwrite the last backup if we're at the same generation
1868          * this happens only at umount
1869          */
1870         root_backup = info->super_for_commit->super_roots + last_backup;
1871         if (btrfs_backup_tree_root_gen(root_backup) ==
1872             btrfs_header_generation(info->tree_root->node))
1873                 next_backup = last_backup;
1874 
1875         root_backup = info->super_for_commit->super_roots + next_backup;
1876 
1877         /*
1878          * make sure all of our padding and empty slots get zero filled
1879          * regardless of which ones we use today
1880          */
1881         memset(root_backup, 0, sizeof(*root_backup));
1882 
1883         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1884 
1885         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1886         btrfs_set_backup_tree_root_gen(root_backup,
1887                                btrfs_header_generation(info->tree_root->node));
1888 
1889         btrfs_set_backup_tree_root_level(root_backup,
1890                                btrfs_header_level(info->tree_root->node));
1891 
1892         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1893         btrfs_set_backup_chunk_root_gen(root_backup,
1894                                btrfs_header_generation(info->chunk_root->node));
1895         btrfs_set_backup_chunk_root_level(root_backup,
1896                                btrfs_header_level(info->chunk_root->node));
1897 
1898         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1899         btrfs_set_backup_extent_root_gen(root_backup,
1900                                btrfs_header_generation(info->extent_root->node));
1901         btrfs_set_backup_extent_root_level(root_backup,
1902                                btrfs_header_level(info->extent_root->node));
1903 
1904         /*
1905          * we might commit during log recovery, which happens before we set
1906          * the fs_root.  Make sure it is valid before we fill it in.
1907          */
1908         if (info->fs_root && info->fs_root->node) {
1909                 btrfs_set_backup_fs_root(root_backup,
1910                                          info->fs_root->node->start);
1911                 btrfs_set_backup_fs_root_gen(root_backup,
1912                                btrfs_header_generation(info->fs_root->node));
1913                 btrfs_set_backup_fs_root_level(root_backup,
1914                                btrfs_header_level(info->fs_root->node));
1915         }
1916 
1917         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1918         btrfs_set_backup_dev_root_gen(root_backup,
1919                                btrfs_header_generation(info->dev_root->node));
1920         btrfs_set_backup_dev_root_level(root_backup,
1921                                        btrfs_header_level(info->dev_root->node));
1922 
1923         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1924         btrfs_set_backup_csum_root_gen(root_backup,
1925                                btrfs_header_generation(info->csum_root->node));
1926         btrfs_set_backup_csum_root_level(root_backup,
1927                                btrfs_header_level(info->csum_root->node));
1928 
1929         btrfs_set_backup_total_bytes(root_backup,
1930                              btrfs_super_total_bytes(info->super_copy));
1931         btrfs_set_backup_bytes_used(root_backup,
1932                              btrfs_super_bytes_used(info->super_copy));
1933         btrfs_set_backup_num_devices(root_backup,
1934                              btrfs_super_num_devices(info->super_copy));
1935 
1936         /*
1937          * if we don't copy this out to the super_copy, it won't get remembered
1938          * for the next commit
1939          */
1940         memcpy(&info->super_copy->super_roots,
1941                &info->super_for_commit->super_roots,
1942                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1943 }
1944 
1945 /*
1946  * this copies info out of the root backup array and back into
1947  * the in-memory super block.  It is meant to help iterate through
1948  * the array, so you send it the number of backups you've already
1949  * tried and the last backup index you used.
1950  *
1951  * this returns -1 when it has tried all the backups
1952  */
1953 static noinline int next_root_backup(struct btrfs_fs_info *info,
1954                                      struct btrfs_super_block *super,
1955                                      int *num_backups_tried, int *backup_index)
1956 {
1957         struct btrfs_root_backup *root_backup;
1958         int newest = *backup_index;
1959 
1960         if (*num_backups_tried == 0) {
1961                 u64 gen = btrfs_super_generation(super);
1962 
1963                 newest = find_newest_super_backup(info, gen);
1964                 if (newest == -1)
1965                         return -1;
1966 
1967                 *backup_index = newest;
1968                 *num_backups_tried = 1;
1969         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1970                 /* we've tried all the backups, all done */
1971                 return -1;
1972         } else {
1973                 /* jump to the next oldest backup */
1974                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1975                         BTRFS_NUM_BACKUP_ROOTS;
1976                 *backup_index = newest;
1977                 *num_backups_tried += 1;
1978         }
1979         root_backup = super->super_roots + newest;
1980 
1981         btrfs_set_super_generation(super,
1982                                    btrfs_backup_tree_root_gen(root_backup));
1983         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1984         btrfs_set_super_root_level(super,
1985                                    btrfs_backup_tree_root_level(root_backup));
1986         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1987 
1988         /*
1989          * fixme: the total bytes and num_devices need to match or we should
1990          * need a fsck
1991          */
1992         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1993         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1994         return 0;
1995 }
1996 
1997 /* helper to cleanup workers */
1998 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1999 {
2000         btrfs_stop_workers(&fs_info->generic_worker);
2001         btrfs_stop_workers(&fs_info->fixup_workers);
2002         btrfs_stop_workers(&fs_info->delalloc_workers);
2003         btrfs_stop_workers(&fs_info->workers);
2004         btrfs_stop_workers(&fs_info->endio_workers);
2005         btrfs_stop_workers(&fs_info->endio_meta_workers);
2006         btrfs_stop_workers(&fs_info->endio_raid56_workers);
2007         btrfs_stop_workers(&fs_info->rmw_workers);
2008         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2009         btrfs_stop_workers(&fs_info->endio_write_workers);
2010         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2011         btrfs_stop_workers(&fs_info->submit_workers);
2012         btrfs_stop_workers(&fs_info->delayed_workers);
2013         btrfs_stop_workers(&fs_info->caching_workers);
2014         btrfs_stop_workers(&fs_info->readahead_workers);
2015         btrfs_stop_workers(&fs_info->flush_workers);
2016         btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2017 }
2018 
2019 static void free_root_extent_buffers(struct btrfs_root *root)
2020 {
2021         if (root) {
2022                 free_extent_buffer(root->node);
2023                 free_extent_buffer(root->commit_root);
2024                 root->node = NULL;
2025                 root->commit_root = NULL;
2026         }
2027 }
2028 
2029 /* helper to cleanup tree roots */
2030 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2031 {
2032         free_root_extent_buffers(info->tree_root);
2033 
2034         free_root_extent_buffers(info->dev_root);
2035         free_root_extent_buffers(info->extent_root);
2036         free_root_extent_buffers(info->csum_root);
2037         free_root_extent_buffers(info->quota_root);
2038         free_root_extent_buffers(info->uuid_root);
2039         if (chunk_root)
2040                 free_root_extent_buffers(info->chunk_root);
2041 }
2042 
2043 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2044 {
2045         int ret;
2046         struct btrfs_root *gang[8];
2047         int i;
2048 
2049         while (!list_empty(&fs_info->dead_roots)) {
2050                 gang[0] = list_entry(fs_info->dead_roots.next,
2051                                      struct btrfs_root, root_list);
2052                 list_del(&gang[0]->root_list);
2053 
2054                 if (gang[0]->in_radix) {
2055                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2056                 } else {
2057                         free_extent_buffer(gang[0]->node);
2058                         free_extent_buffer(gang[0]->commit_root);
2059                         btrfs_put_fs_root(gang[0]);
2060                 }
2061         }
2062 
2063         while (1) {
2064                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2065                                              (void **)gang, 0,
2066                                              ARRAY_SIZE(gang));
2067                 if (!ret)
2068                         break;
2069                 for (i = 0; i < ret; i++)
2070                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2071         }
2072 
2073         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2074                 btrfs_free_log_root_tree(NULL, fs_info);
2075                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2076                                             fs_info->pinned_extents);
2077         }
2078 }
2079 
2080 int open_ctree(struct super_block *sb,
2081                struct btrfs_fs_devices *fs_devices,
2082                char *options)
2083 {
2084         u32 sectorsize;
2085         u32 nodesize;
2086         u32 leafsize;
2087         u32 blocksize;
2088         u32 stripesize;
2089         u64 generation;
2090         u64 features;
2091         struct btrfs_key location;
2092         struct buffer_head *bh;
2093         struct btrfs_super_block *disk_super;
2094         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2095         struct btrfs_root *tree_root;
2096         struct btrfs_root *extent_root;
2097         struct btrfs_root *csum_root;
2098         struct btrfs_root *chunk_root;
2099         struct btrfs_root *dev_root;
2100         struct btrfs_root *quota_root;
2101         struct btrfs_root *uuid_root;
2102         struct btrfs_root *log_tree_root;
2103         int ret;
2104         int err = -EINVAL;
2105         int num_backups_tried = 0;
2106         int backup_index = 0;
2107         bool create_uuid_tree;
2108         bool check_uuid_tree;
2109 
2110         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2111         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2112         if (!tree_root || !chunk_root) {
2113                 err = -ENOMEM;
2114                 goto fail;
2115         }
2116 
2117         ret = init_srcu_struct(&fs_info->subvol_srcu);
2118         if (ret) {
2119                 err = ret;
2120                 goto fail;
2121         }
2122 
2123         ret = setup_bdi(fs_info, &fs_info->bdi);
2124         if (ret) {
2125                 err = ret;
2126                 goto fail_srcu;
2127         }
