2887b8be6fdd26c30794c72c8718db868dd9781f
[linux-3.10.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 "compat.h"
33 #include "ctree.h"
34 #include "disk-io.h"
35 #include "transaction.h"
36 #include "btrfs_inode.h"
37 #include "volumes.h"
38 #include "print-tree.h"
39 #include "async-thread.h"
40 #include "locking.h"
41 #include "tree-log.h"
42 #include "free-space-cache.h"
43
44 static struct extent_io_ops btree_extent_io_ops;
45 static void end_workqueue_fn(struct btrfs_work *work);
46 static void free_fs_root(struct btrfs_root *root);
47 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
48                                     int read_only);
49 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
50 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
51 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
52                                       struct btrfs_root *root);
53 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
54 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
55 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
56                                         struct extent_io_tree *dirty_pages,
57                                         int mark);
58 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
59                                        struct extent_io_tree *pinned_extents);
60 static int btrfs_cleanup_transaction(struct btrfs_root *root);
61
62 /*
63  * end_io_wq structs are used to do processing in task context when an IO is
64  * complete.  This is used during reads to verify checksums, and it is used
65  * by writes to insert metadata for new file extents after IO is complete.
66  */
67 struct end_io_wq {
68         struct bio *bio;
69         bio_end_io_t *end_io;
70         void *private;
71         struct btrfs_fs_info *info;
72         int error;
73         int metadata;
74         struct list_head list;
75         struct btrfs_work work;
76 };
77
78 /*
79  * async submit bios are used to offload expensive checksumming
80  * onto the worker threads.  They checksum file and metadata bios
81  * just before they are sent down the IO stack.
82  */
83 struct async_submit_bio {
84         struct inode *inode;
85         struct bio *bio;
86         struct list_head list;
87         extent_submit_bio_hook_t *submit_bio_start;
88         extent_submit_bio_hook_t *submit_bio_done;
89         int rw;
90         int mirror_num;
91         unsigned long bio_flags;
92         /*
93          * bio_offset is optional, can be used if the pages in the bio
94          * can't tell us where in the file the bio should go
95          */
96         u64 bio_offset;
97         struct btrfs_work work;
98 };
99
100 /* These are used to set the lockdep class on the extent buffer locks.
101  * The class is set by the readpage_end_io_hook after the buffer has
102  * passed csum validation but before the pages are unlocked.
103  *
104  * The lockdep class is also set by btrfs_init_new_buffer on freshly
105  * allocated blocks.
106  *
107  * The class is based on the level in the tree block, which allows lockdep
108  * to know that lower nodes nest inside the locks of higher nodes.
109  *
110  * We also add a check to make sure the highest level of the tree is
111  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
112  * code needs update as well.
113  */
114 #ifdef CONFIG_DEBUG_LOCK_ALLOC
115 # if BTRFS_MAX_LEVEL != 8
116 #  error
117 # endif
118 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
119 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
120         /* leaf */
121         "btrfs-extent-00",
122         "btrfs-extent-01",
123         "btrfs-extent-02",
124         "btrfs-extent-03",
125         "btrfs-extent-04",
126         "btrfs-extent-05",
127         "btrfs-extent-06",
128         "btrfs-extent-07",
129         /* highest possible level */
130         "btrfs-extent-08",
131 };
132 #endif
133
134 /*
135  * extents on the btree inode are pretty simple, there's one extent
136  * that covers the entire device
137  */
138 static struct extent_map *btree_get_extent(struct inode *inode,
139                 struct page *page, size_t page_offset, u64 start, u64 len,
140                 int create)
141 {
142         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
143         struct extent_map *em;
144         int ret;
145
146         read_lock(&em_tree->lock);
147         em = lookup_extent_mapping(em_tree, start, len);
148         if (em) {
149                 em->bdev =
150                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
151                 read_unlock(&em_tree->lock);
152                 goto out;
153         }
154         read_unlock(&em_tree->lock);
155
156         em = alloc_extent_map(GFP_NOFS);
157         if (!em) {
158                 em = ERR_PTR(-ENOMEM);
159                 goto out;
160         }
161         em->start = 0;
162         em->len = (u64)-1;
163         em->block_len = (u64)-1;
164         em->block_start = 0;
165         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
166
167         write_lock(&em_tree->lock);
168         ret = add_extent_mapping(em_tree, em);
169         if (ret == -EEXIST) {
170                 u64 failed_start = em->start;
171                 u64 failed_len = em->len;
172
173                 free_extent_map(em);
174                 em = lookup_extent_mapping(em_tree, start, len);
175                 if (em) {
176                         ret = 0;
177                 } else {
178                         em = lookup_extent_mapping(em_tree, failed_start,
179                                                    failed_len);
180                         ret = -EIO;
181                 }
182         } else if (ret) {
183                 free_extent_map(em);
184                 em = NULL;
185         }
186         write_unlock(&em_tree->lock);
187
188         if (ret)
189                 em = ERR_PTR(ret);
190 out:
191         return em;
192 }
193
194 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
195 {
196         return crc32c(seed, data, len);
197 }
198
199 void btrfs_csum_final(u32 crc, char *result)
200 {
201         *(__le32 *)result = ~cpu_to_le32(crc);
202 }
203
204 /*
205  * compute the csum for a btree block, and either verify it or write it
206  * into the csum field of the block.
207  */
208 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
209                            int verify)
210 {
211         u16 csum_size =
212                 btrfs_super_csum_size(&root->fs_info->super_copy);
213         char *result = NULL;
214         unsigned long len;
215         unsigned long cur_len;
216         unsigned long offset = BTRFS_CSUM_SIZE;
217         char *map_token = NULL;
218         char *kaddr;
219         unsigned long map_start;
220         unsigned long map_len;
221         int err;
222         u32 crc = ~(u32)0;
223         unsigned long inline_result;
224
225         len = buf->len - offset;
226         while (len > 0) {
227                 err = map_private_extent_buffer(buf, offset, 32,
228                                         &map_token, &kaddr,
229                                         &map_start, &map_len, KM_USER0);
230                 if (err)
231                         return 1;
232                 cur_len = min(len, map_len - (offset - map_start));
233                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
234                                       crc, cur_len);
235                 len -= cur_len;
236                 offset += cur_len;
237                 unmap_extent_buffer(buf, map_token, KM_USER0);
238         }
239         if (csum_size > sizeof(inline_result)) {
240                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
241                 if (!result)
242                         return 1;
243         } else {
244                 result = (char *)&inline_result;
245         }
246
247         btrfs_csum_final(crc, result);
248
249         if (verify) {
250                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
251                         u32 val;
252                         u32 found = 0;
253                         memcpy(&found, result, csum_size);
254
255                         read_extent_buffer(buf, &val, 0, csum_size);
256                         if (printk_ratelimit()) {
257                                 printk(KERN_INFO "btrfs: %s checksum verify "
258                                        "failed on %llu wanted %X found %X "
259                                        "level %d\n",
260                                        root->fs_info->sb->s_id,
261                                        (unsigned long long)buf->start, val, found,
262                                        btrfs_header_level(buf));
263                         }
264                         if (result != (char *)&inline_result)
265                                 kfree(result);
266                         return 1;
267                 }
268         } else {
269                 write_extent_buffer(buf, result, 0, csum_size);
270         }
271         if (result != (char *)&inline_result)
272                 kfree(result);
273         return 0;
274 }
275
276 /*
277  * we can't consider a given block up to date unless the transid of the
278  * block matches the transid in the parent node's pointer.  This is how we
279  * detect blocks that either didn't get written at all or got written
280  * in the wrong place.
281  */
282 static int verify_parent_transid(struct extent_io_tree *io_tree,
283                                  struct extent_buffer *eb, u64 parent_transid)
284 {
285         struct extent_state *cached_state = NULL;
286         int ret;
287
288         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
289                 return 0;
290
291         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
292                          0, &cached_state, GFP_NOFS);
293         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
294             btrfs_header_generation(eb) == parent_transid) {
295                 ret = 0;
296                 goto out;
297         }
298         if (printk_ratelimit()) {
299                 printk("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         }
305         ret = 1;
306         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
307 out:
308         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
309                              &cached_state, GFP_NOFS);
310         return ret;
311 }
312
313 /*
314  * helper to read a given tree block, doing retries as required when
315  * the checksums don't match and we have alternate mirrors to try.
