4084959b36fdf82095727139916345d4bac5374e
[linux-3.10.git] / fs / btrfs / disk-io.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <asm/unaligned.h>
33 #include "compat.h"
34 #include "ctree.h"
35 #include "disk-io.h"
36 #include "transaction.h"
37 #include "btrfs_inode.h"
38 #include "volumes.h"
39 #include "print-tree.h"
40 #include "async-thread.h"
41 #include "locking.h"
42 #include "tree-log.h"
43 #include "free-space-cache.h"
44
45 static struct extent_io_ops btree_extent_io_ops;
46 static void end_workqueue_fn(struct btrfs_work *work);
47 static void free_fs_root(struct btrfs_root *root);
48 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
49                                     int read_only);
50 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
51 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
52 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
53                                       struct btrfs_root *root);
54 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
55 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
56 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
57                                         struct extent_io_tree *dirty_pages,
58                                         int mark);
59 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
60                                        struct extent_io_tree *pinned_extents);
61 static int btrfs_cleanup_transaction(struct btrfs_root *root);
62
63 /*
64  * end_io_wq structs are used to do processing in task context when an IO is
65  * complete.  This is used during reads to verify checksums, and it is used
66  * by writes to insert metadata for new file extents after IO is complete.
67  */
68 struct end_io_wq {
69         struct bio *bio;
70         bio_end_io_t *end_io;
71         void *private;
72         struct btrfs_fs_info *info;
73         int error;
74         int metadata;
75         struct list_head list;
76         struct btrfs_work work;
77 };
78
79 /*
80  * async submit bios are used to offload expensive checksumming
81  * onto the worker threads.  They checksum file and metadata bios
82  * just before they are sent down the IO stack.
83  */
84 struct async_submit_bio {
85         struct inode *inode;
86         struct bio *bio;
87         struct list_head list;
88         extent_submit_bio_hook_t *submit_bio_start;
89         extent_submit_bio_hook_t *submit_bio_done;
90         int rw;
91         int mirror_num;
92         unsigned long bio_flags;
93         /*
94          * bio_offset is optional, can be used if the pages in the bio
95          * can't tell us where in the file the bio should go
96          */
97         u64 bio_offset;
98         struct btrfs_work work;
99 };
100
101 /* These are used to set the lockdep class on the extent buffer locks.
102  * The class is set by the readpage_end_io_hook after the buffer has
103  * passed csum validation but before the pages are unlocked.
104  *
105  * The lockdep class is also set by btrfs_init_new_buffer on freshly
106  * allocated blocks.
107  *
108  * The class is based on the level in the tree block, which allows lockdep
109  * to know that lower nodes nest inside the locks of higher nodes.
110  *
111  * We also add a check to make sure the highest level of the tree is
112  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
113  * code needs update as well.
114  */
115 #ifdef CONFIG_DEBUG_LOCK_ALLOC
116 # if BTRFS_MAX_LEVEL != 8
117 #  error
118 # endif
119 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
120 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
121         /* leaf */
122         "btrfs-extent-00",
123         "btrfs-extent-01",
124         "btrfs-extent-02",
125         "btrfs-extent-03",
126         "btrfs-extent-04",
127         "btrfs-extent-05",
128         "btrfs-extent-06",
129         "btrfs-extent-07",
130         /* highest possible level */
131         "btrfs-extent-08",
132 };
133 #endif
134
135 /*
136  * extents on the btree inode are pretty simple, there's one extent
137  * that covers the entire device
138  */
139 static struct extent_map *btree_get_extent(struct inode *inode,
140                 struct page *page, size_t pg_offset, u64 start, u64 len,
141                 int create)
142 {
143         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
144         struct extent_map *em;
145         int ret;
146
147         read_lock(&em_tree->lock);
148         em = lookup_extent_mapping(em_tree, start, len);
149         if (em) {
150                 em->bdev =
151                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
152                 read_unlock(&em_tree->lock);
153                 goto out;
154         }
155         read_unlock(&em_tree->lock);
156
157         em = alloc_extent_map();
158         if (!em) {
159                 em = ERR_PTR(-ENOMEM);
160                 goto out;
161         }
162         em->start = 0;
163         em->len = (u64)-1;
164         em->block_len = (u64)-1;
165         em->block_start = 0;
166         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
167
168         write_lock(&em_tree->lock);
169         ret = add_extent_mapping(em_tree, em);
170         if (ret == -EEXIST) {
171                 u64 failed_start = em->start;
172                 u64 failed_len = em->len;
173
174                 free_extent_map(em);
175                 em = lookup_extent_mapping(em_tree, start, len);
176                 if (em) {
177                         ret = 0;
178                 } else {
179                         em = lookup_extent_mapping(em_tree, failed_start,
180                                                    failed_len);
181                         ret = -EIO;
182                 }
183         } else if (ret) {
184                 free_extent_map(em);
185                 em = NULL;
186         }
187         write_unlock(&em_tree->lock);
188
189         if (ret)
190                 em = ERR_PTR(ret);
191 out:
192         return em;
193 }
194
195 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
196 {
197         return crc32c(seed, data, len);
198 }
199
200 void btrfs_csum_final(u32 crc, char *result)
201 {
202         put_unaligned_le32(~crc, result);
203 }
204
205 /*
206  * compute the csum for a btree block, and either verify it or write it
207  * into the csum field of the block.
208  */
209 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
210                            int verify)
211 {
212         u16 csum_size =
213                 btrfs_super_csum_size(&root->fs_info->super_copy);
214         char *result = NULL;
215         unsigned long len;
216         unsigned long cur_len;
217         unsigned long offset = BTRFS_CSUM_SIZE;
218         char *map_token = NULL;
219         char *kaddr;
220         unsigned long map_start;
221         unsigned long map_len;
222         int err;
223         u32 crc = ~(u32)0;
224         unsigned long inline_result;
225
226         len = buf->len - offset;
227         while (len > 0) {
228                 err = map_private_extent_buffer(buf, offset, 32,
229                                         &map_token, &kaddr,
230                                         &map_start, &map_len, KM_USER0);
231                 if (err)
232                         return 1;
233                 cur_len = min(len, map_len - (offset - map_start));
234                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
235                                       crc, cur_len);
236                 len -= cur_len;
237                 offset += cur_len;
238                 unmap_extent_buffer(buf, map_token, KM_USER0);
239         }
240         if (csum_size > sizeof(inline_result)) {
241                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
242                 if (!result)
243                         return 1;
244         } else {
245                 result = (char *)&inline_result;
246         }
247
248         btrfs_csum_final(crc, result);
249
250         if (verify) {
251                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
252                         u32 val;
253                         u32 found = 0;
254                         memcpy(&found, result, csum_size);
255
256                         read_extent_buffer(buf, &val, 0, csum_size);
257                         if (printk_ratelimit()) {
258                                 printk(KERN_INFO "btrfs: %s checksum verify "
259                                        "failed on %llu wanted %X found %X "
260                                        "level %d\n",
261                                        root->fs_info->sb->s_id,
262                                        (unsigned long long)buf->start, val, found,
263                                        btrfs_header_level(buf));
264                         }
265                         if (result != (char *)&inline_result)
266                                 kfree(result);
267                         return 1;
268                 }
269         } else {
270                 write_extent_buffer(buf, result, 0, csum_size);
271         }
272         if (result != (char *)&inline_result)
273                 kfree(result);
274         return 0;
275 }
276
277 /*
278  * we can't consider a given block up to date unless the transid of the
279  * block matches the transid in the parent node's pointer.  This is how we
280  * detect blocks that either didn't get written at all or got written
281  * in the wrong place.
282  */
283 static int verify_parent_transid(struct extent_io_tree *io_tree,
284                                  struct extent_buffer *eb, u64 parent_transid)
285 {
286         struct extent_state *cached_state = NULL;
287         int ret;
288
289         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
290                 return 0;
291
292         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
293                          0, &cached_state, GFP_NOFS);
294         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
295             btrfs_header_generation(eb) == parent_transid) {
296                 ret = 0;
297                 goto out;
298         }
299         if (printk_ratelimit()) {
300                 printk("parent transid verify failed on %llu wanted %llu "
301                        "found %llu\n",
302                        (unsigned long long)eb->start,
303                        (unsigned long long)parent_transid,
304                        (unsigned long long)btrfs_header_generation(eb));
305         }
306         ret = 1;
307         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
308 out:
309         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
310                              &cached_state, GFP_NOFS);
311         return ret;
312 }
313
314 /*
315  * helper to read a given tree block, doing retries as required when
316  * the checksums don't match and we have alternate mirrors to try.
317  */
318 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
319                                           struct extent_buffer *eb,
320                                           u64 start, u64 parent_transid)
321 {
322         struct extent_io_tree *io_tree;
323         int ret;
324         int num_copies = 0;
325         int mirror_num = 0;
326
327         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
328         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
329         while (1) {
330                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
331                                                btree_get_extent, mirror_num);
332                 if (!ret &&
333                     !verify_parent_transid(io_tree, eb, parent_transid))
334                         return ret;
335
336                 /*
337                  * This buffer's crc is fine, but its contents are corrupted, so
338                  * there is no reason to read the other copies, they won't be
339                  * any less wrong.
