Merge branch 'for-linus' of master.kernel.org:/home/rmk/linux-2.6-arm
[linux-2.6] / 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 /*
46  * end_io_wq structs are used to do processing in task context when an IO is
47  * complete.  This is used during reads to verify checksums, and it is used
48  * by writes to insert metadata for new file extents after IO is complete.
49  */
50 struct end_io_wq {
51         struct bio *bio;
52         bio_end_io_t *end_io;
53         void *private;
54         struct btrfs_fs_info *info;
55         int error;
56         int metadata;
57         struct list_head list;
58         struct btrfs_work work;
59 };
60
61 /*
62  * async submit bios are used to offload expensive checksumming
63  * onto the worker threads.  They checksum file and metadata bios
64  * just before they are sent down the IO stack.
65  */
66 struct async_submit_bio {
67         struct inode *inode;
68         struct bio *bio;
69         struct list_head list;
70         extent_submit_bio_hook_t *submit_bio_start;
71         extent_submit_bio_hook_t *submit_bio_done;
72         int rw;
73         int mirror_num;
74         unsigned long bio_flags;
75         struct btrfs_work work;
76 };
77
78 /* These are used to set the lockdep class on the extent buffer locks.
79  * The class is set by the readpage_end_io_hook after the buffer has
80  * passed csum validation but before the pages are unlocked.
81  *
82  * The lockdep class is also set by btrfs_init_new_buffer on freshly
83  * allocated blocks.
84  *
85  * The class is based on the level in the tree block, which allows lockdep
86  * to know that lower nodes nest inside the locks of higher nodes.
87  *
88  * We also add a check to make sure the highest level of the tree is
89  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
90  * code needs update as well.
91  */
92 #ifdef CONFIG_DEBUG_LOCK_ALLOC
93 # if BTRFS_MAX_LEVEL != 8
94 #  error
95 # endif
96 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
97 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
98         /* leaf */
99         "btrfs-extent-00",
100         "btrfs-extent-01",
101         "btrfs-extent-02",
102         "btrfs-extent-03",
103         "btrfs-extent-04",
104         "btrfs-extent-05",
105         "btrfs-extent-06",
106         "btrfs-extent-07",
107         /* highest possible level */
108         "btrfs-extent-08",
109 };
110 #endif
111
112 /*
113  * extents on the btree inode are pretty simple, there's one extent
114  * that covers the entire device
115  */
116 static struct extent_map *btree_get_extent(struct inode *inode,
117                 struct page *page, size_t page_offset, u64 start, u64 len,
118                 int create)
119 {
120         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
121         struct extent_map *em;
122         int ret;
123
124         spin_lock(&em_tree->lock);
125         em = lookup_extent_mapping(em_tree, start, len);
126         if (em) {
127                 em->bdev =
128                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
129                 spin_unlock(&em_tree->lock);
130                 goto out;
131         }
132         spin_unlock(&em_tree->lock);
133
134         em = alloc_extent_map(GFP_NOFS);
135         if (!em) {
136                 em = ERR_PTR(-ENOMEM);
137                 goto out;
138         }
139         em->start = 0;
140         em->len = (u64)-1;
141         em->block_len = (u64)-1;
142         em->block_start = 0;
143         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
144
145         spin_lock(&em_tree->lock);
146         ret = add_extent_mapping(em_tree, em);
147         if (ret == -EEXIST) {
148                 u64 failed_start = em->start;
149                 u64 failed_len = em->len;
150
151                 free_extent_map(em);
152                 em = lookup_extent_mapping(em_tree, start, len);
153                 if (em) {
154                         ret = 0;
155                 } else {
156                         em = lookup_extent_mapping(em_tree, failed_start,
157                                                    failed_len);
158                         ret = -EIO;
159                 }
160         } else if (ret) {
161                 free_extent_map(em);
162                 em = NULL;
163         }
164         spin_unlock(&em_tree->lock);
165
166         if (ret)
167                 em = ERR_PTR(ret);
168 out:
169         return em;
170 }
171
172 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
173 {
174         return crc32c(seed, data, len);
175 }
176
177 void btrfs_csum_final(u32 crc, char *result)
178 {
179         *(__le32 *)result = ~cpu_to_le32(crc);
180 }
181
182 /*
183  * compute the csum for a btree block, and either verify it or write it
184  * into the csum field of the block.
185  */
186 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
187                            int verify)
188 {
189         u16 csum_size =
190                 btrfs_super_csum_size(&root->fs_info->super_copy);
191         char *result = NULL;
192         unsigned long len;
193         unsigned long cur_len;
194         unsigned long offset = BTRFS_CSUM_SIZE;
195         char *map_token = NULL;
196         char *kaddr;
197         unsigned long map_start;
198         unsigned long map_len;
199         int err;
200         u32 crc = ~(u32)0;
201         unsigned long inline_result;
202
203         len = buf->len - offset;
204         while (len > 0) {
205                 err = map_private_extent_buffer(buf, offset, 32,
206                                         &map_token, &kaddr,
207                                         &map_start, &map_len, KM_USER0);
208                 if (err)
209                         return 1;
210                 cur_len = min(len, map_len - (offset - map_start));
211                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
212                                       crc, cur_len);
213                 len -= cur_len;
214                 offset += cur_len;
215                 unmap_extent_buffer(buf, map_token, KM_USER0);
216         }
217         if (csum_size > sizeof(inline_result)) {
218                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
219                 if (!result)
220                         return 1;
221         } else {
222                 result = (char *)&inline_result;
223         }
224
225         btrfs_csum_final(crc, result);
226
227         if (verify) {
228                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
229                         u32 val;
230                         u32 found = 0;
231                         memcpy(&found, result, csum_size);
232
233                         read_extent_buffer(buf, &val, 0, csum_size);
234                         if (printk_ratelimit()) {
235                                 printk(KERN_INFO "btrfs: %s checksum verify "
236                                        "failed on %llu wanted %X found %X "
237                                        "level %d\n",
238                                        root->fs_info->sb->s_id,
239                                        (unsigned long long)buf->start, val, found,
240                                        btrfs_header_level(buf));
241                         }
242                         if (result != (char *)&inline_result)
243                                 kfree(result);
244                         return 1;
245                 }
246         } else {
247                 write_extent_buffer(buf, result, 0, csum_size);
248         }
249         if (result != (char *)&inline_result)
250                 kfree(result);
251         return 0;
252 }
253
254 /*
255  * we can't consider a given block up to date unless the transid of the
256  * block matches the transid in the parent node's pointer.  This is how we
257  * detect blocks that either didn't get written at all or got written
258  * in the wrong place.
259  */
260 static int verify_parent_transid(struct extent_io_tree *io_tree,
261                                  struct extent_buffer *eb, u64 parent_transid)
262 {
263         int ret;
264
265         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
266                 return 0;
267
268         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
269         if (extent_buffer_uptodate(io_tree, eb) &&
270             btrfs_header_generation(eb) == parent_transid) {
271                 ret = 0;
272                 goto out;
273         }
274         if (printk_ratelimit()) {
275                 printk("parent transid verify failed on %llu wanted %llu "
276                        "found %llu\n",
277                        (unsigned long long)eb->start,
278                        (unsigned long long)parent_transid,
279                        (unsigned long long)btrfs_header_generation(eb));
280         }
281         ret = 1;
282         clear_extent_buffer_uptodate(io_tree, eb);
283 out:
284         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
285                       GFP_NOFS);
286         return ret;
287 }
288
289 /*
290  * helper to read a given tree block, doing retries as required when
291  * the checksums don't match and we have alternate mirrors to try.