316  */
317 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
318                                           struct extent_buffer *eb,
319                                           u64 start, u64 parent_transid)
320 {
321         struct extent_io_tree *io_tree;
322         int ret;
323         int num_copies = 0;
324         int mirror_num = 0;
325
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                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
335                                               eb->start, eb->len);
336                 if (num_copies == 1)
337                         return ret;
338
339                 mirror_num++;
340                 if (mirror_num > num_copies)
341                         return ret;
342         }
343         return -EIO;
344 }
345
346 /*
347  * checksum a dirty tree block before IO.  This has extra checks to make sure
348  * we only fill in the checksum field in the first page of a multi-page block
349  */
350
351 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
352 {
353         struct extent_io_tree *tree;
354         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
355         u64 found_start;
356         unsigned long len;
357         struct extent_buffer *eb;
358         int ret;
359
360         tree = &BTRFS_I(page->mapping->host)->io_tree;
361
362         if (page->private == EXTENT_PAGE_PRIVATE)
363                 goto out;
364         if (!page->private)
365                 goto out;
366         len = page->private >> 2;
367         WARN_ON(len == 0);
368
369         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
370         if (eb == NULL) {
371                 WARN_ON(1);
372                 goto out;
373         }
374         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
375                                              btrfs_header_generation(eb));
376         BUG_ON(ret);
377         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
378
379         found_start = btrfs_header_bytenr(eb);
380         if (found_start != start) {
381                 WARN_ON(1);
382                 goto err;
383         }
384         if (eb->first_page != page) {
385                 WARN_ON(1);
386                 goto err;
387         }
388         if (!PageUptodate(page)) {
389                 WARN_ON(1);
390                 goto err;
391         }
392         csum_tree_block(root, eb, 0);
393 err:
394         free_extent_buffer(eb);
395 out:
396         return 0;
397 }
398
399 static int check_tree_block_fsid(struct btrfs_root *root,
400                                  struct extent_buffer *eb)
401 {
402         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
403         u8 fsid[BTRFS_UUID_SIZE];
404         int ret = 1;
405
406         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
407                            BTRFS_FSID_SIZE);
408         while (fs_devices) {
409                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
410                         ret = 0;
411                         break;
412                 }
413                 fs_devices = fs_devices->seed;
414         }
415         return ret;
416 }
417
418 #ifdef CONFIG_DEBUG_LOCK_ALLOC
419 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
420 {
421         lockdep_set_class_and_name(&eb->lock,
422                            &btrfs_eb_class[level],
423                            btrfs_eb_name[level]);
424 }
425 #endif
426
427 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
428                                struct extent_state *state)
429 {
430         struct extent_io_tree *tree;
431         u64 found_start;
432         int found_level;
433         unsigned long len;
434         struct extent_buffer *eb;
435         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
436         int ret = 0;
437
438         tree = &BTRFS_I(page->mapping->host)->io_tree;
439         if (page->private == EXTENT_PAGE_PRIVATE)
440                 goto out;
441         if (!page->private)
442                 goto out;
443
444         len = page->private >> 2;
445         WARN_ON(len == 0);
446
447         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
448         if (eb == NULL) {
449                 ret = -EIO;
450                 goto out;
451         }
452
453         found_start = btrfs_header_bytenr(eb);
454         if (found_start != start) {
455                 if (printk_ratelimit()) {
456                         printk(KERN_INFO "btrfs bad tree block start "
457                                "%llu %llu\n",
458                                (unsigned long long)found_start,
459                                (unsigned long long)eb->start);
460                 }
461                 ret = -EIO;
462                 goto err;
463         }
464         if (eb->first_page != page) {
465                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
466                        eb->first_page->index, page->index);
467                 WARN_ON(1);
468                 ret = -EIO;
469                 goto err;
470         }
471         if (check_tree_block_fsid(root, eb)) {
472                 if (printk_ratelimit()) {
473                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
474                                (unsigned long long)eb->start);
475                 }
476                 ret = -EIO;
477                 goto err;
478         }
479         found_level = btrfs_header_level(eb);
480
481         btrfs_set_buffer_lockdep_class(eb, found_level);
482
483         ret = csum_tree_block(root, eb, 1);
484         if (ret)
485                 ret = -EIO;
486
487         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
488         end = eb->start + end - 1;
489 err:
490         free_extent_buffer(eb);
491 out:
492         return ret;
493 }
494
495 static void end_workqueue_bio(struct bio *bio, int err)
496 {
497         struct end_io_wq *end_io_wq = bio->bi_private;
498         struct btrfs_fs_info *fs_info;
499
500         fs_info = end_io_wq->info;
501         end_io_wq->error = err;
502         end_io_wq->work.func = end_workqueue_fn;
503         end_io_wq->work.flags = 0;
504
505         if (bio->bi_rw & REQ_WRITE) {
506                 if (end_io_wq->metadata == 1)
507                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
508                                            &end_io_wq->work);
509                 else if (end_io_wq->metadata == 2)
510                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
511                                            &end_io_wq->work);
512                 else
513                         btrfs_queue_worker(&fs_info->endio_write_workers,
514                                            &end_io_wq->work);
515         } else {
516                 if (end_io_wq->metadata)
517                         btrfs_queue_worker(&fs_info->endio_meta_workers,
518                                            &end_io_wq->work);
519                 else
520                         btrfs_queue_worker(&fs_info->endio_workers,
521                                            &end_io_wq->work);
522         }
523 }
524
525 /*
526  * For the metadata arg you want
527  *
528  * 0 - if data
529  * 1 - if normal metadta
530  * 2 - if writing to the free space cache area
531  */
532 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
533                         int metadata)
534 {
535         struct end_io_wq *end_io_wq;
536         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
537         if (!end_io_wq)
538                 return -ENOMEM;
539
540         end_io_wq->private = bio->bi_private;
541         end_io_wq->end_io = bio->bi_end_io;
542         end_io_wq->info = info;
543         end_io_wq->error = 0;
544         end_io_wq->bio = bio;
545         end_io_wq->metadata = metadata;
546
547         bio->bi_private = end_io_wq;
548         bio->bi_end_io = end_workqueue_bio;
549         return 0;
550 }
551
552 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
553 {
554         unsigned long limit = min_t(unsigned long,
555                                     info->workers.max_workers,
556                                     info->fs_devices->open_devices);
557         return 256 * limit;
558 }
559
560 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
561 {
562         return atomic_read(&info->nr_async_bios) >
563                 btrfs_async_submit_limit(info);
564 }
565
566 static void run_one_async_start(struct btrfs_work *work)
567 {
568         struct async_submit_bio *async;
569
570         async = container_of(work, struct  async_submit_bio, work);
571         async->submit_bio_start(async->inode, async->rw, async->bio,
572                                async->mirror_num, async->bio_flags,
573                                async->bio_offset);
574 }
575
576 static void run_one_async_done(struct btrfs_work *work)
577 {
578         struct btrfs_fs_info *fs_info;
579         struct async_submit_bio *async;
580         int limit;
581
582         async = container_of(work, struct  async_submit_bio, work);
583         fs_info = BTRFS_I(async->inode)->root->fs_info;
584
585         limit = btrfs_async_submit_limit(fs_info);
586         limit = limit * 2 / 3;
587
588         atomic_dec(&fs_info->nr_async_submits);
589
590         if (atomic_read(&fs_info->nr_async_submits) < limit &&
591             waitqueue_active(&fs_info->async_submit_wait))
592                 wake_up(&fs_info->async_submit_wait);
593
594         async->submit_bio_done(async->inode, async->rw, async->bio,
595                                async->mirror_num, async->bio_flags,
596                                async->bio_offset);
597 }
598
599 static void run_one_async_free(struct btrfs_work *work)
600 {
601         struct async_submit_bio *async;
602
603         async = container_of(work, struct  async_submit_bio, work);
604         kfree(async);
605 }
606
607 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
608                         int rw, struct bio *bio, int mirror_num,
609                         unsigned long bio_flags,
610                         u64 bio_offset,
611                         extent_submit_bio_hook_t *submit_bio_start,
612                         extent_submit_bio_hook_t *submit_bio_done)
613 {
614         struct async_submit_bio *async;
615
616         async = kmalloc(sizeof(*async), GFP_NOFS);
617         if (!async)
618                 return -ENOMEM;
619
620         async->inode = inode;
621         async->rw = rw;
622         async->bio = bio;
623         async->mirror_num = mirror_num;
624         async->submit_bio_start = submit_bio_start;
625         async->submit_bio_done = submit_bio_done;
626
627         async->work.func = run_one_async_start;
628         async->work.ordered_func = run_one_async_done;
629         async->work.ordered_free = run_one_async_free;
630
631         async->work.flags = 0;
632         async->bio_flags = bio_flags;
633         async->bio_offset = bio_offset;
634
635         atomic_inc(&fs_info->nr_async_submits);
636
637         if (rw & REQ_SYNC)
638                 btrfs_set_work_high_prio(&async->work);
639
640         btrfs_queue_worker(&fs_info->workers, &async->work);
641
642         while (atomic_read(&fs_info->async_submit_draining) &&
643               atomic_read(&fs_info->nr_async_submits)) {
644                 wait_event(fs_info->async_submit_wait,
645                            (atomic_read(&fs_info->nr_async_submits) == 0));
646         }
647
648         return 0;
649 }
650
651 static int btree_csum_one_bio(struct bio *bio)
652 {
653         struct bio_vec *bvec = bio->bi_io_vec;
654         int bio_index = 0;
655         struct btrfs_root *root;
656
657         WARN_ON(bio->bi_vcnt <= 0);
658         while (bio_index < bio->bi_vcnt) {
659                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
660                 csum_dirty_buffer(root, bvec->bv_page);
661                 bio_index++;
662                 bvec++;
663         }
664         return 0;
665 }
666
667 static int __btree_submit_bio_start(struct inode *inode, int rw,
668                                     struct bio *bio, int mirror_num,
669                                     unsigned long bio_flags,
670                                     u64 bio_offset)
671 {
672         /*
673          * when we're called for a write, we're already in the async
674          * submission context.  Just jump into btrfs_map_bio
675          */
676         btree_csum_one_bio(bio);
677         return 0;
678 }
679
680 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
681                                  int mirror_num, unsigned long bio_flags,
682                                  u64 bio_offset)
683 {
684         /*
685          * when we're called for a write, we're already in the async
686          * submission context.  Just jump into btrfs_map_bio
687          */
688         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
689 }
690
691 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
692                                  int mirror_num, unsigned long bio_flags,
693                                  u64 bio_offset)
694 {
695         int ret;
696
697         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
698                                           bio, 1);
699         BUG_ON(ret);
700
701         if (!(rw & REQ_WRITE)) {
702                 /*
703                  * called for a read, do the setup so that checksum validation
704                  * can happen in the async kernel threads
705                  */
706                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
707                                      mirror_num, 0);
708         }
709
710         /*
711          * kthread helpers are used to submit writes so that checksumming
712          * can happen in parallel across all CPUs
713          */
714         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
715                                    inode, rw, bio, mirror_num, 0,
716                                    bio_offset,
717                                    __btree_submit_bio_start,
718                                    __btree_submit_bio_done);
719 }
720
721 #ifdef CONFIG_MIGRATION
722 static int btree_migratepage(struct address_space *mapping,
723                         struct page *newpage, struct page *page)
724 {
725         /*
726          * we can't safely write a btree page from here,
727          * we haven't done the locking hook
728          */
729         if (PageDirty(page))
730                 return -EAGAIN;
731         /*
732          * Buffers may be managed in a filesystem specific way.
733          * We must have no buffers or drop them.
