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

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

Version: ~ [ linux-5.4-rc7 ] ~ [ linux-5.3.11 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.84 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.154 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.201 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.201 ] ~ [ 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-3.16.77 ] ~ [ 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-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ 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/crc32c.h>
 30 #include <linux/slab.h>
 31 #include <linux/migrate.h>
 32 #include <linux/ratelimit.h>
 33 #include <asm/unaligned.h>
 34 #include "compat.h"
 35 #include "ctree.h"
 36 #include "disk-io.h"
 37 #include "transaction.h"
 38 #include "btrfs_inode.h"
 39 #include "volumes.h"
 40 #include "print-tree.h"
 41 #include "async-thread.h"
 42 #include "locking.h"
 43 #include "tree-log.h"
 44 #include "free-space-cache.h"
 45 #include "inode-map.h"
 46 
 47 static struct extent_io_ops btree_extent_io_ops;
 48 static void end_workqueue_fn(struct btrfs_work *work);
 49 static void free_fs_root(struct btrfs_root *root);
 50 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
 51                                     int read_only);
 52 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
 53 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
 54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
 55                                       struct btrfs_root *root);
 56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
 57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
 58 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
 59                                         struct extent_io_tree *dirty_pages,
 60                                         int mark);
 61 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
 62                                        struct extent_io_tree *pinned_extents);
 63 static int btrfs_cleanup_transaction(struct btrfs_root *root);
 64 
 65 /*
 66  * end_io_wq structs are used to do processing in task context when an IO is
 67  * complete.  This is used during reads to verify checksums, and it is used
 68  * by writes to insert metadata for new file extents after IO is complete.
 69  */
 70 struct end_io_wq {
 71         struct bio *bio;
 72         bio_end_io_t *end_io;
 73         void *private;
 74         struct btrfs_fs_info *info;
 75         int error;
 76         int metadata;
 77         struct list_head list;
 78         struct btrfs_work work;
 79 };
 80 
 81 /*
 82  * async submit bios are used to offload expensive checksumming
 83  * onto the worker threads.  They checksum file and metadata bios
 84  * just before they are sent down the IO stack.
 85  */
 86 struct async_submit_bio {
 87         struct inode *inode;
 88         struct bio *bio;
 89         struct list_head list;
 90         extent_submit_bio_hook_t *submit_bio_start;
 91         extent_submit_bio_hook_t *submit_bio_done;
 92         int rw;
 93         int mirror_num;
 94         unsigned long bio_flags;
 95         /*
 96          * bio_offset is optional, can be used if the pages in the bio
 97          * can't tell us where in the file the bio should go
 98          */
 99         u64 bio_offset;
100         struct btrfs_work work;
101 };
102 
103 /*
104  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
105  * eb, the lockdep key is determined by the btrfs_root it belongs to and
106  * the level the eb occupies in the tree.
107  *
108  * Different roots are used for different purposes and may nest inside each
109  * other and they require separate keysets.  As lockdep keys should be
110  * static, assign keysets according to the purpose of the root as indicated
111  * by btrfs_root->objectid.  This ensures that all special purpose roots
112  * have separate keysets.
113  *
114  * Lock-nesting across peer nodes is always done with the immediate parent
115  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
116  * subclass to avoid triggering lockdep warning in such cases.
117  *
118  * The key is set by the readpage_end_io_hook after the buffer has passed
119  * csum validation but before the pages are unlocked.  It is also set by
120  * btrfs_init_new_buffer on freshly allocated blocks.
121  *
122  * We also add a check to make sure the highest level of the tree is the
123  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
124  * needs update as well.
125  */
126 #ifdef CONFIG_DEBUG_LOCK_ALLOC
127 # if BTRFS_MAX_LEVEL != 8
128 #  error
129 # endif
130 
131 static struct btrfs_lockdep_keyset {
132         u64                     id;             /* root objectid */
133         const char              *name_stem;     /* lock name stem */
134         char                    names[BTRFS_MAX_LEVEL + 1][20];
135         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
136 } btrfs_lockdep_keysets[] = {
137         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
138         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
139         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
140         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
141         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
142         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
143         { .id = BTRFS_ORPHAN_OBJECTID,          .name_stem = "orphan"   },
144         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
145         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
146         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
147         { .id = 0,                              .name_stem = "tree"     },
148 };
149 
150 void __init btrfs_init_lockdep(void)
151 {
152         int i, j;
153 
154         /* initialize lockdep class names */
155         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
156                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
157 
158                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
159                         snprintf(ks->names[j], sizeof(ks->names[j]),
160                                  "btrfs-%s-%02d", ks->name_stem, j);
161         }
162 }
163 
164 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
165                                     int level)
166 {
167         struct btrfs_lockdep_keyset *ks;
168 
169         BUG_ON(level >= ARRAY_SIZE(ks->keys));
170 
171         /* find the matching keyset, id 0 is the default entry */
172         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
173                 if (ks->id == objectid)
174                         break;
175 
176         lockdep_set_class_and_name(&eb->lock,
177                                    &ks->keys[level], ks->names[level]);
178 }
179 
180 #endif
181 
182 /*
183  * extents on the btree inode are pretty simple, there's one extent
184  * that covers the entire device
185  */
186 static struct extent_map *btree_get_extent(struct inode *inode,
187                 struct page *page, size_t pg_offset, u64 start, u64 len,
188                 int create)
189 {
190         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
191         struct extent_map *em;
192         int ret;
193 
194         read_lock(&em_tree->lock);
195         em = lookup_extent_mapping(em_tree, start, len);
196         if (em) {
197                 em->bdev =
198                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
199                 read_unlock(&em_tree->lock);
200                 goto out;
201         }
202         read_unlock(&em_tree->lock);
203 
204         em = alloc_extent_map();
205         if (!em) {
206                 em = ERR_PTR(-ENOMEM);
207                 goto out;
208         }
209         em->start = 0;
210         em->len = (u64)-1;
211         em->block_len = (u64)-1;
212         em->block_start = 0;
213         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
214 
215         write_lock(&em_tree->lock);
216         ret = add_extent_mapping(em_tree, em);
217         if (ret == -EEXIST) {
218                 u64 failed_start = em->start;
219                 u64 failed_len = em->len;
220 
221                 free_extent_map(em);
222                 em = lookup_extent_mapping(em_tree, start, len);
223                 if (em) {
224                         ret = 0;
225                 } else {
226                         em = lookup_extent_mapping(em_tree, failed_start,
227                                                    failed_len);
228                         ret = -EIO;
229                 }
230         } else if (ret) {
231                 free_extent_map(em);
232                 em = NULL;
233         }
234         write_unlock(&em_tree->lock);
235 
236         if (ret)
237                 em = ERR_PTR(ret);
238 out:
239         return em;
240 }
241 
242 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
243 {
244         return crc32c(seed, data, len);
245 }
246 
247 void btrfs_csum_final(u32 crc, char *result)
248 {
249         put_unaligned_le32(~crc, result);
250 }
251 
252 /*
253  * compute the csum for a btree block, and either verify it or write it
254  * into the csum field of the block.
255  */
256 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
257                            int verify)
258 {
259         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
260         char *result = NULL;
261         unsigned long len;
262         unsigned long cur_len;
263         unsigned long offset = BTRFS_CSUM_SIZE;
264         char *kaddr;
265         unsigned long map_start;
266         unsigned long map_len;
267         int err;
268         u32 crc = ~(u32)0;
269         unsigned long inline_result;
270 
271         len = buf->len - offset;
272         while (len > 0) {
273                 err = map_private_extent_buffer(buf, offset, 32,
274                                         &kaddr, &map_start, &map_len);
275                 if (err)
276                         return 1;
277                 cur_len = min(len, map_len - (offset - map_start));
278                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
279                                       crc, cur_len);
280                 len -= cur_len;
281                 offset += cur_len;
282         }
283         if (csum_size > sizeof(inline_result)) {
284                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
285                 if (!result)
286                         return 1;
287         } else {
288                 result = (char *)&inline_result;
289         }
290 
291         btrfs_csum_final(crc, result);
292 
293         if (verify) {
294                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
295                         u32 val;
296                         u32 found = 0;
297                         memcpy(&found, result, csum_size);
298 
299                         read_extent_buffer(buf, &val, 0, csum_size);
300                         printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
301                                        "failed on %llu wanted %X found %X "
302                                        "level %d\n",
303                                        root->fs_info->sb->s_id,
304                                        (unsigned long long)buf->start, val, found,
305                                        btrfs_header_level(buf));
306                         if (result != (char *)&inline_result)
307                                 kfree(result);
308                         return 1;
309                 }
310         } else {
311                 write_extent_buffer(buf, result, 0, csum_size);
312         }
313         if (result != (char *)&inline_result)
314                 kfree(result);
315         return 0;
316 }
317 
318 /*
319  * we can't consider a given block up to date unless the transid of the
320  * block matches the transid in the parent node's pointer.  This is how we
321  * detect blocks that either didn't get written at all or got written
322  * in the wrong place.
323  */
324 static int verify_parent_transid(struct extent_io_tree *io_tree,
325                                  struct extent_buffer *eb, u64 parent_transid)
326 {
327         struct extent_state *cached_state = NULL;
328         int ret;
329 
330         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
331                 return 0;
332 
333         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
334                          0, &cached_state, GFP_NOFS);
335         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
336             btrfs_header_generation(eb) == parent_transid) {
337                 ret = 0;
338                 goto out;
339         }
340         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
341                        "found %llu\n",
342                        (unsigned long long)eb->start,
343                        (unsigned long long)parent_transid,
344                        (unsigned long long)btrfs_header_generation(eb));
345         ret = 1;
346         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
347 out:
348         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
349                              &cached_state, GFP_NOFS);
350         return ret;
351 }
352 
353 /*
354  * helper to read a given tree block, doing retries as required when
355  * the checksums don't match and we have alternate mirrors to try.
356  */
357 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
358                                           struct extent_buffer *eb,
359                                           u64 start, u64 parent_transid)
360 {
361         struct extent_io_tree *io_tree;
362         int ret;
363         int num_copies = 0;
364         int mirror_num = 0;
365 
366         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
367         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
368         while (1) {
369                 ret = read_extent_buffer_pages(io_tree, eb, start,
370                                                WAIT_COMPLETE,
371                                                btree_get_extent, mirror_num);
372                 if (!ret &&
373                     !verify_parent_transid(io_tree, eb, parent_transid))
374                         return ret;
375 
376                 /*
377                  * This buffer's crc is fine, but its contents are corrupted, so
378                  * there is no reason to read the other copies, they won't be
379                  * any less wrong.
380                  */
381                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
382                         return ret;
383 
384                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
385                                               eb->start, eb->len);
386                 if (num_copies == 1)
387                         return ret;
388 
389                 mirror_num++;
390                 if (mirror_num > num_copies)
391                         return ret;
392         }
393         return -EIO;
394 }
395 
396 /*
397  * checksum a dirty tree block before IO.  This has extra checks to make sure
398  * we only fill in the checksum field in the first page of a multi-page block
399  */
400 
401 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
402 {
403         struct extent_io_tree *tree;
404         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
405         u64 found_start;
406         unsigned long len;
407         struct extent_buffer *eb;
408         int ret;
409 
410         tree = &BTRFS_I(page->mapping->host)->io_tree;
411 
412         if (page->private == EXTENT_PAGE_PRIVATE) {
413                 WARN_ON(1);
414                 goto out;
415         }
416         if (!page->private) {
417                 WARN_ON(1);
418                 goto out;
419         }
420         len = page->private >> 2;
421         WARN_ON(len == 0);
422 
423         eb = alloc_extent_buffer(tree, start, len, page);
424         if (eb == NULL) {
425                 WARN_ON(1);
426                 goto out;
427         }
428         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
429                                              btrfs_header_generation(eb));
430         BUG_ON(ret);
431         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
432 
433         found_start = btrfs_header_bytenr(eb);
434         if (found_start != start) {
435                 WARN_ON(1);
436                 goto err;
437         }
438         if (eb->first_page != page) {
439                 WARN_ON(1);
440                 goto err;
441         }
442         if (!PageUptodate(page)) {
443                 WARN_ON(1);
444                 goto err;
445         }
446         csum_tree_block(root, eb, 0);
447 err:
448         free_extent_buffer(eb);
449 out:
450         return 0;
451 }
452 
453 static int check_tree_block_fsid(struct btrfs_root *root,
454                                  struct extent_buffer *eb)
455 {
456         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
457         u8 fsid[BTRFS_UUID_SIZE];
458         int ret = 1;
459 
460         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
461                            BTRFS_FSID_SIZE);
462         while (fs_devices) {
463                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
464                         ret = 0;
465                         break;
466                 }
467                 fs_devices = fs_devices->seed;
468         }
469         return ret;
470 }
471 
472 #define CORRUPT(reason, eb, root, slot)                         \
473         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
474                "root=%llu, slot=%d\n", reason,                  \
475                (unsigned long long)btrfs_header_bytenr(eb),     \
476                (unsigned long long)root->objectid, slot)
477 
478 static noinline int check_leaf(struct btrfs_root *root,
479                                struct extent_buffer *leaf)
480 {
481         struct btrfs_key key;
482         struct btrfs_key leaf_key;
483         u32 nritems = btrfs_header_nritems(leaf);
484         int slot;
485 
486         if (nritems == 0)
487                 return 0;
488 
489         /* Check the 0 item */
490         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
491             BTRFS_LEAF_DATA_SIZE(root)) {
492                 CORRUPT("invalid item offset size pair", leaf, root, 0);
493                 return -EIO;
494         }
495 
496         /*
497          * Check to make sure each items keys are in the correct order and their
498          * offsets make sense.  We only have to loop through nritems-1 because
499          * we check the current slot against the next slot, which verifies the
500          * next slot's offset+size makes sense and that the current's slot
501          * offset is correct.
502          */
503         for (slot = 0; slot < nritems - 1; slot++) {
504                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
505                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
506 
507                 /* Make sure the keys are in the right order */
508                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
509                         CORRUPT("bad key order", leaf, root, slot);
510                         return -EIO;
511                 }
512 
513                 /*
514                  * Make sure the offset and ends are right, remember that the
515                  * item data starts at the end of the leaf and grows towards the
516                  * front.
