Btrfs: create special free space cache inode
[linux-3.10.git] / fs / btrfs / disk-io.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include "compat.h"
32 #include "ctree.h"
33 #include "disk-io.h"
34 #include "transaction.h"
35 #include "btrfs_inode.h"
36 #include "volumes.h"
37 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "locking.h"
40 #include "tree-log.h"
41 #include "free-space-cache.h"
42
43 static struct extent_io_ops btree_extent_io_ops;
44 static void end_workqueue_fn(struct btrfs_work *work);
45 static void free_fs_root(struct btrfs_root *root);
46
47 /*
48  * end_io_wq structs are used to do processing in task context when an IO is
49  * complete.  This is used during reads to verify checksums, and it is used
50  * by writes to insert metadata for new file extents after IO is complete.
51  */
52 struct end_io_wq {
53         struct bio *bio;
54         bio_end_io_t *end_io;
55         void *private;
56         struct btrfs_fs_info *info;
57         int error;
58         int metadata;
59         struct list_head list;
60         struct btrfs_work work;
61 };
62
63 /*
64  * async submit bios are used to offload expensive checksumming
65  * onto the worker threads.  They checksum file and metadata bios
66  * just before they are sent down the IO stack.
67  */
68 struct async_submit_bio {
69         struct inode *inode;
70         struct bio *bio;
71         struct list_head list;
72         extent_submit_bio_hook_t *submit_bio_start;
73         extent_submit_bio_hook_t *submit_bio_done;
74         int rw;
75         int mirror_num;
76         unsigned long bio_flags;
77         /*
78          * bio_offset is optional, can be used if the pages in the bio
79          * can't tell us where in the file the bio should go
80          */
81         u64 bio_offset;
82         struct btrfs_work work;
83 };
84
85 /* These are used to set the lockdep class on the extent buffer locks.
86  * The class is set by the readpage_end_io_hook after the buffer has
87  * passed csum validation but before the pages are unlocked.
88  *
89  * The lockdep class is also set by btrfs_init_new_buffer on freshly
90  * allocated blocks.
91  *
92  * The class is based on the level in the tree block, which allows lockdep
93  * to know that lower nodes nest inside the locks of higher nodes.
94  *
95  * We also add a check to make sure the highest level of the tree is
96  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
97  * code needs update as well.
98  */
99 #ifdef CONFIG_DEBUG_LOCK_ALLOC
100 # if BTRFS_MAX_LEVEL != 8
101 #  error
102 # endif
103 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
104 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
105         /* leaf */
106         "btrfs-extent-00",
107         "btrfs-extent-01",
108         "btrfs-extent-02",
109         "btrfs-extent-03",
110         "btrfs-extent-04",
111         "btrfs-extent-05",
112         "btrfs-extent-06",
113         "btrfs-extent-07",
114         /* highest possible level */
115         "btrfs-extent-08",
116 };
117 #endif
118
119 /*
120  * extents on the btree inode are pretty simple, there's one extent
121  * that covers the entire device
122  */
123 static struct extent_map *btree_get_extent(struct inode *inode,
124                 struct page *page, size_t page_offset, u64 start, u64 len,
125                 int create)
126 {
127         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
128         struct extent_map *em;
129         int ret;
130
131         read_lock(&em_tree->lock);
132         em = lookup_extent_mapping(em_tree, start, len);
133         if (em) {
134                 em->bdev =
135                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
136                 read_unlock(&em_tree->lock);
137                 goto out;
138         }
139         read_unlock(&em_tree->lock);
140
141         em = alloc_extent_map(GFP_NOFS);
142         if (!em) {
143                 em = ERR_PTR(-ENOMEM);
144                 goto out;
145         }
146         em->start = 0;
147         em->len = (u64)-1;
148         em->block_len = (u64)-1;
149         em->block_start = 0;
150         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
151
152         write_lock(&em_tree->lock);
153         ret = add_extent_mapping(em_tree, em);
154         if (ret == -EEXIST) {
155                 u64 failed_start = em->start;
156                 u64 failed_len = em->len;
157
158                 free_extent_map(em);
159                 em = lookup_extent_mapping(em_tree, start, len);
160                 if (em) {
161                         ret = 0;
162                 } else {
163                         em = lookup_extent_mapping(em_tree, failed_start,
164                                                    failed_len);
165                         ret = -EIO;
166                 }
167         } else if (ret) {
168                 free_extent_map(em);
169                 em = NULL;
170         }
171         write_unlock(&em_tree->lock);
172
173         if (ret)
174                 em = ERR_PTR(ret);
175 out:
176         return em;
177 }
178
179 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
180 {
181         return crc32c(seed, data, len);
182 }
183
184 void btrfs_csum_final(u32 crc, char *result)
185 {
186         *(__le32 *)result = ~cpu_to_le32(crc);
187 }
188
189 /*
190  * compute the csum for a btree block, and either verify it or write it
191  * into the csum field of the block.
192  */
193 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
194                            int verify)
195 {
196         u16 csum_size =
197                 btrfs_super_csum_size(&root->fs_info->super_copy);
198         char *result = NULL;
199         unsigned long len;
200         unsigned long cur_len;
201         unsigned long offset = BTRFS_CSUM_SIZE;
202         char *map_token = NULL;
203         char *kaddr;
204         unsigned long map_start;
205         unsigned long map_len;
206         int err;
207         u32 crc = ~(u32)0;
208         unsigned long inline_result;
209
210         len = buf->len - offset;
211         while (len > 0) {
212                 err = map_private_extent_buffer(buf, offset, 32,
213                                         &map_token, &kaddr,
214                                         &map_start, &map_len, KM_USER0);
215                 if (err)
216                         return 1;
217                 cur_len = min(len, map_len - (offset - map_start));
218                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
219                                       crc, cur_len);
220                 len -= cur_len;
221                 offset += cur_len;
222                 unmap_extent_buffer(buf, map_token, KM_USER0);
223         }
224         if (csum_size > sizeof(inline_result)) {
225                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
226                 if (!result)
227                         return 1;
228         } else {
229                 result = (char *)&inline_result;
230         }
231
232         btrfs_csum_final(crc, result);
233
234         if (verify) {
235                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
236                         u32 val;
237                         u32 found = 0;
238                         memcpy(&found, result, csum_size);
239
240                         read_extent_buffer(buf, &val, 0, csum_size);
241                         if (printk_ratelimit()) {
242                                 printk(KERN_INFO "btrfs: %s checksum verify "
243                                        "failed on %llu wanted %X found %X "
244                                        "level %d\n",
245                                        root->fs_info->sb->s_id,
246                                        (unsigned long long)buf->start, val, found,
247                                        btrfs_header_level(buf));
248                         }
249                         if (result != (char *)&inline_result)
250                                 kfree(result);
251                         return 1;
252                 }
253         } else {
254                 write_extent_buffer(buf, result, 0, csum_size);
255         }
256         if (result != (char *)&inline_result)
257                 kfree(result);
258         return 0;
259 }
260
261 /*
262  * we can't consider a given block up to date unless the transid of the
263  * block matches the transid in the parent node's pointer.  This is how we
264  * detect blocks that either didn't get written at all or got written
265  * in the wrong place.
266  */
267 static int verify_parent_transid(struct extent_io_tree *io_tree,
268                                  struct extent_buffer *eb, u64 parent_transid)
269 {
270         struct extent_state *cached_state = NULL;
271         int ret;
272
273         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
274                 return 0;
275
276         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
277                          0, &cached_state, GFP_NOFS);
278         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
279             btrfs_header_generation(eb) == parent_transid) {
280                 ret = 0;
281                 goto out;
282         }
283         if (printk_ratelimit()) {
284                 printk("parent transid verify failed on %llu wanted %llu "
285                        "found %llu\n",
286                        (unsigned long long)eb->start,
287                        (unsigned long long)parent_transid,
288                        (unsigned long long)btrfs_header_generation(eb));
289         }
290         ret = 1;
291         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
292 out:
293         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
294                              &cached_state, GFP_NOFS);
295         return ret;
296 }
297
298 /*
299  * helper to read a given tree block, doing retries as required when
300  * the checksums don't match and we have alternate mirrors to try.