734          */
735         if (page_has_private(page) &&
736             !try_to_release_page(page, GFP_KERNEL))
737                 return -EAGAIN;
738         return migrate_page(mapping, newpage, page);
739 }
740 #endif
741
742 static int btree_writepage(struct page *page, struct writeback_control *wbc)
743 {
744         struct extent_io_tree *tree;
745         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
746         struct extent_buffer *eb;
747         int was_dirty;
748
749         tree = &BTRFS_I(page->mapping->host)->io_tree;
750         if (!(current->flags & PF_MEMALLOC)) {
751                 return extent_write_full_page(tree, page,
752                                               btree_get_extent, wbc);
753         }
754
755         redirty_page_for_writepage(wbc, page);
756         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
757         WARN_ON(!eb);
758
759         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
760         if (!was_dirty) {
761                 spin_lock(&root->fs_info->delalloc_lock);
762                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
763                 spin_unlock(&root->fs_info->delalloc_lock);
764         }
765         free_extent_buffer(eb);
766
767         unlock_page(page);
768         return 0;
769 }
770
771 static int btree_writepages(struct address_space *mapping,
772                             struct writeback_control *wbc)
773 {
774         struct extent_io_tree *tree;
775         tree = &BTRFS_I(mapping->host)->io_tree;
776         if (wbc->sync_mode == WB_SYNC_NONE) {
777                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
778                 u64 num_dirty;
779                 unsigned long thresh = 32 * 1024 * 1024;
780
781                 if (wbc->for_kupdate)
782                         return 0;
783
784                 /* this is a bit racy, but that's ok */
785                 num_dirty = root->fs_info->dirty_metadata_bytes;
786                 if (num_dirty < thresh)
787                         return 0;
788         }
789         return extent_writepages(tree, mapping, btree_get_extent, wbc);
790 }
791
792 static int btree_readpage(struct file *file, struct page *page)
793 {
794         struct extent_io_tree *tree;
795         tree = &BTRFS_I(page->mapping->host)->io_tree;
796         return extent_read_full_page(tree, page, btree_get_extent);
797 }
798
799 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
800 {
801         struct extent_io_tree *tree;
802         struct extent_map_tree *map;
803         int ret;
804
805         if (PageWriteback(page) || PageDirty(page))
806                 return 0;
807
808         tree = &BTRFS_I(page->mapping->host)->io_tree;
809         map = &BTRFS_I(page->mapping->host)->extent_tree;
810
811         ret = try_release_extent_state(map, tree, page, gfp_flags);
812         if (!ret)
813                 return 0;
814
815         ret = try_release_extent_buffer(tree, page);
816         if (ret == 1) {
817                 ClearPagePrivate(page);
818                 set_page_private(page, 0);
819                 page_cache_release(page);
820         }
821
822         return ret;
823 }
824
825 static void btree_invalidatepage(struct page *page, unsigned long offset)
826 {
827         struct extent_io_tree *tree;
828         tree = &BTRFS_I(page->mapping->host)->io_tree;
829         extent_invalidatepage(tree, page, offset);
830         btree_releasepage(page, GFP_NOFS);
831         if (PagePrivate(page)) {
832                 printk(KERN_WARNING "btrfs warning page private not zero "
833                        "on page %llu\n", (unsigned long long)page_offset(page));
834                 ClearPagePrivate(page);
835                 set_page_private(page, 0);
836                 page_cache_release(page);
837         }
838 }
839
840 static const struct address_space_operations btree_aops = {
841         .readpage       = btree_readpage,
842         .writepage      = btree_writepage,
843         .writepages     = btree_writepages,
844         .releasepage    = btree_releasepage,
845         .invalidatepage = btree_invalidatepage,
846         .sync_page      = block_sync_page,
847 #ifdef CONFIG_MIGRATION
848         .migratepage    = btree_migratepage,
849 #endif
850 };
851
852 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
853                          u64 parent_transid)
854 {
855         struct extent_buffer *buf = NULL;
856         struct inode *btree_inode = root->fs_info->btree_inode;
857         int ret = 0;
858
859         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
860         if (!buf)
861                 return 0;
862         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
863                                  buf, 0, 0, btree_get_extent, 0);
864         free_extent_buffer(buf);
865         return ret;
866 }
867
868 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
869                                             u64 bytenr, u32 blocksize)
870 {
871         struct inode *btree_inode = root->fs_info->btree_inode;
872         struct extent_buffer *eb;
873         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
874                                 bytenr, blocksize, GFP_NOFS);
875         return eb;
876 }
877
878 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
879                                                  u64 bytenr, u32 blocksize)
880 {
881         struct inode *btree_inode = root->fs_info->btree_inode;
882         struct extent_buffer *eb;
883
884         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
885                                  bytenr, blocksize, NULL, GFP_NOFS);
886         return eb;
887 }
888
889
890 int btrfs_write_tree_block(struct extent_buffer *buf)
891 {
892         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
893                                         buf->start + buf->len - 1);
894 }
895
896 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
897 {
898         return filemap_fdatawait_range(buf->first_page->mapping,
899                                        buf->start, buf->start + buf->len - 1);
900 }
901
902 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
903                                       u32 blocksize, u64 parent_transid)
904 {
905         struct extent_buffer *buf = NULL;
906         int ret;
907
908         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
909         if (!buf)
910                 return NULL;
911
912         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
913
914         if (ret == 0)
915                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
916         return buf;
917
918 }
919
920 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
921                      struct extent_buffer *buf)
922 {
923         struct inode *btree_inode = root->fs_info->btree_inode;
924         if (btrfs_header_generation(buf) ==
925             root->fs_info->running_transaction->transid) {
926                 btrfs_assert_tree_locked(buf);
927
928                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
929                         spin_lock(&root->fs_info->delalloc_lock);
930                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
931                                 root->fs_info->dirty_metadata_bytes -= buf->len;
932                         else
933                                 WARN_ON(1);
934                         spin_unlock(&root->fs_info->delalloc_lock);
935                 }
936
937                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
938                 btrfs_set_lock_blocking(buf);
939                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
940                                           buf);
941         }
942         return 0;
943 }
944
945 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
946                         u32 stripesize, struct btrfs_root *root,
947                         struct btrfs_fs_info *fs_info,
948                         u64 objectid)
949 {
950         root->node = NULL;
951         root->commit_root = NULL;
952         root->sectorsize = sectorsize;
953         root->nodesize = nodesize;
954         root->leafsize = leafsize;
955         root->stripesize = stripesize;
956         root->ref_cows = 0;
957         root->track_dirty = 0;
958         root->in_radix = 0;
959         root->orphan_item_inserted = 0;
960         root->orphan_cleanup_state = 0;
961
962         root->fs_info = fs_info;
963         root->objectid = objectid;
964         root->last_trans = 0;
965         root->highest_objectid = 0;
966         root->name = NULL;
967         root->in_sysfs = 0;
968         root->inode_tree = RB_ROOT;
969         root->block_rsv = NULL;
970         root->orphan_block_rsv = NULL;
971
972         INIT_LIST_HEAD(&root->dirty_list);
973         INIT_LIST_HEAD(&root->orphan_list);
974         INIT_LIST_HEAD(&root->root_list);
975         spin_lock_init(&root->node_lock);
976         spin_lock_init(&root->orphan_lock);
977         spin_lock_init(&root->inode_lock);
978         spin_lock_init(&root->accounting_lock);
979         mutex_init(&root->objectid_mutex);
980         mutex_init(&root->log_mutex);
981         init_waitqueue_head(&root->log_writer_wait);
982         init_waitqueue_head(&root->log_commit_wait[0]);
983         init_waitqueue_head(&root->log_commit_wait[1]);
984         atomic_set(&root->log_commit[0], 0);
985         atomic_set(&root->log_commit[1], 0);
986         atomic_set(&root->log_writers, 0);
987         root->log_batch = 0;
988         root->log_transid = 0;
989         root->last_log_commit = 0;
990         extent_io_tree_init(&root->dirty_log_pages,
991                              fs_info->btree_inode->i_mapping, GFP_NOFS);
992
993         memset(&root->root_key, 0, sizeof(root->root_key));
994         memset(&root->root_item, 0, sizeof(root->root_item));
995         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
996         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
997         root->defrag_trans_start = fs_info->generation;
998         init_completion(&root->kobj_unregister);
999         root->defrag_running = 0;
1000         root->root_key.objectid = objectid;
1001         root->anon_super.s_root = NULL;
1002         root->anon_super.s_dev = 0;
1003         INIT_LIST_HEAD(&root->anon_super.s_list);
1004         INIT_LIST_HEAD(&root->anon_super.s_instances);
1005         init_rwsem(&root->anon_super.s_umount);
1006
1007         return 0;
1008 }
1009
1010 static int find_and_setup_root(struct btrfs_root *tree_root,
1011                                struct btrfs_fs_info *fs_info,
1012                                u64 objectid,
1013                                struct btrfs_root *root)
1014 {
1015         int ret;
1016         u32 blocksize;
1017         u64 generation;
1018
1019         __setup_root(tree_root->nodesize, tree_root->leafsize,
1020                      tree_root->sectorsize, tree_root->stripesize,
1021                      root, fs_info, objectid);
1022         ret = btrfs_find_last_root(tree_root, objectid,
1023                                    &root->root_item, &root->root_key);
1024         if (ret > 0)
1025                 return -ENOENT;
1026         BUG_ON(ret);
1027
1028         generation = btrfs_root_generation(&root->root_item);
1029         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1030         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1031                                      blocksize, generation);
1032         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1033                 free_extent_buffer(root->node);
1034                 return -EIO;
1035         }
1036         root->commit_root = btrfs_root_node(root);
1037         return 0;
1038 }
1039
1040 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1041                                          struct btrfs_fs_info *fs_info)
1042 {
1043         struct btrfs_root *root;
1044         struct btrfs_root *tree_root = fs_info->tree_root;
1045         struct extent_buffer *leaf;
1046
1047         root = kzalloc(sizeof(*root), GFP_NOFS);
1048         if (!root)
1049                 return ERR_PTR(-ENOMEM);
1050
1051         __setup_root(tree_root->nodesize, tree_root->leafsize,