2128 
2129         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2130         if (ret) {
2131                 err = ret;
2132                 goto fail_bdi;
2133         }
2134         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2135                                         (1 + ilog2(nr_cpu_ids));
2136 
2137         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2138         if (ret) {
2139                 err = ret;
2140                 goto fail_dirty_metadata_bytes;
2141         }
2142 
2143         fs_info->btree_inode = new_inode(sb);
2144         if (!fs_info->btree_inode) {
2145                 err = -ENOMEM;
2146                 goto fail_delalloc_bytes;
2147         }
2148 
2149         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2150 
2151         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2152         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2153         INIT_LIST_HEAD(&fs_info->trans_list);
2154         INIT_LIST_HEAD(&fs_info->dead_roots);
2155         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2156         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2157         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2158         spin_lock_init(&fs_info->delalloc_root_lock);
2159         spin_lock_init(&fs_info->trans_lock);
2160         spin_lock_init(&fs_info->fs_roots_radix_lock);
2161         spin_lock_init(&fs_info->delayed_iput_lock);
2162         spin_lock_init(&fs_info->defrag_inodes_lock);
2163         spin_lock_init(&fs_info->free_chunk_lock);
2164         spin_lock_init(&fs_info->tree_mod_seq_lock);
2165         spin_lock_init(&fs_info->super_lock);
2166         spin_lock_init(&fs_info->buffer_lock);
2167         rwlock_init(&fs_info->tree_mod_log_lock);
2168         mutex_init(&fs_info->reloc_mutex);
2169         seqlock_init(&fs_info->profiles_lock);
2170 
2171         init_completion(&fs_info->kobj_unregister);
2172         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2173         INIT_LIST_HEAD(&fs_info->space_info);
2174         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2175         btrfs_mapping_init(&fs_info->mapping_tree);
2176         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2177                              BTRFS_BLOCK_RSV_GLOBAL);
2178         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2179                              BTRFS_BLOCK_RSV_DELALLOC);
2180         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2181         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2182         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2183         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2184                              BTRFS_BLOCK_RSV_DELOPS);
2185         atomic_set(&fs_info->nr_async_submits, 0);
2186         atomic_set(&fs_info->async_delalloc_pages, 0);
2187         atomic_set(&fs_info->async_submit_draining, 0);
2188         atomic_set(&fs_info->nr_async_bios, 0);
2189         atomic_set(&fs_info->defrag_running, 0);
2190         atomic64_set(&fs_info->tree_mod_seq, 0);
2191         fs_info->sb = sb;
2192         fs_info->max_inline = 8192 * 1024;
2193         fs_info->metadata_ratio = 0;
2194         fs_info->defrag_inodes = RB_ROOT;
2195         fs_info->free_chunk_space = 0;
2196         fs_info->tree_mod_log = RB_ROOT;
2197         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2198         fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2199         /* readahead state */
2200         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2201         spin_lock_init(&fs_info->reada_lock);
2202 
2203         fs_info->thread_pool_size = min_t(unsigned long,
2204                                           num_online_cpus() + 2, 8);
2205 
2206         INIT_LIST_HEAD(&fs_info->ordered_roots);
2207         spin_lock_init(&fs_info->ordered_root_lock);
2208         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2209                                         GFP_NOFS);
2210         if (!fs_info->delayed_root) {
2211                 err = -ENOMEM;
2212                 goto fail_iput;
2213         }
2214         btrfs_init_delayed_root(fs_info->delayed_root);
2215 
2216         mutex_init(&fs_info->scrub_lock);
2217         atomic_set(&fs_info->scrubs_running, 0);
2218         atomic_set(&fs_info->scrub_pause_req, 0);
2219         atomic_set(&fs_info->scrubs_paused, 0);
2220         atomic_set(&fs_info->scrub_cancel_req, 0);
2221         init_waitqueue_head(&fs_info->scrub_pause_wait);
2222         fs_info->scrub_workers_refcnt = 0;
2223 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2224         fs_info->check_integrity_print_mask = 0;
2225 #endif
2226 
2227         spin_lock_init(&fs_info->balance_lock);
2228         mutex_init(&fs_info->balance_mutex);
2229         atomic_set(&fs_info->balance_running, 0);
2230         atomic_set(&fs_info->balance_pause_req, 0);
2231         atomic_set(&fs_info->balance_cancel_req, 0);
2232         fs_info->balance_ctl = NULL;
2233         init_waitqueue_head(&fs_info->balance_wait_q);
2234 
2235         sb->s_blocksize = 4096;
2236         sb->s_blocksize_bits = blksize_bits(4096);
2237         sb->s_bdi = &fs_info->bdi;
2238 
2239         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2240         set_nlink(fs_info->btree_inode, 1);
2241         /*
2242          * we set the i_size on the btree inode to the max possible int.
2243          * the real end of the address space is determined by all of
2244          * the devices in the system
2245          */
2246         fs_info->btree_inode->i_size = OFFSET_MAX;
2247         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2248         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2249 
2250         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2251         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2252                              fs_info->btree_inode->i_mapping);
2253         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2254         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2255 
2256         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2257 
2258         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2259         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2260                sizeof(struct btrfs_key));
2261         set_bit(BTRFS_INODE_DUMMY,
2262                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2263         btrfs_insert_inode_hash(fs_info->btree_inode);
2264 
2265         spin_lock_init(&fs_info->block_group_cache_lock);
2266         fs_info->block_group_cache_tree = RB_ROOT;
2267         fs_info->first_logical_byte = (u64)-1;
2268 
2269         extent_io_tree_init(&fs_info->freed_extents[0],
2270                              fs_info->btree_inode->i_mapping);
2271         extent_io_tree_init(&fs_info->freed_extents[1],
2272                              fs_info->btree_inode->i_mapping);
2273         fs_info->pinned_extents = &fs_info->freed_extents[0];
2274         fs_info->do_barriers = 1;
2275 
2276 
2277         mutex_init(&fs_info->ordered_operations_mutex);
2278         mutex_init(&fs_info->ordered_extent_flush_mutex);
2279         mutex_init(&fs_info->tree_log_mutex);
2280         mutex_init(&fs_info->chunk_mutex);
2281         mutex_init(&fs_info->transaction_kthread_mutex);
2282         mutex_init(&fs_info->cleaner_mutex);
2283         mutex_init(&fs_info->volume_mutex);
2284         init_rwsem(&fs_info->extent_commit_sem);
2285         init_rwsem(&fs_info->cleanup_work_sem);
2286         init_rwsem(&fs_info->subvol_sem);
2287         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2288         fs_info->dev_replace.lock_owner = 0;
2289         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2290         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2291         mutex_init(&fs_info->dev_replace.lock_management_lock);
2292         mutex_init(&fs_info->dev_replace.lock);
2293 
2294         spin_lock_init(&fs_info->qgroup_lock);
2295         mutex_init(&fs_info->qgroup_ioctl_lock);
2296         fs_info->qgroup_tree = RB_ROOT;
2297         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2298         fs_info->qgroup_seq = 1;
2299         fs_info->quota_enabled = 0;
2300         fs_info->pending_quota_state = 0;
2301         fs_info->qgroup_ulist = NULL;
2302         mutex_init(&fs_info->qgroup_rescan_lock);
2303 
2304         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2305         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2306 
2307         init_waitqueue_head(&fs_info->transaction_throttle);
2308         init_waitqueue_head(&fs_info->transaction_wait);
2309         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2310         init_waitqueue_head(&fs_info->async_submit_wait);
2311 
2312         ret = btrfs_alloc_stripe_hash_table(fs_info);
2313         if (ret) {
2314                 err = ret;
2315                 goto fail_alloc;
2316         }
2317 
2318         __setup_root(4096, 4096, 4096, 4096, tree_root,
2319                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2320 
2321         invalidate_bdev(fs_devices->latest_bdev);
2322 
2323         /*
2324          * Read super block and check the signature bytes only
2325          */
2326         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2327         if (!bh) {
2328                 err = -EINVAL;
2329                 goto fail_alloc;
2330         }
2331 
2332         /*
2333          * We want to check superblock checksum, the type is stored inside.
2334          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2335          */
2336         if (btrfs_check_super_csum(bh->b_data)) {
2337                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2338                 err = -EINVAL;
2339                 brelse(bh);
2340                 goto fail_alloc;
2341         }
2342 
2343         /*
2344          * super_copy is zeroed at allocation time and we never touch the
2345          * following bytes up to INFO_SIZE, the checksum is calculated from
2346          * the whole block of INFO_SIZE
2347          */
2348         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2349         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2350                sizeof(*fs_info->super_for_commit));
2351         brelse(bh);
2352 
2353         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2354 
2355         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2356         if (ret) {
2357                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2358                 err = -EINVAL;
2359                 goto fail_alloc;
2360         }
2361 
2362         disk_super = fs_info->super_copy;
2363         if (!btrfs_super_root(disk_super))
2364                 goto fail_alloc;
2365 
2366         /* check FS state, whether FS is broken. */
2367         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2368                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2369 
2370         /*
2371          * run through our array of backup supers and setup
2372          * our ring pointer to the oldest one
2373          */
2374         generation = btrfs_super_generation(disk_super);
2375         find_oldest_super_backup(fs_info, generation);
2376 
2377         /*
2378          * In the long term, we'll store the compression type in the super
2379          * block, and it'll be used for per file compression control.
