Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[linux-2.6.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 static atomic_t btrfs_bdi_num = ATOMIC_INIT(0);
48
49 /*
50  * end_io_wq structs are used to do processing in task context when an IO is
51  * complete.  This is used during reads to verify checksums, and it is used
52  * by writes to insert metadata for new file extents after IO is complete.
53  */
54 struct end_io_wq {
55         struct bio *bio;
56         bio_end_io_t *end_io;
57         void *private;
58         struct btrfs_fs_info *info;
59         int error;
60         int metadata;
61         struct list_head list;
62         struct btrfs_work work;
63 };
64
65 /*
66  * async submit bios are used to offload expensive checksumming
67  * onto the worker threads.  They checksum file and metadata bios
68  * just before they are sent down the IO stack.
69  */
70 struct async_submit_bio {
71         struct inode *inode;
72         struct bio *bio;
73         struct list_head list;
74         extent_submit_bio_hook_t *submit_bio_start;
75         extent_submit_bio_hook_t *submit_bio_done;
76         int rw;
77         int mirror_num;
78         unsigned long bio_flags;
79         struct btrfs_work work;
80 };
81
82 /* These are used to set the lockdep class on the extent buffer locks.
83  * The class is set by the readpage_end_io_hook after the buffer has
84  * passed csum validation but before the pages are unlocked.
85  *
86  * The lockdep class is also set by btrfs_init_new_buffer on freshly
87  * allocated blocks.
88  *
89  * The class is based on the level in the tree block, which allows lockdep
90  * to know that lower nodes nest inside the locks of higher nodes.
91  *
92  * We also add a check to make sure the highest level of the tree is
93  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
94  * code needs update as well.
95  */
96 #ifdef CONFIG_DEBUG_LOCK_ALLOC
97 # if BTRFS_MAX_LEVEL != 8
98 #  error
99 # endif
100 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
101 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
102         /* leaf */
103         "btrfs-extent-00",
104         "btrfs-extent-01",
105         "btrfs-extent-02",
106         "btrfs-extent-03",
107         "btrfs-extent-04",
108         "btrfs-extent-05",
109         "btrfs-extent-06",
110         "btrfs-extent-07",
111         /* highest possible level */
112         "btrfs-extent-08",
113 };
114 #endif
115
116 /*
117  * extents on the btree inode are pretty simple, there's one extent
118  * that covers the entire device
119  */
120 static struct extent_map *btree_get_extent(struct inode *inode,
121                 struct page *page, size_t page_offset, u64 start, u64 len,
122                 int create)
123 {
124         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
125         struct extent_map *em;
126         int ret;
127
128         read_lock(&em_tree->lock);
129         em = lookup_extent_mapping(em_tree, start, len);
130         if (em) {
131                 em->bdev =
132                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
133                 read_unlock(&em_tree->lock);
134                 goto out;
135         }
136         read_unlock(&em_tree->lock);
137
138         em = alloc_extent_map(GFP_NOFS);
139         if (!em) {
140                 em = ERR_PTR(-ENOMEM);
141                 goto out;
142         }
143         em->start = 0;
144         em->len = (u64)-1;
145         em->block_len = (u64)-1;
146         em->block_start = 0;
147         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
148
149         write_lock(&em_tree->lock);
150         ret = add_extent_mapping(em_tree, em);
151         if (ret == -EEXIST) {
152                 u64 failed_start = em->start;
153                 u64 failed_len = em->len;
154
155                 free_extent_map(em);
156                 em = lookup_extent_mapping(em_tree, start, len);
157                 if (em) {
158                         ret = 0;
159                 } else {
160                         em = lookup_extent_mapping(em_tree, failed_start,
161                                                    failed_len);
162                         ret = -EIO;
163                 }
164         } else if (ret) {
165                 free_extent_map(em);
166                 em = NULL;
167         }
168         write_unlock(&em_tree->lock);
169
170         if (ret)
171                 em = ERR_PTR(ret);
172 out:
173         return em;
174 }
175
176 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
177 {
178         return crc32c(seed, data, len);
179 }
180
181 void btrfs_csum_final(u32 crc, char *result)
182 {
183         *(__le32 *)result = ~cpu_to_le32(crc);
184 }
185
186 /*
187  * compute the csum for a btree block, and either verify it or write it
188  * into the csum field of the block.
189  */
190 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
191                            int verify)
192 {
193         u16 csum_size =
194                 btrfs_super_csum_size(&root->fs_info->super_copy);
195         char *result = NULL;
196         unsigned long len;
197         unsigned long cur_len;
198         unsigned long offset = BTRFS_CSUM_SIZE;
199         char *map_token = NULL;
200         char *kaddr;
201         unsigned long map_start;
202         unsigned long map_len;
203         int err;
204         u32 crc = ~(u32)0;
205         unsigned long inline_result;
206
207         len = buf->len - offset;
208         while (len > 0) {
209                 err = map_private_extent_buffer(buf, offset, 32,
210                                         &map_token, &kaddr,
211                                         &map_start, &map_len, KM_USER0);
212                 if (err)
213                         return 1;
214                 cur_len = min(len, map_len - (offset - map_start));
215                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
216                                       crc, cur_len);
217                 len -= cur_len;
218                 offset += cur_len;
219                 unmap_extent_buffer(buf, map_token, KM_USER0);
220         }
221         if (csum_size > sizeof(inline_result)) {
222                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
223                 if (!result)
224                         return 1;
225         } else {
226                 result = (char *)&inline_result;
227         }
228
229         btrfs_csum_final(crc, result);
230
231         if (verify) {
232                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
233                         u32 val;
234                         u32 found = 0;
235                         memcpy(&found, result, csum_size);
236
237                         read_extent_buffer(buf, &val, 0, csum_size);
238                         if (printk_ratelimit()) {
239                                 printk(KERN_INFO "btrfs: %s checksum verify "
240                                        "failed on %llu wanted %X found %X "
241                                        "level %d\n",
242                                        root->fs_info->sb->s_id,
243                                        (unsigned long long)buf->start, val, found,
244                                        btrfs_header_level(buf));
245                         }
246                         if (result != (char *)&inline_result)
247                                 kfree(result);
248                         return 1;
249                 }
250         } else {
251                 write_extent_buffer(buf, result, 0, csum_size);
252         }
253         if (result != (char *)&inline_result)
254                 kfree(result);
255         return 0;
256 }
257
258 /*
259  * we can't consider a given block up to date unless the transid of the
260  * block matches the transid in the parent node's pointer.  This is how we
261  * detect blocks that either didn't get written at all or got written
262  * in the wrong place.
263  */
264 static int verify_parent_transid(struct extent_io_tree *io_tree,
265                                  struct extent_buffer *eb, u64 parent_transid)
266 {
267         struct extent_state *cached_state = NULL;
268         int ret;
269
270         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
271                 return 0;
272
273         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
274                          0, &cached_state, GFP_NOFS);
275         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
276             btrfs_header_generation(eb) == parent_transid) {
277                 ret = 0;
278                 goto out;
279         }
280         if (printk_ratelimit()) {
281                 printk("parent transid verify failed on %llu wanted %llu "
282                        "found %llu\n",
283                        (unsigned long long)eb->start,
284                        (unsigned long long)parent_transid,
285                        (unsigned long long)btrfs_header_generation(eb));
286         }
287         ret = 1;
288         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
289 out:
290         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
291                              &cached_state, GFP_NOFS);
292         return ret;
293 }
294
295 /*
296  * helper to read a given tree block, doing retries as required when
297  * the checksums don't match and we have alternate mirrors to try.