292  */
293 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
294                                           struct extent_buffer *eb,
295                                           u64 start, u64 parent_transid)
296 {
297         struct extent_io_tree *io_tree;
298         int ret;
299         int num_copies = 0;
300         int mirror_num = 0;
301
302         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
303         while (1) {
304                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
305                                                btree_get_extent, mirror_num);
306                 if (!ret &&
307                     !verify_parent_transid(io_tree, eb, parent_transid))
308                         return ret;
309
310                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
311                                               eb->start, eb->len);
312                 if (num_copies == 1)
313                         return ret;
314
315                 mirror_num++;
316                 if (mirror_num > num_copies)
317                         return ret;
318         }
319         return -EIO;
320 }
321
322 /*
323  * checksum a dirty tree block before IO.  This has extra checks to make sure
324  * we only fill in the checksum field in the first page of a multi-page block
325  */
326
327 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
328 {
329         struct extent_io_tree *tree;
330         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
331         u64 found_start;
332         int found_level;
333         unsigned long len;
334         struct extent_buffer *eb;
335         int ret;
336
337         tree = &BTRFS_I(page->mapping->host)->io_tree;
338
339         if (page->private == EXTENT_PAGE_PRIVATE)
340                 goto out;
341         if (!page->private)
342                 goto out;
343         len = page->private >> 2;
344         WARN_ON(len == 0);
345
346         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
347         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
348                                              btrfs_header_generation(eb));
349         BUG_ON(ret);
350         found_start = btrfs_header_bytenr(eb);
351         if (found_start != start) {
352                 WARN_ON(1);
353                 goto err;
354         }
355         if (eb->first_page != page) {
356                 WARN_ON(1);
357                 goto err;
358         }
359         if (!PageUptodate(page)) {
360                 WARN_ON(1);
361                 goto err;
362         }
363         found_level = btrfs_header_level(eb);
364
365         csum_tree_block(root, eb, 0);
366 err:
367         free_extent_buffer(eb);
368 out:
369         return 0;
370 }
371
372 static int check_tree_block_fsid(struct btrfs_root *root,
373                                  struct extent_buffer *eb)
374 {
375         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
376         u8 fsid[BTRFS_UUID_SIZE];
377         int ret = 1;
378
379         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
380                            BTRFS_FSID_SIZE);
381         while (fs_devices) {
382                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
383                         ret = 0;
384                         break;
385                 }
386                 fs_devices = fs_devices->seed;
387         }
388         return ret;
389 }
390
391 #ifdef CONFIG_DEBUG_LOCK_ALLOC
392 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
393 {
394         lockdep_set_class_and_name(&eb->lock,
395                            &btrfs_eb_class[level],
396                            btrfs_eb_name[level]);
397 }
398 #endif
399
400 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
401                                struct extent_state *state)
402 {
403         struct extent_io_tree *tree;
404         u64 found_start;
405         int found_level;
406         unsigned long len;
407         struct extent_buffer *eb;
408         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
409         int ret = 0;
410
411         tree = &BTRFS_I(page->mapping->host)->io_tree;
412         if (page->private == EXTENT_PAGE_PRIVATE)
413                 goto out;
414         if (!page->private)
415                 goto out;
416
417         len = page->private >> 2;
418         WARN_ON(len == 0);
419
420         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
421
422         found_start = btrfs_header_bytenr(eb);
423         if (found_start != start) {
424                 if (printk_ratelimit()) {
425                         printk(KERN_INFO "btrfs bad tree block start "
426                                "%llu %llu\n",
427                                (unsigned long long)found_start,
428                                (unsigned long long)eb->start);
429                 }
430                 ret = -EIO;
431                 goto err;
432         }
433         if (eb->first_page != page) {
434                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
435                        eb->first_page->index, page->index);
436                 WARN_ON(1);
437                 ret = -EIO;
438                 goto err;
439         }
440         if (check_tree_block_fsid(root, eb)) {
441                 if (printk_ratelimit()) {
442                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
443                                (unsigned long long)eb->start);
444                 }
445                 ret = -EIO;
446                 goto err;
447         }
448         found_level = btrfs_header_level(eb);
449
450         btrfs_set_buffer_lockdep_class(eb, found_level);
451
452         ret = csum_tree_block(root, eb, 1);
453         if (ret)
454                 ret = -EIO;
455
456         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
457         end = eb->start + end - 1;
458 err:
459         free_extent_buffer(eb);
460 out:
461         return ret;
462 }
463
464 static void end_workqueue_bio(struct bio *bio, int err)
465 {
466         struct end_io_wq *end_io_wq = bio->bi_private;
467         struct btrfs_fs_info *fs_info;
468
469         fs_info = end_io_wq->info;
470         end_io_wq->error = err;
471         end_io_wq->work.func = end_workqueue_fn;
472         end_io_wq->work.flags = 0;
473
474         if (bio->bi_rw & (1 << BIO_RW)) {
475                 if (end_io_wq->metadata)
476                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
477                                            &end_io_wq->work);
478                 else
479                         btrfs_queue_worker(&fs_info->endio_write_workers,
480                                            &end_io_wq->work);
481         } else {
482                 if (end_io_wq->metadata)
483                         btrfs_queue_worker(&fs_info->endio_meta_workers,
484                                            &end_io_wq->work);
485                 else
486                         btrfs_queue_worker(&fs_info->endio_workers,
487                                            &end_io_wq->work);
488         }
489 }
490
491 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
492                         int metadata)
493 {
494         struct end_io_wq *end_io_wq;
495         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
496         if (!end_io_wq)
497                 return -ENOMEM;
498
499         end_io_wq->private = bio->bi_private;
500         end_io_wq->end_io = bio->bi_end_io;
501         end_io_wq->info = info;
502         end_io_wq->error = 0;
503         end_io_wq->bio = bio;
504         end_io_wq->metadata = metadata;
505
506         bio->bi_private = end_io_wq;
507         bio->bi_end_io = end_workqueue_bio;
508         return 0;
509 }
510
511 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
512 {
513         unsigned long limit = min_t(unsigned long,
514                                     info->workers.max_workers,
515                                     info->fs_devices->open_devices);
516         return 256 * limit;
517 }
518
519 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
520 {
521         return atomic_read(&info->nr_async_bios) >
522                 btrfs_async_submit_limit(info);
523 }
524
525 static void run_one_async_start(struct btrfs_work *work)
526 {
527         struct btrfs_fs_info *fs_info;
528         struct async_submit_bio *async;
529
530         async = container_of(work, struct  async_submit_bio, work);
531         fs_info = BTRFS_I(async->inode)->root->fs_info;
532         async->submit_bio_start(async->inode, async->rw, async->bio,
533                                async->mirror_num, async->bio_flags);
534 }
535
536 static void run_one_async_done(struct btrfs_work *work)
537 {
538         struct btrfs_fs_info *fs_info;
539         struct async_submit_bio *async;
540         int limit;
541
542         async = container_of(work, struct  async_submit_bio, work);
543         fs_info = BTRFS_I(async->inode)->root->fs_info;
544
545         limit = btrfs_async_submit_limit(fs_info);
546         limit = limit * 2 / 3;
547
548         atomic_dec(&fs_info->nr_async_submits);
549
550         if (atomic_read(&fs_info->nr_async_submits) < limit &&
551             waitqueue_active(&fs_info->async_submit_wait))
552                 wake_up(&fs_info->async_submit_wait);
553
554         async->submit_bio_done(async->inode, async->rw, async->bio,
555                                async->mirror_num, async->bio_flags);
556 }
557
558 static void run_one_async_free(struct btrfs_work *work)
559 {
560         struct async_submit_bio *async;
561
562         async = container_of(work, struct  async_submit_bio, work);
563         kfree(async);
564 }
565
566 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
567                         int rw, struct bio *bio, int mirror_num,
568                         unsigned long bio_flags,
569                         extent_submit_bio_hook_t *submit_bio_start,
570                         extent_submit_bio_hook_t *submit_bio_done)
571 {
572         struct async_submit_bio *async;
573
574         async = kmalloc(sizeof(*async), GFP_NOFS);
575         if (!async)
576                 return -ENOMEM;
577
578         async->inode = inode;
579         async->rw = rw;
580         async->bio = bio;
581         async->mirror_num = mirror_num;
582         async->submit_bio_start = submit_bio_start;
583         async->submit_bio_done = submit_bio_done;
584
585         async->work.func = run_one_async_start;
586         async->work.ordered_func = run_one_async_done;
587         async->work.ordered_free = run_one_async_free;
588
589         async->work.flags = 0;
590         async->bio_flags = bio_flags;
591
592         atomic_inc(&fs_info->nr_async_submits);
593
594         if (rw & (1 << BIO_RW_SYNCIO))
595                 btrfs_set_work_high_prio(&async->work);
596
597         btrfs_queue_worker(&fs_info->workers, &async->work);
598
599         while (atomic_read(&fs_info->async_submit_draining) &&
600               atomic_read(&fs_info->nr_async_submits)) {
601                 wait_event(fs_info->async_submit_wait,
602                            (atomic_read(&fs_info->nr_async_submits) == 0));
603         }
604
605         return 0;
606 }
607
608 static int btree_csum_one_bio(struct bio *bio)
609 {
610         struct bio_vec *bvec = bio->bi_io_vec;
611         int bio_index = 0;
612         struct btrfs_root *root;
613
614         WARN_ON(bio->bi_vcnt <= 0);
615         while (bio_index < bio->bi_vcnt) {
616                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
617                 csum_dirty_buffer(root, bvec->bv_page);
618                 bio_index++;
619                 bvec++;
620         }
621         return 0;
622 }
623
624 static int __btree_submit_bio_start(struct inode *inode, int rw,
625                                     struct bio *bio, int mirror_num,
626                                     unsigned long bio_flags)
627 {
628         /*
629          * when we're called for a write, we're already in the async
630          * submission context.  Just jump into btrfs_map_bio
631          */
632         btree_csum_one_bio(bio);
633         return 0;
634 }
635
636 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
637                                  int mirror_num, unsigned long bio_flags)
638 {
639         /*
640          * when we're called for a write, we're already in the async
641          * submission context.  Just jump into btrfs_map_bio
642          */
643         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
644 }
645
646 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
647                                  int mirror_num, unsigned long bio_flags)
648 {
649         int ret;
650
651         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
652                                           bio, 1);
653         BUG_ON(ret);
654
655         if (!(rw & (1 << BIO_RW))) {
656                 /*
657                  * called for a read, do the setup so that checksum validation
658                  * can happen in the async kernel threads
659                  */
660                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
661                                      mirror_num, 0);
662         }
663
664         /*
665          * kthread helpers are used to submit writes so that checksumming
666          * can happen in parallel across all CPUs
667          */
668         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
669                                    inode, rw, bio, mirror_num, 0,
670                                    __btree_submit_bio_start,