301  */
302 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
303                                           struct extent_buffer *eb,
304                                           u64 start, u64 parent_transid)
305 {
306         struct extent_io_tree *io_tree;
307         int ret;
308         int num_copies = 0;
309         int mirror_num = 0;
310
311         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
312         while (1) {
313                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
314                                                btree_get_extent, mirror_num);
315                 if (!ret &&
316                     !verify_parent_transid(io_tree, eb, parent_transid))
317                         return ret;
318
319                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
320                                               eb->start, eb->len);
321                 if (num_copies == 1)
322                         return ret;
323
324                 mirror_num++;
325                 if (mirror_num > num_copies)
326                         return ret;
327         }
328         return -EIO;
329 }
330
331 /*
332  * checksum a dirty tree block before IO.  This has extra checks to make sure
333  * we only fill in the checksum field in the first page of a multi-page block
334  */
335
336 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
337 {
338         struct extent_io_tree *tree;
339         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
340         u64 found_start;
341         int found_level;
342         unsigned long len;
343         struct extent_buffer *eb;
344         int ret;
345
346         tree = &BTRFS_I(page->mapping->host)->io_tree;
347
348         if (page->private == EXTENT_PAGE_PRIVATE)
349                 goto out;
350         if (!page->private)
351                 goto out;
352         len = page->private >> 2;
353         WARN_ON(len == 0);
354
355         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
356         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
357                                              btrfs_header_generation(eb));
358         BUG_ON(ret);
359         found_start = btrfs_header_bytenr(eb);
360         if (found_start != start) {
361                 WARN_ON(1);
362                 goto err;
363         }
364         if (eb->first_page != page) {
365                 WARN_ON(1);
366                 goto err;
367         }
368         if (!PageUptodate(page)) {
369                 WARN_ON(1);
370                 goto err;
371         }
372         found_level = btrfs_header_level(eb);
373
374         csum_tree_block(root, eb, 0);
375 err:
376         free_extent_buffer(eb);
377 out:
378         return 0;
379 }
380
381 static int check_tree_block_fsid(struct btrfs_root *root,
382                                  struct extent_buffer *eb)
383 {
384         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
385         u8 fsid[BTRFS_UUID_SIZE];
386         int ret = 1;
387
388         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
389                            BTRFS_FSID_SIZE);
390         while (fs_devices) {
391                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
392                         ret = 0;
393                         break;
394                 }
395                 fs_devices = fs_devices->seed;
396         }
397         return ret;
398 }
399
400 #ifdef CONFIG_DEBUG_LOCK_ALLOC
401 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
402 {
403         lockdep_set_class_and_name(&eb->lock,
404                            &btrfs_eb_class[level],
405                            btrfs_eb_name[level]);
406 }
407 #endif
408
409 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
410                                struct extent_state *state)
411 {
412         struct extent_io_tree *tree;
413         u64 found_start;
414         int found_level;
415         unsigned long len;
416         struct extent_buffer *eb;
417         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
418         int ret = 0;
419
420         tree = &BTRFS_I(page->mapping->host)->io_tree;
421         if (page->private == EXTENT_PAGE_PRIVATE)
422                 goto out;
423         if (!page->private)
424                 goto out;
425
426         len = page->private >> 2;
427         WARN_ON(len == 0);
428
429         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
430
431         found_start = btrfs_header_bytenr(eb);
432         if (found_start != start) {
433                 if (printk_ratelimit()) {
434                         printk(KERN_INFO "btrfs bad tree block start "
435                                "%llu %llu\n",
436                                (unsigned long long)found_start,
437                                (unsigned long long)eb->start);
438                 }
439                 ret = -EIO;
440                 goto err;
441         }
442         if (eb->first_page != page) {
443                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
444                        eb->first_page->index, page->index);
445                 WARN_ON(1);
446                 ret = -EIO;
447                 goto err;
448         }
449         if (check_tree_block_fsid(root, eb)) {
450                 if (printk_ratelimit()) {
451                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
452                                (unsigned long long)eb->start);
453                 }
454                 ret = -EIO;
455                 goto err;
456         }
457         found_level = btrfs_header_level(eb);
458
459         btrfs_set_buffer_lockdep_class(eb, found_level);
460
461         ret = csum_tree_block(root, eb, 1);
462         if (ret)
463                 ret = -EIO;
464
465         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
466         end = eb->start + end - 1;
467 err:
468         free_extent_buffer(eb);
469 out:
470         return ret;
471 }
472
473 static void end_workqueue_bio(struct bio *bio, int err)
474 {
475         struct end_io_wq *end_io_wq = bio->bi_private;
476         struct btrfs_fs_info *fs_info;
477
478         fs_info = end_io_wq->info;
479         end_io_wq->error = err;
480         end_io_wq->work.func = end_workqueue_fn;
481         end_io_wq->work.flags = 0;
482
483         if (bio->bi_rw & REQ_WRITE) {
484                 if (end_io_wq->metadata)
485                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
486                                            &end_io_wq->work);
487                 else
488                         btrfs_queue_worker(&fs_info->endio_write_workers,
489                                            &end_io_wq->work);
490         } else {
491                 if (end_io_wq->metadata)
492                         btrfs_queue_worker(&fs_info->endio_meta_workers,
493                                            &end_io_wq->work);
494                 else
495                         btrfs_queue_worker(&fs_info->endio_workers,
496                                            &end_io_wq->work);
497         }
498 }
499
500 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
501                         int metadata)
502 {
503         struct end_io_wq *end_io_wq;
504         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
505         if (!end_io_wq)
506                 return -ENOMEM;
507
508         end_io_wq->private = bio->bi_private;
509         end_io_wq->end_io = bio->bi_end_io;
510         end_io_wq->info = info;
511         end_io_wq->error = 0;
512         end_io_wq->bio = bio;
513         end_io_wq->metadata = metadata;
514
515         bio->bi_private = end_io_wq;
516         bio->bi_end_io = end_workqueue_bio;
517         return 0;
518 }
519
520 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
521 {
522         unsigned long limit = min_t(unsigned long,
523                                     info->workers.max_workers,
524                                     info->fs_devices->open_devices);
525         return 256 * limit;
526 }
527
528 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
529 {
530         return atomic_read(&info->nr_async_bios) >
531                 btrfs_async_submit_limit(info);
532 }
533
534 static void run_one_async_start(struct btrfs_work *work)
535 {
536         struct btrfs_fs_info *fs_info;
537         struct async_submit_bio *async;
538
539         async = container_of(work, struct  async_submit_bio, work);
540         fs_info = BTRFS_I(async->inode)->root->fs_info;
541         async->submit_bio_start(async->inode, async->rw, async->bio,
542                                async->mirror_num, async->bio_flags,
543                                async->bio_offset);
544 }
545
546 static void run_one_async_done(struct btrfs_work *work)
547 {
548         struct btrfs_fs_info *fs_info;
549         struct async_submit_bio *async;
550         int limit;
551
552         async = container_of(work, struct  async_submit_bio, work);
553         fs_info = BTRFS_I(async->inode)->root->fs_info;
554
555         limit = btrfs_async_submit_limit(fs_info);
556         limit = limit * 2 / 3;
557
558         atomic_dec(&fs_info->nr_async_submits);
559
560         if (atomic_read(&fs_info->nr_async_submits) < limit &&
561             waitqueue_active(&fs_info->async_submit_wait))
562                 wake_up(&fs_info->async_submit_wait);
563
564         async->submit_bio_done(async->inode, async->rw, async->bio,
565                                async->mirror_num, async->bio_flags,
566                                async->bio_offset);
567 }
568
569 static void run_one_async_free(struct btrfs_work *work)
570 {
571         struct async_submit_bio *async;
572
573         async = container_of(work, struct  async_submit_bio, work);
574         kfree(async);
575 }
576
577 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
578                         int rw, struct bio *bio, int mirror_num,
579                         unsigned long bio_flags,
580                         u64 bio_offset,
581                         extent_submit_bio_hook_t *submit_bio_start,
582                         extent_submit_bio_hook_t *submit_bio_done)
583 {
584         struct async_submit_bio *async;
585
586         async = kmalloc(sizeof(*async), GFP_NOFS);
587         if (!async)
588                 return -ENOMEM;
589
590         async->inode = inode;
591         async->rw = rw;
592         async->bio = bio;
593         async->mirror_num = mirror_num;
594         async->submit_bio_start = submit_bio_start;
595         async->submit_bio_done = submit_bio_done;
596
597         async->work.func = run_one_async_start;
598         async->work.ordered_func = run_one_async_done;
599         async->work.ordered_free = run_one_async_free;
600
601         async->work.flags = 0;
602         async->bio_flags = bio_flags;
603         async->bio_offset = bio_offset;
604
605         atomic_inc(&fs_info->nr_async_submits);
606
607         if (rw & REQ_SYNC)
608                 btrfs_set_work_high_prio(&async->work);
609
610         btrfs_queue_worker(&fs_info->workers, &async->work);
611
612         while (atomic_read(&fs_info->async_submit_draining) &&
613               atomic_read(&fs_info->nr_async_submits)) {
614                 wait_event(fs_info->async_submit_wait,
615                            (atomic_read(&fs_info->nr_async_submits) == 0));
616         }
617
618         return 0;
619 }
620
621 static int btree_csum_one_bio(struct bio *bio)
622 {
623         struct bio_vec *bvec = bio->bi_io_vec;
624         int bio_index = 0;
625         struct btrfs_root *root;
626
627         WARN_ON(bio->bi_vcnt <= 0);
628         while (bio_index < bio->bi_vcnt) {
629                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
630                 csum_dirty_buffer(root, bvec->bv_page);
631                 bio_index++;
632                 bvec++;
633         }
634         return 0;
635 }
636
637 static int __btree_submit_bio_start(struct inode *inode, int rw,
638                                     struct bio *bio, int mirror_num,
639                                     unsigned long bio_flags,
640                                     u64 bio_offset)
641 {
642         /*
643          * when we're called for a write, we're already in the async
644          * submission context.  Just jump into btrfs_map_bio
645          */
646         btree_csum_one_bio(bio);
647         return 0;
648 }
649
650 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
651                                  int mirror_num, unsigned long bio_flags,
652                                  u64 bio_offset)
653 {
654         /*
655          * when we're called for a write, we're already in the async
656          * submission context.  Just jump into btrfs_map_bio
657          */
658         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
659 }
660
661 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
662                                  int mirror_num, unsigned long bio_flags,
663                                  u64 bio_offset)
664 {
665         int ret;
666
667         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
668                                           bio, 1);
669         BUG_ON(ret);
670
671         if (!(rw & REQ_WRITE)) {
672                 /*
673                  * called for a read, do the setup so that checksum validation
