234a08404fc210eac38cacb5087ad06339360703
[pandora-kernel.git] / fs / btrfs / disk-io.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <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 /* These are used to set the lockdep class on the extent buffer locks.
104  * The class is set by the readpage_end_io_hook after the buffer has
105  * passed csum validation but before the pages are unlocked.
106  *
107  * The lockdep class is also set by btrfs_init_new_buffer on freshly
108  * allocated blocks.
109  *
110  * The class is based on the level in the tree block, which allows lockdep
111  * to know that lower nodes nest inside the locks of higher nodes.
112  *
113  * We also add a check to make sure the highest level of the tree is
114  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
115  * code needs update as well.
116  */
117 #ifdef CONFIG_DEBUG_LOCK_ALLOC
118 # if BTRFS_MAX_LEVEL != 8
119 #  error
120 # endif
121 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
122 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
123         /* leaf */
124         "btrfs-extent-00",
125         "btrfs-extent-01",
126         "btrfs-extent-02",
127         "btrfs-extent-03",
128         "btrfs-extent-04",
129         "btrfs-extent-05",
130         "btrfs-extent-06",
131         "btrfs-extent-07",
132         /* highest possible level */
133         "btrfs-extent-08",
134 };
135 #endif
136
137 /*
138  * extents on the btree inode are pretty simple, there's one extent
139  * that covers the entire device
140  */
141 static struct extent_map *btree_get_extent(struct inode *inode,
142                 struct page *page, size_t pg_offset, u64 start, u64 len,
143                 int create)
144 {
145         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
146         struct extent_map *em;
147         int ret;
148
149         read_lock(&em_tree->lock);
150         em = lookup_extent_mapping(em_tree, start, len);
151         if (em) {
152                 em->bdev =
153                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
154                 read_unlock(&em_tree->lock);
155                 goto out;
156         }
157         read_unlock(&em_tree->lock);
158
159         em = alloc_extent_map();
160         if (!em) {
161                 em = ERR_PTR(-ENOMEM);
162                 goto out;
163         }
164         em->start = 0;
165         em->len = (u64)-1;
166         em->block_len = (u64)-1;
167         em->block_start = 0;
168         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
169
170         write_lock(&em_tree->lock);
171         ret = add_extent_mapping(em_tree, em);
172         if (ret == -EEXIST) {
173                 u64 failed_start = em->start;
174                 u64 failed_len = em->len;
175
176                 free_extent_map(em);
177                 em = lookup_extent_mapping(em_tree, start, len);
178                 if (em) {
179                         ret = 0;
180                 } else {
181                         em = lookup_extent_mapping(em_tree, failed_start,
182                                                    failed_len);
183                         ret = -EIO;
184                 }
185         } else if (ret) {
186                 free_extent_map(em);
187                 em = NULL;
188         }
189         write_unlock(&em_tree->lock);
190
191         if (ret)
192                 em = ERR_PTR(ret);
193 out:
194         return em;
195 }
196
197 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
198 {
199         return crc32c(seed, data, len);
200 }
201
202 void btrfs_csum_final(u32 crc, char *result)
203 {
204         put_unaligned_le32(~crc, result);
205 }
206
207 /*
208  * compute the csum for a btree block, and either verify it or write it
209  * into the csum field of the block.
210  */
211 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
212                            int verify)
213 {
214         u16 csum_size =
215                 btrfs_super_csum_size(&root->fs_info->super_copy);
216         char *result = NULL;
217         unsigned long len;
218         unsigned long cur_len;
219         unsigned long offset = BTRFS_CSUM_SIZE;
220         char *map_token = NULL;
221         char *kaddr;
222         unsigned long map_start;
223         unsigned long map_len;
224         int err;
225         u32 crc = ~(u32)0;
226         unsigned long inline_result;
227
228         len = buf->len - offset;
229         while (len > 0) {
230                 err = map_private_extent_buffer(buf, offset, 32,
231                                         &map_token, &kaddr,
232                                         &map_start, &map_len, KM_USER0);
233                 if (err)
234                         return 1;
235                 cur_len = min(len, map_len - (offset - map_start));
236                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
237                                       crc, cur_len);
238                 len -= cur_len;
239                 offset += cur_len;
240                 unmap_extent_buffer(buf, map_token, KM_USER0);
241         }
242         if (csum_size > sizeof(inline_result)) {
243                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
244                 if (!result)
245                         return 1;
246         } else {
247                 result = (char *)&inline_result;
248         }
249
250         btrfs_csum_final(crc, result);
251
252         if (verify) {
253                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
254                         u32 val;
255                         u32 found = 0;
256                         memcpy(&found, result, csum_size);
257
258                         read_extent_buffer(buf, &val, 0, csum_size);
259                         printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
260                                        "failed on %llu wanted %X found %X "
261                                        "level %d\n",
262                                        root->fs_info->sb->s_id,
263                                        (unsigned long long)buf->start, val, found,
264                                        btrfs_header_level(buf));
265                         if (result != (char *)&inline_result)
266                                 kfree(result);
267                         return 1;
268                 }
269         } else {
270                 write_extent_buffer(buf, result, 0, csum_size);
271         }
272         if (result != (char *)&inline_result)
273                 kfree(result);
274         return 0;
275 }
276
277 /*
278  * we can't consider a given block up to date unless the transid of the
279  * block matches the transid in the parent node's pointer.  This is how we
280  * detect blocks that either didn't get written at all or got written
281  * in the wrong place.
282  */
283 static int verify_parent_transid(struct extent_io_tree *io_tree,
284                                  struct extent_buffer *eb, u64 parent_transid)
285 {
286         struct extent_state *cached_state = NULL;
287         int ret;
288
289         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
290                 return 0;
291
292         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
293                          0, &cached_state, GFP_NOFS);
294         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
295             btrfs_header_generation(eb) == parent_transid) {
296                 ret = 0;
297                 goto out;
298         }
299         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
300                        "found %llu\n",
301                        (unsigned long long)eb->start,
302                        (unsigned long long)parent_transid,
303                        (unsigned long long)btrfs_header_generation(eb));
304         ret = 1;
305         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
306 out:
307         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
308                              &cached_state, GFP_NOFS);
309         return ret;
310 }
311
312 /*
313  * helper to read a given tree block, doing retries as required when
314  * the checksums don't match and we have alternate mirrors to try.
315  */
316 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
317                                           struct extent_buffer *eb,
318                                           u64 start, u64 parent_transid)
319 {
320         struct extent_io_tree *io_tree;
321         int ret;
322         int num_copies = 0;
323         int mirror_num = 0;
324
325         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
326         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
327         while (1) {
328                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
329                                                btree_get_extent, mirror_num);
330                 if (!ret &&
331                     !verify_parent_transid(io_tree, eb, parent_transid))
332                         return ret;
333
334                 /*
335                  * This buffer's crc is fine, but its contents are corrupted, so
336                  * there is no reason to read the other copies, they won't be
337                  * any less wrong.
338                  */
339                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
340                         return ret;
341
342                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
343                                               eb->start, eb->len);
344                 if (num_copies == 1)
345                         return ret;
346
347                 mirror_num++;
348                 if (mirror_num > num_copies)
349                         return ret;
350         }
351         return -EIO;
352 }
353
354 /*
355  * checksum a dirty tree block before IO.  This has extra checks to make sure
356  * we only fill in the checksum field in the first page of a multi-page block
357  */
358
359 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
360 {
361         struct extent_io_tree *tree;
362         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
363         u64 found_start;
364         unsigned long len;
365         struct extent_buffer *eb;
366         int ret;
367
368         tree = &BTRFS_I(page->mapping->host)->io_tree;
369
370         if (page->private == EXTENT_PAGE_PRIVATE) {
371                 WARN_ON(1);
372                 goto out;
373         }
374         if (!page->private) {
375                 WARN_ON(1);
376                 goto out;
377         }
378         len = page->private >> 2;
379         WARN_ON(len == 0);
380
381         eb = alloc_extent_buffer(tree, start, len, page);
382         if (eb == NULL) {
383                 WARN_ON(1);
384                 goto out;
385         }
386         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
387                                              btrfs_header_generation(eb));
388         BUG_ON(ret);
389         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
390
391         found_start = btrfs_header_bytenr(eb);
392         if (found_start != start) {
393                 WARN_ON(1);
394                 goto err;
395         }
396         if (eb->first_page != page) {
397                 WARN_ON(1);
398                 goto err;
399         }
400         if (!PageUptodate(page)) {
401                 WARN_ON(1);
402                 goto err;
403         }
404         csum_tree_block(root, eb, 0);
405 err:
406         free_extent_buffer(eb);
407 out:
408         return 0;
409 }
410
411 static int check_tree_block_fsid(struct btrfs_root *root,
412                                  struct extent_buffer *eb)
413 {
414         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
415         u8 fsid[BTRFS_UUID_SIZE];
416         int ret = 1;
417
418         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
419                            BTRFS_FSID_SIZE);
420         while (fs_devices) {
421                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
422                         ret = 0;
423                         break;
424                 }
425                 fs_devices = fs_devices->seed;
426         }
427         return ret;
428 }
429
430 #define CORRUPT(reason, eb, root, slot)                         \
431         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
432                "root=%llu, slot=%d\n", reason,                  \
433                (unsigned long long)btrfs_header_bytenr(eb),     \
434                (unsigned long long)root->objectid, slot)
435
436 static noinline int check_leaf(struct btrfs_root *root,
437                                struct extent_buffer *leaf)
438 {
439         struct btrfs_key key;
440         struct btrfs_key leaf_key;
441         u32 nritems = btrfs_header_nritems(leaf);
442         int slot;
443
444         if (nritems == 0)
445                 return 0;
446
447         /* Check the 0 item */
448         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
449             BTRFS_LEAF_DATA_SIZE(root)) {
450                 CORRUPT("invalid item offset size pair", leaf, root, 0);
451                 return -EIO;
452         }
453
454         /*
455          * Check to make sure each items keys are in the correct order and their
456          * offsets make sense.  We only have to loop through nritems-1 because
457          * we check the current slot against the next slot, which verifies the
458          * next slot's offset+size makes sense and that the current's slot
459          * offset is correct.
460          */
461         for (slot = 0; slot < nritems - 1; slot++) {
462                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
463                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
464
465                 /* Make sure the keys are in the right order */
466                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
467                         CORRUPT("bad key order", leaf, root, slot);
468                         return -EIO;
469                 }
470
471                 /*
472                  * Make sure the offset and ends are right, remember that the
473                  * item data starts at the end of the leaf and grows towards the
474                  * front.
475                  */
476                 if (btrfs_item_offset_nr(leaf, slot) !=
477                         btrfs_item_end_nr(leaf, slot + 1)) {
478                         CORRUPT("slot offset bad", leaf, root, slot);
479                         return -EIO;
480                 }
481
482                 /*
483                  * Check to make sure that we don't point outside of the leaf,
484                  * just incase all the items are consistent to eachother, but
485                  * all point outside of the leaf.