2380          */
2381         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2382 
2383         ret = btrfs_parse_options(tree_root, options);
2384         if (ret) {
2385                 err = ret;
2386                 goto fail_alloc;
2387         }
2388 
2389         features = btrfs_super_incompat_flags(disk_super) &
2390                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2391         if (features) {
2392                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2393                        "unsupported optional features (%Lx).\n",
2394                        features);
2395                 err = -EINVAL;
2396                 goto fail_alloc;
2397         }
2398 
2399         if (btrfs_super_leafsize(disk_super) !=
2400             btrfs_super_nodesize(disk_super)) {
2401                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2402                        "blocksizes don't match.  node %d leaf %d\n",
2403                        btrfs_super_nodesize(disk_super),
2404                        btrfs_super_leafsize(disk_super));
2405                 err = -EINVAL;
2406                 goto fail_alloc;
2407         }
2408         if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2409                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2410                        "blocksize (%d) was too large\n",
2411                        btrfs_super_leafsize(disk_super));
2412                 err = -EINVAL;
2413                 goto fail_alloc;
2414         }
2415 
2416         features = btrfs_super_incompat_flags(disk_super);
2417         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2418         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2419                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2420 
2421         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2422                 printk(KERN_INFO "BTRFS: has skinny extents\n");
2423 
2424         /*
2425          * flag our filesystem as having big metadata blocks if
2426          * they are bigger than the page size
2427          */
2428         if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2429                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2430                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2431                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2432         }
2433 
2434         nodesize = btrfs_super_nodesize(disk_super);
2435         leafsize = btrfs_super_leafsize(disk_super);
2436         sectorsize = btrfs_super_sectorsize(disk_super);
2437         stripesize = btrfs_super_stripesize(disk_super);
2438         fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2439         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2440 
2441         /*
2442          * mixed block groups end up with duplicate but slightly offset
2443          * extent buffers for the same range.  It leads to corruptions
2444          */
2445         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2446             (sectorsize != leafsize)) {
2447                 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2448                                 "are not allowed for mixed block groups on %s\n",
2449                                 sb->s_id);
2450                 goto fail_alloc;
2451         }
2452 
2453         /*
2454          * Needn't use the lock because there is no other task which will
2455          * update the flag.
2456          */
2457         btrfs_set_super_incompat_flags(disk_super, features);
2458 
2459         features = btrfs_super_compat_ro_flags(disk_super) &
2460                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2461         if (!(sb->s_flags & MS_RDONLY) && features) {
2462                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2463                        "unsupported option features (%Lx).\n",
2464                        features);
2465                 err = -EINVAL;
2466                 goto fail_alloc;
2467         }
2468 
2469         btrfs_init_workers(&fs_info->generic_worker,
2470                            "genwork", 1, NULL);
2471 
2472         btrfs_init_workers(&fs_info->workers, "worker",
2473                            fs_info->thread_pool_size,
2474                            &fs_info->generic_worker);
2475 
2476         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2477                            fs_info->thread_pool_size, NULL);
2478 
2479         btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2480                            fs_info->thread_pool_size, NULL);
2481 
2482         btrfs_init_workers(&fs_info->submit_workers, "submit",
2483                            min_t(u64, fs_devices->num_devices,
2484                            fs_info->thread_pool_size), NULL);
2485 
2486         btrfs_init_workers(&fs_info->caching_workers, "cache",
2487                            fs_info->thread_pool_size, NULL);
2488 
2489         /* a higher idle thresh on the submit workers makes it much more
2490          * likely that bios will be send down in a sane order to the
2491          * devices
2492          */
2493         fs_info->submit_workers.idle_thresh = 64;
2494 
2495         fs_info->workers.idle_thresh = 16;
2496         fs_info->workers.ordered = 1;
2497 
2498         fs_info->delalloc_workers.idle_thresh = 2;
2499         fs_info->delalloc_workers.ordered = 1;
2500 
2501         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2502                            &fs_info->generic_worker);
2503         btrfs_init_workers(&fs_info->endio_workers, "endio",
2504                            fs_info->thread_pool_size,
2505                            &fs_info->generic_worker);
2506         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2507                            fs_info->thread_pool_size,
2508                            &fs_info->generic_worker);
2509         btrfs_init_workers(&fs_info->endio_meta_write_workers,
2510                            "endio-meta-write", fs_info->thread_pool_size,
2511                            &fs_info->generic_worker);
2512         btrfs_init_workers(&fs_info->endio_raid56_workers,
2513                            "endio-raid56", fs_info->thread_pool_size,
2514                            &fs_info->generic_worker);
2515         btrfs_init_workers(&fs_info->rmw_workers,
2516                            "rmw", fs_info->thread_pool_size,
2517                            &fs_info->generic_worker);
2518         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2519                            fs_info->thread_pool_size,
2520                            &fs_info->generic_worker);
2521         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2522                            1, &fs_info->generic_worker);
2523         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2524                            fs_info->thread_pool_size,
2525                            &fs_info->generic_worker);
2526         btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2527                            fs_info->thread_pool_size,
2528                            &fs_info->generic_worker);
2529         btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2530                            &fs_info->generic_worker);
2531 
2532         /*
2533          * endios are largely parallel and should have a very
2534          * low idle thresh
2535          */
2536         fs_info->endio_workers.idle_thresh = 4;
2537         fs_info->endio_meta_workers.idle_thresh = 4;
2538         fs_info->endio_raid56_workers.idle_thresh = 4;
2539         fs_info->rmw_workers.idle_thresh = 2;
2540 
2541         fs_info->endio_write_workers.idle_thresh = 2;
2542         fs_info->endio_meta_write_workers.idle_thresh = 2;
2543         fs_info->readahead_workers.idle_thresh = 2;
2544 
2545         /*
2546          * btrfs_start_workers can really only fail because of ENOMEM so just
2547          * return -ENOMEM if any of these fail.
2548          */
2549         ret = btrfs_start_workers(&fs_info->workers);
2550         ret |= btrfs_start_workers(&fs_info->generic_worker);
2551         ret |= btrfs_start_workers(&fs_info->submit_workers);
2552         ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2553         ret |= btrfs_start_workers(&fs_info->fixup_workers);
2554         ret |= btrfs_start_workers(&fs_info->endio_workers);
2555         ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2556         ret |= btrfs_start_workers(&fs_info->rmw_workers);
2557         ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2558         ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2559         ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2560         ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2561         ret |= btrfs_start_workers(&fs_info->delayed_workers);
2562         ret |= btrfs_start_workers(&fs_info->caching_workers);
2563         ret |= btrfs_start_workers(&fs_info->readahead_workers);
2564         ret |= btrfs_start_workers(&fs_info->flush_workers);
2565         ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2566         if (ret) {
2567                 err = -ENOMEM;
2568                 goto fail_sb_buffer;
2569         }
2570 
2571         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2572         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2573                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2574 
2575         tree_root->nodesize = nodesize;
2576         tree_root->leafsize = leafsize;
2577         tree_root->sectorsize = sectorsize;
2578         tree_root->stripesize = stripesize;
2579 
2580         sb->s_blocksize = sectorsize;
2581         sb->s_blocksize_bits = blksize_bits(sectorsize);
2582 
2583         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2584                 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2585                 goto fail_sb_buffer;
2586         }
2587 
2588         if (sectorsize != PAGE_SIZE) {
2589                 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2590                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2591                 goto fail_sb_buffer;
2592         }
2593 
2594         mutex_lock(&fs_info->chunk_mutex);
2595         ret = btrfs_read_sys_array(tree_root);
2596         mutex_unlock(&fs_info->chunk_mutex);
2597         if (ret) {
2598                 printk(KERN_WARNING "BTRFS: failed to read the system "
2599                        "array on %s\n", sb->s_id);
2600                 goto fail_sb_buffer;
2601         }
2602 
2603         blocksize = btrfs_level_size(tree_root,
2604                                      btrfs_super_chunk_root_level(disk_super));
2605         generation = btrfs_super_chunk_root_generation(disk_super);
2606 
2607         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2608                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2609 
2610         chunk_root->node = read_tree_block(chunk_root,
2611                                            btrfs_super_chunk_root(disk_super),
2612                                            blocksize, generation);
2613         if (!chunk_root->node ||
2614             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2615                 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2616                        sb->s_id);
2617                 goto fail_tree_roots;
2618         }
2619         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2620         chunk_root->commit_root = btrfs_root_node(chunk_root);