1052                      tree_root->sectorsize, tree_root->stripesize,
1053                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1054
1055         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1056         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1057         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1058         /*
1059          * log trees do not get reference counted because they go away
1060          * before a real commit is actually done.  They do store pointers
1061          * to file data extents, and those reference counts still get
1062          * updated (along with back refs to the log tree).
1063          */
1064         root->ref_cows = 0;
1065
1066         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1067                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1068         if (IS_ERR(leaf)) {
1069                 kfree(root);
1070                 return ERR_CAST(leaf);
1071         }
1072
1073         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1074         btrfs_set_header_bytenr(leaf, leaf->start);
1075         btrfs_set_header_generation(leaf, trans->transid);
1076         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1077         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1078         root->node = leaf;
1079
1080         write_extent_buffer(root->node, root->fs_info->fsid,
1081                             (unsigned long)btrfs_header_fsid(root->node),
1082                             BTRFS_FSID_SIZE);
1083         btrfs_mark_buffer_dirty(root->node);
1084         btrfs_tree_unlock(root->node);
1085         return root;
1086 }
1087
1088 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1089                              struct btrfs_fs_info *fs_info)
1090 {
1091         struct btrfs_root *log_root;
1092
1093         log_root = alloc_log_tree(trans, fs_info);
1094         if (IS_ERR(log_root))
1095                 return PTR_ERR(log_root);
1096         WARN_ON(fs_info->log_root_tree);
1097         fs_info->log_root_tree = log_root;
1098         return 0;
1099 }
1100
1101 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1102                        struct btrfs_root *root)
1103 {
1104         struct btrfs_root *log_root;
1105         struct btrfs_inode_item *inode_item;
1106
1107         log_root = alloc_log_tree(trans, root->fs_info);
1108         if (IS_ERR(log_root))
1109                 return PTR_ERR(log_root);
1110
1111         log_root->last_trans = trans->transid;
1112         log_root->root_key.offset = root->root_key.objectid;
1113
1114         inode_item = &log_root->root_item.inode;
1115         inode_item->generation = cpu_to_le64(1);
1116         inode_item->size = cpu_to_le64(3);
1117         inode_item->nlink = cpu_to_le32(1);
1118         inode_item->nbytes = cpu_to_le64(root->leafsize);
1119         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1120
1121         btrfs_set_root_node(&log_root->root_item, log_root->node);
1122
1123         WARN_ON(root->log_root);
1124         root->log_root = log_root;
1125         root->log_transid = 0;
1126         root->last_log_commit = 0;
1127         return 0;
1128 }
1129
1130 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1131                                                struct btrfs_key *location)
1132 {
1133         struct btrfs_root *root;
1134         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1135         struct btrfs_path *path;
1136         struct extent_buffer *l;
1137         u64 generation;
1138         u32 blocksize;
1139         int ret = 0;
1140
1141         root = kzalloc(sizeof(*root), GFP_NOFS);
1142         if (!root)
1143                 return ERR_PTR(-ENOMEM);
1144         if (location->offset == (u64)-1) {
1145                 ret = find_and_setup_root(tree_root, fs_info,
1146                                           location->objectid, root);
1147                 if (ret) {
1148                         kfree(root);
1149                         return ERR_PTR(ret);
1150                 }
1151                 goto out;
1152         }
1153
1154         __setup_root(tree_root->nodesize, tree_root->leafsize,
1155                      tree_root->sectorsize, tree_root->stripesize,
1156                      root, fs_info, location->objectid);
1157
1158         path = btrfs_alloc_path();
1159         BUG_ON(!path);
1160         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1161         if (ret == 0) {
1162                 l = path->nodes[0];
1163                 read_extent_buffer(l, &root->root_item,
1164                                 btrfs_item_ptr_offset(l, path->slots[0]),
1165                                 sizeof(root->root_item));
1166                 memcpy(&root->root_key, location, sizeof(*location));
1167         }
1168         btrfs_free_path(path);
1169         if (ret) {
1170                 kfree(root);
1171                 if (ret > 0)
1172                         ret = -ENOENT;
1173                 return ERR_PTR(ret);
1174         }
1175
1176         generation = btrfs_root_generation(&root->root_item);
1177         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1178         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1179                                      blocksize, generation);
1180         root->commit_root = btrfs_root_node(root);
1181         BUG_ON(!root->node);
1182 out:
1183         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1184                 root->ref_cows = 1;
1185
1186         return root;
1187 }
1188
1189 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1190                                         u64 root_objectid)
1191 {
1192         struct btrfs_root *root;
1193
1194         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1195                 return fs_info->tree_root;
1196         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1197                 return fs_info->extent_root;
1198
1199         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1200                                  (unsigned long)root_objectid);
1201         return root;
1202 }
1203
1204 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1205                                               struct btrfs_key *location)
1206 {
1207         struct btrfs_root *root;
1208         int ret;
1209
1210         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1211                 return fs_info->tree_root;
1212         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1213                 return fs_info->extent_root;
1214         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1215                 return fs_info->chunk_root;
1216         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1217                 return fs_info->dev_root;
1218         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1219                 return fs_info->csum_root;
1220 again:
1221         spin_lock(&fs_info->fs_roots_radix_lock);
1222         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1223                                  (unsigned long)location->objectid);
1224         spin_unlock(&fs_info->fs_roots_radix_lock);
1225         if (root)
1226                 return root;
1227
1228         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1229         if (IS_ERR(root))
1230                 return root;
1231
1232         set_anon_super(&root->anon_super, NULL);
1233
1234         if (btrfs_root_refs(&root->root_item) == 0) {
1235                 ret = -ENOENT;
1236                 goto fail;
1237         }
1238
1239         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1240         if (ret < 0)
1241                 goto fail;
1242         if (ret == 0)
1243                 root->orphan_item_inserted = 1;
1244
1245         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1246         if (ret)
1247                 goto fail;
1248
1249         spin_lock(&fs_info->fs_roots_radix_lock);
1250         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1251                                 (unsigned long)root->root_key.objectid,
1252                                 root);
1253         if (ret == 0)
1254                 root->in_radix = 1;
1255
1256         spin_unlock(&fs_info->fs_roots_radix_lock);
1257         radix_tree_preload_end();
1258         if (ret) {
1259                 if (ret == -EEXIST) {
1260                         free_fs_root(root);
1261                         goto again;
1262                 }
1263                 goto fail;
1264         }
1265
1266         ret = btrfs_find_dead_roots(fs_info->tree_root,
1267                                     root->root_key.objectid);
1268         WARN_ON(ret);
1269         return root;
1270 fail:
1271         free_fs_root(root);
1272         return ERR_PTR(ret);
1273 }
1274
1275 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1276                                       struct btrfs_key *location,
1277                                       const char *name, int namelen)
1278 {
1279         return btrfs_read_fs_root_no_name(fs_info, location);
1280 #if 0
1281         struct btrfs_root *root;
1282         int ret;
1283
1284         root = btrfs_read_fs_root_no_name(fs_info, location);
1285         if (!root)
1286                 return NULL;
1287
1288         if (root->in_sysfs)
1289                 return root;
1290
1291         ret = btrfs_set_root_name(root, name, namelen);
1292         if (ret) {
1293                 free_extent_buffer(root->node);
1294                 kfree(root);
1295                 return ERR_PTR(ret);
1296         }
1297
1298         ret = btrfs_sysfs_add_root(root);
1299         if (ret) {
1300                 free_extent_buffer(root->node);
1301                 kfree(root->name);
1302                 kfree(root);
1303                 return ERR_PTR(ret);
1304         }
1305         root->in_sysfs = 1;
1306         return root;
1307 #endif
1308 }
1309
1310 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1311 {
1312         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1313         int ret = 0;
1314         struct btrfs_device *device;
1315         struct backing_dev_info *bdi;
1316
1317         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1318                 if (!device->bdev)
1319                         continue;
1320                 bdi = blk_get_backing_dev_info(device->bdev);
1321                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1322                         ret = 1;
1323                         break;
1324                 }
1325         }
1326         return ret;
1327 }
1328
1329 /*
1330  * this unplugs every device on the box, and it is only used when page
1331  * is null
1332  */
1333 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1334 {
1335         struct btrfs_device *device;
1336         struct btrfs_fs_info *info;
1337
1338         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1339         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1340                 if (!device->bdev)
1341                         continue;
1342
1343                 bdi = blk_get_backing_dev_info(device->bdev);
1344                 if (bdi->unplug_io_fn)
1345                         bdi->unplug_io_fn(bdi, page);
1346         }
1347 }
1348
1349 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1350 {
1351         struct inode *inode;
1352         struct extent_map_tree *em_tree;
1353         struct extent_map *em;
1354         struct address_space *mapping;
1355         u64 offset;
1356
1357         /* the generic O_DIRECT read code does this */
1358         if (1 || !page) {
1359                 __unplug_io_fn(bdi, page);
1360                 return;
1361         }
1362
1363         /*
1364          * page->mapping may change at any time.  Get a consistent copy
1365          * and use that for everything below
1366          */
1367         smp_mb();
1368         mapping = page->mapping;
1369         if (!mapping)
1370                 return;
1371
1372         inode = mapping->host;
1373
1374         /*
1375          * don't do the expensive searching for a small number of
1376          * devices
1377          */