671                                    __btree_submit_bio_done);
672 }
673
674 static int btree_writepage(struct page *page, struct writeback_control *wbc)
675 {
676         struct extent_io_tree *tree;
677         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
678         struct extent_buffer *eb;
679         int was_dirty;
680
681         tree = &BTRFS_I(page->mapping->host)->io_tree;
682         if (!(current->flags & PF_MEMALLOC)) {
683                 return extent_write_full_page(tree, page,
684                                               btree_get_extent, wbc);
685         }
686
687         redirty_page_for_writepage(wbc, page);
688         eb = btrfs_find_tree_block(root, page_offset(page),
689                                       PAGE_CACHE_SIZE);
690         WARN_ON(!eb);
691
692         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
693         if (!was_dirty) {
694                 spin_lock(&root->fs_info->delalloc_lock);
695                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
696                 spin_unlock(&root->fs_info->delalloc_lock);
697         }
698         free_extent_buffer(eb);
699
700         unlock_page(page);
701         return 0;
702 }
703
704 static int btree_writepages(struct address_space *mapping,
705                             struct writeback_control *wbc)
706 {
707         struct extent_io_tree *tree;
708         tree = &BTRFS_I(mapping->host)->io_tree;
709         if (wbc->sync_mode == WB_SYNC_NONE) {
710                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
711                 u64 num_dirty;
712                 unsigned long thresh = 32 * 1024 * 1024;
713
714                 if (wbc->for_kupdate)
715                         return 0;
716
717                 /* this is a bit racy, but that's ok */
718                 num_dirty = root->fs_info->dirty_metadata_bytes;
719                 if (num_dirty < thresh)
720                         return 0;
721         }
722         return extent_writepages(tree, mapping, btree_get_extent, wbc);
723 }
724
725 static int btree_readpage(struct file *file, struct page *page)
726 {
727         struct extent_io_tree *tree;
728         tree = &BTRFS_I(page->mapping->host)->io_tree;
729         return extent_read_full_page(tree, page, btree_get_extent);
730 }
731
732 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
733 {
734         struct extent_io_tree *tree;
735         struct extent_map_tree *map;
736         int ret;
737
738         if (PageWriteback(page) || PageDirty(page))
739                 return 0;
740
741         tree = &BTRFS_I(page->mapping->host)->io_tree;
742         map = &BTRFS_I(page->mapping->host)->extent_tree;
743
744         ret = try_release_extent_state(map, tree, page, gfp_flags);
745         if (!ret)
746                 return 0;
747
748         ret = try_release_extent_buffer(tree, page);
749         if (ret == 1) {
750                 ClearPagePrivate(page);
751                 set_page_private(page, 0);
752                 page_cache_release(page);
753         }
754
755         return ret;
756 }
757
758 static void btree_invalidatepage(struct page *page, unsigned long offset)
759 {
760         struct extent_io_tree *tree;
761         tree = &BTRFS_I(page->mapping->host)->io_tree;
762         extent_invalidatepage(tree, page, offset);
763         btree_releasepage(page, GFP_NOFS);
764         if (PagePrivate(page)) {
765                 printk(KERN_WARNING "btrfs warning page private not zero "
766                        "on page %llu\n", (unsigned long long)page_offset(page));
767                 ClearPagePrivate(page);
768                 set_page_private(page, 0);
769                 page_cache_release(page);
770         }
771 }
772
773 static struct address_space_operations btree_aops = {
774         .readpage       = btree_readpage,
775         .writepage      = btree_writepage,
776         .writepages     = btree_writepages,
777         .releasepage    = btree_releasepage,
778         .invalidatepage = btree_invalidatepage,
779         .sync_page      = block_sync_page,
780 };
781
782 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
783                          u64 parent_transid)
784 {
785         struct extent_buffer *buf = NULL;
786         struct inode *btree_inode = root->fs_info->btree_inode;
787         int ret = 0;
788
789         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
790         if (!buf)
791                 return 0;
792         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
793                                  buf, 0, 0, btree_get_extent, 0);
794         free_extent_buffer(buf);
795         return ret;
796 }
797
798 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
799                                             u64 bytenr, u32 blocksize)
800 {
801         struct inode *btree_inode = root->fs_info->btree_inode;
802         struct extent_buffer *eb;
803         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
804                                 bytenr, blocksize, GFP_NOFS);
805         return eb;
806 }
807
808 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
809                                                  u64 bytenr, u32 blocksize)
810 {
811         struct inode *btree_inode = root->fs_info->btree_inode;
812         struct extent_buffer *eb;
813
814         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
815                                  bytenr, blocksize, NULL, GFP_NOFS);
816         return eb;
817 }
818
819
820 int btrfs_write_tree_block(struct extent_buffer *buf)
821 {
822         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
823                                       buf->start + buf->len - 1, WB_SYNC_ALL);
824 }
825
826 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
827 {
828         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
829                                   buf->start, buf->start + buf->len - 1);
830 }
831
832 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
833                                       u32 blocksize, u64 parent_transid)
834 {
835         struct extent_buffer *buf = NULL;
836         struct inode *btree_inode = root->fs_info->btree_inode;
837         struct extent_io_tree *io_tree;
838         int ret;
839
840         io_tree = &BTRFS_I(btree_inode)->io_tree;
841
842         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
843         if (!buf)
844                 return NULL;
845
846         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
847
848         if (ret == 0)
849                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
850         return buf;
851
852 }
853
854 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
855                      struct extent_buffer *buf)
856 {
857         struct inode *btree_inode = root->fs_info->btree_inode;
858         if (btrfs_header_generation(buf) ==
859             root->fs_info->running_transaction->transid) {
860                 btrfs_assert_tree_locked(buf);
861
862                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
863                         spin_lock(&root->fs_info->delalloc_lock);
864                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
865                                 root->fs_info->dirty_metadata_bytes -= buf->len;
866                         else
867                                 WARN_ON(1);
868                         spin_unlock(&root->fs_info->delalloc_lock);
869                 }
870
871                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
872                 btrfs_set_lock_blocking(buf);
873                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
874                                           buf);
875         }
876         return 0;
877 }
878
879 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
880                         u32 stripesize, struct btrfs_root *root,
881                         struct btrfs_fs_info *fs_info,
882                         u64 objectid)
883 {
884         root->node = NULL;
885         root->commit_root = NULL;
886         root->sectorsize = sectorsize;
887         root->nodesize = nodesize;
888         root->leafsize = leafsize;
889         root->stripesize = stripesize;
890         root->ref_cows = 0;
891         root->track_dirty = 0;
892
893         root->fs_info = fs_info;
894         root->objectid = objectid;
895         root->last_trans = 0;
896         root->highest_inode = 0;
897         root->last_inode_alloc = 0;
898         root->name = NULL;
899         root->in_sysfs = 0;
900         root->inode_tree.rb_node = NULL;
901
902         INIT_LIST_HEAD(&root->dirty_list);
903         INIT_LIST_HEAD(&root->orphan_list);
904         INIT_LIST_HEAD(&root->root_list);
905         spin_lock_init(&root->node_lock);
906         spin_lock_init(&root->list_lock);
907         spin_lock_init(&root->inode_lock);
908         mutex_init(&root->objectid_mutex);
909         mutex_init(&root->log_mutex);
910         init_waitqueue_head(&root->log_writer_wait);
911         init_waitqueue_head(&root->log_commit_wait[0]);
912         init_waitqueue_head(&root->log_commit_wait[1]);
913         atomic_set(&root->log_commit[0], 0);
914         atomic_set(&root->log_commit[1], 0);
915         atomic_set(&root->log_writers, 0);
916         root->log_batch = 0;
917         root->log_transid = 0;
918         extent_io_tree_init(&root->dirty_log_pages,
919                              fs_info->btree_inode->i_mapping, GFP_NOFS);
920
921         memset(&root->root_key, 0, sizeof(root->root_key));
922         memset(&root->root_item, 0, sizeof(root->root_item));
923         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
924         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
925         root->defrag_trans_start = fs_info->generation;
926         init_completion(&root->kobj_unregister);
927         root->defrag_running = 0;
928         root->defrag_level = 0;
929         root->root_key.objectid = objectid;
930         root->anon_super.s_root = NULL;
931         root->anon_super.s_dev = 0;
932         INIT_LIST_HEAD(&root->anon_super.s_list);
933         INIT_LIST_HEAD(&root->anon_super.s_instances);
934         init_rwsem(&root->anon_super.s_umount);
935
936         return 0;
937 }
938
939 static int find_and_setup_root(struct btrfs_root *tree_root,
940                                struct btrfs_fs_info *fs_info,
941                                u64 objectid,
942                                struct btrfs_root *root)
943 {
944         int ret;
945         u32 blocksize;
946         u64 generation;
947
948         __setup_root(tree_root->nodesize, tree_root->leafsize,
949                      tree_root->sectorsize, tree_root->stripesize,
950                      root, fs_info, objectid);
951         ret = btrfs_find_last_root(tree_root, objectid,
952                                    &root->root_item, &root->root_key);
953         BUG_ON(ret);
954
955         generation = btrfs_root_generation(&root->root_item);
956         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
957         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
958                                      blocksize, generation);
959         root->commit_root = btrfs_root_node(root);
960         BUG_ON(!root->node);
961         return 0;
962 }
963
964 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
965                              struct btrfs_fs_info *fs_info)
966 {
967         struct extent_buffer *eb;
968         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
969         u64 start = 0;
970         u64 end = 0;
971         int ret;
972
973         if (!log_root_tree)
974                 return 0;
975
976         while (1) {
977                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
978                                     0, &start, &end, EXTENT_DIRTY);
979                 if (ret)
980                         break;
981
982                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
983                                    start, end, GFP_NOFS);
984         }
985         eb = fs_info->log_root_tree->node;
986
987         WARN_ON(btrfs_header_level(eb) != 0);
988         WARN_ON(btrfs_header_nritems(eb) != 0);
989
990         ret = btrfs_free_reserved_extent(fs_info->tree_root,
991                                 eb->start, eb->len);
992         BUG_ON(ret);
993
994         free_extent_buffer(eb);
995         kfree(fs_info->log_root_tree);
996         fs_info->log_root_tree = NULL;
997         return 0;
998 }
999
1000 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1001                                          struct btrfs_fs_info *fs_info)
1002 {
1003         struct btrfs_root *root;
1004         struct btrfs_root *tree_root = fs_info->tree_root;
1005         struct extent_buffer *leaf;
1006
1007         root = kzalloc(sizeof(*root), GFP_NOFS);
1008         if (!root)
1009                 return ERR_PTR(-ENOMEM);
1010
1011         __setup_root(tree_root->nodesize, tree_root->leafsize,
1012                      tree_root->sectorsize, tree_root->stripesize,
1013                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1014
1015         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1016         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1017         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1018         /*
1019          * log trees do not get reference counted because they go away
1020          * before a real commit is actually done.  They do store pointers
1021          * to file data extents, and those reference counts still get
1022          * updated (along with back refs to the log tree).