2621 
2622         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2623            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2624 
2625         ret = btrfs_read_chunk_tree(chunk_root);
2626         if (ret) {
2627                 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2628                        sb->s_id);
2629                 goto fail_tree_roots;
2630         }
2631 
2632         /*
2633          * keep the device that is marked to be the target device for the
2634          * dev_replace procedure
2635          */
2636         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2637 
2638         if (!fs_devices->latest_bdev) {
2639                 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2640                        sb->s_id);
2641                 goto fail_tree_roots;
2642         }
2643 
2644 retry_root_backup:
2645         blocksize = btrfs_level_size(tree_root,
2646                                      btrfs_super_root_level(disk_super));
2647         generation = btrfs_super_generation(disk_super);
2648 
2649         tree_root->node = read_tree_block(tree_root,
2650                                           btrfs_super_root(disk_super),
2651                                           blocksize, generation);
2652         if (!tree_root->node ||
2653             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2654                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2655                        sb->s_id);
2656 
2657                 goto recovery_tree_root;
2658         }
2659 
2660         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2661         tree_root->commit_root = btrfs_root_node(tree_root);
2662         btrfs_set_root_refs(&tree_root->root_item, 1);
2663 
2664         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2665         location.type = BTRFS_ROOT_ITEM_KEY;
2666         location.offset = 0;
2667 
2668         extent_root = btrfs_read_tree_root(tree_root, &location);
2669         if (IS_ERR(extent_root)) {
2670                 ret = PTR_ERR(extent_root);
2671                 goto recovery_tree_root;
2672         }
2673         extent_root->track_dirty = 1;
2674         fs_info->extent_root = extent_root;
2675 
2676         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2677         dev_root = btrfs_read_tree_root(tree_root, &location);
2678         if (IS_ERR(dev_root)) {
2679                 ret = PTR_ERR(dev_root);
2680                 goto recovery_tree_root;
2681         }
2682         dev_root->track_dirty = 1;
2683         fs_info->dev_root = dev_root;
2684         btrfs_init_devices_late(fs_info);
2685 
2686         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2687         csum_root = btrfs_read_tree_root(tree_root, &location);
2688         if (IS_ERR(csum_root)) {
2689                 ret = PTR_ERR(csum_root);
2690                 goto recovery_tree_root;
2691         }
2692         csum_root->track_dirty = 1;
2693         fs_info->csum_root = csum_root;
2694 
2695         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2696         quota_root = btrfs_read_tree_root(tree_root, &location);
2697         if (!IS_ERR(quota_root)) {
2698                 quota_root->track_dirty = 1;
2699                 fs_info->quota_enabled = 1;
2700                 fs_info->pending_quota_state = 1;
2701                 fs_info->quota_root = quota_root;
2702         }
2703 
2704         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2705         uuid_root = btrfs_read_tree_root(tree_root, &location);
2706         if (IS_ERR(uuid_root)) {
2707                 ret = PTR_ERR(uuid_root);
2708                 if (ret != -ENOENT)
2709                         goto recovery_tree_root;
2710                 create_uuid_tree = true;
2711                 check_uuid_tree = false;
2712         } else {
2713                 uuid_root->track_dirty = 1;
2714                 fs_info->uuid_root = uuid_root;
2715                 create_uuid_tree = false;
2716                 check_uuid_tree =
2717                     generation != btrfs_super_uuid_tree_generation(disk_super);
2718         }
2719 
2720         fs_info->generation = generation;
2721         fs_info->last_trans_committed = generation;
2722 
2723         ret = btrfs_recover_balance(fs_info);
2724         if (ret) {
2725                 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2726                 goto fail_block_groups;
2727         }
2728 
2729         ret = btrfs_init_dev_stats(fs_info);
2730         if (ret) {
2731                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2732                        ret);
2733                 goto fail_block_groups;
2734         }
2735 
2736         ret = btrfs_init_dev_replace(fs_info);
2737         if (ret) {
2738                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2739                 goto fail_block_groups;
2740         }
2741 
2742         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2743 
2744         ret = btrfs_sysfs_add_one(fs_info);
2745         if (ret) {
2746                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2747                 goto fail_block_groups;
2748         }
2749 
2750         ret = btrfs_init_space_info(fs_info);
2751         if (ret) {
2752                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2753                 goto fail_sysfs;
2754         }
2755 
2756         ret = btrfs_read_block_groups(extent_root);
2757         if (ret) {
2758                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2759                 goto fail_sysfs;
2760         }
2761         fs_info->num_tolerated_disk_barrier_failures =
2762                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2763         if (fs_info->fs_devices->missing_devices >
2764              fs_info->num_tolerated_disk_barrier_failures &&
2765             !(sb->s_flags & MS_RDONLY)) {
2766                 printk(KERN_WARNING "BTRFS: "
2767                         "too many missing devices, writeable mount is not allowed\n");
2768                 goto fail_sysfs;
2769         }
2770 
2771         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2772                                                "btrfs-cleaner");
2773         if (IS_ERR(fs_info->cleaner_kthread))
2774                 goto fail_sysfs;
2775 
2776         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2777                                                    tree_root,
2778                                                    "btrfs-transaction");
2779         if (IS_ERR(fs_info->transaction_kthread))
2780                 goto fail_cleaner;
2781 
2782         if (!btrfs_test_opt(tree_root, SSD) &&
2783             !btrfs_test_opt(tree_root, NOSSD) &&
2784             !fs_info->fs_devices->rotating) {
2785                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2786                        "mode\n");
2787                 btrfs_set_opt(fs_info->mount_opt, SSD);
2788         }
2789 
2790         /* Set the real inode map cache flag */
2791         if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2792                 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2793 
2794 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2795         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2796                 ret = btrfsic_mount(tree_root, fs_devices,
2797                                     btrfs_test_opt(tree_root,
2798                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2799                                     1 : 0,
2800                                     fs_info->check_integrity_print_mask);
2801                 if (ret)
2802                         printk(KERN_WARNING "BTRFS: failed to initialize"
2803                                " integrity check module %s\n", sb->s_id);
2804         }
2805 #endif
2806         ret = btrfs_read_qgroup_config(fs_info);
2807         if (ret)
2808                 goto fail_trans_kthread;
2809 
2810         /* do not make disk changes in broken FS */
2811         if (btrfs_super_log_root(disk_super) != 0) {
2812                 u64 bytenr = btrfs_super_log_root(disk_super);
2813 
2814                 if (fs_devices->rw_devices == 0) {
2815                         printk(KERN_WARNING "BTRFS: log replay required "
2816                                "on RO media\n");
2817                         err = -EIO;
2818                         goto fail_qgroup;
2819                 }
2820                 blocksize =
2821                      btrfs_level_size(tree_root,
2822                                       btrfs_super_log_root_level(disk_super));
2823 
2824                 log_tree_root = btrfs_alloc_root(fs_info);
2825                 if (!log_tree_root) {
2826                         err = -ENOMEM;
2827                         goto fail_qgroup;
2828                 }
2829 
2830                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2831                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2832 
2833                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2834                                                       blocksize,
2835                                                       generation + 1);
2836                 if (!log_tree_root->node ||
2837                     !extent_buffer_uptodate(log_tree_root->node)) {
2838                         printk(KERN_ERR "BTRFS: failed to read log tree\n");
2839                         free_extent_buffer(log_tree_root->node);
2840                         kfree(log_tree_root);
2841                         goto fail_trans_kthread;
2842                 }
2843                 /* returns with log_tree_root freed on success */
2844                 ret = btrfs_recover_log_trees(log_tree_root);
2845                 if (ret) {
2846                         btrfs_error(tree_root->fs_info, ret,
2847                                     "Failed to recover log tree");
2848                         free_extent_buffer(log_tree_root->node);
2849                         kfree(log_tree_root);
2850                         goto fail_trans_kthread;
2851                 }
2852 
2853                 if (sb->s_flags & MS_RDONLY) {
2854                         ret = btrfs_commit_super(tree_root);
2855                         if (ret)
2856                                 goto fail_trans_kthread;
2857                 }
2858         }
2859 
2860         ret = btrfs_find_orphan_roots(tree_root);
2861         if (ret)
2862                 goto fail_trans_kthread;
2863 
2864         if (!(sb->s_flags & MS_RDONLY)) {
2865                 ret = btrfs_cleanup_fs_roots(fs_info);
2866                 if (ret)
2867                         goto fail_trans_kthread;
2868 
2869                 ret = btrfs_recover_relocation(tree_root);
2870                 if (ret < 0) {
2871                         printk(KERN_WARNING
2872                                "BTRFS: failed to recover relocation\n");
2873                         err = -EINVAL;
2874                         goto fail_qgroup;
2875                 }
2876         }
2877 
2878         location.objectid = BTRFS_FS_TREE_OBJECTID;
2879         location.type = BTRFS_ROOT_ITEM_KEY;
2880         location.offset = 0;
2881 
2882         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2883         if (IS_ERR(fs_info->fs_root)) {
2884                 err = PTR_ERR(fs_info->fs_root);
2885                 goto fail_qgroup;
2886         }
2887 
2888         if (sb->s_flags & MS_RDONLY)