486                  */
487                 if (btrfs_item_end_nr(leaf, slot) >
488                     BTRFS_LEAF_DATA_SIZE(root)) {
489                         CORRUPT("slot end outside of leaf", leaf, root, slot);
490                         return -EIO;
491                 }
492         }
493
494         return 0;
495 }
496
497 #ifdef CONFIG_DEBUG_LOCK_ALLOC
498 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
499 {
500         lockdep_set_class_and_name(&eb->lock,
501                            &btrfs_eb_class[level],
502                            btrfs_eb_name[level]);
503 }
504 #endif
505
506 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
507                                struct extent_state *state)
508 {
509         struct extent_io_tree *tree;
510         u64 found_start;
511         int found_level;
512         unsigned long len;
513         struct extent_buffer *eb;
514         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
515         int ret = 0;
516
517         tree = &BTRFS_I(page->mapping->host)->io_tree;
518         if (page->private == EXTENT_PAGE_PRIVATE)
519                 goto out;
520         if (!page->private)
521                 goto out;
522
523         len = page->private >> 2;
524         WARN_ON(len == 0);
525
526         eb = alloc_extent_buffer(tree, start, len, page);
527         if (eb == NULL) {
528                 ret = -EIO;
529                 goto out;
530         }
531
532         found_start = btrfs_header_bytenr(eb);
533         if (found_start != start) {
534                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
535                                "%llu %llu\n",
536                                (unsigned long long)found_start,
537                                (unsigned long long)eb->start);
538                 ret = -EIO;
539                 goto err;
540         }
541         if (eb->first_page != page) {
542                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
543                        eb->first_page->index, page->index);
544                 WARN_ON(1);
545                 ret = -EIO;
546                 goto err;
547         }
548         if (check_tree_block_fsid(root, eb)) {
549                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
550                                (unsigned long long)eb->start);
551                 ret = -EIO;
552                 goto err;
553         }
554         found_level = btrfs_header_level(eb);
555
556         btrfs_set_buffer_lockdep_class(eb, found_level);
557
558         ret = csum_tree_block(root, eb, 1);
559         if (ret) {
560                 ret = -EIO;
561                 goto err;
562         }
563
564         /*
565          * If this is a leaf block and it is corrupt, set the corrupt bit so
566          * that we don't try and read the other copies of this block, just
567          * return -EIO.
568          */
569         if (found_level == 0 && check_leaf(root, eb)) {
570                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
571                 ret = -EIO;
572         }
573
574         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
575         end = eb->start + end - 1;
576 err:
577         free_extent_buffer(eb);
578 out:
579         return ret;
580 }
581
582 static void end_workqueue_bio(struct bio *bio, int err)
583 {
584         struct end_io_wq *end_io_wq = bio->bi_private;
585         struct btrfs_fs_info *fs_info;
586
587         fs_info = end_io_wq->info;
588         end_io_wq->error = err;
589         end_io_wq->work.func = end_workqueue_fn;
590         end_io_wq->work.flags = 0;
591
592         if (bio->bi_rw & REQ_WRITE) {
593                 if (end_io_wq->metadata == 1)
594                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
595                                            &end_io_wq->work);
596                 else if (end_io_wq->metadata == 2)
597                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
598                                            &end_io_wq->work);
599                 else
600                         btrfs_queue_worker(&fs_info->endio_write_workers,
601                                            &end_io_wq->work);
602         } else {
603                 if (end_io_wq->metadata)
604                         btrfs_queue_worker(&fs_info->endio_meta_workers,
605                                            &end_io_wq->work);
606                 else
607                         btrfs_queue_worker(&fs_info->endio_workers,
608                                            &end_io_wq->work);
609         }
610 }
611
612 /*
613  * For the metadata arg you want
614  *
615  * 0 - if data
616  * 1 - if normal metadta
617  * 2 - if writing to the free space cache area
618  */
619 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
620                         int metadata)
621 {
622         struct end_io_wq *end_io_wq;
623         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
624         if (!end_io_wq)
625                 return -ENOMEM;
626
627         end_io_wq->private = bio->bi_private;
628         end_io_wq->end_io = bio->bi_end_io;
629         end_io_wq->info = info;
630         end_io_wq->error = 0;
631         end_io_wq->bio = bio;
632         end_io_wq->metadata = metadata;
633
634         bio->bi_private = end_io_wq;
635         bio->bi_end_io = end_workqueue_bio;
636         return 0;
637 }
638
639 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
640 {
641         unsigned long limit = min_t(unsigned long,
642                                     info->workers.max_workers,
643                                     info->fs_devices->open_devices);
644         return 256 * limit;
645 }
646
647 static void run_one_async_start(struct btrfs_work *work)
648 {
649         struct async_submit_bio *async;
650
651         async = container_of(work, struct  async_submit_bio, work);
652         async->submit_bio_start(async->inode, async->rw, async->bio,
653                                async->mirror_num, async->bio_flags,
654                                async->bio_offset);
655 }
656
657 static void run_one_async_done(struct btrfs_work *work)
658 {
659         struct btrfs_fs_info *fs_info;
660         struct async_submit_bio *async;
661         int limit;
662
663         async = container_of(work, struct  async_submit_bio, work);
664         fs_info = BTRFS_I(async->inode)->root->fs_info;
665
666         limit = btrfs_async_submit_limit(fs_info);
667         limit = limit * 2 / 3;
668
669         atomic_dec(&fs_info->nr_async_submits);
670
671         if (atomic_read(&fs_info->nr_async_submits) < limit &&
672             waitqueue_active(&fs_info->async_submit_wait))
673                 wake_up(&fs_info->async_submit_wait);
674
675         async->submit_bio_done(async->inode, async->rw, async->bio,
676                                async->mirror_num, async->bio_flags,
677                                async->bio_offset);
678 }
679
680 static void run_one_async_free(struct btrfs_work *work)
681 {
682         struct async_submit_bio *async;
683
684         async = container_of(work, struct  async_submit_bio, work);
685         kfree(async);
686 }
687
688 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
689                         int rw, struct bio *bio, int mirror_num,
690                         unsigned long bio_flags,
691                         u64 bio_offset,
692                         extent_submit_bio_hook_t *submit_bio_start,
693                         extent_submit_bio_hook_t *submit_bio_done)
694 {
695         struct async_submit_bio *async;
696
697         async = kmalloc(sizeof(*async), GFP_NOFS);
698         if (!async)
699                 return -ENOMEM;
700
701         async->inode = inode;
702         async->rw = rw;
703         async->bio = bio;
704         async->mirror_num = mirror_num;
705         async->submit_bio_start = submit_bio_start;
706         async->submit_bio_done = submit_bio_done;
707
708         async->work.func = run_one_async_start;
709         async->work.ordered_func = run_one_async_done;
710         async->work.ordered_free = run_one_async_free;
711
712         async->work.flags = 0;
713         async->bio_flags = bio_flags;
714         async->bio_offset = bio_offset;
715
716         atomic_inc(&fs_info->nr_async_submits);
717
718         if (rw & REQ_SYNC)
719                 btrfs_set_work_high_prio(&async->work);
720
721         btrfs_queue_worker(&fs_info->workers, &async->work);
722
723         while (atomic_read(&fs_info->async_submit_draining) &&
724               atomic_read(&fs_info->nr_async_submits)) {
725                 wait_event(fs_info->async_submit_wait,
726                            (atomic_read(&fs_info->nr_async_submits) == 0));
727         }
728
729         return 0;
730 }
731
732 static int btree_csum_one_bio(struct bio *bio)
733 {
734         struct bio_vec *bvec = bio->bi_io_vec;
735         int bio_index = 0;
736         struct btrfs_root *root;
737
738         WARN_ON(bio->bi_vcnt <= 0);
739         while (bio_index < bio->bi_vcnt) {
740                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
741                 csum_dirty_buffer(root, bvec->bv_page);
742                 bio_index++;
743                 bvec++;
744         }
745         return 0;
746 }
747
748 static int __btree_submit_bio_start(struct inode *inode, int rw,
749                                     struct bio *bio, int mirror_num,
750                                     unsigned long bio_flags,
751                                     u64 bio_offset)
752 {
753         /*
754          * when we're called for a write, we're already in the async
755          * submission context.  Just jump into btrfs_map_bio
756          */
757         btree_csum_one_bio(bio);
758         return 0;
759 }
760
761 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
762                                  int mirror_num, unsigned long bio_flags,
763                                  u64 bio_offset)
764 {
765         /*
766          * when we're called for a write, we're already in the async
767          * submission context.  Just jump into btrfs_map_bio
768          */
769         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
770 }
771
772 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
773                                  int mirror_num, unsigned long bio_flags,
774                                  u64 bio_offset)
775 {
776         int ret;
777
778         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
779                                           bio, 1);
780         BUG_ON(ret);
781
782         if (!(rw & REQ_WRITE)) {
783                 /*
784                  * called for a read, do the setup so that checksum validation
785                  * can happen in the async kernel threads
786                  */
787                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
788                                      mirror_num, 0);
789         }
790
791         /*
792          * kthread helpers are used to submit writes so that checksumming
793          * can happen in parallel across all CPUs
794          */
795         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
796                                    inode, rw, bio, mirror_num, 0,
797                                    bio_offset,
798                                    __btree_submit_bio_start,
799                                    __btree_submit_bio_done);
800 }
801
802 #ifdef CONFIG_MIGRATION
803 static int btree_migratepage(struct address_space *mapping,
804                         struct page *newpage, struct page *page)
805 {
806         /*
807          * we can't safely write a btree page from here,
808          * we haven't done the locking hook
809          */
810         if (PageDirty(page))
811                 return -EAGAIN;
812         /*
813          * Buffers may be managed in a filesystem specific way.
814          * We must have no buffers or drop them.