340                  */
341                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
342                         return ret;
343
344                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
345                                               eb->start, eb->len);
346                 if (num_copies == 1)
347                         return ret;
348
349                 mirror_num++;
350                 if (mirror_num > num_copies)
351                         return ret;
352         }
353         return -EIO;
354 }
355
356 /*
357  * checksum a dirty tree block before IO.  This has extra checks to make sure
358  * we only fill in the checksum field in the first page of a multi-page block
359  */
360
361 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
362 {
363         struct extent_io_tree *tree;
364         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
365         u64 found_start;
366         unsigned long len;
367         struct extent_buffer *eb;
368         int ret;
369
370         tree = &BTRFS_I(page->mapping->host)->io_tree;
371
372         if (page->private == EXTENT_PAGE_PRIVATE) {
373                 WARN_ON(1);
374                 goto out;
375         }
376         if (!page->private) {
377                 WARN_ON(1);
378                 goto out;
379         }
380         len = page->private >> 2;
381         WARN_ON(len == 0);
382
383         eb = alloc_extent_buffer(tree, start, len, page);
384         if (eb == NULL) {
385                 WARN_ON(1);
386                 goto out;
387         }
388         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
389                                              btrfs_header_generation(eb));
390         BUG_ON(ret);
391         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
392
393         found_start = btrfs_header_bytenr(eb);
394         if (found_start != start) {
395                 WARN_ON(1);
396                 goto err;
397         }
398         if (eb->first_page != page) {
399                 WARN_ON(1);
400                 goto err;
401         }
402         if (!PageUptodate(page)) {
403                 WARN_ON(1);
404                 goto err;
405         }
406         csum_tree_block(root, eb, 0);
407 err:
408         free_extent_buffer(eb);
409 out:
410         return 0;
411 }
412
413 static int check_tree_block_fsid(struct btrfs_root *root,
414                                  struct extent_buffer *eb)
415 {
416         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
417         u8 fsid[BTRFS_UUID_SIZE];
418         int ret = 1;
419
420         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
421                            BTRFS_FSID_SIZE);
422         while (fs_devices) {
423                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
424                         ret = 0;
425                         break;
426                 }
427                 fs_devices = fs_devices->seed;
428         }
429         return ret;
430 }
431
432 #define CORRUPT(reason, eb, root, slot)                         \
433         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
434                "root=%llu, slot=%d\n", reason,                  \
435                (unsigned long long)btrfs_header_bytenr(eb),     \
436                (unsigned long long)root->objectid, slot)
437
438 static noinline int check_leaf(struct btrfs_root *root,
439                                struct extent_buffer *leaf)
440 {
441         struct btrfs_key key;
442         struct btrfs_key leaf_key;
443         u32 nritems = btrfs_header_nritems(leaf);
444         int slot;
445
446         if (nritems == 0)
447                 return 0;
448
449         /* Check the 0 item */
450         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
451             BTRFS_LEAF_DATA_SIZE(root)) {
452                 CORRUPT("invalid item offset size pair", leaf, root, 0);
453                 return -EIO;
454         }
455
456         /*
457          * Check to make sure each items keys are in the correct order and their
458          * offsets make sense.  We only have to loop through nritems-1 because
459          * we check the current slot against the next slot, which verifies the
460          * next slot's offset+size makes sense and that the current's slot
461          * offset is correct.
462          */
463         for (slot = 0; slot < nritems - 1; slot++) {
464                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
465                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
466
467                 /* Make sure the keys are in the right order */
468                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
469                         CORRUPT("bad key order", leaf, root, slot);
470                         return -EIO;
471                 }
472
473                 /*
474                  * Make sure the offset and ends are right, remember that the
475                  * item data starts at the end of the leaf and grows towards the
476                  * front.
477                  */
478                 if (btrfs_item_offset_nr(leaf, slot) !=
479                         btrfs_item_end_nr(leaf, slot + 1)) {
480                         CORRUPT("slot offset bad", leaf, root, slot);
481                         return -EIO;
482                 }
483
484                 /*
485                  * Check to make sure that we don't point outside of the leaf,
486                  * just incase all the items are consistent to eachother, but
487                  * all point outside of the leaf.
488                  */
489                 if (btrfs_item_end_nr(leaf, slot) >
490                     BTRFS_LEAF_DATA_SIZE(root)) {
491                         CORRUPT("slot end outside of leaf", leaf, root, slot);
492                         return -EIO;
493                 }
494         }
495
496         return 0;
497 }
498
499 #ifdef CONFIG_DEBUG_LOCK_ALLOC
500 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
501 {
502         lockdep_set_class_and_name(&eb->lock,
503                            &btrfs_eb_class[level],
504                            btrfs_eb_name[level]);
505 }
506 #endif
507
508 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
509                                struct extent_state *state)
510 {
511         struct extent_io_tree *tree;
512         u64 found_start;
513         int found_level;
514         unsigned long len;
515         struct extent_buffer *eb;
516         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
517         int ret = 0;
518
519         tree = &BTRFS_I(page->mapping->host)->io_tree;
520         if (page->private == EXTENT_PAGE_PRIVATE)
521                 goto out;
522         if (!page->private)
523                 goto out;
524
525         len = page->private >> 2;
526         WARN_ON(len == 0);
527
528         eb = alloc_extent_buffer(tree, start, len, page);
529         if (eb == NULL) {
530                 ret = -EIO;
531                 goto out;
532         }
533
534         found_start = btrfs_header_bytenr(eb);
535         if (found_start != start) {
536                 if (printk_ratelimit()) {
537                         printk(KERN_INFO "btrfs bad tree block start "
538                                "%llu %llu\n",
539                                (unsigned long long)found_start,
540                                (unsigned long long)eb->start);
541                 }
542                 ret = -EIO;
543                 goto err;
544         }
545         if (eb->first_page != page) {
546                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
547                        eb->first_page->index, page->index);
548                 WARN_ON(1);
549                 ret = -EIO;
550                 goto err;
551         }
552         if (check_tree_block_fsid(root, eb)) {
553                 if (printk_ratelimit()) {
554                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
555                                (unsigned long long)eb->start);
556                 }
557                 ret = -EIO;
558                 goto err;
559         }
560         found_level = btrfs_header_level(eb);
561
562         btrfs_set_buffer_lockdep_class(eb, found_level);
563
564         ret = csum_tree_block(root, eb, 1);
565         if (ret) {
566                 ret = -EIO;
567                 goto err;
568         }
569
570         /*
571          * If this is a leaf block and it is corrupt, set the corrupt bit so
572          * that we don't try and read the other copies of this block, just
573          * return -EIO.
574          */
575         if (found_level == 0 && check_leaf(root, eb)) {
576                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
577                 ret = -EIO;
578         }
579
580         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
581         end = eb->start + end - 1;
582 err:
583         free_extent_buffer(eb);
584 out:
585         return ret;
586 }
587
588 static void end_workqueue_bio(struct bio *bio, int err)
589 {
590         struct end_io_wq *end_io_wq = bio->bi_private;
591         struct btrfs_fs_info *fs_info;
592
593         fs_info = end_io_wq->info;
594         end_io_wq->error = err;
595         end_io_wq->work.func = end_workqueue_fn;
596         end_io_wq->work.flags = 0;
597
598         if (bio->bi_rw & REQ_WRITE) {
599                 if (end_io_wq->metadata == 1)
600                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
601                                            &end_io_wq->work);
602                 else if (end_io_wq->metadata == 2)
603                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
604                                            &end_io_wq->work);
605                 else
606                         btrfs_queue_worker(&fs_info->endio_write_workers,
607                                            &end_io_wq->work);
608         } else {
609                 if (end_io_wq->metadata)
610                         btrfs_queue_worker(&fs_info->endio_meta_workers,
611                                            &end_io_wq->work);
612                 else
613                         btrfs_queue_worker(&fs_info->endio_workers,
614                                            &end_io_wq->work);
615         }
616 }
617
618 /*
619  * For the metadata arg you want
620  *
621  * 0 - if data
622  * 1 - if normal metadta
623  * 2 - if writing to the free space cache area
624  */
625 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
626                         int metadata)
627 {
628         struct end_io_wq *end_io_wq;
629         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
630         if (!end_io_wq)
631                 return -ENOMEM;
632
633         end_io_wq->private = bio->bi_private;
634         end_io_wq->end_io = bio->bi_end_io;
635         end_io_wq->info = info;
636         end_io_wq->error = 0;
637         end_io_wq->bio = bio;
638         end_io_wq->metadata = metadata;
639
640         bio->bi_private = end_io_wq;
641         bio->bi_end_io = end_workqueue_bio;
642         return 0;
643 }
644
645 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
646 {
647         unsigned long limit = min_t(unsigned long,
648                                     info->workers.max_workers,
649                                     info->fs_devices->open_devices);
650         return 256 * limit;
651 }
652
653 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
654 {
655         return atomic_read(&info->nr_async_bios) >
656                 btrfs_async_submit_limit(info);
657 }
658
659 static void run_one_async_start(struct btrfs_work *work)
660 {
661         struct async_submit_bio *async;
662
663         async = container_of(work, struct  async_submit_bio, work);
664         async->submit_bio_start(async->inode, async->rw, async->bio,
665                                async->mirror_num, async->bio_flags,
666                                async->bio_offset);
667 }
668
669 static void run_one_async_done(struct btrfs_work *work)
670 {
671         struct btrfs_fs_info *fs_info;
672         struct async_submit_bio *async;
673         int limit;
674
675         async = container_of(work, struct  async_submit_bio, work);
676         fs_info = BTRFS_I(async->inode)->root->fs_info;
677
678         limit = btrfs_async_submit_limit(fs_info);
679         limit = limit * 2 / 3;
680
681         atomic_dec(&fs_info->nr_async_submits);
682
683         if (atomic_read(&fs_info->nr_async_submits) < limit &&
684             waitqueue_active(&fs_info->async_submit_wait))
685                 wake_up(&fs_info->async_submit_wait);
686
687         async->submit_bio_done(async->inode, async->rw, async->bio,
688                                async->mirror_num, async->bio_flags,
689                                async->bio_offset);
690 }
691
692 static void run_one_async_free(struct btrfs_work *work)
693 {
694         struct async_submit_bio *async;
695
696         async = container_of(work, struct  async_submit_bio, work);
697         kfree(async);
698 }
699
700 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
701                         int rw, struct bio *bio, int mirror_num,
702                         unsigned long bio_flags,
703                         u64 bio_offset,
704                         extent_submit_bio_hook_t *submit_bio_start,
705                         extent_submit_bio_hook_t *submit_bio_done)
706 {
707         struct async_submit_bio *async;
708
709         async = kmalloc(sizeof(*async), GFP_NOFS);
710         if (!async)
711                 return -ENOMEM;
712
713         async->inode = inode;
714         async->rw = rw;
715         async->bio = bio;
716         async->mirror_num = mirror_num;
717         async->submit_bio_start = submit_bio_start;
718         async->submit_bio_done = submit_bio_done;
719
720         async->work.func = run_one_async_start;
721         async->work.ordered_func = run_one_async_done;
722         async->work.ordered_free = run_one_async_free;
723
724         async->work.flags = 0;
725         async->bio_flags = bio_flags;
726         async->bio_offset = bio_offset;
727
728         atomic_inc(&fs_info->nr_async_submits);
729
730         if (rw & REQ_SYNC)
731                 btrfs_set_work_high_prio(&async->work);
732
733         btrfs_queue_worker(&fs_info->workers, &async->work);
734
735         while (atomic_read(&fs_info->async_submit_draining) &&
736               atomic_read(&fs_info->nr_async_submits)) {
737                 wait_event(fs_info->async_submit_wait,
738                            (atomic_read(&fs_info->nr_async_submits) == 0));
739         }
740
741         return 0;
742 }
743
744 static int btree_csum_one_bio(struct bio *bio)
745 {
746         struct bio_vec *bvec = bio->bi_io_vec;
747         int bio_index = 0;
748         struct btrfs_root *root;
749
750         WARN_ON(bio->bi_vcnt <= 0);
751         while (bio_index < bio->bi_vcnt) {
752                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
753                 csum_dirty_buffer(root, bvec->bv_page);
754                 bio_index++;
755                 bvec++;
756         }
757         return 0;
758 }
759
760 static int __btree_submit_bio_start(struct inode *inode, int rw,
761                                     struct bio *bio, int mirror_num,
762                                     unsigned long bio_flags,
763                                     u64 bio_offset)
764 {
765         /*
766          * when we're called for a write, we're already in the async
767          * submission context.  Just jump into btrfs_map_bio
768          */
769         btree_csum_one_bio(bio);
770         return 0;
771 }
772
773 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
774                                  int mirror_num, unsigned long bio_flags,
775                                  u64 bio_offset)
776 {
777         /*
778          * when we're called for a write, we're already in the async
779          * submission context.  Just jump into btrfs_map_bio
780          */
781         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
782 }
783
784 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
785                                  int mirror_num, unsigned long bio_flags,
786                                  u64 bio_offset)
787 {
788         int ret;
789
790         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
791                                           bio, 1);
792         BUG_ON(ret);
793
794         if (!(rw & REQ_WRITE)) {
795                 /*
796                  * called for a read, do the setup so that checksum validation
797                  * can happen in the async kernel threads
798                  */
799                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
800                                      mirror_num, 0);
801         }
802
803         /*
804          * kthread helpers are used to submit writes so that checksumming
805          * can happen in parallel across all CPUs
806          */
807         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
808                                    inode, rw, bio, mirror_num, 0,
809                                    bio_offset,
810                                    __btree_submit_bio_start,
811                                    __btree_submit_bio_done);
812 }
813
814 #ifdef CONFIG_MIGRATION
815 static int btree_migratepage(struct address_space *mapping,
816                         struct page *newpage, struct page *page)
817 {
818         /*
819          * we can't safely write a btree page from here,
820          * we haven't done the locking hook
821          */
822         if (PageDirty(page))
823                 return -EAGAIN;
824         /*
825          * Buffers may be managed in a filesystem specific way.