1378         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1379                 __unplug_io_fn(bdi, page);
1380                 return;
1381         }
1382
1383         offset = page_offset(page);
1384
1385         em_tree = &BTRFS_I(inode)->extent_tree;
1386         read_lock(&em_tree->lock);
1387         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1388         read_unlock(&em_tree->lock);
1389         if (!em) {
1390                 __unplug_io_fn(bdi, page);
1391                 return;
1392         }
1393
1394         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1395                 free_extent_map(em);
1396                 __unplug_io_fn(bdi, page);
1397                 return;
1398         }
1399         offset = offset - em->start;
1400         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1401                           em->block_start + offset, page);
1402         free_extent_map(em);
1403 }
1404
1405 /*
1406  * If this fails, caller must call bdi_destroy() to get rid of the
1407  * bdi again.
1408  */
1409 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1410 {
1411         int err;
1412
1413         bdi->capabilities = BDI_CAP_MAP_COPY;
1414         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1415         if (err)
1416                 return err;
1417
1418         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1419         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1420         bdi->unplug_io_data     = info;
1421         bdi->congested_fn       = btrfs_congested_fn;
1422         bdi->congested_data     = info;
1423         return 0;
1424 }
1425
1426 static int bio_ready_for_csum(struct bio *bio)
1427 {
1428         u64 length = 0;
1429         u64 buf_len = 0;
1430         u64 start = 0;
1431         struct page *page;
1432         struct extent_io_tree *io_tree = NULL;
1433         struct bio_vec *bvec;
1434         int i;
1435         int ret;
1436
1437         bio_for_each_segment(bvec, bio, i) {
1438                 page = bvec->bv_page;
1439                 if (page->private == EXTENT_PAGE_PRIVATE) {
1440                         length += bvec->bv_len;
1441                         continue;
1442                 }
1443                 if (!page->private) {
1444                         length += bvec->bv_len;
1445                         continue;
1446                 }
1447                 length = bvec->bv_len;
1448                 buf_len = page->private >> 2;
1449                 start = page_offset(page) + bvec->bv_offset;
1450                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1451         }
1452         /* are we fully contained in this bio? */
1453         if (buf_len <= length)
1454                 return 1;
1455
1456         ret = extent_range_uptodate(io_tree, start + length,
1457                                     start + buf_len - 1);
1458         return ret;
1459 }
1460
1461 /*
1462  * called by the kthread helper functions to finally call the bio end_io
1463  * functions.  This is where read checksum verification actually happens
1464  */
1465 static void end_workqueue_fn(struct btrfs_work *work)
1466 {
1467         struct bio *bio;
1468         struct end_io_wq *end_io_wq;
1469         struct btrfs_fs_info *fs_info;
1470         int error;
1471
1472         end_io_wq = container_of(work, struct end_io_wq, work);
1473         bio = end_io_wq->bio;
1474         fs_info = end_io_wq->info;
1475
1476         /* metadata bio reads are special because the whole tree block must
1477          * be checksummed at once.  This makes sure the entire block is in
1478          * ram and up to date before trying to verify things.  For
1479          * blocksize <= pagesize, it is basically a noop
1480          */
1481         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1482             !bio_ready_for_csum(bio)) {
1483                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1484                                    &end_io_wq->work);
1485                 return;
1486         }
1487         error = end_io_wq->error;
1488         bio->bi_private = end_io_wq->private;
1489         bio->bi_end_io = end_io_wq->end_io;
1490         kfree(end_io_wq);
1491         bio_endio(bio, error);
1492 }
1493
1494 static int cleaner_kthread(void *arg)
1495 {
1496         struct btrfs_root *root = arg;
1497
1498         do {
1499                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1500
1501                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1502                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1503                         btrfs_run_delayed_iputs(root);
1504                         btrfs_clean_old_snapshots(root);
1505                         mutex_unlock(&root->fs_info->cleaner_mutex);
1506                 }
1507
1508                 if (freezing(current)) {
1509                         refrigerator();
1510                 } else {
1511                         set_current_state(TASK_INTERRUPTIBLE);
1512                         if (!kthread_should_stop())
1513                                 schedule();
1514                         __set_current_state(TASK_RUNNING);
1515                 }
1516         } while (!kthread_should_stop());
1517         return 0;
1518 }
1519
1520 static int transaction_kthread(void *arg)
1521 {
1522         struct btrfs_root *root = arg;
1523         struct btrfs_trans_handle *trans;
1524         struct btrfs_transaction *cur;
1525         u64 transid;
1526         unsigned long now;
1527         unsigned long delay;
1528         int ret;
1529
1530         do {
1531                 delay = HZ * 30;
1532                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1533                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1534
1535                 spin_lock(&root->fs_info->new_trans_lock);
1536                 cur = root->fs_info->running_transaction;
1537                 if (!cur) {
1538                         spin_unlock(&root->fs_info->new_trans_lock);
1539                         goto sleep;
1540                 }
1541
1542                 now = get_seconds();
1543                 if (!cur->blocked &&
1544                     (now < cur->start_time || now - cur->start_time < 30)) {
1545                         spin_unlock(&root->fs_info->new_trans_lock);
1546                         delay = HZ * 5;
1547                         goto sleep;
1548                 }
1549                 transid = cur->transid;
1550                 spin_unlock(&root->fs_info->new_trans_lock);
1551
1552                 trans = btrfs_join_transaction(root, 1);
1553                 if (transid == trans->transid) {
1554                         ret = btrfs_commit_transaction(trans, root);
1555                         BUG_ON(ret);
1556                 } else {
1557                         btrfs_end_transaction(trans, root);
1558                 }
1559 sleep:
1560                 wake_up_process(root->fs_info->cleaner_kthread);
1561                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1562
1563                 if (freezing(current)) {
1564                         refrigerator();
1565                 } else {
1566                         set_current_state(TASK_INTERRUPTIBLE);
1567                         if (!kthread_should_stop() &&
1568                             !btrfs_transaction_blocked(root->fs_info))
1569                                 schedule_timeout(delay);
1570                         __set_current_state(TASK_RUNNING);
1571                 }
1572         } while (!kthread_should_stop());
1573         return 0;
1574 }
1575
1576 struct btrfs_root *open_ctree(struct super_block *sb,
1577                               struct btrfs_fs_devices *fs_devices,
1578                               char *options)
1579 {
1580         u32 sectorsize;
1581         u32 nodesize;
1582         u32 leafsize;
1583         u32 blocksize;
1584         u32 stripesize;
1585         u64 generation;
1586         u64 features;
1587         struct btrfs_key location;
1588         struct buffer_head *bh;
1589         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1590                                                  GFP_NOFS);
1591         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1592                                                  GFP_NOFS);
1593         struct btrfs_root *tree_root = btrfs_sb(sb);
1594         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1595         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1596                                                 GFP_NOFS);
1597         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1598                                               GFP_NOFS);
1599         struct btrfs_root *log_tree_root;
1600
1601         int ret;
1602         int err = -EINVAL;
1603
1604         struct btrfs_super_block *disk_super;
1605
1606         if (!extent_root || !tree_root || !fs_info ||
1607             !chunk_root || !dev_root || !csum_root) {
1608                 err = -ENOMEM;
1609                 goto fail;
1610         }
1611
1612         ret = init_srcu_struct(&fs_info->subvol_srcu);
1613         if (ret) {
1614                 err = ret;
1615                 goto fail;
1616         }
1617
1618         ret = setup_bdi(fs_info, &fs_info->bdi);
1619         if (ret) {
1620                 err = ret;
1621                 goto fail_srcu;
1622         }
1623
1624         fs_info->btree_inode = new_inode(sb);
1625         if (!fs_info->btree_inode) {
1626                 err = -ENOMEM;
1627                 goto fail_bdi;
1628         }
1629
1630         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1631         INIT_LIST_HEAD(&fs_info->trans_list);
1632         INIT_LIST_HEAD(&fs_info->dead_roots);
1633         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1634         INIT_LIST_HEAD(&fs_info->hashers);
1635         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1636         INIT_LIST_HEAD(&fs_info->ordered_operations);
1637         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1638         spin_lock_init(&fs_info->delalloc_lock);
1639         spin_lock_init(&fs_info->new_trans_lock);
1640         spin_lock_init(&fs_info->ref_cache_lock);
1641         spin_lock_init(&fs_info->fs_roots_radix_lock);
1642         spin_lock_init(&fs_info->delayed_iput_lock);
1643
1644         init_completion(&fs_info->kobj_unregister);
1645         fs_info->tree_root = tree_root;
1646         fs_info->extent_root = extent_root;
1647         fs_info->csum_root = csum_root;
1648         fs_info->chunk_root = chunk_root;
1649         fs_info->dev_root = dev_root;
1650         fs_info->fs_devices = fs_devices;
1651         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1652         INIT_LIST_HEAD(&fs_info->space_info);
1653         btrfs_mapping_init(&fs_info->mapping_tree);
1654         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1655         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1656         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1657         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1658         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1659         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1660         mutex_init(&fs_info->durable_block_rsv_mutex);
1661         atomic_set(&fs_info->nr_async_submits, 0);
1662         atomic_set(&fs_info->async_delalloc_pages, 0);
1663         atomic_set(&fs_info->async_submit_draining, 0);
1664         atomic_set(&fs_info->nr_async_bios, 0);
1665         fs_info->sb = sb;
1666         fs_info->max_inline = 8192 * 1024;
1667         fs_info->metadata_ratio = 0;
1668
1669         fs_info->thread_pool_size = min_t(unsigned long,
1670                                           num_online_cpus() + 2, 8);
1671
1672         INIT_LIST_HEAD(&fs_info->ordered_extents);
1673         spin_lock_init(&fs_info->ordered_extent_lock);
1674
1675         sb->s_blocksize = 4096;
1676         sb->s_blocksize_bits = blksize_bits(4096);
1677         sb->s_bdi = &fs_info->bdi;
1678
1679         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1680         fs_info->btree_inode->i_nlink = 1;
1681         /*
1682          * we set the i_size on the btree inode to the max possible int.