1023          */
1024         root->ref_cows = 0;
1025
1026         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1027                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1028         if (IS_ERR(leaf)) {
1029                 kfree(root);
1030                 return ERR_CAST(leaf);
1031         }
1032
1033         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1034         btrfs_set_header_bytenr(leaf, leaf->start);
1035         btrfs_set_header_generation(leaf, trans->transid);
1036         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1037         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1038         root->node = leaf;
1039
1040         write_extent_buffer(root->node, root->fs_info->fsid,
1041                             (unsigned long)btrfs_header_fsid(root->node),
1042                             BTRFS_FSID_SIZE);
1043         btrfs_mark_buffer_dirty(root->node);
1044         btrfs_tree_unlock(root->node);
1045         return root;
1046 }
1047
1048 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1049                              struct btrfs_fs_info *fs_info)
1050 {
1051         struct btrfs_root *log_root;
1052
1053         log_root = alloc_log_tree(trans, fs_info);
1054         if (IS_ERR(log_root))
1055                 return PTR_ERR(log_root);
1056         WARN_ON(fs_info->log_root_tree);
1057         fs_info->log_root_tree = log_root;
1058         return 0;
1059 }
1060
1061 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1062                        struct btrfs_root *root)
1063 {
1064         struct btrfs_root *log_root;
1065         struct btrfs_inode_item *inode_item;
1066
1067         log_root = alloc_log_tree(trans, root->fs_info);
1068         if (IS_ERR(log_root))
1069                 return PTR_ERR(log_root);
1070
1071         log_root->last_trans = trans->transid;
1072         log_root->root_key.offset = root->root_key.objectid;
1073
1074         inode_item = &log_root->root_item.inode;
1075         inode_item->generation = cpu_to_le64(1);
1076         inode_item->size = cpu_to_le64(3);
1077         inode_item->nlink = cpu_to_le32(1);
1078         inode_item->nbytes = cpu_to_le64(root->leafsize);
1079         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1080
1081         btrfs_set_root_node(&log_root->root_item, log_root->node);
1082
1083         WARN_ON(root->log_root);
1084         root->log_root = log_root;
1085         root->log_transid = 0;
1086         return 0;
1087 }
1088
1089 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1090                                                struct btrfs_key *location)
1091 {
1092         struct btrfs_root *root;
1093         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1094         struct btrfs_path *path;
1095         struct extent_buffer *l;
1096         u64 highest_inode;
1097         u64 generation;
1098         u32 blocksize;
1099         int ret = 0;
1100
1101         root = kzalloc(sizeof(*root), GFP_NOFS);
1102         if (!root)
1103                 return ERR_PTR(-ENOMEM);
1104         if (location->offset == (u64)-1) {
1105                 ret = find_and_setup_root(tree_root, fs_info,
1106                                           location->objectid, root);
1107                 if (ret) {
1108                         kfree(root);
1109                         return ERR_PTR(ret);
1110                 }
1111                 goto insert;
1112         }
1113
1114         __setup_root(tree_root->nodesize, tree_root->leafsize,
1115                      tree_root->sectorsize, tree_root->stripesize,
1116                      root, fs_info, location->objectid);
1117
1118         path = btrfs_alloc_path();
1119         BUG_ON(!path);
1120         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1121         if (ret != 0) {
1122                 if (ret > 0)
1123                         ret = -ENOENT;
1124                 goto out;
1125         }
1126         l = path->nodes[0];
1127         read_extent_buffer(l, &root->root_item,
1128                btrfs_item_ptr_offset(l, path->slots[0]),
1129                sizeof(root->root_item));
1130         memcpy(&root->root_key, location, sizeof(*location));
1131         ret = 0;
1132 out:
1133         btrfs_release_path(root, path);
1134         btrfs_free_path(path);
1135         if (ret) {
1136                 kfree(root);
1137                 return ERR_PTR(ret);
1138         }
1139         generation = btrfs_root_generation(&root->root_item);
1140         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1141         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1142                                      blocksize, generation);
1143         root->commit_root = btrfs_root_node(root);
1144         BUG_ON(!root->node);
1145 insert:
1146         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1147                 root->ref_cows = 1;
1148                 ret = btrfs_find_highest_inode(root, &highest_inode);
1149                 if (ret == 0) {
1150                         root->highest_inode = highest_inode;
1151                         root->last_inode_alloc = highest_inode;
1152                 }
1153         }
1154         return root;
1155 }
1156
1157 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1158                                         u64 root_objectid)
1159 {
1160         struct btrfs_root *root;
1161
1162         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1163                 return fs_info->tree_root;
1164         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1165                 return fs_info->extent_root;
1166
1167         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1168                                  (unsigned long)root_objectid);
1169         return root;
1170 }
1171
1172 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1173                                               struct btrfs_key *location)
1174 {
1175         struct btrfs_root *root;
1176         int ret;
1177
1178         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1179                 return fs_info->tree_root;
1180         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1181                 return fs_info->extent_root;
1182         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1183                 return fs_info->chunk_root;
1184         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1185                 return fs_info->dev_root;
1186         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1187                 return fs_info->csum_root;
1188
1189         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1190                                  (unsigned long)location->objectid);
1191         if (root)
1192                 return root;
1193
1194         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1195         if (IS_ERR(root))
1196                 return root;
1197
1198         set_anon_super(&root->anon_super, NULL);
1199
1200         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1201                                 (unsigned long)root->root_key.objectid,
1202                                 root);
1203         if (ret) {
1204                 free_extent_buffer(root->node);
1205                 kfree(root);
1206                 return ERR_PTR(ret);
1207         }
1208         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1209                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1210                                             root->root_key.objectid);
1211                 BUG_ON(ret);
1212                 btrfs_orphan_cleanup(root);
1213         }
1214         return root;
1215 }
1216
1217 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1218                                       struct btrfs_key *location,
1219                                       const char *name, int namelen)
1220 {
1221         struct btrfs_root *root;
1222         int ret;
1223
1224         root = btrfs_read_fs_root_no_name(fs_info, location);
1225         if (!root)
1226                 return NULL;
1227
1228         if (root->in_sysfs)
1229                 return root;
1230
1231         ret = btrfs_set_root_name(root, name, namelen);
1232         if (ret) {
1233                 free_extent_buffer(root->node);
1234                 kfree(root);
1235                 return ERR_PTR(ret);
1236         }
1237 #if 0
1238         ret = btrfs_sysfs_add_root(root);
1239         if (ret) {
1240                 free_extent_buffer(root->node);
1241                 kfree(root->name);
1242                 kfree(root);
1243                 return ERR_PTR(ret);
1244         }
1245 #endif
1246         root->in_sysfs = 1;
1247         return root;
1248 }
1249
1250 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1251 {
1252         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1253         int ret = 0;
1254         struct btrfs_device *device;
1255         struct backing_dev_info *bdi;
1256
1257         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1258                 if (!device->bdev)
1259                         continue;
1260                 bdi = blk_get_backing_dev_info(device->bdev);
1261                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1262                         ret = 1;
1263                         break;
1264                 }
1265         }
1266         return ret;
1267 }
1268
1269 /*
1270  * this unplugs every device on the box, and it is only used when page
1271  * is null
1272  */
1273 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1274 {
1275         struct btrfs_device *device;
1276         struct btrfs_fs_info *info;
1277
1278         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1279         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1280                 if (!device->bdev)
1281                         continue;
1282
1283                 bdi = blk_get_backing_dev_info(device->bdev);
1284                 if (bdi->unplug_io_fn)
1285                         bdi->unplug_io_fn(bdi, page);
1286         }
1287 }
1288
1289 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1290 {
1291         struct inode *inode;
1292         struct extent_map_tree *em_tree;
1293         struct extent_map *em;
1294         struct address_space *mapping;
1295         u64 offset;
1296
1297         /* the generic O_DIRECT read code does this */
1298         if (1 || !page) {
1299                 __unplug_io_fn(bdi, page);
1300                 return;
1301         }
1302
1303         /*
1304          * page->mapping may change at any time.  Get a consistent copy
1305          * and use that for everything below
1306          */
1307         smp_mb();
1308         mapping = page->mapping;
1309         if (!mapping)
1310                 return;
1311
1312         inode = mapping->host;
1313
1314         /*
1315          * don't do the expensive searching for a small number of
1316          * devices
1317          */