815          */
816         if (page_has_private(page) &&
817             !try_to_release_page(page, GFP_KERNEL))
818                 return -EAGAIN;
819         return migrate_page(mapping, newpage, page);
820 }
821 #endif
822
823 static int btree_writepage(struct page *page, struct writeback_control *wbc)
824 {
825         struct extent_io_tree *tree;
826         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
827         struct extent_buffer *eb;
828         int was_dirty;
829
830         tree = &BTRFS_I(page->mapping->host)->io_tree;
831         if (!(current->flags & PF_MEMALLOC)) {
832                 return extent_write_full_page(tree, page,
833                                               btree_get_extent, wbc);
834         }
835
836         redirty_page_for_writepage(wbc, page);
837         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
838         WARN_ON(!eb);
839
840         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
841         if (!was_dirty) {
842                 spin_lock(&root->fs_info->delalloc_lock);
843                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
844                 spin_unlock(&root->fs_info->delalloc_lock);
845         }
846         free_extent_buffer(eb);
847
848         unlock_page(page);
849         return 0;
850 }
851
852 static int btree_writepages(struct address_space *mapping,
853                             struct writeback_control *wbc)
854 {
855         struct extent_io_tree *tree;
856         tree = &BTRFS_I(mapping->host)->io_tree;
857         if (wbc->sync_mode == WB_SYNC_NONE) {
858                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
859                 u64 num_dirty;
860                 unsigned long thresh = 32 * 1024 * 1024;
861
862                 if (wbc->for_kupdate)
863                         return 0;
864
865                 /* this is a bit racy, but that's ok */
866                 num_dirty = root->fs_info->dirty_metadata_bytes;
867                 if (num_dirty < thresh)
868                         return 0;
869         }
870         return extent_writepages(tree, mapping, btree_get_extent, wbc);
871 }
872
873 static int btree_readpage(struct file *file, struct page *page)
874 {
875         struct extent_io_tree *tree;
876         tree = &BTRFS_I(page->mapping->host)->io_tree;
877         return extent_read_full_page(tree, page, btree_get_extent);
878 }
879
880 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
881 {
882         struct extent_io_tree *tree;
883         struct extent_map_tree *map;
884         int ret;
885
886         if (PageWriteback(page) || PageDirty(page))
887                 return 0;
888
889         tree = &BTRFS_I(page->mapping->host)->io_tree;
890         map = &BTRFS_I(page->mapping->host)->extent_tree;
891
892         ret = try_release_extent_state(map, tree, page, gfp_flags);
893         if (!ret)
894                 return 0;
895
896         ret = try_release_extent_buffer(tree, page);
897         if (ret == 1) {
898                 ClearPagePrivate(page);
899                 set_page_private(page, 0);
900                 page_cache_release(page);
901         }
902
903         return ret;
904 }
905
906 static void btree_invalidatepage(struct page *page, unsigned long offset)
907 {
908         struct extent_io_tree *tree;
909         tree = &BTRFS_I(page->mapping->host)->io_tree;
910         extent_invalidatepage(tree, page, offset);
911         btree_releasepage(page, GFP_NOFS);
912         if (PagePrivate(page)) {
913                 printk(KERN_WARNING "btrfs warning page private not zero "
914                        "on page %llu\n", (unsigned long long)page_offset(page));
915                 ClearPagePrivate(page);
916                 set_page_private(page, 0);
917                 page_cache_release(page);
918         }
919 }
920
921 static const struct address_space_operations btree_aops = {
922         .readpage       = btree_readpage,
923         .writepage      = btree_writepage,
924         .writepages     = btree_writepages,
925         .releasepage    = btree_releasepage,
926         .invalidatepage = btree_invalidatepage,
927 #ifdef CONFIG_MIGRATION
928         .migratepage    = btree_migratepage,
929 #endif
930 };
931
932 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
933                          u64 parent_transid)
934 {
935         struct extent_buffer *buf = NULL;
936         struct inode *btree_inode = root->fs_info->btree_inode;
937         int ret = 0;
938
939         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
940         if (!buf)
941                 return 0;
942         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
943                                  buf, 0, 0, btree_get_extent, 0);
944         free_extent_buffer(buf);
945         return ret;
946 }
947
948 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
949                                             u64 bytenr, u32 blocksize)
950 {
951         struct inode *btree_inode = root->fs_info->btree_inode;
952         struct extent_buffer *eb;
953         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
954                                 bytenr, blocksize);
955         return eb;
956 }
957
958 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
959                                                  u64 bytenr, u32 blocksize)
960 {
961         struct inode *btree_inode = root->fs_info->btree_inode;
962         struct extent_buffer *eb;
963
964         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
965                                  bytenr, blocksize, NULL);
966         return eb;
967 }
968
969
970 int btrfs_write_tree_block(struct extent_buffer *buf)
971 {
972         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
973                                         buf->start + buf->len - 1);
974 }
975
976 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
977 {
978         return filemap_fdatawait_range(buf->first_page->mapping,
979                                        buf->start, buf->start + buf->len - 1);
980 }
981
982 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
983                                       u32 blocksize, u64 parent_transid)
984 {
985         struct extent_buffer *buf = NULL;
986         int ret;
987
988         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
989         if (!buf)
990                 return NULL;
991
992         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
993
994         if (ret == 0)
995                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
996         return buf;
997
998 }
999
1000 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1001                      struct extent_buffer *buf)
1002 {
1003         struct inode *btree_inode = root->fs_info->btree_inode;
1004         if (btrfs_header_generation(buf) ==
1005             root->fs_info->running_transaction->transid) {
1006                 btrfs_assert_tree_locked(buf);
1007
1008                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1009                         spin_lock(&root->fs_info->delalloc_lock);
1010                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1011                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1012                         else
1013                                 WARN_ON(1);
1014                         spin_unlock(&root->fs_info->delalloc_lock);
1015                 }
1016
1017                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1018                 btrfs_set_lock_blocking(buf);
1019                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1020                                           buf);
1021         }
1022         return 0;
1023 }
1024
1025 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1026                         u32 stripesize, struct btrfs_root *root,
1027                         struct btrfs_fs_info *fs_info,
1028                         u64 objectid)
1029 {
1030         root->node = NULL;
1031         root->commit_root = NULL;
1032         root->sectorsize = sectorsize;
1033         root->nodesize = nodesize;
1034         root->leafsize = leafsize;
1035         root->stripesize = stripesize;
1036         root->ref_cows = 0;
1037         root->track_dirty = 0;
1038         root->in_radix = 0;
1039         root->orphan_item_inserted = 0;
1040         root->orphan_cleanup_state = 0;
1041
1042         root->fs_info = fs_info;
1043         root->objectid = objectid;
1044         root->last_trans = 0;
1045         root->highest_objectid = 0;
1046         root->name = NULL;
1047         root->inode_tree = RB_ROOT;
1048         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1049         root->block_rsv = NULL;
1050         root->orphan_block_rsv = NULL;
1051
1052         INIT_LIST_HEAD(&root->dirty_list);
1053         INIT_LIST_HEAD(&root->orphan_list);
1054         INIT_LIST_HEAD(&root->root_list);
1055         spin_lock_init(&root->orphan_lock);
1056         spin_lock_init(&root->inode_lock);
1057         spin_lock_init(&root->accounting_lock);
1058         mutex_init(&root->objectid_mutex);
1059         mutex_init(&root->log_mutex);
1060         init_waitqueue_head(&root->log_writer_wait);
1061         init_waitqueue_head(&root->log_commit_wait[0]);
1062         init_waitqueue_head(&root->log_commit_wait[1]);
1063         atomic_set(&root->log_commit[0], 0);
1064         atomic_set(&root->log_commit[1], 0);
1065         atomic_set(&root->log_writers, 0);
1066         root->log_batch = 0;
1067         root->log_transid = 0;
1068         root->last_log_commit = 0;
1069         extent_io_tree_init(&root->dirty_log_pages,
1070                              fs_info->btree_inode->i_mapping);
1071
1072         memset(&root->root_key, 0, sizeof(root->root_key));
1073         memset(&root->root_item, 0, sizeof(root->root_item));
1074         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1075         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1076         root->defrag_trans_start = fs_info->generation;
1077         init_completion(&root->kobj_unregister);
1078         root->defrag_running = 0;
1079         root->root_key.objectid = objectid;
1080         root->anon_super.s_root = NULL;
1081         root->anon_super.s_dev = 0;
1082         INIT_LIST_HEAD(&root->anon_super.s_list);
1083         INIT_LIST_HEAD(&root->anon_super.s_instances);
1084         init_rwsem(&root->anon_super.s_umount);
1085
1086         return 0;
1087 }
1088
1089 static int find_and_setup_root(struct btrfs_root *tree_root,
1090                                struct btrfs_fs_info *fs_info,
1091                                u64 objectid,
1092                                struct btrfs_root *root)
1093 {
1094         int ret;
1095         u32 blocksize;
1096         u64 generation;
1097
1098         __setup_root(tree_root->nodesize, tree_root->leafsize,
1099                      tree_root->sectorsize, tree_root->stripesize,
1100                      root, fs_info, objectid);
1101         ret = btrfs_find_last_root(tree_root, objectid,
1102                                    &root->root_item, &root->root_key);
1103         if (ret > 0)
1104                 return -ENOENT;
1105         BUG_ON(ret);
1106
1107         generation = btrfs_root_generation(&root->root_item);
1108         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1109         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1110                                      blocksize, generation);
1111         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1112                 free_extent_buffer(root->node);
1113                 return -EIO;
1114         }
1115         root->commit_root = btrfs_root_node(root);
1116         return 0;
1117 }
1118
1119 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1120                                          struct btrfs_fs_info *fs_info)
1121 {
1122         struct btrfs_root *root;
1123         struct btrfs_root *tree_root = fs_info->tree_root;
1124         struct extent_buffer *leaf;
1125
1126         root = kzalloc(sizeof(*root), GFP_NOFS);
1127         if (!root)
1128                 return ERR_PTR(-ENOMEM);
1129
1130         __setup_root(tree_root->nodesize, tree_root->leafsize,
1131                      tree_root->sectorsize, tree_root->stripesize,
1132                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1133
1134         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1135         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1136         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1137         /*
1138          * log trees do not get reference counted because they go away
1139          * before a real commit is actually done.  They do store pointers
1140          * to file data extents, and those reference counts still get
1141          * updated (along with back refs to the log tree).
1142          */
1143         root->ref_cows = 0;
1144
1145         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1146                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1147         if (IS_ERR(leaf)) {
1148                 kfree(root);
1149                 return ERR_CAST(leaf);
1150         }
1151
1152         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1153         btrfs_set_header_bytenr(leaf, leaf->start);
1154         btrfs_set_header_generation(leaf, trans->transid);
1155         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1156         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1157         root->node = leaf;
1158
1159         write_extent_buffer(root->node, root->fs_info->fsid,
1160                             (unsigned long)btrfs_header_fsid(root->node),
1161                             BTRFS_FSID_SIZE);
1162         btrfs_mark_buffer_dirty(root->node);
1163         btrfs_tree_unlock(root->node);
1164         return root;
1165 }
1166
1167 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1168                              struct btrfs_fs_info *fs_info)
1169 {
1170         struct btrfs_root *log_root;
1171
1172         log_root = alloc_log_tree(trans, fs_info);
1173         if (IS_ERR(log_root))
1174                 return PTR_ERR(log_root);
1175         WARN_ON(fs_info->log_root_tree);
1176         fs_info->log_root_tree = log_root;
1177         return 0;
1178 }
1179
1180 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1181                        struct btrfs_root *root)
1182 {
1183         struct btrfs_root *log_root;
1184         struct btrfs_inode_item *inode_item;
1185
1186         log_root = alloc_log_tree(trans, root->fs_info);
1187         if (IS_ERR(log_root))
1188                 return PTR_ERR(log_root);
1189
1190         log_root->last_trans = trans->transid;
1191         log_root->root_key.offset = root->root_key.objectid;
1192
1193         inode_item = &log_root->root_item.inode;
1194         inode_item->generation = cpu_to_le64(1);
1195         inode_item->size = cpu_to_le64(3);
1196         inode_item->nlink = cpu_to_le32(1);
1197         inode_item->nbytes = cpu_to_le64(root->leafsize);
1198         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1199
1200         btrfs_set_root_node(&log_root->root_item, log_root->node);
1201
1202         WARN_ON(root->log_root);
1203         root->log_root = log_root;
1204         root->log_transid = 0;
1205         root->last_log_commit = 0;
1206         return 0;
1207 }
1208
1209 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1210                                                struct btrfs_key *location)
1211 {
1212         struct btrfs_root *root;
1213         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1214         struct btrfs_path *path;
1215         struct extent_buffer *l;
1216         u64 generation;
1217         u32 blocksize;
1218         int ret = 0;
1219
1220         root = kzalloc(sizeof(*root), GFP_NOFS);
1221         if (!root)
1222                 return ERR_PTR(-ENOMEM);
1223         if (location->offset == (u64)-1) {
1224                 ret = find_and_setup_root(tree_root, fs_info,
1225                                           location->objectid, root);
1226                 if (ret) {
1227                         kfree(root);
1228                         return ERR_PTR(ret);
1229                 }
1230                 goto out;
1231         }
1232
1233         __setup_root(tree_root->nodesize, tree_root->leafsize,
1234                      tree_root->sectorsize, tree_root->stripesize,
1235                      root, fs_info, location->objectid);
1236
1237         path = btrfs_alloc_path();
1238         if (!path) {
1239                 kfree(root);
1240                 return ERR_PTR(-ENOMEM);
1241         }
1242         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1243         if (ret == 0) {
1244                 l = path->nodes[0];
1245                 read_extent_buffer(l, &root->root_item,
1246                                 btrfs_item_ptr_offset(l, path->slots[0]),
1247                                 sizeof(root->root_item));
1248                 memcpy(&root->root_key, location, sizeof(*location));
1249         }
1250         btrfs_free_path(path);
1251         if (ret) {
1252                 kfree(root);
1253                 if (ret > 0)
1254                         ret = -ENOENT;
1255                 return ERR_PTR(ret);
1256         }
1257
1258         generation = btrfs_root_generation(&root->root_item);
1259         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1260         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1261                                      blocksize, generation);
1262         root->commit_root = btrfs_root_node(root);
1263         BUG_ON(!root->node);
1264 out:
1265         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1266                 root->ref_cows = 1;
1267                 btrfs_check_and_init_root_item(&root->root_item);
1268         }
1269
1270         return root;
1271 }
1272
1273 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1274                                               struct btrfs_key *location)
1275 {
1276         struct btrfs_root *root;
1277         int ret;
1278
1279         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1280                 return fs_info->tree_root;
1281         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1282                 return fs_info->extent_root;
1283         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1284                 return fs_info->chunk_root;
1285         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1286                 return fs_info->dev_root;
1287         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1288                 return fs_info->csum_root;
1289 again:
1290         spin_lock(&fs_info->fs_roots_radix_lock);
1291         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1292                                  (unsigned long)location->objectid);
1293         spin_unlock(&fs_info->fs_roots_radix_lock);
1294         if (root)
1295                 return root;
1296
1297         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1298         if (IS_ERR(root))
1299                 return root;
1300
1301         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1302         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1303                                         GFP_NOFS);
1304         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1305                 ret = -ENOMEM;
1306                 goto fail;
1307         }
1308
1309         btrfs_init_free_ino_ctl(root);
1310         mutex_init(&root->fs_commit_mutex);
1311         spin_lock_init(&root->cache_lock);
1312         init_waitqueue_head(&root->cache_wait);
1313
1314         ret = set_anon_super(&root->anon_super, NULL);
1315         if (ret)
1316                 goto fail;
1317
1318         if (btrfs_root_refs(&root->root_item) == 0) {
1319                 ret = -ENOENT;
1320                 goto fail;
1321         }
1322
1323         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1324         if (ret < 0)
1325                 goto fail;
1326         if (ret == 0)
1327                 root->orphan_item_inserted = 1;
1328
1329         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1330         if (ret)
1331                 goto fail;
1332
1333         spin_lock(&fs_info->fs_roots_radix_lock);
1334         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1335                                 (unsigned long)root->root_key.objectid,
1336                                 root);
1337         if (ret == 0)
1338                 root->in_radix = 1;
1339
1340         spin_unlock(&fs_info->fs_roots_radix_lock);
1341         radix_tree_preload_end();
1342         if (ret) {
1343                 if (ret == -EEXIST) {
1344                         free_fs_root(root);
1345                         goto again;
1346                 }
1347                 goto fail;
1348         }
1349
1350         ret = btrfs_find_dead_roots(fs_info->tree_root,
1351                                     root->root_key.objectid);
1352         WARN_ON(ret);
1353         return root;
1354 fail:
1355         free_fs_root(root);
1356         return ERR_PTR(ret);
1357 }
1358
1359 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1360 {
1361         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1362         int ret = 0;
1363         struct btrfs_device *device;
1364         struct backing_dev_info *bdi;
1365
1366         rcu_read_lock();
1367         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1368                 if (!device->bdev)
1369                         continue;
1370                 bdi = blk_get_backing_dev_info(device->bdev);
1371                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1372                         ret = 1;
1373                         break;
1374                 }
1375         }
1376         rcu_read_unlock();
1377         return ret;
1378 }
1379
1380 /*
1381  * If this fails, caller must call bdi_destroy() to get rid of the
1382  * bdi again.