674                  * can happen in the async kernel threads
675                  */
676                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
677                                      mirror_num, 0);
678         }
679
680         /*
681          * kthread helpers are used to submit writes so that checksumming
682          * can happen in parallel across all CPUs
683          */
684         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
685                                    inode, rw, bio, mirror_num, 0,
686                                    bio_offset,
687                                    __btree_submit_bio_start,
688                                    __btree_submit_bio_done);
689 }
690
691 static int btree_writepage(struct page *page, struct writeback_control *wbc)
692 {
693         struct extent_io_tree *tree;
694         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
695         struct extent_buffer *eb;
696         int was_dirty;
697
698         tree = &BTRFS_I(page->mapping->host)->io_tree;
699         if (!(current->flags & PF_MEMALLOC)) {
700                 return extent_write_full_page(tree, page,
701                                               btree_get_extent, wbc);
702         }
703
704         redirty_page_for_writepage(wbc, page);
705         eb = btrfs_find_tree_block(root, page_offset(page),
706                                       PAGE_CACHE_SIZE);
707         WARN_ON(!eb);
708
709         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
710         if (!was_dirty) {
711                 spin_lock(&root->fs_info->delalloc_lock);
712                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
713                 spin_unlock(&root->fs_info->delalloc_lock);
714         }
715         free_extent_buffer(eb);
716
717         unlock_page(page);
718         return 0;
719 }
720
721 static int btree_writepages(struct address_space *mapping,
722                             struct writeback_control *wbc)
723 {
724         struct extent_io_tree *tree;
725         tree = &BTRFS_I(mapping->host)->io_tree;
726         if (wbc->sync_mode == WB_SYNC_NONE) {
727                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
728                 u64 num_dirty;
729                 unsigned long thresh = 32 * 1024 * 1024;
730
731                 if (wbc->for_kupdate)
732                         return 0;
733
734                 /* this is a bit racy, but that's ok */
735                 num_dirty = root->fs_info->dirty_metadata_bytes;
736                 if (num_dirty < thresh)
737                         return 0;
738         }
739         return extent_writepages(tree, mapping, btree_get_extent, wbc);
740 }
741
742 static int btree_readpage(struct file *file, struct page *page)
743 {
744         struct extent_io_tree *tree;
745         tree = &BTRFS_I(page->mapping->host)->io_tree;
746         return extent_read_full_page(tree, page, btree_get_extent);
747 }
748
749 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
750 {
751         struct extent_io_tree *tree;
752         struct extent_map_tree *map;
753         int ret;
754
755         if (PageWriteback(page) || PageDirty(page))
756                 return 0;
757
758         tree = &BTRFS_I(page->mapping->host)->io_tree;
759         map = &BTRFS_I(page->mapping->host)->extent_tree;
760
761         ret = try_release_extent_state(map, tree, page, gfp_flags);
762         if (!ret)
763                 return 0;
764
765         ret = try_release_extent_buffer(tree, page);
766         if (ret == 1) {
767                 ClearPagePrivate(page);
768                 set_page_private(page, 0);
769                 page_cache_release(page);
770         }
771
772         return ret;
773 }
774
775 static void btree_invalidatepage(struct page *page, unsigned long offset)
776 {
777         struct extent_io_tree *tree;
778         tree = &BTRFS_I(page->mapping->host)->io_tree;
779         extent_invalidatepage(tree, page, offset);
780         btree_releasepage(page, GFP_NOFS);
781         if (PagePrivate(page)) {
782                 printk(KERN_WARNING "btrfs warning page private not zero "
783                        "on page %llu\n", (unsigned long long)page_offset(page));
784                 ClearPagePrivate(page);
785                 set_page_private(page, 0);
786                 page_cache_release(page);
787         }
788 }
789
790 static const struct address_space_operations btree_aops = {
791         .readpage       = btree_readpage,
792         .writepage      = btree_writepage,
793         .writepages     = btree_writepages,
794         .releasepage    = btree_releasepage,
795         .invalidatepage = btree_invalidatepage,
796         .sync_page      = block_sync_page,
797 };
798
799 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
800                          u64 parent_transid)
801 {
802         struct extent_buffer *buf = NULL;
803         struct inode *btree_inode = root->fs_info->btree_inode;
804         int ret = 0;
805
806         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
807         if (!buf)
808                 return 0;
809         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
810                                  buf, 0, 0, btree_get_extent, 0);
811         free_extent_buffer(buf);
812         return ret;
813 }
814
815 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
816                                             u64 bytenr, u32 blocksize)
817 {
818         struct inode *btree_inode = root->fs_info->btree_inode;
819         struct extent_buffer *eb;
820         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
821                                 bytenr, blocksize, GFP_NOFS);
822         return eb;
823 }
824
825 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
826                                                  u64 bytenr, u32 blocksize)
827 {
828         struct inode *btree_inode = root->fs_info->btree_inode;
829         struct extent_buffer *eb;
830
831         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
832                                  bytenr, blocksize, NULL, GFP_NOFS);
833         return eb;
834 }
835
836
837 int btrfs_write_tree_block(struct extent_buffer *buf)
838 {
839         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
840                                         buf->start + buf->len - 1);
841 }
842
843 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
844 {
845         return filemap_fdatawait_range(buf->first_page->mapping,
846                                        buf->start, buf->start + buf->len - 1);
847 }
848
849 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
850                                       u32 blocksize, u64 parent_transid)
851 {
852         struct extent_buffer *buf = NULL;
853         struct inode *btree_inode = root->fs_info->btree_inode;
854         struct extent_io_tree *io_tree;
855         int ret;
856
857         io_tree = &BTRFS_I(btree_inode)->io_tree;
858
859         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
860         if (!buf)
861                 return NULL;
862
863         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
864
865         if (ret == 0)
866                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
867         return buf;
868
869 }
870
871 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
872                      struct extent_buffer *buf)
873 {
874         struct inode *btree_inode = root->fs_info->btree_inode;
875         if (btrfs_header_generation(buf) ==
876             root->fs_info->running_transaction->transid) {
877                 btrfs_assert_tree_locked(buf);
878
879                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
880                         spin_lock(&root->fs_info->delalloc_lock);
881                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
882                                 root->fs_info->dirty_metadata_bytes -= buf->len;
883                         else
884                                 WARN_ON(1);
885                         spin_unlock(&root->fs_info->delalloc_lock);
886                 }
887
888                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
889                 btrfs_set_lock_blocking(buf);
890                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
891                                           buf);
892         }
893         return 0;
894 }
895
896 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
897                         u32 stripesize, struct btrfs_root *root,
898                         struct btrfs_fs_info *fs_info,
899                         u64 objectid)
900 {
901         root->node = NULL;
902         root->commit_root = NULL;
903         root->sectorsize = sectorsize;
904         root->nodesize = nodesize;
905         root->leafsize = leafsize;
906         root->stripesize = stripesize;
907         root->ref_cows = 0;
908         root->track_dirty = 0;
909         root->in_radix = 0;
910         root->orphan_item_inserted = 0;
911         root->orphan_cleanup_state = 0;
912
913         root->fs_info = fs_info;
914         root->objectid = objectid;
915         root->last_trans = 0;
916         root->highest_objectid = 0;
917         root->name = NULL;
918         root->in_sysfs = 0;
919         root->inode_tree = RB_ROOT;
920         root->block_rsv = NULL;
921         root->orphan_block_rsv = NULL;
922
923         INIT_LIST_HEAD(&root->dirty_list);
924         INIT_LIST_HEAD(&root->orphan_list);
925         INIT_LIST_HEAD(&root->root_list);
926         spin_lock_init(&root->node_lock);
927         spin_lock_init(&root->orphan_lock);
928         spin_lock_init(&root->inode_lock);
929         spin_lock_init(&root->accounting_lock);
930         mutex_init(&root->objectid_mutex);
931         mutex_init(&root->log_mutex);
932         init_waitqueue_head(&root->log_writer_wait);
933         init_waitqueue_head(&root->log_commit_wait[0]);
934         init_waitqueue_head(&root->log_commit_wait[1]);
935         atomic_set(&root->log_commit[0], 0);
936         atomic_set(&root->log_commit[1], 0);
937         atomic_set(&root->log_writers, 0);
938         root->log_batch = 0;
939         root->log_transid = 0;
940         root->last_log_commit = 0;
941         extent_io_tree_init(&root->dirty_log_pages,
942                              fs_info->btree_inode->i_mapping, GFP_NOFS);
943
944         memset(&root->root_key, 0, sizeof(root->root_key));
945         memset(&root->root_item, 0, sizeof(root->root_item));
946         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
947         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
948         root->defrag_trans_start = fs_info->generation;
949         init_completion(&root->kobj_unregister);
950         root->defrag_running = 0;
951         root->root_key.objectid = objectid;
952         root->anon_super.s_root = NULL;
953         root->anon_super.s_dev = 0;
954         INIT_LIST_HEAD(&root->anon_super.s_list);
955         INIT_LIST_HEAD(&root->anon_super.s_instances);
956         init_rwsem(&root->anon_super.s_umount);
957
958         return 0;
959 }
960
961 static int find_and_setup_root(struct btrfs_root *tree_root,
962                                struct btrfs_fs_info *fs_info,
963                                u64 objectid,
964                                struct btrfs_root *root)
965 {
966         int ret;
967         u32 blocksize;
968         u64 generation;
969
970         __setup_root(tree_root->nodesize, tree_root->leafsize,
971                      tree_root->sectorsize, tree_root->stripesize,
972                      root, fs_info, objectid);
973         ret = btrfs_find_last_root(tree_root, objectid,
974                                    &root->root_item, &root->root_key);
975         if (ret > 0)
976                 return -ENOENT;
977         BUG_ON(ret);
978
979         generation = btrfs_root_generation(&root->root_item);
980         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
981         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
982                                      blocksize, generation);
983         BUG_ON(!root->node);
984         root->commit_root = btrfs_root_node(root);
985         return 0;
986 }
987
988 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
989                                          struct btrfs_fs_info *fs_info)
990 {
991         struct btrfs_root *root;
992         struct btrfs_root *tree_root = fs_info->tree_root;
993         struct extent_buffer *leaf;
994
995         root = kzalloc(sizeof(*root), GFP_NOFS);
996         if (!root)
997                 return ERR_PTR(-ENOMEM);
998
999         __setup_root(tree_root->nodesize, tree_root->leafsize,
1000                      tree_root->sectorsize, tree_root->stripesize,
1001                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1002
1003         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1004         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1005         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1006         /*
1007          * log trees do not get reference counted because they go away
1008          * before a real commit is actually done.  They do store pointers
1009          * to file data extents, and those reference counts still get
1010          * updated (along with back refs to the log tree).
