Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs...
[~shefty/rdma-dev.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 "compat.h"
32 #include "ctree.h"
33 #include "disk-io.h"
34 #include "transaction.h"
35 #include "btrfs_inode.h"
36 #include "volumes.h"
37 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "locking.h"
40 #include "tree-log.h"
41 #include "free-space-cache.h"
42
43 static struct extent_io_ops btree_extent_io_ops;
44 static void end_workqueue_fn(struct btrfs_work *work);
45 static void free_fs_root(struct btrfs_root *root);
46
47 /*
48  * end_io_wq structs are used to do processing in task context when an IO is
49  * complete.  This is used during reads to verify checksums, and it is used
50  * by writes to insert metadata for new file extents after IO is complete.
51  */
52 struct end_io_wq {
53         struct bio *bio;
54         bio_end_io_t *end_io;
55         void *private;
56         struct btrfs_fs_info *info;
57         int error;
58         int metadata;
59         struct list_head list;
60         struct btrfs_work work;
61 };
62
63 /*
64  * async submit bios are used to offload expensive checksumming
65  * onto the worker threads.  They checksum file and metadata bios
66  * just before they are sent down the IO stack.
67  */
68 struct async_submit_bio {
69         struct inode *inode;
70         struct bio *bio;
71         struct list_head list;
72         extent_submit_bio_hook_t *submit_bio_start;
73         extent_submit_bio_hook_t *submit_bio_done;
74         int rw;
75         int mirror_num;
76         unsigned long bio_flags;
77         /*
78          * bio_offset is optional, can be used if the pages in the bio
79          * can't tell us where in the file the bio should go
80          */
81         u64 bio_offset;
82         struct btrfs_work work;
83 };
84
85 /* These are used to set the lockdep class on the extent buffer locks.
86  * The class is set by the readpage_end_io_hook after the buffer has
87  * passed csum validation but before the pages are unlocked.
88  *
89  * The lockdep class is also set by btrfs_init_new_buffer on freshly
90  * allocated blocks.
91  *
92  * The class is based on the level in the tree block, which allows lockdep
93  * to know that lower nodes nest inside the locks of higher nodes.
94  *
95  * We also add a check to make sure the highest level of the tree is
96  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
97  * code needs update as well.
98  */
99 #ifdef CONFIG_DEBUG_LOCK_ALLOC
100 # if BTRFS_MAX_LEVEL != 8
101 #  error
102 # endif
103 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
104 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
105         /* leaf */
106         "btrfs-extent-00",
107         "btrfs-extent-01",
108         "btrfs-extent-02",
109         "btrfs-extent-03",
110         "btrfs-extent-04",
111         "btrfs-extent-05",
112         "btrfs-extent-06",
113         "btrfs-extent-07",
114         /* highest possible level */
115         "btrfs-extent-08",
116 };
117 #endif
118
119 /*
120  * extents on the btree inode are pretty simple, there's one extent
121  * that covers the entire device
122  */
123 static struct extent_map *btree_get_extent(struct inode *inode,
124                 struct page *page, size_t page_offset, u64 start, u64 len,
125                 int create)
126 {
127         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
128         struct extent_map *em;
129         int ret;
130
131         read_lock(&em_tree->lock);
132         em = lookup_extent_mapping(em_tree, start, len);
133         if (em) {
134                 em->bdev =
135                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
136                 read_unlock(&em_tree->lock);
137                 goto out;
138         }
139         read_unlock(&em_tree->lock);
140
141         em = alloc_extent_map(GFP_NOFS);
142         if (!em) {
143                 em = ERR_PTR(-ENOMEM);
144                 goto out;
145         }
146         em->start = 0;
147         em->len = (u64)-1;
148         em->block_len = (u64)-1;
149         em->block_start = 0;
150         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
151
152         write_lock(&em_tree->lock);
153         ret = add_extent_mapping(em_tree, em);
154         if (ret == -EEXIST) {
155                 u64 failed_start = em->start;
156                 u64 failed_len = em->len;
157
158                 free_extent_map(em);
159                 em = lookup_extent_mapping(em_tree, start, len);
160                 if (em) {
161                         ret = 0;
162                 } else {
163                         em = lookup_extent_mapping(em_tree, failed_start,
164                                                    failed_len);
165                         ret = -EIO;
166                 }
167         } else if (ret) {
168                 free_extent_map(em);
169                 em = NULL;
170         }
171         write_unlock(&em_tree->lock);
172
173         if (ret)
174                 em = ERR_PTR(ret);
175 out:
176         return em;
177 }
178
179 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
180 {
181         return crc32c(seed, data, len);
182 }
183
184 void btrfs_csum_final(u32 crc, char *result)
185 {
186         *(__le32 *)result = ~cpu_to_le32(crc);
187 }
188
189 /*
190  * compute the csum for a btree block, and either verify it or write it
191  * into the csum field of the block.
192  */
193 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
194                            int verify)
195 {
196         u16 csum_size =
197                 btrfs_super_csum_size(&root->fs_info->super_copy);
198         char *result = NULL;
199         unsigned long len;
200         unsigned long cur_len;
201         unsigned long offset = BTRFS_CSUM_SIZE;
202         char *map_token = NULL;
203         char *kaddr;
204         unsigned long map_start;
205         unsigned long map_len;
206         int err;
207         u32 crc = ~(u32)0;
208         unsigned long inline_result;
209
210         len = buf->len - offset;
211         while (len > 0) {
212                 err = map_private_extent_buffer(buf, offset, 32,
213                                         &map_token, &kaddr,
214                                         &map_start, &map_len, KM_USER0);
215                 if (err)
216                         return 1;
217                 cur_len = min(len, map_len - (offset - map_start));
218                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
219                                       crc, cur_len);
220                 len -= cur_len;
221                 offset += cur_len;
222                 unmap_extent_buffer(buf, map_token, KM_USER0);
223         }
224         if (csum_size > sizeof(inline_result)) {
225                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
226                 if (!result)
227                         return 1;
228         } else {
229                 result = (char *)&inline_result;
230         }
231
232         btrfs_csum_final(crc, result);
233
234         if (verify) {
235                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
236                         u32 val;
237                         u32 found = 0;
238                         memcpy(&found, result, csum_size);
239
240                         read_extent_buffer(buf, &val, 0, csum_size);
241                         if (printk_ratelimit()) {
242                                 printk(KERN_INFO "btrfs: %s checksum verify "
243                                        "failed on %llu wanted %X found %X "
244                                        "level %d\n",
245                                        root->fs_info->sb->s_id,
246                                        (unsigned long long)buf->start, val, found,
247                                        btrfs_header_level(buf));
248                         }
249                         if (result != (char *)&inline_result)
250                                 kfree(result);
251                         return 1;
252                 }
253         } else {
254                 write_extent_buffer(buf, result, 0, csum_size);
255         }
256         if (result != (char *)&inline_result)
257                 kfree(result);
258         return 0;
259 }
260
261 /*
262  * we can't consider a given block up to date unless the transid of the
263  * block matches the transid in the parent node's pointer.  This is how we
264  * detect blocks that either didn't get written at all or got written
265  * in the wrong place.
266  */
267 static int verify_parent_transid(struct extent_io_tree *io_tree,
268                                  struct extent_buffer *eb, u64 parent_transid)
269 {
270         struct extent_state *cached_state = NULL;
271         int ret;
272
273         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
274                 return 0;
275
276         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
277                          0, &cached_state, GFP_NOFS);
278         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
279             btrfs_header_generation(eb) == parent_transid) {
280                 ret = 0;
281                 goto out;
282         }
283         if (printk_ratelimit()) {
284                 printk("parent transid verify failed on %llu wanted %llu "
285                        "found %llu\n",
286                        (unsigned long long)eb->start,
287                        (unsigned long long)parent_transid,
288                        (unsigned long long)btrfs_header_generation(eb));
289         }
290         ret = 1;
291         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
292 out:
293         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
294                              &cached_state, GFP_NOFS);
295         return ret;
296 }
297
298 /*
299  * helper to read a given tree block, doing retries as required when
300  * the checksums don't match and we have alternate mirrors to try.