1318         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1319                 __unplug_io_fn(bdi, page);
1320                 return;
1321         }
1322
1323         offset = page_offset(page);
1324
1325         em_tree = &BTRFS_I(inode)->extent_tree;
1326         spin_lock(&em_tree->lock);
1327         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1328         spin_unlock(&em_tree->lock);
1329         if (!em) {
1330                 __unplug_io_fn(bdi, page);
1331                 return;
1332         }
1333
1334         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1335                 free_extent_map(em);
1336                 __unplug_io_fn(bdi, page);
1337                 return;
1338         }
1339         offset = offset - em->start;
1340         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1341                           em->block_start + offset, page);
1342         free_extent_map(em);
1343 }
1344
1345 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1346 {
1347         bdi_init(bdi);
1348         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1349         bdi->state              = 0;
1350         bdi->capabilities       = default_backing_dev_info.capabilities;
1351         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1352         bdi->unplug_io_data     = info;
1353         bdi->congested_fn       = btrfs_congested_fn;
1354         bdi->congested_data     = info;
1355         return 0;
1356 }
1357
1358 static int bio_ready_for_csum(struct bio *bio)
1359 {
1360         u64 length = 0;
1361         u64 buf_len = 0;
1362         u64 start = 0;
1363         struct page *page;
1364         struct extent_io_tree *io_tree = NULL;
1365         struct btrfs_fs_info *info = NULL;
1366         struct bio_vec *bvec;
1367         int i;
1368         int ret;
1369
1370         bio_for_each_segment(bvec, bio, i) {
1371                 page = bvec->bv_page;
1372                 if (page->private == EXTENT_PAGE_PRIVATE) {
1373                         length += bvec->bv_len;
1374                         continue;
1375                 }
1376                 if (!page->private) {
1377                         length += bvec->bv_len;
1378                         continue;
1379                 }
1380                 length = bvec->bv_len;
1381                 buf_len = page->private >> 2;
1382                 start = page_offset(page) + bvec->bv_offset;
1383                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1384                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1385         }
1386         /* are we fully contained in this bio? */
1387         if (buf_len <= length)
1388                 return 1;
1389
1390         ret = extent_range_uptodate(io_tree, start + length,
1391                                     start + buf_len - 1);
1392         return ret;
1393 }
1394
1395 /*
1396  * called by the kthread helper functions to finally call the bio end_io
1397  * functions.  This is where read checksum verification actually happens
1398  */
1399 static void end_workqueue_fn(struct btrfs_work *work)
1400 {
1401         struct bio *bio;
1402         struct end_io_wq *end_io_wq;
1403         struct btrfs_fs_info *fs_info;
1404         int error;
1405
1406         end_io_wq = container_of(work, struct end_io_wq, work);
1407         bio = end_io_wq->bio;
1408         fs_info = end_io_wq->info;
1409
1410         /* metadata bio reads are special because the whole tree block must
1411          * be checksummed at once.  This makes sure the entire block is in
1412          * ram and up to date before trying to verify things.  For
1413          * blocksize <= pagesize, it is basically a noop
1414          */
1415         if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1416             !bio_ready_for_csum(bio)) {
1417                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1418                                    &end_io_wq->work);
1419                 return;
1420         }
1421         error = end_io_wq->error;
1422         bio->bi_private = end_io_wq->private;
1423         bio->bi_end_io = end_io_wq->end_io;
1424         kfree(end_io_wq);
1425         bio_endio(bio, error);
1426 }
1427
1428 static int cleaner_kthread(void *arg)
1429 {
1430         struct btrfs_root *root = arg;
1431
1432         do {
1433                 smp_mb();
1434                 if (root->fs_info->closing)
1435                         break;
1436
1437                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1438                 mutex_lock(&root->fs_info->cleaner_mutex);
1439                 btrfs_clean_old_snapshots(root);
1440                 mutex_unlock(&root->fs_info->cleaner_mutex);
1441
1442                 if (freezing(current)) {
1443                         refrigerator();
1444                 } else {
1445                         smp_mb();
1446                         if (root->fs_info->closing)
1447                                 break;
1448                         set_current_state(TASK_INTERRUPTIBLE);
1449                         schedule();
1450                         __set_current_state(TASK_RUNNING);
1451                 }
1452         } while (!kthread_should_stop());
1453         return 0;
1454 }
1455
1456 static int transaction_kthread(void *arg)
1457 {
1458         struct btrfs_root *root = arg;
1459         struct btrfs_trans_handle *trans;
1460         struct btrfs_transaction *cur;
1461         unsigned long now;
1462         unsigned long delay;
1463         int ret;
1464
1465         do {
1466                 smp_mb();
1467                 if (root->fs_info->closing)
1468                         break;
1469
1470                 delay = HZ * 30;
1471                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1472                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1473
1474                 mutex_lock(&root->fs_info->trans_mutex);
1475                 cur = root->fs_info->running_transaction;
1476                 if (!cur) {
1477                         mutex_unlock(&root->fs_info->trans_mutex);
1478                         goto sleep;
1479                 }
1480
1481                 now = get_seconds();
1482                 if (now < cur->start_time || now - cur->start_time < 30) {
1483                         mutex_unlock(&root->fs_info->trans_mutex);
1484                         delay = HZ * 5;
1485                         goto sleep;
1486                 }
1487                 mutex_unlock(&root->fs_info->trans_mutex);
1488                 trans = btrfs_start_transaction(root, 1);
1489                 ret = btrfs_commit_transaction(trans, root);
1490
1491 sleep:
1492                 wake_up_process(root->fs_info->cleaner_kthread);
1493                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1494
1495                 if (freezing(current)) {
1496                         refrigerator();
1497                 } else {
1498                         if (root->fs_info->closing)
1499                                 break;
1500                         set_current_state(TASK_INTERRUPTIBLE);
1501                         schedule_timeout(delay);
1502                         __set_current_state(TASK_RUNNING);
1503                 }
1504         } while (!kthread_should_stop());
1505         return 0;
1506 }
1507
1508 struct btrfs_root *open_ctree(struct super_block *sb,
1509                               struct btrfs_fs_devices *fs_devices,
1510                               char *options)
1511 {
1512         u32 sectorsize;
1513         u32 nodesize;
1514         u32 leafsize;
1515         u32 blocksize;
1516         u32 stripesize;
1517         u64 generation;
1518         u64 features;
1519         struct btrfs_key location;
1520         struct buffer_head *bh;
1521         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1522                                                  GFP_NOFS);
1523         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1524                                                  GFP_NOFS);
1525         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1526                                                GFP_NOFS);
1527         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1528                                                 GFP_NOFS);
1529         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1530                                                 GFP_NOFS);
1531         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1532                                               GFP_NOFS);
1533         struct btrfs_root *log_tree_root;
1534
1535         int ret;
1536         int err = -EINVAL;
1537
1538         struct btrfs_super_block *disk_super;
1539
1540         if (!extent_root || !tree_root || !fs_info ||
1541             !chunk_root || !dev_root || !csum_root) {
1542                 err = -ENOMEM;
1543                 goto fail;
1544         }
1545         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1546         INIT_LIST_HEAD(&fs_info->trans_list);
1547         INIT_LIST_HEAD(&fs_info->dead_roots);
1548         INIT_LIST_HEAD(&fs_info->hashers);
1549         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1550         INIT_LIST_HEAD(&fs_info->ordered_operations);
1551         spin_lock_init(&fs_info->delalloc_lock);
1552         spin_lock_init(&fs_info->new_trans_lock);
1553         spin_lock_init(&fs_info->ref_cache_lock);
1554
1555         init_completion(&fs_info->kobj_unregister);
1556         fs_info->tree_root = tree_root;
1557         fs_info->extent_root = extent_root;
1558         fs_info->csum_root = csum_root;
1559         fs_info->chunk_root = chunk_root;
1560         fs_info->dev_root = dev_root;
1561         fs_info->fs_devices = fs_devices;
1562         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1563         INIT_LIST_HEAD(&fs_info->space_info);
1564         btrfs_mapping_init(&fs_info->mapping_tree);
1565         atomic_set(&fs_info->nr_async_submits, 0);
1566         atomic_set(&fs_info->async_delalloc_pages, 0);
1567         atomic_set(&fs_info->async_submit_draining, 0);
1568         atomic_set(&fs_info->nr_async_bios, 0);
1569         fs_info->sb = sb;
1570         fs_info->max_extent = (u64)-1;
1571         fs_info->max_inline = 8192 * 1024;
1572         setup_bdi(fs_info, &fs_info->bdi);
1573         fs_info->btree_inode = new_inode(sb);
1574         fs_info->btree_inode->i_ino = 1;
1575         fs_info->btree_inode->i_nlink = 1;
1576         fs_info->metadata_ratio = 8;
1577
1578         fs_info->thread_pool_size = min_t(unsigned long,
1579                                           num_online_cpus() + 2, 8);
1580
1581         INIT_LIST_HEAD(&fs_info->ordered_extents);
1582         spin_lock_init(&fs_info->ordered_extent_lock);
1583
1584         sb->s_blocksize = 4096;
1585         sb->s_blocksize_bits = blksize_bits(4096);
1586
1587         /*
1588          * we set the i_size on the btree inode to the max possible int.