1383  */
1384 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1385 {
1386         int err;
1387
1388         bdi->capabilities = BDI_CAP_MAP_COPY;
1389         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1390         if (err)
1391                 return err;
1392
1393         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1394         bdi->congested_fn       = btrfs_congested_fn;
1395         bdi->congested_data     = info;
1396         return 0;
1397 }
1398
1399 static int bio_ready_for_csum(struct bio *bio)
1400 {
1401         u64 length = 0;
1402         u64 buf_len = 0;
1403         u64 start = 0;
1404         struct page *page;
1405         struct extent_io_tree *io_tree = NULL;
1406         struct bio_vec *bvec;
1407         int i;
1408         int ret;
1409
1410         bio_for_each_segment(bvec, bio, i) {
1411                 page = bvec->bv_page;
1412                 if (page->private == EXTENT_PAGE_PRIVATE) {
1413                         length += bvec->bv_len;
1414                         continue;
1415                 }
1416                 if (!page->private) {
1417                         length += bvec->bv_len;
1418                         continue;
1419                 }
1420                 length = bvec->bv_len;
1421                 buf_len = page->private >> 2;
1422                 start = page_offset(page) + bvec->bv_offset;
1423                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1424         }
1425         /* are we fully contained in this bio? */
1426         if (buf_len <= length)
1427                 return 1;
1428
1429         ret = extent_range_uptodate(io_tree, start + length,
1430                                     start + buf_len - 1);
1431         return ret;
1432 }
1433
1434 /*
1435  * called by the kthread helper functions to finally call the bio end_io
1436  * functions.  This is where read checksum verification actually happens
1437  */
1438 static void end_workqueue_fn(struct btrfs_work *work)
1439 {
1440         struct bio *bio;
1441         struct end_io_wq *end_io_wq;
1442         struct btrfs_fs_info *fs_info;
1443         int error;
1444
1445         end_io_wq = container_of(work, struct end_io_wq, work);
1446         bio = end_io_wq->bio;
1447         fs_info = end_io_wq->info;
1448
1449         /* metadata bio reads are special because the whole tree block must
1450          * be checksummed at once.  This makes sure the entire block is in
1451          * ram and up to date before trying to verify things.  For
1452          * blocksize <= pagesize, it is basically a noop
1453          */
1454         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1455             !bio_ready_for_csum(bio)) {
1456                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1457                                    &end_io_wq->work);
1458                 return;
1459         }
1460         error = end_io_wq->error;
1461         bio->bi_private = end_io_wq->private;
1462         bio->bi_end_io = end_io_wq->end_io;
1463         kfree(end_io_wq);
1464         bio_endio(bio, error);
1465 }
1466
1467 static int cleaner_kthread(void *arg)
1468 {
1469         struct btrfs_root *root = arg;
1470
1471         do {
1472                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1473
1474                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1475                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1476                         btrfs_run_delayed_iputs(root);
1477                         btrfs_clean_old_snapshots(root);
1478                         mutex_unlock(&root->fs_info->cleaner_mutex);
1479                         btrfs_run_defrag_inodes(root->fs_info);
1480                 }
1481
1482                 if (freezing(current)) {
1483                         refrigerator();
1484                 } else {
1485                         set_current_state(TASK_INTERRUPTIBLE);
1486                         if (!kthread_should_stop())
1487                                 schedule();
1488                         __set_current_state(TASK_RUNNING);
1489                 }
1490         } while (!kthread_should_stop());
1491         return 0;
1492 }
1493
1494 static int transaction_kthread(void *arg)
1495 {
1496         struct btrfs_root *root = arg;
1497         struct btrfs_trans_handle *trans;
1498         struct btrfs_transaction *cur;
1499         u64 transid;
1500         unsigned long now;
1501         unsigned long delay;
1502         int ret;
1503
1504         do {
1505                 delay = HZ * 30;
1506                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1507                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1508
1509                 spin_lock(&root->fs_info->trans_lock);
1510                 cur = root->fs_info->running_transaction;
1511                 if (!cur) {
1512                         spin_unlock(&root->fs_info->trans_lock);
1513                         goto sleep;
1514                 }
1515
1516                 now = get_seconds();
1517                 if (!cur->blocked &&
1518                     (now < cur->start_time || now - cur->start_time < 30)) {
1519                         spin_unlock(&root->fs_info->trans_lock);
1520                         delay = HZ * 5;
1521                         goto sleep;
1522                 }
1523                 transid = cur->transid;
1524                 spin_unlock(&root->fs_info->trans_lock);
1525
1526                 trans = btrfs_join_transaction(root);
1527                 BUG_ON(IS_ERR(trans));
1528                 if (transid == trans->transid) {
1529                         ret = btrfs_commit_transaction(trans, root);
1530                         BUG_ON(ret);
1531                 } else {
1532                         btrfs_end_transaction(trans, root);
1533                 }
1534 sleep:
1535                 wake_up_process(root->fs_info->cleaner_kthread);
1536                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1537
1538                 if (freezing(current)) {
1539                         refrigerator();
1540                 } else {
1541                         set_current_state(TASK_INTERRUPTIBLE);
1542                         if (!kthread_should_stop() &&
1543                             !btrfs_transaction_blocked(root->fs_info))
1544                                 schedule_timeout(delay);
1545                         __set_current_state(TASK_RUNNING);
1546                 }
1547         } while (!kthread_should_stop());
1548         return 0;
1549 }
1550
1551 struct btrfs_root *open_ctree(struct super_block *sb,
1552                               struct btrfs_fs_devices *fs_devices,
1553                               char *options)
1554 {
1555         u32 sectorsize;
1556         u32 nodesize;
1557         u32 leafsize;
1558         u32 blocksize;
1559         u32 stripesize;
1560         u64 generation;
1561         u64 features;
1562         struct btrfs_key location;
1563         struct buffer_head *bh;
1564         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1565                                                  GFP_NOFS);
1566         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1567                                                  GFP_NOFS);
1568         struct btrfs_root *tree_root = btrfs_sb(sb);
1569         struct btrfs_fs_info *fs_info = NULL;
1570         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1571                                                 GFP_NOFS);
1572         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1573                                               GFP_NOFS);
1574         struct btrfs_root *log_tree_root;
1575
1576         int ret;
1577         int err = -EINVAL;
1578
1579         struct btrfs_super_block *disk_super;
1580
1581         if (!extent_root || !tree_root || !tree_root->fs_info ||
1582             !chunk_root || !dev_root || !csum_root) {
1583                 err = -ENOMEM;
1584                 goto fail;
1585         }
1586         fs_info = tree_root->fs_info;
1587
1588         ret = init_srcu_struct(&fs_info->subvol_srcu);
1589         if (ret) {
1590                 err = ret;
1591                 goto fail;
1592         }
1593
1594         ret = setup_bdi(fs_info, &fs_info->bdi);
1595         if (ret) {
1596                 err = ret;
1597                 goto fail_srcu;
1598         }
1599
1600         fs_info->btree_inode = new_inode(sb);
1601         if (!fs_info->btree_inode) {
1602                 err = -ENOMEM;
1603                 goto fail_bdi;
1604         }
1605
1606         fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1607
1608         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1609         INIT_LIST_HEAD(&fs_info->trans_list);
1610         INIT_LIST_HEAD(&fs_info->dead_roots);
1611         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1612         INIT_LIST_HEAD(&fs_info->hashers);
1613         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1614         INIT_LIST_HEAD(&fs_info->ordered_operations);
1615         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1616         spin_lock_init(&fs_info->delalloc_lock);
1617         spin_lock_init(&fs_info->trans_lock);
1618         spin_lock_init(&fs_info->ref_cache_lock);
1619         spin_lock_init(&fs_info->fs_roots_radix_lock);
1620         spin_lock_init(&fs_info->delayed_iput_lock);
1621         spin_lock_init(&fs_info->defrag_inodes_lock);
1622         mutex_init(&fs_info->reloc_mutex);
1623
1624         init_completion(&fs_info->kobj_unregister);
1625         fs_info->tree_root = tree_root;
1626         fs_info->extent_root = extent_root;
1627         fs_info->csum_root = csum_root;
1628         fs_info->chunk_root = chunk_root;
1629         fs_info->dev_root = dev_root;
1630         fs_info->fs_devices = fs_devices;
1631         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1632         INIT_LIST_HEAD(&fs_info->space_info);
1633         btrfs_mapping_init(&fs_info->mapping_tree);
1634         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1635         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1636         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1637         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1638         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1639         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1640         mutex_init(&fs_info->durable_block_rsv_mutex);
1641         atomic_set(&fs_info->nr_async_submits, 0);
1642         atomic_set(&fs_info->async_delalloc_pages, 0);
1643         atomic_set(&fs_info->async_submit_draining, 0);
1644         atomic_set(&fs_info->nr_async_bios, 0);
1645         atomic_set(&fs_info->defrag_running, 0);
1646         fs_info->sb = sb;
1647         fs_info->max_inline = 8192 * 1024;
1648         fs_info->metadata_ratio = 0;
1649         fs_info->defrag_inodes = RB_ROOT;
1650         fs_info->trans_no_join = 0;
1651
1652         fs_info->thread_pool_size = min_t(unsigned long,
1653                                           num_online_cpus() + 2, 8);
1654
1655         INIT_LIST_HEAD(&fs_info->ordered_extents);
1656         spin_lock_init(&fs_info->ordered_extent_lock);
1657         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1658                                         GFP_NOFS);
1659         if (!fs_info->delayed_root) {
1660                 err = -ENOMEM;
1661                 goto fail_iput;
1662         }
1663         btrfs_init_delayed_root(fs_info->delayed_root);
1664
1665         mutex_init(&fs_info->scrub_lock);
1666         atomic_set(&fs_info->scrubs_running, 0);
1667         atomic_set(&fs_info->scrub_pause_req, 0);
1668         atomic_set(&fs_info->scrubs_paused, 0);
1669         atomic_set(&fs_info->scrub_cancel_req, 0);
1670         init_waitqueue_head(&fs_info->scrub_pause_wait);
1671         init_rwsem(&fs_info->scrub_super_lock);
1672         fs_info->scrub_workers_refcnt = 0;
1673
1674         sb->s_blocksize = 4096;
1675         sb->s_blocksize_bits = blksize_bits(4096);
1676         sb->s_bdi = &fs_info->bdi;
1677
1678         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1679         fs_info->btree_inode->i_nlink = 1;
1680         /*
1681          * we set the i_size on the btree inode to the max possible int.