301  */
302 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
303                                           struct extent_buffer *eb,
304                                           u64 start, u64 parent_transid)
305 {
306         struct extent_io_tree *io_tree;
307         int ret;
308         int num_copies = 0;
309         int mirror_num = 0;
310
311         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
312         while (1) {
313                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
314                                                btree_get_extent, mirror_num);
315                 if (!ret &&
316                     !verify_parent_transid(io_tree, eb, parent_transid))
317                         return ret;
318
319                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
320                                               eb->start, eb->len);
321                 if (num_copies == 1)
322                         return ret;
323
324                 mirror_num++;
325                 if (mirror_num > num_copies)
326                         return ret;
327         }
328         return -EIO;
329 }
330
331 /*
332  * checksum a dirty tree block before IO.  This has extra checks to make sure
333  * we only fill in the checksum field in the first page of a multi-page block
334  */
335
336 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
337 {
338         struct extent_io_tree *tree;
339         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
340         u64 found_start;
341         unsigned long len;
342         struct extent_buffer *eb;
343         int ret;
344
345         tree = &BTRFS_I(page->mapping->host)->io_tree;
346
347         if (page->private == EXTENT_PAGE_PRIVATE)
348                 goto out;
349         if (!page->private)
350                 goto out;
351         len = page->private >> 2;
352         WARN_ON(len == 0);
353
354         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
355         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
356                                              btrfs_header_generation(eb));
357         BUG_ON(ret);
358         found_start = btrfs_header_bytenr(eb);
359         if (found_start != start) {
360                 WARN_ON(1);
361                 goto err;
362         }
363         if (eb->first_page != page) {
364                 WARN_ON(1);
365                 goto err;
366         }
367         if (!PageUptodate(page)) {
368                 WARN_ON(1);
369                 goto err;
370         }
371         csum_tree_block(root, eb, 0);
372 err:
373         free_extent_buffer(eb);
374 out:
375         return 0;
376 }
377
378 static int check_tree_block_fsid(struct btrfs_root *root,
379                                  struct extent_buffer *eb)
380 {
381         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
382         u8 fsid[BTRFS_UUID_SIZE];
383         int ret = 1;
384
385         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
386                            BTRFS_FSID_SIZE);
387         while (fs_devices) {
388                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
389                         ret = 0;
390                         break;
391                 }
392                 fs_devices = fs_devices->seed;
393         }
394         return ret;
395 }
396
397 #ifdef CONFIG_DEBUG_LOCK_ALLOC
398 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
399 {
400         lockdep_set_class_and_name(&eb->lock,
401                            &btrfs_eb_class[level],
402                            btrfs_eb_name[level]);
403 }
404 #endif
405
406 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
407                                struct extent_state *state)
408 {
409         struct extent_io_tree *tree;
410         u64 found_start;
411         int found_level;
412         unsigned long len;
413         struct extent_buffer *eb;
414         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
415         int ret = 0;
416
417         tree = &BTRFS_I(page->mapping->host)->io_tree;
418         if (page->private == EXTENT_PAGE_PRIVATE)
419                 goto out;
420         if (!page->private)
421                 goto out;
422
423         len = page->private >> 2;
424         WARN_ON(len == 0);
425
426         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
427
428         found_start = btrfs_header_bytenr(eb);
429         if (found_start != start) {
430                 if (printk_ratelimit()) {
431                         printk(KERN_INFO "btrfs bad tree block start "
432                                "%llu %llu\n",
433                                (unsigned long long)found_start,
434                                (unsigned long long)eb->start);
435                 }
436                 ret = -EIO;
437                 goto err;
438         }
439         if (eb->first_page != page) {
440                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
441                        eb->first_page->index, page->index);
442                 WARN_ON(1);
443                 ret = -EIO;
444                 goto err;
445         }
446         if (check_tree_block_fsid(root, eb)) {
447                 if (printk_ratelimit()) {
448                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
449                                (unsigned long long)eb->start);
450                 }
451                 ret = -EIO;
452                 goto err;
453         }
454         found_level = btrfs_header_level(eb);
455
456         btrfs_set_buffer_lockdep_class(eb, found_level);
457
458         ret = csum_tree_block(root, eb, 1);
459         if (ret)
460                 ret = -EIO;
461
462         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
463         end = eb->start + end - 1;
464 err:
465         free_extent_buffer(eb);
466 out:
467         return ret;
468 }
469
470 static void end_workqueue_bio(struct bio *bio, int err)
471 {
472         struct end_io_wq *end_io_wq = bio->bi_private;
473         struct btrfs_fs_info *fs_info;
474
475         fs_info = end_io_wq->info;
476         end_io_wq->error = err;
477         end_io_wq->work.func = end_workqueue_fn;
478         end_io_wq->work.flags = 0;
479
480         if (bio->bi_rw & REQ_WRITE) {
481                 if (end_io_wq->metadata == 1)
482                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
483                                            &end_io_wq->work);
484                 else if (end_io_wq->metadata == 2)
485                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
486                                            &end_io_wq->work);
487                 else
488                         btrfs_queue_worker(&fs_info->endio_write_workers,
489                                            &end_io_wq->work);
490         } else {
491                 if (end_io_wq->metadata)
492                         btrfs_queue_worker(&fs_info->endio_meta_workers,
493                                            &end_io_wq->work);
494                 else
495                         btrfs_queue_worker(&fs_info->endio_workers,
496                                            &end_io_wq->work);
497         }
498 }
499
500 /*
501  * For the metadata arg you want
502  *
503  * 0 - if data
504  * 1 - if normal metadta
505  * 2 - if writing to the free space cache area
506  */
507 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
508                         int metadata)
509 {
510         struct end_io_wq *end_io_wq;
511         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
512         if (!end_io_wq)
513                 return -ENOMEM;
514
515         end_io_wq->private = bio->bi_private;
516         end_io_wq->end_io = bio->bi_end_io;
517         end_io_wq->info = info;
518         end_io_wq->error = 0;
519         end_io_wq->bio = bio;
520         end_io_wq->metadata = metadata;
521
522         bio->bi_private = end_io_wq;
523         bio->bi_end_io = end_workqueue_bio;
524         return 0;
525 }
526
527 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
528 {
529         unsigned long limit = min_t(unsigned long,
530                                     info->workers.max_workers,
531                                     info->fs_devices->open_devices);
532         return 256 * limit;
533 }
534
535 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
536 {
537         return atomic_read(&info->nr_async_bios) >
538                 btrfs_async_submit_limit(info);
539 }
540
541 static void run_one_async_start(struct btrfs_work *work)
542 {
543         struct async_submit_bio *async;
544
545         async = container_of(work, struct  async_submit_bio, work);
546         async->submit_bio_start(async->inode, async->rw, async->bio,
547                                async->mirror_num, async->bio_flags,
548                                async->bio_offset);
549 }
550
551 static void run_one_async_done(struct btrfs_work *work)
552 {
553         struct btrfs_fs_info *fs_info;
554         struct async_submit_bio *async;
555         int limit;
556
557         async = container_of(work, struct  async_submit_bio, work);
558         fs_info = BTRFS_I(async->inode)->root->fs_info;
559
560         limit = btrfs_async_submit_limit(fs_info);
561         limit = limit * 2 / 3;
562
563         atomic_dec(&fs_info->nr_async_submits);
564
565         if (atomic_read(&fs_info->nr_async_submits) < limit &&
566             waitqueue_active(&fs_info->async_submit_wait))
567                 wake_up(&fs_info->async_submit_wait);
568
569         async->submit_bio_done(async->inode, async->rw, async->bio,
570                                async->mirror_num, async->bio_flags,
571                                async->bio_offset);
572 }
573
574 static void run_one_async_free(struct btrfs_work *work)
575 {
576         struct async_submit_bio *async;
577
578         async = container_of(work, struct  async_submit_bio, work);
579         kfree(async);
580 }
581
582 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
583                         int rw, struct bio *bio, int mirror_num,
584                         unsigned long bio_flags,
585                         u64 bio_offset,
586                         extent_submit_bio_hook_t *submit_bio_start,
587                         extent_submit_bio_hook_t *submit_bio_done)
588 {
589         struct async_submit_bio *async;
590
591         async = kmalloc(sizeof(*async), GFP_NOFS);
592         if (!async)
593                 return -ENOMEM;
594
595         async->inode = inode;
596         async->rw = rw;
597         async->bio = bio;
598         async->mirror_num = mirror_num;
599         async->submit_bio_start = submit_bio_start;
600         async->submit_bio_done = submit_bio_done;
601
602         async->work.func = run_one_async_start;
603         async->work.ordered_func = run_one_async_done;
604         async->work.ordered_free = run_one_async_free;
605
606         async->work.flags = 0;
607         async->bio_flags = bio_flags;
608         async->bio_offset = bio_offset;
609
610         atomic_inc(&fs_info->nr_async_submits);
611
612         if (rw & REQ_SYNC)
613                 btrfs_set_work_high_prio(&async->work);
614
615         btrfs_queue_worker(&fs_info->workers, &async->work);
616
617         while (atomic_read(&fs_info->async_submit_draining) &&
618               atomic_read(&fs_info->nr_async_submits)) {
619                 wait_event(fs_info->async_submit_wait,
620                            (atomic_read(&fs_info->nr_async_submits) == 0));
621         }
622
623         return 0;
624 }
625
626 static int btree_csum_one_bio(struct bio *bio)
627 {
628         struct bio_vec *bvec = bio->bi_io_vec;
629         int bio_index = 0;
630         struct btrfs_root *root;
631
632         WARN_ON(bio->bi_vcnt <= 0);
633         while (bio_index < bio->bi_vcnt) {
634                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
635                 csum_dirty_buffer(root, bvec->bv_page);
636                 bio_index++;
637                 bvec++;
638         }
639         return 0;
640 }
641
642 static int __btree_submit_bio_start(struct inode *inode, int rw,
643                                     struct bio *bio, int mirror_num,
644                                     unsigned long bio_flags,
645                                     u64 bio_offset)
646 {
647         /*
648          * when we're called for a write, we're already in the async
649          * submission context.  Just jump into btrfs_map_bio
650          */
651         btree_csum_one_bio(bio);
652         return 0;
653 }
654
655 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
656                                  int mirror_num, unsigned long bio_flags,
657                                  u64 bio_offset)
658 {
659         /*
660          * when we're called for a write, we're already in the async
661          * submission context.  Just jump into btrfs_map_bio
662          */
663         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
664 }
665
666 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
667                                  int mirror_num, unsigned long bio_flags,
668                                  u64 bio_offset)
669 {
670         int ret;
671
672         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
673                                           bio, 1);