1589          * the real end of the address space is determined by all of
1590          * the devices in the system
1591          */
1592         fs_info->btree_inode->i_size = OFFSET_MAX;
1593         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1594         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1595
1596         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1597         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1598                              fs_info->btree_inode->i_mapping,
1599                              GFP_NOFS);
1600         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1601                              GFP_NOFS);
1602
1603         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1604
1605         spin_lock_init(&fs_info->block_group_cache_lock);
1606         fs_info->block_group_cache_tree.rb_node = NULL;
1607
1608         extent_io_tree_init(&fs_info->pinned_extents,
1609                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1610         fs_info->do_barriers = 1;
1611
1612         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1613         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1614                sizeof(struct btrfs_key));
1615         insert_inode_hash(fs_info->btree_inode);
1616
1617         mutex_init(&fs_info->trans_mutex);
1618         mutex_init(&fs_info->ordered_operations_mutex);
1619         mutex_init(&fs_info->tree_log_mutex);
1620         mutex_init(&fs_info->drop_mutex);
1621         mutex_init(&fs_info->chunk_mutex);
1622         mutex_init(&fs_info->transaction_kthread_mutex);
1623         mutex_init(&fs_info->cleaner_mutex);
1624         mutex_init(&fs_info->volume_mutex);
1625         mutex_init(&fs_info->tree_reloc_mutex);
1626
1627         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1628         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1629
1630         init_waitqueue_head(&fs_info->transaction_throttle);
1631         init_waitqueue_head(&fs_info->transaction_wait);
1632         init_waitqueue_head(&fs_info->async_submit_wait);
1633
1634         __setup_root(4096, 4096, 4096, 4096, tree_root,
1635                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1636
1637
1638         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1639         if (!bh)
1640                 goto fail_iput;
1641
1642         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1643         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1644                sizeof(fs_info->super_for_commit));
1645         brelse(bh);
1646
1647         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1648
1649         disk_super = &fs_info->super_copy;
1650         if (!btrfs_super_root(disk_super))
1651                 goto fail_iput;
1652
1653         ret = btrfs_parse_options(tree_root, options);
1654         if (ret) {
1655                 err = ret;
1656                 goto fail_iput;
1657         }
1658
1659         features = btrfs_super_incompat_flags(disk_super) &
1660                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1661         if (features) {
1662                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1663                        "unsupported optional features (%Lx).\n",
1664                        (unsigned long long)features);
1665                 err = -EINVAL;
1666                 goto fail_iput;
1667         }
1668
1669         features = btrfs_super_incompat_flags(disk_super);
1670         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1671                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1672                 btrfs_set_super_incompat_flags(disk_super, features);
1673         }
1674
1675         features = btrfs_super_compat_ro_flags(disk_super) &
1676                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1677         if (!(sb->s_flags & MS_RDONLY) && features) {
1678                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1679                        "unsupported option features (%Lx).\n",
1680                        (unsigned long long)features);
1681                 err = -EINVAL;
1682                 goto fail_iput;
1683         }
1684
1685         /*
1686          * we need to start all the end_io workers up front because the
1687          * queue work function gets called at interrupt time, and so it
1688          * cannot dynamically grow.
1689          */
1690         btrfs_init_workers(&fs_info->workers, "worker",
1691                            fs_info->thread_pool_size);
1692
1693         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1694                            fs_info->thread_pool_size);
1695
1696         btrfs_init_workers(&fs_info->submit_workers, "submit",
1697                            min_t(u64, fs_devices->num_devices,
1698                            fs_info->thread_pool_size));
1699
1700         /* a higher idle thresh on the submit workers makes it much more
1701          * likely that bios will be send down in a sane order to the
1702          * devices
1703          */
1704         fs_info->submit_workers.idle_thresh = 64;
1705
1706         fs_info->workers.idle_thresh = 16;
1707         fs_info->workers.ordered = 1;
1708
1709         fs_info->delalloc_workers.idle_thresh = 2;
1710         fs_info->delalloc_workers.ordered = 1;
1711
1712         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1713         btrfs_init_workers(&fs_info->endio_workers, "endio",
1714                            fs_info->thread_pool_size);
1715         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1716                            fs_info->thread_pool_size);
1717         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1718                            "endio-meta-write", fs_info->thread_pool_size);
1719         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1720                            fs_info->thread_pool_size);
1721
1722         /*
1723          * endios are largely parallel and should have a very
1724          * low idle thresh
1725          */
1726         fs_info->endio_workers.idle_thresh = 4;
1727         fs_info->endio_meta_workers.idle_thresh = 4;
1728
1729         fs_info->endio_write_workers.idle_thresh = 64;
1730         fs_info->endio_meta_write_workers.idle_thresh = 64;
1731
1732         btrfs_start_workers(&fs_info->workers, 1);
1733         btrfs_start_workers(&fs_info->submit_workers, 1);
1734         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1735         btrfs_start_workers(&fs_info->fixup_workers, 1);
1736         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1737         btrfs_start_workers(&fs_info->endio_meta_workers,
1738                             fs_info->thread_pool_size);
1739         btrfs_start_workers(&fs_info->endio_meta_write_workers,
1740                             fs_info->thread_pool_size);
1741         btrfs_start_workers(&fs_info->endio_write_workers,
1742                             fs_info->thread_pool_size);
1743
1744         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1745         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1746                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1747
1748         nodesize = btrfs_super_nodesize(disk_super);
1749         leafsize = btrfs_super_leafsize(disk_super);
1750         sectorsize = btrfs_super_sectorsize(disk_super);
1751         stripesize = btrfs_super_stripesize(disk_super);
1752         tree_root->nodesize = nodesize;
1753         tree_root->leafsize = leafsize;
1754         tree_root->sectorsize = sectorsize;
1755         tree_root->stripesize = stripesize;
1756
1757         sb->s_blocksize = sectorsize;
1758         sb->s_blocksize_bits = blksize_bits(sectorsize);
1759
1760         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1761                     sizeof(disk_super->magic))) {
1762                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1763                 goto fail_sb_buffer;
1764         }
1765
1766         mutex_lock(&fs_info->chunk_mutex);
1767         ret = btrfs_read_sys_array(tree_root);
1768         mutex_unlock(&fs_info->chunk_mutex);
1769         if (ret) {
1770                 printk(KERN_WARNING "btrfs: failed to read the system "
1771                        "array on %s\n", sb->s_id);
1772                 goto fail_sb_buffer;
1773         }
1774
1775         blocksize = btrfs_level_size(tree_root,
1776                                      btrfs_super_chunk_root_level(disk_super));
1777         generation = btrfs_super_chunk_root_generation(disk_super);
1778
1779         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1780                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1781
1782         chunk_root->node = read_tree_block(chunk_root,
1783                                            btrfs_super_chunk_root(disk_super),
1784                                            blocksize, generation);
1785         BUG_ON(!chunk_root->node);
1786         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1787         chunk_root->commit_root = btrfs_root_node(chunk_root);
1788
1789         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1790            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1791            BTRFS_UUID_SIZE);
1792
1793         mutex_lock(&fs_info->chunk_mutex);
1794         ret = btrfs_read_chunk_tree(chunk_root);
1795         mutex_unlock(&fs_info->chunk_mutex);
1796         if (ret) {
1797                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1798                        sb->s_id);
1799                 goto fail_chunk_root;
1800         }
1801
1802         btrfs_close_extra_devices(fs_devices);
1803
1804         blocksize = btrfs_level_size(tree_root,
1805                                      btrfs_super_root_level(disk_super));
1806         generation = btrfs_super_generation(disk_super);
1807
1808         tree_root->node = read_tree_block(tree_root,
1809                                           btrfs_super_root(disk_super),
1810                                           blocksize, generation);
1811         if (!tree_root->node)
1812                 goto fail_chunk_root;
1813         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1814         tree_root->commit_root = btrfs_root_node(tree_root);
1815
1816         ret = find_and_setup_root(tree_root, fs_info,
1817                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1818         if (ret)
1819                 goto fail_tree_root;
1820         extent_root->track_dirty = 1;
1821
1822         ret = find_and_setup_root(tree_root, fs_info,
1823                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1824         if (ret)
1825                 goto fail_extent_root;
1826         dev_root->track_dirty = 1;
1827
1828         ret = find_and_setup_root(tree_root, fs_info,
1829                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1830         if (ret)
1831                 goto fail_dev_root;
1832
1833         csum_root->track_dirty = 1;
1834
1835         btrfs_read_block_groups(extent_root);
1836
1837         fs_info->generation = generation;
1838         fs_info->last_trans_committed = generation;
1839         fs_info->data_alloc_profile = (u64)-1;
1840         fs_info->metadata_alloc_profile = (u64)-1;
1841         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1842         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1843                                                "btrfs-cleaner");
1844         if (IS_ERR(fs_info->cleaner_kthread))
1845                 goto fail_csum_root;
1846
1847         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1848                                                    tree_root,
1849                                                    "btrfs-transaction");
1850         if (IS_ERR(fs_info->transaction_kthread))
1851                 goto fail_cleaner;
1852
1853         if (!btrfs_test_opt(tree_root, SSD) &&
1854             !btrfs_test_opt(tree_root, NOSSD) &&
1855             !fs_info->fs_devices->rotating) {
1856                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1857                        "mode\n");
1858                 btrfs_set_opt(fs_info->mount_opt, SSD);
1859         }
1860