826          * We must have no buffers or drop them.
827          */
828         if (page_has_private(page) &&
829             !try_to_release_page(page, GFP_KERNEL))
830                 return -EAGAIN;
831         return migrate_page(mapping, newpage, page);
832 }
833 #endif
834
835 static int btree_writepage(struct page *page, struct writeback_control *wbc)
836 {
837         struct extent_io_tree *tree;
838         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
839         struct extent_buffer *eb;
840         int was_dirty;
841
842         tree = &BTRFS_I(page->mapping->host)->io_tree;
843         if (!(current->flags & PF_MEMALLOC)) {
844                 return extent_write_full_page(tree, page,
845                                               btree_get_extent, wbc);
846         }
847
848         redirty_page_for_writepage(wbc, page);
849         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
850         WARN_ON(!eb);
851
852         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
853         if (!was_dirty) {
854                 spin_lock(&root->fs_info->delalloc_lock);
855                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
856                 spin_unlock(&root->fs_info->delalloc_lock);
857         }
858         free_extent_buffer(eb);
859
860         unlock_page(page);
861         return 0;
862 }
863
864 static int btree_writepages(struct address_space *mapping,
865                             struct writeback_control *wbc)
866 {
867         struct extent_io_tree *tree;
868         tree = &BTRFS_I(mapping->host)->io_tree;
869         if (wbc->sync_mode == WB_SYNC_NONE) {
870                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
871                 u64 num_dirty;
872                 unsigned long thresh = 32 * 1024 * 1024;
873
874                 if (wbc->for_kupdate)
875                         return 0;
876
877                 /* this is a bit racy, but that's ok */
878                 num_dirty = root->fs_info->dirty_metadata_bytes;
879                 if (num_dirty < thresh)
880                         return 0;
881         }
882         return extent_writepages(tree, mapping, btree_get_extent, wbc);
883 }
884
885 static int btree_readpage(struct file *file, struct page *page)
886 {
887         struct extent_io_tree *tree;
888         tree = &BTRFS_I(page->mapping->host)->io_tree;
889         return extent_read_full_page(tree, page, btree_get_extent);
890 }
891
892 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
893 {
894         struct extent_io_tree *tree;
895         struct extent_map_tree *map;
896         int ret;
897
898         if (PageWriteback(page) || PageDirty(page))
899                 return 0;
900
901         tree = &BTRFS_I(page->mapping->host)->io_tree;
902         map = &BTRFS_I(page->mapping->host)->extent_tree;
903
904         ret = try_release_extent_state(map, tree, page, gfp_flags);
905         if (!ret)
906                 return 0;
907
908         ret = try_release_extent_buffer(tree, page);
909         if (ret == 1) {
910                 ClearPagePrivate(page);
911                 set_page_private(page, 0);
912                 page_cache_release(page);
913         }
914
915         return ret;
916 }
917
918 static void btree_invalidatepage(struct page *page, unsigned long offset)
919 {
920         struct extent_io_tree *tree;
921         tree = &BTRFS_I(page->mapping->host)->io_tree;
922         extent_invalidatepage(tree, page, offset);
923         btree_releasepage(page, GFP_NOFS);
924         if (PagePrivate(page)) {
925                 printk(KERN_WARNING "btrfs warning page private not zero "
926                        "on page %llu\n", (unsigned long long)page_offset(page));
927                 ClearPagePrivate(page);
928                 set_page_private(page, 0);
929                 page_cache_release(page);
930         }
931 }
932
933 static const struct address_space_operations btree_aops = {
934         .readpage       = btree_readpage,
935         .writepage      = btree_writepage,
936         .writepages     = btree_writepages,
937         .releasepage    = btree_releasepage,
938         .invalidatepage = btree_invalidatepage,
939 #ifdef CONFIG_MIGRATION
940         .migratepage    = btree_migratepage,
941 #endif
942 };
943
944 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
945                          u64 parent_transid)
946 {
947         struct extent_buffer *buf = NULL;
948         struct inode *btree_inode = root->fs_info->btree_inode;
949         int ret = 0;
950
951         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
952         if (!buf)
953                 return 0;
954         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
955                                  buf, 0, 0, btree_get_extent, 0);
956         free_extent_buffer(buf);
957         return ret;
958 }
959
960 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
961                                             u64 bytenr, u32 blocksize)
962 {
963         struct inode *btree_inode = root->fs_info->btree_inode;
964         struct extent_buffer *eb;
965         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
966                                 bytenr, blocksize);
967         return eb;
968 }
969
970 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
971                                                  u64 bytenr, u32 blocksize)
972 {
973         struct inode *btree_inode = root->fs_info->btree_inode;
974         struct extent_buffer *eb;
975
976         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
977                                  bytenr, blocksize, NULL);
978         return eb;
979 }
980
981
982 int btrfs_write_tree_block(struct extent_buffer *buf)
983 {
984         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
985                                         buf->start + buf->len - 1);
986 }
987
988 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
989 {
990         return filemap_fdatawait_range(buf->first_page->mapping,
991                                        buf->start, buf->start + buf->len - 1);
992 }
993
994 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
995                                       u32 blocksize, u64 parent_transid)
996 {
997         struct extent_buffer *buf = NULL;
998         int ret;
999
1000         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1001         if (!buf)
1002                 return NULL;
1003
1004         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1005
1006         if (ret == 0)
1007                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1008         return buf;
1009
1010 }
1011
1012 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1013                      struct extent_buffer *buf)
1014 {
1015         struct inode *btree_inode = root->fs_info->btree_inode;
1016         if (btrfs_header_generation(buf) ==
1017             root->fs_info->running_transaction->transid) {
1018                 btrfs_assert_tree_locked(buf);
1019
1020                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1021                         spin_lock(&root->fs_info->delalloc_lock);
1022                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1023                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1024                         else
1025                                 WARN_ON(1);
1026                         spin_unlock(&root->fs_info->delalloc_lock);
1027                 }
1028
1029                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1030                 btrfs_set_lock_blocking(buf);
1031                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1032                                           buf);
1033         }
1034         return 0;
1035 }
1036
1037 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1038                         u32 stripesize, struct btrfs_root *root,
1039                         struct btrfs_fs_info *fs_info,
1040                         u64 objectid)
1041 {
1042         root->node = NULL;
1043         root->commit_root = NULL;
1044         root->sectorsize = sectorsize;
1045         root->nodesize = nodesize;
1046         root->leafsize = leafsize;
1047         root->stripesize = stripesize;
1048         root->ref_cows = 0;
1049         root->track_dirty = 0;
1050         root->in_radix = 0;
1051         root->orphan_item_inserted = 0;
1052         root->orphan_cleanup_state = 0;
1053
1054         root->fs_info = fs_info;
1055         root->objectid = objectid;
1056         root->last_trans = 0;
1057         root->highest_objectid = 0;
1058         root->name = NULL;
1059         root->in_sysfs = 0;
1060         root->inode_tree = RB_ROOT;
1061         root->block_rsv = NULL;
1062         root->orphan_block_rsv = NULL;
1063
1064         INIT_LIST_HEAD(&root->dirty_list);
1065         INIT_LIST_HEAD(&root->orphan_list);
1066         INIT_LIST_HEAD(&root->root_list);
1067         spin_lock_init(&root->node_lock);
1068         spin_lock_init(&root->orphan_lock);
1069         spin_lock_init(&root->inode_lock);
1070         spin_lock_init(&root->accounting_lock);
1071         mutex_init(&root->objectid_mutex);
1072         mutex_init(&root->log_mutex);
1073         init_waitqueue_head(&root->log_writer_wait);
1074         init_waitqueue_head(&root->log_commit_wait[0]);
1075         init_waitqueue_head(&root->log_commit_wait[1]);
1076         atomic_set(&root->log_commit[0], 0);
1077         atomic_set(&root->log_commit[1], 0);
1078         atomic_set(&root->log_writers, 0);
1079         root->log_batch = 0;
1080         root->log_transid = 0;
1081         root->last_log_commit = 0;
1082         extent_io_tree_init(&root->dirty_log_pages,
1083                              fs_info->btree_inode->i_mapping);
1084
1085         memset(&root->root_key, 0, sizeof(root->root_key));
1086         memset(&root->root_item, 0, sizeof(root->root_item));
1087         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1088         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1089         root->defrag_trans_start = fs_info->generation;
1090         init_completion(&root->kobj_unregister);
1091         root->defrag_running = 0;
1092         root->root_key.objectid = objectid;
1093         root->anon_super.s_root = NULL;
1094         root->anon_super.s_dev = 0;
1095         INIT_LIST_HEAD(&root->anon_super.s_list);
1096         INIT_LIST_HEAD(&root->anon_super.s_instances);
1097         init_rwsem(&root->anon_super.s_umount);
1098
1099         return 0;
1100 }
1101
1102 static int find_and_setup_root(struct btrfs_root *tree_root,
1103                                struct btrfs_fs_info *fs_info,
1104                                u64 objectid,
1105                                struct btrfs_root *root)
1106 {
1107         int ret;
1108         u32 blocksize;
1109         u64 generation;
1110
1111         __setup_root(tree_root->nodesize, tree_root->leafsize,
1112                      tree_root->sectorsize, tree_root->stripesize,
1113                      root, fs_info, objectid);
1114         ret = btrfs_find_last_root(tree_root, objectid,
1115                                    &root->root_item, &root->root_key);
1116         if (ret > 0)
1117                 return -ENOENT;
1118         BUG_ON(ret);
1119
1120         generation = btrfs_root_generation(&root->root_item);
1121         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1122         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1123                                      blocksize, generation);
1124         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1125                 free_extent_buffer(root->node);
1126                 return -EIO;
1127         }
1128         root->commit_root = btrfs_root_node(root);
1129         return 0;
1130 }
1131
1132 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1133                                          struct btrfs_fs_info *fs_info)
1134 {
1135         struct btrfs_root *root;
1136         struct btrfs_root *tree_root = fs_info->tree_root;
1137         struct extent_buffer *leaf;
1138
1139         root = kzalloc(sizeof(*root), GFP_NOFS);
1140         if (!root)
1141                 return ERR_PTR(-ENOMEM);
1142
1143         __setup_root(tree_root->nodesize, tree_root->leafsize,
1144                      tree_root->sectorsize, tree_root->stripesize,
1145                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1146
1147         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1148         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1149         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1150         /*
1151          * log trees do not get reference counted because they go away
1152          * before a real commit is actually done.  They do store pointers
1153          * to file data extents, and those reference counts still get
1154          * updated (along with back refs to the log tree).