2889                 return 0;
2890 
2891         down_read(&fs_info->cleanup_work_sem);
2892         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2893             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2894                 up_read(&fs_info->cleanup_work_sem);
2895                 close_ctree(tree_root);
2896                 return ret;
2897         }
2898         up_read(&fs_info->cleanup_work_sem);
2899 
2900         ret = btrfs_resume_balance_async(fs_info);
2901         if (ret) {
2902                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2903                 close_ctree(tree_root);
2904                 return ret;
2905         }
2906 
2907         ret = btrfs_resume_dev_replace_async(fs_info);
2908         if (ret) {
2909                 pr_warn("BTRFS: failed to resume dev_replace\n");
2910                 close_ctree(tree_root);
2911                 return ret;
2912         }
2913 
2914         btrfs_qgroup_rescan_resume(fs_info);
2915 
2916         if (create_uuid_tree) {
2917                 pr_info("BTRFS: creating UUID tree\n");
2918                 ret = btrfs_create_uuid_tree(fs_info);
2919                 if (ret) {
2920                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
2921                                 ret);
2922                         close_ctree(tree_root);
2923                         return ret;
2924                 }
2925         } else if (check_uuid_tree ||
2926                    btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2927                 pr_info("BTRFS: checking UUID tree\n");
2928                 ret = btrfs_check_uuid_tree(fs_info);
2929                 if (ret) {
2930                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
2931                                 ret);
2932                         close_ctree(tree_root);
2933                         return ret;
2934                 }
2935         } else {
2936                 fs_info->update_uuid_tree_gen = 1;
2937         }
2938 
2939         return 0;
2940 
2941 fail_qgroup:
2942         btrfs_free_qgroup_config(fs_info);
2943 fail_trans_kthread:
2944         kthread_stop(fs_info->transaction_kthread);
2945         btrfs_cleanup_transaction(fs_info->tree_root);
2946         del_fs_roots(fs_info);
2947 fail_cleaner:
2948         kthread_stop(fs_info->cleaner_kthread);
2949 
2950         /*
2951          * make sure we're done with the btree inode before we stop our
2952          * kthreads
2953          */
2954         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2955 
2956 fail_sysfs:
2957         btrfs_sysfs_remove_one(fs_info);
2958 
2959 fail_block_groups:
2960         btrfs_put_block_group_cache(fs_info);
2961         btrfs_free_block_groups(fs_info);
2962 
2963 fail_tree_roots:
2964         free_root_pointers(fs_info, 1);
2965         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2966 
2967 fail_sb_buffer:
2968         btrfs_stop_all_workers(fs_info);
2969 fail_alloc:
2970 fail_iput:
2971         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2972 
2973         iput(fs_info->btree_inode);
2974 fail_delalloc_bytes:
2975         percpu_counter_destroy(&fs_info->delalloc_bytes);
2976 fail_dirty_metadata_bytes:
2977         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2978 fail_bdi:
2979         bdi_destroy(&fs_info->bdi);
2980 fail_srcu:
2981         cleanup_srcu_struct(&fs_info->subvol_srcu);
2982 fail:
2983         btrfs_free_stripe_hash_table(fs_info);
2984         btrfs_close_devices(fs_info->fs_devices);
2985         return err;
2986 
2987 recovery_tree_root:
2988         if (!btrfs_test_opt(tree_root, RECOVERY))
2989                 goto fail_tree_roots;
2990 
2991         free_root_pointers(fs_info, 0);
2992 
2993         /* don't use the log in recovery mode, it won't be valid */
2994         btrfs_set_super_log_root(disk_super, 0);
2995 
2996         /* we can't trust the free space cache either */
2997         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2998 
2999         ret = next_root_backup(fs_info, fs_info->super_copy,
3000                                &num_backups_tried, &backup_index);
3001         if (ret == -1)
3002                 goto fail_block_groups;
3003         goto retry_root_backup;
3004 }
3005 
3006 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3007 {
3008         if (uptodate) {
3009                 set_buffer_uptodate(bh);
3010         } else {
3011                 struct btrfs_device *device = (struct btrfs_device *)
3012                         bh->b_private;
3013 
3014                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3015                                           "I/O error on %s\n",
3016                                           rcu_str_deref(device->name));
3017                 /* note, we dont' set_buffer_write_io_error because we have
3018                  * our own ways of dealing with the IO errors
3019                  */
3020                 clear_buffer_uptodate(bh);
3021                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3022         }
3023         unlock_buffer(bh);
3024         put_bh(bh);
3025 }
3026 
3027 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3028 {
3029         struct buffer_head *bh;
3030         struct buffer_head *latest = NULL;
3031         struct btrfs_super_block *super;
3032         int i;
3033         u64 transid = 0;
3034         u64 bytenr;
3035 
3036         /* we would like to check all the supers, but that would make
3037          * a btrfs mount succeed after a mkfs from a different FS.
3038          * So, we need to add a special mount option to scan for
3039          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3040          */
3041         for (i = 0; i < 1; i++) {
3042                 bytenr = btrfs_sb_offset(i);
3043                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3044                                         i_size_read(bdev->bd_inode))
3045                         break;
3046                 bh = __bread(bdev, bytenr / 4096,
3047                                         BTRFS_SUPER_INFO_SIZE);
3048                 if (!bh)
3049                         continue;
3050 
3051                 super = (struct btrfs_super_block *)bh->b_data;
3052                 if (btrfs_super_bytenr(super) != bytenr ||
3053                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3054                         brelse(bh);
3055                         continue;
3056                 }
3057 
3058                 if (!latest || btrfs_super_generation(super) > transid) {
3059                         brelse(latest);
3060                         latest = bh;
3061                         transid = btrfs_super_generation(super);
3062                 } else {
3063                         brelse(bh);
3064                 }
3065         }
3066         return latest;
3067 }
3068 
3069 /*
3070  * this should be called twice, once with wait == 0 and
3071  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3072  * we write are pinned.
3073  *
3074  * They are released when wait == 1 is done.
3075  * max_mirrors must be the same for both runs, and it indicates how
3076  * many supers on this one device should be written.
3077  *
3078  * max_mirrors == 0 means to write them all.
3079  */
3080 static int write_dev_supers(struct btrfs_device *device,
3081                             struct btrfs_super_block *sb,
3082                             int do_barriers, int wait, int max_mirrors)
3083 {
3084         struct buffer_head *bh;
3085         int i;
3086         int ret;
3087         int errors = 0;
3088         u32 crc;
3089         u64 bytenr;
3090 
3091         if (max_mirrors == 0)
3092                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3093 
3094         for (i = 0; i < max_mirrors; i++) {
3095                 bytenr = btrfs_sb_offset(i);
3096                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3097                         break;
3098 
3099                 if (wait) {
3100                         bh = __find_get_block(device->bdev, bytenr / 4096,
3101                                               BTRFS_SUPER_INFO_SIZE);
3102                         if (!bh) {
3103                                 errors++;
3104                                 continue;
3105                         }
3106                         wait_on_buffer(bh);
3107                         if (!buffer_uptodate(bh))
3108                                 errors++;
3109 
3110                         /* drop our reference */
3111                         brelse(bh);
3112 
3113                         /* drop the reference from the wait == 0 run */
3114                         brelse(bh);
3115                         continue;
3116                 } else {
3117                         btrfs_set_super_bytenr(sb, bytenr);
3118 
3119                         crc = ~(u32)0;
3120                         crc = btrfs_csum_data((char *)sb +
3121                                               BTRFS_CSUM_SIZE, crc,
3122                                               BTRFS_SUPER_INFO_SIZE -
3123                                               BTRFS_CSUM_SIZE);
3124                         btrfs_csum_final(crc, sb->csum);
3125 
3126                         /*
3127                          * one reference for us, and we leave it for the
3128                          * caller
3129                          */
3130                         bh = __getblk(device->bdev, bytenr / 4096,
3131                                       BTRFS_SUPER_INFO_SIZE);
3132                         if (!bh) {
3133                                 printk(KERN_ERR "BTRFS: couldn't get super "
3134                                        "buffer head for bytenr %Lu\n", bytenr);
3135                                 errors++;
3136                                 continue;
3137                         }
3138 
3139                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3140 
3141                         /* one reference for submit_bh */
3142                         get_bh(bh);
3143 
3144                         set_buffer_uptodate(bh);
3145                         lock_buffer(bh);
3146                         bh->b_end_io = btrfs_end_buffer_write_sync;
3147                         bh->b_private = device;
3148                 }
3149 
3150                 /*
3151                  * we fua the first super.  The others we allow
3152                  * to go down lazy.
3153                  */
3154                 if (i == 0)
3155                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3156                 else
3157                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3158                 if (ret)
3159                         errors++;
3160         }
3161         return errors < i ? 0 : -1;
3162 }
3163 
3164 /*
3165  * endio for the write_dev_flush, this will wake anyone waiting
3166  * for the barrier when it is done
3167  */
3168 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3169 {
3170         if (err) {
3171                 if (err == -EOPNOTSUPP)
3172                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3173                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3174         }
3175         if (bio->bi_private)
3176                 complete(bio->bi_private);
3177         bio_put(bio);
3178 }
3179 
3180 /*
3181  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3182  * sent down.  With wait == 1, it waits for the previous flush.