674         BUG_ON(ret);
675
676         if (!(rw & REQ_WRITE)) {
677                 /*
678                  * called for a read, do the setup so that checksum validation
679                  * can happen in the async kernel threads
680                  */
681                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
682                                      mirror_num, 0);
683         }
684
685         /*
686          * kthread helpers are used to submit writes so that checksumming
687          * can happen in parallel across all CPUs
688          */
689         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
690                                    inode, rw, bio, mirror_num, 0,
691                                    bio_offset,
692                                    __btree_submit_bio_start,
693                                    __btree_submit_bio_done);
694 }
695
696 static int btree_writepage(struct page *page, struct writeback_control *wbc)
697 {
698         struct extent_io_tree *tree;
699         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
700         struct extent_buffer *eb;
701         int was_dirty;
702
703         tree = &BTRFS_I(page->mapping->host)->io_tree;
704         if (!(current->flags & PF_MEMALLOC)) {
705                 return extent_write_full_page(tree, page,
706                                               btree_get_extent, wbc);
707         }
708
709         redirty_page_for_writepage(wbc, page);
710         eb = btrfs_find_tree_block(root, page_offset(page),
711                                       PAGE_CACHE_SIZE);
712         WARN_ON(!eb);
713
714         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
715         if (!was_dirty) {
716                 spin_lock(&root->fs_info->delalloc_lock);
717                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
718                 spin_unlock(&root->fs_info->delalloc_lock);
719         }
720         free_extent_buffer(eb);
721
722         unlock_page(page);
723         return 0;
724 }
725
726 static int btree_writepages(struct address_space *mapping,
727                             struct writeback_control *wbc)
728 {
729         struct extent_io_tree *tree;
730         tree = &BTRFS_I(mapping->host)->io_tree;
731         if (wbc->sync_mode == WB_SYNC_NONE) {
732                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
733                 u64 num_dirty;
734                 unsigned long thresh = 32 * 1024 * 1024;
735
736                 if (wbc->for_kupdate)
737                         return 0;
738
739                 /* this is a bit racy, but that's ok */
740                 num_dirty = root->fs_info->dirty_metadata_bytes;
741                 if (num_dirty < thresh)
742                         return 0;
743         }
744         return extent_writepages(tree, mapping, btree_get_extent, wbc);
745 }
746
747 static int btree_readpage(struct file *file, struct page *page)
748 {
749         struct extent_io_tree *tree;
750         tree = &BTRFS_I(page->mapping->host)->io_tree;
751         return extent_read_full_page(tree, page, btree_get_extent);
752 }
753
754 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
755 {
756         struct extent_io_tree *tree;
757         struct extent_map_tree *map;
758         int ret;
759
760         if (PageWriteback(page) || PageDirty(page))
761                 return 0;
762
763         tree = &BTRFS_I(page->mapping->host)->io_tree;
764         map = &BTRFS_I(page->mapping->host)->extent_tree;
765
766         ret = try_release_extent_state(map, tree, page, gfp_flags);
767         if (!ret)
768                 return 0;
769
770         ret = try_release_extent_buffer(tree, page);
771         if (ret == 1) {
772                 ClearPagePrivate(page);
773                 set_page_private(page, 0);
774                 page_cache_release(page);
775         }
776
777         return ret;
778 }
779
780 static void btree_invalidatepage(struct page *page, unsigned long offset)
781 {
782         struct extent_io_tree *tree;
783         tree = &BTRFS_I(page->mapping->host)->io_tree;
784         extent_invalidatepage(tree, page, offset);
785         btree_releasepage(page, GFP_NOFS);
786         if (PagePrivate(page)) {
787                 printk(KERN_WARNING "btrfs warning page private not zero "
788                        "on page %llu\n", (unsigned long long)page_offset(page));
789                 ClearPagePrivate(page);
790                 set_page_private(page, 0);
791                 page_cache_release(page);
792         }
793 }
794
795 static const struct address_space_operations btree_aops = {
796         .readpage       = btree_readpage,
797         .writepage      = btree_writepage,
798         .writepages     = btree_writepages,
799         .releasepage    = btree_releasepage,
800         .invalidatepage = btree_invalidatepage,
801         .sync_page      = block_sync_page,
802 };
803
804 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
805                          u64 parent_transid)
806 {
807         struct extent_buffer *buf = NULL;
808         struct inode *btree_inode = root->fs_info->btree_inode;
809         int ret = 0;
810
811         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
812         if (!buf)
813                 return 0;
814         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
815                                  buf, 0, 0, btree_get_extent, 0);
816         free_extent_buffer(buf);
817         return ret;
818 }
819
820 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
821                                             u64 bytenr, u32 blocksize)
822 {
823         struct inode *btree_inode = root->fs_info->btree_inode;
824         struct extent_buffer *eb;
825         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
826                                 bytenr, blocksize, GFP_NOFS);
827         return eb;
828 }
829
830 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
831                                                  u64 bytenr, u32 blocksize)
832 {
833         struct inode *btree_inode = root->fs_info->btree_inode;
834         struct extent_buffer *eb;
835
836         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
837                                  bytenr, blocksize, NULL, GFP_NOFS);
838         return eb;
839 }
840
841
842 int btrfs_write_tree_block(struct extent_buffer *buf)
843 {
844         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
845                                         buf->start + buf->len - 1);
846 }
847
848 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
849 {
850         return filemap_fdatawait_range(buf->first_page->mapping,
851                                        buf->start, buf->start + buf->len - 1);
852 }
853
854 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
855                                       u32 blocksize, u64 parent_transid)
856 {
857         struct extent_buffer *buf = NULL;
858         int ret;
859
860         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
861         if (!buf)
862                 return NULL;
863
864         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
865
866         if (ret == 0)
867                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
868         return buf;
869
870 }
871
872 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
873                      struct extent_buffer *buf)
874 {
875         struct inode *btree_inode = root->fs_info->btree_inode;
876         if (btrfs_header_generation(buf) ==
877             root->fs_info->running_transaction->transid) {
878                 btrfs_assert_tree_locked(buf);
879
880                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
881                         spin_lock(&root->fs_info->delalloc_lock);
882                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
883                                 root->fs_info->dirty_metadata_bytes -= buf->len;
884                         else
885                                 WARN_ON(1);
886                         spin_unlock(&root->fs_info->delalloc_lock);
887                 }
888
889                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
890                 btrfs_set_lock_blocking(buf);
891                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
892                                           buf);
893         }
894         return 0;
895 }
896
897 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
898                         u32 stripesize, struct btrfs_root *root,
899                         struct btrfs_fs_info *fs_info,
900                         u64 objectid)
901 {
902         root->node = NULL;
903         root->commit_root = NULL;
904         root->sectorsize = sectorsize;
905         root->nodesize = nodesize;
906         root->leafsize = leafsize;
907         root->stripesize = stripesize;
908         root->ref_cows = 0;
909         root->track_dirty = 0;
910         root->in_radix = 0;
911         root->orphan_item_inserted = 0;
912         root->orphan_cleanup_state = 0;
913
914         root->fs_info = fs_info;
915         root->objectid = objectid;
916         root->last_trans = 0;
917         root->highest_objectid = 0;
918         root->name = NULL;
919         root->in_sysfs = 0;
920         root->inode_tree = RB_ROOT;
921         root->block_rsv = NULL;
922         root->orphan_block_rsv = NULL;
923
924         INIT_LIST_HEAD(&root->dirty_list);
925         INIT_LIST_HEAD(&root->orphan_list);
926         INIT_LIST_HEAD(&root->root_list);
927         spin_lock_init(&root->node_lock);
928         spin_lock_init(&root->orphan_lock);
929         spin_lock_init(&root->inode_lock);
930         spin_lock_init(&root->accounting_lock);
931         mutex_init(&root->objectid_mutex);
932         mutex_init(&root->log_mutex);
933         init_waitqueue_head(&root->log_writer_wait);
934         init_waitqueue_head(&root->log_commit_wait[0]);
935         init_waitqueue_head(&root->log_commit_wait[1]);
936         atomic_set(&root->log_commit[0], 0);
937         atomic_set(&root->log_commit[1], 0);
938         atomic_set(&root->log_writers, 0);
939         root->log_batch = 0;
940         root->log_transid = 0;
941         root->last_log_commit = 0;
942         extent_io_tree_init(&root->dirty_log_pages,
943                              fs_info->btree_inode->i_mapping, GFP_NOFS);
944
945         memset(&root->root_key, 0, sizeof(root->root_key));
946         memset(&root->root_item, 0, sizeof(root->root_item));
947         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
948         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
949         root->defrag_trans_start = fs_info->generation;
950         init_completion(&root->kobj_unregister);
951         root->defrag_running = 0;
952         root->root_key.objectid = objectid;
953         root->anon_super.s_root = NULL;
954         root->anon_super.s_dev = 0;
955         INIT_LIST_HEAD(&root->anon_super.s_list);
956         INIT_LIST_HEAD(&root->anon_super.s_instances);
957         init_rwsem(&root->anon_super.s_umount);
958
959         return 0;
960 }
961
962 static int find_and_setup_root(struct btrfs_root *tree_root,
963                                struct btrfs_fs_info *fs_info,
964                                u64 objectid,
965                                struct btrfs_root *root)
966 {
967         int ret;
968         u32 blocksize;
969         u64 generation;
970
971         __setup_root(tree_root->nodesize, tree_root->leafsize,
972                      tree_root->sectorsize, tree_root->stripesize,
973                      root, fs_info, objectid);
974         ret = btrfs_find_last_root(tree_root, objectid,
975                                    &root->root_item, &root->root_key);
976         if (ret > 0)
977                 return -ENOENT;
978         BUG_ON(ret);
979
980         generation = btrfs_root_generation(&root->root_item);
981         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
982         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
983                                      blocksize, generation);
984         BUG_ON(!root->node);
985         root->commit_root = btrfs_root_node(root);
986         return 0;
987 }
988
989 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
990                                          struct btrfs_fs_info *fs_info)
991 {
992         struct btrfs_root *root;
993         struct btrfs_root *tree_root = fs_info->tree_root;
994         struct extent_buffer *leaf;
995
996         root = kzalloc(sizeof(*root), GFP_NOFS);
997         if (!root)
998                 return ERR_PTR(-ENOMEM);
999
1000         __setup_root(tree_root->nodesize, tree_root->leafsize,
1001                      tree_root->sectorsize, tree_root->stripesize,
1002                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1003
1004         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1005         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1006         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1007         /*
1008          * log trees do not get reference counted because they go away
1009          * before a real commit is actually done.  They do store pointers
1010          * to file data extents, and those reference counts still get
1011          * updated (along with back refs to the log tree).