1682          * the real end of the address space is determined by all of
1683          * the devices in the system
1684          */
1685         fs_info->btree_inode->i_size = OFFSET_MAX;
1686         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1687         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1688
1689         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1690         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1691                              fs_info->btree_inode->i_mapping);
1692         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1693
1694         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1695
1696         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1697         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1698                sizeof(struct btrfs_key));
1699         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1700         insert_inode_hash(fs_info->btree_inode);
1701
1702         spin_lock_init(&fs_info->block_group_cache_lock);
1703         fs_info->block_group_cache_tree = RB_ROOT;
1704
1705         extent_io_tree_init(&fs_info->freed_extents[0],
1706                              fs_info->btree_inode->i_mapping);
1707         extent_io_tree_init(&fs_info->freed_extents[1],
1708                              fs_info->btree_inode->i_mapping);
1709         fs_info->pinned_extents = &fs_info->freed_extents[0];
1710         fs_info->do_barriers = 1;
1711
1712
1713         mutex_init(&fs_info->ordered_operations_mutex);
1714         mutex_init(&fs_info->tree_log_mutex);
1715         mutex_init(&fs_info->chunk_mutex);
1716         mutex_init(&fs_info->transaction_kthread_mutex);
1717         mutex_init(&fs_info->cleaner_mutex);
1718         mutex_init(&fs_info->volume_mutex);
1719         init_rwsem(&fs_info->extent_commit_sem);
1720         init_rwsem(&fs_info->cleanup_work_sem);
1721         init_rwsem(&fs_info->subvol_sem);
1722
1723         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1724         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1725
1726         init_waitqueue_head(&fs_info->transaction_throttle);
1727         init_waitqueue_head(&fs_info->transaction_wait);
1728         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1729         init_waitqueue_head(&fs_info->async_submit_wait);
1730
1731         __setup_root(4096, 4096, 4096, 4096, tree_root,
1732                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1733
1734         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1735         if (!bh) {
1736                 err = -EINVAL;
1737                 goto fail_alloc;
1738         }
1739
1740         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1741         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1742                sizeof(fs_info->super_for_commit));
1743         brelse(bh);
1744
1745         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1746
1747         disk_super = &fs_info->super_copy;
1748         if (!btrfs_super_root(disk_super))
1749                 goto fail_alloc;
1750
1751         /* check FS state, whether FS is broken. */
1752         fs_info->fs_state |= btrfs_super_flags(disk_super);
1753
1754         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1755
1756         /*
1757          * In the long term, we'll store the compression type in the super
1758          * block, and it'll be used for per file compression control.
1759          */
1760         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1761
1762         ret = btrfs_parse_options(tree_root, options);
1763         if (ret) {
1764                 err = ret;
1765                 goto fail_alloc;
1766         }
1767
1768         features = btrfs_super_incompat_flags(disk_super) &
1769                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1770         if (features) {
1771                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1772                        "unsupported optional features (%Lx).\n",
1773                        (unsigned long long)features);
1774                 err = -EINVAL;
1775                 goto fail_alloc;
1776         }
1777
1778         features = btrfs_super_incompat_flags(disk_super);
1779         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1780         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1781                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1782         btrfs_set_super_incompat_flags(disk_super, features);
1783
1784         features = btrfs_super_compat_ro_flags(disk_super) &
1785                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1786         if (!(sb->s_flags & MS_RDONLY) && features) {
1787                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1788                        "unsupported option features (%Lx).\n",
1789                        (unsigned long long)features);
1790                 err = -EINVAL;
1791                 goto fail_alloc;
1792         }
1793
1794         btrfs_init_workers(&fs_info->generic_worker,
1795                            "genwork", 1, NULL);
1796
1797         btrfs_init_workers(&fs_info->workers, "worker",
1798                            fs_info->thread_pool_size,
1799                            &fs_info->generic_worker);
1800
1801         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1802                            fs_info->thread_pool_size,
1803                            &fs_info->generic_worker);
1804
1805         btrfs_init_workers(&fs_info->submit_workers, "submit",
1806                            min_t(u64, fs_devices->num_devices,
1807                            fs_info->thread_pool_size),
1808                            &fs_info->generic_worker);
1809
1810         btrfs_init_workers(&fs_info->caching_workers, "cache",
1811                            2, &fs_info->generic_worker);
1812
1813         /* a higher idle thresh on the submit workers makes it much more
1814          * likely that bios will be send down in a sane order to the
1815          * devices
1816          */
1817         fs_info->submit_workers.idle_thresh = 64;
1818
1819         fs_info->workers.idle_thresh = 16;
1820         fs_info->workers.ordered = 1;
1821
1822         fs_info->delalloc_workers.idle_thresh = 2;
1823         fs_info->delalloc_workers.ordered = 1;
1824
1825         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1826                            &fs_info->generic_worker);
1827         btrfs_init_workers(&fs_info->endio_workers, "endio",
1828                            fs_info->thread_pool_size,
1829                            &fs_info->generic_worker);
1830         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1831                            fs_info->thread_pool_size,
1832                            &fs_info->generic_worker);
1833         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1834                            "endio-meta-write", fs_info->thread_pool_size,
1835                            &fs_info->generic_worker);
1836         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1837                            fs_info->thread_pool_size,
1838                            &fs_info->generic_worker);
1839         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1840                            1, &fs_info->generic_worker);
1841         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1842                            fs_info->thread_pool_size,
1843                            &fs_info->generic_worker);
1844
1845         /*
1846          * endios are largely parallel and should have a very
1847          * low idle thresh
1848          */
1849         fs_info->endio_workers.idle_thresh = 4;
1850         fs_info->endio_meta_workers.idle_thresh = 4;
1851
1852         fs_info->endio_write_workers.idle_thresh = 2;
1853         fs_info->endio_meta_write_workers.idle_thresh = 2;
1854
1855         btrfs_start_workers(&fs_info->workers, 1);
1856         btrfs_start_workers(&fs_info->generic_worker, 1);
1857         btrfs_start_workers(&fs_info->submit_workers, 1);
1858         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1859         btrfs_start_workers(&fs_info->fixup_workers, 1);
1860         btrfs_start_workers(&fs_info->endio_workers, 1);
1861         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1862         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1863         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1864         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1865         btrfs_start_workers(&fs_info->delayed_workers, 1);
1866         btrfs_start_workers(&fs_info->caching_workers, 1);
1867
1868         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1869         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1870                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1871
1872         nodesize = btrfs_super_nodesize(disk_super);
1873         leafsize = btrfs_super_leafsize(disk_super);
1874         sectorsize = btrfs_super_sectorsize(disk_super);
1875         stripesize = btrfs_super_stripesize(disk_super);
1876         tree_root->nodesize = nodesize;
1877         tree_root->leafsize = leafsize;
1878         tree_root->sectorsize = sectorsize;
1879         tree_root->stripesize = stripesize;
1880
1881         sb->s_blocksize = sectorsize;
1882         sb->s_blocksize_bits = blksize_bits(sectorsize);
1883
1884         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1885                     sizeof(disk_super->magic))) {
1886                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1887                 goto fail_sb_buffer;
1888         }
1889
1890         mutex_lock(&fs_info->chunk_mutex);
1891         ret = btrfs_read_sys_array(tree_root);
1892         mutex_unlock(&fs_info->chunk_mutex);
1893         if (ret) {
1894                 printk(KERN_WARNING "btrfs: failed to read the system "
1895                        "array on %s\n", sb->s_id);
1896                 goto fail_sb_buffer;
1897         }
1898
1899         blocksize = btrfs_level_size(tree_root,
1900                                      btrfs_super_chunk_root_level(disk_super));
1901         generation = btrfs_super_chunk_root_generation(disk_super);
1902
1903         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1904                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1905
1906         chunk_root->node = read_tree_block(chunk_root,
1907                                            btrfs_super_chunk_root(disk_super),
1908                                            blocksize, generation);
1909         BUG_ON(!chunk_root->node);
1910         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1911                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1912                        sb->s_id);
1913                 goto fail_chunk_root;
1914         }
1915         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1916         chunk_root->commit_root = btrfs_root_node(chunk_root);
1917
1918         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1919            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1920            BTRFS_UUID_SIZE);
1921
1922         mutex_lock(&fs_info->chunk_mutex);
1923         ret = btrfs_read_chunk_tree(chunk_root);
1924         mutex_unlock(&fs_info->chunk_mutex);
1925         if (ret) {
1926                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1927                        sb->s_id);
1928                 goto fail_chunk_root;
1929         }
1930
1931         btrfs_close_extra_devices(fs_devices);
1932
1933         blocksize = btrfs_level_size(tree_root,
1934                                      btrfs_super_root_level(disk_super));
1935         generation = btrfs_super_generation(disk_super);
1936
1937         tree_root->node = read_tree_block(tree_root,
1938                                           btrfs_super_root(disk_super),
1939                                           blocksize, generation);
1940         if (!tree_root->node)
1941                 goto fail_chunk_root;
1942         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1943                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1944                        sb->s_id);
1945                 goto fail_tree_root;
1946         }
1947         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1948         tree_root->commit_root = btrfs_root_node(tree_root);
1949
1950         ret = find_and_setup_root(tree_root, fs_info,
1951                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1952         if (ret)
1953                 goto fail_tree_root;
1954         extent_root->track_dirty = 1;
1955
1956         ret = find_and_setup_root(tree_root, fs_info,
1957                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1958         if (ret)
1959                 goto fail_extent_root;
1960         dev_root->track_dirty = 1;
1961
1962         ret = find_and_setup_root(tree_root, fs_info,
1963                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1964         if (ret)
1965                 goto fail_dev_root;
1966
1967         csum_root->track_dirty = 1;
1968
1969         fs_info->generation = generation;
1970         fs_info->last_trans_committed = generation;
1971         fs_info->data_alloc_profile = (u64)-1;
1972         fs_info->metadata_alloc_profile = (u64)-1;
1973         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1974
1975         ret = btrfs_init_space_info(fs_info);
1976         if (ret) {
1977                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
1978                 goto fail_block_groups;
1979         }
1980
1981         ret = btrfs_read_block_groups(extent_root);
1982         if (ret) {
1983                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1984                 goto fail_block_groups;
1985         }
1986
1987         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1988                                                "btrfs-cleaner");
1989         if (IS_ERR(fs_info->cleaner_kthread))
1990                 goto fail_block_groups;
1991
1992         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1993                                                    tree_root,
1994                                                    "btrfs-transaction");
1995         if (IS_ERR(fs_info->transaction_kthread))