1683          * the real end of the address space is determined by all of
1684          * the devices in the system
1685          */
1686         fs_info->btree_inode->i_size = OFFSET_MAX;
1687         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1688         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1689
1690         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1691         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1692                              fs_info->btree_inode->i_mapping,
1693                              GFP_NOFS);
1694         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1695                              GFP_NOFS);
1696
1697         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1698
1699         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1700         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1701                sizeof(struct btrfs_key));
1702         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1703         insert_inode_hash(fs_info->btree_inode);
1704
1705         spin_lock_init(&fs_info->block_group_cache_lock);
1706         fs_info->block_group_cache_tree = RB_ROOT;
1707
1708         extent_io_tree_init(&fs_info->freed_extents[0],
1709                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1710         extent_io_tree_init(&fs_info->freed_extents[1],
1711                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1712         fs_info->pinned_extents = &fs_info->freed_extents[0];
1713         fs_info->do_barriers = 1;
1714
1715
1716         mutex_init(&fs_info->trans_mutex);
1717         mutex_init(&fs_info->ordered_operations_mutex);
1718         mutex_init(&fs_info->tree_log_mutex);
1719         mutex_init(&fs_info->chunk_mutex);
1720         mutex_init(&fs_info->transaction_kthread_mutex);
1721         mutex_init(&fs_info->cleaner_mutex);
1722         mutex_init(&fs_info->volume_mutex);
1723         init_rwsem(&fs_info->extent_commit_sem);
1724         init_rwsem(&fs_info->cleanup_work_sem);
1725         init_rwsem(&fs_info->subvol_sem);
1726
1727         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1728         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1729
1730         init_waitqueue_head(&fs_info->transaction_throttle);
1731         init_waitqueue_head(&fs_info->transaction_wait);
1732         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1733         init_waitqueue_head(&fs_info->async_submit_wait);
1734
1735         __setup_root(4096, 4096, 4096, 4096, tree_root,
1736                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1737
1738         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1739         if (!bh) {
1740                 err = -EINVAL;
1741                 goto fail_iput;
1742         }
1743
1744         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1745         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1746                sizeof(fs_info->super_for_commit));
1747         brelse(bh);
1748
1749         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1750
1751         disk_super = &fs_info->super_copy;
1752         if (!btrfs_super_root(disk_super))
1753                 goto fail_iput;
1754
1755         /* check FS state, whether FS is broken. */
1756         fs_info->fs_state |= btrfs_super_flags(disk_super);
1757
1758         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1759
1760         ret = btrfs_parse_options(tree_root, options);
1761         if (ret) {
1762                 err = ret;
1763                 goto fail_iput;
1764         }
1765
1766         features = btrfs_super_incompat_flags(disk_super) &
1767                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1768         if (features) {
1769                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1770                        "unsupported optional features (%Lx).\n",
1771                        (unsigned long long)features);
1772                 err = -EINVAL;
1773                 goto fail_iput;
1774         }
1775
1776         features = btrfs_super_incompat_flags(disk_super);
1777         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1778         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1779                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1780         btrfs_set_super_incompat_flags(disk_super, features);
1781
1782         features = btrfs_super_compat_ro_flags(disk_super) &
1783                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1784         if (!(sb->s_flags & MS_RDONLY) && features) {
1785                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1786                        "unsupported option features (%Lx).\n",
1787                        (unsigned long long)features);
1788                 err = -EINVAL;
1789                 goto fail_iput;
1790         }
1791
1792         btrfs_init_workers(&fs_info->generic_worker,
1793                            "genwork", 1, NULL);
1794
1795         btrfs_init_workers(&fs_info->workers, "worker",
1796                            fs_info->thread_pool_size,
1797                            &fs_info->generic_worker);
1798
1799         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1800                            fs_info->thread_pool_size,
1801                            &fs_info->generic_worker);
1802
1803         btrfs_init_workers(&fs_info->submit_workers, "submit",
1804                            min_t(u64, fs_devices->num_devices,
1805                            fs_info->thread_pool_size),
1806                            &fs_info->generic_worker);
1807
1808         /* a higher idle thresh on the submit workers makes it much more
1809          * likely that bios will be send down in a sane order to the
1810          * devices
1811          */
1812         fs_info->submit_workers.idle_thresh = 64;
1813
1814         fs_info->workers.idle_thresh = 16;
1815         fs_info->workers.ordered = 1;
1816
1817         fs_info->delalloc_workers.idle_thresh = 2;
1818         fs_info->delalloc_workers.ordered = 1;
1819
1820         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1821                            &fs_info->generic_worker);
1822         btrfs_init_workers(&fs_info->endio_workers, "endio",
1823                            fs_info->thread_pool_size,
1824                            &fs_info->generic_worker);
1825         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1826                            fs_info->thread_pool_size,
1827                            &fs_info->generic_worker);
1828         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1829                            "endio-meta-write", fs_info->thread_pool_size,
1830                            &fs_info->generic_worker);
1831         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1832                            fs_info->thread_pool_size,
1833                            &fs_info->generic_worker);
1834         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1835                            1, &fs_info->generic_worker);
1836
1837         /*
1838          * endios are largely parallel and should have a very
1839          * low idle thresh
1840          */
1841         fs_info->endio_workers.idle_thresh = 4;
1842         fs_info->endio_meta_workers.idle_thresh = 4;
1843
1844         fs_info->endio_write_workers.idle_thresh = 2;
1845         fs_info->endio_meta_write_workers.idle_thresh = 2;
1846
1847         btrfs_start_workers(&fs_info->workers, 1);
1848         btrfs_start_workers(&fs_info->generic_worker, 1);
1849         btrfs_start_workers(&fs_info->submit_workers, 1);
1850         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1851         btrfs_start_workers(&fs_info->fixup_workers, 1);
1852         btrfs_start_workers(&fs_info->endio_workers, 1);
1853         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1854         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1855         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1856         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1857
1858         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1859         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1860                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1861
1862         nodesize = btrfs_super_nodesize(disk_super);
1863         leafsize = btrfs_super_leafsize(disk_super);
1864         sectorsize = btrfs_super_sectorsize(disk_super);
1865         stripesize = btrfs_super_stripesize(disk_super);
1866         tree_root->nodesize = nodesize;
1867         tree_root->leafsize = leafsize;
1868         tree_root->sectorsize = sectorsize;
1869         tree_root->stripesize = stripesize;
1870
1871         sb->s_blocksize = sectorsize;
1872         sb->s_blocksize_bits = blksize_bits(sectorsize);
1873
1874         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1875                     sizeof(disk_super->magic))) {
1876                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1877                 goto fail_sb_buffer;
1878         }
1879
1880         mutex_lock(&fs_info->chunk_mutex);
1881         ret = btrfs_read_sys_array(tree_root);
1882         mutex_unlock(&fs_info->chunk_mutex);
1883         if (ret) {
1884                 printk(KERN_WARNING "btrfs: failed to read the system "
1885                        "array on %s\n", sb->s_id);
1886                 goto fail_sb_buffer;
1887         }
1888
1889         blocksize = btrfs_level_size(tree_root,
1890                                      btrfs_super_chunk_root_level(disk_super));
1891         generation = btrfs_super_chunk_root_generation(disk_super);
1892
1893         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1894                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1895
1896         chunk_root->node = read_tree_block(chunk_root,
1897                                            btrfs_super_chunk_root(disk_super),
1898                                            blocksize, generation);
1899         BUG_ON(!chunk_root->node);
1900         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1901                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1902                        sb->s_id);
1903                 goto fail_chunk_root;
1904         }
1905         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1906         chunk_root->commit_root = btrfs_root_node(chunk_root);
1907
1908         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1909            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1910            BTRFS_UUID_SIZE);
1911
1912         mutex_lock(&fs_info->chunk_mutex);
1913         ret = btrfs_read_chunk_tree(chunk_root);
1914         mutex_unlock(&fs_info->chunk_mutex);
1915         if (ret) {
1916                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1917                        sb->s_id);
1918                 goto fail_chunk_root;
1919         }
1920
1921         btrfs_close_extra_devices(fs_devices);
1922
1923         blocksize = btrfs_level_size(tree_root,
1924                                      btrfs_super_root_level(disk_super));
1925         generation = btrfs_super_generation(disk_super);
1926
1927         tree_root->node = read_tree_block(tree_root,
1928                                           btrfs_super_root(disk_super),
1929                                           blocksize, generation);
1930         if (!tree_root->node)
1931                 goto fail_chunk_root;
1932         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1933                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1934                        sb->s_id);
1935                 goto fail_tree_root;
1936         }
1937         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1938         tree_root->commit_root = btrfs_root_node(tree_root);
1939
1940         ret = find_and_setup_root(tree_root, fs_info,
1941                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1942         if (ret)
1943                 goto fail_tree_root;
1944         extent_root->track_dirty = 1;
1945
1946         ret = find_and_setup_root(tree_root, fs_info,
1947                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1948         if (ret)
1949                 goto fail_extent_root;
1950         dev_root->track_dirty = 1;
1951
1952         ret = find_and_setup_root(tree_root, fs_info,
1953                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1954         if (ret)
1955                 goto fail_dev_root;
1956
1957         csum_root->track_dirty = 1;
1958
1959         fs_info->generation = generation;
1960         fs_info->last_trans_committed = generation;
1961         fs_info->data_alloc_profile = (u64)-1;
1962         fs_info->metadata_alloc_profile = (u64)-1;
1963         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1964
1965         ret = btrfs_read_block_groups(extent_root);
1966         if (ret) {
1967                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1968                 goto fail_block_groups;
1969         }
1970
1971         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1972                                                "btrfs-cleaner");
1973         if (IS_ERR(fs_info->cleaner_kthread))
1974                 goto fail_block_groups;
1975
1976         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1977                                                    tree_root,
1978                                                    "btrfs-transaction");
1979         if (IS_ERR(fs_info->transaction_kthread))
1980                 goto fail_cleaner;
1981
1982         if (!btrfs_test_opt(tree_root, SSD) &&
1983             !btrfs_test_opt(tree_root, NOSSD) &&
1984             !fs_info->fs_devices->rotating) {
1985                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1986                        "mode\n");
1987                 btrfs_set_opt(fs_info->mount_opt, SSD);
1988         }
1989
1990         /* do not make disk changes in broken FS */
1991         if (btrfs_super_log_root(disk_super) != 0 &&
1992             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
1993                 u64 bytenr = btrfs_super_log_root(disk_super);
1994
1995                 if (fs_devices->rw_devices == 0) {
1996                         printk(KERN_WARNING "Btrfs log replay required "
1997                                "on RO media\n");
1998                         err = -EIO;
1999                         goto fail_trans_kthread;
2000                 }
2001                 blocksize =
2002                      btrfs_level_size(tree_root,
2003                                       btrfs_super_log_root_level(disk_super));
2004
2005                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2006                 if (!log_tree_root) {
2007                         err = -ENOMEM;
2008                         goto fail_trans_kthread;
2009                 }
2010
2011                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2012                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2013
2014                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2015                                                       blocksize,
2016                                                       generation + 1);
2017                 ret = btrfs_recover_log_trees(log_tree_root);
2018                 BUG_ON(ret);
2019
2020                 if (sb->s_flags & MS_RDONLY) {
2021                         ret =  btrfs_commit_super(tree_root);
2022                         BUG_ON(ret);
2023                 }
2024         }
2025
2026         ret = btrfs_find_orphan_roots(tree_root);
2027         BUG_ON(ret);
2028
2029         if (!(sb->s_flags & MS_RDONLY)) {
2030                 ret = btrfs_cleanup_fs_roots(fs_info);
2031                 BUG_ON(ret);
2032
2033                 ret = btrfs_recover_relocation(tree_root);
2034                 if (ret < 0) {
2035                         printk(KERN_WARNING
2036                                "btrfs: failed to recover relocation\n");
2037                         err = -EINVAL;
2038                         goto fail_trans_kthread;
2039                 }
2040         }
2041
2042         location.objectid = BTRFS_FS_TREE_OBJECTID;
2043         location.type = BTRFS_ROOT_ITEM_KEY;
2044         location.offset = (u64)-1;
2045
2046         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2047         if (!fs_info->fs_root)
2048                 goto fail_trans_kthread;
2049         if (IS_ERR(fs_info->fs_root)) {
2050                 err = PTR_ERR(fs_info->fs_root);
2051                 goto fail_trans_kthread;
2052         }
2053
2054         if (!(sb->s_flags & MS_RDONLY)) {
2055                 down_read(&fs_info->cleanup_work_sem);
2056                 btrfs_orphan_cleanup(fs_info->fs_root);
2057                 btrfs_orphan_cleanup(fs_info->tree_root);
2058                 up_read(&fs_info->cleanup_work_sem);
2059         }
2060
2061         return tree_root;
2062
2063 fail_trans_kthread:
2064         kthread_stop(fs_info->transaction_kthread);
2065 fail_cleaner:
2066         kthread_stop(fs_info->cleaner_kthread);
2067
2068         /*
2069          * make sure we're done with the btree inode before we stop our
2070          * kthreads
2071          */
2072         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2073         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2074
2075 fail_block_groups:
2076         btrfs_free_block_groups(fs_info);
2077         free_extent_buffer(csum_root->node);
2078         free_extent_buffer(csum_root->commit_root);
2079 fail_dev_root:
2080         free_extent_buffer(dev_root->node);
2081         free_extent_buffer(dev_root->commit_root);
2082 fail_extent_root:
2083         free_extent_buffer(extent_root->node);
2084         free_extent_buffer(extent_root->commit_root);
2085 fail_tree_root:
2086         free_extent_buffer(tree_root->node);
2087         free_extent_buffer(tree_root->commit_root);
2088 fail_chunk_root:
2089         free_extent_buffer(chunk_root->node);
2090         free_extent_buffer(chunk_root->commit_root);
2091 fail_sb_buffer:
2092         btrfs_stop_workers(&fs_info->generic_worker);
2093         btrfs_stop_workers(&fs_info->fixup_workers);
2094         btrfs_stop_workers(&fs_info->delalloc_workers);
2095         btrfs_stop_workers(&fs_info->workers);
2096         btrfs_stop_workers(&fs_info->endio_workers);
2097         btrfs_stop_workers(&fs_info->endio_meta_workers);
2098         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2099         btrfs_stop_workers(&fs_info->endio_write_workers);
2100         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2101         btrfs_stop_workers(&fs_info->submit_workers);
2102 fail_iput:
2103         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2104         iput(fs_info->btree_inode);
2105
2106         btrfs_close_devices(fs_info->fs_devices);
2107         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2108 fail_bdi:
2109         bdi_destroy(&fs_info->bdi);
2110 fail_srcu:
2111         cleanup_srcu_struct(&fs_info->subvol_srcu);
2112 fail:
2113         kfree(extent_root);
2114         kfree(tree_root);
2115         kfree(fs_info);
2116         kfree(chunk_root);
2117         kfree(dev_root);
2118         kfree(csum_root);
2119         return ERR_PTR(err);
2120 }
2121
2122 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2123 {
2124         char b[BDEVNAME_SIZE];
2125
2126         if (uptodate) {
2127                 set_buffer_uptodate(bh);
2128         } else {
2129                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2130                         printk(KERN_WARNING "lost page write due to "
2131                                         "I/O error on %s\n",
2132                                        bdevname(bh->b_bdev, b));
2133                 }
2134                 /* note, we dont' set_buffer_write_io_error because we have
2135                  * our own ways of dealing with the IO errors
2136                  */
2137                 clear_buffer_uptodate(bh);
2138         }
2139         unlock_buffer(bh);
2140         put_bh(bh);
2141 }
2142
2143 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2144 {
2145         struct buffer_head *bh;
2146         struct buffer_head *latest = NULL;
2147         struct btrfs_super_block *super;
2148         int i;
2149         u64 transid = 0;
2150         u64 bytenr;
2151
2152         /* we would like to check all the supers, but that would make
2153          * a btrfs mount succeed after a mkfs from a different FS.
2154          * So, we need to add a special mount option to scan for
2155          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2156          */
2157         for (i = 0; i < 1; i++) {
2158                 bytenr = btrfs_sb_offset(i);
2159                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2160                         break;
2161                 bh = __bread(bdev, bytenr / 4096, 4096);
2162                 if (!bh)
2163                         continue;
2164
2165                 super = (struct btrfs_super_block *)bh->b_data;
2166                 if (btrfs_super_bytenr(super) != bytenr ||
2167                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2168                             sizeof(super->magic))) {
2169                         brelse(bh);
2170                         continue;
2171                 }
2172
2173                 if (!latest || btrfs_super_generation(super) > transid) {
2174                         brelse(latest);
2175                         latest = bh;
2176                         transid = btrfs_super_generation(super);
2177                 } else {
2178                         brelse(bh);
2179                 }
2180         }
2181         return latest;
2182 }
2183
2184 /*
2185  * this should be called twice, once with wait == 0 and
2186  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2187  * we write are pinned.
2188  *
2189  * They are released when wait == 1 is done.
2190  * max_mirrors must be the same for both runs, and it indicates how
2191  * many supers on this one device should be written.
2192  *
2193  * max_mirrors == 0 means to write them all.