3183  *
3184  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3185  * capable
3186  */
3187 static int write_dev_flush(struct btrfs_device *device, int wait)
3188 {
3189         struct bio *bio;
3190         int ret = 0;
3191 
3192         if (device->nobarriers)
3193                 return 0;
3194 
3195         if (wait) {
3196                 bio = device->flush_bio;
3197                 if (!bio)
3198                         return 0;
3199 
3200                 wait_for_completion(&device->flush_wait);
3201 
3202                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3203                         printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3204                                       rcu_str_deref(device->name));
3205                         device->nobarriers = 1;
3206                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3207                         ret = -EIO;
3208                         btrfs_dev_stat_inc_and_print(device,
3209                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3210                 }
3211 
3212                 /* drop the reference from the wait == 0 run */
3213                 bio_put(bio);
3214                 device->flush_bio = NULL;
3215 
3216                 return ret;
3217         }
3218 
3219         /*
3220          * one reference for us, and we leave it for the
3221          * caller
3222          */
3223         device->flush_bio = NULL;
3224         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3225         if (!bio)
3226                 return -ENOMEM;
3227 
3228         bio->bi_end_io = btrfs_end_empty_barrier;
3229         bio->bi_bdev = device->bdev;
3230         init_completion(&device->flush_wait);
3231         bio->bi_private = &device->flush_wait;
3232         device->flush_bio = bio;
3233 
3234         bio_get(bio);
3235         btrfsic_submit_bio(WRITE_FLUSH, bio);
3236 
3237         return 0;
3238 }
3239 
3240 /*
3241  * send an empty flush down to each device in parallel,
3242  * then wait for them
3243  */
3244 static int barrier_all_devices(struct btrfs_fs_info *info)
3245 {
3246         struct list_head *head;
3247         struct btrfs_device *dev;
3248         int errors_send = 0;
3249         int errors_wait = 0;
3250         int ret;
3251 
3252         /* send down all the barriers */
3253         head = &info->fs_devices->devices;
3254         list_for_each_entry_rcu(dev, head, dev_list) {
3255                 if (dev->missing)
3256                         continue;
3257                 if (!dev->bdev) {
3258                         errors_send++;
3259                         continue;
3260                 }
3261                 if (!dev->in_fs_metadata || !dev->writeable)
3262                         continue;
3263 
3264                 ret = write_dev_flush(dev, 0);
3265                 if (ret)
3266                         errors_send++;
3267         }
3268 
3269         /* wait for all the barriers */
3270         list_for_each_entry_rcu(dev, head, dev_list) {
3271                 if (dev->missing)
3272                         continue;
3273                 if (!dev->bdev) {
3274                         errors_wait++;
3275                         continue;
3276                 }
3277                 if (!dev->in_fs_metadata || !dev->writeable)
3278                         continue;
3279 
3280                 ret = write_dev_flush(dev, 1);
3281                 if (ret)
3282                         errors_wait++;
3283         }
3284         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3285             errors_wait > info->num_tolerated_disk_barrier_failures)
3286                 return -EIO;
3287         return 0;
3288 }
3289 
3290 int btrfs_calc_num_tolerated_disk_barrier_failures(
3291         struct btrfs_fs_info *fs_info)
3292 {
3293         struct btrfs_ioctl_space_info space;
3294         struct btrfs_space_info *sinfo;
3295         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3296                        BTRFS_BLOCK_GROUP_SYSTEM,
3297                        BTRFS_BLOCK_GROUP_METADATA,
3298                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3299         int num_types = 4;
3300         int i;
3301         int c;
3302         int num_tolerated_disk_barrier_failures =
3303                 (int)fs_info->fs_devices->num_devices;
3304 
3305         for (i = 0; i < num_types; i++) {
3306                 struct btrfs_space_info *tmp;
3307 
3308                 sinfo = NULL;
3309                 rcu_read_lock();
3310                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3311                         if (tmp->flags == types[i]) {
3312                                 sinfo = tmp;
3313                                 break;
3314                         }
3315                 }
3316                 rcu_read_unlock();
3317 
3318                 if (!sinfo)
3319                         continue;
3320 
3321                 down_read(&sinfo->groups_sem);
3322                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3323                         if (!list_empty(&sinfo->block_groups[c])) {
3324                                 u64 flags;
3325 
3326                                 btrfs_get_block_group_info(
3327                                         &sinfo->block_groups[c], &space);
3328                                 if (space.total_bytes == 0 ||
3329                                     space.used_bytes == 0)
3330                                         continue;
3331                                 flags = space.flags;
3332                                 /*
3333                                  * return
3334                                  * 0: if dup, single or RAID0 is configured for
3335                                  *    any of metadata, system or data, else
3336                                  * 1: if RAID5 is configured, or if RAID1 or
3337                                  *    RAID10 is configured and only two mirrors
3338                                  *    are used, else
3339                                  * 2: if RAID6 is configured, else
3340                                  * num_mirrors - 1: if RAID1 or RAID10 is
3341                                  *                  configured and more than
3342                                  *                  2 mirrors are used.
3343                                  */
3344                                 if (num_tolerated_disk_barrier_failures > 0 &&
3345                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3346                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3347                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3348                                       == 0)))
3349                                         num_tolerated_disk_barrier_failures = 0;
3350                                 else if (num_tolerated_disk_barrier_failures > 1) {
3351                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3352                                             BTRFS_BLOCK_GROUP_RAID5 |
3353                                             BTRFS_BLOCK_GROUP_RAID10)) {
3354                                                 num_tolerated_disk_barrier_failures = 1;
3355                                         } else if (flags &
3356                                                    BTRFS_BLOCK_GROUP_RAID6) {
3357                                                 num_tolerated_disk_barrier_failures = 2;
3358                                         }
3359                                 }
3360                         }
3361                 }
3362                 up_read(&sinfo->groups_sem);
3363         }
3364 
3365         return num_tolerated_disk_barrier_failures;
3366 }
3367 
3368 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3369 {
3370         struct list_head *head;
3371         struct btrfs_device *dev;
3372         struct btrfs_super_block *sb;
3373         struct btrfs_dev_item *dev_item;
3374         int ret;
3375         int do_barriers;
3376         int max_errors;
3377         int total_errors = 0;
3378         u64 flags;
3379 
3380         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3381         backup_super_roots(root->fs_info);
3382 
3383         sb = root->fs_info->super_for_commit;
3384         dev_item = &sb->dev_item;
3385 
3386         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3387         head = &root->fs_info->fs_devices->devices;
3388         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3389 
3390         if (do_barriers) {
3391                 ret = barrier_all_devices(root->fs_info);
3392                 if (ret) {
3393                         mutex_unlock(
3394                                 &root->fs_info->fs_devices->device_list_mutex);
3395                         btrfs_error(root->fs_info, ret,
3396                                     "errors while submitting device barriers.");
3397                         return ret;
3398                 }
3399         }
3400 
3401         list_for_each_entry_rcu(dev, head, dev_list) {
3402                 if (!dev->bdev) {
3403                         total_errors++;
3404                         continue;
3405                 }
3406                 if (!dev->in_fs_metadata || !dev->writeable)
3407                         continue;
3408 
3409                 btrfs_set_stack_device_generation(dev_item, 0);
3410                 btrfs_set_stack_device_type(dev_item, dev->type);
3411                 btrfs_set_stack_device_id(dev_item, dev->devid);
3412                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3413                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3414                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3415                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3416                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3417                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3418                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3419 
3420                 flags = btrfs_super_flags(sb);
3421                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3422 
3423                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3424                 if (ret)
3425                         total_errors++;
3426         }
3427         if (total_errors > max_errors) {
3428                 btrfs_err(root->fs_info, "%d errors while writing supers",
3429                        total_errors);
3430                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3431 
3432                 /* FUA is masked off if unsupported and can't be the reason */
3433                 btrfs_error(root->fs_info, -EIO,
3434                             "%d errors while writing supers", total_errors);
3435                 return -EIO;
3436         }
3437 
3438         total_errors = 0;
3439         list_for_each_entry_rcu(dev, head, dev_list) {
3440                 if (!dev->bdev)
3441                         continue;
3442                 if (!dev->in_fs_metadata || !dev->writeable)
3443                         continue;
3444 
3445                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3446                 if (ret)
3447                         total_errors++;
3448         }
3449         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3450         if (total_errors > max_errors) {
3451                 btrfs_error(root->fs_info, -EIO,
3452                             "%d errors while writing supers", total_errors);
3453                 return -EIO;
3454         }
3455         return 0;
3456 }
3457 
3458 int write_ctree_super(struct btrfs_trans_handle *trans,
3459                       struct btrfs_root *root, int max_mirrors)
3460 {
3461         return write_all_supers(root, max_mirrors);
3462 }
3463 