1155          */
1156         root->ref_cows = 0;
1157
1158         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1159                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1160         if (IS_ERR(leaf)) {
1161                 kfree(root);
1162                 return ERR_CAST(leaf);
1163         }
1164
1165         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1166         btrfs_set_header_bytenr(leaf, leaf->start);
1167         btrfs_set_header_generation(leaf, trans->transid);
1168         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1169         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1170         root->node = leaf;
1171
1172         write_extent_buffer(root->node, root->fs_info->fsid,
1173                             (unsigned long)btrfs_header_fsid(root->node),
1174                             BTRFS_FSID_SIZE);
1175         btrfs_mark_buffer_dirty(root->node);
1176         btrfs_tree_unlock(root->node);
1177         return root;
1178 }
1179
1180 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1181                              struct btrfs_fs_info *fs_info)
1182 {
1183         struct btrfs_root *log_root;
1184
1185         log_root = alloc_log_tree(trans, fs_info);
1186         if (IS_ERR(log_root))
1187                 return PTR_ERR(log_root);
1188         WARN_ON(fs_info->log_root_tree);
1189         fs_info->log_root_tree = log_root;
1190         return 0;
1191 }
1192
1193 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1194                        struct btrfs_root *root)
1195 {
1196         struct btrfs_root *log_root;
1197         struct btrfs_inode_item *inode_item;
1198
1199         log_root = alloc_log_tree(trans, root->fs_info);
1200         if (IS_ERR(log_root))
1201                 return PTR_ERR(log_root);
1202
1203         log_root->last_trans = trans->transid;
1204         log_root->root_key.offset = root->root_key.objectid;
1205
1206         inode_item = &log_root->root_item.inode;
1207         inode_item->generation = cpu_to_le64(1);
1208         inode_item->size = cpu_to_le64(3);
1209         inode_item->nlink = cpu_to_le32(1);
1210         inode_item->nbytes = cpu_to_le64(root->leafsize);
1211         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1212
1213         btrfs_set_root_node(&log_root->root_item, log_root->node);
1214
1215         WARN_ON(root->log_root);
1216         root->log_root = log_root;
1217         root->log_transid = 0;
1218         root->last_log_commit = 0;
1219         return 0;
1220 }
1221
1222 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1223                                                struct btrfs_key *location)
1224 {
1225         struct btrfs_root *root;
1226         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1227         struct btrfs_path *path;
1228         struct extent_buffer *l;
1229         u64 generation;
1230         u32 blocksize;
1231         int ret = 0;
1232
1233         root = kzalloc(sizeof(*root), GFP_NOFS);
1234         if (!root)
1235                 return ERR_PTR(-ENOMEM);
1236         if (location->offset == (u64)-1) {
1237                 ret = find_and_setup_root(tree_root, fs_info,
1238                                           location->objectid, root);
1239                 if (ret) {
1240                         kfree(root);
1241                         return ERR_PTR(ret);
1242                 }
1243                 goto out;
1244         }
1245
1246         __setup_root(tree_root->nodesize, tree_root->leafsize,
1247                      tree_root->sectorsize, tree_root->stripesize,
1248                      root, fs_info, location->objectid);
1249
1250         path = btrfs_alloc_path();
1251         if (!path) {
1252                 kfree(root);
1253                 return ERR_PTR(-ENOMEM);
1254         }
1255         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1256         if (ret == 0) {
1257                 l = path->nodes[0];
1258                 read_extent_buffer(l, &root->root_item,
1259                                 btrfs_item_ptr_offset(l, path->slots[0]),
1260                                 sizeof(root->root_item));
1261                 memcpy(&root->root_key, location, sizeof(*location));
1262         }
1263         btrfs_free_path(path);
1264         if (ret) {
1265                 kfree(root);
1266                 if (ret > 0)
1267                         ret = -ENOENT;
1268                 return ERR_PTR(ret);
1269         }
1270
1271         generation = btrfs_root_generation(&root->root_item);
1272         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1273         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1274                                      blocksize, generation);
1275         root->commit_root = btrfs_root_node(root);
1276         BUG_ON(!root->node);
1277 out:
1278         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1279                 root->ref_cows = 1;
1280                 btrfs_check_and_init_root_item(&root->root_item);
1281         }
1282
1283         return root;
1284 }
1285
1286 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1287                                         u64 root_objectid)
1288 {
1289         struct btrfs_root *root;
1290
1291         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1292                 return fs_info->tree_root;
1293         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1294                 return fs_info->extent_root;
1295
1296         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1297                                  (unsigned long)root_objectid);
1298         return root;
1299 }
1300
1301 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1302                                               struct btrfs_key *location)
1303 {
1304         struct btrfs_root *root;
1305         int ret;
1306
1307         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1308                 return fs_info->tree_root;
1309         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1310                 return fs_info->extent_root;
1311         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1312                 return fs_info->chunk_root;
1313         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1314                 return fs_info->dev_root;
1315         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1316                 return fs_info->csum_root;
1317 again:
1318         spin_lock(&fs_info->fs_roots_radix_lock);
1319         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1320                                  (unsigned long)location->objectid);
1321         spin_unlock(&fs_info->fs_roots_radix_lock);
1322         if (root)
1323                 return root;
1324
1325         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1326         if (IS_ERR(root))
1327                 return root;
1328
1329         set_anon_super(&root->anon_super, NULL);
1330
1331         if (btrfs_root_refs(&root->root_item) == 0) {
1332                 ret = -ENOENT;
1333                 goto fail;
1334         }
1335
1336         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1337         if (ret < 0)
1338                 goto fail;
1339         if (ret == 0)
1340                 root->orphan_item_inserted = 1;
1341
1342         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1343         if (ret)
1344                 goto fail;
1345
1346         spin_lock(&fs_info->fs_roots_radix_lock);
1347         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1348                                 (unsigned long)root->root_key.objectid,
1349                                 root);
1350         if (ret == 0)
1351                 root->in_radix = 1;
1352
1353         spin_unlock(&fs_info->fs_roots_radix_lock);
1354         radix_tree_preload_end();
1355         if (ret) {
1356                 if (ret == -EEXIST) {
1357                         free_fs_root(root);
1358                         goto again;
1359                 }
1360                 goto fail;
1361         }
1362
1363         ret = btrfs_find_dead_roots(fs_info->tree_root,
1364                                     root->root_key.objectid);
1365         WARN_ON(ret);
1366         return root;
1367 fail:
1368         free_fs_root(root);
1369         return ERR_PTR(ret);
1370 }
1371
1372 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1373                                       struct btrfs_key *location,
1374                                       const char *name, int namelen)
1375 {
1376         return btrfs_read_fs_root_no_name(fs_info, location);
1377 #if 0
1378         struct btrfs_root *root;
1379         int ret;
1380
1381         root = btrfs_read_fs_root_no_name(fs_info, location);
1382         if (!root)
1383                 return NULL;
1384
1385         if (root->in_sysfs)
1386                 return root;
1387
1388         ret = btrfs_set_root_name(root, name, namelen);
1389         if (ret) {
1390                 free_extent_buffer(root->node);
1391                 kfree(root);
1392                 return ERR_PTR(ret);
1393         }
1394
1395         ret = btrfs_sysfs_add_root(root);
1396         if (ret) {
1397                 free_extent_buffer(root->node);
1398                 kfree(root->name);
1399                 kfree(root);
1400                 return ERR_PTR(ret);
1401         }
1402         root->in_sysfs = 1;
1403         return root;
1404 #endif
1405 }
1406
1407 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1408 {
1409         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1410         int ret = 0;
1411         struct btrfs_device *device;
1412         struct backing_dev_info *bdi;
1413
1414         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1415                 if (!device->bdev)
1416                         continue;
1417                 bdi = blk_get_backing_dev_info(device->bdev);
1418                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1419                         ret = 1;
1420                         break;
1421                 }
1422         }
1423         return ret;
1424 }
1425
1426 /*
1427  * If this fails, caller must call bdi_destroy() to get rid of the
1428  * bdi again.