2194  */
2195 static int write_dev_supers(struct btrfs_device *device,
2196                             struct btrfs_super_block *sb,
2197                             int do_barriers, int wait, int max_mirrors)
2198 {
2199         struct buffer_head *bh;
2200         int i;
2201         int ret;
2202         int errors = 0;
2203         u32 crc;
2204         u64 bytenr;
2205         int last_barrier = 0;
2206
2207         if (max_mirrors == 0)
2208                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2209
2210         /* make sure only the last submit_bh does a barrier */
2211         if (do_barriers) {
2212                 for (i = 0; i < max_mirrors; i++) {
2213                         bytenr = btrfs_sb_offset(i);
2214                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2215                             device->total_bytes)
2216                                 break;
2217                         last_barrier = i;
2218                 }
2219         }
2220
2221         for (i = 0; i < max_mirrors; i++) {
2222                 bytenr = btrfs_sb_offset(i);
2223                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2224                         break;
2225
2226                 if (wait) {
2227                         bh = __find_get_block(device->bdev, bytenr / 4096,
2228                                               BTRFS_SUPER_INFO_SIZE);
2229                         BUG_ON(!bh);
2230                         wait_on_buffer(bh);
2231                         if (!buffer_uptodate(bh))
2232                                 errors++;
2233
2234                         /* drop our reference */
2235                         brelse(bh);
2236
2237                         /* drop the reference from the wait == 0 run */
2238                         brelse(bh);
2239                         continue;
2240                 } else {
2241                         btrfs_set_super_bytenr(sb, bytenr);
2242
2243                         crc = ~(u32)0;
2244                         crc = btrfs_csum_data(NULL, (char *)sb +
2245                                               BTRFS_CSUM_SIZE, crc,
2246                                               BTRFS_SUPER_INFO_SIZE -
2247                                               BTRFS_CSUM_SIZE);
2248                         btrfs_csum_final(crc, sb->csum);
2249
2250                         /*
2251                          * one reference for us, and we leave it for the
2252                          * caller
2253                          */
2254                         bh = __getblk(device->bdev, bytenr / 4096,
2255                                       BTRFS_SUPER_INFO_SIZE);
2256                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2257
2258                         /* one reference for submit_bh */
2259                         get_bh(bh);
2260
2261                         set_buffer_uptodate(bh);
2262                         lock_buffer(bh);
2263                         bh->b_end_io = btrfs_end_buffer_write_sync;
2264                 }
2265
2266                 if (i == last_barrier && do_barriers && device->barriers) {
2267                         ret = submit_bh(WRITE_BARRIER, bh);
2268                         if (ret == -EOPNOTSUPP) {
2269                                 printk("btrfs: disabling barriers on dev %s\n",
2270                                        device->name);
2271                                 set_buffer_uptodate(bh);
2272                                 device->barriers = 0;
2273                                 /* one reference for submit_bh */
2274                                 get_bh(bh);
2275                                 lock_buffer(bh);
2276                                 ret = submit_bh(WRITE_SYNC, bh);
2277                         }
2278                 } else {
2279                         ret = submit_bh(WRITE_SYNC, bh);
2280                 }
2281
2282                 if (ret)
2283                         errors++;
2284         }
2285         return errors < i ? 0 : -1;
2286 }
2287
2288 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2289 {
2290         struct list_head *head;
2291         struct btrfs_device *dev;
2292         struct btrfs_super_block *sb;
2293         struct btrfs_dev_item *dev_item;
2294         int ret;
2295         int do_barriers;
2296         int max_errors;
2297         int total_errors = 0;
2298         u64 flags;
2299
2300         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2301         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2302
2303         sb = &root->fs_info->super_for_commit;
2304         dev_item = &sb->dev_item;
2305
2306         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2307         head = &root->fs_info->fs_devices->devices;
2308         list_for_each_entry(dev, head, dev_list) {
2309                 if (!dev->bdev) {
2310                         total_errors++;
2311                         continue;
2312                 }
2313                 if (!dev->in_fs_metadata || !dev->writeable)
2314                         continue;
2315
2316                 btrfs_set_stack_device_generation(dev_item, 0);
2317                 btrfs_set_stack_device_type(dev_item, dev->type);
2318                 btrfs_set_stack_device_id(dev_item, dev->devid);
2319                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2320                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2321                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2322                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2323                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2324                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2325                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2326
2327                 flags = btrfs_super_flags(sb);
2328                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2329
2330                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2331                 if (ret)
2332                         total_errors++;
2333         }
2334         if (total_errors > max_errors) {
2335                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2336                        total_errors);
2337                 BUG();
2338         }
2339
2340         total_errors = 0;
2341         list_for_each_entry(dev, head, dev_list) {
2342                 if (!dev->bdev)
2343                         continue;
2344                 if (!dev->in_fs_metadata || !dev->writeable)
2345                         continue;
2346
2347                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2348                 if (ret)
2349                         total_errors++;
2350         }
2351         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2352         if (total_errors > max_errors) {
2353                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2354                        total_errors);
2355                 BUG();
2356         }
2357         return 0;
2358 }
2359
2360 int write_ctree_super(struct btrfs_trans_handle *trans,
2361                       struct btrfs_root *root, int max_mirrors)
2362 {
2363         int ret;
2364
2365         ret = write_all_supers(root, max_mirrors);
2366         return ret;
2367 }
2368
2369 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2370 {
2371         spin_lock(&fs_info->fs_roots_radix_lock);
2372         radix_tree_delete(&fs_info->fs_roots_radix,
2373                           (unsigned long)root->root_key.objectid);
2374         spin_unlock(&fs_info->fs_roots_radix_lock);
2375
2376         if (btrfs_root_refs(&root->root_item) == 0)
2377                 synchronize_srcu(&fs_info->subvol_srcu);
2378
2379         free_fs_root(root);
2380         return 0;
2381 }
2382
2383 static void free_fs_root(struct btrfs_root *root)
2384 {
2385         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2386         if (root->anon_super.s_dev) {
2387                 down_write(&root->anon_super.s_umount);
2388                 kill_anon_super(&root->anon_super);
2389         }
2390         free_extent_buffer(root->node);
2391         free_extent_buffer(root->commit_root);
2392         kfree(root->name);
2393         kfree(root);
2394 }
2395
2396 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2397 {
2398         int ret;
2399         struct btrfs_root *gang[8];
2400         int i;
2401
2402         while (!list_empty(&fs_info->dead_roots)) {
2403                 gang[0] = list_entry(fs_info->dead_roots.next,
2404                                      struct btrfs_root, root_list);
2405                 list_del(&gang[0]->root_list);
2406
2407                 if (gang[0]->in_radix) {
2408                         btrfs_free_fs_root(fs_info, gang[0]);
2409                 } else {
2410                         free_extent_buffer(gang[0]->node);
2411                         free_extent_buffer(gang[0]->commit_root);
2412                         kfree(gang[0]);
2413                 }
2414         }
2415
2416         while (1) {
2417                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2418                                              (void **)gang, 0,
2419                                              ARRAY_SIZE(gang));
2420                 if (!ret)
2421                         break;
2422                 for (i = 0; i < ret; i++)
2423                         btrfs_free_fs_root(fs_info, gang[i]);
2424         }
2425         return 0;
2426 }
2427
2428 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2429 {
2430         u64 root_objectid = 0;
2431         struct btrfs_root *gang[8];
2432         int i;
2433         int ret;
2434
2435         while (1) {
2436                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2437                                              (void **)gang, root_objectid,
2438                                              ARRAY_SIZE(gang));
2439                 if (!ret)
2440                         break;
2441
2442                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2443                 for (i = 0; i < ret; i++) {
2444                         root_objectid = gang[i]->root_key.objectid;
2445                         btrfs_orphan_cleanup(gang[i]);
2446                 }
2447                 root_objectid++;
2448         }
2449         return 0;
2450 }
2451
2452 int btrfs_commit_super(struct btrfs_root *root)
2453 {
2454         struct btrfs_trans_handle *trans;
2455         int ret;
2456
2457         mutex_lock(&root->fs_info->cleaner_mutex);
2458         btrfs_run_delayed_iputs(root);
2459         btrfs_clean_old_snapshots(root);
2460         mutex_unlock(&root->fs_info->cleaner_mutex);
2461
2462         /* wait until ongoing cleanup work done */
2463         down_write(&root->fs_info->cleanup_work_sem);
2464         up_write(&root->fs_info->cleanup_work_sem);
2465
2466         trans = btrfs_join_transaction(root, 1);
2467         ret = btrfs_commit_transaction(trans, root);
2468         BUG_ON(ret);
2469         /* run commit again to drop the original snapshot */
2470         trans = btrfs_join_transaction(root, 1);
2471         btrfs_commit_transaction(trans, root);
2472         ret = btrfs_write_and_wait_transaction(NULL, root);
2473         BUG_ON(ret);
2474
2475         ret = write_ctree_super(NULL, root, 0);
2476         return ret;
2477 }
2478
2479 int close_ctree(struct btrfs_root *root)
2480 {
2481         struct btrfs_fs_info *fs_info = root->fs_info;
2482         int ret;
2483
2484         fs_info->closing = 1;
2485         smp_mb();
2486
2487         btrfs_put_block_group_cache(fs_info);
2488
2489         /*
2490          * Here come 2 situations when btrfs is broken to flip readonly:
2491          *
2492          * 1. when btrfs flips readonly somewhere else before
2493          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2494          * and btrfs will skip to write sb directly to keep
2495          * ERROR state on disk.
2496          *
2497          * 2. when btrfs flips readonly just in btrfs_commit_super,
2498          * and in such case, btrfs cannnot write sb via btrfs_commit_super,
2499          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2500          * btrfs will cleanup all FS resources first and write sb then.
2501          */
2502         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2503                 ret = btrfs_commit_super(root);
2504                 if (ret)
2505                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2506         }
2507
2508         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2509                 ret = btrfs_error_commit_super(root);
2510                 if (ret)
2511                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2512         }
2513
2514         kthread_stop(root->fs_info->transaction_kthread);
2515         kthread_stop(root->fs_info->cleaner_kthread);
2516
2517         fs_info->closing = 2;
2518         smp_mb();
2519
2520         if (fs_info->delalloc_bytes) {
2521                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2522                        (unsigned long long)fs_info->delalloc_bytes);
2523         }
2524         if (fs_info->total_ref_cache_size) {
2525                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2526                        (unsigned long long)fs_info->total_ref_cache_size);
2527         }
2528
2529         free_extent_buffer(fs_info->extent_root->node);
2530         free_extent_buffer(fs_info->extent_root->commit_root);
2531         free_extent_buffer(fs_info->tree_root->node);
2532         free_extent_buffer(fs_info->tree_root->commit_root);
2533         free_extent_buffer(root->fs_info->chunk_root->node);
2534         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2535         free_extent_buffer(root->fs_info->dev_root->node);
2536         free_extent_buffer(root->fs_info->dev_root->commit_root);
2537         free_extent_buffer(root->fs_info->csum_root->node);
2538         free_extent_buffer(root->fs_info->csum_root->commit_root);
2539
2540         btrfs_free_block_groups(root->fs_info);
2541
2542         del_fs_roots(fs_info);
2543
2544         iput(fs_info->btree_inode);
2545
2546         btrfs_stop_workers(&fs_info->generic_worker);
2547         btrfs_stop_workers(&fs_info->fixup_workers);
2548         btrfs_stop_workers(&fs_info->delalloc_workers);
2549         btrfs_stop_workers(&fs_info->workers);
2550         btrfs_stop_workers(&fs_info->endio_workers);
2551         btrfs_stop_workers(&fs_info->endio_meta_workers);
2552         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2553         btrfs_stop_workers(&fs_info->endio_write_workers);
2554         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2555         btrfs_stop_workers(&fs_info->submit_workers);
2556
2557         btrfs_close_devices(fs_info->fs_devices);
2558         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2559
2560         bdi_destroy(&fs_info->bdi);
2561         cleanup_srcu_struct(&fs_info->subvol_srcu);
2562
2563         kfree(fs_info->extent_root);
2564         kfree(fs_info->tree_root);
2565         kfree(fs_info->chunk_root);
2566         kfree(fs_info->dev_root);
2567         kfree(fs_info->csum_root);
2568         kfree(fs_info);
2569
2570         return 0;
2571 }
2572
2573 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2574 {
2575         int ret;
2576         struct inode *btree_inode = buf->first_page->mapping->host;
2577
2578         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2579                                      NULL);
2580         if (!ret)
2581                 return ret;
2582
2583         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2584                                     parent_transid);
2585         return !ret;
2586 }
2587
2588 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2589 {
2590         struct inode *btree_inode = buf->first_page->mapping->host;
2591         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2592                                           buf);
2593 }
2594
2595 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2596 {
2597         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2598         u64 transid = btrfs_header_generation(buf);
2599         struct inode *btree_inode = root->fs_info->btree_inode;
2600         int was_dirty;
2601
2602         btrfs_assert_tree_locked(buf);