1996                 goto fail_cleaner;
1997
1998         if (!btrfs_test_opt(tree_root, SSD) &&
1999             !btrfs_test_opt(tree_root, NOSSD) &&
2000             !fs_info->fs_devices->rotating) {
2001                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2002                        "mode\n");
2003                 btrfs_set_opt(fs_info->mount_opt, SSD);
2004         }
2005
2006         /* do not make disk changes in broken FS */
2007         if (btrfs_super_log_root(disk_super) != 0 &&
2008             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2009                 u64 bytenr = btrfs_super_log_root(disk_super);
2010
2011                 if (fs_devices->rw_devices == 0) {
2012                         printk(KERN_WARNING "Btrfs log replay required "
2013                                "on RO media\n");
2014                         err = -EIO;
2015                         goto fail_trans_kthread;
2016                 }
2017                 blocksize =
2018                      btrfs_level_size(tree_root,
2019                                       btrfs_super_log_root_level(disk_super));
2020
2021                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2022                 if (!log_tree_root) {
2023                         err = -ENOMEM;
2024                         goto fail_trans_kthread;
2025                 }
2026
2027                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2028                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2029
2030                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2031                                                       blocksize,
2032                                                       generation + 1);
2033                 ret = btrfs_recover_log_trees(log_tree_root);
2034                 BUG_ON(ret);
2035
2036                 if (sb->s_flags & MS_RDONLY) {
2037                         ret =  btrfs_commit_super(tree_root);
2038                         BUG_ON(ret);
2039                 }
2040         }
2041
2042         ret = btrfs_find_orphan_roots(tree_root);
2043         BUG_ON(ret);
2044
2045         if (!(sb->s_flags & MS_RDONLY)) {
2046                 ret = btrfs_cleanup_fs_roots(fs_info);
2047                 BUG_ON(ret);
2048
2049                 ret = btrfs_recover_relocation(tree_root);
2050                 if (ret < 0) {
2051                         printk(KERN_WARNING
2052                                "btrfs: failed to recover relocation\n");
2053                         err = -EINVAL;
2054                         goto fail_trans_kthread;
2055                 }
2056         }
2057
2058         location.objectid = BTRFS_FS_TREE_OBJECTID;
2059         location.type = BTRFS_ROOT_ITEM_KEY;
2060         location.offset = (u64)-1;
2061
2062         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2063         if (!fs_info->fs_root)
2064                 goto fail_trans_kthread;
2065         if (IS_ERR(fs_info->fs_root)) {
2066                 err = PTR_ERR(fs_info->fs_root);
2067                 goto fail_trans_kthread;
2068         }
2069
2070         if (!(sb->s_flags & MS_RDONLY)) {
2071                 down_read(&fs_info->cleanup_work_sem);
2072                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2073                 if (!err)
2074                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2075                 up_read(&fs_info->cleanup_work_sem);
2076                 if (err) {
2077                         close_ctree(tree_root);
2078                         return ERR_PTR(err);
2079                 }
2080         }
2081
2082         return tree_root;
2083
2084 fail_trans_kthread:
2085         kthread_stop(fs_info->transaction_kthread);
2086 fail_cleaner:
2087         kthread_stop(fs_info->cleaner_kthread);
2088
2089         /*
2090          * make sure we're done with the btree inode before we stop our
2091          * kthreads
2092          */
2093         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2094         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2095
2096 fail_block_groups:
2097         btrfs_free_block_groups(fs_info);
2098         free_extent_buffer(csum_root->node);
2099         free_extent_buffer(csum_root->commit_root);
2100 fail_dev_root:
2101         free_extent_buffer(dev_root->node);
2102         free_extent_buffer(dev_root->commit_root);
2103 fail_extent_root:
2104         free_extent_buffer(extent_root->node);
2105         free_extent_buffer(extent_root->commit_root);
2106 fail_tree_root:
2107         free_extent_buffer(tree_root->node);
2108         free_extent_buffer(tree_root->commit_root);
2109 fail_chunk_root:
2110         free_extent_buffer(chunk_root->node);
2111         free_extent_buffer(chunk_root->commit_root);
2112 fail_sb_buffer:
2113         btrfs_stop_workers(&fs_info->generic_worker);
2114         btrfs_stop_workers(&fs_info->fixup_workers);
2115         btrfs_stop_workers(&fs_info->delalloc_workers);
2116         btrfs_stop_workers(&fs_info->workers);
2117         btrfs_stop_workers(&fs_info->endio_workers);
2118         btrfs_stop_workers(&fs_info->endio_meta_workers);
2119         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2120         btrfs_stop_workers(&fs_info->endio_write_workers);
2121         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2122         btrfs_stop_workers(&fs_info->submit_workers);
2123         btrfs_stop_workers(&fs_info->delayed_workers);
2124         btrfs_stop_workers(&fs_info->caching_workers);
2125 fail_alloc:
2126         kfree(fs_info->delayed_root);
2127 fail_iput:
2128         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2129         iput(fs_info->btree_inode);
2130
2131         btrfs_close_devices(fs_info->fs_devices);
2132         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2133 fail_bdi:
2134         bdi_destroy(&fs_info->bdi);
2135 fail_srcu:
2136         cleanup_srcu_struct(&fs_info->subvol_srcu);
2137 fail:
2138         kfree(extent_root);
2139         kfree(tree_root);
2140         kfree(fs_info);
2141         kfree(chunk_root);
2142         kfree(dev_root);
2143         kfree(csum_root);
2144         return ERR_PTR(err);
2145 }
2146
2147 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2148 {
2149         char b[BDEVNAME_SIZE];
2150
2151         if (uptodate) {
2152                 set_buffer_uptodate(bh);
2153         } else {
2154                 printk_ratelimited(KERN_WARNING "lost page write due to "
2155                                         "I/O error on %s\n",
2156                                        bdevname(bh->b_bdev, b));
2157                 /* note, we dont' set_buffer_write_io_error because we have
2158                  * our own ways of dealing with the IO errors
2159                  */
2160                 clear_buffer_uptodate(bh);
2161         }
2162         unlock_buffer(bh);
2163         put_bh(bh);
2164 }
2165
2166 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2167 {
2168         struct buffer_head *bh;
2169         struct buffer_head *latest = NULL;
2170         struct btrfs_super_block *super;
2171         int i;
2172         u64 transid = 0;
2173         u64 bytenr;
2174
2175         /* we would like to check all the supers, but that would make
2176          * a btrfs mount succeed after a mkfs from a different FS.
2177          * So, we need to add a special mount option to scan for
2178          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2179          */
2180         for (i = 0; i < 1; i++) {
2181                 bytenr = btrfs_sb_offset(i);
2182                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2183                         break;
2184                 bh = __bread(bdev, bytenr / 4096, 4096);
2185                 if (!bh)
2186                         continue;
2187
2188                 super = (struct btrfs_super_block *)bh->b_data;
2189                 if (btrfs_super_bytenr(super) != bytenr ||
2190                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2191                             sizeof(super->magic))) {
2192                         brelse(bh);
2193                         continue;
2194                 }
2195
2196                 if (!latest || btrfs_super_generation(super) > transid) {
2197                         brelse(latest);
2198                         latest = bh;
2199                         transid = btrfs_super_generation(super);
2200                 } else {
2201                         brelse(bh);
2202                 }
2203         }
2204         return latest;
2205 }
2206
2207 /*
2208  * this should be called twice, once with wait == 0 and
2209  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2210  * we write are pinned.
2211  *
2212  * They are released when wait == 1 is done.
2213  * max_mirrors must be the same for both runs, and it indicates how
2214  * many supers on this one device should be written.
2215  *
2216  * max_mirrors == 0 means to write them all.
2217  */
2218 static int write_dev_supers(struct btrfs_device *device,
2219                             struct btrfs_super_block *sb,
2220                             int do_barriers, int wait, int max_mirrors)
2221 {
2222         struct buffer_head *bh;
2223         int i;
2224         int ret;
2225         int errors = 0;
2226         u32 crc;
2227         u64 bytenr;
2228         int last_barrier = 0;
2229
2230         if (max_mirrors == 0)
2231                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2232
2233         /* make sure only the last submit_bh does a barrier */
2234         if (do_barriers) {
2235                 for (i = 0; i < max_mirrors; i++) {
2236                         bytenr = btrfs_sb_offset(i);
2237                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2238                             device->total_bytes)
2239                                 break;
2240                         last_barrier = i;
2241                 }
2242         }
2243
2244         for (i = 0; i < max_mirrors; i++) {
2245                 bytenr = btrfs_sb_offset(i);
2246                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2247                         break;
2248
2249                 if (wait) {
2250                         bh = __find_get_block(device->bdev, bytenr / 4096,
2251                                               BTRFS_SUPER_INFO_SIZE);
2252                         BUG_ON(!bh);
2253                         wait_on_buffer(bh);
2254                         if (!buffer_uptodate(bh))
2255                                 errors++;
2256
2257                         /* drop our reference */
2258                         brelse(bh);
2259
2260                         /* drop the reference from the wait == 0 run */
2261                         brelse(bh);
2262                         continue;
2263                 } else {
2264                         btrfs_set_super_bytenr(sb, bytenr);
2265
2266                         crc = ~(u32)0;
2267                         crc = btrfs_csum_data(NULL, (char *)sb +
2268                                               BTRFS_CSUM_SIZE, crc,
2269                                               BTRFS_SUPER_INFO_SIZE -
2270                                               BTRFS_CSUM_SIZE);
2271                         btrfs_csum_final(crc, sb->csum);
2272
2273                         /*
2274                          * one reference for us, and we leave it for the
2275                          * caller
2276                          */
2277                         bh = __getblk(device->bdev, bytenr / 4096,
2278                                       BTRFS_SUPER_INFO_SIZE);
2279                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2280
2281                         /* one reference for submit_bh */
2282                         get_bh(bh);
2283
2284                         set_buffer_uptodate(bh);
2285                         lock_buffer(bh);
2286                         bh->b_end_io = btrfs_end_buffer_write_sync;
2287                 }
2288
2289                 if (i == last_barrier && do_barriers)
2290                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2291                 else
2292                         ret = submit_bh(WRITE_SYNC, bh);
2293
2294                 if (ret)
2295                         errors++;
2296         }
2297         return errors < i ? 0 : -1;
2298 }
2299
2300 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2301 {
2302         struct list_head *head;
2303         struct btrfs_device *dev;
2304         struct btrfs_super_block *sb;
2305         struct btrfs_dev_item *dev_item;
2306         int ret;
2307         int do_barriers;
2308         int max_errors;
2309         int total_errors = 0;
2310         u64 flags;
2311
2312         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2313         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2314
2315         sb = &root->fs_info->super_for_commit;
2316         dev_item = &sb->dev_item;
2317
2318         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2319         head = &root->fs_info->fs_devices->devices;
2320         list_for_each_entry_rcu(dev, head, dev_list) {
2321                 if (!dev->bdev) {
2322                         total_errors++;
2323                         continue;
2324                 }
2325                 if (!dev->in_fs_metadata || !dev->writeable)
2326                         continue;
2327
2328                 btrfs_set_stack_device_generation(dev_item, 0);
2329                 btrfs_set_stack_device_type(dev_item, dev->type);
2330                 btrfs_set_stack_device_id(dev_item, dev->devid);
2331                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2332                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2333                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2334                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2335                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2336                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2337                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2338
2339                 flags = btrfs_super_flags(sb);
2340                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2341
2342                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2343                 if (ret)
2344                         total_errors++;