298  */
299 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
300                                           struct extent_buffer *eb,
301                                           u64 start, u64 parent_transid)
302 {
303         struct extent_io_tree *io_tree;
304         int ret;
305         int num_copies = 0;
306         int mirror_num = 0;
307
308         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
309         while (1) {
310                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
311                                                btree_get_extent, mirror_num);
312                 if (!ret &&
313                     !verify_parent_transid(io_tree, eb, parent_transid))
314                         return ret;
315
316                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
317                                               eb->start, eb->len);
318                 if (num_copies == 1)
319                         return ret;
320
321                 mirror_num++;
322                 if (mirror_num > num_copies)
323                         return ret;
324         }
325         return -EIO;
326 }
327
328 /*
329  * checksum a dirty tree block before IO.  This has extra checks to make sure
330  * we only fill in the checksum field in the first page of a multi-page block
331  */
332
333 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
334 {
335         struct extent_io_tree *tree;
336         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
337         u64 found_start;
338         int found_level;
339         unsigned long len;
340         struct extent_buffer *eb;
341         int ret;
342
343         tree = &BTRFS_I(page->mapping->host)->io_tree;
344
345         if (page->private == EXTENT_PAGE_PRIVATE)
346                 goto out;
347         if (!page->private)
348                 goto out;
349         len = page->private >> 2;
350         WARN_ON(len == 0);
351
352         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
353         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
354                                              btrfs_header_generation(eb));
355         BUG_ON(ret);
356         found_start = btrfs_header_bytenr(eb);
357         if (found_start != start) {
358                 WARN_ON(1);
359                 goto err;
360         }
361         if (eb->first_page != page) {
362                 WARN_ON(1);
363                 goto err;
364         }
365         if (!PageUptodate(page)) {
366                 WARN_ON(1);
367                 goto err;
368         }
369         found_level = btrfs_header_level(eb);
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 & (1 << BIO_RW)) {
481                 if (end_io_wq->metadata)
482                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
483                                            &end_io_wq->work);
484                 else
485                         btrfs_queue_worker(&fs_info->endio_write_workers,
486                                            &end_io_wq->work);
487         } else {
488                 if (end_io_wq->metadata)
489                         btrfs_queue_worker(&fs_info->endio_meta_workers,
490                                            &end_io_wq->work);
491                 else
492                         btrfs_queue_worker(&fs_info->endio_workers,
493                                            &end_io_wq->work);
494         }
495 }
496
497 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
498                         int metadata)
499 {
500         struct end_io_wq *end_io_wq;
501         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
502         if (!end_io_wq)
503                 return -ENOMEM;
504
505         end_io_wq->private = bio->bi_private;
506         end_io_wq->end_io = bio->bi_end_io;
507         end_io_wq->info = info;
508         end_io_wq->error = 0;
509         end_io_wq->bio = bio;
510         end_io_wq->metadata = metadata;
511
512         bio->bi_private = end_io_wq;
513         bio->bi_end_io = end_workqueue_bio;
514         return 0;
515 }
516
517 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
518 {
519         unsigned long limit = min_t(unsigned long,
520                                     info->workers.max_workers,
521                                     info->fs_devices->open_devices);
522         return 256 * limit;
523 }
524
525 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
526 {
527         return atomic_read(&info->nr_async_bios) >
528                 btrfs_async_submit_limit(info);
529 }
530
531 static void run_one_async_start(struct btrfs_work *work)
532 {
533         struct btrfs_fs_info *fs_info;
534         struct async_submit_bio *async;
535
536         async = container_of(work, struct  async_submit_bio, work);
537         fs_info = BTRFS_I(async->inode)->root->fs_info;
538         async->submit_bio_start(async->inode, async->rw, async->bio,
539                                async->mirror_num, async->bio_flags);
540 }
541
542 static void run_one_async_done(struct btrfs_work *work)
543 {
544         struct btrfs_fs_info *fs_info;
545         struct async_submit_bio *async;
546         int limit;
547
548         async = container_of(work, struct  async_submit_bio, work);
549         fs_info = BTRFS_I(async->inode)->root->fs_info;
550
551         limit = btrfs_async_submit_limit(fs_info);
552         limit = limit * 2 / 3;
553
554         atomic_dec(&fs_info->nr_async_submits);
555
556         if (atomic_read(&fs_info->nr_async_submits) < limit &&
557             waitqueue_active(&fs_info->async_submit_wait))
558                 wake_up(&fs_info->async_submit_wait);
559
560         async->submit_bio_done(async->inode, async->rw, async->bio,
561                                async->mirror_num, async->bio_flags);
562 }
563
564 static void run_one_async_free(struct btrfs_work *work)
565 {
566         struct async_submit_bio *async;
567
568         async = container_of(work, struct  async_submit_bio, work);
569         kfree(async);
570 }
571
572 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
573                         int rw, struct bio *bio, int mirror_num,
574                         unsigned long bio_flags,
575                         extent_submit_bio_hook_t *submit_bio_start,
576                         extent_submit_bio_hook_t *submit_bio_done)
577 {
578         struct async_submit_bio *async;
579
580         async = kmalloc(sizeof(*async), GFP_NOFS);
581         if (!async)
582                 return -ENOMEM;
583
584         async->inode = inode;
585         async->rw = rw;
586         async->bio = bio;
587         async->mirror_num = mirror_num;
588         async->submit_bio_start = submit_bio_start;
589         async->submit_bio_done = submit_bio_done;
590
591         async->work.func = run_one_async_start;
592         async->work.ordered_func = run_one_async_done;
593         async->work.ordered_free = run_one_async_free;
594
595         async->work.flags = 0;
596         async->bio_flags = bio_flags;
597
598         atomic_inc(&fs_info->nr_async_submits);
599
600         if (rw & (1 << BIO_RW_SYNCIO))
601                 btrfs_set_work_high_prio(&async->work);
602
603         btrfs_queue_worker(&fs_info->workers, &async->work);
604
605         while (atomic_read(&fs_info->async_submit_draining) &&
606               atomic_read(&fs_info->nr_async_submits)) {
607                 wait_event(fs_info->async_submit_wait,
608                            (atomic_read(&fs_info->nr_async_submits) == 0));
609         }
610
611         return 0;
612 }
613
614 static int btree_csum_one_bio(struct bio *bio)
615 {
616         struct bio_vec *bvec = bio->bi_io_vec;
617         int bio_index = 0;
618         struct btrfs_root *root;
619
620         WARN_ON(bio->bi_vcnt <= 0);
621         while (bio_index < bio->bi_vcnt) {
622                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
623                 csum_dirty_buffer(root, bvec->bv_page);
624                 bio_index++;
625                 bvec++;
626         }
627         return 0;
628 }
629
630 static int __btree_submit_bio_start(struct inode *inode, int rw,
631                                     struct bio *bio, int mirror_num,
632                                     unsigned long bio_flags)
633 {
634         /*
635          * when we're called for a write, we're already in the async
636          * submission context.  Just jump into btrfs_map_bio
637          */
638         btree_csum_one_bio(bio);
639         return 0;
640 }
641
642 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
643                                  int mirror_num, unsigned long bio_flags)
644 {
645         /*
646          * when we're called for a write, we're already in the async
647          * submission context.  Just jump into btrfs_map_bio
648          */
649         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
650 }
651
652 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
653                                  int mirror_num, unsigned long bio_flags)
654 {
655         int ret;
656
657         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
658                                           bio, 1);
659         BUG_ON(ret);
660
661         if (!(rw & (1 << BIO_RW))) {
662                 /*
663                  * called for a read, do the setup so that checksum validation
664                  * can happen in the async kernel threads
665                  */
666                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
667                                      mirror_num, 0);
668         }
669
670         /*
671          * kthread helpers are used to submit writes so that checksumming
672          * can happen in parallel across all CPUs
673          */
674         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
675                                    inode, rw, bio, mirror_num, 0,
676                                    __btree_submit_bio_start,
677                                    __btree_submit_bio_done);
678 }
679
680 static int btree_writepage(struct page *page, struct writeback_control *wbc)
681 {
682         struct extent_io_tree *tree;
683         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
684         struct extent_buffer *eb;
685         int was_dirty;
686
687         tree = &BTRFS_I(page->mapping->host)->io_tree;
688         if (!(current->flags & PF_MEMALLOC)) {
689                 return extent_write_full_page(tree, page,
690                                               btree_get_extent, wbc);
691         }
692
693         redirty_page_for_writepage(wbc, page);
694         eb = btrfs_find_tree_block(root, page_offset(page),
695                                       PAGE_CACHE_SIZE);
696         WARN_ON(!eb);
697
698         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
699         if (!was_dirty) {
700                 spin_lock(&root->fs_info->delalloc_lock);
701                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
702                 spin_unlock(&root->fs_info->delalloc_lock);
703         }
704         free_extent_buffer(eb);
705
706         unlock_page(page);
707         return 0;
708 }
709
710 static int btree_writepages(struct address_space *mapping,
711                             struct writeback_control *wbc)
712 {
713         struct extent_io_tree *tree;
714         tree = &BTRFS_I(mapping->host)->io_tree;
715         if (wbc->sync_mode == WB_SYNC_NONE) {
716                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
717                 u64 num_dirty;
718                 unsigned long thresh = 32 * 1024 * 1024;
719
720                 if (wbc->for_kupdate)
721                         return 0;
722
723                 /* this is a bit racy, but that's ok */
724                 num_dirty = root->fs_info->dirty_metadata_bytes;
725                 if (num_dirty < thresh)
726                         return 0;