2603         if (transid != root->fs_info->generation) {
2604                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2605                        "found %llu running %llu\n",
2606                         (unsigned long long)buf->start,
2607                         (unsigned long long)transid,
2608                         (unsigned long long)root->fs_info->generation);
2609                 WARN_ON(1);
2610         }
2611         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2612                                             buf);
2613         if (!was_dirty) {
2614                 spin_lock(&root->fs_info->delalloc_lock);
2615                 root->fs_info->dirty_metadata_bytes += buf->len;
2616                 spin_unlock(&root->fs_info->delalloc_lock);
2617         }
2618 }
2619
2620 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2621 {
2622         /*
2623          * looks as though older kernels can get into trouble with
2624          * this code, they end up stuck in balance_dirty_pages forever
2625          */
2626         u64 num_dirty;
2627         unsigned long thresh = 32 * 1024 * 1024;
2628
2629         if (current->flags & PF_MEMALLOC)
2630                 return;
2631
2632         num_dirty = root->fs_info->dirty_metadata_bytes;
2633
2634         if (num_dirty > thresh) {
2635                 balance_dirty_pages_ratelimited_nr(
2636                                    root->fs_info->btree_inode->i_mapping, 1);
2637         }
2638         return;
2639 }
2640
2641 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2642 {
2643         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2644         int ret;
2645         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2646         if (ret == 0)
2647                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2648         return ret;
2649 }
2650
2651 int btree_lock_page_hook(struct page *page)
2652 {
2653         struct inode *inode = page->mapping->host;
2654         struct btrfs_root *root = BTRFS_I(inode)->root;
2655         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2656         struct extent_buffer *eb;
2657         unsigned long len;
2658         u64 bytenr = page_offset(page);
2659
2660         if (page->private == EXTENT_PAGE_PRIVATE)
2661                 goto out;
2662
2663         len = page->private >> 2;
2664         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2665         if (!eb)
2666                 goto out;
2667
2668         btrfs_tree_lock(eb);
2669         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2670
2671         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2672                 spin_lock(&root->fs_info->delalloc_lock);
2673                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2674                         root->fs_info->dirty_metadata_bytes -= eb->len;
2675                 else
2676                         WARN_ON(1);
2677                 spin_unlock(&root->fs_info->delalloc_lock);
2678         }
2679
2680         btrfs_tree_unlock(eb);
2681         free_extent_buffer(eb);
2682 out:
2683         lock_page(page);
2684         return 0;
2685 }
2686
2687 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2688                               int read_only)
2689 {
2690         if (read_only)
2691                 return;
2692
2693         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2694                 printk(KERN_WARNING "warning: mount fs with errors, "
2695                        "running btrfsck is recommended\n");
2696 }
2697
2698 int btrfs_error_commit_super(struct btrfs_root *root)
2699 {
2700         int ret;
2701
2702         mutex_lock(&root->fs_info->cleaner_mutex);
2703         btrfs_run_delayed_iputs(root);
2704         mutex_unlock(&root->fs_info->cleaner_mutex);
2705
2706         down_write(&root->fs_info->cleanup_work_sem);
2707         up_write(&root->fs_info->cleanup_work_sem);
2708
2709         /* cleanup FS via transaction */
2710         btrfs_cleanup_transaction(root);
2711
2712         ret = write_ctree_super(NULL, root, 0);
2713
2714         return ret;
2715 }
2716
2717 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2718 {
2719         struct btrfs_inode *btrfs_inode;
2720         struct list_head splice;
2721
2722         INIT_LIST_HEAD(&splice);
2723
2724         mutex_lock(&root->fs_info->ordered_operations_mutex);
2725         spin_lock(&root->fs_info->ordered_extent_lock);
2726
2727         list_splice_init(&root->fs_info->ordered_operations, &splice);
2728         while (!list_empty(&splice)) {
2729                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2730                                          ordered_operations);
2731
2732                 list_del_init(&btrfs_inode->ordered_operations);
2733
2734                 btrfs_invalidate_inodes(btrfs_inode->root);
2735         }
2736
2737         spin_unlock(&root->fs_info->ordered_extent_lock);
2738         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2739
2740         return 0;
2741 }
2742
2743 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2744 {
2745         struct list_head splice;
2746         struct btrfs_ordered_extent *ordered;
2747         struct inode *inode;
2748
2749         INIT_LIST_HEAD(&splice);
2750
2751         spin_lock(&root->fs_info->ordered_extent_lock);
2752
2753         list_splice_init(&root->fs_info->ordered_extents, &splice);
2754         while (!list_empty(&splice)) {
2755                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2756                                      root_extent_list);
2757
2758                 list_del_init(&ordered->root_extent_list);
2759                 atomic_inc(&ordered->refs);
2760
2761                 /* the inode may be getting freed (in sys_unlink path). */
2762                 inode = igrab(ordered->inode);
2763
2764                 spin_unlock(&root->fs_info->ordered_extent_lock);
2765                 if (inode)
2766                         iput(inode);
2767
2768                 atomic_set(&ordered->refs, 1);
2769                 btrfs_put_ordered_extent(ordered);
2770
2771                 spin_lock(&root->fs_info->ordered_extent_lock);
2772         }
2773
2774         spin_unlock(&root->fs_info->ordered_extent_lock);
2775
2776         return 0;
2777 }
2778
2779 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2780                                       struct btrfs_root *root)
2781 {
2782         struct rb_node *node;
2783         struct btrfs_delayed_ref_root *delayed_refs;
2784         struct btrfs_delayed_ref_node *ref;
2785         int ret = 0;
2786
2787         delayed_refs = &trans->delayed_refs;
2788
2789         spin_lock(&delayed_refs->lock);
2790         if (delayed_refs->num_entries == 0) {
2791                 printk(KERN_INFO "delayed_refs has NO entry\n");
2792                 return ret;
2793         }
2794
2795         node = rb_first(&delayed_refs->root);
2796         while (node) {
2797                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2798                 node = rb_next(node);
2799
2800                 ref->in_tree = 0;
2801                 rb_erase(&ref->rb_node, &delayed_refs->root);
2802                 delayed_refs->num_entries--;
2803
2804                 atomic_set(&ref->refs, 1);
2805                 if (btrfs_delayed_ref_is_head(ref)) {
2806                         struct btrfs_delayed_ref_head *head;
2807
2808                         head = btrfs_delayed_node_to_head(ref);
2809                         mutex_lock(&head->mutex);
2810                         kfree(head->extent_op);
2811                         delayed_refs->num_heads--;
2812                         if (list_empty(&head->cluster))
2813                                 delayed_refs->num_heads_ready--;
2814                         list_del_init(&head->cluster);
2815                         mutex_unlock(&head->mutex);
2816                 }
2817
2818                 spin_unlock(&delayed_refs->lock);
2819                 btrfs_put_delayed_ref(ref);
2820
2821                 cond_resched();
2822                 spin_lock(&delayed_refs->lock);
2823         }
2824
2825         spin_unlock(&delayed_refs->lock);
2826
2827         return ret;
2828 }
2829
2830 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2831 {
2832         struct btrfs_pending_snapshot *snapshot;
2833         struct list_head splice;
2834
2835         INIT_LIST_HEAD(&splice);
2836
2837         list_splice_init(&t->pending_snapshots, &splice);
2838
2839         while (!list_empty(&splice)) {
2840                 snapshot = list_entry(splice.next,
2841                                       struct btrfs_pending_snapshot,
2842                                       list);
2843
2844                 list_del_init(&snapshot->list);
2845
2846                 kfree(snapshot);
2847         }
2848
2849         return 0;
2850 }
2851
2852 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2853 {
2854         struct btrfs_inode *btrfs_inode;
2855         struct list_head splice;
2856
2857         INIT_LIST_HEAD(&splice);
2858
2859         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2860
2861         spin_lock(&root->fs_info->delalloc_lock);
2862
2863         while (!list_empty(&splice)) {
2864                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2865                                     delalloc_inodes);
2866
2867                 list_del_init(&btrfs_inode->delalloc_inodes);
2868
2869                 btrfs_invalidate_inodes(btrfs_inode->root);
2870         }
2871
2872         spin_unlock(&root->fs_info->delalloc_lock);
2873
2874         return 0;
2875 }
2876
2877 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2878                                         struct extent_io_tree *dirty_pages,
2879                                         int mark)
2880 {
2881         int ret;
2882         struct page *page;
2883         struct inode *btree_inode = root->fs_info->btree_inode;
2884         struct extent_buffer *eb;
2885         u64 start = 0;
2886         u64 end;
2887         u64 offset;
2888         unsigned long index;
2889
2890         while (1) {
2891                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2892                                             mark);
2893                 if (ret)
2894                         break;
2895
2896                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2897                 while (start <= end) {
2898                         index = start >> PAGE_CACHE_SHIFT;
2899                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2900                         page = find_get_page(btree_inode->i_mapping, index);
2901                         if (!page)
2902                                 continue;
2903                         offset = page_offset(page);
2904
2905                         spin_lock(&dirty_pages->buffer_lock);
2906                         eb = radix_tree_lookup(
2907                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2908                                                offset >> PAGE_CACHE_SHIFT);
2909                         spin_unlock(&dirty_pages->buffer_lock);
2910                         if (eb) {
2911                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2912                                                          &eb->bflags);
2913                                 atomic_set(&eb->refs, 1);
2914                         }
2915                         if (PageWriteback(page))
2916                                 end_page_writeback(page);
2917
2918                         lock_page(page);
2919                         if (PageDirty(page)) {
2920                                 clear_page_dirty_for_io(page);
2921                                 spin_lock_irq(&page->mapping->tree_lock);
2922                                 radix_tree_tag_clear(&page->mapping->page_tree,
2923                                                         page_index(page),
2924                                                         PAGECACHE_TAG_DIRTY);
2925                                 spin_unlock_irq(&page->mapping->tree_lock);
2926                         }
2927
2928                         page->mapping->a_ops->invalidatepage(page, 0);
2929                         unlock_page(page);
2930                 }
2931         }
2932
2933         return ret;
2934 }
2935
2936 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
2937                                        struct extent_io_tree *pinned_extents)
2938 {
2939         struct extent_io_tree *unpin;
2940         u64 start;
2941         u64 end;
2942         int ret;
2943
2944         unpin = pinned_extents;
2945         while (1) {
2946                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2947                                             EXTENT_DIRTY);
2948                 if (ret)
2949                         break;
2950
2951                 /* opt_discard */
2952                 ret = btrfs_error_discard_extent(root, start, end + 1 - start);
2953
2954                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2955                 btrfs_error_unpin_extent_range(root, start, end);
2956                 cond_resched();
2957         }
2958
2959         return 0;
2960 }
2961
2962 static int btrfs_cleanup_transaction(struct btrfs_root *root)
2963 {
2964         struct btrfs_transaction *t;
2965         LIST_HEAD(list);
2966
2967         WARN_ON(1);
2968
2969         mutex_lock(&root->fs_info->trans_mutex);
2970         mutex_lock(&root->fs_info->transaction_kthread_mutex);
2971
2972         list_splice_init(&root->fs_info->trans_list, &list);
2973         while (!list_empty(&list)) {
2974                 t = list_entry(list.next, struct btrfs_transaction, list);
2975                 if (!t)
2976                         break;
2977
2978                 btrfs_destroy_ordered_operations(root);
2979
2980                 btrfs_destroy_ordered_extents(root);
2981
2982                 btrfs_destroy_delayed_refs(t, root);
2983
2984                 btrfs_block_rsv_release(root,
2985                                         &root->fs_info->trans_block_rsv,
2986                                         t->dirty_pages.dirty_bytes);
2987
2988                 /* FIXME: cleanup wait for commit */
2989                 t->in_commit = 1;
2990                 t->blocked = 1;
2991                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
2992                         wake_up(&root->fs_info->transaction_blocked_wait);
2993
2994                 t->blocked = 0;
2995                 if (waitqueue_active(&root->fs_info->transaction_wait))
2996                         wake_up(&root->fs_info->transaction_wait);
2997                 mutex_unlock(&root->fs_info->trans_mutex);
2998
2999                 mutex_lock(&root->fs_info->trans_mutex);
3000                 t->commit_done = 1;
3001                 if (waitqueue_active(&t->commit_wait))
3002                         wake_up(&t->commit_wait);
3003                 mutex_unlock(&root->fs_info->trans_mutex);
3004
3005                 mutex_lock(&root->fs_info->trans_mutex);
3006
3007                 btrfs_destroy_pending_snapshots(t);
3008
3009                 btrfs_destroy_delalloc_inodes(root);
3010
3011                 spin_lock(&root->fs_info->new_trans_lock);
3012                 root->fs_info->running_transaction = NULL;
3013                 spin_unlock(&root->fs_info->new_trans_lock);
3014
3015                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3016                                              EXTENT_DIRTY);
3017
3018                 btrfs_destroy_pinned_extent(root,
3019                                             root->fs_info->pinned_extents);
3020
3021                 t->use_count = 0;
3022                 list_del_init(&t->list);
3023                 memset(t, 0, sizeof(*t));
3024                 kmem_cache_free(btrfs_transaction_cachep, t);
3025         }
3026
3027         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3028         mutex_unlock(&root->fs_info->trans_mutex);
3029
3030         return 0;
3031 }
3032
3033 static struct extent_io_ops btree_extent_io_ops = {
3034         .write_cache_pages_lock_hook = btree_lock_page_hook,
3035         .readpage_end_io_hook = btree_readpage_end_io_hook,
3036         .submit_bio_hook = btree_submit_bio_hook,
3037         /* note we're sharing with inode.c for the merge bio hook */
3038         .merge_bio_hook = btrfs_merge_bio_hook,
3039 };