1012          */
1013         root->ref_cows = 0;
1014
1015         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1016                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1017         if (IS_ERR(leaf)) {
1018                 kfree(root);
1019                 return ERR_CAST(leaf);
1020         }
1021
1022         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1023         btrfs_set_header_bytenr(leaf, leaf->start);
1024         btrfs_set_header_generation(leaf, trans->transid);
1025         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1026         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1027         root->node = leaf;
1028
1029         write_extent_buffer(root->node, root->fs_info->fsid,
1030                             (unsigned long)btrfs_header_fsid(root->node),
1031                             BTRFS_FSID_SIZE);
1032         btrfs_mark_buffer_dirty(root->node);
1033         btrfs_tree_unlock(root->node);
1034         return root;
1035 }
1036
1037 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1038                              struct btrfs_fs_info *fs_info)
1039 {
1040         struct btrfs_root *log_root;
1041
1042         log_root = alloc_log_tree(trans, fs_info);
1043         if (IS_ERR(log_root))
1044                 return PTR_ERR(log_root);
1045         WARN_ON(fs_info->log_root_tree);
1046         fs_info->log_root_tree = log_root;
1047         return 0;
1048 }
1049
1050 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1051                        struct btrfs_root *root)
1052 {
1053         struct btrfs_root *log_root;
1054         struct btrfs_inode_item *inode_item;
1055
1056         log_root = alloc_log_tree(trans, root->fs_info);
1057         if (IS_ERR(log_root))
1058                 return PTR_ERR(log_root);
1059
1060         log_root->last_trans = trans->transid;
1061         log_root->root_key.offset = root->root_key.objectid;
1062
1063         inode_item = &log_root->root_item.inode;
1064         inode_item->generation = cpu_to_le64(1);
1065         inode_item->size = cpu_to_le64(3);
1066         inode_item->nlink = cpu_to_le32(1);
1067         inode_item->nbytes = cpu_to_le64(root->leafsize);
1068         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1069
1070         btrfs_set_root_node(&log_root->root_item, log_root->node);
1071
1072         WARN_ON(root->log_root);
1073         root->log_root = log_root;
1074         root->log_transid = 0;
1075         root->last_log_commit = 0;
1076         return 0;
1077 }
1078
1079 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1080                                                struct btrfs_key *location)
1081 {
1082         struct btrfs_root *root;
1083         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1084         struct btrfs_path *path;
1085         struct extent_buffer *l;
1086         u64 generation;
1087         u32 blocksize;
1088         int ret = 0;
1089
1090         root = kzalloc(sizeof(*root), GFP_NOFS);
1091         if (!root)
1092                 return ERR_PTR(-ENOMEM);
1093         if (location->offset == (u64)-1) {
1094                 ret = find_and_setup_root(tree_root, fs_info,
1095                                           location->objectid, root);
1096                 if (ret) {
1097                         kfree(root);
1098                         return ERR_PTR(ret);
1099                 }
1100                 goto out;
1101         }
1102
1103         __setup_root(tree_root->nodesize, tree_root->leafsize,
1104                      tree_root->sectorsize, tree_root->stripesize,
1105                      root, fs_info, location->objectid);
1106
1107         path = btrfs_alloc_path();
1108         BUG_ON(!path);
1109         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1110         if (ret == 0) {
1111                 l = path->nodes[0];
1112                 read_extent_buffer(l, &root->root_item,
1113                                 btrfs_item_ptr_offset(l, path->slots[0]),
1114                                 sizeof(root->root_item));
1115                 memcpy(&root->root_key, location, sizeof(*location));
1116         }
1117         btrfs_free_path(path);
1118         if (ret) {
1119                 if (ret > 0)
1120                         ret = -ENOENT;
1121                 return ERR_PTR(ret);
1122         }
1123
1124         generation = btrfs_root_generation(&root->root_item);
1125         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1126         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1127                                      blocksize, generation);
1128         root->commit_root = btrfs_root_node(root);
1129         BUG_ON(!root->node);
1130 out:
1131         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1132                 root->ref_cows = 1;
1133
1134         return root;
1135 }
1136
1137 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1138                                         u64 root_objectid)
1139 {
1140         struct btrfs_root *root;
1141
1142         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1143                 return fs_info->tree_root;
1144         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1145                 return fs_info->extent_root;
1146
1147         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1148                                  (unsigned long)root_objectid);
1149         return root;
1150 }
1151
1152 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1153                                               struct btrfs_key *location)
1154 {
1155         struct btrfs_root *root;
1156         int ret;
1157
1158         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1159                 return fs_info->tree_root;
1160         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1161                 return fs_info->extent_root;
1162         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1163                 return fs_info->chunk_root;
1164         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1165                 return fs_info->dev_root;
1166         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1167                 return fs_info->csum_root;
1168 again:
1169         spin_lock(&fs_info->fs_roots_radix_lock);
1170         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1171                                  (unsigned long)location->objectid);
1172         spin_unlock(&fs_info->fs_roots_radix_lock);
1173         if (root)
1174                 return root;
1175
1176         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1177         if (IS_ERR(root))
1178                 return root;
1179
1180         set_anon_super(&root->anon_super, NULL);
1181
1182         if (btrfs_root_refs(&root->root_item) == 0) {
1183                 ret = -ENOENT;
1184                 goto fail;
1185         }
1186
1187         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1188         if (ret < 0)
1189                 goto fail;
1190         if (ret == 0)
1191                 root->orphan_item_inserted = 1;
1192
1193         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1194         if (ret)
1195                 goto fail;
1196
1197         spin_lock(&fs_info->fs_roots_radix_lock);
1198         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1199                                 (unsigned long)root->root_key.objectid,
1200                                 root);
1201         if (ret == 0)
1202                 root->in_radix = 1;
1203
1204         spin_unlock(&fs_info->fs_roots_radix_lock);
1205         radix_tree_preload_end();
1206         if (ret) {
1207                 if (ret == -EEXIST) {
1208                         free_fs_root(root);
1209                         goto again;
1210                 }
1211                 goto fail;
1212         }
1213
1214         ret = btrfs_find_dead_roots(fs_info->tree_root,
1215                                     root->root_key.objectid);
1216         WARN_ON(ret);
1217         return root;
1218 fail:
1219         free_fs_root(root);
1220         return ERR_PTR(ret);
1221 }
1222
1223 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1224                                       struct btrfs_key *location,
1225                                       const char *name, int namelen)
1226 {
1227         return btrfs_read_fs_root_no_name(fs_info, location);
1228 #if 0
1229         struct btrfs_root *root;
1230         int ret;
1231
1232         root = btrfs_read_fs_root_no_name(fs_info, location);
1233         if (!root)
1234                 return NULL;
1235
1236         if (root->in_sysfs)
1237                 return root;
1238
1239         ret = btrfs_set_root_name(root, name, namelen);
1240         if (ret) {
1241                 free_extent_buffer(root->node);
1242                 kfree(root);
1243                 return ERR_PTR(ret);
1244         }
1245
1246         ret = btrfs_sysfs_add_root(root);
1247         if (ret) {
1248                 free_extent_buffer(root->node);
1249                 kfree(root->name);
1250                 kfree(root);
1251                 return ERR_PTR(ret);
1252         }
1253         root->in_sysfs = 1;
1254         return root;
1255 #endif
1256 }
1257
1258 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1259 {
1260         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1261         int ret = 0;
1262         struct btrfs_device *device;
1263         struct backing_dev_info *bdi;
1264
1265         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1266                 if (!device->bdev)
1267                         continue;
1268                 bdi = blk_get_backing_dev_info(device->bdev);
1269                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1270                         ret = 1;
1271                         break;
1272                 }
1273         }
1274         return ret;
1275 }
1276
1277 /*
1278  * this unplugs every device on the box, and it is only used when page
1279  * is null
1280  */
1281 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1282 {
1283         struct btrfs_device *device;
1284         struct btrfs_fs_info *info;
1285
1286         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1287         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1288                 if (!device->bdev)
1289                         continue;
1290
1291                 bdi = blk_get_backing_dev_info(device->bdev);
1292                 if (bdi->unplug_io_fn)
1293                         bdi->unplug_io_fn(bdi, page);
1294         }
1295 }
1296
1297 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1298 {
1299         struct inode *inode;
1300         struct extent_map_tree *em_tree;
1301         struct extent_map *em;
1302         struct address_space *mapping;
1303         u64 offset;
1304
1305         /* the generic O_DIRECT read code does this */
1306         if (1 || !page) {
1307                 __unplug_io_fn(bdi, page);
1308                 return;
1309         }
1310
1311         /*
1312          * page->mapping may change at any time.  Get a consistent copy
1313          * and use that for everything below
1314          */
1315         smp_mb();
1316         mapping = page->mapping;
1317         if (!mapping)
1318                 return;
1319
1320         inode = mapping->host;
1321
1322         /*
1323          * don't do the expensive searching for a small number of
1324          * devices
1325          */
1326         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1327                 __unplug_io_fn(bdi, page);
1328                 return;
1329         }
1330
1331         offset = page_offset(page);
1332
1333         em_tree = &BTRFS_I(inode)->extent_tree;
1334         read_lock(&em_tree->lock);
1335         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1336         read_unlock(&em_tree->lock);
1337         if (!em) {
1338                 __unplug_io_fn(bdi, page);
1339                 return;
1340         }
1341
1342         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1343                 free_extent_map(em);
1344                 __unplug_io_fn(bdi, page);
1345                 return;
1346         }
1347         offset = offset - em->start;
1348         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1349                           em->block_start + offset, page);
1350         free_extent_map(em);
1351 }
1352
1353 /*
1354  * If this fails, caller must call bdi_destroy() to get rid of the
1355  * bdi again.