727         }
728         return extent_writepages(tree, mapping, btree_get_extent, wbc);
729 }
730
731 static int btree_readpage(struct file *file, struct page *page)
732 {
733         struct extent_io_tree *tree;
734         tree = &BTRFS_I(page->mapping->host)->io_tree;
735         return extent_read_full_page(tree, page, btree_get_extent);
736 }
737
738 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
739 {
740         struct extent_io_tree *tree;
741         struct extent_map_tree *map;
742         int ret;
743
744         if (PageWriteback(page) || PageDirty(page))
745                 return 0;
746
747         tree = &BTRFS_I(page->mapping->host)->io_tree;
748         map = &BTRFS_I(page->mapping->host)->extent_tree;
749
750         ret = try_release_extent_state(map, tree, page, gfp_flags);
751         if (!ret)
752                 return 0;
753
754         ret = try_release_extent_buffer(tree, page);
755         if (ret == 1) {
756                 ClearPagePrivate(page);
757                 set_page_private(page, 0);
758                 page_cache_release(page);
759         }
760
761         return ret;
762 }
763
764 static void btree_invalidatepage(struct page *page, unsigned long offset)
765 {
766         struct extent_io_tree *tree;
767         tree = &BTRFS_I(page->mapping->host)->io_tree;
768         extent_invalidatepage(tree, page, offset);
769         btree_releasepage(page, GFP_NOFS);
770         if (PagePrivate(page)) {
771                 printk(KERN_WARNING "btrfs warning page private not zero "
772                        "on page %llu\n", (unsigned long long)page_offset(page));
773                 ClearPagePrivate(page);
774                 set_page_private(page, 0);
775                 page_cache_release(page);
776         }
777 }
778
779 static const struct address_space_operations btree_aops = {
780         .readpage       = btree_readpage,
781         .writepage      = btree_writepage,
782         .writepages     = btree_writepages,
783         .releasepage    = btree_releasepage,
784         .invalidatepage = btree_invalidatepage,
785         .sync_page      = block_sync_page,
786 };
787
788 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
789                          u64 parent_transid)
790 {
791         struct extent_buffer *buf = NULL;
792         struct inode *btree_inode = root->fs_info->btree_inode;
793         int ret = 0;
794
795         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
796         if (!buf)
797                 return 0;
798         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
799                                  buf, 0, 0, btree_get_extent, 0);
800         free_extent_buffer(buf);
801         return ret;
802 }
803
804 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
805                                             u64 bytenr, u32 blocksize)
806 {
807         struct inode *btree_inode = root->fs_info->btree_inode;
808         struct extent_buffer *eb;
809         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
810                                 bytenr, blocksize, GFP_NOFS);
811         return eb;
812 }
813
814 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
815                                                  u64 bytenr, u32 blocksize)
816 {
817         struct inode *btree_inode = root->fs_info->btree_inode;
818         struct extent_buffer *eb;
819
820         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
821                                  bytenr, blocksize, NULL, GFP_NOFS);
822         return eb;
823 }
824
825
826 int btrfs_write_tree_block(struct extent_buffer *buf)
827 {
828         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
829                                         buf->start + buf->len - 1);
830 }
831
832 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
833 {
834         return filemap_fdatawait_range(buf->first_page->mapping,
835                                        buf->start, buf->start + buf->len - 1);
836 }
837
838 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
839                                       u32 blocksize, u64 parent_transid)
840 {
841         struct extent_buffer *buf = NULL;
842         struct inode *btree_inode = root->fs_info->btree_inode;
843         struct extent_io_tree *io_tree;
844         int ret;
845
846         io_tree = &BTRFS_I(btree_inode)->io_tree;
847
848         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
849         if (!buf)
850                 return NULL;
851
852         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
853
854         if (ret == 0)
855                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
856         return buf;
857
858 }
859
860 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
861                      struct extent_buffer *buf)
862 {
863         struct inode *btree_inode = root->fs_info->btree_inode;
864         if (btrfs_header_generation(buf) ==
865             root->fs_info->running_transaction->transid) {
866                 btrfs_assert_tree_locked(buf);
867
868                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
869                         spin_lock(&root->fs_info->delalloc_lock);
870                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
871                                 root->fs_info->dirty_metadata_bytes -= buf->len;
872                         else
873                                 WARN_ON(1);
874                         spin_unlock(&root->fs_info->delalloc_lock);
875                 }
876
877                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
878                 btrfs_set_lock_blocking(buf);
879                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
880                                           buf);
881         }
882         return 0;
883 }
884
885 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
886                         u32 stripesize, struct btrfs_root *root,
887                         struct btrfs_fs_info *fs_info,
888                         u64 objectid)
889 {
890         root->node = NULL;
891         root->commit_root = NULL;
892         root->sectorsize = sectorsize;
893         root->nodesize = nodesize;
894         root->leafsize = leafsize;
895         root->stripesize = stripesize;
896         root->ref_cows = 0;
897         root->track_dirty = 0;
898         root->in_radix = 0;
899         root->clean_orphans = 0;
900
901         root->fs_info = fs_info;
902         root->objectid = objectid;
903         root->last_trans = 0;
904         root->highest_objectid = 0;
905         root->name = NULL;
906         root->in_sysfs = 0;
907         root->inode_tree = RB_ROOT;
908
909         INIT_LIST_HEAD(&root->dirty_list);
910         INIT_LIST_HEAD(&root->orphan_list);
911         INIT_LIST_HEAD(&root->root_list);
912         spin_lock_init(&root->node_lock);
913         spin_lock_init(&root->list_lock);
914         spin_lock_init(&root->inode_lock);
915         mutex_init(&root->objectid_mutex);
916         mutex_init(&root->log_mutex);
917         init_waitqueue_head(&root->log_writer_wait);
918         init_waitqueue_head(&root->log_commit_wait[0]);
919         init_waitqueue_head(&root->log_commit_wait[1]);
920         atomic_set(&root->log_commit[0], 0);
921         atomic_set(&root->log_commit[1], 0);
922         atomic_set(&root->log_writers, 0);
923         root->log_batch = 0;
924         root->log_transid = 0;
925         root->last_log_commit = 0;
926         extent_io_tree_init(&root->dirty_log_pages,
927                              fs_info->btree_inode->i_mapping, GFP_NOFS);
928
929         memset(&root->root_key, 0, sizeof(root->root_key));
930         memset(&root->root_item, 0, sizeof(root->root_item));
931         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
932         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
933         root->defrag_trans_start = fs_info->generation;
934         init_completion(&root->kobj_unregister);
935         root->defrag_running = 0;
936         root->root_key.objectid = objectid;
937         root->anon_super.s_root = NULL;
938         root->anon_super.s_dev = 0;
939         INIT_LIST_HEAD(&root->anon_super.s_list);
940         INIT_LIST_HEAD(&root->anon_super.s_instances);
941         init_rwsem(&root->anon_super.s_umount);
942
943         return 0;
944 }
945
946 static int find_and_setup_root(struct btrfs_root *tree_root,
947                                struct btrfs_fs_info *fs_info,
948                                u64 objectid,
949                                struct btrfs_root *root)
950 {
951         int ret;
952         u32 blocksize;
953         u64 generation;
954
955         __setup_root(tree_root->nodesize, tree_root->leafsize,
956                      tree_root->sectorsize, tree_root->stripesize,
957                      root, fs_info, objectid);
958         ret = btrfs_find_last_root(tree_root, objectid,
959                                    &root->root_item, &root->root_key);
960         if (ret > 0)
961                 return -ENOENT;
962         BUG_ON(ret);
963
964         generation = btrfs_root_generation(&root->root_item);
965         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
966         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
967                                      blocksize, generation);
968         BUG_ON(!root->node);
969         root->commit_root = btrfs_root_node(root);
970         return 0;
971 }
972
973 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
974                              struct btrfs_fs_info *fs_info)
975 {
976         struct extent_buffer *eb;
977         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
978         u64 start = 0;
979         u64 end = 0;
980         int ret;
981
982         if (!log_root_tree)
983                 return 0;
984
985         while (1) {
986                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
987                                 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
988                 if (ret)
989                         break;
990
991                 clear_extent_bits(&log_root_tree->dirty_log_pages, start, end,
992                                   EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
993         }
994         eb = fs_info->log_root_tree->node;
995
996         WARN_ON(btrfs_header_level(eb) != 0);
997         WARN_ON(btrfs_header_nritems(eb) != 0);
998
999         ret = btrfs_free_reserved_extent(fs_info->tree_root,
1000                                 eb->start, eb->len);
1001         BUG_ON(ret);
1002
1003         free_extent_buffer(eb);
1004         kfree(fs_info->log_root_tree);
1005         fs_info->log_root_tree = NULL;
1006         return 0;
1007 }
1008
1009 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1010                                          struct btrfs_fs_info *fs_info)
1011 {
1012         struct btrfs_root *root;
1013         struct btrfs_root *tree_root = fs_info->tree_root;
1014         struct extent_buffer *leaf;
1015
1016         root = kzalloc(sizeof(*root), GFP_NOFS);
1017         if (!root)
1018                 return ERR_PTR(-ENOMEM);
1019
1020         __setup_root(tree_root->nodesize, tree_root->leafsize,
1021                      tree_root->sectorsize, tree_root->stripesize,
1022                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1023
1024         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1025         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1026         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1027         /*
1028          * log trees do not get reference counted because they go away
1029          * before a real commit is actually done.  They do store pointers
1030          * to file data extents, and those reference counts still get
1031          * updated (along with back refs to the log tree).