1011          */
1012         root->ref_cows = 0;
1013
1014         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1015                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1016         if (IS_ERR(leaf)) {
1017                 kfree(root);
1018                 return ERR_CAST(leaf);
1019         }
1020
1021         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1022         btrfs_set_header_bytenr(leaf, leaf->start);
1023         btrfs_set_header_generation(leaf, trans->transid);
1024         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1025         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1026         root->node = leaf;
1027
1028         write_extent_buffer(root->node, root->fs_info->fsid,
1029                             (unsigned long)btrfs_header_fsid(root->node),
1030                             BTRFS_FSID_SIZE);
1031         btrfs_mark_buffer_dirty(root->node);
1032         btrfs_tree_unlock(root->node);
1033         return root;
1034 }
1035
1036 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1037                              struct btrfs_fs_info *fs_info)
1038 {
1039         struct btrfs_root *log_root;
1040
1041         log_root = alloc_log_tree(trans, fs_info);
1042         if (IS_ERR(log_root))
1043                 return PTR_ERR(log_root);
1044         WARN_ON(fs_info->log_root_tree);
1045         fs_info->log_root_tree = log_root;
1046         return 0;
1047 }
1048
1049 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1050                        struct btrfs_root *root)
1051 {
1052         struct btrfs_root *log_root;
1053         struct btrfs_inode_item *inode_item;
1054
1055         log_root = alloc_log_tree(trans, root->fs_info);
1056         if (IS_ERR(log_root))
1057                 return PTR_ERR(log_root);
1058
1059         log_root->last_trans = trans->transid;
1060         log_root->root_key.offset = root->root_key.objectid;
1061
1062         inode_item = &log_root->root_item.inode;
1063         inode_item->generation = cpu_to_le64(1);
1064         inode_item->size = cpu_to_le64(3);
1065         inode_item->nlink = cpu_to_le32(1);
1066         inode_item->nbytes = cpu_to_le64(root->leafsize);
1067         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1068
1069         btrfs_set_root_node(&log_root->root_item, log_root->node);
1070
1071         WARN_ON(root->log_root);
1072         root->log_root = log_root;
1073         root->log_transid = 0;
1074         root->last_log_commit = 0;
1075         return 0;
1076 }
1077
1078 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1079                                                struct btrfs_key *location)
1080 {
1081         struct btrfs_root *root;
1082         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1083         struct btrfs_path *path;
1084         struct extent_buffer *l;
1085         u64 generation;
1086         u32 blocksize;
1087         int ret = 0;
1088
1089         root = kzalloc(sizeof(*root), GFP_NOFS);
1090         if (!root)
1091                 return ERR_PTR(-ENOMEM);
1092         if (location->offset == (u64)-1) {
1093                 ret = find_and_setup_root(tree_root, fs_info,
1094                                           location->objectid, root);
1095                 if (ret) {
1096                         kfree(root);
1097                         return ERR_PTR(ret);
1098                 }
1099                 goto out;
1100         }
1101
1102         __setup_root(tree_root->nodesize, tree_root->leafsize,
1103                      tree_root->sectorsize, tree_root->stripesize,
1104                      root, fs_info, location->objectid);
1105
1106         path = btrfs_alloc_path();
1107         BUG_ON(!path);
1108         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1109         if (ret == 0) {
1110                 l = path->nodes[0];
1111                 read_extent_buffer(l, &root->root_item,
1112                                 btrfs_item_ptr_offset(l, path->slots[0]),
1113                                 sizeof(root->root_item));
1114                 memcpy(&root->root_key, location, sizeof(*location));
1115         }
1116         btrfs_free_path(path);
1117         if (ret) {
1118                 if (ret > 0)
1119                         ret = -ENOENT;
1120                 return ERR_PTR(ret);
1121         }
1122
1123         generation = btrfs_root_generation(&root->root_item);
1124         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1125         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1126                                      blocksize, generation);
1127         root->commit_root = btrfs_root_node(root);
1128         BUG_ON(!root->node);
1129 out:
1130         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1131                 root->ref_cows = 1;
1132
1133         return root;
1134 }
1135
1136 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1137                                         u64 root_objectid)
1138 {
1139         struct btrfs_root *root;
1140
1141         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1142                 return fs_info->tree_root;
1143         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1144                 return fs_info->extent_root;
1145
1146         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1147                                  (unsigned long)root_objectid);
1148         return root;
1149 }
1150
1151 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1152                                               struct btrfs_key *location)
1153 {
1154         struct btrfs_root *root;
1155         int ret;
1156
1157         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1158                 return fs_info->tree_root;
1159         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1160                 return fs_info->extent_root;
1161         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1162                 return fs_info->chunk_root;
1163         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1164                 return fs_info->dev_root;
1165         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1166                 return fs_info->csum_root;
1167 again:
1168         spin_lock(&fs_info->fs_roots_radix_lock);
1169         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1170                                  (unsigned long)location->objectid);
1171         spin_unlock(&fs_info->fs_roots_radix_lock);
1172         if (root)
1173                 return root;
1174
1175         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1176         if (IS_ERR(root))
1177                 return root;
1178
1179         set_anon_super(&root->anon_super, NULL);
1180
1181         if (btrfs_root_refs(&root->root_item) == 0) {
1182                 ret = -ENOENT;
1183                 goto fail;
1184         }
1185
1186         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1187         if (ret < 0)
1188                 goto fail;
1189         if (ret == 0)
1190                 root->orphan_item_inserted = 1;
1191
1192         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1193         if (ret)
1194                 goto fail;
1195
1196         spin_lock(&fs_info->fs_roots_radix_lock);
1197         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1198                                 (unsigned long)root->root_key.objectid,
1199                                 root);
1200         if (ret == 0)
1201                 root->in_radix = 1;
1202
1203         spin_unlock(&fs_info->fs_roots_radix_lock);
1204         radix_tree_preload_end();
1205         if (ret) {
1206                 if (ret == -EEXIST) {
1207                         free_fs_root(root);
1208                         goto again;
1209                 }
1210                 goto fail;
1211         }
1212
1213         ret = btrfs_find_dead_roots(fs_info->tree_root,
1214                                     root->root_key.objectid);
1215         WARN_ON(ret);
1216         return root;
1217 fail:
1218         free_fs_root(root);
1219         return ERR_PTR(ret);
1220 }
1221
1222 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1223                                       struct btrfs_key *location,
1224                                       const char *name, int namelen)
1225 {
1226         return btrfs_read_fs_root_no_name(fs_info, location);
1227 #if 0
1228         struct btrfs_root *root;
1229         int ret;
1230
1231         root = btrfs_read_fs_root_no_name(fs_info, location);
1232         if (!root)
1233                 return NULL;
1234
1235         if (root->in_sysfs)
1236                 return root;
1237
1238         ret = btrfs_set_root_name(root, name, namelen);
1239         if (ret) {
1240                 free_extent_buffer(root->node);
1241                 kfree(root);
1242                 return ERR_PTR(ret);
1243         }
1244
1245         ret = btrfs_sysfs_add_root(root);
1246         if (ret) {
1247                 free_extent_buffer(root->node);
1248                 kfree(root->name);
1249                 kfree(root);
1250                 return ERR_PTR(ret);
1251         }
1252         root->in_sysfs = 1;
1253         return root;
1254 #endif
1255 }
1256
1257 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1258 {
1259         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1260         int ret = 0;
1261         struct btrfs_device *device;
1262         struct backing_dev_info *bdi;
1263
1264         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1265                 if (!device->bdev)
1266                         continue;
1267                 bdi = blk_get_backing_dev_info(device->bdev);
1268                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1269                         ret = 1;
1270                         break;
1271                 }
1272         }
1273         return ret;
1274 }
1275
1276 /*
1277  * this unplugs every device on the box, and it is only used when page
1278  * is null
1279  */
1280 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1281 {
1282         struct btrfs_device *device;
1283         struct btrfs_fs_info *info;
1284
1285         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1286         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1287                 if (!device->bdev)
1288                         continue;
1289
1290                 bdi = blk_get_backing_dev_info(device->bdev);
1291                 if (bdi->unplug_io_fn)
1292                         bdi->unplug_io_fn(bdi, page);
1293         }
1294 }
1295
1296 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1297 {
1298         struct inode *inode;
1299         struct extent_map_tree *em_tree;
1300         struct extent_map *em;
1301         struct address_space *mapping;
1302         u64 offset;
1303
1304         /* the generic O_DIRECT read code does this */
1305         if (1 || !page) {
1306                 __unplug_io_fn(bdi, page);
1307                 return;
1308         }
1309
1310         /*
1311          * page->mapping may change at any time.  Get a consistent copy
1312          * and use that for everything below
1313          */
1314         smp_mb();
1315         mapping = page->mapping;
1316         if (!mapping)
1317                 return;
1318
1319         inode = mapping->host;
1320
1321         /*
1322          * don't do the expensive searching for a small number of
1323          * devices
1324          */
1325         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1326                 __unplug_io_fn(bdi, page);
1327                 return;
1328         }
1329
1330         offset = page_offset(page);
1331
1332         em_tree = &BTRFS_I(inode)->extent_tree;
1333         read_lock(&em_tree->lock);
1334         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1335         read_unlock(&em_tree->lock);
1336         if (!em) {
1337                 __unplug_io_fn(bdi, page);
1338                 return;
1339         }
1340
1341         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1342                 free_extent_map(em);
1343                 __unplug_io_fn(bdi, page);
1344                 return;
1345         }
1346         offset = offset - em->start;
1347         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1348                           em->block_start + offset, page);
1349         free_extent_map(em);
1350 }
1351
1352 /*
1353  * If this fails, caller must call bdi_destroy() to get rid of the
1354  * bdi again.