517                  */
518                 if (btrfs_item_offset_nr(leaf, slot) !=
519                         btrfs_item_end_nr(leaf, slot + 1)) {
520                         CORRUPT("slot offset bad", leaf, root, slot);
521                         return -EIO;
522                 }
523 
524                 /*
525                  * Check to make sure that we don't point outside of the leaf,
526                  * just incase all the items are consistent to eachother, but
527                  * all point outside of the leaf.
528                  */
529                 if (btrfs_item_end_nr(leaf, slot) >
530                     BTRFS_LEAF_DATA_SIZE(root)) {
531                         CORRUPT("slot end outside of leaf", leaf, root, slot);
532                         return -EIO;
533                 }
534         }
535 
536         return 0;
537 }
538 
539 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
540                                struct extent_state *state)
541 {
542         struct extent_io_tree *tree;
543         u64 found_start;
544         int found_level;
545         unsigned long len;
546         struct extent_buffer *eb;
547         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
548         int ret = 0;
549 
550         tree = &BTRFS_I(page->mapping->host)->io_tree;
551         if (page->private == EXTENT_PAGE_PRIVATE)
552                 goto out;
553         if (!page->private)
554                 goto out;
555 
556         len = page->private >> 2;
557         WARN_ON(len == 0);
558 
559         eb = alloc_extent_buffer(tree, start, len, page);
560         if (eb == NULL) {
561                 ret = -EIO;
562                 goto out;
563         }
564 
565         found_start = btrfs_header_bytenr(eb);
566         if (found_start != start) {
567                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
568                                "%llu %llu\n",
569                                (unsigned long long)found_start,
570                                (unsigned long long)eb->start);
571                 ret = -EIO;
572                 goto err;
573         }
574         if (eb->first_page != page) {
575                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
576                        eb->first_page->index, page->index);
577                 WARN_ON(1);
578                 ret = -EIO;
579                 goto err;
580         }
581         if (check_tree_block_fsid(root, eb)) {
582                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
583                                (unsigned long long)eb->start);
584                 ret = -EIO;
585                 goto err;
586         }
587         found_level = btrfs_header_level(eb);
588 
589         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
590                                        eb, found_level);
591 
592         ret = csum_tree_block(root, eb, 1);
593         if (ret) {
594                 ret = -EIO;
595                 goto err;
596         }
597 
598         /*
599          * If this is a leaf block and it is corrupt, set the corrupt bit so
600          * that we don't try and read the other copies of this block, just
601          * return -EIO.
602          */
603         if (found_level == 0 && check_leaf(root, eb)) {
604                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
605                 ret = -EIO;
606         }
607 
608         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
609         end = eb->start + end - 1;
610 err:
611         if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
612                 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
613                 btree_readahead_hook(root, eb, eb->start, ret);
614         }
615 
616         free_extent_buffer(eb);
617 out:
618         return ret;
619 }
620 
621 static int btree_io_failed_hook(struct bio *failed_bio,
622                          struct page *page, u64 start, u64 end,
623                          int mirror_num, struct extent_state *state)
624 {
625         struct extent_io_tree *tree;
626         unsigned long len;
627         struct extent_buffer *eb;
628         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
629 
630         tree = &BTRFS_I(page->mapping->host)->io_tree;
631         if (page->private == EXTENT_PAGE_PRIVATE)
632                 goto out;
633         if (!page->private)
634                 goto out;
635 
636         len = page->private >> 2;
637         WARN_ON(len == 0);
638 
639         eb = alloc_extent_buffer(tree, start, len, page);
640         if (eb == NULL)
641                 goto out;
642 
643         if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
644                 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
645                 btree_readahead_hook(root, eb, eb->start, -EIO);
646         }
647         free_extent_buffer(eb);
648 
649 out:
650         return -EIO;    /* we fixed nothing */
651 }
652 
653 static void end_workqueue_bio(struct bio *bio, int err)
654 {
655         struct end_io_wq *end_io_wq = bio->bi_private;
656         struct btrfs_fs_info *fs_info;
657 
658         fs_info = end_io_wq->info;
659         end_io_wq->error = err;
660         end_io_wq->work.func = end_workqueue_fn;
661         end_io_wq->work.flags = 0;
662 
663         if (bio->bi_rw & REQ_WRITE) {
664                 if (end_io_wq->metadata == 1)
665                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
666                                            &end_io_wq->work);
667                 else if (end_io_wq->metadata == 2)
668                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
669                                            &end_io_wq->work);
670                 else
671                         btrfs_queue_worker(&fs_info->endio_write_workers,
672                                            &end_io_wq->work);
673         } else {
674                 if (end_io_wq->metadata)
675                         btrfs_queue_worker(&fs_info->endio_meta_workers,
676                                            &end_io_wq->work);
677                 else
678                         btrfs_queue_worker(&fs_info->endio_workers,
679                                            &end_io_wq->work);
680         }
681 }
682 
683 /*
684  * For the metadata arg you want
685  *
686  * 0 - if data
687  * 1 - if normal metadta
688  * 2 - if writing to the free space cache area
689  */
690 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
691                         int metadata)
692 {
693         struct end_io_wq *end_io_wq;
694         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
695         if (!end_io_wq)
696                 return -ENOMEM;
697 
698         end_io_wq->private = bio->bi_private;
699         end_io_wq->end_io = bio->bi_end_io;
700         end_io_wq->info = info;
701         end_io_wq->error = 0;
702         end_io_wq->bio = bio;
703         end_io_wq->metadata = metadata;
704 
705         bio->bi_private = end_io_wq;
706         bio->bi_end_io = end_workqueue_bio;
707         return 0;
708 }
709 
710 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
711 {
712         unsigned long limit = min_t(unsigned long,
713                                     info->workers.max_workers,
714                                     info->fs_devices->open_devices);
715         return 256 * limit;
716 }
717 
718 static void run_one_async_start(struct btrfs_work *work)
719 {
720         struct async_submit_bio *async;
721 
722         async = container_of(work, struct  async_submit_bio, work);
723         async->submit_bio_start(async->inode, async->rw, async->bio,
724                                async->mirror_num, async->bio_flags,
725                                async->bio_offset);
726 }
727 
728 static void run_one_async_done(struct btrfs_work *work)
729 {
730         struct btrfs_fs_info *fs_info;
731         struct async_submit_bio *async;
732         int limit;
733 
734         async = container_of(work, struct  async_submit_bio, work);
735         fs_info = BTRFS_I(async->inode)->root->fs_info;
736 
737         limit = btrfs_async_submit_limit(fs_info);
738         limit = limit * 2 / 3;
739 
740         atomic_dec(&fs_info->nr_async_submits);
741 
742         if (atomic_read(&fs_info->nr_async_submits) < limit &&
743             waitqueue_active(&fs_info->async_submit_wait))
744                 wake_up(&fs_info->async_submit_wait);
745 
746         async->submit_bio_done(async->inode, async->rw, async->bio,
747                                async->mirror_num, async->bio_flags,
748                                async->bio_offset);
749 }
750 
751 static void run_one_async_free(struct btrfs_work *work)
752 {
753         struct async_submit_bio *async;
754 
755         async = container_of(work, struct  async_submit_bio, work);
756         kfree(async);
757 }
758 
759 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
760                         int rw, struct bio *bio, int mirror_num,
761                         unsigned long bio_flags,
762                         u64 bio_offset,
763                         extent_submit_bio_hook_t *submit_bio_start,
764                         extent_submit_bio_hook_t *submit_bio_done)
765 {
766         struct async_submit_bio *async;
767 
768         async = kmalloc(sizeof(*async), GFP_NOFS);
769         if (!async)
770                 return -ENOMEM;
771 
772         async->inode = inode;
773         async->rw = rw;
774         async->bio = bio;
775         async->mirror_num = mirror_num;
776         async->submit_bio_start = submit_bio_start;
777         async->submit_bio_done = submit_bio_done;
778 
779         async->work.func = run_one_async_start;
780         async->work.ordered_func = run_one_async_done;
781         async->work.ordered_free = run_one_async_free;
782 
783         async->work.flags = 0;
784         async->bio_flags = bio_flags;
785         async->bio_offset = bio_offset;
786 
787         atomic_inc(&fs_info->nr_async_submits);
788 
789         if (rw & REQ_SYNC)
790                 btrfs_set_work_high_prio(&async->work);
791 
792         btrfs_queue_worker(&fs_info->workers, &async->work);
793 
794         while (atomic_read(&fs_info->async_submit_draining) &&
795               atomic_read(&fs_info->nr_async_submits)) {
796                 wait_event(fs_info->async_submit_wait,
797                            (atomic_read(&fs_info->nr_async_submits) == 0));
798         }
799 
800         return 0;
801 }
802 
803 static int btree_csum_one_bio(struct bio *bio)
804 {
805         struct bio_vec *bvec = bio->bi_io_vec;
806         int bio_index = 0;
807         struct btrfs_root *root;
808 
809         WARN_ON(bio->bi_vcnt <= 0);
810         while (bio_index < bio->bi_vcnt) {
811                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
812                 csum_dirty_buffer(root, bvec->bv_page);
813                 bio_index++;
814                 bvec++;
815         }
816         return 0;
817 }
818 
819 static int __btree_submit_bio_start(struct inode *inode, int rw,
820                                     struct bio *bio, int mirror_num,
821                                     unsigned long bio_flags,
822                                     u64 bio_offset)
823 {
824         /*
825          * when we're called for a write, we're already in the async
826          * submission context.  Just jump into btrfs_map_bio
827          */
828         btree_csum_one_bio(bio);
829         return 0;
830 }
831 
832 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
833                                  int mirror_num, unsigned long bio_flags,
834                                  u64 bio_offset)
835 {
836         /*
837          * when we're called for a write, we're already in the async
838          * submission context.  Just jump into btrfs_map_bio
839          */
840         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
841 }
842 
843 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
844                                  int mirror_num, unsigned long bio_flags,
845                                  u64 bio_offset)
846 {
847         int ret;
848 
849         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
850                                           bio, 1);
851         BUG_ON(ret);
852 
853         if (!(rw & REQ_WRITE)) {
854                 /*
855                  * called for a read, do the setup so that checksum validation
856                  * can happen in the async kernel threads
857                  */
858                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
859                                      mirror_num, 0);
860         }
861 
862         /*
863          * kthread helpers are used to submit writes so that checksumming
864          * can happen in parallel across all CPUs
865          */
866         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
867                                    inode, rw, bio, mirror_num, 0,
868                                    bio_offset,
869                                    __btree_submit_bio_start,
870                                    __btree_submit_bio_done);
871 }
872 
873 #ifdef CONFIG_MIGRATION
874 static int btree_migratepage(struct address_space *mapping,
875                         struct page *newpage, struct page *page,
876                         enum migrate_mode mode)
877 {
878         /*
879          * we can't safely write a btree page from here,
880          * we haven't done the locking hook
881          */
882         if (PageDirty(page))
883                 return -EAGAIN;
884         /*
885          * Buffers may be managed in a filesystem specific way.
886          * We must have no buffers or drop them.