1032          */
1033         root->ref_cows = 0;
1034
1035         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1036                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1037         if (IS_ERR(leaf)) {
1038                 kfree(root);
1039                 return ERR_CAST(leaf);
1040         }
1041
1042         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1043         btrfs_set_header_bytenr(leaf, leaf->start);
1044         btrfs_set_header_generation(leaf, trans->transid);
1045         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1046         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1047         root->node = leaf;
1048
1049         write_extent_buffer(root->node, root->fs_info->fsid,
1050                             (unsigned long)btrfs_header_fsid(root->node),
1051                             BTRFS_FSID_SIZE);
1052         btrfs_mark_buffer_dirty(root->node);
1053         btrfs_tree_unlock(root->node);
1054         return root;
1055 }
1056
1057 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1058                              struct btrfs_fs_info *fs_info)
1059 {
1060         struct btrfs_root *log_root;
1061
1062         log_root = alloc_log_tree(trans, fs_info);
1063         if (IS_ERR(log_root))
1064                 return PTR_ERR(log_root);
1065         WARN_ON(fs_info->log_root_tree);
1066         fs_info->log_root_tree = log_root;
1067         return 0;
1068 }
1069
1070 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1071                        struct btrfs_root *root)
1072 {
1073         struct btrfs_root *log_root;
1074         struct btrfs_inode_item *inode_item;
1075
1076         log_root = alloc_log_tree(trans, root->fs_info);
1077         if (IS_ERR(log_root))
1078                 return PTR_ERR(log_root);
1079
1080         log_root->last_trans = trans->transid;
1081         log_root->root_key.offset = root->root_key.objectid;
1082
1083         inode_item = &log_root->root_item.inode;
1084         inode_item->generation = cpu_to_le64(1);
1085         inode_item->size = cpu_to_le64(3);
1086         inode_item->nlink = cpu_to_le32(1);
1087         inode_item->nbytes = cpu_to_le64(root->leafsize);
1088         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1089
1090         btrfs_set_root_node(&log_root->root_item, log_root->node);
1091
1092         WARN_ON(root->log_root);
1093         root->log_root = log_root;
1094         root->log_transid = 0;
1095         root->last_log_commit = 0;
1096         return 0;
1097 }
1098
1099 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1100                                                struct btrfs_key *location)
1101 {
1102         struct btrfs_root *root;
1103         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1104         struct btrfs_path *path;
1105         struct extent_buffer *l;
1106         u64 generation;
1107         u32 blocksize;
1108         int ret = 0;
1109
1110         root = kzalloc(sizeof(*root), GFP_NOFS);
1111         if (!root)
1112                 return ERR_PTR(-ENOMEM);
1113         if (location->offset == (u64)-1) {
1114                 ret = find_and_setup_root(tree_root, fs_info,
1115                                           location->objectid, root);
1116                 if (ret) {
1117                         kfree(root);
1118                         return ERR_PTR(ret);
1119                 }
1120                 goto out;
1121         }
1122
1123         __setup_root(tree_root->nodesize, tree_root->leafsize,
1124                      tree_root->sectorsize, tree_root->stripesize,
1125                      root, fs_info, location->objectid);
1126
1127         path = btrfs_alloc_path();
1128         BUG_ON(!path);
1129         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1130         if (ret == 0) {
1131                 l = path->nodes[0];
1132                 read_extent_buffer(l, &root->root_item,
1133                                 btrfs_item_ptr_offset(l, path->slots[0]),
1134                                 sizeof(root->root_item));
1135                 memcpy(&root->root_key, location, sizeof(*location));
1136         }
1137         btrfs_free_path(path);
1138         if (ret) {
1139                 if (ret > 0)
1140                         ret = -ENOENT;
1141                 return ERR_PTR(ret);
1142         }
1143
1144         generation = btrfs_root_generation(&root->root_item);
1145         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1146         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1147                                      blocksize, generation);
1148         root->commit_root = btrfs_root_node(root);
1149         BUG_ON(!root->node);
1150 out:
1151         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1152                 root->ref_cows = 1;
1153
1154         return root;
1155 }
1156
1157 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1158                                         u64 root_objectid)
1159 {
1160         struct btrfs_root *root;
1161
1162         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1163                 return fs_info->tree_root;
1164         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1165                 return fs_info->extent_root;
1166
1167         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1168                                  (unsigned long)root_objectid);
1169         return root;
1170 }
1171
1172 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1173                                               struct btrfs_key *location)
1174 {
1175         struct btrfs_root *root;
1176         int ret;
1177
1178         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1179                 return fs_info->tree_root;
1180         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1181                 return fs_info->extent_root;
1182         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1183                 return fs_info->chunk_root;
1184         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1185                 return fs_info->dev_root;
1186         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1187                 return fs_info->csum_root;
1188 again:
1189         spin_lock(&fs_info->fs_roots_radix_lock);
1190         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1191                                  (unsigned long)location->objectid);
1192         spin_unlock(&fs_info->fs_roots_radix_lock);
1193         if (root)
1194                 return root;
1195
1196         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1197         if (ret == 0)
1198                 ret = -ENOENT;
1199         if (ret < 0)
1200                 return ERR_PTR(ret);
1201
1202         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1203         if (IS_ERR(root))
1204                 return root;
1205
1206         WARN_ON(btrfs_root_refs(&root->root_item) == 0);
1207         set_anon_super(&root->anon_super, NULL);
1208
1209         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1210         if (ret)
1211                 goto fail;
1212
1213         spin_lock(&fs_info->fs_roots_radix_lock);
1214         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1215                                 (unsigned long)root->root_key.objectid,
1216                                 root);
1217         if (ret == 0) {
1218                 root->in_radix = 1;
1219                 root->clean_orphans = 1;
1220         }
1221         spin_unlock(&fs_info->fs_roots_radix_lock);
1222         radix_tree_preload_end();
1223         if (ret) {
1224                 if (ret == -EEXIST) {
1225                         free_fs_root(root);
1226                         goto again;
1227                 }
1228                 goto fail;
1229         }
1230
1231         ret = btrfs_find_dead_roots(fs_info->tree_root,
1232                                     root->root_key.objectid);
1233         WARN_ON(ret);
1234         return root;
1235 fail:
1236         free_fs_root(root);
1237         return ERR_PTR(ret);
1238 }
1239
1240 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1241                                       struct btrfs_key *location,
1242                                       const char *name, int namelen)
1243 {
1244         return btrfs_read_fs_root_no_name(fs_info, location);
1245 #if 0
1246         struct btrfs_root *root;
1247         int ret;
1248
1249         root = btrfs_read_fs_root_no_name(fs_info, location);
1250         if (!root)
1251                 return NULL;
1252
1253         if (root->in_sysfs)
1254                 return root;
1255
1256         ret = btrfs_set_root_name(root, name, namelen);
1257         if (ret) {
1258                 free_extent_buffer(root->node);
1259                 kfree(root);
1260                 return ERR_PTR(ret);
1261         }
1262
1263         ret = btrfs_sysfs_add_root(root);
1264         if (ret) {
1265                 free_extent_buffer(root->node);
1266                 kfree(root->name);
1267                 kfree(root);
1268                 return ERR_PTR(ret);
1269         }
1270         root->in_sysfs = 1;
1271         return root;
1272 #endif
1273 }
1274
1275 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1276 {
1277         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1278         int ret = 0;
1279         struct btrfs_device *device;
1280         struct backing_dev_info *bdi;
1281
1282         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1283                 if (!device->bdev)
1284                         continue;
1285                 bdi = blk_get_backing_dev_info(device->bdev);
1286                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1287                         ret = 1;
1288                         break;
1289                 }
1290         }
1291         return ret;
1292 }
1293
1294 /*
1295  * this unplugs every device on the box, and it is only used when page
1296  * is null
1297  */
1298 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1299 {
1300         struct btrfs_device *device;
1301         struct btrfs_fs_info *info;
1302
1303         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1304         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1305                 if (!device->bdev)
1306                         continue;
1307
1308                 bdi = blk_get_backing_dev_info(device->bdev);
1309                 if (bdi->unplug_io_fn)
1310                         bdi->unplug_io_fn(bdi, page);
1311         }
1312 }
1313
1314 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1315 {
1316         struct inode *inode;
1317         struct extent_map_tree *em_tree;
1318         struct extent_map *em;
1319         struct address_space *mapping;
1320         u64 offset;
1321
1322         /* the generic O_DIRECT read code does this */
1323         if (1 || !page) {
1324                 __unplug_io_fn(bdi, page);
1325                 return;
1326         }
1327
1328         /*
1329          * page->mapping may change at any time.  Get a consistent copy
1330          * and use that for everything below
1331          */
1332         smp_mb();
1333         mapping = page->mapping;
1334         if (!mapping)
1335                 return;
1336
1337         inode = mapping->host;
1338
1339         /*
1340          * don't do the expensive searching for a small number of
1341          * devices
1342          */
1343         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1344                 __unplug_io_fn(bdi, page);
1345                 return;
1346         }
1347
1348         offset = page_offset(page);
1349
1350         em_tree = &BTRFS_I(inode)->extent_tree;
1351         read_lock(&em_tree->lock);
1352         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1353         read_unlock(&em_tree->lock);
1354         if (!em) {
1355                 __unplug_io_fn(bdi, page);
1356                 return;
1357         }
1358
1359         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1360                 free_extent_map(em);
1361                 __unplug_io_fn(bdi, page);
1362                 return;
1363         }
1364         offset = offset - em->start;
1365         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1366                           em->block_start + offset, page);
1367         free_extent_map(em);
1368 }
1369
1370 /*
1371  * If this fails, caller must call bdi_destroy() to get rid of the
1372  * bdi again.