1429  */
1430 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1431 {
1432         int err;
1433
1434         bdi->capabilities = BDI_CAP_MAP_COPY;
1435         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1436         if (err)
1437                 return err;
1438
1439         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1440         bdi->congested_fn       = btrfs_congested_fn;
1441         bdi->congested_data     = info;
1442         return 0;
1443 }
1444
1445 static int bio_ready_for_csum(struct bio *bio)
1446 {
1447         u64 length = 0;
1448         u64 buf_len = 0;
1449         u64 start = 0;
1450         struct page *page;
1451         struct extent_io_tree *io_tree = NULL;
1452         struct bio_vec *bvec;
1453         int i;
1454         int ret;
1455
1456         bio_for_each_segment(bvec, bio, i) {
1457                 page = bvec->bv_page;
1458                 if (page->private == EXTENT_PAGE_PRIVATE) {
1459                         length += bvec->bv_len;
1460                         continue;
1461                 }
1462                 if (!page->private) {
1463                         length += bvec->bv_len;
1464                         continue;
1465                 }
1466                 length = bvec->bv_len;
1467                 buf_len = page->private >> 2;
1468                 start = page_offset(page) + bvec->bv_offset;
1469                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1470         }
1471         /* are we fully contained in this bio? */
1472         if (buf_len <= length)
1473                 return 1;
1474
1475         ret = extent_range_uptodate(io_tree, start + length,
1476                                     start + buf_len - 1);
1477         return ret;
1478 }
1479
1480 /*
1481  * called by the kthread helper functions to finally call the bio end_io
1482  * functions.  This is where read checksum verification actually happens
1483  */
1484 static void end_workqueue_fn(struct btrfs_work *work)
1485 {
1486         struct bio *bio;
1487         struct end_io_wq *end_io_wq;
1488         struct btrfs_fs_info *fs_info;
1489         int error;
1490
1491         end_io_wq = container_of(work, struct end_io_wq, work);
1492         bio = end_io_wq->bio;
1493         fs_info = end_io_wq->info;
1494
1495         /* metadata bio reads are special because the whole tree block must
1496          * be checksummed at once.  This makes sure the entire block is in
1497          * ram and up to date before trying to verify things.  For
1498          * blocksize <= pagesize, it is basically a noop
1499          */
1500         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1501             !bio_ready_for_csum(bio)) {
1502                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1503                                    &end_io_wq->work);
1504                 return;
1505         }
1506         error = end_io_wq->error;
1507         bio->bi_private = end_io_wq->private;
1508         bio->bi_end_io = end_io_wq->end_io;
1509         kfree(end_io_wq);
1510         bio_endio(bio, error);
1511 }
1512
1513 static int cleaner_kthread(void *arg)
1514 {
1515         struct btrfs_root *root = arg;
1516
1517         do {
1518                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1519
1520                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1521                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1522                         btrfs_run_delayed_iputs(root);
1523                         btrfs_clean_old_snapshots(root);
1524                         mutex_unlock(&root->fs_info->cleaner_mutex);
1525                 }
1526
1527                 if (freezing(current)) {
1528                         refrigerator();
1529                 } else {
1530                         set_current_state(TASK_INTERRUPTIBLE);
1531                         if (!kthread_should_stop())
1532                                 schedule();
1533                         __set_current_state(TASK_RUNNING);
1534                 }
1535         } while (!kthread_should_stop());
1536         return 0;
1537 }
1538
1539 static int transaction_kthread(void *arg)
1540 {
1541         struct btrfs_root *root = arg;
1542         struct btrfs_trans_handle *trans;
1543         struct btrfs_transaction *cur;
1544         u64 transid;
1545         unsigned long now;
1546         unsigned long delay;
1547         int ret;
1548
1549         do {
1550                 delay = HZ * 30;
1551                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1552                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1553
1554                 spin_lock(&root->fs_info->new_trans_lock);
1555                 cur = root->fs_info->running_transaction;
1556                 if (!cur) {
1557                         spin_unlock(&root->fs_info->new_trans_lock);
1558                         goto sleep;
1559                 }
1560
1561                 now = get_seconds();
1562                 if (!cur->blocked &&
1563                     (now < cur->start_time || now - cur->start_time < 30)) {
1564                         spin_unlock(&root->fs_info->new_trans_lock);
1565                         delay = HZ * 5;
1566                         goto sleep;
1567                 }
1568                 transid = cur->transid;
1569                 spin_unlock(&root->fs_info->new_trans_lock);
1570
1571                 trans = btrfs_join_transaction(root, 1);
1572                 BUG_ON(IS_ERR(trans));
1573                 if (transid == trans->transid) {
1574                         ret = btrfs_commit_transaction(trans, root);
1575                         BUG_ON(ret);
1576                 } else {
1577                         btrfs_end_transaction(trans, root);
1578                 }
1579 sleep:
1580                 wake_up_process(root->fs_info->cleaner_kthread);
1581                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1582
1583                 if (freezing(current)) {
1584                         refrigerator();
1585                 } else {
1586                         set_current_state(TASK_INTERRUPTIBLE);
1587                         if (!kthread_should_stop() &&
1588                             !btrfs_transaction_blocked(root->fs_info))
1589                                 schedule_timeout(delay);
1590                         __set_current_state(TASK_RUNNING);
1591                 }
1592         } while (!kthread_should_stop());
1593         return 0;
1594 }
1595
1596 struct btrfs_root *open_ctree(struct super_block *sb,
1597                               struct btrfs_fs_devices *fs_devices,
1598                               char *options)
1599 {
1600         u32 sectorsize;
1601         u32 nodesize;
1602         u32 leafsize;
1603         u32 blocksize;
1604         u32 stripesize;
1605         u64 generation;
1606         u64 features;
1607         struct btrfs_key location;
1608         struct buffer_head *bh;
1609         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1610                                                  GFP_NOFS);
1611         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1612                                                  GFP_NOFS);
1613         struct btrfs_root *tree_root = btrfs_sb(sb);
1614         struct btrfs_fs_info *fs_info = NULL;
1615         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1616                                                 GFP_NOFS);
1617         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1618                                               GFP_NOFS);
1619         struct btrfs_root *log_tree_root;
1620
1621         int ret;
1622         int err = -EINVAL;
1623
1624         struct btrfs_super_block *disk_super;
1625
1626         if (!extent_root || !tree_root || !tree_root->fs_info ||
1627             !chunk_root || !dev_root || !csum_root) {
1628                 err = -ENOMEM;
1629                 goto fail;
1630         }
1631         fs_info = tree_root->fs_info;
1632
1633         ret = init_srcu_struct(&fs_info->subvol_srcu);
1634         if (ret) {
1635                 err = ret;
1636                 goto fail;
1637         }
1638
1639         ret = setup_bdi(fs_info, &fs_info->bdi);
1640         if (ret) {
1641                 err = ret;
1642                 goto fail_srcu;
1643         }
1644
1645         fs_info->btree_inode = new_inode(sb);
1646         if (!fs_info->btree_inode) {
1647                 err = -ENOMEM;
1648                 goto fail_bdi;
1649         }
1650
1651         fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1652
1653         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1654         INIT_LIST_HEAD(&fs_info->trans_list);
1655         INIT_LIST_HEAD(&fs_info->dead_roots);
1656         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1657         INIT_LIST_HEAD(&fs_info->hashers);
1658         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1659         INIT_LIST_HEAD(&fs_info->ordered_operations);
1660         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1661         spin_lock_init(&fs_info->delalloc_lock);
1662         spin_lock_init(&fs_info->new_trans_lock);
1663         spin_lock_init(&fs_info->ref_cache_lock);
1664         spin_lock_init(&fs_info->fs_roots_radix_lock);
1665         spin_lock_init(&fs_info->delayed_iput_lock);
1666
1667         init_completion(&fs_info->kobj_unregister);
1668         fs_info->tree_root = tree_root;
1669         fs_info->extent_root = extent_root;
1670         fs_info->csum_root = csum_root;
1671         fs_info->chunk_root = chunk_root;
1672         fs_info->dev_root = dev_root;
1673         fs_info->fs_devices = fs_devices;
1674         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1675         INIT_LIST_HEAD(&fs_info->space_info);
1676         btrfs_mapping_init(&fs_info->mapping_tree);
1677         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1678         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1679         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1680         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1681         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1682         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1683         mutex_init(&fs_info->durable_block_rsv_mutex);
1684         atomic_set(&fs_info->nr_async_submits, 0);
1685         atomic_set(&fs_info->async_delalloc_pages, 0);
1686         atomic_set(&fs_info->async_submit_draining, 0);
1687         atomic_set(&fs_info->nr_async_bios, 0);
1688         fs_info->sb = sb;
1689         fs_info->max_inline = 8192 * 1024;
1690         fs_info->metadata_ratio = 0;
1691
1692         fs_info->thread_pool_size = min_t(unsigned long,
1693                                           num_online_cpus() + 2, 8);
1694
1695         INIT_LIST_HEAD(&fs_info->ordered_extents);
1696         spin_lock_init(&fs_info->ordered_extent_lock);
1697
1698         sb->s_blocksize = 4096;
1699         sb->s_blocksize_bits = blksize_bits(4096);
1700         sb->s_bdi = &fs_info->bdi;
1701
1702         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1703         fs_info->btree_inode->i_nlink = 1;
1704         /*
1705          * we set the i_size on the btree inode to the max possible int.
1706          * the real end of the address space is determined by all of
1707          * the devices in the system
1708          */
1709         fs_info->btree_inode->i_size = OFFSET_MAX;
1710         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1711         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1712
1713         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1714         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1715                              fs_info->btree_inode->i_mapping);
1716         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1717
1718         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1719
1720         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1721         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1722                sizeof(struct btrfs_key));
1723         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1724         insert_inode_hash(fs_info->btree_inode);
1725
1726         spin_lock_init(&fs_info->block_group_cache_lock);
1727         fs_info->block_group_cache_tree = RB_ROOT;
1728
1729         extent_io_tree_init(&fs_info->freed_extents[0],
1730                              fs_info->btree_inode->i_mapping);
1731         extent_io_tree_init(&fs_info->freed_extents[1],
1732                              fs_info->btree_inode->i_mapping);
1733         fs_info->pinned_extents = &fs_info->freed_extents[0];
1734         fs_info->do_barriers = 1;
1735
1736
1737         mutex_init(&fs_info->trans_mutex);
1738         mutex_init(&fs_info->ordered_operations_mutex);
1739         mutex_init(&fs_info->tree_log_mutex);
1740         mutex_init(&fs_info->chunk_mutex);
1741         mutex_init(&fs_info->transaction_kthread_mutex);
1742         mutex_init(&fs_info->cleaner_mutex);
1743         mutex_init(&fs_info->volume_mutex);
1744         init_rwsem(&fs_info->extent_commit_sem);
1745         init_rwsem(&fs_info->cleanup_work_sem);
1746         init_rwsem(&fs_info->subvol_sem);
1747
1748         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1749         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1750
1751         init_waitqueue_head(&fs_info->transaction_throttle);
1752         init_waitqueue_head(&fs_info->transaction_wait);
1753         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1754         init_waitqueue_head(&fs_info->async_submit_wait);
1755
1756         __setup_root(4096, 4096, 4096, 4096, tree_root,
1757                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1758
1759         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1760         if (!bh) {
1761                 err = -EINVAL;
1762                 goto fail_iput;
1763         }
1764
1765         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1766         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1767                sizeof(fs_info->super_for_commit));
1768         brelse(bh);
1769
1770         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1771
1772         disk_super = &fs_info->super_copy;
1773         if (!btrfs_super_root(disk_super))
1774                 goto fail_iput;
1775
1776         /* check FS state, whether FS is broken. */
1777         fs_info->fs_state |= btrfs_super_flags(disk_super);
1778
1779         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1780
1781         /*
1782          * In the long term, we'll store the compression type in the super
1783          * block, and it'll be used for per file compression control.