887          */
888         if (page_has_private(page) &&
889             !try_to_release_page(page, GFP_KERNEL))
890                 return -EAGAIN;
891         return migrate_page(mapping, newpage, page, mode);
892 }
893 #endif
894 
895 static int btree_writepage(struct page *page, struct writeback_control *wbc)
896 {
897         struct extent_io_tree *tree;
898         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
899         struct extent_buffer *eb;
900         int was_dirty;
901 
902         tree = &BTRFS_I(page->mapping->host)->io_tree;
903         if (!(current->flags & PF_MEMALLOC)) {
904                 return extent_write_full_page(tree, page,
905                                               btree_get_extent, wbc);
906         }
907 
908         redirty_page_for_writepage(wbc, page);
909         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
910         WARN_ON(!eb);
911 
912         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
913         if (!was_dirty) {
914                 spin_lock(&root->fs_info->delalloc_lock);
915                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
916                 spin_unlock(&root->fs_info->delalloc_lock);
917         }
918         free_extent_buffer(eb);
919 
920         unlock_page(page);
921         return 0;
922 }
923 
924 static int btree_writepages(struct address_space *mapping,
925                             struct writeback_control *wbc)
926 {
927         struct extent_io_tree *tree;
928         tree = &BTRFS_I(mapping->host)->io_tree;
929         if (wbc->sync_mode == WB_SYNC_NONE) {
930                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
931                 u64 num_dirty;
932                 unsigned long thresh = 32 * 1024 * 1024;
933 
934                 if (wbc->for_kupdate)
935                         return 0;
936 
937                 /* this is a bit racy, but that's ok */
938                 num_dirty = root->fs_info->dirty_metadata_bytes;
939                 if (num_dirty < thresh)
940                         return 0;
941         }
942         return extent_writepages(tree, mapping, btree_get_extent, wbc);
943 }
944 
945 static int btree_readpage(struct file *file, struct page *page)
946 {
947         struct extent_io_tree *tree;
948         tree = &BTRFS_I(page->mapping->host)->io_tree;
949         return extent_read_full_page(tree, page, btree_get_extent, 0);
950 }
951 
952 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
953 {
954         struct extent_io_tree *tree;
955         struct extent_map_tree *map;
956         int ret;
957 
958         if (PageWriteback(page) || PageDirty(page))
959                 return 0;
960 
961         tree = &BTRFS_I(page->mapping->host)->io_tree;
962         map = &BTRFS_I(page->mapping->host)->extent_tree;
963 
964         ret = try_release_extent_state(map, tree, page, gfp_flags);
965         if (!ret)
966                 return 0;
967 
968         ret = try_release_extent_buffer(tree, page);
969         if (ret == 1) {
970                 ClearPagePrivate(page);
971                 set_page_private(page, 0);
972                 page_cache_release(page);
973         }
974 
975         return ret;
976 }
977 
978 static void btree_invalidatepage(struct page *page, unsigned long offset)
979 {
980         struct extent_io_tree *tree;
981         tree = &BTRFS_I(page->mapping->host)->io_tree;
982         extent_invalidatepage(tree, page, offset);
983         btree_releasepage(page, GFP_NOFS);
984         if (PagePrivate(page)) {
985                 printk(KERN_WARNING "btrfs warning page private not zero "
986                        "on page %llu\n", (unsigned long long)page_offset(page));
987                 ClearPagePrivate(page);
988                 set_page_private(page, 0);
989                 page_cache_release(page);
990         }
991 }
992 
993 static const struct address_space_operations btree_aops = {
994         .readpage       = btree_readpage,
995         .writepage      = btree_writepage,
996         .writepages     = btree_writepages,
997         .releasepage    = btree_releasepage,
998         .invalidatepage = btree_invalidatepage,
999 #ifdef CONFIG_MIGRATION
1000         .migratepage    = btree_migratepage,
1001 #endif
1002 };
1003 
1004 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1005                          u64 parent_transid)
1006 {
1007         struct extent_buffer *buf = NULL;
1008         struct inode *btree_inode = root->fs_info->btree_inode;
1009         int ret = 0;
1010 
1011         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1012         if (!buf)
1013                 return 0;
1014         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1015                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1016         free_extent_buffer(buf);
1017         return ret;
1018 }
1019 
1020 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1021                          int mirror_num, struct extent_buffer **eb)
1022 {
1023         struct extent_buffer *buf = NULL;
1024         struct inode *btree_inode = root->fs_info->btree_inode;
1025         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1026         int ret;
1027 
1028         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1029         if (!buf)
1030                 return 0;
1031 
1032         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1033 
1034         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1035                                        btree_get_extent, mirror_num);
1036         if (ret) {
1037                 free_extent_buffer(buf);
1038                 return ret;
1039         }
1040 
1041         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1042                 free_extent_buffer(buf);
1043                 return -EIO;
1044         } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1045                 *eb = buf;
1046         } else {
1047                 free_extent_buffer(buf);
1048         }
1049         return 0;
1050 }
1051 
1052 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1053                                             u64 bytenr, u32 blocksize)
1054 {
1055         struct inode *btree_inode = root->fs_info->btree_inode;
1056         struct extent_buffer *eb;
1057         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1058                                 bytenr, blocksize);
1059         return eb;
1060 }
1061 
1062 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1063                                                  u64 bytenr, u32 blocksize)
1064 {
1065         struct inode *btree_inode = root->fs_info->btree_inode;
1066         struct extent_buffer *eb;
1067 
1068         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1069                                  bytenr, blocksize, NULL);
1070         return eb;
1071 }
1072 
1073 
1074 int btrfs_write_tree_block(struct extent_buffer *buf)
1075 {
1076         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1077                                         buf->start + buf->len - 1);
1078 }
1079 
1080 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1081 {
1082         return filemap_fdatawait_range(buf->first_page->mapping,
1083                                        buf->start, buf->start + buf->len - 1);
1084 }
1085 
1086 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1087                                       u32 blocksize, u64 parent_transid)
1088 {
1089         struct extent_buffer *buf = NULL;
1090         int ret;
1091 
1092         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1093         if (!buf)
1094                 return NULL;
1095 
1096         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1097 
1098         if (ret == 0)
1099                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1100         return buf;
1101 
1102 }
1103 
1104 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1105                      struct extent_buffer *buf)
1106 {
1107         struct inode *btree_inode = root->fs_info->btree_inode;
1108         if (btrfs_header_generation(buf) ==
1109             root->fs_info->running_transaction->transid) {
1110                 btrfs_assert_tree_locked(buf);
1111 
1112                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1113                         spin_lock(&root->fs_info->delalloc_lock);
1114                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1115                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1116                         else
1117                                 WARN_ON(1);
1118                         spin_unlock(&root->fs_info->delalloc_lock);
1119                 }
1120 
1121                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1122                 btrfs_set_lock_blocking(buf);
1123                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1124                                           buf);
1125         }
1126         return 0;
1127 }
1128 
1129 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1130                         u32 stripesize, struct btrfs_root *root,
1131                         struct btrfs_fs_info *fs_info,
1132                         u64 objectid)
1133 {
1134         root->node = NULL;
1135         root->commit_root = NULL;
1136         root->sectorsize = sectorsize;
1137         root->nodesize = nodesize;
1138         root->leafsize = leafsize;
1139         root->stripesize = stripesize;
1140         root->ref_cows = 0;
1141         root->track_dirty = 0;
1142         root->in_radix = 0;
1143         root->orphan_item_inserted = 0;
1144         root->orphan_cleanup_state = 0;
1145 
1146         root->fs_info = fs_info;
1147         root->objectid = objectid;
1148         root->last_trans = 0;
1149         root->highest_objectid = 0;
1150         root->name = NULL;
1151         root->inode_tree = RB_ROOT;
1152         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1153         root->block_rsv = NULL;
1154         root->orphan_block_rsv = NULL;
1155 
1156         INIT_LIST_HEAD(&root->dirty_list);
1157         INIT_LIST_HEAD(&root->orphan_list);
1158         INIT_LIST_HEAD(&root->root_list);
1159         spin_lock_init(&root->orphan_lock);
1160         spin_lock_init(&root->inode_lock);
1161         spin_lock_init(&root->accounting_lock);
1162         mutex_init(&root->objectid_mutex);
1163         mutex_init(&root->log_mutex);
1164         init_waitqueue_head(&root->log_writer_wait);
1165         init_waitqueue_head(&root->log_commit_wait[0]);
1166         init_waitqueue_head(&root->log_commit_wait[1]);
1167         atomic_set(&root->log_commit[0], 0);
1168         atomic_set(&root->log_commit[1], 0);
1169         atomic_set(&root->log_writers, 0);
1170         root->log_batch = 0;
1171         root->log_transid = 0;
1172         root->last_log_commit = 0;
1173         extent_io_tree_init(&root->dirty_log_pages,
1174                              fs_info->btree_inode->i_mapping);
1175 
1176         memset(&root->root_key, 0, sizeof(root->root_key));
1177         memset(&root->root_item, 0, sizeof(root->root_item));
1178         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1179         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1180         root->defrag_trans_start = fs_info->generation;
1181         init_completion(&root->kobj_unregister);
1182         root->defrag_running = 0;
1183         root->root_key.objectid = objectid;
1184         root->anon_dev = 0;
1185         return 0;
1186 }
1187 
1188 static int find_and_setup_root(struct btrfs_root *tree_root,
1189                                struct btrfs_fs_info *fs_info,
1190                                u64 objectid,
1191                                struct btrfs_root *root)
1192 {
1193         int ret;
1194         u32 blocksize;
1195         u64 generation;
1196 
1197         __setup_root(tree_root->nodesize, tree_root->leafsize,
1198                      tree_root->sectorsize, tree_root->stripesize,
1199                      root, fs_info, objectid);
1200         ret = btrfs_find_last_root(tree_root, objectid,
1201                                    &root->root_item, &root->root_key);
1202         if (ret > 0)
1203                 return -ENOENT;
1204         BUG_ON(ret);
1205 
1206         generation = btrfs_root_generation(&root->root_item);
1207         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1208         root->commit_root = NULL;
1209         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1210                                      blocksize, generation);
1211         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1212                 free_extent_buffer(root->node);
1213                 root->node = NULL;
1214                 return -EIO;
1215         }
1216         root->commit_root = btrfs_root_node(root);
1217         return 0;
1218 }
1219 
1220 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1221                                          struct btrfs_fs_info *fs_info)
1222 {
1223         struct btrfs_root *root;
1224         struct btrfs_root *tree_root = fs_info->tree_root;
1225         struct extent_buffer *leaf;
1226 
1227         root = kzalloc(sizeof(*root), GFP_NOFS);
1228         if (!root)
1229                 return ERR_PTR(-ENOMEM);
1230 
1231         __setup_root(tree_root->nodesize, tree_root->leafsize,
1232                      tree_root->sectorsize, tree_root->stripesize,
1233                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1234 
1235         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1236         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1237         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1238         /*
1239          * log trees do not get reference counted because they go away
1240          * before a real commit is actually done.  They do store pointers
1241          * to file data extents, and those reference counts still get
1242          * updated (along with back refs to the log tree).
1243          */
1244         root->ref_cows = 0;
1245 
1246         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1247                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1248         if (IS_ERR(leaf)) {
1249                 kfree(root);
1250                 return ERR_CAST(leaf);
1251         }
1252 
1253         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1254         btrfs_set_header_bytenr(leaf, leaf->start);
1255         btrfs_set_header_generation(leaf, trans->transid);
1256         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1257         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1258         root->node = leaf;
1259 
1260         write_extent_buffer(root->node, root->fs_info->fsid,
1261                             (unsigned long)btrfs_header_fsid(root->node),
1262                             BTRFS_FSID_SIZE);
1263         btrfs_mark_buffer_dirty(root->node);
1264         btrfs_tree_unlock(root->node);
1265         return root;
1266 }
1267 
1268 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1269                              struct btrfs_fs_info *fs_info)
1270 {
1271         struct btrfs_root *log_root;
1272 
1273         log_root = alloc_log_tree(trans, fs_info);
1274         if (IS_ERR(log_root))
1275                 return PTR_ERR(log_root);
1276         WARN_ON(fs_info->log_root_tree);
1277         fs_info->log_root_tree = log_root;
1278         return 0;
1279 }
1280 
1281 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1282                        struct btrfs_root *root)
1283 {
1284         struct btrfs_root *log_root;
1285         struct btrfs_inode_item *inode_item;
1286 
1287         log_root = alloc_log_tree(trans, root->fs_info);
1288         if (IS_ERR(log_root))
1289                 return PTR_ERR(log_root);
1290 
1291         log_root->last_trans = trans->transid;
1292         log_root->root_key.offset = root->root_key.objectid;
1293 
1294         inode_item = &log_root->root_item.inode;
1295         inode_item->generation = cpu_to_le64(1);
1296         inode_item->size = cpu_to_le64(3);
1297         inode_item->nlink = cpu_to_le32(1);
1298         inode_item->nbytes = cpu_to_le64(root->leafsize);
1299         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1300 
1301         btrfs_set_root_node(&log_root->root_item, log_root->node);
1302 
1303         WARN_ON(root->log_root);
1304         root->log_root = log_root;
1305         root->log_transid = 0;
1306         root->last_log_commit = 0;
1307         return 0;
1308 }
1309 
1310 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1311                                                struct btrfs_key *location)
1312 {
1313         struct btrfs_root *root;
1314         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1315         struct btrfs_path *path;
1316         struct extent_buffer *l;
1317         u64 generation;
1318         u32 blocksize;
1319         int ret = 0;
1320 
1321         root = kzalloc(sizeof(*root), GFP_NOFS);
1322         if (!root)
1323                 return ERR_PTR(-ENOMEM);
1324         if (location->offset == (u64)-1) {
1325                 ret = find_and_setup_root(tree_root, fs_info,
1326                                           location->objectid, root);
1327                 if (ret) {
1328                         kfree(root);
1329                         return ERR_PTR(ret);
1330                 }
1331                 goto out;
1332         }
1333 
1334         __setup_root(tree_root->nodesize, tree_root->leafsize,
1335                      tree_root->sectorsize, tree_root->stripesize,
1336                      root, fs_info, location->objectid);
1337 
1338         path = btrfs_alloc_path();
1339         if (!path) {
1340                 kfree(root);
1341                 return ERR_PTR(-ENOMEM);
1342         }
1343         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1344         if (ret == 0) {
1345                 l = path->nodes[0];
1346                 read_extent_buffer(l, &root->root_item,
1347                                 btrfs_item_ptr_offset(l, path->slots[0]),
1348                                 sizeof(root->root_item));
1349                 memcpy(&root->root_key, location, sizeof(*location));
1350         }
1351         btrfs_free_path(path);
1352         if (ret) {
1353                 kfree(root);
1354                 if (ret > 0)
1355                         ret = -ENOENT;
1356                 return ERR_PTR(ret);
1357         }
1358 
1359         generation = btrfs_root_generation(&root->root_item);
1360         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1361         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1362                                      blocksize, generation);
1363         root->commit_root = btrfs_root_node(root);
1364         BUG_ON(!root->node);
1365 out:
1366         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1367                 root->ref_cows = 1;
1368                 btrfs_check_and_init_root_item(&root->root_item);
1369         }
1370 
1371         return root;
1372 }
1373 
1374 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1375                                               struct btrfs_key *location)
1376 {
1377         struct btrfs_root *root;
1378         int ret;
1379 
1380         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1381                 return fs_info->tree_root;
1382         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1383                 return fs_info->extent_root;
1384         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1385                 return fs_info->chunk_root;
1386         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1387                 return fs_info->dev_root;
1388         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1389                 return fs_info->csum_root;
1390 again:
1391         spin_lock(&fs_info->fs_roots_radix_lock);
1392         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1393                                  (unsigned long)location->objectid);
1394         spin_unlock(&fs_info->fs_roots_radix_lock);
1395         if (root)
1396                 return root;
1397 
1398         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1399         if (IS_ERR(root))
1400                 return root;
1401 
1402         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1403         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1404                                         GFP_NOFS);
1405         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1406                 ret = -ENOMEM;
1407                 goto fail;
1408         }
1409 
1410         btrfs_init_free_ino_ctl(root);
1411         mutex_init(&root->fs_commit_mutex);
1412         spin_lock_init(&root->cache_lock);
1413         init_waitqueue_head(&root->cache_wait);
1414 
1415         ret = get_anon_bdev(&root->anon_dev);
1416         if (ret)
1417                 goto fail;
1418 
1419         if (btrfs_root_refs(&root->root_item) == 0) {
1420                 ret = -ENOENT;
1421                 goto fail;
1422         }
1423 
1424         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1425         if (ret < 0)
1426                 goto fail;
1427         if (ret == 0)
1428                 root->orphan_item_inserted = 1;
1429 
1430         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1431         if (ret)
1432                 goto fail;
1433 
1434         spin_lock(&fs_info->fs_roots_radix_lock);
1435         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1436                                 (unsigned long)root->root_key.objectid,
1437                                 root);
1438         if (ret == 0)
1439                 root->in_radix = 1;
1440 
1441         spin_unlock(&fs_info->fs_roots_radix_lock);
1442         radix_tree_preload_end();
1443         if (ret) {
1444                 if (ret == -EEXIST) {
1445                         free_fs_root(root);
1446                         goto again;
1447                 }
1448                 goto fail;
1449         }
1450 
1451         ret = btrfs_find_dead_roots(fs_info->tree_root,
1452                                     root->root_key.objectid);
1453         WARN_ON(ret);
1454         return root;
1455 fail:
1456         free_fs_root(root);
1457         return ERR_PTR(ret);
1458 }
1459 
1460 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1461 {
1462         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1463         int ret = 0;
1464         struct btrfs_device *device;
1465         struct backing_dev_info *bdi;
1466 
1467         rcu_read_lock();
1468         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1469                 if (!device->bdev)
1470                         continue;
1471                 bdi = blk_get_backing_dev_info(device->bdev);
1472                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1473                         ret = 1;
1474                         break;
1475                 }
1476         }
1477         rcu_read_unlock();
1478         return ret;
1479 }
1480 
1481 /*
1482  * If this fails, caller must call bdi_destroy() to get rid of the
1483  * bdi again.