1373  */
1374 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1375 {
1376         int err;
1377
1378         bdi->name = "btrfs";
1379         bdi->capabilities = BDI_CAP_MAP_COPY;
1380         err = bdi_init(bdi);
1381         if (err)
1382                 return err;
1383
1384         err = bdi_register(bdi, NULL, "btrfs-%d",
1385                                 atomic_inc_return(&btrfs_bdi_num));
1386         if (err) {
1387                 bdi_destroy(bdi);
1388                 return err;
1389         }
1390
1391         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1392         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1393         bdi->unplug_io_data     = info;
1394         bdi->congested_fn       = btrfs_congested_fn;
1395         bdi->congested_data     = info;
1396         return 0;
1397 }
1398
1399 static int bio_ready_for_csum(struct bio *bio)
1400 {
1401         u64 length = 0;
1402         u64 buf_len = 0;
1403         u64 start = 0;
1404         struct page *page;
1405         struct extent_io_tree *io_tree = NULL;
1406         struct btrfs_fs_info *info = NULL;
1407         struct bio_vec *bvec;
1408         int i;
1409         int ret;
1410
1411         bio_for_each_segment(bvec, bio, i) {
1412                 page = bvec->bv_page;
1413                 if (page->private == EXTENT_PAGE_PRIVATE) {
1414                         length += bvec->bv_len;
1415                         continue;
1416                 }
1417                 if (!page->private) {
1418                         length += bvec->bv_len;
1419                         continue;
1420                 }
1421                 length = bvec->bv_len;
1422                 buf_len = page->private >> 2;
1423                 start = page_offset(page) + bvec->bv_offset;
1424                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1425                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1426         }
1427         /* are we fully contained in this bio? */
1428         if (buf_len <= length)
1429                 return 1;
1430
1431         ret = extent_range_uptodate(io_tree, start + length,
1432                                     start + buf_len - 1);
1433         return ret;
1434 }
1435
1436 /*
1437  * called by the kthread helper functions to finally call the bio end_io
1438  * functions.  This is where read checksum verification actually happens
1439  */
1440 static void end_workqueue_fn(struct btrfs_work *work)
1441 {
1442         struct bio *bio;
1443         struct end_io_wq *end_io_wq;
1444         struct btrfs_fs_info *fs_info;
1445         int error;
1446
1447         end_io_wq = container_of(work, struct end_io_wq, work);
1448         bio = end_io_wq->bio;
1449         fs_info = end_io_wq->info;
1450
1451         /* metadata bio reads are special because the whole tree block must
1452          * be checksummed at once.  This makes sure the entire block is in
1453          * ram and up to date before trying to verify things.  For
1454          * blocksize <= pagesize, it is basically a noop
1455          */
1456         if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1457             !bio_ready_for_csum(bio)) {
1458                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1459                                    &end_io_wq->work);
1460                 return;
1461         }
1462         error = end_io_wq->error;
1463         bio->bi_private = end_io_wq->private;
1464         bio->bi_end_io = end_io_wq->end_io;
1465         kfree(end_io_wq);
1466         bio_endio(bio, error);
1467 }
1468
1469 static int cleaner_kthread(void *arg)
1470 {
1471         struct btrfs_root *root = arg;
1472
1473         do {
1474                 smp_mb();
1475                 if (root->fs_info->closing)
1476                         break;
1477
1478                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1479
1480                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1481                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1482                         btrfs_run_delayed_iputs(root);
1483                         btrfs_clean_old_snapshots(root);
1484                         mutex_unlock(&root->fs_info->cleaner_mutex);
1485                 }
1486
1487                 if (freezing(current)) {
1488                         refrigerator();
1489                 } else {
1490                         smp_mb();
1491                         if (root->fs_info->closing)
1492                                 break;
1493                         set_current_state(TASK_INTERRUPTIBLE);
1494                         schedule();
1495                         __set_current_state(TASK_RUNNING);
1496                 }
1497         } while (!kthread_should_stop());
1498         return 0;
1499 }
1500
1501 static int transaction_kthread(void *arg)
1502 {
1503         struct btrfs_root *root = arg;
1504         struct btrfs_trans_handle *trans;
1505         struct btrfs_transaction *cur;
1506         unsigned long now;
1507         unsigned long delay;
1508         int ret;
1509
1510         do {
1511                 smp_mb();
1512                 if (root->fs_info->closing)
1513                         break;
1514
1515                 delay = HZ * 30;
1516                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1517                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1518
1519                 mutex_lock(&root->fs_info->trans_mutex);
1520                 cur = root->fs_info->running_transaction;
1521                 if (!cur) {
1522                         mutex_unlock(&root->fs_info->trans_mutex);
1523                         goto sleep;
1524                 }
1525
1526                 now = get_seconds();
1527                 if (now < cur->start_time || now - cur->start_time < 30) {
1528                         mutex_unlock(&root->fs_info->trans_mutex);
1529                         delay = HZ * 5;
1530                         goto sleep;
1531                 }
1532                 mutex_unlock(&root->fs_info->trans_mutex);
1533                 trans = btrfs_start_transaction(root, 1);
1534                 ret = btrfs_commit_transaction(trans, root);
1535
1536 sleep:
1537                 wake_up_process(root->fs_info->cleaner_kthread);
1538                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1539
1540                 if (freezing(current)) {
1541                         refrigerator();
1542                 } else {
1543                         if (root->fs_info->closing)
1544                                 break;
1545                         set_current_state(TASK_INTERRUPTIBLE);
1546                         schedule_timeout(delay);
1547                         __set_current_state(TASK_RUNNING);
1548                 }
1549         } while (!kthread_should_stop());
1550         return 0;
1551 }
1552
1553 struct btrfs_root *open_ctree(struct super_block *sb,
1554                               struct btrfs_fs_devices *fs_devices,
1555                               char *options)
1556 {
1557         u32 sectorsize;
1558         u32 nodesize;
1559         u32 leafsize;
1560         u32 blocksize;
1561         u32 stripesize;
1562         u64 generation;
1563         u64 features;
1564         struct btrfs_key location;
1565         struct buffer_head *bh;
1566         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1567                                                  GFP_NOFS);
1568         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1569                                                  GFP_NOFS);
1570         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1571                                                GFP_NOFS);
1572         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1573                                                 GFP_NOFS);
1574         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1575                                                 GFP_NOFS);
1576         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1577                                               GFP_NOFS);
1578         struct btrfs_root *log_tree_root;
1579
1580         int ret;
1581         int err = -EINVAL;
1582
1583         struct btrfs_super_block *disk_super;
1584
1585         if (!extent_root || !tree_root || !fs_info ||
1586             !chunk_root || !dev_root || !csum_root) {
1587                 err = -ENOMEM;
1588                 goto fail;
1589         }
1590
1591         ret = init_srcu_struct(&fs_info->subvol_srcu);
1592         if (ret) {
1593                 err = ret;
1594                 goto fail;
1595         }
1596
1597         ret = setup_bdi(fs_info, &fs_info->bdi);
1598         if (ret) {
1599                 err = ret;
1600                 goto fail_srcu;
1601         }
1602
1603         fs_info->btree_inode = new_inode(sb);
1604         if (!fs_info->btree_inode) {
1605                 err = -ENOMEM;
1606                 goto fail_bdi;
1607         }
1608
1609         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1610         INIT_LIST_HEAD(&fs_info->trans_list);
1611         INIT_LIST_HEAD(&fs_info->dead_roots);
1612         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1613         INIT_LIST_HEAD(&fs_info->hashers);
1614         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1615         INIT_LIST_HEAD(&fs_info->ordered_operations);
1616         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1617         spin_lock_init(&fs_info->delalloc_lock);
1618         spin_lock_init(&fs_info->new_trans_lock);
1619         spin_lock_init(&fs_info->ref_cache_lock);
1620         spin_lock_init(&fs_info->fs_roots_radix_lock);
1621         spin_lock_init(&fs_info->delayed_iput_lock);
1622
1623         init_completion(&fs_info->kobj_unregister);
1624         fs_info->tree_root = tree_root;
1625         fs_info->extent_root = extent_root;
1626         fs_info->csum_root = csum_root;
1627         fs_info->chunk_root = chunk_root;
1628         fs_info->dev_root = dev_root;
1629         fs_info->fs_devices = fs_devices;
1630         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1631         INIT_LIST_HEAD(&fs_info->space_info);
1632         btrfs_mapping_init(&fs_info->mapping_tree);
1633         atomic_set(&fs_info->nr_async_submits, 0);
1634         atomic_set(&fs_info->async_delalloc_pages, 0);
1635         atomic_set(&fs_info->async_submit_draining, 0);
1636         atomic_set(&fs_info->nr_async_bios, 0);
1637         fs_info->sb = sb;
1638         fs_info->max_inline = 8192 * 1024;
1639         fs_info->metadata_ratio = 0;
1640
1641         fs_info->thread_pool_size = min_t(unsigned long,
1642                                           num_online_cpus() + 2, 8);
1643
1644         INIT_LIST_HEAD(&fs_info->ordered_extents);
1645         spin_lock_init(&fs_info->ordered_extent_lock);
1646
1647         sb->s_blocksize = 4096;
1648         sb->s_blocksize_bits = blksize_bits(4096);
1649         sb->s_bdi = &fs_info->bdi;
1650
1651         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1652         fs_info->btree_inode->i_nlink = 1;
1653         /*
1654          * we set the i_size on the btree inode to the max possible int.