1861         if (btrfs_super_log_root(disk_super) != 0) {
1862                 u64 bytenr = btrfs_super_log_root(disk_super);
1863
1864                 if (fs_devices->rw_devices == 0) {
1865                         printk(KERN_WARNING "Btrfs log replay required "
1866                                "on RO media\n");
1867                         err = -EIO;
1868                         goto fail_trans_kthread;
1869                 }
1870                 blocksize =
1871                      btrfs_level_size(tree_root,
1872                                       btrfs_super_log_root_level(disk_super));
1873
1874                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1875                                                       GFP_NOFS);
1876
1877                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1878                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1879
1880                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1881                                                       blocksize,
1882                                                       generation + 1);
1883                 ret = btrfs_recover_log_trees(log_tree_root);
1884                 BUG_ON(ret);
1885
1886                 if (sb->s_flags & MS_RDONLY) {
1887                         ret =  btrfs_commit_super(tree_root);
1888                         BUG_ON(ret);
1889                 }
1890         }
1891
1892         if (!(sb->s_flags & MS_RDONLY)) {
1893                 ret = btrfs_recover_relocation(tree_root);
1894                 BUG_ON(ret);
1895         }
1896
1897         location.objectid = BTRFS_FS_TREE_OBJECTID;
1898         location.type = BTRFS_ROOT_ITEM_KEY;
1899         location.offset = (u64)-1;
1900
1901         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1902         if (!fs_info->fs_root)
1903                 goto fail_trans_kthread;
1904
1905         return tree_root;
1906
1907 fail_trans_kthread:
1908         kthread_stop(fs_info->transaction_kthread);
1909 fail_cleaner:
1910         kthread_stop(fs_info->cleaner_kthread);
1911
1912         /*
1913          * make sure we're done with the btree inode before we stop our
1914          * kthreads
1915          */
1916         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1917         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1918
1919 fail_csum_root:
1920         free_extent_buffer(csum_root->node);
1921         free_extent_buffer(csum_root->commit_root);
1922 fail_dev_root:
1923         free_extent_buffer(dev_root->node);
1924         free_extent_buffer(dev_root->commit_root);
1925 fail_extent_root:
1926         free_extent_buffer(extent_root->node);
1927         free_extent_buffer(extent_root->commit_root);
1928 fail_tree_root:
1929         free_extent_buffer(tree_root->node);
1930         free_extent_buffer(tree_root->commit_root);
1931 fail_chunk_root:
1932         free_extent_buffer(chunk_root->node);
1933         free_extent_buffer(chunk_root->commit_root);
1934 fail_sb_buffer:
1935         btrfs_stop_workers(&fs_info->fixup_workers);
1936         btrfs_stop_workers(&fs_info->delalloc_workers);
1937         btrfs_stop_workers(&fs_info->workers);
1938         btrfs_stop_workers(&fs_info->endio_workers);
1939         btrfs_stop_workers(&fs_info->endio_meta_workers);
1940         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1941         btrfs_stop_workers(&fs_info->endio_write_workers);
1942         btrfs_stop_workers(&fs_info->submit_workers);
1943 fail_iput:
1944         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1945         iput(fs_info->btree_inode);
1946
1947         btrfs_close_devices(fs_info->fs_devices);
1948         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1949         bdi_destroy(&fs_info->bdi);
1950
1951 fail:
1952         kfree(extent_root);
1953         kfree(tree_root);
1954         kfree(fs_info);
1955         kfree(chunk_root);
1956         kfree(dev_root);
1957         kfree(csum_root);
1958         return ERR_PTR(err);
1959 }
1960
1961 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1962 {
1963         char b[BDEVNAME_SIZE];
1964
1965         if (uptodate) {
1966                 set_buffer_uptodate(bh);
1967         } else {
1968                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1969                         printk(KERN_WARNING "lost page write due to "
1970                                         "I/O error on %s\n",
1971                                        bdevname(bh->b_bdev, b));
1972                 }
1973                 /* note, we dont' set_buffer_write_io_error because we have
1974                  * our own ways of dealing with the IO errors
1975                  */
1976                 clear_buffer_uptodate(bh);
1977         }
1978         unlock_buffer(bh);
1979         put_bh(bh);
1980 }
1981
1982 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1983 {
1984         struct buffer_head *bh;
1985         struct buffer_head *latest = NULL;
1986         struct btrfs_super_block *super;
1987         int i;
1988         u64 transid = 0;
1989         u64 bytenr;
1990
1991         /* we would like to check all the supers, but that would make
1992          * a btrfs mount succeed after a mkfs from a different FS.
1993          * So, we need to add a special mount option to scan for
1994          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1995          */
1996         for (i = 0; i < 1; i++) {
1997                 bytenr = btrfs_sb_offset(i);
1998                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
1999                         break;
2000                 bh = __bread(bdev, bytenr / 4096, 4096);
2001                 if (!bh)
2002                         continue;
2003
2004                 super = (struct btrfs_super_block *)bh->b_data;
2005                 if (btrfs_super_bytenr(super) != bytenr ||
2006                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2007                             sizeof(super->magic))) {
2008                         brelse(bh);
2009                         continue;
2010                 }
2011
2012                 if (!latest || btrfs_super_generation(super) > transid) {
2013                         brelse(latest);
2014                         latest = bh;
2015                         transid = btrfs_super_generation(super);
2016                 } else {
2017                         brelse(bh);
2018                 }
2019         }
2020         return latest;
2021 }
2022
2023 /*
2024  * this should be called twice, once with wait == 0 and
2025  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2026  * we write are pinned.
2027  *
2028  * They are released when wait == 1 is done.
2029  * max_mirrors must be the same for both runs, and it indicates how
2030  * many supers on this one device should be written.
2031  *
2032  * max_mirrors == 0 means to write them all.
2033  */
2034 static int write_dev_supers(struct btrfs_device *device,
2035                             struct btrfs_super_block *sb,
2036                             int do_barriers, int wait, int max_mirrors)
2037 {
2038         struct buffer_head *bh;
2039         int i;
2040         int ret;
2041         int errors = 0;
2042         u32 crc;
2043         u64 bytenr;
2044         int last_barrier = 0;
2045
2046         if (max_mirrors == 0)
2047                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2048
2049         /* make sure only the last submit_bh does a barrier */
2050         if (do_barriers) {
2051                 for (i = 0; i < max_mirrors; i++) {
2052                         bytenr = btrfs_sb_offset(i);
2053                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2054                             device->total_bytes)
2055                                 break;
2056                         last_barrier = i;
2057                 }
2058         }
2059
2060         for (i = 0; i < max_mirrors; i++) {
2061                 bytenr = btrfs_sb_offset(i);
2062                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2063                         break;
2064
2065                 if (wait) {
2066                         bh = __find_get_block(device->bdev, bytenr / 4096,
2067                                               BTRFS_SUPER_INFO_SIZE);
2068                         BUG_ON(!bh);
2069                         wait_on_buffer(bh);
2070                         if (!buffer_uptodate(bh))
2071                                 errors++;
2072
2073                         /* drop our reference */
2074                         brelse(bh);
2075
2076                         /* drop the reference from the wait == 0 run */
2077                         brelse(bh);
2078                         continue;
2079                 } else {
2080                         btrfs_set_super_bytenr(sb, bytenr);
2081
2082                         crc = ~(u32)0;
2083                         crc = btrfs_csum_data(NULL, (char *)sb +
2084                                               BTRFS_CSUM_SIZE, crc,
2085                                               BTRFS_SUPER_INFO_SIZE -
2086                                               BTRFS_CSUM_SIZE);
2087                         btrfs_csum_final(crc, sb->csum);
2088
2089                         /*
2090                          * one reference for us, and we leave it for the
2091                          * caller
2092                          */
2093                         bh = __getblk(device->bdev, bytenr / 4096,
2094                                       BTRFS_SUPER_INFO_SIZE);
2095                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2096
2097                         /* one reference for submit_bh */
2098                         get_bh(bh);
2099
2100                         set_buffer_uptodate(bh);
2101                         lock_buffer(bh);
2102                         bh->b_end_io = btrfs_end_buffer_write_sync;
2103                 }
2104
2105                 if (i == last_barrier && do_barriers && device->barriers) {
2106                         ret = submit_bh(WRITE_BARRIER, bh);
2107                         if (ret == -EOPNOTSUPP) {
2108                                 printk("btrfs: disabling barriers on dev %s\n",
2109                                        device->name);
2110                                 set_buffer_uptodate(bh);
2111                                 device->barriers = 0;
2112                                 /* one reference for submit_bh */
2113                                 get_bh(bh);
2114                                 lock_buffer(bh);
2115                                 ret = submit_bh(WRITE_SYNC, bh);
2116                         }
2117                 } else {
2118                         ret = submit_bh(WRITE_SYNC, bh);
2119                 }
2120
2121                 if (ret)
2122                         errors++;
2123         }
2124         return errors < i ? 0 : -1;
2125 }
2126
2127 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2128 {
2129         struct list_head *head;
2130         struct btrfs_device *dev;
2131         struct btrfs_super_block *sb;
2132         struct btrfs_dev_item *dev_item;
2133         int ret;
2134         int do_barriers;
2135         int max_errors;
2136         int total_errors = 0;
2137         u64 flags;
2138
2139         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2140         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2141
2142         sb = &root->fs_info->super_for_commit;
2143         dev_item = &sb->dev_item;
2144
2145         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2146         head = &root->fs_info->fs_devices->devices;
2147         list_for_each_entry(dev, head, dev_list) {
2148                 if (!dev->bdev) {
2149                         total_errors++;
2150                         continue;
2151                 }
2152                 if (!dev->in_fs_metadata || !dev->writeable)
2153                         continue;
2154
2155                 btrfs_set_stack_device_generation(dev_item, 0);
2156                 btrfs_set_stack_device_type(dev_item, dev->type);
2157                 btrfs_set_stack_device_id(dev_item, dev->devid);