3464 /* Drop a fs root from the radix tree and free it. */
3465 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3466                                   struct btrfs_root *root)
3467 {
3468         spin_lock(&fs_info->fs_roots_radix_lock);
3469         radix_tree_delete(&fs_info->fs_roots_radix,
3470                           (unsigned long)root->root_key.objectid);
3471         spin_unlock(&fs_info->fs_roots_radix_lock);
3472 
3473         if (btrfs_root_refs(&root->root_item) == 0)
3474                 synchronize_srcu(&fs_info->subvol_srcu);
3475 
3476         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3477                 btrfs_free_log(NULL, root);
3478 
3479         __btrfs_remove_free_space_cache(root->free_ino_pinned);
3480         __btrfs_remove_free_space_cache(root->free_ino_ctl);
3481         free_fs_root(root);
3482 }
3483 
3484 static void free_fs_root(struct btrfs_root *root)
3485 {
3486         iput(root->cache_inode);
3487         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3488         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3489         root->orphan_block_rsv = NULL;
3490         if (root->anon_dev)
3491                 free_anon_bdev(root->anon_dev);
3492         free_extent_buffer(root->node);
3493         free_extent_buffer(root->commit_root);
3494         kfree(root->free_ino_ctl);
3495         kfree(root->free_ino_pinned);
3496         kfree(root->name);
3497         btrfs_put_fs_root(root);
3498 }
3499 
3500 void btrfs_free_fs_root(struct btrfs_root *root)
3501 {
3502         free_fs_root(root);
3503 }
3504 
3505 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3506 {
3507         u64 root_objectid = 0;
3508         struct btrfs_root *gang[8];
3509         int i;
3510         int ret;
3511 
3512         while (1) {
3513                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3514                                              (void **)gang, root_objectid,
3515                                              ARRAY_SIZE(gang));
3516                 if (!ret)
3517                         break;
3518 
3519                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3520                 for (i = 0; i < ret; i++) {
3521                         int err;
3522 
3523                         root_objectid = gang[i]->root_key.objectid;
3524                         err = btrfs_orphan_cleanup(gang[i]);
3525                         if (err)
3526                                 return err;
3527                 }
3528                 root_objectid++;
3529         }
3530         return 0;
3531 }
3532 
3533 int btrfs_commit_super(struct btrfs_root *root)
3534 {
3535         struct btrfs_trans_handle *trans;
3536 
3537         mutex_lock(&root->fs_info->cleaner_mutex);
3538         btrfs_run_delayed_iputs(root);
3539         mutex_unlock(&root->fs_info->cleaner_mutex);
3540         wake_up_process(root->fs_info->cleaner_kthread);
3541 
3542         /* wait until ongoing cleanup work done */
3543         down_write(&root->fs_info->cleanup_work_sem);
3544         up_write(&root->fs_info->cleanup_work_sem);
3545 
3546         trans = btrfs_join_transaction(root);
3547         if (IS_ERR(trans))
3548                 return PTR_ERR(trans);
3549         return btrfs_commit_transaction(trans, root);
3550 }
3551 
3552 int close_ctree(struct btrfs_root *root)
3553 {
3554         struct btrfs_fs_info *fs_info = root->fs_info;
3555         int ret;
3556 
3557         fs_info->closing = 1;
3558         smp_mb();
3559 
3560         /* wait for the uuid_scan task to finish */
3561         down(&fs_info->uuid_tree_rescan_sem);
3562         /* avoid complains from lockdep et al., set sem back to initial state */
3563         up(&fs_info->uuid_tree_rescan_sem);
3564 
3565         /* pause restriper - we want to resume on mount */
3566         btrfs_pause_balance(fs_info);
3567 
3568         btrfs_dev_replace_suspend_for_unmount(fs_info);
3569 
3570         btrfs_scrub_cancel(fs_info);
3571 
3572         /* wait for any defraggers to finish */
3573         wait_event(fs_info->transaction_wait,
3574                    (atomic_read(&fs_info->defrag_running) == 0));
3575 
3576         /* clear out the rbtree of defraggable inodes */
3577         btrfs_cleanup_defrag_inodes(fs_info);
3578 
3579         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3580                 ret = btrfs_commit_super(root);
3581                 if (ret)
3582                         btrfs_err(root->fs_info, "commit super ret %d", ret);
3583         }
3584 
3585         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3586                 btrfs_error_commit_super(root);
3587 
3588         kthread_stop(fs_info->transaction_kthread);
3589         kthread_stop(fs_info->cleaner_kthread);
3590 
3591         fs_info->closing = 2;
3592         smp_mb();
3593 
3594         btrfs_free_qgroup_config(root->fs_info);
3595 
3596         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3597                 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3598                        percpu_counter_sum(&fs_info->delalloc_bytes));
3599         }
3600 
3601         btrfs_sysfs_remove_one(fs_info);
3602 
3603         del_fs_roots(fs_info);
3604 
3605         btrfs_put_block_group_cache(fs_info);
3606 
3607         btrfs_free_block_groups(fs_info);
3608 
3609         /*
3610          * we must make sure there is not any read request to
3611          * submit after we stopping all workers.
3612          */
3613         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3614         btrfs_stop_all_workers(fs_info);
3615 
3616         free_root_pointers(fs_info, 1);
3617 
3618         iput(fs_info->btree_inode);
3619 
3620 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3621         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3622                 btrfsic_unmount(root, fs_info->fs_devices);
3623 #endif
3624 
3625         btrfs_close_devices(fs_info->fs_devices);
3626         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3627 
3628         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3629         percpu_counter_destroy(&fs_info->delalloc_bytes);
3630         bdi_destroy(&fs_info->bdi);
3631         cleanup_srcu_struct(&fs_info->subvol_srcu);
3632 
3633         btrfs_free_stripe_hash_table(fs_info);
3634 
3635         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3636         root->orphan_block_rsv = NULL;
3637 
3638         return 0;
3639 }
3640 
3641 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3642                           int atomic)
3643 {
3644         int ret;
3645         struct inode *btree_inode = buf->pages[0]->mapping->host;
3646 
3647         ret = extent_buffer_uptodate(buf);
3648         if (!ret)
3649                 return ret;
3650 
3651         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3652                                     parent_transid, atomic);
3653         if (ret == -EAGAIN)
3654                 return ret;
3655         return !ret;
3656 }
3657 
3658 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3659 {
3660         return set_extent_buffer_uptodate(buf);
3661 }
3662 
3663 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3664 {
3665         struct btrfs_root *root;
3666         u64 transid = btrfs_header_generation(buf);
3667         int was_dirty;
3668 
3669 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3670         /*
3671          * This is a fast path so only do this check if we have sanity tests
3672          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3673          * outside of the sanity tests.
3674          */
3675         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3676                 return;
3677 #endif
3678         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3679         btrfs_assert_tree_locked(buf);
3680         if (transid != root->fs_info->generation)
3681                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3682                        "found %llu running %llu\n",
3683                         buf->start, transid, root->fs_info->generation);
3684         was_dirty = set_extent_buffer_dirty(buf);
3685         if (!was_dirty)
3686                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3687                                      buf->len,
3688                                      root->fs_info->dirty_metadata_batch);
3689 }
3690 
3691 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3692                                         int flush_delayed)
3693 {
3694         /*
3695          * looks as though older kernels can get into trouble with
3696          * this code, they end up stuck in balance_dirty_pages forever
3697          */
3698         int ret;
3699 
3700         if (current->flags & PF_MEMALLOC)
3701                 return;
3702 
3703         if (flush_delayed)
3704                 btrfs_balance_delayed_items(root);
3705 
3706         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3707                                      BTRFS_DIRTY_METADATA_THRESH);
3708         if (ret > 0) {
3709                 balance_dirty_pages_ratelimited(
3710                                    root->fs_info->btree_inode->i_mapping);
3711         }
3712         return;
3713 }
3714 
3715 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3716 {
3717         __btrfs_btree_balance_dirty(root, 1);
3718 }
3719 
3720 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3721 {
3722         __btrfs_btree_balance_dirty(root, 0);
3723 }
3724 
3725 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3726 {
3727         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3728         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3729 }
3730 
3731 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3732                               int read_only)
3733 {
3734         /*
3735          * Placeholder for checks
3736          */
3737         return 0;
3738 }
3739 
3740 static void btrfs_error_commit_super(struct btrfs_root *root)
3741 {
3742         mutex_lock(&root->fs_info->cleaner_mutex);
3743         btrfs_run_delayed_iputs(root);
3744         mutex_unlock(&root->fs_info->cleaner_mutex);
3745 
3746         down_write(&root->fs_info->cleanup_work_sem);
3747         up_write(&root->fs_info->cleanup_work_sem);
3748 
3749         /* cleanup FS via transaction */
3750         btrfs_cleanup_transaction(root);
3751 }
3752 
3753 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3754                                              struct btrfs_root *root)
3755 {
3756         struct btrfs_inode *btrfs_inode;
3757         struct list_head splice;
3758 
3759         INIT_LIST_HEAD(&splice);
3760 
3761         mutex_lock(&root->fs_info->ordered_operations_mutex);
3762         spin_lock(&root->fs_info->ordered_root_lock);
3763 
3764         list_splice_init(&t->ordered_operations, &splice);
3765         while (!list_empty(&splice)) {
3766                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3767                                          ordered_operations);
3768 
3769                 list_del_init(&btrfs_inode->ordered_operations);
3770                 spin_unlock(&root->fs_info->ordered_root_lock);
3771 
3772                 btrfs_invalidate_inodes(btrfs_inode->root);
3773 
3774                 spin_lock(&root->fs_info->ordered_root_lock);
3775         }
3776 
3777         spin_unlock(&root->fs_info->ordered_root_lock);
3778         mutex_unlock(&root->fs_info->ordered_operations_mutex);
3779 }
3780 
3781 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3782 {
3783         struct btrfs_ordered_extent *ordered;
3784 
3785         spin_lock(&root->ordered_extent_lock);
3786         /*
3787          * This will just short circuit the ordered completion stuff which will
3788          * make sure the ordered extent gets properly cleaned up.