1655          * the real end of the address space is determined by all of
1656          * the devices in the system
1657          */
1658         fs_info->btree_inode->i_size = OFFSET_MAX;
1659         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1660         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1661
1662         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1663         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1664                              fs_info->btree_inode->i_mapping,
1665                              GFP_NOFS);
1666         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1667                              GFP_NOFS);
1668
1669         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1670
1671         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1672         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1673                sizeof(struct btrfs_key));
1674         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1675         insert_inode_hash(fs_info->btree_inode);
1676
1677         spin_lock_init(&fs_info->block_group_cache_lock);
1678         fs_info->block_group_cache_tree = RB_ROOT;
1679
1680         extent_io_tree_init(&fs_info->freed_extents[0],
1681                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1682         extent_io_tree_init(&fs_info->freed_extents[1],
1683                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1684         fs_info->pinned_extents = &fs_info->freed_extents[0];
1685         fs_info->do_barriers = 1;
1686
1687
1688         mutex_init(&fs_info->trans_mutex);
1689         mutex_init(&fs_info->ordered_operations_mutex);
1690         mutex_init(&fs_info->tree_log_mutex);
1691         mutex_init(&fs_info->chunk_mutex);
1692         mutex_init(&fs_info->transaction_kthread_mutex);
1693         mutex_init(&fs_info->cleaner_mutex);
1694         mutex_init(&fs_info->volume_mutex);
1695         init_rwsem(&fs_info->extent_commit_sem);
1696         init_rwsem(&fs_info->cleanup_work_sem);
1697         init_rwsem(&fs_info->subvol_sem);
1698
1699         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1700         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1701
1702         init_waitqueue_head(&fs_info->transaction_throttle);
1703         init_waitqueue_head(&fs_info->transaction_wait);
1704         init_waitqueue_head(&fs_info->async_submit_wait);
1705
1706         __setup_root(4096, 4096, 4096, 4096, tree_root,
1707                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1708
1709
1710         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1711         if (!bh)
1712                 goto fail_iput;
1713
1714         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1715         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1716                sizeof(fs_info->super_for_commit));
1717         brelse(bh);
1718
1719         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1720
1721         disk_super = &fs_info->super_copy;
1722         if (!btrfs_super_root(disk_super))
1723                 goto fail_iput;
1724
1725         ret = btrfs_parse_options(tree_root, options);
1726         if (ret) {
1727                 err = ret;
1728                 goto fail_iput;
1729         }
1730
1731         features = btrfs_super_incompat_flags(disk_super) &
1732                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1733         if (features) {
1734                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1735                        "unsupported optional features (%Lx).\n",
1736                        (unsigned long long)features);
1737                 err = -EINVAL;
1738                 goto fail_iput;
1739         }
1740
1741         features = btrfs_super_incompat_flags(disk_super);
1742         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1743                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1744                 btrfs_set_super_incompat_flags(disk_super, features);
1745         }
1746
1747         features = btrfs_super_compat_ro_flags(disk_super) &
1748                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1749         if (!(sb->s_flags & MS_RDONLY) && features) {
1750                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1751                        "unsupported option features (%Lx).\n",
1752                        (unsigned long long)features);
1753                 err = -EINVAL;
1754                 goto fail_iput;
1755         }
1756
1757         btrfs_init_workers(&fs_info->generic_worker,
1758                            "genwork", 1, NULL);
1759
1760         btrfs_init_workers(&fs_info->workers, "worker",
1761                            fs_info->thread_pool_size,
1762                            &fs_info->generic_worker);
1763
1764         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1765                            fs_info->thread_pool_size,
1766                            &fs_info->generic_worker);
1767
1768         btrfs_init_workers(&fs_info->submit_workers, "submit",
1769                            min_t(u64, fs_devices->num_devices,
1770                            fs_info->thread_pool_size),
1771                            &fs_info->generic_worker);
1772         btrfs_init_workers(&fs_info->enospc_workers, "enospc",
1773                            fs_info->thread_pool_size,
1774                            &fs_info->generic_worker);
1775
1776         /* a higher idle thresh on the submit workers makes it much more
1777          * likely that bios will be send down in a sane order to the
1778          * devices
1779          */
1780         fs_info->submit_workers.idle_thresh = 64;
1781
1782         fs_info->workers.idle_thresh = 16;
1783         fs_info->workers.ordered = 1;
1784
1785         fs_info->delalloc_workers.idle_thresh = 2;
1786         fs_info->delalloc_workers.ordered = 1;
1787
1788         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1789                            &fs_info->generic_worker);
1790         btrfs_init_workers(&fs_info->endio_workers, "endio",
1791                            fs_info->thread_pool_size,
1792                            &fs_info->generic_worker);
1793         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1794                            fs_info->thread_pool_size,
1795                            &fs_info->generic_worker);
1796         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1797                            "endio-meta-write", fs_info->thread_pool_size,
1798                            &fs_info->generic_worker);
1799         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1800                            fs_info->thread_pool_size,
1801                            &fs_info->generic_worker);
1802
1803         /*
1804          * endios are largely parallel and should have a very
1805          * low idle thresh
1806          */
1807         fs_info->endio_workers.idle_thresh = 4;
1808         fs_info->endio_meta_workers.idle_thresh = 4;
1809
1810         fs_info->endio_write_workers.idle_thresh = 2;
1811         fs_info->endio_meta_write_workers.idle_thresh = 2;
1812
1813         btrfs_start_workers(&fs_info->workers, 1);
1814         btrfs_start_workers(&fs_info->generic_worker, 1);
1815         btrfs_start_workers(&fs_info->submit_workers, 1);
1816         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1817         btrfs_start_workers(&fs_info->fixup_workers, 1);
1818         btrfs_start_workers(&fs_info->endio_workers, 1);
1819         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1820         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1821         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1822         btrfs_start_workers(&fs_info->enospc_workers, 1);
1823
1824         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1825         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1826                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1827
1828         nodesize = btrfs_super_nodesize(disk_super);
1829         leafsize = btrfs_super_leafsize(disk_super);
1830         sectorsize = btrfs_super_sectorsize(disk_super);
1831         stripesize = btrfs_super_stripesize(disk_super);
1832         tree_root->nodesize = nodesize;
1833         tree_root->leafsize = leafsize;
1834         tree_root->sectorsize = sectorsize;
1835         tree_root->stripesize = stripesize;
1836
1837         sb->s_blocksize = sectorsize;
1838         sb->s_blocksize_bits = blksize_bits(sectorsize);
1839
1840         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1841                     sizeof(disk_super->magic))) {
1842                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1843                 goto fail_sb_buffer;
1844         }
1845
1846         mutex_lock(&fs_info->chunk_mutex);
1847         ret = btrfs_read_sys_array(tree_root);
1848         mutex_unlock(&fs_info->chunk_mutex);
1849         if (ret) {
1850                 printk(KERN_WARNING "btrfs: failed to read the system "
1851                        "array on %s\n", sb->s_id);
1852                 goto fail_sb_buffer;
1853         }
1854
1855         blocksize = btrfs_level_size(tree_root,
1856                                      btrfs_super_chunk_root_level(disk_super));
1857         generation = btrfs_super_chunk_root_generation(disk_super);
1858
1859         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1860                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1861
1862         chunk_root->node = read_tree_block(chunk_root,
1863                                            btrfs_super_chunk_root(disk_super),
1864                                            blocksize, generation);
1865         BUG_ON(!chunk_root->node);
1866         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1867                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1868                        sb->s_id);
1869                 goto fail_chunk_root;
1870         }
1871         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1872         chunk_root->commit_root = btrfs_root_node(chunk_root);
1873
1874         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1875            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1876            BTRFS_UUID_SIZE);
1877
1878         mutex_lock(&fs_info->chunk_mutex);
1879         ret = btrfs_read_chunk_tree(chunk_root);
1880         mutex_unlock(&fs_info->chunk_mutex);
1881         if (ret) {
1882                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1883                        sb->s_id);
1884                 goto fail_chunk_root;
1885         }
1886
1887         btrfs_close_extra_devices(fs_devices);
1888
1889         blocksize = btrfs_level_size(tree_root,
1890                                      btrfs_super_root_level(disk_super));
1891         generation = btrfs_super_generation(disk_super);
1892
1893         tree_root->node = read_tree_block(tree_root,
1894                                           btrfs_super_root(disk_super),
1895                                           blocksize, generation);
1896         if (!tree_root->node)
1897                 goto fail_chunk_root;
1898         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1899                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1900                        sb->s_id);
1901                 goto fail_tree_root;
1902         }
1903         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1904         tree_root->commit_root = btrfs_root_node(tree_root);
1905
1906         ret = find_and_setup_root(tree_root, fs_info,
1907                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1908         if (ret)
1909                 goto fail_tree_root;
1910         extent_root->track_dirty = 1;
1911
1912         ret = find_and_setup_root(tree_root, fs_info,
1913                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1914         if (ret)
1915                 goto fail_extent_root;
1916         dev_root->track_dirty = 1;
1917
1918         ret = find_and_setup_root(tree_root, fs_info,
1919                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1920         if (ret)
1921                 goto fail_dev_root;
1922
1923         csum_root->track_dirty = 1;
1924
1925         ret = btrfs_read_block_groups(extent_root);
1926         if (ret) {
1927                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1928                 goto fail_block_groups;
1929         }
1930
1931         fs_info->generation = generation;
1932         fs_info->last_trans_committed = generation;
1933         fs_info->data_alloc_profile = (u64)-1;
1934         fs_info->metadata_alloc_profile = (u64)-1;
1935         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1936         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1937                                                "btrfs-cleaner");
1938         if (IS_ERR(fs_info->cleaner_kthread))
1939                 goto fail_block_groups;
1940
1941         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1942                                                    tree_root,
1943                                                    "btrfs-transaction");
1944         if (IS_ERR(fs_info->transaction_kthread))
1945                 goto fail_cleaner;
1946
1947         if (!btrfs_test_opt(tree_root, SSD) &&
1948             !btrfs_test_opt(tree_root, NOSSD) &&