2158                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2159                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2160                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2161                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2162                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2163                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2164                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2165
2166                 flags = btrfs_super_flags(sb);
2167                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2168
2169                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2170                 if (ret)
2171                         total_errors++;
2172         }
2173         if (total_errors > max_errors) {
2174                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2175                        total_errors);
2176                 BUG();
2177         }
2178
2179         total_errors = 0;
2180         list_for_each_entry(dev, head, dev_list) {
2181                 if (!dev->bdev)
2182                         continue;
2183                 if (!dev->in_fs_metadata || !dev->writeable)
2184                         continue;
2185
2186                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2187                 if (ret)
2188                         total_errors++;
2189         }
2190         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2191         if (total_errors > max_errors) {
2192                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2193                        total_errors);
2194                 BUG();
2195         }
2196         return 0;
2197 }
2198
2199 int write_ctree_super(struct btrfs_trans_handle *trans,
2200                       struct btrfs_root *root, int max_mirrors)
2201 {
2202         int ret;
2203
2204         ret = write_all_supers(root, max_mirrors);
2205         return ret;
2206 }
2207
2208 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2209 {
2210         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2211         radix_tree_delete(&fs_info->fs_roots_radix,
2212                           (unsigned long)root->root_key.objectid);
2213         if (root->anon_super.s_dev) {
2214                 down_write(&root->anon_super.s_umount);
2215                 kill_anon_super(&root->anon_super);
2216         }
2217         if (root->node)
2218                 free_extent_buffer(root->node);
2219         if (root->commit_root)
2220                 free_extent_buffer(root->commit_root);
2221         kfree(root->name);
2222         kfree(root);
2223         return 0;
2224 }
2225
2226 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2227 {
2228         int ret;
2229         struct btrfs_root *gang[8];
2230         int i;
2231
2232         while (1) {
2233                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2234                                              (void **)gang, 0,
2235                                              ARRAY_SIZE(gang));
2236                 if (!ret)
2237                         break;
2238                 for (i = 0; i < ret; i++)
2239                         btrfs_free_fs_root(fs_info, gang[i]);
2240         }
2241         return 0;
2242 }
2243
2244 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2245 {
2246         u64 root_objectid = 0;
2247         struct btrfs_root *gang[8];
2248         int i;
2249         int ret;
2250
2251         while (1) {
2252                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2253                                              (void **)gang, root_objectid,
2254                                              ARRAY_SIZE(gang));
2255                 if (!ret)
2256                         break;
2257
2258                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2259                 for (i = 0; i < ret; i++) {
2260                         root_objectid = gang[i]->root_key.objectid;
2261                         ret = btrfs_find_dead_roots(fs_info->tree_root,
2262                                                     root_objectid);
2263                         BUG_ON(ret);
2264                         btrfs_orphan_cleanup(gang[i]);
2265                 }
2266                 root_objectid++;
2267         }
2268         return 0;
2269 }
2270
2271 int btrfs_commit_super(struct btrfs_root *root)
2272 {
2273         struct btrfs_trans_handle *trans;
2274         int ret;
2275
2276         mutex_lock(&root->fs_info->cleaner_mutex);
2277         btrfs_clean_old_snapshots(root);
2278         mutex_unlock(&root->fs_info->cleaner_mutex);
2279         trans = btrfs_start_transaction(root, 1);
2280         ret = btrfs_commit_transaction(trans, root);
2281         BUG_ON(ret);
2282         /* run commit again to drop the original snapshot */
2283         trans = btrfs_start_transaction(root, 1);
2284         btrfs_commit_transaction(trans, root);
2285         ret = btrfs_write_and_wait_transaction(NULL, root);
2286         BUG_ON(ret);
2287
2288         ret = write_ctree_super(NULL, root, 0);
2289         return ret;
2290 }
2291
2292 int close_ctree(struct btrfs_root *root)
2293 {
2294         struct btrfs_fs_info *fs_info = root->fs_info;
2295         int ret;
2296
2297         fs_info->closing = 1;
2298         smp_mb();
2299
2300         kthread_stop(root->fs_info->transaction_kthread);
2301         kthread_stop(root->fs_info->cleaner_kthread);
2302
2303         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2304                 ret =  btrfs_commit_super(root);
2305                 if (ret)
2306                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2307         }
2308
2309         if (fs_info->delalloc_bytes) {
2310                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2311                        (unsigned long long)fs_info->delalloc_bytes);
2312         }
2313         if (fs_info->total_ref_cache_size) {
2314                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2315                        (unsigned long long)fs_info->total_ref_cache_size);
2316         }
2317
2318         free_extent_buffer(fs_info->extent_root->node);
2319         free_extent_buffer(fs_info->extent_root->commit_root);
2320         free_extent_buffer(fs_info->tree_root->node);
2321         free_extent_buffer(fs_info->tree_root->commit_root);
2322         free_extent_buffer(root->fs_info->chunk_root->node);
2323         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2324         free_extent_buffer(root->fs_info->dev_root->node);
2325         free_extent_buffer(root->fs_info->dev_root->commit_root);
2326         free_extent_buffer(root->fs_info->csum_root->node);
2327         free_extent_buffer(root->fs_info->csum_root->commit_root);
2328
2329         btrfs_free_block_groups(root->fs_info);
2330
2331         del_fs_roots(fs_info);
2332
2333         iput(fs_info->btree_inode);
2334
2335         btrfs_stop_workers(&fs_info->fixup_workers);
2336         btrfs_stop_workers(&fs_info->delalloc_workers);
2337         btrfs_stop_workers(&fs_info->workers);
2338         btrfs_stop_workers(&fs_info->endio_workers);
2339         btrfs_stop_workers(&fs_info->endio_meta_workers);
2340         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2341         btrfs_stop_workers(&fs_info->endio_write_workers);
2342         btrfs_stop_workers(&fs_info->submit_workers);
2343
2344         btrfs_close_devices(fs_info->fs_devices);
2345         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2346
2347         bdi_destroy(&fs_info->bdi);
2348
2349         kfree(fs_info->extent_root);
2350         kfree(fs_info->tree_root);
2351         kfree(fs_info->chunk_root);
2352         kfree(fs_info->dev_root);
2353         kfree(fs_info->csum_root);
2354         return 0;
2355 }
2356
2357 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2358 {
2359         int ret;
2360         struct inode *btree_inode = buf->first_page->mapping->host;
2361
2362         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2363         if (!ret)
2364                 return ret;
2365
2366         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2367                                     parent_transid);
2368         return !ret;
2369 }
2370
2371 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2372 {
2373         struct inode *btree_inode = buf->first_page->mapping->host;
2374         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2375                                           buf);
2376 }
2377
2378 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2379 {
2380         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2381         u64 transid = btrfs_header_generation(buf);
2382         struct inode *btree_inode = root->fs_info->btree_inode;
2383         int was_dirty;
2384
2385         btrfs_assert_tree_locked(buf);
2386         if (transid != root->fs_info->generation) {
2387                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2388                        "found %llu running %llu\n",
2389                         (unsigned long long)buf->start,
2390                         (unsigned long long)transid,
2391                         (unsigned long long)root->fs_info->generation);
2392                 WARN_ON(1);
2393         }
2394         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2395                                             buf);
2396         if (!was_dirty) {
2397                 spin_lock(&root->fs_info->delalloc_lock);
2398                 root->fs_info->dirty_metadata_bytes += buf->len;
2399                 spin_unlock(&root->fs_info->delalloc_lock);
2400         }
2401 }
2402
2403 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2404 {
2405         /*
2406          * looks as though older kernels can get into trouble with
2407          * this code, they end up stuck in balance_dirty_pages forever
2408          */
2409         u64 num_dirty;
2410         unsigned long thresh = 32 * 1024 * 1024;
2411
2412         if (current->flags & PF_MEMALLOC)
2413                 return;
2414
2415         num_dirty = root->fs_info->dirty_metadata_bytes;
2416
2417         if (num_dirty > thresh) {
2418                 balance_dirty_pages_ratelimited_nr(
2419                                    root->fs_info->btree_inode->i_mapping, 1);
2420         }
2421         return;
2422 }
2423
2424 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2425 {
2426         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2427         int ret;
2428         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2429         if (ret == 0)
2430                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2431         return ret;
2432 }
2433
2434 int btree_lock_page_hook(struct page *page)
2435 {
2436         struct inode *inode = page->mapping->host;
2437         struct btrfs_root *root = BTRFS_I(inode)->root;
2438         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2439         struct extent_buffer *eb;
2440         unsigned long len;
2441         u64 bytenr = page_offset(page);
2442
2443         if (page->private == EXTENT_PAGE_PRIVATE)
2444                 goto out;
2445
2446         len = page->private >> 2;
2447         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2448         if (!eb)
2449                 goto out;
2450
2451         btrfs_tree_lock(eb);
2452         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2453
2454         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2455                 spin_lock(&root->fs_info->delalloc_lock);
2456                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2457                         root->fs_info->dirty_metadata_bytes -= eb->len;
2458                 else
2459                         WARN_ON(1);
2460                 spin_unlock(&root->fs_info->delalloc_lock);
2461         }
2462
2463         btrfs_tree_unlock(eb);
2464         free_extent_buffer(eb);
2465 out:
2466         lock_page(page);
2467         return 0;
2468 }
2469
2470 static struct extent_io_ops btree_extent_io_ops = {
2471         .write_cache_pages_lock_hook = btree_lock_page_hook,
2472         .readpage_end_io_hook = btree_readpage_end_io_hook,
2473         .submit_bio_hook = btree_submit_bio_hook,
2474         /* note we're sharing with inode.c for the merge bio hook */
2475         .merge_bio_hook = btrfs_merge_bio_hook,
2476 };