1484  */
1485 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1486 {
1487         int err;
1488 
1489         bdi->capabilities = BDI_CAP_MAP_COPY;
1490         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1491         if (err)
1492                 return err;
1493 
1494         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1495         bdi->congested_fn       = btrfs_congested_fn;
1496         bdi->congested_data     = info;
1497         return 0;
1498 }
1499 
1500 static int bio_ready_for_csum(struct bio *bio)
1501 {
1502         u64 length = 0;
1503         u64 buf_len = 0;
1504         u64 start = 0;
1505         struct page *page;
1506         struct extent_io_tree *io_tree = NULL;
1507         struct bio_vec *bvec;
1508         int i;
1509         int ret;
1510 
1511         bio_for_each_segment(bvec, bio, i) {
1512                 page = bvec->bv_page;
1513                 if (page->private == EXTENT_PAGE_PRIVATE) {
1514                         length += bvec->bv_len;
1515                         continue;
1516                 }
1517                 if (!page->private) {
1518                         length += bvec->bv_len;
1519                         continue;
1520                 }
1521                 length = bvec->bv_len;
1522                 buf_len = page->private >> 2;
1523                 start = page_offset(page) + bvec->bv_offset;
1524                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1525         }
1526         /* are we fully contained in this bio? */
1527         if (buf_len <= length)
1528                 return 1;
1529 
1530         ret = extent_range_uptodate(io_tree, start + length,
1531                                     start + buf_len - 1);
1532         return ret;
1533 }
1534 
1535 /*
1536  * called by the kthread helper functions to finally call the bio end_io
1537  * functions.  This is where read checksum verification actually happens
1538  */
1539 static void end_workqueue_fn(struct btrfs_work *work)
1540 {
1541         struct bio *bio;
1542         struct end_io_wq *end_io_wq;
1543         struct btrfs_fs_info *fs_info;
1544         int error;
1545 
1546         end_io_wq = container_of(work, struct end_io_wq, work);
1547         bio = end_io_wq->bio;
1548         fs_info = end_io_wq->info;
1549 
1550         /* metadata bio reads are special because the whole tree block must
1551          * be checksummed at once.  This makes sure the entire block is in
1552          * ram and up to date before trying to verify things.  For
1553          * blocksize <= pagesize, it is basically a noop
1554          */
1555         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1556             !bio_ready_for_csum(bio)) {
1557                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1558                                    &end_io_wq->work);
1559                 return;
1560         }
1561         error = end_io_wq->error;
1562         bio->bi_private = end_io_wq->private;
1563         bio->bi_end_io = end_io_wq->end_io;
1564         kfree(end_io_wq);
1565         bio_endio(bio, error);
1566 }
1567 
1568 static int cleaner_kthread(void *arg)
1569 {
1570         struct btrfs_root *root = arg;
1571 
1572         do {
1573                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1574 
1575                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1576                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1577                         btrfs_run_delayed_iputs(root);
1578                         btrfs_clean_old_snapshots(root);
1579                         mutex_unlock(&root->fs_info->cleaner_mutex);
1580                         btrfs_run_defrag_inodes(root->fs_info);
1581                 }
1582 
1583                 if (freezing(current)) {
1584                         refrigerator();
1585                 } else {
1586                         set_current_state(TASK_INTERRUPTIBLE);
1587                         if (!kthread_should_stop())
1588                                 schedule();
1589                         __set_current_state(TASK_RUNNING);
1590                 }
1591         } while (!kthread_should_stop());
1592         return 0;
1593 }
1594 
1595 static int transaction_kthread(void *arg)
1596 {
1597         struct btrfs_root *root = arg;
1598         struct btrfs_trans_handle *trans;
1599         struct btrfs_transaction *cur;
1600         u64 transid;
1601         unsigned long now;
1602         unsigned long delay;
1603         int ret;
1604 
1605         do {
1606                 delay = HZ * 30;
1607                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1608                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1609 
1610                 spin_lock(&root->fs_info->trans_lock);
1611                 cur = root->fs_info->running_transaction;
1612                 if (!cur) {
1613                         spin_unlock(&root->fs_info->trans_lock);
1614                         goto sleep;
1615                 }
1616 
1617                 now = get_seconds();
1618                 if (!cur->blocked &&
1619                     (now < cur->start_time || now - cur->start_time < 30)) {
1620                         spin_unlock(&root->fs_info->trans_lock);
1621                         delay = HZ * 5;
1622                         goto sleep;
1623                 }
1624                 transid = cur->transid;
1625                 spin_unlock(&root->fs_info->trans_lock);
1626 
1627                 trans = btrfs_join_transaction(root);
1628                 BUG_ON(IS_ERR(trans));
1629                 if (transid == trans->transid) {
1630                         ret = btrfs_commit_transaction(trans, root);
1631                         BUG_ON(ret);
1632                 } else {
1633                         btrfs_end_transaction(trans, root);
1634                 }
1635 sleep:
1636                 wake_up_process(root->fs_info->cleaner_kthread);
1637                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1638 
1639                 if (freezing(current)) {
1640                         refrigerator();
1641                 } else {
1642                         set_current_state(TASK_INTERRUPTIBLE);
1643                         if (!kthread_should_stop() &&
1644                             !btrfs_transaction_blocked(root->fs_info))
1645                                 schedule_timeout(delay);
1646                         __set_current_state(TASK_RUNNING);
1647                 }
1648         } while (!kthread_should_stop());
1649         return 0;
1650 }
1651 
1652 /*
1653  * this will find the highest generation in the array of
1654  * root backups.  The index of the highest array is returned,
1655  * or -1 if we can't find anything.
1656  *
1657  * We check to make sure the array is valid by comparing the
1658  * generation of the latest  root in the array with the generation
1659  * in the super block.  If they don't match we pitch it.
1660  */
1661 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1662 {
1663         u64 cur;
1664         int newest_index = -1;
1665         struct btrfs_root_backup *root_backup;
1666         int i;
1667 
1668         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1669                 root_backup = info->super_copy->super_roots + i;
1670                 cur = btrfs_backup_tree_root_gen(root_backup);
1671                 if (cur == newest_gen)
1672                         newest_index = i;
1673         }
1674 
1675         /* check to see if we actually wrapped around */
1676         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1677                 root_backup = info->super_copy->super_roots;
1678                 cur = btrfs_backup_tree_root_gen(root_backup);
1679                 if (cur == newest_gen)
1680                         newest_index = 0;
1681         }
1682         return newest_index;
1683 }
1684 
1685 
1686 /*
1687  * find the oldest backup so we know where to store new entries
1688  * in the backup array.  This will set the backup_root_index
1689  * field in the fs_info struct
1690  */
1691 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1692                                      u64 newest_gen)
1693 {
1694         int newest_index = -1;
1695 
1696         newest_index = find_newest_super_backup(info, newest_gen);
1697         /* if there was garbage in there, just move along */
1698         if (newest_index == -1) {
1699                 info->backup_root_index = 0;
1700         } else {
1701                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1702         }
1703 }
1704 
1705 /*
1706  * copy all the root pointers into the super backup array.
1707  * this will bump the backup pointer by one when it is
1708  * done
1709  */
1710 static void backup_super_roots(struct btrfs_fs_info *info)
1711 {
1712         int next_backup;
1713         struct btrfs_root_backup *root_backup;
1714         int last_backup;
1715 
1716         next_backup = info->backup_root_index;
1717         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1718                 BTRFS_NUM_BACKUP_ROOTS;
1719 
1720         /*
1721          * just overwrite the last backup if we're at the same generation
1722          * this happens only at umount
1723          */
1724         root_backup = info->super_for_commit->super_roots + last_backup;
1725         if (btrfs_backup_tree_root_gen(root_backup) ==
1726             btrfs_header_generation(info->tree_root->node))
1727                 next_backup = last_backup;
1728 
1729         root_backup = info->super_for_commit->super_roots + next_backup;
1730 
1731         /*
1732          * make sure all of our padding and empty slots get zero filled
1733          * regardless of which ones we use today
1734          */
1735         memset(root_backup, 0, sizeof(*root_backup));
1736 
1737         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1738 
1739         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1740         btrfs_set_backup_tree_root_gen(root_backup,
1741                                btrfs_header_generation(info->tree_root->node));
1742 
1743         btrfs_set_backup_tree_root_level(root_backup,
1744                                btrfs_header_level(info->tree_root->node));
1745 
1746         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1747         btrfs_set_backup_chunk_root_gen(root_backup,
1748                                btrfs_header_generation(info->chunk_root->node));
1749         btrfs_set_backup_chunk_root_level(root_backup,
1750                                btrfs_header_level(info->chunk_root->node));
1751 
1752         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1753         btrfs_set_backup_extent_root_gen(root_backup,
1754                                btrfs_header_generation(info->extent_root->node));
1755         btrfs_set_backup_extent_root_level(root_backup,
1756                                btrfs_header_level(info->extent_root->node));
1757 
1758         /*
1759          * we might commit during log recovery, which happens before we set
1760          * the fs_root.  Make sure it is valid before we fill it in.
1761          */
1762         if (info->fs_root && info->fs_root->node) {
1763                 btrfs_set_backup_fs_root(root_backup,
1764                                          info->fs_root->node->start);
1765                 btrfs_set_backup_fs_root_gen(root_backup,
1766                                btrfs_header_generation(info->fs_root->node));
1767                 btrfs_set_backup_fs_root_level(root_backup,
1768                                btrfs_header_level(info->fs_root->node));
1769         }
1770 
1771         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1772         btrfs_set_backup_dev_root_gen(root_backup,
1773                                btrfs_header_generation(info->dev_root->node));
1774         btrfs_set_backup_dev_root_level(root_backup,
1775                                        btrfs_header_level(info->dev_root->node));
1776 
1777         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1778         btrfs_set_backup_csum_root_gen(root_backup,
1779                                btrfs_header_generation(info->csum_root->node));
1780         btrfs_set_backup_csum_root_level(root_backup,
1781                                btrfs_header_level(info->csum_root->node));
1782 
1783         btrfs_set_backup_total_bytes(root_backup,
1784                              btrfs_super_total_bytes(info->super_copy));
1785         btrfs_set_backup_bytes_used(root_backup,
1786                              btrfs_super_bytes_used(info->super_copy));
1787         btrfs_set_backup_num_devices(root_backup,
1788                              btrfs_super_num_devices(info->super_copy));
1789 
1790         /*
1791          * if we don't copy this out to the super_copy, it won't get remembered
1792          * for the next commit
1793          */
1794         memcpy(&info->super_copy->super_roots,
1795                &info->super_for_commit->super_roots,
1796                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1797 }
1798 
1799 /*
1800  * this copies info out of the root backup array and back into
1801  * the in-memory super block.  It is meant to help iterate through
1802  * the array, so you send it the number of backups you've already
1803  * tried and the last backup index you used.