1949             !fs_info->fs_devices->rotating) {
1950                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1951                        "mode\n");
1952                 btrfs_set_opt(fs_info->mount_opt, SSD);
1953         }
1954
1955         if (btrfs_super_log_root(disk_super) != 0) {
1956                 u64 bytenr = btrfs_super_log_root(disk_super);
1957
1958                 if (fs_devices->rw_devices == 0) {
1959                         printk(KERN_WARNING "Btrfs log replay required "
1960                                "on RO media\n");
1961                         err = -EIO;
1962                         goto fail_trans_kthread;
1963                 }
1964                 blocksize =
1965                      btrfs_level_size(tree_root,
1966                                       btrfs_super_log_root_level(disk_super));
1967
1968                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1969                                                       GFP_NOFS);
1970
1971                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1972                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1973
1974                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1975                                                       blocksize,
1976                                                       generation + 1);
1977                 ret = btrfs_recover_log_trees(log_tree_root);
1978                 BUG_ON(ret);
1979
1980                 if (sb->s_flags & MS_RDONLY) {
1981                         ret =  btrfs_commit_super(tree_root);
1982                         BUG_ON(ret);
1983                 }
1984         }
1985
1986         ret = btrfs_find_orphan_roots(tree_root);
1987         BUG_ON(ret);
1988
1989         if (!(sb->s_flags & MS_RDONLY)) {
1990                 ret = btrfs_recover_relocation(tree_root);
1991                 if (ret < 0) {
1992                         printk(KERN_WARNING
1993                                "btrfs: failed to recover relocation\n");
1994                         err = -EINVAL;
1995                         goto fail_trans_kthread;
1996                 }
1997         }
1998
1999         location.objectid = BTRFS_FS_TREE_OBJECTID;
2000         location.type = BTRFS_ROOT_ITEM_KEY;
2001         location.offset = (u64)-1;
2002
2003         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2004         if (!fs_info->fs_root)
2005                 goto fail_trans_kthread;
2006
2007         if (!(sb->s_flags & MS_RDONLY)) {
2008                 down_read(&fs_info->cleanup_work_sem);
2009                 btrfs_orphan_cleanup(fs_info->fs_root);
2010                 up_read(&fs_info->cleanup_work_sem);
2011         }
2012
2013         return tree_root;
2014
2015 fail_trans_kthread:
2016         kthread_stop(fs_info->transaction_kthread);
2017 fail_cleaner:
2018         kthread_stop(fs_info->cleaner_kthread);
2019
2020         /*
2021          * make sure we're done with the btree inode before we stop our
2022          * kthreads
2023          */
2024         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2025         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2026
2027 fail_block_groups:
2028         btrfs_free_block_groups(fs_info);
2029         free_extent_buffer(csum_root->node);
2030         free_extent_buffer(csum_root->commit_root);
2031 fail_dev_root:
2032         free_extent_buffer(dev_root->node);
2033         free_extent_buffer(dev_root->commit_root);
2034 fail_extent_root:
2035         free_extent_buffer(extent_root->node);
2036         free_extent_buffer(extent_root->commit_root);
2037 fail_tree_root:
2038         free_extent_buffer(tree_root->node);
2039         free_extent_buffer(tree_root->commit_root);
2040 fail_chunk_root:
2041         free_extent_buffer(chunk_root->node);
2042         free_extent_buffer(chunk_root->commit_root);
2043 fail_sb_buffer:
2044         btrfs_stop_workers(&fs_info->generic_worker);
2045         btrfs_stop_workers(&fs_info->fixup_workers);
2046         btrfs_stop_workers(&fs_info->delalloc_workers);
2047         btrfs_stop_workers(&fs_info->workers);
2048         btrfs_stop_workers(&fs_info->endio_workers);
2049         btrfs_stop_workers(&fs_info->endio_meta_workers);
2050         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2051         btrfs_stop_workers(&fs_info->endio_write_workers);
2052         btrfs_stop_workers(&fs_info->submit_workers);
2053         btrfs_stop_workers(&fs_info->enospc_workers);
2054 fail_iput:
2055         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2056         iput(fs_info->btree_inode);
2057
2058         btrfs_close_devices(fs_info->fs_devices);
2059         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2060 fail_bdi:
2061         bdi_destroy(&fs_info->bdi);
2062 fail_srcu:
2063         cleanup_srcu_struct(&fs_info->subvol_srcu);
2064 fail:
2065         kfree(extent_root);
2066         kfree(tree_root);
2067         kfree(fs_info);
2068         kfree(chunk_root);
2069         kfree(dev_root);
2070         kfree(csum_root);
2071         return ERR_PTR(err);
2072 }
2073
2074 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2075 {
2076         char b[BDEVNAME_SIZE];
2077
2078         if (uptodate) {
2079                 set_buffer_uptodate(bh);
2080         } else {
2081                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2082                         printk(KERN_WARNING "lost page write due to "
2083                                         "I/O error on %s\n",
2084                                        bdevname(bh->b_bdev, b));
2085                 }
2086                 /* note, we dont' set_buffer_write_io_error because we have
2087                  * our own ways of dealing with the IO errors
2088                  */
2089                 clear_buffer_uptodate(bh);
2090         }
2091         unlock_buffer(bh);
2092         put_bh(bh);
2093 }
2094
2095 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2096 {
2097         struct buffer_head *bh;
2098         struct buffer_head *latest = NULL;
2099         struct btrfs_super_block *super;
2100         int i;
2101         u64 transid = 0;
2102         u64 bytenr;
2103
2104         /* we would like to check all the supers, but that would make
2105          * a btrfs mount succeed after a mkfs from a different FS.
2106          * So, we need to add a special mount option to scan for
2107          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2108          */
2109         for (i = 0; i < 1; i++) {
2110                 bytenr = btrfs_sb_offset(i);
2111                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2112                         break;
2113                 bh = __bread(bdev, bytenr / 4096, 4096);
2114                 if (!bh)
2115                         continue;
2116
2117                 super = (struct btrfs_super_block *)bh->b_data;
2118                 if (btrfs_super_bytenr(super) != bytenr ||
2119                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2120                             sizeof(super->magic))) {
2121                         brelse(bh);
2122                         continue;
2123                 }
2124
2125                 if (!latest || btrfs_super_generation(super) > transid) {
2126                         brelse(latest);
2127                         latest = bh;
2128                         transid = btrfs_super_generation(super);
2129                 } else {
2130                         brelse(bh);
2131                 }
2132         }
2133         return latest;
2134 }
2135
2136 /*
2137  * this should be called twice, once with wait == 0 and
2138  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2139  * we write are pinned.
2140  *
2141  * They are released when wait == 1 is done.
2142  * max_mirrors must be the same for both runs, and it indicates how
2143  * many supers on this one device should be written.
2144  *
2145  * max_mirrors == 0 means to write them all.
2146  */
2147 static int write_dev_supers(struct btrfs_device *device,
2148                             struct btrfs_super_block *sb,
2149                             int do_barriers, int wait, int max_mirrors)
2150 {
2151         struct buffer_head *bh;
2152         int i;
2153         int ret;
2154         int errors = 0;
2155         u32 crc;
2156         u64 bytenr;
2157         int last_barrier = 0;
2158
2159         if (max_mirrors == 0)
2160                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2161
2162         /* make sure only the last submit_bh does a barrier */
2163         if (do_barriers) {
2164                 for (i = 0; i < max_mirrors; i++) {
2165                         bytenr = btrfs_sb_offset(i);
2166                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2167                             device->total_bytes)
2168                                 break;
2169                         last_barrier = i;
2170                 }
2171         }
2172
2173         for (i = 0; i < max_mirrors; i++) {
2174                 bytenr = btrfs_sb_offset(i);
2175                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2176                         break;
2177
2178                 if (wait) {
2179                         bh = __find_get_block(device->bdev, bytenr / 4096,
2180                                               BTRFS_SUPER_INFO_SIZE);
2181                         BUG_ON(!bh);
2182                         wait_on_buffer(bh);
2183                         if (!buffer_uptodate(bh))
2184                                 errors++;
2185
2186                         /* drop our reference */
2187                         brelse(bh);
2188
2189                         /* drop the reference from the wait == 0 run */
2190                         brelse(bh);
2191                         continue;
2192                 } else {
2193                         btrfs_set_super_bytenr(sb, bytenr);
2194
2195                         crc = ~(u32)0;
2196                         crc = btrfs_csum_data(NULL, (char *)sb +
2197                                               BTRFS_CSUM_SIZE, crc,
2198                                               BTRFS_SUPER_INFO_SIZE -
2199                                               BTRFS_CSUM_SIZE);
2200                         btrfs_csum_final(crc, sb->csum);
2201
2202                         /*
2203                          * one reference for us, and we leave it for the
2204                          * caller
2205                          */
2206                         bh = __getblk(device->bdev, bytenr / 4096,
2207                                       BTRFS_SUPER_INFO_SIZE);
2208                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2209
2210                         /* one reference for submit_bh */
2211                         get_bh(bh);
2212
2213                         set_buffer_uptodate(bh);
2214                         lock_buffer(bh);
2215                         bh->b_end_io = btrfs_end_buffer_write_sync;
2216                 }
2217
2218                 if (i == last_barrier && do_barriers && device->barriers) {
2219                         ret = submit_bh(WRITE_BARRIER, bh);
2220                         if (ret == -EOPNOTSUPP) {
2221                                 printk("btrfs: disabling barriers on dev %s\n",
2222                                        device->name);
2223                                 set_buffer_uptodate(bh);
2224                                 device->barriers = 0;
2225                                 /* one reference for submit_bh */
2226                                 get_bh(bh);
2227                                 lock_buffer(bh);
2228                                 ret = submit_bh(WRITE_SYNC, bh);
2229                         }
2230                 } else {
2231                         ret = submit_bh(WRITE_SYNC, bh);
2232                 }
2233
2234                 if (ret)
2235                         errors++;
2236         }
2237         return errors < i ? 0 : -1;
2238 }
2239
2240 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2241 {
2242         struct list_head *head;
2243         struct btrfs_device *dev;
2244         struct btrfs_super_block *sb;
2245         struct btrfs_dev_item *dev_item;
2246         int ret;
2247         int do_barriers;
2248         int max_errors;
2249         int total_errors = 0;
2250         u64 flags;
2251
2252         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2253         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2254
2255         sb = &root->fs_info->super_for_commit;
2256         dev_item = &sb->dev_item;
2257
2258         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2259         head = &root->fs_info->fs_devices->devices;
2260         list_for_each_entry(dev, head, dev_list) {
2261                 if (!dev->bdev) {
2262                         total_errors++;
2263                         continue;
2264                 }
2265                 if (!dev->in_fs_metadata || !dev->writeable)
2266                         continue;
2267
2268                 btrfs_set_stack_device_generation(dev_item, 0);
2269                 btrfs_set_stack_device_type(dev_item, dev->type);
2270                 btrfs_set_stack_device_id(dev_item, dev->devid);