1355  */
1356 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1357 {
1358         int err;
1359
1360         bdi->capabilities = BDI_CAP_MAP_COPY;
1361         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1362         if (err)
1363                 return err;
1364
1365         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1366         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1367         bdi->unplug_io_data     = info;
1368         bdi->congested_fn       = btrfs_congested_fn;
1369         bdi->congested_data     = info;
1370         return 0;
1371 }
1372
1373 static int bio_ready_for_csum(struct bio *bio)
1374 {
1375         u64 length = 0;
1376         u64 buf_len = 0;
1377         u64 start = 0;
1378         struct page *page;
1379         struct extent_io_tree *io_tree = NULL;
1380         struct btrfs_fs_info *info = NULL;
1381         struct bio_vec *bvec;
1382         int i;
1383         int ret;
1384
1385         bio_for_each_segment(bvec, bio, i) {
1386                 page = bvec->bv_page;
1387                 if (page->private == EXTENT_PAGE_PRIVATE) {
1388                         length += bvec->bv_len;
1389                         continue;
1390                 }
1391                 if (!page->private) {
1392                         length += bvec->bv_len;
1393                         continue;
1394                 }
1395                 length = bvec->bv_len;
1396                 buf_len = page->private >> 2;
1397                 start = page_offset(page) + bvec->bv_offset;
1398                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1399                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1400         }
1401         /* are we fully contained in this bio? */
1402         if (buf_len <= length)
1403                 return 1;
1404
1405         ret = extent_range_uptodate(io_tree, start + length,
1406                                     start + buf_len - 1);
1407         return ret;
1408 }
1409
1410 /*
1411  * called by the kthread helper functions to finally call the bio end_io
1412  * functions.  This is where read checksum verification actually happens
1413  */
1414 static void end_workqueue_fn(struct btrfs_work *work)
1415 {
1416         struct bio *bio;
1417         struct end_io_wq *end_io_wq;
1418         struct btrfs_fs_info *fs_info;
1419         int error;
1420
1421         end_io_wq = container_of(work, struct end_io_wq, work);
1422         bio = end_io_wq->bio;
1423         fs_info = end_io_wq->info;
1424
1425         /* metadata bio reads are special because the whole tree block must
1426          * be checksummed at once.  This makes sure the entire block is in
1427          * ram and up to date before trying to verify things.  For
1428          * blocksize <= pagesize, it is basically a noop
1429          */
1430         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1431             !bio_ready_for_csum(bio)) {
1432                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1433                                    &end_io_wq->work);
1434                 return;
1435         }
1436         error = end_io_wq->error;
1437         bio->bi_private = end_io_wq->private;
1438         bio->bi_end_io = end_io_wq->end_io;
1439         kfree(end_io_wq);
1440         bio_endio(bio, error);
1441 }
1442
1443 static int cleaner_kthread(void *arg)
1444 {
1445         struct btrfs_root *root = arg;
1446
1447         do {
1448                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1449
1450                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1451                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1452                         btrfs_run_delayed_iputs(root);
1453                         btrfs_clean_old_snapshots(root);
1454                         mutex_unlock(&root->fs_info->cleaner_mutex);
1455                 }
1456
1457                 if (freezing(current)) {
1458                         refrigerator();
1459                 } else {
1460                         set_current_state(TASK_INTERRUPTIBLE);
1461                         if (!kthread_should_stop())
1462                                 schedule();
1463                         __set_current_state(TASK_RUNNING);
1464                 }
1465         } while (!kthread_should_stop());
1466         return 0;
1467 }
1468
1469 static int transaction_kthread(void *arg)
1470 {
1471         struct btrfs_root *root = arg;
1472         struct btrfs_trans_handle *trans;
1473         struct btrfs_transaction *cur;
1474         u64 transid;
1475         unsigned long now;
1476         unsigned long delay;
1477         int ret;
1478
1479         do {
1480                 delay = HZ * 30;
1481                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1482                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1483
1484                 spin_lock(&root->fs_info->new_trans_lock);
1485                 cur = root->fs_info->running_transaction;
1486                 if (!cur) {
1487                         spin_unlock(&root->fs_info->new_trans_lock);
1488                         goto sleep;
1489                 }
1490
1491                 now = get_seconds();
1492                 if (!cur->blocked &&
1493                     (now < cur->start_time || now - cur->start_time < 30)) {
1494                         spin_unlock(&root->fs_info->new_trans_lock);
1495                         delay = HZ * 5;
1496                         goto sleep;
1497                 }
1498                 transid = cur->transid;
1499                 spin_unlock(&root->fs_info->new_trans_lock);
1500
1501                 trans = btrfs_join_transaction(root, 1);
1502                 if (transid == trans->transid) {
1503                         ret = btrfs_commit_transaction(trans, root);
1504                         BUG_ON(ret);
1505                 } else {
1506                         btrfs_end_transaction(trans, root);
1507                 }
1508 sleep:
1509                 wake_up_process(root->fs_info->cleaner_kthread);
1510                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1511
1512                 if (freezing(current)) {
1513                         refrigerator();
1514                 } else {
1515                         set_current_state(TASK_INTERRUPTIBLE);
1516                         if (!kthread_should_stop() &&
1517                             !btrfs_transaction_blocked(root->fs_info))
1518                                 schedule_timeout(delay);
1519                         __set_current_state(TASK_RUNNING);
1520                 }
1521         } while (!kthread_should_stop());
1522         return 0;
1523 }
1524
1525 struct btrfs_root *open_ctree(struct super_block *sb,
1526                               struct btrfs_fs_devices *fs_devices,
1527                               char *options)
1528 {
1529         u32 sectorsize;
1530         u32 nodesize;
1531         u32 leafsize;
1532         u32 blocksize;
1533         u32 stripesize;
1534         u64 generation;
1535         u64 features;
1536         struct btrfs_key location;
1537         struct buffer_head *bh;
1538         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1539                                                  GFP_NOFS);
1540         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1541                                                  GFP_NOFS);
1542         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1543                                                GFP_NOFS);
1544         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1545                                                 GFP_NOFS);
1546         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1547                                                 GFP_NOFS);
1548         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1549                                               GFP_NOFS);
1550         struct btrfs_root *log_tree_root;
1551
1552         int ret;
1553         int err = -EINVAL;
1554
1555         struct btrfs_super_block *disk_super;
1556
1557         if (!extent_root || !tree_root || !fs_info ||
1558             !chunk_root || !dev_root || !csum_root) {
1559                 err = -ENOMEM;
1560                 goto fail;
1561         }
1562
1563         ret = init_srcu_struct(&fs_info->subvol_srcu);
1564         if (ret) {
1565                 err = ret;
1566                 goto fail;
1567         }
1568
1569         ret = setup_bdi(fs_info, &fs_info->bdi);
1570         if (ret) {
1571                 err = ret;
1572                 goto fail_srcu;
1573         }
1574
1575         fs_info->btree_inode = new_inode(sb);
1576         if (!fs_info->btree_inode) {
1577                 err = -ENOMEM;
1578                 goto fail_bdi;
1579         }
1580
1581         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1582         INIT_LIST_HEAD(&fs_info->trans_list);
1583         INIT_LIST_HEAD(&fs_info->dead_roots);
1584         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1585         INIT_LIST_HEAD(&fs_info->hashers);
1586         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1587         INIT_LIST_HEAD(&fs_info->ordered_operations);
1588         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1589         spin_lock_init(&fs_info->delalloc_lock);
1590         spin_lock_init(&fs_info->new_trans_lock);
1591         spin_lock_init(&fs_info->ref_cache_lock);
1592         spin_lock_init(&fs_info->fs_roots_radix_lock);
1593         spin_lock_init(&fs_info->delayed_iput_lock);
1594
1595         init_completion(&fs_info->kobj_unregister);
1596         fs_info->tree_root = tree_root;
1597         fs_info->extent_root = extent_root;
1598         fs_info->csum_root = csum_root;
1599         fs_info->chunk_root = chunk_root;
1600         fs_info->dev_root = dev_root;
1601         fs_info->fs_devices = fs_devices;
1602         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1603         INIT_LIST_HEAD(&fs_info->space_info);
1604         btrfs_mapping_init(&fs_info->mapping_tree);
1605         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1606         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1607         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1608         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1609         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1610         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1611         mutex_init(&fs_info->durable_block_rsv_mutex);
1612         atomic_set(&fs_info->nr_async_submits, 0);
1613         atomic_set(&fs_info->async_delalloc_pages, 0);
1614         atomic_set(&fs_info->async_submit_draining, 0);
1615         atomic_set(&fs_info->nr_async_bios, 0);
1616         fs_info->sb = sb;
1617         fs_info->max_inline = 8192 * 1024;
1618         fs_info->metadata_ratio = 0;
1619
1620         fs_info->thread_pool_size = min_t(unsigned long,
1621                                           num_online_cpus() + 2, 8);
1622
1623         INIT_LIST_HEAD(&fs_info->ordered_extents);
1624         spin_lock_init(&fs_info->ordered_extent_lock);
1625
1626         sb->s_blocksize = 4096;
1627         sb->s_blocksize_bits = blksize_bits(4096);
1628         sb->s_bdi = &fs_info->bdi;
1629
1630         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1631         fs_info->btree_inode->i_nlink = 1;
1632         /*
1633          * we set the i_size on the btree inode to the max possible int.