1804  *
1805  * this returns -1 when it has tried all the backups
1806  */
1807 static noinline int next_root_backup(struct btrfs_fs_info *info,
1808                                      struct btrfs_super_block *super,
1809                                      int *num_backups_tried, int *backup_index)
1810 {
1811         struct btrfs_root_backup *root_backup;
1812         int newest = *backup_index;
1813 
1814         if (*num_backups_tried == 0) {
1815                 u64 gen = btrfs_super_generation(super);
1816 
1817                 newest = find_newest_super_backup(info, gen);
1818                 if (newest == -1)
1819                         return -1;
1820 
1821                 *backup_index = newest;
1822                 *num_backups_tried = 1;
1823         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1824                 /* we've tried all the backups, all done */
1825                 return -1;
1826         } else {
1827                 /* jump to the next oldest backup */
1828                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1829                         BTRFS_NUM_BACKUP_ROOTS;
1830                 *backup_index = newest;
1831                 *num_backups_tried += 1;
1832         }
1833         root_backup = super->super_roots + newest;
1834 
1835         btrfs_set_super_generation(super,
1836                                    btrfs_backup_tree_root_gen(root_backup));
1837         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1838         btrfs_set_super_root_level(super,
1839                                    btrfs_backup_tree_root_level(root_backup));
1840         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1841 
1842         /*
1843          * fixme: the total bytes and num_devices need to match or we should
1844          * need a fsck
1845          */
1846         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1847         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1848         return 0;
1849 }
1850 
1851 /* helper to cleanup tree roots */
1852 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1853 {
1854         free_extent_buffer(info->tree_root->node);
1855         free_extent_buffer(info->tree_root->commit_root);
1856         free_extent_buffer(info->dev_root->node);
1857         free_extent_buffer(info->dev_root->commit_root);
1858         free_extent_buffer(info->extent_root->node);
1859         free_extent_buffer(info->extent_root->commit_root);
1860         free_extent_buffer(info->csum_root->node);
1861         free_extent_buffer(info->csum_root->commit_root);
1862 
1863         info->tree_root->node = NULL;
1864         info->tree_root->commit_root = NULL;
1865         info->dev_root->node = NULL;
1866         info->dev_root->commit_root = NULL;
1867         info->extent_root->node = NULL;
1868         info->extent_root->commit_root = NULL;
1869         info->csum_root->node = NULL;
1870         info->csum_root->commit_root = NULL;
1871 
1872         if (chunk_root) {
1873                 free_extent_buffer(info->chunk_root->node);
1874                 free_extent_buffer(info->chunk_root->commit_root);
1875                 info->chunk_root->node = NULL;
1876                 info->chunk_root->commit_root = NULL;
1877         }
1878 }
1879 
1880 
1881 struct btrfs_root *open_ctree(struct super_block *sb,
1882                               struct btrfs_fs_devices *fs_devices,
1883                               char *options)
1884 {
1885         u32 sectorsize;
1886         u32 nodesize;
1887         u32 leafsize;
1888         u32 blocksize;
1889         u32 stripesize;
1890         u64 generation;
1891         u64 features;
1892         struct btrfs_key location;
1893         struct buffer_head *bh;
1894         struct btrfs_super_block *disk_super;
1895         struct btrfs_root *tree_root = btrfs_sb(sb);
1896         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1897         struct btrfs_root *extent_root;
1898         struct btrfs_root *csum_root;
1899         struct btrfs_root *chunk_root;
1900         struct btrfs_root *dev_root;
1901         struct btrfs_root *log_tree_root;
1902         int ret;
1903         int err = -EINVAL;
1904         int num_backups_tried = 0;
1905         int backup_index = 0;
1906 
1907         extent_root = fs_info->extent_root =
1908                 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1909         csum_root = fs_info->csum_root =
1910                 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1911         chunk_root = fs_info->chunk_root =
1912                 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1913         dev_root = fs_info->dev_root =
1914                 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1915 
1916         if (!extent_root || !csum_root || !chunk_root || !dev_root) {
1917                 err = -ENOMEM;
1918                 goto fail;
1919         }
1920 
1921         ret = init_srcu_struct(&fs_info->subvol_srcu);
1922         if (ret) {
1923                 err = ret;
1924                 goto fail;
1925         }
1926 
1927         ret = setup_bdi(fs_info, &fs_info->bdi);
1928         if (ret) {
1929                 err = ret;
1930                 goto fail_srcu;
1931         }
1932 
1933         fs_info->btree_inode = new_inode(sb);
1934         if (!fs_info->btree_inode) {
1935                 err = -ENOMEM;
1936                 goto fail_bdi;
1937         }
1938 
1939         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1940 
1941         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1942         INIT_LIST_HEAD(&fs_info->trans_list);
1943         INIT_LIST_HEAD(&fs_info->dead_roots);
1944         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1945         INIT_LIST_HEAD(&fs_info->hashers);
1946         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1947         INIT_LIST_HEAD(&fs_info->ordered_operations);
1948         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1949         spin_lock_init(&fs_info->delalloc_lock);
1950         spin_lock_init(&fs_info->trans_lock);
1951         spin_lock_init(&fs_info->ref_cache_lock);
1952         spin_lock_init(&fs_info->fs_roots_radix_lock);
1953         spin_lock_init(&fs_info->delayed_iput_lock);
1954         spin_lock_init(&fs_info->defrag_inodes_lock);
1955         spin_lock_init(&fs_info->free_chunk_lock);
1956         mutex_init(&fs_info->reloc_mutex);
1957 
1958         init_completion(&fs_info->kobj_unregister);
1959         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1960         INIT_LIST_HEAD(&fs_info->space_info);
1961         btrfs_mapping_init(&fs_info->mapping_tree);
1962         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1963         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1964         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1965         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1966         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1967         btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1968         atomic_set(&fs_info->nr_async_submits, 0);
1969         atomic_set(&fs_info->async_delalloc_pages, 0);
1970         atomic_set(&fs_info->async_submit_draining, 0);
1971         atomic_set(&fs_info->nr_async_bios, 0);
1972         atomic_set(&fs_info->defrag_running, 0);
1973         fs_info->sb = sb;
1974         fs_info->max_inline = 8192 * 1024;
1975         fs_info->metadata_ratio = 0;
1976         fs_info->defrag_inodes = RB_ROOT;
1977         fs_info->trans_no_join = 0;
1978         fs_info->free_chunk_space = 0;
1979 
1980         /* readahead state */
1981         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1982         spin_lock_init(&fs_info->reada_lock);
1983 
1984         fs_info->thread_pool_size = min_t(unsigned long,
1985                                           num_online_cpus() + 2, 8);
1986 
1987         INIT_LIST_HEAD(&fs_info->ordered_extents);
1988         spin_lock_init(&fs_info->ordered_extent_lock);
1989         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1990                                         GFP_NOFS);
1991         if (!fs_info->delayed_root) {
1992                 err = -ENOMEM;
1993                 goto fail_iput;
1994         }
1995         btrfs_init_delayed_root(fs_info->delayed_root);
1996 
1997         mutex_init(&fs_info->scrub_lock);
1998         atomic_set(&fs_info->scrubs_running, 0);
1999         atomic_set(&fs_info->scrub_pause_req, 0);
2000         atomic_set(&fs_info->scrubs_paused, 0);
2001         atomic_set(&fs_info->scrub_cancel_req, 0);
2002         init_waitqueue_head(&fs_info->scrub_pause_wait);
2003         init_rwsem(&fs_info->scrub_super_lock);
2004         fs_info->scrub_workers_refcnt = 0;
2005 
2006         sb->s_blocksize = 4096;
2007         sb->s_blocksize_bits = blksize_bits(4096);
2008         sb->s_bdi = &fs_info->bdi;
2009 
2010         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2011         set_nlink(fs_info->btree_inode, 1);
2012         /*
2013          * we set the i_size on the btree inode to the max possible int.
2014          * the real end of the address space is determined by all of
2015          * the devices in the system
2016          */
2017         fs_info->btree_inode->i_size = OFFSET_MAX;
2018         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2019         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2020 
2021         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2022         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2023                              fs_info->btree_inode->i_mapping);
2024         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2025 
2026         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2027 
2028         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2029         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2030                sizeof(struct btrfs_key));
2031         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
2032         insert_inode_hash(fs_info->btree_inode);
2033 
2034         spin_lock_init(&fs_info->block_group_cache_lock);
2035         fs_info->block_group_cache_tree = RB_ROOT;
2036 
2037         extent_io_tree_init(&fs_info->freed_extents[0],
2038                              fs_info->btree_inode->i_mapping);
2039         extent_io_tree_init(&fs_info->freed_extents[1],
2040                              fs_info->btree_inode->i_mapping);
2041         fs_info->pinned_extents = &fs_info->freed_extents[0];
2042         fs_info->do_barriers = 1;
2043 
2044 
2045         mutex_init(&fs_info->ordered_operations_mutex);
2046         mutex_init(&fs_info->tree_log_mutex);
2047         mutex_init(&fs_info->chunk_mutex);
2048         mutex_init(&fs_info->transaction_kthread_mutex);
2049         mutex_init(&fs_info->cleaner_mutex);
2050         mutex_init(&fs_info->volume_mutex);
2051         init_rwsem(&fs_info->extent_commit_sem);
2052         init_rwsem(&fs_info->cleanup_work_sem);
2053         init_rwsem(&fs_info->subvol_sem);
2054 
2055         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2056         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2057 
2058         init_waitqueue_head(&fs_info->transaction_throttle);
2059         init_waitqueue_head(&fs_info->transaction_wait);
2060         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2061         init_waitqueue_head(&fs_info->async_submit_wait);
2062 
2063         __setup_root(4096, 4096, 4096, 4096, tree_root,
2064                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2065 
2066         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2067         if (!bh) {
2068                 err = -EINVAL;
2069                 goto fail_alloc;
2070         }
2071 
2072         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2073         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2074                sizeof(*fs_info->super_for_commit));
2075         brelse(bh);
2076 
2077         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2078 
2079         disk_super = fs_info->super_copy;
2080         if (!btrfs_super_root(disk_super))
2081                 goto fail_alloc;
2082 
2083         /* check FS state, whether FS is broken. */
2084         fs_info->fs_state |= btrfs_super_flags(disk_super);
2085 
2086         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2087 
2088         /*
2089          * run through our array of backup supers and setup
2090          * our ring pointer to the oldest one
2091          */
2092         generation = btrfs_super_generation(disk_super);
2093         find_oldest_super_backup(fs_info, generation);
2094 
2095         /*
2096          * In the long term, we'll store the compression type in the super
2097          * block, and it'll be used for per file compression control.
2098          */
2099         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2100 
2101         ret = btrfs_parse_options(tree_root, options);
2102         if (ret) {
2103                 err = ret;
2104                 goto fail_alloc;
2105         }
2106 
2107         features = btrfs_super_incompat_flags(disk_super) &
2108                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2109         if (features) {
2110                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2111                        "unsupported optional features (%Lx).\n",
2112                        (unsigned long long)features);
2113                 err = -EINVAL;
2114                 goto fail_alloc;
2115         }
2116 
2117         features = btrfs_super_incompat_flags(disk_super);
2118         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2119         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2120                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2121         btrfs_set_super_incompat_flags(disk_super, features);
2122 
2123         features = btrfs_super_compat_ro_flags(disk_super) &
2124                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2125         if (!(sb->s_flags & MS_RDONLY) && features) {
2126                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2127                        "unsupported option features (%Lx).\n",
2128                        (unsigned long long)features);
2129                 err = -EINVAL;
2130                 goto fail_alloc;
2131         }
2132 
2133         btrfs_init_workers(&fs_info->generic_worker,
2134                            "genwork", 1, NULL);
2135 
2136         btrfs_init_workers(&fs_info->workers, "worker",
2137                            fs_info->thread_pool_size,
2138                            &fs_info->generic_worker);
2139 
2140         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2141                            fs_info->thread_pool_size,
2142                            &fs_info->generic_worker);
2143 
2144         btrfs_init_workers(&fs_info->submit_workers, "submit",
2145                            min_t(u64, fs_devices->num_devices,
2146                            fs_info->thread_pool_size),
2147                            &fs_info->generic_worker);
2148 
2149         btrfs_init_workers(&fs_info->caching_workers, "cache",
2150                            2, &fs_info->generic_worker);
2151 
2152         /* a higher idle thresh on the submit workers makes it much more
2153          * likely that bios will be send down in a sane order to the
2154          * devices
2155          */
2156         fs_info->submit_workers.idle_thresh = 64;
2157 
2158         fs_info->workers.idle_thresh = 16;
2159         fs_info->workers.ordered = 1;
2160 
2161         fs_info->delalloc_workers.idle_thresh = 2;
2162         fs_info->delalloc_workers.ordered = 1;
2163 
2164         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2165                            &fs_info->generic_worker);
2166         btrfs_init_workers(&fs_info->endio_workers, "endio",
2167                            fs_info->thread_pool_size,
2168                            &fs_info->generic_worker);
2169         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2170                            fs_info->thread_pool_size,
2171                            &fs_info->generic_worker);
2172         btrfs_init_workers(&fs_info->endio_meta_write_workers,
2173                            "endio-meta-write", fs_info->thread_pool_size,
2174                            &fs_info->generic_worker);
2175         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2176                            fs_info->thread_pool_size,
2177                            &fs_info->generic_worker);
2178         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2179                            1, &fs_info->generic_worker);
2180         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2181                            fs_info->thread_pool_size,
2182                            &fs_info->generic_worker);
2183         btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2184                            fs_info->thread_pool_size,
2185                            &fs_info->generic_worker);
2186 
2187         /*
2188          * endios are largely parallel and should have a very
2189          * low idle thresh
2190          */
2191         fs_info->endio_workers.idle_thresh = 4;
2192         fs_info->endio_meta_workers.idle_thresh = 4;
2193 
2194         fs_info->endio_write_workers.idle_thresh = 2;
2195         fs_info->endio_meta_write_workers.idle_thresh = 2;
2196         fs_info->readahead_workers.idle_thresh = 2;
2197 
2198         /*
2199          * btrfs_start_workers can really only fail because of ENOMEM so just
2200          * return -ENOMEM if any of these fail.