1356  */
1357 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1358 {
1359         int err;
1360
1361         bdi->capabilities = BDI_CAP_MAP_COPY;
1362         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1363         if (err)
1364                 return err;
1365
1366         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1367         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1368         bdi->unplug_io_data     = info;
1369         bdi->congested_fn       = btrfs_congested_fn;
1370         bdi->congested_data     = info;
1371         return 0;
1372 }
1373
1374 static int bio_ready_for_csum(struct bio *bio)
1375 {
1376         u64 length = 0;
1377         u64 buf_len = 0;
1378         u64 start = 0;
1379         struct page *page;
1380         struct extent_io_tree *io_tree = NULL;
1381         struct bio_vec *bvec;
1382         int i;
1383         int ret;
1384
1385         bio_for_each_segment(bvec, bio, i) {
1386                 page = bvec->bv_page;
1387                 if (page->private == EXTENT_PAGE_PRIVATE) {
1388                         length += bvec->bv_len;
1389                         continue;
1390                 }
1391                 if (!page->private) {
1392                         length += bvec->bv_len;
1393                         continue;
1394                 }
1395                 length = bvec->bv_len;
1396                 buf_len = page->private >> 2;
1397                 start = page_offset(page) + bvec->bv_offset;
1398                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1399         }
1400         /* are we fully contained in this bio? */
1401         if (buf_len <= length)
1402                 return 1;
1403
1404         ret = extent_range_uptodate(io_tree, start + length,
1405                                     start + buf_len - 1);
1406         return ret;
1407 }
1408
1409 /*
1410  * called by the kthread helper functions to finally call the bio end_io
1411  * functions.  This is where read checksum verification actually happens
1412  */
1413 static void end_workqueue_fn(struct btrfs_work *work)
1414 {
1415         struct bio *bio;
1416         struct end_io_wq *end_io_wq;
1417         struct btrfs_fs_info *fs_info;
1418         int error;
1419
1420         end_io_wq = container_of(work, struct end_io_wq, work);
1421         bio = end_io_wq->bio;
1422         fs_info = end_io_wq->info;
1423
1424         /* metadata bio reads are special because the whole tree block must
1425          * be checksummed at once.  This makes sure the entire block is in
1426          * ram and up to date before trying to verify things.  For
1427          * blocksize <= pagesize, it is basically a noop
1428          */
1429         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1430             !bio_ready_for_csum(bio)) {
1431                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1432                                    &end_io_wq->work);
1433                 return;
1434         }
1435         error = end_io_wq->error;
1436         bio->bi_private = end_io_wq->private;
1437         bio->bi_end_io = end_io_wq->end_io;
1438         kfree(end_io_wq);
1439         bio_endio(bio, error);
1440 }
1441
1442 static int cleaner_kthread(void *arg)
1443 {
1444         struct btrfs_root *root = arg;
1445
1446         do {
1447                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1448
1449                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1450                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1451                         btrfs_run_delayed_iputs(root);
1452                         btrfs_clean_old_snapshots(root);
1453                         mutex_unlock(&root->fs_info->cleaner_mutex);
1454                 }
1455
1456                 if (freezing(current)) {
1457                         refrigerator();
1458                 } else {
1459                         set_current_state(TASK_INTERRUPTIBLE);
1460                         if (!kthread_should_stop())
1461                                 schedule();
1462                         __set_current_state(TASK_RUNNING);
1463                 }
1464         } while (!kthread_should_stop());
1465         return 0;
1466 }
1467
1468 static int transaction_kthread(void *arg)
1469 {
1470         struct btrfs_root *root = arg;
1471         struct btrfs_trans_handle *trans;
1472         struct btrfs_transaction *cur;
1473         u64 transid;
1474         unsigned long now;
1475         unsigned long delay;
1476         int ret;
1477
1478         do {
1479                 delay = HZ * 30;
1480                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1481                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1482
1483                 spin_lock(&root->fs_info->new_trans_lock);
1484                 cur = root->fs_info->running_transaction;
1485                 if (!cur) {
1486                         spin_unlock(&root->fs_info->new_trans_lock);
1487                         goto sleep;
1488                 }
1489
1490                 now = get_seconds();
1491                 if (!cur->blocked &&
1492                     (now < cur->start_time || now - cur->start_time < 30)) {
1493                         spin_unlock(&root->fs_info->new_trans_lock);
1494                         delay = HZ * 5;
1495                         goto sleep;
1496                 }
1497                 transid = cur->transid;
1498                 spin_unlock(&root->fs_info->new_trans_lock);
1499
1500                 trans = btrfs_join_transaction(root, 1);
1501                 if (transid == trans->transid) {
1502                         ret = btrfs_commit_transaction(trans, root);
1503                         BUG_ON(ret);
1504                 } else {
1505                         btrfs_end_transaction(trans, root);
1506                 }
1507 sleep:
1508                 wake_up_process(root->fs_info->cleaner_kthread);
1509                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1510
1511                 if (freezing(current)) {
1512                         refrigerator();
1513                 } else {
1514                         set_current_state(TASK_INTERRUPTIBLE);
1515                         if (!kthread_should_stop() &&
1516                             !btrfs_transaction_blocked(root->fs_info))
1517                                 schedule_timeout(delay);
1518                         __set_current_state(TASK_RUNNING);
1519                 }
1520         } while (!kthread_should_stop());
1521         return 0;
1522 }
1523
1524 struct btrfs_root *open_ctree(struct super_block *sb,
1525                               struct btrfs_fs_devices *fs_devices,
1526                               char *options)
1527 {
1528         u32 sectorsize;
1529         u32 nodesize;
1530         u32 leafsize;
1531         u32 blocksize;
1532         u32 stripesize;
1533         u64 generation;
1534         u64 features;
1535         struct btrfs_key location;
1536         struct buffer_head *bh;
1537         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1538                                                  GFP_NOFS);
1539         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1540                                                  GFP_NOFS);
1541         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1542                                                GFP_NOFS);
1543         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1544                                                 GFP_NOFS);
1545         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1546                                                 GFP_NOFS);
1547         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1548                                               GFP_NOFS);
1549         struct btrfs_root *log_tree_root;
1550
1551         int ret;
1552         int err = -EINVAL;
1553
1554         struct btrfs_super_block *disk_super;
1555
1556         if (!extent_root || !tree_root || !fs_info ||
1557             !chunk_root || !dev_root || !csum_root) {
1558                 err = -ENOMEM;
1559                 goto fail;
1560         }
1561
1562         ret = init_srcu_struct(&fs_info->subvol_srcu);
1563         if (ret) {
1564                 err = ret;
1565                 goto fail;
1566         }
1567
1568         ret = setup_bdi(fs_info, &fs_info->bdi);
1569         if (ret) {
1570                 err = ret;
1571                 goto fail_srcu;
1572         }
1573
1574         fs_info->btree_inode = new_inode(sb);
1575         if (!fs_info->btree_inode) {
1576                 err = -ENOMEM;
1577                 goto fail_bdi;
1578         }
1579
1580         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1581         INIT_LIST_HEAD(&fs_info->trans_list);
1582         INIT_LIST_HEAD(&fs_info->dead_roots);
1583         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1584         INIT_LIST_HEAD(&fs_info->hashers);
1585         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1586         INIT_LIST_HEAD(&fs_info->ordered_operations);
1587         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1588         spin_lock_init(&fs_info->delalloc_lock);
1589         spin_lock_init(&fs_info->new_trans_lock);
1590         spin_lock_init(&fs_info->ref_cache_lock);
1591         spin_lock_init(&fs_info->fs_roots_radix_lock);
1592         spin_lock_init(&fs_info->delayed_iput_lock);
1593
1594         init_completion(&fs_info->kobj_unregister);
1595         fs_info->tree_root = tree_root;
1596         fs_info->extent_root = extent_root;
1597         fs_info->csum_root = csum_root;
1598         fs_info->chunk_root = chunk_root;
1599         fs_info->dev_root = dev_root;
1600         fs_info->fs_devices = fs_devices;
1601         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1602         INIT_LIST_HEAD(&fs_info->space_info);
1603         btrfs_mapping_init(&fs_info->mapping_tree);
1604         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1605         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1606         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1607         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1608         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1609         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1610         mutex_init(&fs_info->durable_block_rsv_mutex);
1611         atomic_set(&fs_info->nr_async_submits, 0);
1612         atomic_set(&fs_info->async_delalloc_pages, 0);
1613         atomic_set(&fs_info->async_submit_draining, 0);
1614         atomic_set(&fs_info->nr_async_bios, 0);
1615         fs_info->sb = sb;
1616         fs_info->max_inline = 8192 * 1024;
1617         fs_info->metadata_ratio = 0;
1618
1619         fs_info->thread_pool_size = min_t(unsigned long,
1620                                           num_online_cpus() + 2, 8);
1621
1622         INIT_LIST_HEAD(&fs_info->ordered_extents);
1623         spin_lock_init(&fs_info->ordered_extent_lock);
1624
1625         sb->s_blocksize = 4096;
1626         sb->s_blocksize_bits = blksize_bits(4096);
1627         sb->s_bdi = &fs_info->bdi;
1628
1629         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1630         fs_info->btree_inode->i_nlink = 1;
1631         /*
1632          * we set the i_size on the btree inode to the max possible int.