1634          * the real end of the address space is determined by all of
1635          * the devices in the system
1636          */
1637         fs_info->btree_inode->i_size = OFFSET_MAX;
1638         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1639         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1640
1641         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1642         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1643                              fs_info->btree_inode->i_mapping,
1644                              GFP_NOFS);
1645         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1646                              GFP_NOFS);
1647
1648         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1649
1650         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1651         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1652                sizeof(struct btrfs_key));
1653         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1654         insert_inode_hash(fs_info->btree_inode);
1655
1656         spin_lock_init(&fs_info->block_group_cache_lock);
1657         fs_info->block_group_cache_tree = RB_ROOT;
1658
1659         extent_io_tree_init(&fs_info->freed_extents[0],
1660                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1661         extent_io_tree_init(&fs_info->freed_extents[1],
1662                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1663         fs_info->pinned_extents = &fs_info->freed_extents[0];
1664         fs_info->do_barriers = 1;
1665
1666
1667         mutex_init(&fs_info->trans_mutex);
1668         mutex_init(&fs_info->ordered_operations_mutex);
1669         mutex_init(&fs_info->tree_log_mutex);
1670         mutex_init(&fs_info->chunk_mutex);
1671         mutex_init(&fs_info->transaction_kthread_mutex);
1672         mutex_init(&fs_info->cleaner_mutex);
1673         mutex_init(&fs_info->volume_mutex);
1674         init_rwsem(&fs_info->extent_commit_sem);
1675         init_rwsem(&fs_info->cleanup_work_sem);
1676         init_rwsem(&fs_info->subvol_sem);
1677
1678         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1679         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1680
1681         init_waitqueue_head(&fs_info->transaction_throttle);
1682         init_waitqueue_head(&fs_info->transaction_wait);
1683         init_waitqueue_head(&fs_info->async_submit_wait);
1684
1685         __setup_root(4096, 4096, 4096, 4096, tree_root,
1686                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1687
1688         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1689         if (!bh)
1690                 goto fail_iput;
1691
1692         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1693         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1694                sizeof(fs_info->super_for_commit));
1695         brelse(bh);
1696
1697         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1698
1699         disk_super = &fs_info->super_copy;
1700         if (!btrfs_super_root(disk_super))
1701                 goto fail_iput;
1702
1703         ret = btrfs_parse_options(tree_root, options);
1704         if (ret) {
1705                 err = ret;
1706                 goto fail_iput;
1707         }
1708
1709         features = btrfs_super_incompat_flags(disk_super) &
1710                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1711         if (features) {
1712                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1713                        "unsupported optional features (%Lx).\n",
1714                        (unsigned long long)features);
1715                 err = -EINVAL;
1716                 goto fail_iput;
1717         }
1718
1719         features = btrfs_super_incompat_flags(disk_super);
1720         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1721                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1722                 btrfs_set_super_incompat_flags(disk_super, features);
1723         }
1724
1725         features = btrfs_super_compat_ro_flags(disk_super) &
1726                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1727         if (!(sb->s_flags & MS_RDONLY) && features) {
1728                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1729                        "unsupported option features (%Lx).\n",
1730                        (unsigned long long)features);
1731                 err = -EINVAL;
1732                 goto fail_iput;
1733         }
1734
1735         btrfs_init_workers(&fs_info->generic_worker,
1736                            "genwork", 1, NULL);
1737
1738         btrfs_init_workers(&fs_info->workers, "worker",
1739                            fs_info->thread_pool_size,
1740                            &fs_info->generic_worker);
1741
1742         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1743                            fs_info->thread_pool_size,
1744                            &fs_info->generic_worker);
1745
1746         btrfs_init_workers(&fs_info->submit_workers, "submit",
1747                            min_t(u64, fs_devices->num_devices,
1748                            fs_info->thread_pool_size),
1749                            &fs_info->generic_worker);
1750
1751         /* a higher idle thresh on the submit workers makes it much more
1752          * likely that bios will be send down in a sane order to the
1753          * devices
1754          */
1755         fs_info->submit_workers.idle_thresh = 64;
1756
1757         fs_info->workers.idle_thresh = 16;
1758         fs_info->workers.ordered = 1;
1759
1760         fs_info->delalloc_workers.idle_thresh = 2;
1761         fs_info->delalloc_workers.ordered = 1;
1762
1763         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1764                            &fs_info->generic_worker);
1765         btrfs_init_workers(&fs_info->endio_workers, "endio",
1766                            fs_info->thread_pool_size,
1767                            &fs_info->generic_worker);
1768         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1769                            fs_info->thread_pool_size,
1770                            &fs_info->generic_worker);
1771         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1772                            "endio-meta-write", fs_info->thread_pool_size,
1773                            &fs_info->generic_worker);
1774         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1775                            fs_info->thread_pool_size,
1776                            &fs_info->generic_worker);
1777
1778         /*
1779          * endios are largely parallel and should have a very
1780          * low idle thresh
1781          */
1782         fs_info->endio_workers.idle_thresh = 4;
1783         fs_info->endio_meta_workers.idle_thresh = 4;
1784
1785         fs_info->endio_write_workers.idle_thresh = 2;
1786         fs_info->endio_meta_write_workers.idle_thresh = 2;
1787
1788         btrfs_start_workers(&fs_info->workers, 1);
1789         btrfs_start_workers(&fs_info->generic_worker, 1);
1790         btrfs_start_workers(&fs_info->submit_workers, 1);
1791         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1792         btrfs_start_workers(&fs_info->fixup_workers, 1);
1793         btrfs_start_workers(&fs_info->endio_workers, 1);
1794         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1795         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1796         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1797
1798         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1799         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1800                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1801
1802         nodesize = btrfs_super_nodesize(disk_super);
1803         leafsize = btrfs_super_leafsize(disk_super);
1804         sectorsize = btrfs_super_sectorsize(disk_super);
1805         stripesize = btrfs_super_stripesize(disk_super);
1806         tree_root->nodesize = nodesize;
1807         tree_root->leafsize = leafsize;
1808         tree_root->sectorsize = sectorsize;
1809         tree_root->stripesize = stripesize;
1810
1811         sb->s_blocksize = sectorsize;
1812         sb->s_blocksize_bits = blksize_bits(sectorsize);
1813
1814         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1815                     sizeof(disk_super->magic))) {
1816                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1817                 goto fail_sb_buffer;
1818         }
1819
1820         mutex_lock(&fs_info->chunk_mutex);
1821         ret = btrfs_read_sys_array(tree_root);
1822         mutex_unlock(&fs_info->chunk_mutex);
1823         if (ret) {
1824                 printk(KERN_WARNING "btrfs: failed to read the system "
1825                        "array on %s\n", sb->s_id);
1826                 goto fail_sb_buffer;
1827         }
1828
1829         blocksize = btrfs_level_size(tree_root,
1830                                      btrfs_super_chunk_root_level(disk_super));
1831         generation = btrfs_super_chunk_root_generation(disk_super);
1832
1833         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1834                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1835
1836         chunk_root->node = read_tree_block(chunk_root,
1837                                            btrfs_super_chunk_root(disk_super),
1838                                            blocksize, generation);
1839         BUG_ON(!chunk_root->node);
1840         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1841                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1842                        sb->s_id);
1843                 goto fail_chunk_root;
1844         }
1845         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1846         chunk_root->commit_root = btrfs_root_node(chunk_root);
1847
1848         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1849            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1850            BTRFS_UUID_SIZE);
1851
1852         mutex_lock(&fs_info->chunk_mutex);
1853         ret = btrfs_read_chunk_tree(chunk_root);
1854         mutex_unlock(&fs_info->chunk_mutex);
1855         if (ret) {
1856                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1857                        sb->s_id);
1858                 goto fail_chunk_root;
1859         }
1860
1861         btrfs_close_extra_devices(fs_devices);
1862
1863         blocksize = btrfs_level_size(tree_root,
1864                                      btrfs_super_root_level(disk_super));
1865         generation = btrfs_super_generation(disk_super);
1866
1867         tree_root->node = read_tree_block(tree_root,
1868                                           btrfs_super_root(disk_super),
1869                                           blocksize, generation);
1870         if (!tree_root->node)
1871                 goto fail_chunk_root;
1872         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1873                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1874                        sb->s_id);
1875                 goto fail_tree_root;
1876         }
1877         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1878         tree_root->commit_root = btrfs_root_node(tree_root);
1879
1880         ret = find_and_setup_root(tree_root, fs_info,
1881                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1882         if (ret)
1883                 goto fail_tree_root;
1884         extent_root->track_dirty = 1;
1885
1886         ret = find_and_setup_root(tree_root, fs_info,
1887                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1888         if (ret)
1889                 goto fail_extent_root;
1890         dev_root->track_dirty = 1;
1891
1892         ret = find_and_setup_root(tree_root, fs_info,
1893                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1894         if (ret)
1895                 goto fail_dev_root;
1896
1897         csum_root->track_dirty = 1;
1898
1899         fs_info->generation = generation;
1900         fs_info->last_trans_committed = generation;
1901         fs_info->data_alloc_profile = (u64)-1;
1902         fs_info->metadata_alloc_profile = (u64)-1;
1903         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1904
1905         ret = btrfs_read_block_groups(extent_root);
1906         if (ret) {
1907                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1908                 goto fail_block_groups;
1909         }
1910
1911         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1912                                                "btrfs-cleaner");
1913         if (IS_ERR(fs_info->cleaner_kthread))
1914                 goto fail_block_groups;
1915
1916         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1917                                                    tree_root,
1918                                                    "btrfs-transaction");
1919         if (IS_ERR(fs_info->transaction_kthread))
1920                 goto fail_cleaner;
1921
1922         if (!btrfs_test_opt(tree_root, SSD) &&
1923             !btrfs_test_opt(tree_root, NOSSD) &&
1924             !fs_info->fs_devices->rotating) {
1925                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1926                        "mode\n");
1927                 btrfs_set_opt(fs_info->mount_opt, SSD);
1928         }
1929
1930         if (btrfs_super_log_root(disk_super) != 0) {