2201          */
2202         ret = btrfs_start_workers(&fs_info->workers);
2203         ret |= btrfs_start_workers(&fs_info->generic_worker);
2204         ret |= btrfs_start_workers(&fs_info->submit_workers);
2205         ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2206         ret |= btrfs_start_workers(&fs_info->fixup_workers);
2207         ret |= btrfs_start_workers(&fs_info->endio_workers);
2208         ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2209         ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2210         ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2211         ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2212         ret |= btrfs_start_workers(&fs_info->delayed_workers);
2213         ret |= btrfs_start_workers(&fs_info->caching_workers);
2214         ret |= btrfs_start_workers(&fs_info->readahead_workers);
2215         if (ret) {
2216                 ret = -ENOMEM;
2217                 goto fail_sb_buffer;
2218         }
2219 
2220         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2221         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2222                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2223 
2224         nodesize = btrfs_super_nodesize(disk_super);
2225         leafsize = btrfs_super_leafsize(disk_super);
2226         sectorsize = btrfs_super_sectorsize(disk_super);
2227         stripesize = btrfs_super_stripesize(disk_super);
2228         tree_root->nodesize = nodesize;
2229         tree_root->leafsize = leafsize;
2230         tree_root->sectorsize = sectorsize;
2231         tree_root->stripesize = stripesize;
2232 
2233         sb->s_blocksize = sectorsize;
2234         sb->s_blocksize_bits = blksize_bits(sectorsize);
2235 
2236         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2237                     sizeof(disk_super->magic))) {
2238                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2239                 goto fail_sb_buffer;
2240         }
2241 
2242         mutex_lock(&fs_info->chunk_mutex);
2243         ret = btrfs_read_sys_array(tree_root);
2244         mutex_unlock(&fs_info->chunk_mutex);
2245         if (ret) {
2246                 printk(KERN_WARNING "btrfs: failed to read the system "
2247                        "array on %s\n", sb->s_id);
2248                 goto fail_sb_buffer;
2249         }
2250 
2251         blocksize = btrfs_level_size(tree_root,
2252                                      btrfs_super_chunk_root_level(disk_super));
2253         generation = btrfs_super_chunk_root_generation(disk_super);
2254 
2255         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2256                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2257 
2258         chunk_root->node = read_tree_block(chunk_root,
2259                                            btrfs_super_chunk_root(disk_super),
2260                                            blocksize, generation);
2261         BUG_ON(!chunk_root->node);
2262         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2263                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2264                        sb->s_id);
2265                 goto fail_tree_roots;
2266         }
2267         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2268         chunk_root->commit_root = btrfs_root_node(chunk_root);
2269 
2270         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2271            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2272            BTRFS_UUID_SIZE);
2273 
2274         mutex_lock(&fs_info->chunk_mutex);
2275         ret = btrfs_read_chunk_tree(chunk_root);
2276         mutex_unlock(&fs_info->chunk_mutex);
2277         if (ret) {
2278                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2279                        sb->s_id);
2280                 goto fail_tree_roots;
2281         }
2282 
2283         btrfs_close_extra_devices(fs_devices);
2284 
2285 retry_root_backup:
2286         blocksize = btrfs_level_size(tree_root,
2287                                      btrfs_super_root_level(disk_super));
2288         generation = btrfs_super_generation(disk_super);
2289 
2290         tree_root->node = read_tree_block(tree_root,
2291                                           btrfs_super_root(disk_super),
2292                                           blocksize, generation);
2293         if (!tree_root->node ||
2294             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2295                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2296                        sb->s_id);
2297 
2298                 goto recovery_tree_root;
2299         }
2300 
2301         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2302         tree_root->commit_root = btrfs_root_node(tree_root);
2303 
2304         ret = find_and_setup_root(tree_root, fs_info,
2305                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2306         if (ret)
2307                 goto recovery_tree_root;
2308         extent_root->track_dirty = 1;
2309 
2310         ret = find_and_setup_root(tree_root, fs_info,
2311                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2312         if (ret)
2313                 goto recovery_tree_root;
2314         dev_root->track_dirty = 1;
2315 
2316         ret = find_and_setup_root(tree_root, fs_info,
2317                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2318         if (ret)
2319                 goto recovery_tree_root;
2320 
2321         csum_root->track_dirty = 1;
2322 
2323         fs_info->generation = generation;
2324         fs_info->last_trans_committed = generation;
2325         fs_info->data_alloc_profile = (u64)-1;
2326         fs_info->metadata_alloc_profile = (u64)-1;
2327         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2328 
2329         ret = btrfs_init_space_info(fs_info);
2330         if (ret) {
2331                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2332                 goto fail_block_groups;
2333         }
2334 
2335         ret = btrfs_read_block_groups(extent_root);
2336         if (ret) {
2337                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2338                 goto fail_block_groups;
2339         }
2340 
2341         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2342                                                "btrfs-cleaner");
2343         if (IS_ERR(fs_info->cleaner_kthread))
2344                 goto fail_block_groups;
2345 
2346         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2347                                                    tree_root,
2348                                                    "btrfs-transaction");
2349         if (IS_ERR(fs_info->transaction_kthread))
2350                 goto fail_cleaner;
2351 
2352         if (!btrfs_test_opt(tree_root, SSD) &&
2353             !btrfs_test_opt(tree_root, NOSSD) &&
2354             !fs_info->fs_devices->rotating) {
2355                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2356                        "mode\n");
2357                 btrfs_set_opt(fs_info->mount_opt, SSD);
2358         }
2359 
2360         /* do not make disk changes in broken FS */
2361         if (btrfs_super_log_root(disk_super) != 0 &&
2362             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2363                 u64 bytenr = btrfs_super_log_root(disk_super);
2364 
2365                 if (fs_devices->rw_devices == 0) {
2366                         printk(KERN_WARNING "Btrfs log replay required "
2367                                "on RO media\n");
2368                         err = -EIO;
2369                         goto fail_trans_kthread;
2370                 }
2371                 blocksize =
2372                      btrfs_level_size(tree_root,
2373                                       btrfs_super_log_root_level(disk_super));
2374 
2375                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2376                 if (!log_tree_root) {
2377                         err = -ENOMEM;
2378                         goto fail_trans_kthread;
2379                 }
2380 
2381                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2382                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2383 
2384                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2385                                                       blocksize,
2386                                                       generation + 1);
2387                 ret = btrfs_recover_log_trees(log_tree_root);
2388                 BUG_ON(ret);
2389 
2390                 if (sb->s_flags & MS_RDONLY) {
2391                         ret =  btrfs_commit_super(tree_root);
2392                         BUG_ON(ret);
2393                 }
2394         }
2395 
2396         ret = btrfs_find_orphan_roots(tree_root);
2397         BUG_ON(ret);
2398 
2399         if (!(sb->s_flags & MS_RDONLY)) {
2400                 ret = btrfs_cleanup_fs_roots(fs_info);
2401                 BUG_ON(ret);
2402 
2403                 ret = btrfs_recover_relocation(tree_root);
2404                 if (ret < 0) {
2405                         printk(KERN_WARNING
2406                                "btrfs: failed to recover relocation\n");
2407                         err = -EINVAL;
2408                         goto fail_trans_kthread;
2409                 }
2410         }
2411 
2412         location.objectid = BTRFS_FS_TREE_OBJECTID;
2413         location.type = BTRFS_ROOT_ITEM_KEY;
2414         location.offset = (u64)-1;
2415 
2416         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2417         if (!fs_info->fs_root)
2418                 goto fail_trans_kthread;
2419         if (IS_ERR(fs_info->fs_root)) {
2420                 err = PTR_ERR(fs_info->fs_root);
2421                 goto fail_trans_kthread;
2422         }
2423 
2424         if (!(sb->s_flags & MS_RDONLY)) {
2425                 down_read(&fs_info->cleanup_work_sem);
2426                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2427                 if (!err)
2428                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2429                 up_read(&fs_info->cleanup_work_sem);
2430                 if (err) {
2431                         close_ctree(tree_root);
2432                         return ERR_PTR(err);
2433                 }
2434         }
2435 
2436         return tree_root;
2437 
2438 fail_trans_kthread:
2439         kthread_stop(fs_info->transaction_kthread);
2440 fail_cleaner:
2441         kthread_stop(fs_info->cleaner_kthread);
2442 
2443         /*
2444          * make sure we're done with the btree inode before we stop our
2445          * kthreads
2446          */
2447         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2448         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2449 
2450 fail_block_groups:
2451         btrfs_free_block_groups(fs_info);
2452 
2453 fail_tree_roots:
2454         free_root_pointers(fs_info, 1);
2455 
2456 fail_sb_buffer:
2457         btrfs_stop_workers(&fs_info->generic_worker);
2458         btrfs_stop_workers(&fs_info->readahead_workers);
2459         btrfs_stop_workers(&fs_info->fixup_workers);
2460         btrfs_stop_workers(&fs_info->delalloc_workers);
2461         btrfs_stop_workers(&fs_info->workers);
2462         btrfs_stop_workers(&fs_info->endio_workers);
2463         btrfs_stop_workers(&fs_info->endio_meta_workers);
2464         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2465         btrfs_stop_workers(&fs_info->endio_write_workers);
2466         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2467         btrfs_stop_workers(&fs_info->submit_workers);
2468         btrfs_stop_workers(&fs_info->delayed_workers);
2469         btrfs_stop_workers(&fs_info->caching_workers);
2470 fail_alloc:
2471 fail_iput:
2472         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2473 
2474         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2475         iput(fs_info->btree_inode);
2476 fail_bdi:
2477         bdi_destroy(&fs_info->bdi);
2478 fail_srcu:
2479         cleanup_srcu_struct(&fs_info->subvol_srcu);
2480 fail:
2481         btrfs_close_devices(fs_info->fs_devices);
2482         free_fs_info(fs_info);
2483         return ERR_PTR(err);
2484 
2485 recovery_tree_root:
2486         if (!btrfs_test_opt(tree_root, RECOVERY))
2487                 goto fail_tree_roots;
2488 
2489         free_root_pointers(fs_info, 0);
2490 
2491         /* don't use the log in recovery mode, it won't be valid */
2492         btrfs_set_super_log_root(disk_super, 0);
2493 
2494         /* we can't trust the free space cache either */
2495         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2496 
2497         ret = next_root_backup(fs_info, fs_info->super_copy,
2498                                &num_backups_tried, &backup_index);
2499         if (ret == -1)
2500                 goto fail_block_groups;
2501         goto retry_root_backup;
2502 }
2503 
2504 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2505 {
2506         char b[BDEVNAME_SIZE];
2507 
2508         if (uptodate) {
2509                 set_buffer_uptodate(bh);
2510         } else {
2511                 printk_ratelimited(KERN_WARNING "lost page write due to "
2512                                         "I/O error on %s\n",
2513                                        bdevname(bh->b_bdev, b));
2514                 /* note, we dont' set_buffer_write_io_error because we have
2515                  * our own ways of dealing with the IO errors
2516                  */
2517                 clear_buffer_uptodate(bh);
2518         }
2519         unlock_buffer(bh);
2520         put_bh(bh);
2521 }
2522 
2523 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2524 {
2525         struct buffer_head *bh;
2526         struct buffer_head *latest = NULL;
2527         struct btrfs_super_block *super;
2528         int i;
2529         u64 transid = 0;
2530         u64 bytenr;
2531 
2532         /* we would like to check all the supers, but that would make
2533          * a btrfs mount succeed after a mkfs from a different FS.
2534          * So, we need to add a special mount option to scan for
2535          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2536          */
2537         for (i = 0; i < 1; i++) {
2538                 bytenr = btrfs_sb_offset(i);
2539                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2540                         break;
2541                 bh = __bread(bdev, bytenr / 4096, 4096);
2542                 if (!bh)
2543                         continue;
2544 
2545                 super = (struct btrfs_super_block *)bh->b_data;
2546                 if (btrfs_super_bytenr(super) != bytenr ||
2547                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2548                             sizeof(super->magic))) {
2549                         brelse(bh);
2550                         continue;
2551                 }
2552 
2553                 if (!latest || btrfs_super_generation(super) > transid) {
2554                         brelse(latest);
2555                         latest = bh;
2556                         transid = btrfs_super_generation(super);
2557                 } else {
2558                         brelse(bh);
2559                 }
2560         }
2561         return latest;
2562 }
2563 
2564 /*
2565  * this should be called twice, once with wait == 0 and
2566  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2567  * we write are pinned.
2568  *
2569  * They are released when wait == 1 is done.
2570  * max_mirrors must be the same for both runs, and it indicates how
2571  * many supers on this one device should be written.
2572  *
2573  * max_mirrors == 0 means to write them all.
2574  */
2575 static int write_dev_supers(struct btrfs_device *device,
2576                             struct btrfs_super_block *sb,
2577                             int do_barriers, int wait, int max_mirrors)
2578 {
2579         struct buffer_head *bh;
2580         int i;
2581         int ret;
2582         int errors = 0;
2583         u32 crc;
2584         u64 bytenr;
2585 
2586         if (max_mirrors == 0)
2587                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2588 
2589         for (i = 0; i < max_mirrors; i++) {
2590                 bytenr = btrfs_sb_offset(i);
2591                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2592                         break;
2593 
2594                 if (wait) {
2595                         bh = __find_get_block(device->bdev, bytenr / 4096,
2596                                               BTRFS_SUPER_INFO_SIZE);
2597                         BUG_ON(!bh);
2598                         wait_on_buffer(bh);
2599                         if (!buffer_uptodate(bh))
2600                                 errors++;
2601 
2602                         /* drop our reference */
2603                         brelse(bh);
2604 
2605                         /* drop the reference from the wait == 0 run */
2606                         brelse(bh);
2607                         continue;
2608                 } else {
2609                         btrfs_set_super_bytenr(sb, bytenr);
2610 
2611                         crc = ~(u32)0;
2612                         crc = btrfs_csum_data(NULL, (char *)sb +
2613                                               BTRFS_CSUM_SIZE, crc,
2614                                               BTRFS_SUPER_INFO_SIZE -
2615                                               BTRFS_CSUM_SIZE);
2616                         btrfs_csum_final(crc, sb->csum);
2617 
2618                         /*
2619                          * one reference for us, and we leave it for the
2620                          * caller
2621                          */
2622                         bh = __getblk(device->bdev, bytenr / 4096,
2623                                       BTRFS_SUPER_INFO_SIZE);
2624                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2625 
2626                         /* one reference for submit_bh */
2627                         get_bh(bh);
2628 
2629                         set_buffer_uptodate(bh);
2630                         lock_buffer(bh);
2631                         bh->b_end_io = btrfs_end_buffer_write_sync;
2632                 }
2633 
2634                 /*
2635                  * we fua the first super.  The others we allow
2636                  * to go down lazy.
2637                  */
2638                 ret = submit_bh(WRITE_FUA, bh);
2639                 if (ret)
2640                         errors++;
2641         }
2642         return errors < i ? 0 : -1;
2643 }
2644 
2645 /*
2646  * endio for the write_dev_flush, this will wake anyone waiting
2647  * for the barrier when it is done
2648  */
2649 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2650 {
2651         if (err) {
2652                 if (err == -EOPNOTSUPP)
2653                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2654                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2655         }
2656         if (bio->bi_private)
2657                 complete(bio->bi_private);
2658         bio_put(bio);
2659 }
2660 
2661 /*
2662  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
2663  * sent down.  With wait == 1, it waits for the previous flush.