1784          */
1785         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1786
1787         ret = btrfs_parse_options(tree_root, options);
1788         if (ret) {
1789                 err = ret;
1790                 goto fail_iput;
1791         }
1792
1793         features = btrfs_super_incompat_flags(disk_super) &
1794                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1795         if (features) {
1796                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1797                        "unsupported optional features (%Lx).\n",
1798                        (unsigned long long)features);
1799                 err = -EINVAL;
1800                 goto fail_iput;
1801         }
1802
1803         features = btrfs_super_incompat_flags(disk_super);
1804         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1805         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1806                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1807         btrfs_set_super_incompat_flags(disk_super, features);
1808
1809         features = btrfs_super_compat_ro_flags(disk_super) &
1810                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1811         if (!(sb->s_flags & MS_RDONLY) && features) {
1812                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1813                        "unsupported option features (%Lx).\n",
1814                        (unsigned long long)features);
1815                 err = -EINVAL;
1816                 goto fail_iput;
1817         }
1818
1819         btrfs_init_workers(&fs_info->generic_worker,
1820                            "genwork", 1, NULL);
1821
1822         btrfs_init_workers(&fs_info->workers, "worker",
1823                            fs_info->thread_pool_size,
1824                            &fs_info->generic_worker);
1825
1826         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1827                            fs_info->thread_pool_size,
1828                            &fs_info->generic_worker);
1829
1830         btrfs_init_workers(&fs_info->submit_workers, "submit",
1831                            min_t(u64, fs_devices->num_devices,
1832                            fs_info->thread_pool_size),
1833                            &fs_info->generic_worker);
1834
1835         /* a higher idle thresh on the submit workers makes it much more
1836          * likely that bios will be send down in a sane order to the
1837          * devices
1838          */
1839         fs_info->submit_workers.idle_thresh = 64;
1840
1841         fs_info->workers.idle_thresh = 16;
1842         fs_info->workers.ordered = 1;
1843
1844         fs_info->delalloc_workers.idle_thresh = 2;
1845         fs_info->delalloc_workers.ordered = 1;
1846
1847         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1848                            &fs_info->generic_worker);
1849         btrfs_init_workers(&fs_info->endio_workers, "endio",
1850                            fs_info->thread_pool_size,
1851                            &fs_info->generic_worker);
1852         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1853                            fs_info->thread_pool_size,
1854                            &fs_info->generic_worker);
1855         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1856                            "endio-meta-write", fs_info->thread_pool_size,
1857                            &fs_info->generic_worker);
1858         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1859                            fs_info->thread_pool_size,
1860                            &fs_info->generic_worker);
1861         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1862                            1, &fs_info->generic_worker);
1863
1864         /*
1865          * endios are largely parallel and should have a very
1866          * low idle thresh
1867          */
1868         fs_info->endio_workers.idle_thresh = 4;
1869         fs_info->endio_meta_workers.idle_thresh = 4;
1870
1871         fs_info->endio_write_workers.idle_thresh = 2;
1872         fs_info->endio_meta_write_workers.idle_thresh = 2;
1873
1874         btrfs_start_workers(&fs_info->workers, 1);
1875         btrfs_start_workers(&fs_info->generic_worker, 1);
1876         btrfs_start_workers(&fs_info->submit_workers, 1);
1877         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1878         btrfs_start_workers(&fs_info->fixup_workers, 1);
1879         btrfs_start_workers(&fs_info->endio_workers, 1);
1880         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1881         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1882         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1883         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1884
1885         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1886         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1887                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1888
1889         nodesize = btrfs_super_nodesize(disk_super);
1890         leafsize = btrfs_super_leafsize(disk_super);
1891         sectorsize = btrfs_super_sectorsize(disk_super);
1892         stripesize = btrfs_super_stripesize(disk_super);
1893         tree_root->nodesize = nodesize;
1894         tree_root->leafsize = leafsize;
1895         tree_root->sectorsize = sectorsize;
1896         tree_root->stripesize = stripesize;
1897
1898         sb->s_blocksize = sectorsize;
1899         sb->s_blocksize_bits = blksize_bits(sectorsize);
1900
1901         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1902                     sizeof(disk_super->magic))) {
1903                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1904                 goto fail_sb_buffer;
1905         }
1906
1907         mutex_lock(&fs_info->chunk_mutex);
1908         ret = btrfs_read_sys_array(tree_root);
1909         mutex_unlock(&fs_info->chunk_mutex);
1910         if (ret) {
1911                 printk(KERN_WARNING "btrfs: failed to read the system "
1912                        "array on %s\n", sb->s_id);
1913                 goto fail_sb_buffer;
1914         }
1915
1916         blocksize = btrfs_level_size(tree_root,
1917                                      btrfs_super_chunk_root_level(disk_super));
1918         generation = btrfs_super_chunk_root_generation(disk_super);
1919
1920         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1921                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1922
1923         chunk_root->node = read_tree_block(chunk_root,
1924                                            btrfs_super_chunk_root(disk_super),
1925                                            blocksize, generation);
1926         BUG_ON(!chunk_root->node);
1927         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1928                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1929                        sb->s_id);
1930                 goto fail_chunk_root;
1931         }
1932         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1933         chunk_root->commit_root = btrfs_root_node(chunk_root);
1934
1935         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1936            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1937            BTRFS_UUID_SIZE);
1938
1939         mutex_lock(&fs_info->chunk_mutex);
1940         ret = btrfs_read_chunk_tree(chunk_root);
1941         mutex_unlock(&fs_info->chunk_mutex);
1942         if (ret) {
1943                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1944                        sb->s_id);
1945                 goto fail_chunk_root;
1946         }
1947
1948         btrfs_close_extra_devices(fs_devices);
1949
1950         blocksize = btrfs_level_size(tree_root,
1951                                      btrfs_super_root_level(disk_super));
1952         generation = btrfs_super_generation(disk_super);
1953
1954         tree_root->node = read_tree_block(tree_root,
1955                                           btrfs_super_root(disk_super),
1956                                           blocksize, generation);
1957         if (!tree_root->node)
1958                 goto fail_chunk_root;
1959         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1960                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1961                        sb->s_id);
1962                 goto fail_tree_root;
1963         }
1964         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1965         tree_root->commit_root = btrfs_root_node(tree_root);
1966
1967         ret = find_and_setup_root(tree_root, fs_info,
1968                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1969         if (ret)
1970                 goto fail_tree_root;
1971         extent_root->track_dirty = 1;
1972
1973         ret = find_and_setup_root(tree_root, fs_info,
1974                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1975         if (ret)
1976                 goto fail_extent_root;
1977         dev_root->track_dirty = 1;
1978
1979         ret = find_and_setup_root(tree_root, fs_info,
1980                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1981         if (ret)
1982                 goto fail_dev_root;
1983
1984         csum_root->track_dirty = 1;
1985
1986         fs_info->generation = generation;
1987         fs_info->last_trans_committed = generation;
1988         fs_info->data_alloc_profile = (u64)-1;
1989         fs_info->metadata_alloc_profile = (u64)-1;
1990         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1991
1992         ret = btrfs_init_space_info(fs_info);
1993         if (ret) {
1994                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
1995                 goto fail_block_groups;
1996         }
1997
1998         ret = btrfs_read_block_groups(extent_root);
1999         if (ret) {
2000                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2001                 goto fail_block_groups;
2002         }
2003
2004         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2005                                                "btrfs-cleaner");
2006         if (IS_ERR(fs_info->cleaner_kthread))
2007                 goto fail_block_groups;
2008
2009         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2010                                                    tree_root,
2011                                                    "btrfs-transaction");
2012         if (IS_ERR(fs_info->transaction_kthread))
2013                 goto fail_cleaner;
2014
2015         if (!btrfs_test_opt(tree_root, SSD) &&
2016             !btrfs_test_opt(tree_root, NOSSD) &&
2017             !fs_info->fs_devices->rotating) {
2018                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2019                        "mode\n");
2020                 btrfs_set_opt(fs_info->mount_opt, SSD);
2021         }
2022
2023         /* do not make disk changes in broken FS */
2024         if (btrfs_super_log_root(disk_super) != 0 &&
2025             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2026                 u64 bytenr = btrfs_super_log_root(disk_super);
2027
2028                 if (fs_devices->rw_devices == 0) {
2029                         printk(KERN_WARNING "Btrfs log replay required "
2030                                "on RO media\n");
2031                         err = -EIO;
2032                         goto fail_trans_kthread;
2033                 }
2034                 blocksize =
2035                      btrfs_level_size(tree_root,
2036                                       btrfs_super_log_root_level(disk_super));
2037
2038                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2039                 if (!log_tree_root) {
2040                         err = -ENOMEM;
2041                         goto fail_trans_kthread;
2042                 }
2043
2044                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2045                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2046
2047                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2048                                                       blocksize,
2049                                                       generation + 1);
2050                 ret = btrfs_recover_log_trees(log_tree_root);
2051                 BUG_ON(ret);
2052
2053                 if (sb->s_flags & MS_RDONLY) {
2054                         ret =  btrfs_commit_super(tree_root);
2055                         BUG_ON(ret);
2056                 }
2057         }
2058
2059         ret = btrfs_find_orphan_roots(tree_root);
2060         BUG_ON(ret);
2061
2062         if (!(sb->s_flags & MS_RDONLY)) {
2063                 ret = btrfs_cleanup_fs_roots(fs_info);
2064                 BUG_ON(ret);
2065
2066                 ret = btrfs_recover_relocation(tree_root);
2067                 if (ret < 0) {
2068                         printk(KERN_WARNING
2069                                "btrfs: failed to recover relocation\n");
2070                         err = -EINVAL;
2071                         goto fail_trans_kthread;
2072                 }
2073         }
2074
2075         location.objectid = BTRFS_FS_TREE_OBJECTID;
2076         location.type = BTRFS_ROOT_ITEM_KEY;
2077         location.offset = (u64)-1;
2078
2079         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2080         if (!fs_info->fs_root)
2081                 goto fail_trans_kthread;
2082         if (IS_ERR(fs_info->fs_root)) {
2083                 err = PTR_ERR(fs_info->fs_root);
2084                 goto fail_trans_kthread;
2085         }
2086
2087         if (!(sb->s_flags & MS_RDONLY)) {
2088                 down_read(&fs_info->cleanup_work_sem);
2089                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2090                 if (!err)
2091                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2092                 up_read(&fs_info->cleanup_work_sem);
2093                 if (err) {
2094                         close_ctree(tree_root);
2095                         return ERR_PTR(err);
2096                 }
2097         }
2098
2099         return tree_root;
2100
2101 fail_trans_kthread:
2102         kthread_stop(fs_info->transaction_kthread);
2103 fail_cleaner:
2104         kthread_stop(fs_info->cleaner_kthread);
2105
2106         /*
2107          * make sure we're done with the btree inode before we stop our
2108          * kthreads
2109          */
2110         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2111         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2112
2113 fail_block_groups:
2114         btrfs_free_block_groups(fs_info);
2115         free_extent_buffer(csum_root->node);
2116         free_extent_buffer(csum_root->commit_root);
2117 fail_dev_root:
2118         free_extent_buffer(dev_root->node);
2119         free_extent_buffer(dev_root->commit_root);
2120 fail_extent_root:
2121         free_extent_buffer(extent_root->node);
2122         free_extent_buffer(extent_root->commit_root);
2123 fail_tree_root:
2124         free_extent_buffer(tree_root->node);
2125         free_extent_buffer(tree_root->commit_root);
2126 fail_chunk_root:
2127         free_extent_buffer(chunk_root->node);
2128         free_extent_buffer(chunk_root->commit_root);
2129 fail_sb_buffer:
2130         btrfs_stop_workers(&fs_info->generic_worker);
2131         btrfs_stop_workers(&fs_info->fixup_workers);
2132         btrfs_stop_workers(&fs_info->delalloc_workers);