1931                 u64 bytenr = btrfs_super_log_root(disk_super);
1932
1933                 if (fs_devices->rw_devices == 0) {
1934                         printk(KERN_WARNING "Btrfs log replay required "
1935                                "on RO media\n");
1936                         err = -EIO;
1937                         goto fail_trans_kthread;
1938                 }
1939                 blocksize =
1940                      btrfs_level_size(tree_root,
1941                                       btrfs_super_log_root_level(disk_super));
1942
1943                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1944                 if (!log_tree_root) {
1945                         err = -ENOMEM;
1946                         goto fail_trans_kthread;
1947                 }
1948
1949                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1950                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1951
1952                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1953                                                       blocksize,
1954                                                       generation + 1);
1955                 ret = btrfs_recover_log_trees(log_tree_root);
1956                 BUG_ON(ret);
1957
1958                 if (sb->s_flags & MS_RDONLY) {
1959                         ret =  btrfs_commit_super(tree_root);
1960                         BUG_ON(ret);
1961                 }
1962         }
1963
1964         ret = btrfs_find_orphan_roots(tree_root);
1965         BUG_ON(ret);
1966
1967         if (!(sb->s_flags & MS_RDONLY)) {
1968                 ret = btrfs_cleanup_fs_roots(fs_info);
1969                 BUG_ON(ret);
1970
1971                 ret = btrfs_recover_relocation(tree_root);
1972                 if (ret < 0) {
1973                         printk(KERN_WARNING
1974                                "btrfs: failed to recover relocation\n");
1975                         err = -EINVAL;
1976                         goto fail_trans_kthread;
1977                 }
1978         }
1979
1980         location.objectid = BTRFS_FS_TREE_OBJECTID;
1981         location.type = BTRFS_ROOT_ITEM_KEY;
1982         location.offset = (u64)-1;
1983
1984         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1985         if (!fs_info->fs_root)
1986                 goto fail_trans_kthread;
1987         if (IS_ERR(fs_info->fs_root)) {
1988                 err = PTR_ERR(fs_info->fs_root);
1989                 goto fail_trans_kthread;
1990         }
1991
1992         if (!(sb->s_flags & MS_RDONLY)) {
1993                 down_read(&fs_info->cleanup_work_sem);
1994                 btrfs_orphan_cleanup(fs_info->fs_root);
1995                 btrfs_orphan_cleanup(fs_info->tree_root);
1996                 up_read(&fs_info->cleanup_work_sem);
1997         }
1998
1999         return tree_root;
2000
2001 fail_trans_kthread:
2002         kthread_stop(fs_info->transaction_kthread);
2003 fail_cleaner:
2004         kthread_stop(fs_info->cleaner_kthread);
2005
2006         /*
2007          * make sure we're done with the btree inode before we stop our
2008          * kthreads
2009          */
2010         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2011         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2012
2013 fail_block_groups:
2014         btrfs_free_block_groups(fs_info);
2015         free_extent_buffer(csum_root->node);
2016         free_extent_buffer(csum_root->commit_root);
2017 fail_dev_root:
2018         free_extent_buffer(dev_root->node);
2019         free_extent_buffer(dev_root->commit_root);
2020 fail_extent_root:
2021         free_extent_buffer(extent_root->node);
2022         free_extent_buffer(extent_root->commit_root);
2023 fail_tree_root:
2024         free_extent_buffer(tree_root->node);
2025         free_extent_buffer(tree_root->commit_root);
2026 fail_chunk_root:
2027         free_extent_buffer(chunk_root->node);
2028         free_extent_buffer(chunk_root->commit_root);
2029 fail_sb_buffer:
2030         btrfs_stop_workers(&fs_info->generic_worker);
2031         btrfs_stop_workers(&fs_info->fixup_workers);
2032         btrfs_stop_workers(&fs_info->delalloc_workers);
2033         btrfs_stop_workers(&fs_info->workers);
2034         btrfs_stop_workers(&fs_info->endio_workers);
2035         btrfs_stop_workers(&fs_info->endio_meta_workers);
2036         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2037         btrfs_stop_workers(&fs_info->endio_write_workers);
2038         btrfs_stop_workers(&fs_info->submit_workers);
2039 fail_iput:
2040         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2041         iput(fs_info->btree_inode);
2042
2043         btrfs_close_devices(fs_info->fs_devices);
2044         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2045 fail_bdi:
2046         bdi_destroy(&fs_info->bdi);
2047 fail_srcu:
2048         cleanup_srcu_struct(&fs_info->subvol_srcu);
2049 fail:
2050         kfree(extent_root);
2051         kfree(tree_root);
2052         kfree(fs_info);
2053         kfree(chunk_root);
2054         kfree(dev_root);
2055         kfree(csum_root);
2056         return ERR_PTR(err);
2057 }
2058
2059 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2060 {
2061         char b[BDEVNAME_SIZE];
2062
2063         if (uptodate) {
2064                 set_buffer_uptodate(bh);
2065         } else {
2066                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2067                         printk(KERN_WARNING "lost page write due to "
2068                                         "I/O error on %s\n",
2069                                        bdevname(bh->b_bdev, b));
2070                 }
2071                 /* note, we dont' set_buffer_write_io_error because we have
2072                  * our own ways of dealing with the IO errors
2073                  */
2074                 clear_buffer_uptodate(bh);
2075         }
2076         unlock_buffer(bh);
2077         put_bh(bh);
2078 }
2079
2080 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2081 {
2082         struct buffer_head *bh;
2083         struct buffer_head *latest = NULL;
2084         struct btrfs_super_block *super;
2085         int i;
2086         u64 transid = 0;
2087         u64 bytenr;
2088
2089         /* we would like to check all the supers, but that would make
2090          * a btrfs mount succeed after a mkfs from a different FS.
2091          * So, we need to add a special mount option to scan for
2092          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2093          */
2094         for (i = 0; i < 1; i++) {
2095                 bytenr = btrfs_sb_offset(i);
2096                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2097                         break;
2098                 bh = __bread(bdev, bytenr / 4096, 4096);
2099                 if (!bh)
2100                         continue;
2101
2102                 super = (struct btrfs_super_block *)bh->b_data;
2103                 if (btrfs_super_bytenr(super) != bytenr ||
2104                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2105                             sizeof(super->magic))) {
2106                         brelse(bh);
2107                         continue;
2108                 }
2109
2110                 if (!latest || btrfs_super_generation(super) > transid) {
2111                         brelse(latest);
2112                         latest = bh;
2113                         transid = btrfs_super_generation(super);
2114                 } else {
2115                         brelse(bh);
2116                 }
2117         }
2118         return latest;
2119 }
2120
2121 /*
2122  * this should be called twice, once with wait == 0 and
2123  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2124  * we write are pinned.
2125  *
2126  * They are released when wait == 1 is done.
2127  * max_mirrors must be the same for both runs, and it indicates how
2128  * many supers on this one device should be written.
2129  *
2130  * max_mirrors == 0 means to write them all.
2131  */
2132 static int write_dev_supers(struct btrfs_device *device,
2133                             struct btrfs_super_block *sb,
2134                             int do_barriers, int wait, int max_mirrors)
2135 {
2136         struct buffer_head *bh;
2137         int i;
2138         int ret;
2139         int errors = 0;
2140         u32 crc;
2141         u64 bytenr;
2142         int last_barrier = 0;
2143
2144         if (max_mirrors == 0)
2145                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2146
2147         /* make sure only the last submit_bh does a barrier */
2148         if (do_barriers) {
2149                 for (i = 0; i < max_mirrors; i++) {
2150                         bytenr = btrfs_sb_offset(i);
2151                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2152                             device->total_bytes)
2153                                 break;
2154                         last_barrier = i;
2155                 }
2156         }
2157
2158         for (i = 0; i < max_mirrors; i++) {
2159                 bytenr = btrfs_sb_offset(i);
2160                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2161                         break;
2162
2163                 if (wait) {
2164                         bh = __find_get_block(device->bdev, bytenr / 4096,
2165                                               BTRFS_SUPER_INFO_SIZE);
2166                         BUG_ON(!bh);
2167                         wait_on_buffer(bh);
2168                         if (!buffer_uptodate(bh))
2169                                 errors++;
2170
2171                         /* drop our reference */
2172                         brelse(bh);
2173
2174                         /* drop the reference from the wait == 0 run */
2175                         brelse(bh);
2176                         continue;
2177                 } else {
2178                         btrfs_set_super_bytenr(sb, bytenr);
2179
2180                         crc = ~(u32)0;
2181                         crc = btrfs_csum_data(NULL, (char *)sb +
2182                                               BTRFS_CSUM_SIZE, crc,
2183                                               BTRFS_SUPER_INFO_SIZE -
2184                                               BTRFS_CSUM_SIZE);
2185                         btrfs_csum_final(crc, sb->csum);
2186
2187                         /*
2188                          * one reference for us, and we leave it for the
2189                          * caller
2190                          */
2191                         bh = __getblk(device->bdev, bytenr / 4096,
2192                                       BTRFS_SUPER_INFO_SIZE);
2193                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2194
2195                         /* one reference for submit_bh */
2196                         get_bh(bh);
2197
2198                         set_buffer_uptodate(bh);
2199                         lock_buffer(bh);
2200                         bh->b_end_io = btrfs_end_buffer_write_sync;
2201                 }
2202
2203                 if (i == last_barrier && do_barriers && device->barriers) {
2204                         ret = submit_bh(WRITE_BARRIER, bh);
2205                         if (ret == -EOPNOTSUPP) {
2206                                 printk("btrfs: disabling barriers on dev %s\n",
2207                                        device->name);
2208                                 set_buffer_uptodate(bh);
2209                                 device->barriers = 0;
2210                                 /* one reference for submit_bh */
2211                                 get_bh(bh);
2212                                 lock_buffer(bh);
2213                                 ret = submit_bh(WRITE_SYNC, bh);
2214                         }
2215                 } else {
2216                         ret = submit_bh(WRITE_SYNC, bh);
2217                 }
2218
2219                 if (ret)
2220                         errors++;
2221         }
2222         return errors < i ? 0 : -1;
2223 }
2224
2225 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2226 {
2227         struct list_head *head;
2228         struct btrfs_device *dev;
2229         struct btrfs_super_block *sb;
2230         struct btrfs_dev_item *dev_item;
2231         int ret;
2232         int do_barriers;
2233         int max_errors;
2234         int total_errors = 0;
2235         u64 flags;
2236
2237         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2238         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2239
2240         sb = &root->fs_info->super_for_commit;
2241         dev_item = &sb->dev_item;
2242
2243         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2244         head = &root->fs_info->fs_devices->devices;
2245         list_for_each_entry(dev, head, dev_list) {
2246                 if (!dev->bdev) {
2247                         total_errors++;
2248                         continue;
2249                 }
2250                 if (!dev->in_fs_metadata || !dev->writeable)
2251                         continue;
2252
2253                 btrfs_set_stack_device_generation(dev_item, 0);
2254                 btrfs_set_stack_device_type(dev_item, dev->type);
2255                 btrfs_set_stack_device_id(dev_item, dev->devid);