2664  *
2665  * any device where the flush fails with eopnotsupp are flagged as not-barrier
2666  * capable
2667  */
2668 static int write_dev_flush(struct btrfs_device *device, int wait)
2669 {
2670         struct bio *bio;
2671         int ret = 0;
2672 
2673         if (device->nobarriers)
2674                 return 0;
2675 
2676         if (wait) {
2677                 bio = device->flush_bio;
2678                 if (!bio)
2679                         return 0;
2680 
2681                 wait_for_completion(&device->flush_wait);
2682 
2683                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2684                         printk("btrfs: disabling barriers on dev %s\n",
2685                                device->name);
2686                         device->nobarriers = 1;
2687                 }
2688                 if (!bio_flagged(bio, BIO_UPTODATE)) {
2689                         ret = -EIO;
2690                 }
2691 
2692                 /* drop the reference from the wait == 0 run */
2693                 bio_put(bio);
2694                 device->flush_bio = NULL;
2695 
2696                 return ret;
2697         }
2698 
2699         /*
2700          * one reference for us, and we leave it for the
2701          * caller
2702          */
2703         device->flush_bio = NULL;;
2704         bio = bio_alloc(GFP_NOFS, 0);
2705         if (!bio)
2706                 return -ENOMEM;
2707 
2708         bio->bi_end_io = btrfs_end_empty_barrier;
2709         bio->bi_bdev = device->bdev;
2710         init_completion(&device->flush_wait);
2711         bio->bi_private = &device->flush_wait;
2712         device->flush_bio = bio;
2713 
2714         bio_get(bio);
2715         submit_bio(WRITE_FLUSH, bio);
2716 
2717         return 0;
2718 }
2719 
2720 /*
2721  * send an empty flush down to each device in parallel,
2722  * then wait for them
2723  */
2724 static int barrier_all_devices(struct btrfs_fs_info *info)
2725 {
2726         struct list_head *head;
2727         struct btrfs_device *dev;
2728         int errors = 0;
2729         int ret;
2730 
2731         /* send down all the barriers */
2732         head = &info->fs_devices->devices;
2733         list_for_each_entry_rcu(dev, head, dev_list) {
2734                 if (dev->missing)
2735                         continue;
2736                 if (!dev->bdev) {
2737                         errors++;
2738                         continue;
2739                 }
2740                 if (!dev->in_fs_metadata || !dev->writeable)
2741                         continue;
2742 
2743                 ret = write_dev_flush(dev, 0);
2744                 if (ret)
2745                         errors++;
2746         }
2747 
2748         /* wait for all the barriers */
2749         list_for_each_entry_rcu(dev, head, dev_list) {
2750                 if (dev->missing)
2751                         continue;
2752                 if (!dev->bdev) {
2753                         errors++;
2754                         continue;
2755                 }
2756                 if (!dev->in_fs_metadata || !dev->writeable)
2757                         continue;
2758 
2759                 ret = write_dev_flush(dev, 1);
2760                 if (ret)
2761                         errors++;
2762         }
2763         if (errors)
2764                 return -EIO;
2765         return 0;
2766 }
2767 
2768 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2769 {
2770         struct list_head *head;
2771         struct btrfs_device *dev;
2772         struct btrfs_super_block *sb;
2773         struct btrfs_dev_item *dev_item;
2774         int ret;
2775         int do_barriers;
2776         int max_errors;
2777         int total_errors = 0;
2778         u64 flags;
2779 
2780         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2781         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2782         backup_super_roots(root->fs_info);
2783 
2784         sb = root->fs_info->super_for_commit;
2785         dev_item = &sb->dev_item;
2786 
2787         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2788         head = &root->fs_info->fs_devices->devices;
2789 
2790         if (do_barriers)
2791                 barrier_all_devices(root->fs_info);
2792 
2793         list_for_each_entry_rcu(dev, head, dev_list) {
2794                 if (!dev->bdev) {
2795                         total_errors++;
2796                         continue;
2797                 }
2798                 if (!dev->in_fs_metadata || !dev->writeable)
2799                         continue;
2800 
2801                 btrfs_set_stack_device_generation(dev_item, 0);
2802                 btrfs_set_stack_device_type(dev_item, dev->type);
2803                 btrfs_set_stack_device_id(dev_item, dev->devid);
2804                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2805                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2806                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2807                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2808                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2809                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2810                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2811 
2812                 flags = btrfs_super_flags(sb);
2813                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2814 
2815                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2816                 if (ret)
2817                         total_errors++;
2818         }
2819         if (total_errors > max_errors) {
2820                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2821                        total_errors);
2822                 BUG();
2823         }
2824 
2825         total_errors = 0;
2826         list_for_each_entry_rcu(dev, head, dev_list) {
2827                 if (!dev->bdev)
2828                         continue;
2829                 if (!dev->in_fs_metadata || !dev->writeable)
2830                         continue;
2831 
2832                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2833                 if (ret)
2834                         total_errors++;
2835         }
2836         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2837         if (total_errors > max_errors) {
2838                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2839                        total_errors);
2840                 BUG();
2841         }
2842         return 0;
2843 }
2844 
2845 int write_ctree_super(struct btrfs_trans_handle *trans,
2846                       struct btrfs_root *root, int max_mirrors)
2847 {
2848         int ret;
2849 
2850         ret = write_all_supers(root, max_mirrors);
2851         return ret;
2852 }
2853 
2854 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2855 {
2856         spin_lock(&fs_info->fs_roots_radix_lock);
2857         radix_tree_delete(&fs_info->fs_roots_radix,
2858                           (unsigned long)root->root_key.objectid);
2859         spin_unlock(&fs_info->fs_roots_radix_lock);
2860 
2861         if (btrfs_root_refs(&root->root_item) == 0)
2862                 synchronize_srcu(&fs_info->subvol_srcu);
2863 
2864         __btrfs_remove_free_space_cache(root->free_ino_pinned);
2865         __btrfs_remove_free_space_cache(root->free_ino_ctl);
2866         free_fs_root(root);
2867         return 0;
2868 }
2869 
2870 static void free_fs_root(struct btrfs_root *root)
2871 {
2872         iput(root->cache_inode);
2873         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2874         if (root->anon_dev)
2875                 free_anon_bdev(root->anon_dev);
2876         free_extent_buffer(root->node);
2877         free_extent_buffer(root->commit_root);
2878         kfree(root->free_ino_ctl);
2879         kfree(root->free_ino_pinned);
2880         kfree(root->name);
2881         kfree(root);
2882 }
2883 
2884 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2885 {
2886         int ret;
2887         struct btrfs_root *gang[8];
2888         int i;
2889 
2890         while (!list_empty(&fs_info->dead_roots)) {
2891                 gang[0] = list_entry(fs_info->dead_roots.next,
2892                                      struct btrfs_root, root_list);
2893                 list_del(&gang[0]->root_list);
2894 
2895                 if (gang[0]->in_radix) {
2896                         btrfs_free_fs_root(fs_info, gang[0]);
2897                 } else {
2898                         free_extent_buffer(gang[0]->node);
2899                         free_extent_buffer(gang[0]->commit_root);
2900                         kfree(gang[0]);
2901                 }
2902         }
2903 
2904         while (1) {
2905                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2906                                              (void **)gang, 0,
2907                                              ARRAY_SIZE(gang));
2908                 if (!ret)
2909                         break;
2910                 for (i = 0; i < ret; i++)
2911                         btrfs_free_fs_root(fs_info, gang[i]);
2912         }
2913         return 0;
2914 }
2915 
2916 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2917 {
2918         u64 root_objectid = 0;
2919         struct btrfs_root *gang[8];
2920         int i;
2921         int ret;
2922 
2923         while (1) {
2924                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2925                                              (void **)gang, root_objectid,
2926                                              ARRAY_SIZE(gang));
2927                 if (!ret)
2928                         break;
2929 
2930                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2931                 for (i = 0; i < ret; i++) {
2932                         int err;
2933 
2934                         root_objectid = gang[i]->root_key.objectid;
2935                         err = btrfs_orphan_cleanup(gang[i]);
2936                         if (err)
2937                                 return err;
2938                 }
2939                 root_objectid++;
2940         }
2941         return 0;
2942 }
2943 
2944 int btrfs_commit_super(struct btrfs_root *root)
2945 {
2946         struct btrfs_trans_handle *trans;
2947         int ret;
2948 
2949         mutex_lock(&root->fs_info->cleaner_mutex);
2950         btrfs_run_delayed_iputs(root);
2951         btrfs_clean_old_snapshots(root);
2952         mutex_unlock(&root->fs_info->cleaner_mutex);
2953 
2954         /* wait until ongoing cleanup work done */
2955         down_write(&root->fs_info->cleanup_work_sem);
2956         up_write(&root->fs_info->cleanup_work_sem);
2957 
2958         trans = btrfs_join_transaction(root);
2959         if (IS_ERR(trans))
2960                 return PTR_ERR(trans);
2961         ret = btrfs_commit_transaction(trans, root);
2962         BUG_ON(ret);
2963         /* run commit again to drop the original snapshot */
2964         trans = btrfs_join_transaction(root);
2965         if (IS_ERR(trans))
2966                 return PTR_ERR(trans);
2967         btrfs_commit_transaction(trans, root);
2968         ret = btrfs_write_and_wait_transaction(NULL, root);
2969         BUG_ON(ret);
2970 
2971         ret = write_ctree_super(NULL, root, 0);
2972         return ret;
2973 }
2974 
2975 int close_ctree(struct btrfs_root *root)
2976 {
2977         struct btrfs_fs_info *fs_info = root->fs_info;
2978         int ret;
2979 
2980         fs_info->closing = 1;
2981         smp_mb();
2982 
2983         btrfs_scrub_cancel(root);
2984 
2985         /* wait for any defraggers to finish */
2986         wait_event(fs_info->transaction_wait,
2987                    (atomic_read(&fs_info->defrag_running) == 0));
2988 
2989         /* clear out the rbtree of defraggable inodes */
2990         btrfs_run_defrag_inodes(root->fs_info);
2991 
2992         /*
2993          * Here come 2 situations when btrfs is broken to flip readonly:
2994          *
2995          * 1. when btrfs flips readonly somewhere else before
2996          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2997          * and btrfs will skip to write sb directly to keep
2998          * ERROR state on disk.
2999          *
3000          * 2. when btrfs flips readonly just in btrfs_commit_super,
3001          * and in such case, btrfs cannot write sb via btrfs_commit_super,
3002          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3003          * btrfs will cleanup all FS resources first and write sb then.
3004          */
3005         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3006                 ret = btrfs_commit_super(root);
3007                 if (ret)
3008                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3009         }
3010 
3011         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3012                 ret = btrfs_error_commit_super(root);
3013                 if (ret)
3014                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3015         }
3016 
3017         btrfs_put_block_group_cache(fs_info);
3018 
3019         kthread_stop(root->fs_info->transaction_kthread);
3020         kthread_stop(root->fs_info->cleaner_kthread);
3021 
3022         fs_info->closing = 2;
3023         smp_mb();
3024 
3025         if (fs_info->delalloc_bytes) {
3026                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3027                        (unsigned long long)fs_info->delalloc_bytes);
3028         }
3029         if (fs_info->total_ref_cache_size) {
3030                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3031                        (unsigned long long)fs_info->total_ref_cache_size);
3032         }
3033 
3034         free_extent_buffer(fs_info->extent_root->node);
3035         free_extent_buffer(fs_info->extent_root->commit_root);
3036         free_extent_buffer(fs_info->tree_root->node);
3037         free_extent_buffer(fs_info->tree_root->commit_root);
3038         free_extent_buffer(root->fs_info->chunk_root->node);
3039         free_extent_buffer(root->fs_info->chunk_root->commit_root);
3040         free_extent_buffer(root->fs_info->dev_root->node);
3041         free_extent_buffer(root->fs_info->dev_root->commit_root);
3042         free_extent_buffer(root->fs_info->csum_root->node);
3043         free_extent_buffer(root->fs_info->csum_root->commit_root);
3044 
3045         btrfs_free_block_groups(root->fs_info);
3046 
3047         del_fs_roots(fs_info);
3048 
3049         iput(fs_info->btree_inode);
3050 
3051         btrfs_stop_workers(&fs_info->generic_worker);
3052         btrfs_stop_workers(&fs_info->fixup_workers);
3053         btrfs_stop_workers(&fs_info->delalloc_workers);
3054         btrfs_stop_workers(&fs_info->workers);
3055         btrfs_stop_workers(&fs_info->endio_workers);
3056         btrfs_stop_workers(&fs_info->endio_meta_workers);
3057         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3058         btrfs_stop_workers(&fs_info->endio_write_workers);
3059         btrfs_stop_workers(&fs_info->endio_freespace_worker);
3060         btrfs_stop_workers(&fs_info->submit_workers);
3061         btrfs_stop_workers(&fs_info->delayed_workers);
3062         btrfs_stop_workers(&fs_info->caching_workers);
3063         btrfs_stop_workers(&fs_info->readahead_workers);
3064 
3065         btrfs_close_devices(fs_info->fs_devices);
3066         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3067 
3068         bdi_destroy(&fs_info->bdi);
3069         cleanup_srcu_struct(&fs_info->subvol_srcu);
3070 
3071         free_fs_info(fs_info);
3072 
3073         return 0;
3074 }
3075 
3076 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
3077 {
3078         int ret;
3079         struct inode *btree_inode = buf->first_page->mapping->host;
3080 
3081         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
3082                                      NULL);
3083         if (!ret)
3084                 return ret;
3085 
3086         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3087                                     parent_transid);
3088         return !ret;
3089 }
3090 
3091 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3092 {
3093         struct inode *btree_inode = buf->first_page->mapping->host;
3094         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
3095                                           buf);
3096 }
3097 
3098 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3099 {
3100         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3101         u64 transid = btrfs_header_generation(buf);
3102         struct inode *btree_inode = root->fs_info->btree_inode;
3103         int was_dirty;
3104 
3105         btrfs_assert_tree_locked(buf);
3106         if (transid != root->fs_info->generation) {
3107                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3108                        "found %llu running %llu\n",
3109                         (unsigned long long)buf->start,
3110                         (unsigned long long)transid,
3111                         (unsigned long long)root->fs_info->generation);
3112                 WARN_ON(1);
3113         }
