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