3789          */
3790         list_for_each_entry(ordered, &root->ordered_extents,
3791                             root_extent_list)
3792                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3793         spin_unlock(&root->ordered_extent_lock);
3794 }
3795 
3796 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3797 {
3798         struct btrfs_root *root;
3799         struct list_head splice;
3800 
3801         INIT_LIST_HEAD(&splice);
3802 
3803         spin_lock(&fs_info->ordered_root_lock);
3804         list_splice_init(&fs_info->ordered_roots, &splice);
3805         while (!list_empty(&splice)) {
3806                 root = list_first_entry(&splice, struct btrfs_root,
3807                                         ordered_root);
3808                 list_move_tail(&root->ordered_root,
3809                                &fs_info->ordered_roots);
3810 
3811                 btrfs_destroy_ordered_extents(root);
3812 
3813                 cond_resched_lock(&fs_info->ordered_root_lock);
3814         }
3815         spin_unlock(&fs_info->ordered_root_lock);
3816 }
3817 
3818 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3819                                       struct btrfs_root *root)
3820 {
3821         struct rb_node *node;
3822         struct btrfs_delayed_ref_root *delayed_refs;
3823         struct btrfs_delayed_ref_node *ref;
3824         int ret = 0;
3825 
3826         delayed_refs = &trans->delayed_refs;
3827 
3828         spin_lock(&delayed_refs->lock);
3829         if (atomic_read(&delayed_refs->num_entries) == 0) {
3830                 spin_unlock(&delayed_refs->lock);
3831                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3832                 return ret;
3833         }
3834 
3835         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3836                 struct btrfs_delayed_ref_head *head;
3837                 bool pin_bytes = false;
3838 
3839                 head = rb_entry(node, struct btrfs_delayed_ref_head,
3840                                 href_node);
3841                 if (!mutex_trylock(&head->mutex)) {
3842                         atomic_inc(&head->node.refs);
3843                         spin_unlock(&delayed_refs->lock);
3844 
3845                         mutex_lock(&head->mutex);
3846                         mutex_unlock(&head->mutex);
3847                         btrfs_put_delayed_ref(&head->node);
3848                         spin_lock(&delayed_refs->lock);
3849                         continue;
3850                 }
3851                 spin_lock(&head->lock);
3852                 while ((node = rb_first(&head->ref_root)) != NULL) {
3853                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
3854                                        rb_node);
3855                         ref->in_tree = 0;
3856                         rb_erase(&ref->rb_node, &head->ref_root);
3857                         atomic_dec(&delayed_refs->num_entries);
3858                         btrfs_put_delayed_ref(ref);
3859                 }
3860                 if (head->must_insert_reserved)
3861                         pin_bytes = true;
3862                 btrfs_free_delayed_extent_op(head->extent_op);
3863                 delayed_refs->num_heads--;
3864                 if (head->processing == 0)
3865                         delayed_refs->num_heads_ready--;
3866                 atomic_dec(&delayed_refs->num_entries);
3867                 head->node.in_tree = 0;
3868                 rb_erase(&head->href_node, &delayed_refs->href_root);
3869                 spin_unlock(&head->lock);
3870                 spin_unlock(&delayed_refs->lock);
3871                 mutex_unlock(&head->mutex);
3872 
3873                 if (pin_bytes)
3874                         btrfs_pin_extent(root, head->node.bytenr,
3875                                          head->node.num_bytes, 1);
3876                 btrfs_put_delayed_ref(&head->node);
3877                 cond_resched();
3878                 spin_lock(&delayed_refs->lock);
3879         }
3880 
3881         spin_unlock(&delayed_refs->lock);
3882 
3883         return ret;
3884 }
3885 
3886 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3887 {
3888         struct btrfs_inode *btrfs_inode;
3889         struct list_head splice;
3890 
3891         INIT_LIST_HEAD(&splice);
3892 
3893         spin_lock(&root->delalloc_lock);
3894         list_splice_init(&root->delalloc_inodes, &splice);
3895 
3896         while (!list_empty(&splice)) {
3897                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3898                                                delalloc_inodes);
3899 
3900                 list_del_init(&btrfs_inode->delalloc_inodes);
3901                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3902                           &btrfs_inode->runtime_flags);
3903                 spin_unlock(&root->delalloc_lock);
3904 
3905                 btrfs_invalidate_inodes(btrfs_inode->root);
3906 
3907                 spin_lock(&root->delalloc_lock);
3908         }
3909 
3910         spin_unlock(&root->delalloc_lock);
3911 }
3912 
3913 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3914 {
3915         struct btrfs_root *root;
3916         struct list_head splice;
3917 
3918         INIT_LIST_HEAD(&splice);
3919 
3920         spin_lock(&fs_info->delalloc_root_lock);
3921         list_splice_init(&fs_info->delalloc_roots, &splice);
3922         while (!list_empty(&splice)) {
3923                 root = list_first_entry(&splice, struct btrfs_root,
3924                                          delalloc_root);
3925                 list_del_init(&root->delalloc_root);
3926                 root = btrfs_grab_fs_root(root);
3927                 BUG_ON(!root);
3928                 spin_unlock(&fs_info->delalloc_root_lock);
3929 
3930                 btrfs_destroy_delalloc_inodes(root);
3931                 btrfs_put_fs_root(root);
3932 
3933                 spin_lock(&fs_info->delalloc_root_lock);
3934         }
3935         spin_unlock(&fs_info->delalloc_root_lock);
3936 }
3937 
3938 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3939                                         struct extent_io_tree *dirty_pages,
3940                                         int mark)
3941 {
3942         int ret;
3943         struct extent_buffer *eb;
3944         u64 start = 0;
3945         u64 end;
3946 
3947         while (1) {
3948                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3949                                             mark, NULL);
3950                 if (ret)
3951                         break;
3952 
3953                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3954                 while (start <= end) {
3955                         eb = btrfs_find_tree_block(root, start,
3956                                                    root->leafsize);
3957                         start += root->leafsize;
3958                         if (!eb)
3959                                 continue;
3960                         wait_on_extent_buffer_writeback(eb);
3961 
3962                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3963                                                &eb->bflags))
3964                                 clear_extent_buffer_dirty(eb);
3965                         free_extent_buffer_stale(eb);
3966                 }
3967         }
3968 
3969         return ret;
3970 }
3971 
3972 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3973                                        struct extent_io_tree *pinned_extents)
3974 {
3975         struct extent_io_tree *unpin;
3976         u64 start;
3977         u64 end;
3978         int ret;
3979         bool loop = true;
3980 
3981         unpin = pinned_extents;
3982 again:
3983         while (1) {
3984                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3985                                             EXTENT_DIRTY, NULL);
3986                 if (ret)
3987                         break;
3988 
3989                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3990                 btrfs_error_unpin_extent_range(root, start, end);
3991                 cond_resched();
3992         }
3993 
3994         if (loop) {
3995                 if (unpin == &root->fs_info->freed_extents[0])
3996                         unpin = &root->fs_info->freed_extents[1];
3997                 else
3998                         unpin = &root->fs_info->freed_extents[0];
3999                 loop = false;
4000                 goto again;
4001         }
4002 
4003         return 0;
4004 }
4005 
4006 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4007                                    struct btrfs_root *root)
4008 {
4009         btrfs_destroy_ordered_operations(cur_trans, root);
4010 
4011         btrfs_destroy_delayed_refs(cur_trans, root);
4012 
4013         cur_trans->state = TRANS_STATE_COMMIT_START;
4014         wake_up(&root->fs_info->transaction_blocked_wait);
4015 
4016         cur_trans->state = TRANS_STATE_UNBLOCKED;
4017         wake_up(&root->fs_info->transaction_wait);
4018 
4019         btrfs_destroy_delayed_inodes(root);
4020         btrfs_assert_delayed_root_empty(root);
4021 
4022         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4023                                      EXTENT_DIRTY);
4024         btrfs_destroy_pinned_extent(root,
4025                                     root->fs_info->pinned_extents);
4026 
4027         cur_trans->state =TRANS_STATE_COMPLETED;
4028         wake_up(&cur_trans->commit_wait);
4029 
4030         /*
4031         memset(cur_trans, 0, sizeof(*cur_trans));
4032         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4033         */
4034 }
4035 
4036 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4037 {
4038         struct btrfs_transaction *t;
4039 
4040         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4041 
4042         spin_lock(&root->fs_info->trans_lock);
4043         while (!list_empty(&root->fs_info->trans_list)) {
4044                 t = list_first_entry(&root->fs_info->trans_list,
4045                                      struct btrfs_transaction, list);
4046                 if (t->state >= TRANS_STATE_COMMIT_START) {
4047                         atomic_inc(&t->use_count);
4048                         spin_unlock(&root->fs_info->trans_lock);
4049                         btrfs_wait_for_commit(root, t->transid);
4050                         btrfs_put_transaction(t);
4051                         spin_lock(&root->fs_info->trans_lock);
4052                         continue;
4053                 }
4054                 if (t == root->fs_info->running_transaction) {
4055                         t->state = TRANS_STATE_COMMIT_DOING;
4056                         spin_unlock(&root->fs_info->trans_lock);
4057                         /*
4058                          * We wait for 0 num_writers since we don't hold a trans
4059                          * handle open currently for this transaction.
4060                          */
4061                         wait_event(t->writer_wait,
4062                                    atomic_read(&t->num_writers) == 0);
4063                 } else {
4064                         spin_unlock(&root->fs_info->trans_lock);
4065                 }
4066                 btrfs_cleanup_one_transaction(t, root);
4067 
4068                 spin_lock(&root->fs_info->trans_lock);
4069                 if (t == root->fs_info->running_transaction)
4070                         root->fs_info->running_transaction = NULL;
4071                 list_del_init(&t->list);
4072                 spin_unlock(&root->fs_info->trans_lock);
4073 
4074                 btrfs_put_transaction(t);
4075                 trace_btrfs_transaction_commit(root);
4076                 spin_lock(&root->fs_info->trans_lock);
4077         }
4078         spin_unlock(&root->fs_info->trans_lock);
4079         btrfs_destroy_all_ordered_extents(root->fs_info);
4080         btrfs_destroy_delayed_inodes(root);
4081         btrfs_assert_delayed_root_empty(root);
4082         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4083         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4084         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4085 
4086         return 0;
4087 }
4088 
4089 static struct extent_io_ops btree_extent_io_ops = {
4090         .readpage_end_io_hook = btree_readpage_end_io_hook,
4091         .readpage_io_failed_hook = btree_io_failed_hook,
4092         .submit_bio_hook = btree_submit_bio_hook,
4093         /* note we're sharing with inode.c for the merge bio hook */
4094         .merge_bio_hook = btrfs_merge_bio_hook,
4095 };
4096 

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

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

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

osdn.jp