3114         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
3115                                             buf);
3116         if (!was_dirty) {
3117                 spin_lock(&root->fs_info->delalloc_lock);
3118                 root->fs_info->dirty_metadata_bytes += buf->len;
3119                 spin_unlock(&root->fs_info->delalloc_lock);
3120         }
3121 }
3122 
3123 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3124 {
3125         /*
3126          * looks as though older kernels can get into trouble with
3127          * this code, they end up stuck in balance_dirty_pages forever
3128          */
3129         u64 num_dirty;
3130         unsigned long thresh = 32 * 1024 * 1024;
3131 
3132         if (current->flags & PF_MEMALLOC)
3133                 return;
3134 
3135         btrfs_balance_delayed_items(root);
3136 
3137         num_dirty = root->fs_info->dirty_metadata_bytes;
3138 
3139         if (num_dirty > thresh) {
3140                 balance_dirty_pages_ratelimited_nr(
3141                                    root->fs_info->btree_inode->i_mapping, 1);
3142         }
3143         return;
3144 }
3145 
3146 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3147 {
3148         /*
3149          * looks as though older kernels can get into trouble with
3150          * this code, they end up stuck in balance_dirty_pages forever
3151          */
3152         u64 num_dirty;
3153         unsigned long thresh = 32 * 1024 * 1024;
3154 
3155         if (current->flags & PF_MEMALLOC)
3156                 return;
3157 
3158         num_dirty = root->fs_info->dirty_metadata_bytes;
3159 
3160         if (num_dirty > thresh) {
3161                 balance_dirty_pages_ratelimited_nr(
3162                                    root->fs_info->btree_inode->i_mapping, 1);
3163         }
3164         return;
3165 }
3166 
3167 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3168 {
3169         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3170         int ret;
3171         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3172         if (ret == 0)
3173                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
3174         return ret;
3175 }
3176 
3177 static int btree_lock_page_hook(struct page *page, void *data,
3178                                 void (*flush_fn)(void *))
3179 {
3180         struct inode *inode = page->mapping->host;
3181         struct btrfs_root *root = BTRFS_I(inode)->root;
3182         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3183         struct extent_buffer *eb;
3184         unsigned long len;
3185         u64 bytenr = page_offset(page);
3186 
3187         if (page->private == EXTENT_PAGE_PRIVATE)
3188                 goto out;
3189 
3190         len = page->private >> 2;
3191         eb = find_extent_buffer(io_tree, bytenr, len);
3192         if (!eb)
3193                 goto out;
3194 
3195         if (!btrfs_try_tree_write_lock(eb)) {
3196                 flush_fn(data);
3197                 btrfs_tree_lock(eb);
3198         }
3199         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3200 
3201         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3202                 spin_lock(&root->fs_info->delalloc_lock);
3203                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3204                         root->fs_info->dirty_metadata_bytes -= eb->len;
3205                 else
3206                         WARN_ON(1);
3207                 spin_unlock(&root->fs_info->delalloc_lock);
3208         }
3209 
3210         btrfs_tree_unlock(eb);
3211         free_extent_buffer(eb);
3212 out:
3213         if (!trylock_page(page)) {
3214                 flush_fn(data);
3215                 lock_page(page);
3216         }
3217         return 0;
3218 }
3219 
3220 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3221                               int read_only)
3222 {
3223         if (read_only)
3224                 return;
3225 
3226         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3227                 printk(KERN_WARNING "warning: mount fs with errors, "
3228                        "running btrfsck is recommended\n");
3229 }
3230 
3231 int btrfs_error_commit_super(struct btrfs_root *root)
3232 {
3233         int ret;
3234 
3235         mutex_lock(&root->fs_info->cleaner_mutex);
3236         btrfs_run_delayed_iputs(root);
3237         mutex_unlock(&root->fs_info->cleaner_mutex);
3238 
3239         down_write(&root->fs_info->cleanup_work_sem);
3240         up_write(&root->fs_info->cleanup_work_sem);
3241 
3242         /* cleanup FS via transaction */
3243         btrfs_cleanup_transaction(root);
3244 
3245         ret = write_ctree_super(NULL, root, 0);
3246 
3247         return ret;
3248 }
3249 
3250 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3251 {
3252         struct btrfs_inode *btrfs_inode;
3253         struct list_head splice;
3254 
3255         INIT_LIST_HEAD(&splice);
3256 
3257         mutex_lock(&root->fs_info->ordered_operations_mutex);
3258         spin_lock(&root->fs_info->ordered_extent_lock);
3259 
3260         list_splice_init(&root->fs_info->ordered_operations, &splice);
3261         while (!list_empty(&splice)) {
3262                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3263                                          ordered_operations);
3264 
3265                 list_del_init(&btrfs_inode->ordered_operations);
3266 
3267                 btrfs_invalidate_inodes(btrfs_inode->root);
3268         }
3269 
3270         spin_unlock(&root->fs_info->ordered_extent_lock);
3271         mutex_unlock(&root->fs_info->ordered_operations_mutex);
3272 
3273         return 0;
3274 }
3275 
3276 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3277 {
3278         struct list_head splice;
3279         struct btrfs_ordered_extent *ordered;
3280         struct inode *inode;
3281 
3282         INIT_LIST_HEAD(&splice);
3283 
3284         spin_lock(&root->fs_info->ordered_extent_lock);
3285 
3286         list_splice_init(&root->fs_info->ordered_extents, &splice);
3287         while (!list_empty(&splice)) {
3288                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3289                                      root_extent_list);
3290 
3291                 list_del_init(&ordered->root_extent_list);
3292                 atomic_inc(&ordered->refs);
3293 
3294                 /* the inode may be getting freed (in sys_unlink path). */
3295                 inode = igrab(ordered->inode);
3296 
3297                 spin_unlock(&root->fs_info->ordered_extent_lock);
3298                 if (inode)
3299                         iput(inode);
3300 
3301                 atomic_set(&ordered->refs, 1);
3302                 btrfs_put_ordered_extent(ordered);
3303 
3304                 spin_lock(&root->fs_info->ordered_extent_lock);
3305         }
3306 
3307         spin_unlock(&root->fs_info->ordered_extent_lock);
3308 
3309         return 0;
3310 }
3311 
3312 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3313                                       struct btrfs_root *root)
3314 {
3315         struct rb_node *node;
3316         struct btrfs_delayed_ref_root *delayed_refs;
3317         struct btrfs_delayed_ref_node *ref;
3318         int ret = 0;
3319 
3320         delayed_refs = &trans->delayed_refs;
3321 
3322         spin_lock(&delayed_refs->lock);
3323         if (delayed_refs->num_entries == 0) {
3324                 spin_unlock(&delayed_refs->lock);
3325                 printk(KERN_INFO "delayed_refs has NO entry\n");
3326                 return ret;
3327         }
3328 
3329         node = rb_first(&delayed_refs->root);
3330         while (node) {
3331                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3332                 node = rb_next(node);
3333 
3334                 ref->in_tree = 0;
3335                 rb_erase(&ref->rb_node, &delayed_refs->root);
3336                 delayed_refs->num_entries--;
3337 
3338                 atomic_set(&ref->refs, 1);
3339                 if (btrfs_delayed_ref_is_head(ref)) {
3340                         struct btrfs_delayed_ref_head *head;
3341 
3342                         head = btrfs_delayed_node_to_head(ref);
3343                         mutex_lock(&head->mutex);
3344                         kfree(head->extent_op);
3345                         delayed_refs->num_heads--;
3346                         if (list_empty(&head->cluster))
3347                                 delayed_refs->num_heads_ready--;
3348                         list_del_init(&head->cluster);
3349                         mutex_unlock(&head->mutex);
3350                 }
3351 
3352                 spin_unlock(&delayed_refs->lock);
3353                 btrfs_put_delayed_ref(ref);
3354 
3355                 cond_resched();
3356                 spin_lock(&delayed_refs->lock);
3357         }
3358 
3359         spin_unlock(&delayed_refs->lock);
3360 
3361         return ret;
3362 }
3363 
3364 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3365 {
3366         struct btrfs_pending_snapshot *snapshot;
3367         struct list_head splice;
3368 
3369         INIT_LIST_HEAD(&splice);
3370 
3371         list_splice_init(&t->pending_snapshots, &splice);
3372 
3373         while (!list_empty(&splice)) {
3374                 snapshot = list_entry(splice.next,
3375                                       struct btrfs_pending_snapshot,
3376                                       list);
3377 
3378                 list_del_init(&snapshot->list);
3379 
3380                 kfree(snapshot);
3381         }
3382 
3383         return 0;
3384 }
3385 
3386 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3387 {
3388         struct btrfs_inode *btrfs_inode;
3389         struct list_head splice;
3390 
3391         INIT_LIST_HEAD(&splice);
3392 
3393         spin_lock(&root->fs_info->delalloc_lock);
3394         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3395 
3396         while (!list_empty(&splice)) {
3397                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3398                                     delalloc_inodes);
3399 
3400                 list_del_init(&btrfs_inode->delalloc_inodes);
3401 
3402                 btrfs_invalidate_inodes(btrfs_inode->root);
3403         }
3404 
3405         spin_unlock(&root->fs_info->delalloc_lock);
3406 
3407         return 0;
3408 }
3409 
3410 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3411                                         struct extent_io_tree *dirty_pages,
3412                                         int mark)
3413 {
3414         int ret;
3415         struct page *page;
3416         struct inode *btree_inode = root->fs_info->btree_inode;
3417         struct extent_buffer *eb;
3418         u64 start = 0;
3419         u64 end;
3420         u64 offset;
3421         unsigned long index;
3422 
3423         while (1) {
3424                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3425                                             mark);
3426                 if (ret)
3427                         break;
3428 
3429                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3430                 while (start <= end) {
3431                         index = start >> PAGE_CACHE_SHIFT;
3432                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3433                         page = find_get_page(btree_inode->i_mapping, index);
3434                         if (!page)
3435                                 continue;
3436                         offset = page_offset(page);
3437 
3438                         spin_lock(&dirty_pages->buffer_lock);
3439                         eb = radix_tree_lookup(
3440                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3441                                                offset >> PAGE_CACHE_SHIFT);
3442                         spin_unlock(&dirty_pages->buffer_lock);
3443                         if (eb) {
3444                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3445                                                          &eb->bflags);
3446                                 atomic_set(&eb->refs, 1);
3447                         }
3448                         if (PageWriteback(page))
3449                                 end_page_writeback(page);
3450 
3451                         lock_page(page);
3452                         if (PageDirty(page)) {
3453                                 clear_page_dirty_for_io(page);
3454                                 spin_lock_irq(&page->mapping->tree_lock);
3455                                 radix_tree_tag_clear(&page->mapping->page_tree,
3456                                                         page_index(page),
3457                                                         PAGECACHE_TAG_DIRTY);
3458                                 spin_unlock_irq(&page->mapping->tree_lock);
3459                         }
3460 
3461                         page->mapping->a_ops->invalidatepage(page, 0);
3462                         unlock_page(page);
3463                 }
3464         }
3465 
3466         return ret;
3467 }
3468 
3469 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3470                                        struct extent_io_tree *pinned_extents)
3471 {
3472         struct extent_io_tree *unpin;
3473         u64 start;
3474         u64 end;
3475         int ret;
3476 
3477         unpin = pinned_extents;
3478         while (1) {
3479                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3480                                             EXTENT_DIRTY);
3481                 if (ret)
3482                         break;
3483 
3484                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3485                 btrfs_error_unpin_extent_range(root, start, end);
3486                 cond_resched();
3487         }
3488 
3489         return 0;
3490 }
3491 
3492 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3493 {
3494         struct btrfs_transaction *t;
3495         LIST_HEAD(list);
3496 
3497         WARN_ON(1);
3498 
3499         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3500 
3501         spin_lock(&root->fs_info->trans_lock);
3502         list_splice_init(&root->fs_info->trans_list, &list);
3503         root->fs_info->trans_no_join = 1;
3504         spin_unlock(&root->fs_info->trans_lock);
3505 
3506         while (!list_empty(&list)) {
3507                 t = list_entry(list.next, struct btrfs_transaction, list);
3508                 if (!t)
3509                         break;
3510 
3511                 btrfs_destroy_ordered_operations(root);
3512 
3513                 btrfs_destroy_ordered_extents(root);
3514 
3515                 btrfs_destroy_delayed_refs(t, root);
3516 
3517                 btrfs_block_rsv_release(root,
3518                                         &root->fs_info->trans_block_rsv,
3519                                         t->dirty_pages.dirty_bytes);
3520 
3521                 /* FIXME: cleanup wait for commit */
3522                 t->in_commit = 1;
3523                 t->blocked = 1;
3524                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3525                         wake_up(&root->fs_info->transaction_blocked_wait);
3526 
3527                 t->blocked = 0;
3528                 if (waitqueue_active(&root->fs_info->transaction_wait))
3529                         wake_up(&root->fs_info->transaction_wait);
3530 
3531                 t->commit_done = 1;
3532                 if (waitqueue_active(&t->commit_wait))
3533                         wake_up(&t->commit_wait);
3534 
3535                 btrfs_destroy_pending_snapshots(t);
3536 
3537                 btrfs_destroy_delalloc_inodes(root);
3538 
3539                 spin_lock(&root->fs_info->trans_lock);
3540                 root->fs_info->running_transaction = NULL;
3541                 spin_unlock(&root->fs_info->trans_lock);
3542 
3543                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3544                                              EXTENT_DIRTY);
3545 
3546                 btrfs_destroy_pinned_extent(root,
3547                                             root->fs_info->pinned_extents);
3548 
3549                 atomic_set(&t->use_count, 0);
3550                 list_del_init(&t->list);
3551                 memset(t, 0, sizeof(*t));
3552                 kmem_cache_free(btrfs_transaction_cachep, t);
3553         }
3554 
3555         spin_lock(&root->fs_info->trans_lock);
3556         root->fs_info->trans_no_join = 0;
3557         spin_unlock(&root->fs_info->trans_lock);
3558         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3559 
3560         return 0;
3561 }
3562 
3563 static struct extent_io_ops btree_extent_io_ops = {
3564         .write_cache_pages_lock_hook = btree_lock_page_hook,
3565         .readpage_end_io_hook = btree_readpage_end_io_hook,
3566         .readpage_io_failed_hook = btree_io_failed_hook,
3567         .submit_bio_hook = btree_submit_bio_hook,
3568         /* note we're sharing with inode.c for the merge bio hook */
3569         .merge_bio_hook = btrfs_merge_bio_hook,
3570 };
3571 

~ [ 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