2345         }
2346         if (total_errors > max_errors) {
2347                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2348                        total_errors);
2349                 BUG();
2350         }
2351
2352         total_errors = 0;
2353         list_for_each_entry_rcu(dev, head, dev_list) {
2354                 if (!dev->bdev)
2355                         continue;
2356                 if (!dev->in_fs_metadata || !dev->writeable)
2357                         continue;
2358
2359                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2360                 if (ret)
2361                         total_errors++;
2362         }
2363         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2364         if (total_errors > max_errors) {
2365                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2366                        total_errors);
2367                 BUG();
2368         }
2369         return 0;
2370 }
2371
2372 int write_ctree_super(struct btrfs_trans_handle *trans,
2373                       struct btrfs_root *root, int max_mirrors)
2374 {
2375         int ret;
2376
2377         ret = write_all_supers(root, max_mirrors);
2378         return ret;
2379 }
2380
2381 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2382 {
2383         spin_lock(&fs_info->fs_roots_radix_lock);
2384         radix_tree_delete(&fs_info->fs_roots_radix,
2385                           (unsigned long)root->root_key.objectid);
2386         spin_unlock(&fs_info->fs_roots_radix_lock);
2387
2388         if (btrfs_root_refs(&root->root_item) == 0)
2389                 synchronize_srcu(&fs_info->subvol_srcu);
2390
2391         __btrfs_remove_free_space_cache(root->free_ino_pinned);
2392         __btrfs_remove_free_space_cache(root->free_ino_ctl);
2393         free_fs_root(root);
2394         return 0;
2395 }
2396
2397 static void free_fs_root(struct btrfs_root *root)
2398 {
2399         iput(root->cache_inode);
2400         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2401         if (root->anon_super.s_dev) {
2402                 down_write(&root->anon_super.s_umount);
2403                 kill_anon_super(&root->anon_super);
2404         }
2405         free_extent_buffer(root->node);
2406         free_extent_buffer(root->commit_root);
2407         kfree(root->free_ino_ctl);
2408         kfree(root->free_ino_pinned);
2409         kfree(root->name);
2410         kfree(root);
2411 }
2412
2413 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2414 {
2415         int ret;
2416         struct btrfs_root *gang[8];
2417         int i;
2418
2419         while (!list_empty(&fs_info->dead_roots)) {
2420                 gang[0] = list_entry(fs_info->dead_roots.next,
2421                                      struct btrfs_root, root_list);
2422                 list_del(&gang[0]->root_list);
2423
2424                 if (gang[0]->in_radix) {
2425                         btrfs_free_fs_root(fs_info, gang[0]);
2426                 } else {
2427                         free_extent_buffer(gang[0]->node);
2428                         free_extent_buffer(gang[0]->commit_root);
2429                         kfree(gang[0]);
2430                 }
2431         }
2432
2433         while (1) {
2434                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2435                                              (void **)gang, 0,
2436                                              ARRAY_SIZE(gang));
2437                 if (!ret)
2438                         break;
2439                 for (i = 0; i < ret; i++)
2440                         btrfs_free_fs_root(fs_info, gang[i]);
2441         }
2442         return 0;
2443 }
2444
2445 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2446 {
2447         u64 root_objectid = 0;
2448         struct btrfs_root *gang[8];
2449         int i;
2450         int ret;
2451
2452         while (1) {
2453                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2454                                              (void **)gang, root_objectid,
2455                                              ARRAY_SIZE(gang));
2456                 if (!ret)
2457                         break;
2458
2459                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2460                 for (i = 0; i < ret; i++) {
2461                         int err;
2462
2463                         root_objectid = gang[i]->root_key.objectid;
2464                         err = btrfs_orphan_cleanup(gang[i]);
2465                         if (err)
2466                                 return err;
2467                 }
2468                 root_objectid++;
2469         }
2470         return 0;
2471 }
2472
2473 int btrfs_commit_super(struct btrfs_root *root)
2474 {
2475         struct btrfs_trans_handle *trans;
2476         int ret;
2477
2478         mutex_lock(&root->fs_info->cleaner_mutex);
2479         btrfs_run_delayed_iputs(root);
2480         btrfs_clean_old_snapshots(root);
2481         mutex_unlock(&root->fs_info->cleaner_mutex);
2482
2483         /* wait until ongoing cleanup work done */
2484         down_write(&root->fs_info->cleanup_work_sem);
2485         up_write(&root->fs_info->cleanup_work_sem);
2486
2487         trans = btrfs_join_transaction(root);
2488         if (IS_ERR(trans))
2489                 return PTR_ERR(trans);
2490         ret = btrfs_commit_transaction(trans, root);
2491         BUG_ON(ret);
2492         /* run commit again to drop the original snapshot */
2493         trans = btrfs_join_transaction(root);
2494         if (IS_ERR(trans))
2495                 return PTR_ERR(trans);
2496         btrfs_commit_transaction(trans, root);
2497         ret = btrfs_write_and_wait_transaction(NULL, root);
2498         BUG_ON(ret);
2499
2500         ret = write_ctree_super(NULL, root, 0);
2501         return ret;
2502 }
2503
2504 int close_ctree(struct btrfs_root *root)
2505 {
2506         struct btrfs_fs_info *fs_info = root->fs_info;
2507         int ret;
2508
2509         fs_info->closing = 1;
2510         smp_mb();
2511
2512         btrfs_scrub_cancel(root);
2513
2514         /* wait for any defraggers to finish */
2515         wait_event(fs_info->transaction_wait,
2516                    (atomic_read(&fs_info->defrag_running) == 0));
2517
2518         /* clear out the rbtree of defraggable inodes */
2519         btrfs_run_defrag_inodes(root->fs_info);
2520
2521         btrfs_put_block_group_cache(fs_info);
2522
2523         /*
2524          * Here come 2 situations when btrfs is broken to flip readonly:
2525          *
2526          * 1. when btrfs flips readonly somewhere else before
2527          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2528          * and btrfs will skip to write sb directly to keep
2529          * ERROR state on disk.
2530          *
2531          * 2. when btrfs flips readonly just in btrfs_commit_super,
2532          * and in such case, btrfs cannot write sb via btrfs_commit_super,
2533          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2534          * btrfs will cleanup all FS resources first and write sb then.
2535          */
2536         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2537                 ret = btrfs_commit_super(root);
2538                 if (ret)
2539                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2540         }
2541
2542         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2543                 ret = btrfs_error_commit_super(root);
2544                 if (ret)
2545                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2546         }
2547
2548         kthread_stop(root->fs_info->transaction_kthread);
2549         kthread_stop(root->fs_info->cleaner_kthread);
2550
2551         fs_info->closing = 2;
2552         smp_mb();
2553
2554         if (fs_info->delalloc_bytes) {
2555                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2556                        (unsigned long long)fs_info->delalloc_bytes);
2557         }
2558         if (fs_info->total_ref_cache_size) {
2559                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2560                        (unsigned long long)fs_info->total_ref_cache_size);
2561         }
2562
2563         free_extent_buffer(fs_info->extent_root->node);
2564         free_extent_buffer(fs_info->extent_root->commit_root);
2565         free_extent_buffer(fs_info->tree_root->node);
2566         free_extent_buffer(fs_info->tree_root->commit_root);
2567         free_extent_buffer(root->fs_info->chunk_root->node);
2568         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2569         free_extent_buffer(root->fs_info->dev_root->node);
2570         free_extent_buffer(root->fs_info->dev_root->commit_root);
2571         free_extent_buffer(root->fs_info->csum_root->node);
2572         free_extent_buffer(root->fs_info->csum_root->commit_root);
2573
2574         btrfs_free_block_groups(root->fs_info);
2575
2576         del_fs_roots(fs_info);
2577
2578         iput(fs_info->btree_inode);
2579         kfree(fs_info->delayed_root);
2580
2581         btrfs_stop_workers(&fs_info->generic_worker);
2582         btrfs_stop_workers(&fs_info->fixup_workers);
2583         btrfs_stop_workers(&fs_info->delalloc_workers);
2584         btrfs_stop_workers(&fs_info->workers);
2585         btrfs_stop_workers(&fs_info->endio_workers);
2586         btrfs_stop_workers(&fs_info->endio_meta_workers);
2587         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2588         btrfs_stop_workers(&fs_info->endio_write_workers);
2589         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2590         btrfs_stop_workers(&fs_info->submit_workers);
2591         btrfs_stop_workers(&fs_info->delayed_workers);
2592         btrfs_stop_workers(&fs_info->caching_workers);
2593
2594         btrfs_close_devices(fs_info->fs_devices);
2595         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2596
2597         bdi_destroy(&fs_info->bdi);
2598         cleanup_srcu_struct(&fs_info->subvol_srcu);
2599
2600         kfree(fs_info->extent_root);
2601         kfree(fs_info->tree_root);
2602         kfree(fs_info->chunk_root);
2603         kfree(fs_info->dev_root);
2604         kfree(fs_info->csum_root);
2605         kfree(fs_info);
2606
2607         return 0;
2608 }
2609
2610 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2611 {
2612         int ret;
2613         struct inode *btree_inode = buf->first_page->mapping->host;
2614
2615         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2616                                      NULL);
2617         if (!ret)
2618                 return ret;
2619
2620         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2621                                     parent_transid);
2622         return !ret;
2623 }
2624
2625 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2626 {
2627         struct inode *btree_inode = buf->first_page->mapping->host;
2628         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2629                                           buf);
2630 }
2631
2632 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2633 {
2634         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2635         u64 transid = btrfs_header_generation(buf);
2636         struct inode *btree_inode = root->fs_info->btree_inode;
2637         int was_dirty;
2638
2639         btrfs_assert_tree_locked(buf);
2640         if (transid != root->fs_info->generation) {
2641                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2642                        "found %llu running %llu\n",
2643                         (unsigned long long)buf->start,
2644                         (unsigned long long)transid,
2645                         (unsigned long long)root->fs_info->generation);
2646                 WARN_ON(1);
2647         }
2648         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2649                                             buf);
2650         if (!was_dirty) {
2651                 spin_lock(&root->fs_info->delalloc_lock);
2652                 root->fs_info->dirty_metadata_bytes += buf->len;
2653                 spin_unlock(&root->fs_info->delalloc_lock);
2654         }
2655 }
2656
2657 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2658 {
2659         /*
2660          * looks as though older kernels can get into trouble with
2661          * this code, they end up stuck in balance_dirty_pages forever
2662          */
2663         u64 num_dirty;
2664         unsigned long thresh = 32 * 1024 * 1024;
2665
2666         if (current->flags & PF_MEMALLOC)
2667                 return;
2668
2669         btrfs_balance_delayed_items(root);
2670
2671         num_dirty = root->fs_info->dirty_metadata_bytes;
2672
2673         if (num_dirty > thresh) {
2674                 balance_dirty_pages_ratelimited_nr(
2675                                    root->fs_info->btree_inode->i_mapping, 1);
2676         }
2677         return;
2678 }
2679
2680 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2681 {
2682         /*
2683          * looks as though older kernels can get into trouble with
2684          * this code, they end up stuck in balance_dirty_pages forever
2685          */
2686         u64 num_dirty;
2687         unsigned long thresh = 32 * 1024 * 1024;
2688
2689         if (current->flags & PF_MEMALLOC)
2690                 return;
2691
2692         num_dirty = root->fs_info->dirty_metadata_bytes;
2693
2694         if (num_dirty > thresh) {
2695                 balance_dirty_pages_ratelimited_nr(
2696                                    root->fs_info->btree_inode->i_mapping, 1);
2697         }
2698         return;
2699 }
2700
2701 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2702 {
2703         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2704         int ret;
2705         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2706         if (ret == 0)
2707                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2708         return ret;
2709 }
2710
2711 int btree_lock_page_hook(struct page *page)
2712 {
2713         struct inode *inode = page->mapping->host;
2714         struct btrfs_root *root = BTRFS_I(inode)->root;
2715         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2716         struct extent_buffer *eb;
2717         unsigned long len;
2718         u64 bytenr = page_offset(page);
2719