2133         btrfs_stop_workers(&fs_info->workers);
2134         btrfs_stop_workers(&fs_info->endio_workers);
2135         btrfs_stop_workers(&fs_info->endio_meta_workers);
2136         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2137         btrfs_stop_workers(&fs_info->endio_write_workers);
2138         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2139         btrfs_stop_workers(&fs_info->submit_workers);
2140 fail_iput:
2141         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2142         iput(fs_info->btree_inode);
2143
2144         btrfs_close_devices(fs_info->fs_devices);
2145         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2146 fail_bdi:
2147         bdi_destroy(&fs_info->bdi);
2148 fail_srcu:
2149         cleanup_srcu_struct(&fs_info->subvol_srcu);
2150 fail:
2151         kfree(extent_root);
2152         kfree(tree_root);
2153         kfree(fs_info);
2154         kfree(chunk_root);
2155         kfree(dev_root);
2156         kfree(csum_root);
2157         return ERR_PTR(err);
2158 }
2159
2160 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2161 {
2162         char b[BDEVNAME_SIZE];
2163
2164         if (uptodate) {
2165                 set_buffer_uptodate(bh);
2166         } else {
2167                 if (printk_ratelimit()) {
2168                         printk(KERN_WARNING "lost page write due to "
2169                                         "I/O error on %s\n",
2170                                        bdevname(bh->b_bdev, b));
2171                 }
2172                 /* note, we dont' set_buffer_write_io_error because we have
2173                  * our own ways of dealing with the IO errors
2174                  */
2175                 clear_buffer_uptodate(bh);
2176         }
2177         unlock_buffer(bh);
2178         put_bh(bh);
2179 }
2180
2181 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2182 {
2183         struct buffer_head *bh;
2184         struct buffer_head *latest = NULL;
2185         struct btrfs_super_block *super;
2186         int i;
2187         u64 transid = 0;
2188         u64 bytenr;
2189
2190         /* we would like to check all the supers, but that would make
2191          * a btrfs mount succeed after a mkfs from a different FS.
2192          * So, we need to add a special mount option to scan for
2193          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2194          */
2195         for (i = 0; i < 1; i++) {
2196                 bytenr = btrfs_sb_offset(i);
2197                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2198                         break;
2199                 bh = __bread(bdev, bytenr / 4096, 4096);
2200                 if (!bh)
2201                         continue;
2202
2203                 super = (struct btrfs_super_block *)bh->b_data;
2204                 if (btrfs_super_bytenr(super) != bytenr ||
2205                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2206                             sizeof(super->magic))) {
2207                         brelse(bh);
2208                         continue;
2209                 }
2210
2211                 if (!latest || btrfs_super_generation(super) > transid) {
2212                         brelse(latest);
2213                         latest = bh;
2214                         transid = btrfs_super_generation(super);
2215                 } else {
2216                         brelse(bh);
2217                 }
2218         }
2219         return latest;
2220 }
2221
2222 /*
2223  * this should be called twice, once with wait == 0 and
2224  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2225  * we write are pinned.
2226  *
2227  * They are released when wait == 1 is done.
2228  * max_mirrors must be the same for both runs, and it indicates how
2229  * many supers on this one device should be written.
2230  *
2231  * max_mirrors == 0 means to write them all.
2232  */
2233 static int write_dev_supers(struct btrfs_device *device,
2234                             struct btrfs_super_block *sb,
2235                             int do_barriers, int wait, int max_mirrors)
2236 {
2237         struct buffer_head *bh;
2238         int i;
2239         int ret;
2240         int errors = 0;
2241         u32 crc;
2242         u64 bytenr;
2243         int last_barrier = 0;
2244
2245         if (max_mirrors == 0)
2246                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2247
2248         /* make sure only the last submit_bh does a barrier */
2249         if (do_barriers) {
2250                 for (i = 0; i < max_mirrors; i++) {
2251                         bytenr = btrfs_sb_offset(i);
2252                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2253                             device->total_bytes)
2254                                 break;
2255                         last_barrier = i;
2256                 }
2257         }
2258
2259         for (i = 0; i < max_mirrors; i++) {
2260                 bytenr = btrfs_sb_offset(i);
2261                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2262                         break;
2263
2264                 if (wait) {
2265                         bh = __find_get_block(device->bdev, bytenr / 4096,
2266                                               BTRFS_SUPER_INFO_SIZE);
2267                         BUG_ON(!bh);
2268                         wait_on_buffer(bh);
2269                         if (!buffer_uptodate(bh))
2270                                 errors++;
2271
2272                         /* drop our reference */
2273                         brelse(bh);
2274
2275                         /* drop the reference from the wait == 0 run */
2276                         brelse(bh);
2277                         continue;
2278                 } else {
2279                         btrfs_set_super_bytenr(sb, bytenr);
2280
2281                         crc = ~(u32)0;
2282                         crc = btrfs_csum_data(NULL, (char *)sb +
2283                                               BTRFS_CSUM_SIZE, crc,
2284                                               BTRFS_SUPER_INFO_SIZE -
2285                                               BTRFS_CSUM_SIZE);
2286                         btrfs_csum_final(crc, sb->csum);
2287
2288                         /*
2289                          * one reference for us, and we leave it for the
2290                          * caller
2291                          */
2292                         bh = __getblk(device->bdev, bytenr / 4096,
2293                                       BTRFS_SUPER_INFO_SIZE);
2294                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2295
2296                         /* one reference for submit_bh */
2297                         get_bh(bh);
2298
2299                         set_buffer_uptodate(bh);
2300                         lock_buffer(bh);
2301                         bh->b_end_io = btrfs_end_buffer_write_sync;
2302                 }
2303
2304                 if (i == last_barrier && do_barriers)
2305                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2306                 else
2307                         ret = submit_bh(WRITE_SYNC, bh);
2308
2309                 if (ret)
2310                         errors++;
2311         }
2312         return errors < i ? 0 : -1;
2313 }
2314
2315 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2316 {
2317         struct list_head *head;
2318         struct btrfs_device *dev;
2319         struct btrfs_super_block *sb;
2320         struct btrfs_dev_item *dev_item;
2321         int ret;
2322         int do_barriers;
2323         int max_errors;
2324         int total_errors = 0;
2325         u64 flags;
2326
2327         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2328         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2329
2330         sb = &root->fs_info->super_for_commit;
2331         dev_item = &sb->dev_item;
2332
2333         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2334         head = &root->fs_info->fs_devices->devices;
2335         list_for_each_entry(dev, head, dev_list) {
2336                 if (!dev->bdev) {
2337                         total_errors++;
2338                         continue;
2339                 }
2340                 if (!dev->in_fs_metadata || !dev->writeable)
2341                         continue;
2342
2343                 btrfs_set_stack_device_generation(dev_item, 0);
2344                 btrfs_set_stack_device_type(dev_item, dev->type);
2345                 btrfs_set_stack_device_id(dev_item, dev->devid);
2346                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2347                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2348                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2349                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2350                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2351                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2352                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2353
2354                 flags = btrfs_super_flags(sb);
2355                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2356
2357                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2358                 if (ret)
2359                         total_errors++;
2360         }
2361         if (total_errors > max_errors) {
2362                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2363                        total_errors);
2364                 BUG();
2365         }
2366
2367         total_errors = 0;
2368         list_for_each_entry(dev, head, dev_list) {
2369                 if (!dev->bdev)
2370                         continue;
2371                 if (!dev->in_fs_metadata || !dev->writeable)
2372                         continue;
2373
2374                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2375                 if (ret)
2376                         total_errors++;
2377         }
2378         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2379         if (total_errors > max_errors) {
2380                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2381                        total_errors);
2382                 BUG();
2383         }
2384         return 0;
2385 }
2386
2387 int write_ctree_super(struct btrfs_trans_handle *trans,
2388                       struct btrfs_root *root, int max_mirrors)
2389 {
2390         int ret;
2391
2392         ret = write_all_supers(root, max_mirrors);
2393         return ret;
2394 }
2395
2396 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2397 {
2398         spin_lock(&fs_info->fs_roots_radix_lock);
2399         radix_tree_delete(&fs_info->fs_roots_radix,
2400                           (unsigned long)root->root_key.objectid);
2401         spin_unlock(&fs_info->fs_roots_radix_lock);
2402
2403         if (btrfs_root_refs(&root->root_item) == 0)
2404                 synchronize_srcu(&fs_info->subvol_srcu);
2405
2406         free_fs_root(root);
2407         return 0;
2408 }
2409
2410 static void free_fs_root(struct btrfs_root *root)
2411 {
2412         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2413         if (root->anon_super.s_dev) {
2414                 down_write(&root->anon_super.s_umount);
2415                 kill_anon_super(&root->anon_super);
2416         }
2417         free_extent_buffer(root->node);
2418         free_extent_buffer(root->commit_root);
2419         kfree(root->name);
2420         kfree(root);
2421 }
2422
2423 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2424 {
2425         int ret;
2426         struct btrfs_root *gang[8];
2427         int i;
2428
2429         while (!list_empty(&fs_info->dead_roots)) {
2430                 gang[0] = list_entry(fs_info->dead_roots.next,
2431                                      struct btrfs_root, root_list);
2432                 list_del(&gang[0]->root_list);
2433
2434                 if (gang[0]->in_radix) {
2435                         btrfs_free_fs_root(fs_info, gang[0]);
2436                 } else {
2437                         free_extent_buffer(gang[0]->node);
2438                         free_extent_buffer(gang[0]->commit_root);
2439                         kfree(gang[0]);
2440                 }
2441         }
2442
2443         while (1) {
2444                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2445                                              (void **)gang, 0,
2446                                              ARRAY_SIZE(gang));
2447                 if (!ret)
2448                         break;
2449                 for (i = 0; i < ret; i++)
2450                         btrfs_free_fs_root(fs_info, gang[i]);
2451         }
2452         return 0;
2453 }
2454
2455 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2456 {
2457         u64 root_objectid = 0;
2458         struct btrfs_root *gang[8];
2459         int i;
2460         int ret;
2461
2462         while (1) {
2463                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2464                                              (void **)gang, root_objectid,
2465                                              ARRAY_SIZE(gang));
2466                 if (!ret)
2467                         break;
2468
2469                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2470                 for (i = 0; i < ret; i++) {
2471                         int err;
2472
2473                         root_objectid = gang[i]->root_key.objectid;
2474                         err = btrfs_orphan_cleanup(gang[i]);
2475                         if (err)
2476                                 return err;
2477                 }
2478                 root_objectid++;
2479         }
2480         return 0;
2481 }
2482
2483 int btrfs_commit_super(struct btrfs_root *root)
2484 {
2485         struct btrfs_trans_handle *trans;
2486         int ret;
2487
2488         mutex_lock(&root->fs_info->cleaner_mutex);
2489         btrfs_run_delayed_iputs(root);
2490         btrfs_clean_old_snapshots(root);
2491         mutex_unlock(&root->fs_info->cleaner_mutex);
2492
2493         /* wait until ongoing cleanup work done */
2494         down_write(&root->fs_info->cleanup_work_sem);
2495         up_write(&root->fs_info->cleanup_work_sem);
2496
2497         trans = btrfs_join_transaction(root, 1);
2498         if (IS_ERR(trans))
2499                 return PTR_ERR(trans);
2500         ret = btrfs_commit_transaction(trans, root);
2501         BUG_ON(ret);
2502         /* run commit again to drop the original snapshot */
2503         trans = btrfs_join_transaction(root, 1);
2504         if (IS_ERR(trans))
2505                 return PTR_ERR(trans);
2506         btrfs_commit_transaction(trans, root);
2507         ret = btrfs_write_and_wait_transaction(NULL, root);
2508         BUG_ON(ret);
2509
2510         ret = write_ctree_super(NULL, root, 0);
2511         return ret;
2512 }
2513
2514 int close_ctree(struct btrfs_root *root)
2515 {
2516         struct btrfs_fs_info *fs_info = root->fs_info;
2517         int ret;
2518
2519         fs_info->closing = 1;
2520         smp_mb();
2521
2522         btrfs_put_block_group_cache(fs_info);
2523
2524         /*
2525          * Here come 2 situations when btrfs is broken to flip readonly:
2526          *
2527          * 1. when btrfs flips readonly somewhere else before
2528          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2529          * and btrfs will skip to write sb directly to keep
2530          * ERROR state on disk.