2271                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2272                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2273                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2274                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2275                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2276                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2277                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2278
2279                 flags = btrfs_super_flags(sb);
2280                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2281
2282                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2283                 if (ret)
2284                         total_errors++;
2285         }
2286         if (total_errors > max_errors) {
2287                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2288                        total_errors);
2289                 BUG();
2290         }
2291
2292         total_errors = 0;
2293         list_for_each_entry(dev, head, dev_list) {
2294                 if (!dev->bdev)
2295                         continue;
2296                 if (!dev->in_fs_metadata || !dev->writeable)
2297                         continue;
2298
2299                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2300                 if (ret)
2301                         total_errors++;
2302         }
2303         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2304         if (total_errors > max_errors) {
2305                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2306                        total_errors);
2307                 BUG();
2308         }
2309         return 0;
2310 }
2311
2312 int write_ctree_super(struct btrfs_trans_handle *trans,
2313                       struct btrfs_root *root, int max_mirrors)
2314 {
2315         int ret;
2316
2317         ret = write_all_supers(root, max_mirrors);
2318         return ret;
2319 }
2320
2321 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2322 {
2323         spin_lock(&fs_info->fs_roots_radix_lock);
2324         radix_tree_delete(&fs_info->fs_roots_radix,
2325                           (unsigned long)root->root_key.objectid);
2326         spin_unlock(&fs_info->fs_roots_radix_lock);
2327
2328         if (btrfs_root_refs(&root->root_item) == 0)
2329                 synchronize_srcu(&fs_info->subvol_srcu);
2330
2331         free_fs_root(root);
2332         return 0;
2333 }
2334
2335 static void free_fs_root(struct btrfs_root *root)
2336 {
2337         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2338         if (root->anon_super.s_dev) {
2339                 down_write(&root->anon_super.s_umount);
2340                 kill_anon_super(&root->anon_super);
2341         }
2342         free_extent_buffer(root->node);
2343         free_extent_buffer(root->commit_root);
2344         kfree(root->name);
2345         kfree(root);
2346 }
2347
2348 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2349 {
2350         int ret;
2351         struct btrfs_root *gang[8];
2352         int i;
2353
2354         while (!list_empty(&fs_info->dead_roots)) {
2355                 gang[0] = list_entry(fs_info->dead_roots.next,
2356                                      struct btrfs_root, root_list);
2357                 list_del(&gang[0]->root_list);
2358
2359                 if (gang[0]->in_radix) {
2360                         btrfs_free_fs_root(fs_info, gang[0]);
2361                 } else {
2362                         free_extent_buffer(gang[0]->node);
2363                         free_extent_buffer(gang[0]->commit_root);
2364                         kfree(gang[0]);
2365                 }
2366         }
2367
2368         while (1) {
2369                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2370                                              (void **)gang, 0,
2371                                              ARRAY_SIZE(gang));
2372                 if (!ret)
2373                         break;
2374                 for (i = 0; i < ret; i++)
2375                         btrfs_free_fs_root(fs_info, gang[i]);
2376         }
2377         return 0;
2378 }
2379
2380 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2381 {
2382         u64 root_objectid = 0;
2383         struct btrfs_root *gang[8];
2384         int i;
2385         int ret;
2386
2387         while (1) {
2388                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2389                                              (void **)gang, root_objectid,
2390                                              ARRAY_SIZE(gang));
2391                 if (!ret)
2392                         break;
2393
2394                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2395                 for (i = 0; i < ret; i++) {
2396                         root_objectid = gang[i]->root_key.objectid;
2397                         btrfs_orphan_cleanup(gang[i]);
2398                 }
2399                 root_objectid++;
2400         }
2401         return 0;
2402 }
2403
2404 int btrfs_commit_super(struct btrfs_root *root)
2405 {
2406         struct btrfs_trans_handle *trans;
2407         int ret;
2408
2409         mutex_lock(&root->fs_info->cleaner_mutex);
2410         btrfs_run_delayed_iputs(root);
2411         btrfs_clean_old_snapshots(root);
2412         mutex_unlock(&root->fs_info->cleaner_mutex);
2413
2414         /* wait until ongoing cleanup work done */
2415         down_write(&root->fs_info->cleanup_work_sem);
2416         up_write(&root->fs_info->cleanup_work_sem);
2417
2418         trans = btrfs_start_transaction(root, 1);
2419         ret = btrfs_commit_transaction(trans, root);
2420         BUG_ON(ret);
2421         /* run commit again to drop the original snapshot */
2422         trans = btrfs_start_transaction(root, 1);
2423         btrfs_commit_transaction(trans, root);
2424         ret = btrfs_write_and_wait_transaction(NULL, root);
2425         BUG_ON(ret);
2426
2427         ret = write_ctree_super(NULL, root, 0);
2428         return ret;
2429 }
2430
2431 int close_ctree(struct btrfs_root *root)
2432 {
2433         struct btrfs_fs_info *fs_info = root->fs_info;
2434         int ret;
2435
2436         fs_info->closing = 1;
2437         smp_mb();
2438
2439         kthread_stop(root->fs_info->transaction_kthread);
2440         kthread_stop(root->fs_info->cleaner_kthread);
2441
2442         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2443                 ret =  btrfs_commit_super(root);
2444                 if (ret)
2445                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2446         }
2447
2448         fs_info->closing = 2;
2449         smp_mb();
2450
2451         if (fs_info->delalloc_bytes) {
2452                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2453                        (unsigned long long)fs_info->delalloc_bytes);
2454         }
2455         if (fs_info->total_ref_cache_size) {
2456                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2457                        (unsigned long long)fs_info->total_ref_cache_size);
2458         }
2459
2460         free_extent_buffer(fs_info->extent_root->node);
2461         free_extent_buffer(fs_info->extent_root->commit_root);
2462         free_extent_buffer(fs_info->tree_root->node);
2463         free_extent_buffer(fs_info->tree_root->commit_root);
2464         free_extent_buffer(root->fs_info->chunk_root->node);
2465         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2466         free_extent_buffer(root->fs_info->dev_root->node);
2467         free_extent_buffer(root->fs_info->dev_root->commit_root);
2468         free_extent_buffer(root->fs_info->csum_root->node);
2469         free_extent_buffer(root->fs_info->csum_root->commit_root);
2470
2471         btrfs_free_block_groups(root->fs_info);
2472
2473         del_fs_roots(fs_info);
2474
2475         iput(fs_info->btree_inode);
2476
2477         btrfs_stop_workers(&fs_info->generic_worker);
2478         btrfs_stop_workers(&fs_info->fixup_workers);
2479         btrfs_stop_workers(&fs_info->delalloc_workers);
2480         btrfs_stop_workers(&fs_info->workers);
2481         btrfs_stop_workers(&fs_info->endio_workers);
2482         btrfs_stop_workers(&fs_info->endio_meta_workers);
2483         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2484         btrfs_stop_workers(&fs_info->endio_write_workers);
2485         btrfs_stop_workers(&fs_info->submit_workers);
2486         btrfs_stop_workers(&fs_info->enospc_workers);
2487
2488         btrfs_close_devices(fs_info->fs_devices);
2489         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2490
2491         bdi_destroy(&fs_info->bdi);
2492         cleanup_srcu_struct(&fs_info->subvol_srcu);
2493
2494         kfree(fs_info->extent_root);
2495         kfree(fs_info->tree_root);
2496         kfree(fs_info->chunk_root);
2497         kfree(fs_info->dev_root);
2498         kfree(fs_info->csum_root);
2499         return 0;
2500 }
2501
2502 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2503 {
2504         int ret;
2505         struct inode *btree_inode = buf->first_page->mapping->host;
2506
2507         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2508                                      NULL);
2509         if (!ret)
2510                 return ret;
2511
2512         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2513                                     parent_transid);
2514         return !ret;
2515 }
2516
2517 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2518 {
2519         struct inode *btree_inode = buf->first_page->mapping->host;
2520         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2521                                           buf);
2522 }
2523
2524 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2525 {
2526         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2527         u64 transid = btrfs_header_generation(buf);
2528         struct inode *btree_inode = root->fs_info->btree_inode;
2529         int was_dirty;
2530
2531         btrfs_assert_tree_locked(buf);
2532         if (transid != root->fs_info->generation) {
2533                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2534                        "found %llu running %llu\n",
2535                         (unsigned long long)buf->start,
2536                         (unsigned long long)transid,
2537                         (unsigned long long)root->fs_info->generation);
2538                 WARN_ON(1);
2539         }
2540         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2541                                             buf);
2542         if (!was_dirty) {
2543                 spin_lock(&root->fs_info->delalloc_lock);
2544                 root->fs_info->dirty_metadata_bytes += buf->len;
2545                 spin_unlock(&root->fs_info->delalloc_lock);
2546         }
2547 }
2548
2549 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2550 {
2551         /*
2552          * looks as though older kernels can get into trouble with
2553          * this code, they end up stuck in balance_dirty_pages forever
2554          */
2555         u64 num_dirty;
2556         unsigned long thresh = 32 * 1024 * 1024;
2557
2558         if (current->flags & PF_MEMALLOC)
2559                 return;
2560
2561         num_dirty = root->fs_info->dirty_metadata_bytes;
2562
2563         if (num_dirty > thresh) {
2564                 balance_dirty_pages_ratelimited_nr(
2565                                    root->fs_info->btree_inode->i_mapping, 1);
2566         }
2567         return;
2568 }
2569
2570 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2571 {
2572         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2573         int ret;
2574         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2575         if (ret == 0)
2576                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2577         return ret;
2578 }
2579
2580 int btree_lock_page_hook(struct page *page)
2581 {
2582         struct inode *inode = page->mapping->host;
2583         struct btrfs_root *root = BTRFS_I(inode)->root;
2584         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2585         struct extent_buffer *eb;
2586         unsigned long len;
2587         u64 bytenr = page_offset(page);
2588
2589         if (page->private == EXTENT_PAGE_PRIVATE)
2590                 goto out;
2591
2592         len = page->private >> 2;
2593         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2594         if (!eb)
2595                 goto out;
2596
2597         btrfs_tree_lock(eb);
2598         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2599
2600         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2601                 spin_lock(&root->fs_info->delalloc_lock);
2602                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2603                         root->fs_info->dirty_metadata_bytes -= eb->len;
2604                 else
2605                         WARN_ON(1);
2606                 spin_unlock(&root->fs_info->delalloc_lock);
2607         }
2608
2609         btrfs_tree_unlock(eb);
2610         free_extent_buffer(eb);
2611 out:
2612         lock_page(page);
2613         return 0;
2614 }
2615
2616 static struct extent_io_ops btree_extent_io_ops = {
2617         .write_cache_pages_lock_hook = btree_lock_page_hook,
2618         .readpage_end_io_hook = btree_readpage_end_io_hook,
2619         .submit_bio_hook = btree_submit_bio_hook,
2620         /* note we're sharing with inode.c for the merge bio hook */
2621         .merge_bio_hook = btrfs_merge_bio_hook,
2622 };