1633          * the real end of the address space is determined by all of
1634          * the devices in the system
1635          */
1636         fs_info->btree_inode->i_size = OFFSET_MAX;
1637         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1638         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1639
1640         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1641         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1642                              fs_info->btree_inode->i_mapping,
1643                              GFP_NOFS);
1644         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1645                              GFP_NOFS);
1646
1647         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1648
1649         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1650         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1651                sizeof(struct btrfs_key));
1652         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1653         insert_inode_hash(fs_info->btree_inode);
1654
1655         spin_lock_init(&fs_info->block_group_cache_lock);
1656         fs_info->block_group_cache_tree = RB_ROOT;
1657
1658         extent_io_tree_init(&fs_info->freed_extents[0],
1659                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1660         extent_io_tree_init(&fs_info->freed_extents[1],
1661                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1662         fs_info->pinned_extents = &fs_info->freed_extents[0];
1663         fs_info->do_barriers = 1;
1664
1665
1666         mutex_init(&fs_info->trans_mutex);
1667         mutex_init(&fs_info->ordered_operations_mutex);
1668         mutex_init(&fs_info->tree_log_mutex);
1669         mutex_init(&fs_info->chunk_mutex);
1670         mutex_init(&fs_info->transaction_kthread_mutex);
1671         mutex_init(&fs_info->cleaner_mutex);
1672         mutex_init(&fs_info->volume_mutex);
1673         init_rwsem(&fs_info->extent_commit_sem);
1674         init_rwsem(&fs_info->cleanup_work_sem);
1675         init_rwsem(&fs_info->subvol_sem);
1676
1677         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1678         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1679
1680         init_waitqueue_head(&fs_info->transaction_throttle);
1681         init_waitqueue_head(&fs_info->transaction_wait);
1682         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1683         init_waitqueue_head(&fs_info->async_submit_wait);
1684
1685         __setup_root(4096, 4096, 4096, 4096, tree_root,
1686                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1687
1688         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1689         if (!bh)
1690                 goto fail_iput;
1691
1692         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1693         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1694                sizeof(fs_info->super_for_commit));
1695         brelse(bh);
1696
1697         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1698
1699         disk_super = &fs_info->super_copy;
1700         if (!btrfs_super_root(disk_super))
1701                 goto fail_iput;
1702
1703         ret = btrfs_parse_options(tree_root, options);
1704         if (ret) {
1705                 err = ret;
1706                 goto fail_iput;
1707         }
1708
1709         features = btrfs_super_incompat_flags(disk_super) &
1710                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1711         if (features) {
1712                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1713                        "unsupported optional features (%Lx).\n",
1714                        (unsigned long long)features);
1715                 err = -EINVAL;
1716                 goto fail_iput;
1717         }
1718
1719         features = btrfs_super_incompat_flags(disk_super);
1720         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1721                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1722                 btrfs_set_super_incompat_flags(disk_super, features);
1723         }
1724
1725         features = btrfs_super_compat_ro_flags(disk_super) &
1726                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1727         if (!(sb->s_flags & MS_RDONLY) && features) {
1728                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1729                        "unsupported option features (%Lx).\n",
1730                        (unsigned long long)features);
1731                 err = -EINVAL;
1732                 goto fail_iput;
1733         }
1734
1735         btrfs_init_workers(&fs_info->generic_worker,
1736                            "genwork", 1, NULL);
1737
1738         btrfs_init_workers(&fs_info->workers, "worker",
1739                            fs_info->thread_pool_size,
1740                            &fs_info->generic_worker);
1741
1742         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1743                            fs_info->thread_pool_size,
1744                            &fs_info->generic_worker);
1745
1746         btrfs_init_workers(&fs_info->submit_workers, "submit",
1747                            min_t(u64, fs_devices->num_devices,
1748                            fs_info->thread_pool_size),
1749                            &fs_info->generic_worker);
1750
1751         /* a higher idle thresh on the submit workers makes it much more
1752          * likely that bios will be send down in a sane order to the
1753          * devices
1754          */
1755         fs_info->submit_workers.idle_thresh = 64;
1756
1757         fs_info->workers.idle_thresh = 16;
1758         fs_info->workers.ordered = 1;
1759
1760         fs_info->delalloc_workers.idle_thresh = 2;
1761         fs_info->delalloc_workers.ordered = 1;
1762
1763         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1764                            &fs_info->generic_worker);
1765         btrfs_init_workers(&fs_info->endio_workers, "endio",
1766                            fs_info->thread_pool_size,
1767                            &fs_info->generic_worker);
1768         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1769                            fs_info->thread_pool_size,
1770                            &fs_info->generic_worker);
1771         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1772                            "endio-meta-write", fs_info->thread_pool_size,
1773                            &fs_info->generic_worker);
1774         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1775                            fs_info->thread_pool_size,
1776                            &fs_info->generic_worker);
1777         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1778                            1, &fs_info->generic_worker);
1779
1780         /*
1781          * endios are largely parallel and should have a very
1782          * low idle thresh
1783          */
1784         fs_info->endio_workers.idle_thresh = 4;
1785         fs_info->endio_meta_workers.idle_thresh = 4;
1786
1787         fs_info->endio_write_workers.idle_thresh = 2;
1788         fs_info->endio_meta_write_workers.idle_thresh = 2;
1789
1790         btrfs_start_workers(&fs_info->workers, 1);
1791         btrfs_start_workers(&fs_info->generic_worker, 1);
1792         btrfs_start_workers(&fs_info->submit_workers, 1);
1793         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1794         btrfs_start_workers(&fs_info->fixup_workers, 1);
1795         btrfs_start_workers(&fs_info->endio_workers, 1);
1796         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1797         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1798         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1799         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1800
1801         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1802         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1803                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1804
1805         nodesize = btrfs_super_nodesize(disk_super);
1806         leafsize = btrfs_super_leafsize(disk_super);
1807         sectorsize = btrfs_super_sectorsize(disk_super);
1808         stripesize = btrfs_super_stripesize(disk_super);
1809         tree_root->nodesize = nodesize;
1810         tree_root->leafsize = leafsize;
1811         tree_root->sectorsize = sectorsize;
1812         tree_root->stripesize = stripesize;
1813
1814         sb->s_blocksize = sectorsize;
1815         sb->s_blocksize_bits = blksize_bits(sectorsize);
1816
1817         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1818                     sizeof(disk_super->magic))) {
1819                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1820                 goto fail_sb_buffer;
1821         }
1822
1823         mutex_lock(&fs_info->chunk_mutex);
1824         ret = btrfs_read_sys_array(tree_root);
1825         mutex_unlock(&fs_info->chunk_mutex);
1826         if (ret) {
1827                 printk(KERN_WARNING "btrfs: failed to read the system "
1828                        "array on %s\n", sb->s_id);
1829                 goto fail_sb_buffer;
1830         }
1831
1832         blocksize = btrfs_level_size(tree_root,
1833                                      btrfs_super_chunk_root_level(disk_super));
1834         generation = btrfs_super_chunk_root_generation(disk_super);
1835
1836         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1837                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1838
1839         chunk_root->node = read_tree_block(chunk_root,
1840                                            btrfs_super_chunk_root(disk_super),
1841                                            blocksize, generation);
1842         BUG_ON(!chunk_root->node);
1843         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1844                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1845                        sb->s_id);
1846                 goto fail_chunk_root;
1847         }
1848         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1849         chunk_root->commit_root = btrfs_root_node(chunk_root);
1850
1851         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1852            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1853            BTRFS_UUID_SIZE);
1854
1855         mutex_lock(&fs_info->chunk_mutex);
1856         ret = btrfs_read_chunk_tree(chunk_root);
1857         mutex_unlock(&fs_info->chunk_mutex);
1858         if (ret) {
1859                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1860                        sb->s_id);
1861                 goto fail_chunk_root;
1862         }
1863
1864         btrfs_close_extra_devices(fs_devices);
1865
1866         blocksize = btrfs_level_size(tree_root,
1867                                      btrfs_super_root_level(disk_super));
1868         generation = btrfs_super_generation(disk_super);
1869
1870         tree_root->node = read_tree_block(tree_root,
1871                                           btrfs_super_root(disk_super),
1872                                           blocksize, generation);
1873         if (!tree_root->node)
1874                 goto fail_chunk_root;
1875         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1876                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1877                        sb->s_id);
1878                 goto fail_tree_root;
1879         }
1880         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1881         tree_root->commit_root = btrfs_root_node(tree_root);
1882
1883         ret = find_and_setup_root(tree_root, fs_info,
1884                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1885         if (ret)
1886                 goto fail_tree_root;
1887         extent_root->track_dirty = 1;
1888
1889         ret = find_and_setup_root(tree_root, fs_info,
1890                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1891         if (ret)
1892                 goto fail_extent_root;
1893         dev_root->track_dirty = 1;
1894
1895         ret = find_and_setup_root(tree_root, fs_info,
1896                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1897         if (ret)
1898                 goto fail_dev_root;
1899
1900         csum_root->track_dirty = 1;
1901
1902         fs_info->generation = generation;
1903         fs_info->last_trans_committed = generation;
1904         fs_info->data_alloc_profile = (u64)-1;
1905         fs_info->metadata_alloc_profile = (u64)-1;
1906         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1907
1908         ret = btrfs_read_block_groups(extent_root);
1909         if (ret) {
1910                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1911                 goto fail_block_groups;
1912         }
1913
1914         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1915                                                "btrfs-cleaner");
1916         if (IS_ERR(fs_info->cleaner_kthread))
1917                 goto fail_block_groups;
1918
1919         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1920                                                    tree_root,
1921                                                    "btrfs-transaction");
1922         if (IS_ERR(fs_info->transaction_kthread))
1923                 goto fail_cleaner;
1924
1925         if (!btrfs_test_opt(tree_root, SSD) &&