2720         if (page->private == EXTENT_PAGE_PRIVATE)
2721                 goto out;
2722
2723         len = page->private >> 2;
2724         eb = find_extent_buffer(io_tree, bytenr, len);
2725         if (!eb)
2726                 goto out;
2727
2728         btrfs_tree_lock(eb);
2729         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2730
2731         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2732                 spin_lock(&root->fs_info->delalloc_lock);
2733                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2734                         root->fs_info->dirty_metadata_bytes -= eb->len;
2735                 else
2736                         WARN_ON(1);
2737                 spin_unlock(&root->fs_info->delalloc_lock);
2738         }
2739
2740         btrfs_tree_unlock(eb);
2741         free_extent_buffer(eb);
2742 out:
2743         lock_page(page);
2744         return 0;
2745 }
2746
2747 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2748                               int read_only)
2749 {
2750         if (read_only)
2751                 return;
2752
2753         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2754                 printk(KERN_WARNING "warning: mount fs with errors, "
2755                        "running btrfsck is recommended\n");
2756 }
2757
2758 int btrfs_error_commit_super(struct btrfs_root *root)
2759 {
2760         int ret;
2761
2762         mutex_lock(&root->fs_info->cleaner_mutex);
2763         btrfs_run_delayed_iputs(root);
2764         mutex_unlock(&root->fs_info->cleaner_mutex);
2765
2766         down_write(&root->fs_info->cleanup_work_sem);
2767         up_write(&root->fs_info->cleanup_work_sem);
2768
2769         /* cleanup FS via transaction */
2770         btrfs_cleanup_transaction(root);
2771
2772         ret = write_ctree_super(NULL, root, 0);
2773
2774         return ret;
2775 }
2776
2777 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2778 {
2779         struct btrfs_inode *btrfs_inode;
2780         struct list_head splice;
2781
2782         INIT_LIST_HEAD(&splice);
2783
2784         mutex_lock(&root->fs_info->ordered_operations_mutex);
2785         spin_lock(&root->fs_info->ordered_extent_lock);
2786
2787         list_splice_init(&root->fs_info->ordered_operations, &splice);
2788         while (!list_empty(&splice)) {
2789                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2790                                          ordered_operations);
2791
2792                 list_del_init(&btrfs_inode->ordered_operations);
2793
2794                 btrfs_invalidate_inodes(btrfs_inode->root);
2795         }
2796
2797         spin_unlock(&root->fs_info->ordered_extent_lock);
2798         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2799
2800         return 0;
2801 }
2802
2803 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2804 {
2805         struct list_head splice;
2806         struct btrfs_ordered_extent *ordered;
2807         struct inode *inode;
2808
2809         INIT_LIST_HEAD(&splice);
2810
2811         spin_lock(&root->fs_info->ordered_extent_lock);
2812
2813         list_splice_init(&root->fs_info->ordered_extents, &splice);
2814         while (!list_empty(&splice)) {
2815                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2816                                      root_extent_list);
2817
2818                 list_del_init(&ordered->root_extent_list);
2819                 atomic_inc(&ordered->refs);
2820
2821                 /* the inode may be getting freed (in sys_unlink path). */
2822                 inode = igrab(ordered->inode);
2823
2824                 spin_unlock(&root->fs_info->ordered_extent_lock);
2825                 if (inode)
2826                         iput(inode);
2827
2828                 atomic_set(&ordered->refs, 1);
2829                 btrfs_put_ordered_extent(ordered);
2830
2831                 spin_lock(&root->fs_info->ordered_extent_lock);
2832         }
2833
2834         spin_unlock(&root->fs_info->ordered_extent_lock);
2835
2836         return 0;
2837 }
2838
2839 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2840                                       struct btrfs_root *root)
2841 {
2842         struct rb_node *node;
2843         struct btrfs_delayed_ref_root *delayed_refs;
2844         struct btrfs_delayed_ref_node *ref;
2845         int ret = 0;
2846
2847         delayed_refs = &trans->delayed_refs;
2848
2849         spin_lock(&delayed_refs->lock);
2850         if (delayed_refs->num_entries == 0) {
2851                 spin_unlock(&delayed_refs->lock);
2852                 printk(KERN_INFO "delayed_refs has NO entry\n");
2853                 return ret;
2854         }
2855
2856         node = rb_first(&delayed_refs->root);
2857         while (node) {
2858                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2859                 node = rb_next(node);
2860
2861                 ref->in_tree = 0;
2862                 rb_erase(&ref->rb_node, &delayed_refs->root);
2863                 delayed_refs->num_entries--;
2864
2865                 atomic_set(&ref->refs, 1);
2866                 if (btrfs_delayed_ref_is_head(ref)) {
2867                         struct btrfs_delayed_ref_head *head;
2868
2869                         head = btrfs_delayed_node_to_head(ref);
2870                         mutex_lock(&head->mutex);
2871                         kfree(head->extent_op);
2872                         delayed_refs->num_heads--;
2873                         if (list_empty(&head->cluster))
2874                                 delayed_refs->num_heads_ready--;
2875                         list_del_init(&head->cluster);
2876                         mutex_unlock(&head->mutex);
2877                 }
2878
2879                 spin_unlock(&delayed_refs->lock);
2880                 btrfs_put_delayed_ref(ref);
2881
2882                 cond_resched();
2883                 spin_lock(&delayed_refs->lock);
2884         }
2885
2886         spin_unlock(&delayed_refs->lock);
2887
2888         return ret;
2889 }
2890
2891 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2892 {
2893         struct btrfs_pending_snapshot *snapshot;
2894         struct list_head splice;
2895
2896         INIT_LIST_HEAD(&splice);
2897
2898         list_splice_init(&t->pending_snapshots, &splice);
2899
2900         while (!list_empty(&splice)) {
2901                 snapshot = list_entry(splice.next,
2902                                       struct btrfs_pending_snapshot,
2903                                       list);
2904
2905                 list_del_init(&snapshot->list);
2906
2907                 kfree(snapshot);
2908         }
2909
2910         return 0;
2911 }
2912
2913 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2914 {
2915         struct btrfs_inode *btrfs_inode;
2916         struct list_head splice;
2917
2918         INIT_LIST_HEAD(&splice);
2919
2920         spin_lock(&root->fs_info->delalloc_lock);
2921         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2922
2923         while (!list_empty(&splice)) {
2924                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2925                                     delalloc_inodes);
2926
2927                 list_del_init(&btrfs_inode->delalloc_inodes);
2928
2929                 btrfs_invalidate_inodes(btrfs_inode->root);
2930         }
2931
2932         spin_unlock(&root->fs_info->delalloc_lock);
2933
2934         return 0;
2935 }
2936
2937 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2938                                         struct extent_io_tree *dirty_pages,
2939                                         int mark)
2940 {
2941         int ret;
2942         struct page *page;
2943         struct inode *btree_inode = root->fs_info->btree_inode;
2944         struct extent_buffer *eb;
2945         u64 start = 0;
2946         u64 end;
2947         u64 offset;
2948         unsigned long index;
2949
2950         while (1) {
2951                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2952                                             mark);
2953                 if (ret)
2954                         break;
2955
2956                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2957                 while (start <= end) {
2958                         index = start >> PAGE_CACHE_SHIFT;
2959                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2960                         page = find_get_page(btree_inode->i_mapping, index);
2961                         if (!page)
2962                                 continue;
2963                         offset = page_offset(page);
2964
2965                         spin_lock(&dirty_pages->buffer_lock);
2966                         eb = radix_tree_lookup(
2967                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2968                                                offset >> PAGE_CACHE_SHIFT);
2969                         spin_unlock(&dirty_pages->buffer_lock);
2970                         if (eb) {
2971                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2972                                                          &eb->bflags);
2973                                 atomic_set(&eb->refs, 1);
2974                         }
2975                         if (PageWriteback(page))
2976                                 end_page_writeback(page);
2977
2978                         lock_page(page);
2979                         if (PageDirty(page)) {
2980                                 clear_page_dirty_for_io(page);
2981                                 spin_lock_irq(&page->mapping->tree_lock);
2982                                 radix_tree_tag_clear(&page->mapping->page_tree,
2983                                                         page_index(page),
2984                                                         PAGECACHE_TAG_DIRTY);
2985                                 spin_unlock_irq(&page->mapping->tree_lock);
2986                         }
2987
2988                         page->mapping->a_ops->invalidatepage(page, 0);
2989                         unlock_page(page);
2990                 }
2991         }
2992
2993         return ret;
2994 }
2995
2996 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
2997                                        struct extent_io_tree *pinned_extents)
2998 {
2999         struct extent_io_tree *unpin;
3000         u64 start;
3001         u64 end;
3002         int ret;
3003
3004         unpin = pinned_extents;
3005         while (1) {
3006                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3007                                             EXTENT_DIRTY);
3008                 if (ret)
3009                         break;
3010
3011                 /* opt_discard */
3012                 if (btrfs_test_opt(root, DISCARD))
3013                         ret = btrfs_error_discard_extent(root, start,
3014                                                          end + 1 - start,
3015                                                          NULL);
3016
3017                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3018                 btrfs_error_unpin_extent_range(root, start, end);
3019                 cond_resched();
3020         }
3021
3022         return 0;
3023 }
3024
3025 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3026 {
3027         struct btrfs_transaction *t;
3028         LIST_HEAD(list);
3029
3030         WARN_ON(1);
3031
3032         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3033
3034         spin_lock(&root->fs_info->trans_lock);
3035         list_splice_init(&root->fs_info->trans_list, &list);
3036         root->fs_info->trans_no_join = 1;
3037         spin_unlock(&root->fs_info->trans_lock);
3038
3039         while (!list_empty(&list)) {
3040                 t = list_entry(list.next, struct btrfs_transaction, list);
3041                 if (!t)
3042                         break;
3043
3044                 btrfs_destroy_ordered_operations(root);
3045
3046                 btrfs_destroy_ordered_extents(root);
3047
3048                 btrfs_destroy_delayed_refs(t, root);
3049
3050                 btrfs_block_rsv_release(root,
3051                                         &root->fs_info->trans_block_rsv,
3052                                         t->dirty_pages.dirty_bytes);
3053
3054                 /* FIXME: cleanup wait for commit */
3055                 t->in_commit = 1;
3056                 t->blocked = 1;
3057                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3058                         wake_up(&root->fs_info->transaction_blocked_wait);
3059
3060                 t->blocked = 0;
3061                 if (waitqueue_active(&root->fs_info->transaction_wait))
3062                         wake_up(&root->fs_info->transaction_wait);
3063
3064                 t->commit_done = 1;
3065                 if (waitqueue_active(&t->commit_wait))
3066                         wake_up(&t->commit_wait);
3067
3068                 btrfs_destroy_pending_snapshots(t);
3069
3070                 btrfs_destroy_delalloc_inodes(root);
3071
3072                 spin_lock(&root->fs_info->trans_lock);
3073                 root->fs_info->running_transaction = NULL;
3074                 spin_unlock(&root->fs_info->trans_lock);
3075
3076                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3077                                              EXTENT_DIRTY);
3078
3079                 btrfs_destroy_pinned_extent(root,
3080                                             root->fs_info->pinned_extents);
3081
3082                 atomic_set(&t->use_count, 0);
3083                 list_del_init(&t->list);
3084                 memset(t, 0, sizeof(*t));
3085                 kmem_cache_free(btrfs_transaction_cachep, t);
3086         }
3087
3088         spin_lock(&root->fs_info->trans_lock);
3089         root->fs_info->trans_no_join = 0;
3090         spin_unlock(&root->fs_info->trans_lock);
3091         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3092
3093         return 0;
3094 }
3095
3096 static struct extent_io_ops btree_extent_io_ops = {
3097         .write_cache_pages_lock_hook = btree_lock_page_hook,
3098         .readpage_end_io_hook = btree_readpage_end_io_hook,
3099         .submit_bio_hook = btree_submit_bio_hook,
3100         /* note we're sharing with inode.c for the merge bio hook */
3101         .merge_bio_hook = btrfs_merge_bio_hook,
3102 };