2256                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2257                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2258                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2259                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2260                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2261                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2262                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2263
2264                 flags = btrfs_super_flags(sb);
2265                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2266
2267                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2268                 if (ret)
2269                         total_errors++;
2270         }
2271         if (total_errors > max_errors) {
2272                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2273                        total_errors);
2274                 BUG();
2275         }
2276
2277         total_errors = 0;
2278         list_for_each_entry(dev, head, dev_list) {
2279                 if (!dev->bdev)
2280                         continue;
2281                 if (!dev->in_fs_metadata || !dev->writeable)
2282                         continue;
2283
2284                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2285                 if (ret)
2286                         total_errors++;
2287         }
2288         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2289         if (total_errors > max_errors) {
2290                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2291                        total_errors);
2292                 BUG();
2293         }
2294         return 0;
2295 }
2296
2297 int write_ctree_super(struct btrfs_trans_handle *trans,
2298                       struct btrfs_root *root, int max_mirrors)
2299 {
2300         int ret;
2301
2302         ret = write_all_supers(root, max_mirrors);
2303         return ret;
2304 }
2305
2306 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2307 {
2308         spin_lock(&fs_info->fs_roots_radix_lock);
2309         radix_tree_delete(&fs_info->fs_roots_radix,
2310                           (unsigned long)root->root_key.objectid);
2311         spin_unlock(&fs_info->fs_roots_radix_lock);
2312
2313         if (btrfs_root_refs(&root->root_item) == 0)
2314                 synchronize_srcu(&fs_info->subvol_srcu);
2315
2316         free_fs_root(root);
2317         return 0;
2318 }
2319
2320 static void free_fs_root(struct btrfs_root *root)
2321 {
2322         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2323         if (root->anon_super.s_dev) {
2324                 down_write(&root->anon_super.s_umount);
2325                 kill_anon_super(&root->anon_super);
2326         }
2327         free_extent_buffer(root->node);
2328         free_extent_buffer(root->commit_root);
2329         kfree(root->name);
2330         kfree(root);
2331 }
2332
2333 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2334 {
2335         int ret;
2336         struct btrfs_root *gang[8];
2337         int i;
2338
2339         while (!list_empty(&fs_info->dead_roots)) {
2340                 gang[0] = list_entry(fs_info->dead_roots.next,
2341                                      struct btrfs_root, root_list);
2342                 list_del(&gang[0]->root_list);
2343
2344                 if (gang[0]->in_radix) {
2345                         btrfs_free_fs_root(fs_info, gang[0]);
2346                 } else {
2347                         free_extent_buffer(gang[0]->node);
2348                         free_extent_buffer(gang[0]->commit_root);
2349                         kfree(gang[0]);
2350                 }
2351         }
2352
2353         while (1) {
2354                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2355                                              (void **)gang, 0,
2356                                              ARRAY_SIZE(gang));
2357                 if (!ret)
2358                         break;
2359                 for (i = 0; i < ret; i++)
2360                         btrfs_free_fs_root(fs_info, gang[i]);
2361         }
2362         return 0;
2363 }
2364
2365 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2366 {
2367         u64 root_objectid = 0;
2368         struct btrfs_root *gang[8];
2369         int i;
2370         int ret;
2371
2372         while (1) {
2373                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2374                                              (void **)gang, root_objectid,
2375                                              ARRAY_SIZE(gang));
2376                 if (!ret)
2377                         break;
2378
2379                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2380                 for (i = 0; i < ret; i++) {
2381                         root_objectid = gang[i]->root_key.objectid;
2382                         btrfs_orphan_cleanup(gang[i]);
2383                 }
2384                 root_objectid++;
2385         }
2386         return 0;
2387 }
2388
2389 int btrfs_commit_super(struct btrfs_root *root)
2390 {
2391         struct btrfs_trans_handle *trans;
2392         int ret;
2393
2394         mutex_lock(&root->fs_info->cleaner_mutex);
2395         btrfs_run_delayed_iputs(root);
2396         btrfs_clean_old_snapshots(root);
2397         mutex_unlock(&root->fs_info->cleaner_mutex);
2398
2399         /* wait until ongoing cleanup work done */
2400         down_write(&root->fs_info->cleanup_work_sem);
2401         up_write(&root->fs_info->cleanup_work_sem);
2402
2403         trans = btrfs_join_transaction(root, 1);
2404         ret = btrfs_commit_transaction(trans, root);
2405         BUG_ON(ret);
2406         /* run commit again to drop the original snapshot */
2407         trans = btrfs_join_transaction(root, 1);
2408         btrfs_commit_transaction(trans, root);
2409         ret = btrfs_write_and_wait_transaction(NULL, root);
2410         BUG_ON(ret);
2411
2412         ret = write_ctree_super(NULL, root, 0);
2413         return ret;
2414 }
2415
2416 int close_ctree(struct btrfs_root *root)
2417 {
2418         struct btrfs_fs_info *fs_info = root->fs_info;
2419         int ret;
2420
2421         fs_info->closing = 1;
2422         smp_mb();
2423
2424         btrfs_put_block_group_cache(fs_info);
2425         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2426                 ret =  btrfs_commit_super(root);
2427                 if (ret)
2428                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2429         }
2430
2431         kthread_stop(root->fs_info->transaction_kthread);
2432         kthread_stop(root->fs_info->cleaner_kthread);
2433
2434         fs_info->closing = 2;
2435         smp_mb();
2436
2437         if (fs_info->delalloc_bytes) {
2438                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2439                        (unsigned long long)fs_info->delalloc_bytes);
2440         }
2441         if (fs_info->total_ref_cache_size) {
2442                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2443                        (unsigned long long)fs_info->total_ref_cache_size);
2444         }
2445
2446         free_extent_buffer(fs_info->extent_root->node);
2447         free_extent_buffer(fs_info->extent_root->commit_root);
2448         free_extent_buffer(fs_info->tree_root->node);
2449         free_extent_buffer(fs_info->tree_root->commit_root);
2450         free_extent_buffer(root->fs_info->chunk_root->node);
2451         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2452         free_extent_buffer(root->fs_info->dev_root->node);
2453         free_extent_buffer(root->fs_info->dev_root->commit_root);
2454         free_extent_buffer(root->fs_info->csum_root->node);
2455         free_extent_buffer(root->fs_info->csum_root->commit_root);
2456
2457         btrfs_free_block_groups(root->fs_info);
2458
2459         del_fs_roots(fs_info);
2460
2461         iput(fs_info->btree_inode);
2462
2463         btrfs_stop_workers(&fs_info->generic_worker);
2464         btrfs_stop_workers(&fs_info->fixup_workers);
2465         btrfs_stop_workers(&fs_info->delalloc_workers);
2466         btrfs_stop_workers(&fs_info->workers);
2467         btrfs_stop_workers(&fs_info->endio_workers);
2468         btrfs_stop_workers(&fs_info->endio_meta_workers);
2469         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2470         btrfs_stop_workers(&fs_info->endio_write_workers);
2471         btrfs_stop_workers(&fs_info->submit_workers);
2472
2473         btrfs_close_devices(fs_info->fs_devices);
2474         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2475
2476         bdi_destroy(&fs_info->bdi);
2477         cleanup_srcu_struct(&fs_info->subvol_srcu);
2478
2479         kfree(fs_info->extent_root);
2480         kfree(fs_info->tree_root);
2481         kfree(fs_info->chunk_root);
2482         kfree(fs_info->dev_root);
2483         kfree(fs_info->csum_root);
2484         return 0;
2485 }
2486
2487 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2488 {
2489         int ret;
2490         struct inode *btree_inode = buf->first_page->mapping->host;
2491
2492         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2493                                      NULL);
2494         if (!ret)
2495                 return ret;
2496
2497         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2498                                     parent_transid);
2499         return !ret;
2500 }
2501
2502 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2503 {
2504         struct inode *btree_inode = buf->first_page->mapping->host;
2505         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2506                                           buf);
2507 }
2508
2509 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2510 {
2511         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2512         u64 transid = btrfs_header_generation(buf);
2513         struct inode *btree_inode = root->fs_info->btree_inode;
2514         int was_dirty;
2515
2516         btrfs_assert_tree_locked(buf);
2517         if (transid != root->fs_info->generation) {
2518                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2519                        "found %llu running %llu\n",
2520                         (unsigned long long)buf->start,
2521                         (unsigned long long)transid,
2522                         (unsigned long long)root->fs_info->generation);
2523                 WARN_ON(1);
2524         }
2525         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2526                                             buf);
2527         if (!was_dirty) {
2528                 spin_lock(&root->fs_info->delalloc_lock);
2529                 root->fs_info->dirty_metadata_bytes += buf->len;
2530                 spin_unlock(&root->fs_info->delalloc_lock);
2531         }
2532 }
2533
2534 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2535 {
2536         /*
2537          * looks as though older kernels can get into trouble with
2538          * this code, they end up stuck in balance_dirty_pages forever
2539          */
2540         u64 num_dirty;
2541         unsigned long thresh = 32 * 1024 * 1024;
2542
2543         if (current->flags & PF_MEMALLOC)
2544                 return;
2545
2546         num_dirty = root->fs_info->dirty_metadata_bytes;
2547
2548         if (num_dirty > thresh) {
2549                 balance_dirty_pages_ratelimited_nr(
2550                                    root->fs_info->btree_inode->i_mapping, 1);
2551         }
2552         return;
2553 }
2554
2555 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2556 {
2557         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2558         int ret;
2559         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2560         if (ret == 0)
2561                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2562         return ret;
2563 }
2564
2565 int btree_lock_page_hook(struct page *page)
2566 {
2567         struct inode *inode = page->mapping->host;
2568         struct btrfs_root *root = BTRFS_I(inode)->root;
2569         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2570         struct extent_buffer *eb;
2571         unsigned long len;
2572         u64 bytenr = page_offset(page);
2573
2574         if (page->private == EXTENT_PAGE_PRIVATE)
2575                 goto out;
2576
2577         len = page->private >> 2;
2578         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2579         if (!eb)
2580                 goto out;
2581
2582         btrfs_tree_lock(eb);
2583         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2584
2585         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2586                 spin_lock(&root->fs_info->delalloc_lock);
2587                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2588                         root->fs_info->dirty_metadata_bytes -= eb->len;
2589                 else
2590                         WARN_ON(1);
2591                 spin_unlock(&root->fs_info->delalloc_lock);
2592         }
2593
2594         btrfs_tree_unlock(eb);
2595         free_extent_buffer(eb);
2596 out:
2597         lock_page(page);
2598         return 0;
2599 }
2600
2601 static struct extent_io_ops btree_extent_io_ops = {
2602         .write_cache_pages_lock_hook = btree_lock_page_hook,
2603         .readpage_end_io_hook = btree_readpage_end_io_hook,
2604         .submit_bio_hook = btree_submit_bio_hook,
2605         /* note we're sharing with inode.c for the merge bio hook */
2606         .merge_bio_hook = btrfs_merge_bio_hook,
2607 };