1926             !btrfs_test_opt(tree_root, NOSSD) &&
1927             !fs_info->fs_devices->rotating) {
1928                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1929                        "mode\n");
1930                 btrfs_set_opt(fs_info->mount_opt, SSD);
1931         }
1932
1933         if (btrfs_super_log_root(disk_super) != 0) {
1934                 u64 bytenr = btrfs_super_log_root(disk_super);
1935
1936                 if (fs_devices->rw_devices == 0) {
1937                         printk(KERN_WARNING "Btrfs log replay required "
1938                                "on RO media\n");
1939                         err = -EIO;
1940                         goto fail_trans_kthread;
1941                 }
1942                 blocksize =
1943                      btrfs_level_size(tree_root,
1944                                       btrfs_super_log_root_level(disk_super));
1945
1946                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1947                 if (!log_tree_root) {
1948                         err = -ENOMEM;
1949                         goto fail_trans_kthread;
1950                 }
1951
1952                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1953                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1954
1955                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1956                                                       blocksize,
1957                                                       generation + 1);
1958                 ret = btrfs_recover_log_trees(log_tree_root);
1959                 BUG_ON(ret);
1960
1961                 if (sb->s_flags & MS_RDONLY) {
1962                         ret =  btrfs_commit_super(tree_root);
1963                         BUG_ON(ret);
1964                 }
1965         }
1966
1967         ret = btrfs_find_orphan_roots(tree_root);
1968         BUG_ON(ret);
1969
1970         if (!(sb->s_flags & MS_RDONLY)) {
1971                 ret = btrfs_cleanup_fs_roots(fs_info);
1972                 BUG_ON(ret);
1973
1974                 ret = btrfs_recover_relocation(tree_root);
1975                 if (ret < 0) {
1976                         printk(KERN_WARNING
1977                                "btrfs: failed to recover relocation\n");
1978                         err = -EINVAL;
1979                         goto fail_trans_kthread;
1980                 }
1981         }
1982
1983         location.objectid = BTRFS_FS_TREE_OBJECTID;
1984         location.type = BTRFS_ROOT_ITEM_KEY;
1985         location.offset = (u64)-1;
1986
1987         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1988         if (!fs_info->fs_root)
1989                 goto fail_trans_kthread;
1990         if (IS_ERR(fs_info->fs_root)) {
1991                 err = PTR_ERR(fs_info->fs_root);
1992                 goto fail_trans_kthread;
1993         }
1994
1995         if (!(sb->s_flags & MS_RDONLY)) {
1996                 down_read(&fs_info->cleanup_work_sem);
1997                 btrfs_orphan_cleanup(fs_info->fs_root);
1998                 btrfs_orphan_cleanup(fs_info->tree_root);
1999                 up_read(&fs_info->cleanup_work_sem);
2000         }
2001
2002         return tree_root;
2003
2004 fail_trans_kthread:
2005         kthread_stop(fs_info->transaction_kthread);
2006 fail_cleaner:
2007         kthread_stop(fs_info->cleaner_kthread);
2008
2009         /*
2010          * make sure we're done with the btree inode before we stop our
2011          * kthreads
2012          */
2013         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2014         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2015
2016 fail_block_groups:
2017         btrfs_free_block_groups(fs_info);
2018         free_extent_buffer(csum_root->node);
2019         free_extent_buffer(csum_root->commit_root);
2020 fail_dev_root:
2021         free_extent_buffer(dev_root->node);
2022         free_extent_buffer(dev_root->commit_root);
2023 fail_extent_root:
2024         free_extent_buffer(extent_root->node);
2025         free_extent_buffer(extent_root->commit_root);
2026 fail_tree_root:
2027         free_extent_buffer(tree_root->node);
2028         free_extent_buffer(tree_root->commit_root);
2029 fail_chunk_root:
2030         free_extent_buffer(chunk_root->node);
2031         free_extent_buffer(chunk_root->commit_root);
2032 fail_sb_buffer:
2033         btrfs_stop_workers(&fs_info->generic_worker);
2034         btrfs_stop_workers(&fs_info->fixup_workers);
2035         btrfs_stop_workers(&fs_info->delalloc_workers);
2036         btrfs_stop_workers(&fs_info->workers);
2037         btrfs_stop_workers(&fs_info->endio_workers);
2038         btrfs_stop_workers(&fs_info->endio_meta_workers);
2039         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2040         btrfs_stop_workers(&fs_info->endio_write_workers);
2041         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2042         btrfs_stop_workers(&fs_info->submit_workers);
2043 fail_iput:
2044         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2045         iput(fs_info->btree_inode);
2046
2047         btrfs_close_devices(fs_info->fs_devices);
2048         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2049 fail_bdi:
2050         bdi_destroy(&fs_info->bdi);
2051 fail_srcu:
2052         cleanup_srcu_struct(&fs_info->subvol_srcu);
2053 fail:
2054         kfree(extent_root);
2055         kfree(tree_root);
2056         kfree(fs_info);
2057         kfree(chunk_root);
2058         kfree(dev_root);
2059         kfree(csum_root);
2060         return ERR_PTR(err);
2061 }
2062
2063 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2064 {
2065         char b[BDEVNAME_SIZE];
2066
2067         if (uptodate) {
2068                 set_buffer_uptodate(bh);
2069         } else {
2070                 if (printk_ratelimit()) {
2071                         printk(KERN_WARNING "lost page write due to "
2072                                         "I/O error on %s\n",
2073                                        bdevname(bh->b_bdev, b));
2074                 }
2075                 /* note, we dont' set_buffer_write_io_error because we have
2076                  * our own ways of dealing with the IO errors
2077                  */
2078                 clear_buffer_uptodate(bh);
2079         }
2080         unlock_buffer(bh);
2081         put_bh(bh);
2082 }
2083
2084 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2085 {
2086         struct buffer_head *bh;
2087         struct buffer_head *latest = NULL;
2088         struct btrfs_super_block *super;
2089         int i;
2090         u64 transid = 0;
2091         u64 bytenr;
2092
2093         /* we would like to check all the supers, but that would make
2094          * a btrfs mount succeed after a mkfs from a different FS.
2095          * So, we need to add a special mount option to scan for
2096          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2097          */
2098         for (i = 0; i < 1; i++) {
2099                 bytenr = btrfs_sb_offset(i);
2100                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2101                         break;
2102                 bh = __bread(bdev, bytenr / 4096, 4096);
2103                 if (!bh)
2104                         continue;
2105
2106                 super = (struct btrfs_super_block *)bh->b_data;
2107                 if (btrfs_super_bytenr(super) != bytenr ||
2108                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2109                             sizeof(super->magic))) {
2110                         brelse(bh);
2111                         continue;
2112                 }
2113
2114                 if (!latest || btrfs_super_generation(super) > transid) {
2115                         brelse(latest);
2116                         latest = bh;
2117                         transid = btrfs_super_generation(super);
2118                 } else {
2119                         brelse(bh);
2120                 }
2121         }
2122         return latest;
2123 }
2124
2125 /*
2126  * this should be called twice, once with wait == 0 and
2127  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2128  * we write are pinned.
2129  *
2130  * They are released when wait == 1 is done.
2131  * max_mirrors must be the same for both runs, and it indicates how
2132  * many supers on this one device should be written.
2133  *
2134  * max_mirrors == 0 means to write them all.
2135  */
2136 static int write_dev_supers(struct btrfs_device *device,
2137                             struct btrfs_super_block *sb,
2138                             int do_barriers, int wait, int max_mirrors)
2139 {
2140         struct buffer_head *bh;
2141         int i;
2142         int ret;
2143         int errors = 0;
2144         u32 crc;
2145         u64 bytenr;
2146         int last_barrier = 0;
2147
2148         if (max_mirrors == 0)
2149                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2150
2151         /* make sure only the last submit_bh does a barrier */
2152         if (do_barriers) {
2153                 for (i = 0; i < max_mirrors; i++) {
2154                         bytenr = btrfs_sb_offset(i);
2155                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2156                             device->total_bytes)
2157                                 break;
2158                         last_barrier = i;
2159                 }
2160         }
2161
2162         for (i = 0; i < max_mirrors; i++) {
2163                 bytenr = btrfs_sb_offset(i);
2164                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2165                         break;
2166
2167                 if (wait) {
2168                         bh = __find_get_block(device->bdev, bytenr / 4096,
2169                                               BTRFS_SUPER_INFO_SIZE);
2170                         BUG_ON(!bh);
2171                         wait_on_buffer(bh);
2172                         if (!buffer_uptodate(bh))
2173                                 errors++;
2174
2175                         /* drop our reference */
2176                         brelse(bh);
2177
2178                         /* drop the reference from the wait == 0 run */
2179                         brelse(bh);
2180                         continue;
2181                 } else {
2182                         btrfs_set_super_bytenr(sb, bytenr);
2183
2184                         crc = ~(u32)0;
2185                         crc = btrfs_csum_data(NULL, (char *)sb +
2186                                               BTRFS_CSUM_SIZE, crc,
2187                                               BTRFS_SUPER_INFO_SIZE -
2188                                               BTRFS_CSUM_SIZE);
2189                         btrfs_csum_final(crc, sb->csum);
2190
2191                         /*
2192                          * one reference for us, and we leave it for the
2193                          * caller
2194                          */
2195                         bh = __getblk(device->bdev, bytenr / 4096,
2196                                       BTRFS_SUPER_INFO_SIZE);
2197                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2198
2199                         /* one reference for submit_bh */
2200                         get_bh(bh);
2201
2202                         set_buffer_uptodate(bh);
2203                         lock_buffer(bh);
2204                         bh->b_end_io = btrfs_end_buffer_write_sync;
2205                 }
2206
2207                 if (i == last_barrier && do_barriers)
2208                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2209                 else
2210                         ret = submit_bh(WRITE_SYNC, bh);
2211
2212                 if (ret)
2213                         errors++;
2214         }
2215         return errors < i ? 0 : -1;
2216 }
2217
2218 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2219 {
2220         struct list_head *head;
2221         struct btrfs_device *dev;
2222         struct btrfs_super_block *sb;
2223         struct btrfs_dev_item *dev_item;
2224         int ret;
2225         int do_barriers;
2226         int max_errors;
2227         int total_errors = 0;
2228         u64 flags;
2229
2230         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2231         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2232
2233         sb = &root->fs_info->super_for_commit;
2234         dev_item = &sb->dev_item;
2235
2236         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2237         head = &root->fs_info->fs_devices->devices;
2238         list_for_each_entry(dev, head, dev_list) {
2239                 if (!dev->bdev) {
2240                         total_errors++;
2241                         continue;
2242                 }
2243                 if (!dev->in_fs_metadata || !dev->writeable)
2244                         continue;
2245
2246                 btrfs_set_stack_device_generation(dev_item, 0);
2247                 btrfs_set_stack_device_type(dev_item, dev->type);
2248                 btrfs_set_stack_device_id(dev_item, dev->devid);
2249                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2250                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2251                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2252                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2253                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2254                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2255                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2256
2257                 flags = btrfs_super_flags(sb);
2258                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2259
2260                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2261                 if (ret)
2262                         total_errors++;
2263         }
2264         if (total_errors > max_errors) {
2265                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2266                        total_errors);
2267                 BUG();
2268         }
2269
2270         total_errors = 0;
2271         list_for_each_entry(dev, head, dev_list) {
2272                 if (!dev->bdev)
2273                         continue;
2274                 if (!dev->in_fs_metadata || !dev->writeable)
2275                         continue;
2276
2277                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2278                 if (ret)
2279                         total_errors++;
2280         }
2281         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2282         if (total_errors > max_errors) {
2283                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2284                        total_errors);
2285                 BUG();
2286         }
2287         return 0;
2288 }
2289
2290 int write_ctree_super(struct btrfs_trans_handle *trans,
2291                       struct btrfs_root *root, int max_mirrors)
2292 {
2293         int ret;
2294
2295         ret = write_all_supers(root, max_mirrors);
2296         return ret;
2297 }
2298
2299 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2300 {
2301         spin_lock(&fs_info->fs_roots_radix_lock);
2302         radix_tree_delete(&fs_info->fs_roots_radix,
2303                           (unsigned long)root->root_key.objectid);
2304         spin_unlock(&fs_info->fs_roots_radix_lock);
2305
2306         if (btrfs_root_refs(&root->root_item) == 0)
2307                 synchronize_srcu(&fs_info->subvol_srcu);
2308
2309         free_fs_root(root);
2310         return 0;
2311 }
2312
2313 static void free_fs_root(struct btrfs_root *root)
2314 {
2315         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2316         if (root->anon_super.s_dev) {
2317                 down_write(&root->anon_super.s_umount);
2318                 kill_anon_super(&root->anon_super);
2319         }
2320         free_extent_buffer(root->node);
2321         free_extent_buffer(root->commit_root);
2322         kfree(root->name);
2323         kfree(root);
2324 }
2325
2326 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2327 {
2328         int ret;
2329         struct btrfs_root *gang[8];
2330         int i;
2331
2332         while (!list_empty(&fs_info->dead_roots)) {
2333                 gang[0] = list_entry(fs_info->dead_roots.next,
2334                                      struct btrfs_root, root_list);
2335                 list_del(&gang[0]->root_list);
2336
2337                 if (gang[0]->in_radix) {
2338                         btrfs_free_fs_root(fs_info, gang[0]);
2339                 } else {
2340                         free_extent_buffer(gang[0]->node);
2341                         free_extent_buffer(gang[0]->commit_root);
2342                         kfree(gang[0]);
2343                 }
2344         }
2345
2346         while (1) {
2347                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2348                                              (void **)gang, 0,
2349                                              ARRAY_SIZE(gang));
2350                 if (!ret)
2351                         break;
2352                 for (i = 0; i < ret; i++)
2353                         btrfs_free_fs_root(fs_info, gang[i]);
2354         }
2355         return 0;
2356 }
2357
2358 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2359 {
2360         u64 root_objectid = 0;
2361         struct btrfs_root *gang[8];
2362         int i;
2363         int ret;
2364
2365         while (1) {
2366                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2367                                              (void **)gang, root_objectid,
2368                                              ARRAY_SIZE(gang));
2369                 if (!ret)
2370                         break;
2371
2372                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2373                 for (i = 0; i < ret; i++) {
2374                         root_objectid = gang[i]->root_key.objectid;
2375                         btrfs_orphan_cleanup(gang[i]);
2376                 }
2377                 root_objectid++;
2378         }
2379         return 0;
2380 }
2381
2382 int btrfs_commit_super(struct btrfs_root *root)
2383 {
2384         struct btrfs_trans_handle *trans;
2385         int ret;
2386
2387         mutex_lock(&root->fs_info->cleaner_mutex);
2388         btrfs_run_delayed_iputs(root);
2389         btrfs_clean_old_snapshots(root);
2390         mutex_unlock(&root->fs_info->cleaner_mutex);
2391
2392         /* wait until ongoing cleanup work done */
2393         down_write(&root->fs_info->cleanup_work_sem);
2394         up_write(&root->fs_info->cleanup_work_sem);
2395
2396         trans = btrfs_join_transaction(root, 1);
2397         ret = btrfs_commit_transaction(trans, root);
2398         BUG_ON(ret);
2399         /* run commit again to drop the original snapshot */
2400         trans = btrfs_join_transaction(root, 1);
2401         btrfs_commit_transaction(trans, root);
2402         ret = btrfs_write_and_wait_transaction(NULL, root);
2403         BUG_ON(ret);
2404
2405         ret = write_ctree_super(NULL, root, 0);
2406         return ret;
2407 }
2408
2409 int close_ctree(struct btrfs_root *root)
2410 {
2411         struct btrfs_fs_info *fs_info = root->fs_info;
2412         int ret;
2413
2414         fs_info->closing = 1;
2415         smp_mb();
2416
2417         btrfs_put_block_group_cache(fs_info);
2418         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2419                 ret =  btrfs_commit_super(root);
2420                 if (ret)
2421                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2422         }
2423
2424         kthread_stop(root->fs_info->transaction_kthread);
2425         kthread_stop(root->fs_info->cleaner_kthread);
2426
2427         fs_info->closing = 2;
2428         smp_mb();
2429
2430         if (fs_info->delalloc_bytes) {
2431                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2432                        (unsigned long long)fs_info->delalloc_bytes);
2433         }
2434         if (fs_info->total_ref_cache_size) {
2435                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2436                        (unsigned long long)fs_info->total_ref_cache_size);
2437         }
2438
2439         free_extent_buffer(fs_info->extent_root->node);
2440         free_extent_buffer(fs_info->extent_root->commit_root);
2441         free_extent_buffer(fs_info->tree_root->node);
2442         free_extent_buffer(fs_info->tree_root->commit_root);
2443         free_extent_buffer(root->fs_info->chunk_root->node);
2444         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2445         free_extent_buffer(root->fs_info->dev_root->node);
2446         free_extent_buffer(root->fs_info->dev_root->commit_root);
2447         free_extent_buffer(root->fs_info->csum_root->node);
2448         free_extent_buffer(root->fs_info->csum_root->commit_root);
2449
2450         btrfs_free_block_groups(root->fs_info);
2451
2452         del_fs_roots(fs_info);
2453
2454         iput(fs_info->btree_inode);
2455
2456         btrfs_stop_workers(&fs_info->generic_worker);
2457         btrfs_stop_workers(&fs_info->fixup_workers);
2458         btrfs_stop_workers(&fs_info->delalloc_workers);
2459         btrfs_stop_workers(&fs_info->workers);
2460         btrfs_stop_workers(&fs_info->endio_workers);
2461         btrfs_stop_workers(&fs_info->endio_meta_workers);
2462         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2463         btrfs_stop_workers(&fs_info->endio_write_workers);
2464         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2465         btrfs_stop_workers(&fs_info->submit_workers);
2466
2467         btrfs_close_devices(fs_info->fs_devices);
2468         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2469
2470         bdi_destroy(&fs_info->bdi);
2471         cleanup_srcu_struct(&fs_info->subvol_srcu);
2472
2473         kfree(fs_info->extent_root);
2474         kfree(fs_info->tree_root);
2475         kfree(fs_info->chunk_root);
2476         kfree(fs_info->dev_root);
2477         kfree(fs_info->csum_root);
2478         return 0;
2479 }
2480
2481 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2482 {
2483         int ret;
2484         struct inode *btree_inode = buf->first_page->mapping->host;
2485
2486         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2487                                      NULL);
2488         if (!ret)
2489                 return ret;
2490
2491         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2492                                     parent_transid);
2493         return !ret;
2494 }
2495
2496 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2497 {
2498         struct inode *btree_inode = buf->first_page->mapping->host;
2499         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2500                                           buf);
2501 }
2502
2503 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2504 {
2505         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2506         u64 transid = btrfs_header_generation(buf);
2507         struct inode *btree_inode = root->fs_info->btree_inode;
2508         int was_dirty;
2509
2510         btrfs_assert_tree_locked(buf);
2511         if (transid != root->fs_info->generation) {
2512                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2513                        "found %llu running %llu\n",
2514                         (unsigned long long)buf->start,
2515                         (unsigned long long)transid,
2516                         (unsigned long long)root->fs_info->generation);
2517                 WARN_ON(1);
2518         }
2519         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2520                                             buf);
2521         if (!was_dirty) {
2522                 spin_lock(&root->fs_info->delalloc_lock);
2523                 root->fs_info->dirty_metadata_bytes += buf->len;
2524                 spin_unlock(&root->fs_info->delalloc_lock);
2525         }
2526 }
2527
2528 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2529 {
2530         /*
2531          * looks as though older kernels can get into trouble with
2532          * this code, they end up stuck in balance_dirty_pages forever
2533          */
2534         u64 num_dirty;
2535         unsigned long thresh = 32 * 1024 * 1024;
2536
2537         if (current->flags & PF_MEMALLOC)
2538                 return;
2539
2540         num_dirty = root->fs_info->dirty_metadata_bytes;
2541
2542         if (num_dirty > thresh) {
2543                 balance_dirty_pages_ratelimited_nr(
2544                                    root->fs_info->btree_inode->i_mapping, 1);
2545         }
2546         return;
2547 }
2548
2549 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2550 {
2551         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2552         int ret;
2553         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2554         if (ret == 0)
2555                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2556         return ret;
2557 }
2558
2559 int btree_lock_page_hook(struct page *page)
2560 {
2561         struct inode *inode = page->mapping->host;
2562         struct btrfs_root *root = BTRFS_I(inode)->root;
2563         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2564         struct extent_buffer *eb;
2565         unsigned long len;
2566         u64 bytenr = page_offset(page);
2567
2568         if (page->private == EXTENT_PAGE_PRIVATE)
2569                 goto out;
2570
2571         len = page->private >> 2;
2572         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2573         if (!eb)
2574                 goto out;
2575
2576         btrfs_tree_lock(eb);
2577         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2578
2579         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2580                 spin_lock(&root->fs_info->delalloc_lock);
2581                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2582                         root->fs_info->dirty_metadata_bytes -= eb->len;
2583                 else
2584                         WARN_ON(1);
2585                 spin_unlock(&root->fs_info->delalloc_lock);
2586         }
2587
2588         btrfs_tree_unlock(eb);
2589         free_extent_buffer(eb);
2590 out:
2591         lock_page(page);
2592         return 0;
2593 }
2594
2595 static struct extent_io_ops btree_extent_io_ops = {
2596         .write_cache_pages_lock_hook = btree_lock_page_hook,
2597         .readpage_end_io_hook = btree_readpage_end_io_hook,
2598         .submit_bio_hook = btree_submit_bio_hook,
2599         /* note we're sharing with inode.c for the merge bio hook */
2600         .merge_bio_hook = btrfs_merge_bio_hook,
2601 };