2531          *
2532          * 2. when btrfs flips readonly just in btrfs_commit_super,
2533          * and in such case, btrfs cannot write sb via btrfs_commit_super,
2534          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2535          * btrfs will cleanup all FS resources first and write sb then.
2536          */
2537         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2538                 ret = btrfs_commit_super(root);
2539                 if (ret)
2540                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2541         }
2542
2543         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2544                 ret = btrfs_error_commit_super(root);
2545                 if (ret)
2546                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2547         }
2548
2549         kthread_stop(root->fs_info->transaction_kthread);
2550         kthread_stop(root->fs_info->cleaner_kthread);
2551
2552         fs_info->closing = 2;
2553         smp_mb();
2554
2555         if (fs_info->delalloc_bytes) {
2556                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2557                        (unsigned long long)fs_info->delalloc_bytes);
2558         }
2559         if (fs_info->total_ref_cache_size) {
2560                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2561                        (unsigned long long)fs_info->total_ref_cache_size);
2562         }
2563
2564         free_extent_buffer(fs_info->extent_root->node);
2565         free_extent_buffer(fs_info->extent_root->commit_root);
2566         free_extent_buffer(fs_info->tree_root->node);
2567         free_extent_buffer(fs_info->tree_root->commit_root);
2568         free_extent_buffer(root->fs_info->chunk_root->node);
2569         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2570         free_extent_buffer(root->fs_info->dev_root->node);
2571         free_extent_buffer(root->fs_info->dev_root->commit_root);
2572         free_extent_buffer(root->fs_info->csum_root->node);
2573         free_extent_buffer(root->fs_info->csum_root->commit_root);
2574
2575         btrfs_free_block_groups(root->fs_info);
2576
2577         del_fs_roots(fs_info);
2578
2579         iput(fs_info->btree_inode);
2580
2581         btrfs_stop_workers(&fs_info->generic_worker);
2582         btrfs_stop_workers(&fs_info->fixup_workers);
2583         btrfs_stop_workers(&fs_info->delalloc_workers);
2584         btrfs_stop_workers(&fs_info->workers);
2585         btrfs_stop_workers(&fs_info->endio_workers);
2586         btrfs_stop_workers(&fs_info->endio_meta_workers);
2587         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2588         btrfs_stop_workers(&fs_info->endio_write_workers);
2589         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2590         btrfs_stop_workers(&fs_info->submit_workers);
2591
2592         btrfs_close_devices(fs_info->fs_devices);
2593         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2594
2595         bdi_destroy(&fs_info->bdi);
2596         cleanup_srcu_struct(&fs_info->subvol_srcu);
2597
2598         kfree(fs_info->extent_root);
2599         kfree(fs_info->tree_root);
2600         kfree(fs_info->chunk_root);
2601         kfree(fs_info->dev_root);
2602         kfree(fs_info->csum_root);
2603         kfree(fs_info);
2604
2605         return 0;
2606 }
2607
2608 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2609 {
2610         int ret;
2611         struct inode *btree_inode = buf->first_page->mapping->host;
2612
2613         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2614                                      NULL);
2615         if (!ret)
2616                 return ret;
2617
2618         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2619                                     parent_transid);
2620         return !ret;
2621 }
2622
2623 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2624 {
2625         struct inode *btree_inode = buf->first_page->mapping->host;
2626         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2627                                           buf);
2628 }
2629
2630 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2631 {
2632         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2633         u64 transid = btrfs_header_generation(buf);
2634         struct inode *btree_inode = root->fs_info->btree_inode;
2635         int was_dirty;
2636
2637         btrfs_assert_tree_locked(buf);
2638         if (transid != root->fs_info->generation) {
2639                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2640                        "found %llu running %llu\n",
2641                         (unsigned long long)buf->start,
2642                         (unsigned long long)transid,
2643                         (unsigned long long)root->fs_info->generation);
2644                 WARN_ON(1);
2645         }
2646         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2647                                             buf);
2648         if (!was_dirty) {
2649                 spin_lock(&root->fs_info->delalloc_lock);
2650                 root->fs_info->dirty_metadata_bytes += buf->len;
2651                 spin_unlock(&root->fs_info->delalloc_lock);
2652         }
2653 }
2654
2655 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2656 {
2657         /*
2658          * looks as though older kernels can get into trouble with
2659          * this code, they end up stuck in balance_dirty_pages forever
2660          */
2661         u64 num_dirty;
2662         unsigned long thresh = 32 * 1024 * 1024;
2663
2664         if (current->flags & PF_MEMALLOC)
2665                 return;
2666
2667         num_dirty = root->fs_info->dirty_metadata_bytes;
2668
2669         if (num_dirty > thresh) {
2670                 balance_dirty_pages_ratelimited_nr(
2671                                    root->fs_info->btree_inode->i_mapping, 1);
2672         }
2673         return;
2674 }
2675
2676 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2677 {
2678         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2679         int ret;
2680         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2681         if (ret == 0)
2682                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2683         return ret;
2684 }
2685
2686 int btree_lock_page_hook(struct page *page)
2687 {
2688         struct inode *inode = page->mapping->host;
2689         struct btrfs_root *root = BTRFS_I(inode)->root;
2690         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2691         struct extent_buffer *eb;
2692         unsigned long len;
2693         u64 bytenr = page_offset(page);
2694
2695         if (page->private == EXTENT_PAGE_PRIVATE)
2696                 goto out;
2697
2698         len = page->private >> 2;
2699         eb = find_extent_buffer(io_tree, bytenr, len);
2700         if (!eb)
2701                 goto out;
2702
2703         btrfs_tree_lock(eb);
2704         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2705
2706         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2707                 spin_lock(&root->fs_info->delalloc_lock);
2708                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2709                         root->fs_info->dirty_metadata_bytes -= eb->len;
2710                 else
2711                         WARN_ON(1);
2712                 spin_unlock(&root->fs_info->delalloc_lock);
2713         }
2714
2715         btrfs_tree_unlock(eb);
2716         free_extent_buffer(eb);
2717 out:
2718         lock_page(page);
2719         return 0;
2720 }
2721
2722 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2723                               int read_only)
2724 {
2725         if (read_only)
2726                 return;
2727
2728         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2729                 printk(KERN_WARNING "warning: mount fs with errors, "
2730                        "running btrfsck is recommended\n");
2731 }
2732
2733 int btrfs_error_commit_super(struct btrfs_root *root)
2734 {
2735         int ret;
2736
2737         mutex_lock(&root->fs_info->cleaner_mutex);
2738         btrfs_run_delayed_iputs(root);
2739         mutex_unlock(&root->fs_info->cleaner_mutex);
2740
2741         down_write(&root->fs_info->cleanup_work_sem);
2742         up_write(&root->fs_info->cleanup_work_sem);
2743
2744         /* cleanup FS via transaction */
2745         btrfs_cleanup_transaction(root);
2746
2747         ret = write_ctree_super(NULL, root, 0);
2748
2749         return ret;
2750 }
2751
2752 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2753 {
2754         struct btrfs_inode *btrfs_inode;
2755         struct list_head splice;
2756
2757         INIT_LIST_HEAD(&splice);
2758
2759         mutex_lock(&root->fs_info->ordered_operations_mutex);
2760         spin_lock(&root->fs_info->ordered_extent_lock);
2761
2762         list_splice_init(&root->fs_info->ordered_operations, &splice);
2763         while (!list_empty(&splice)) {
2764                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2765                                          ordered_operations);
2766
2767                 list_del_init(&btrfs_inode->ordered_operations);
2768
2769                 btrfs_invalidate_inodes(btrfs_inode->root);
2770         }
2771
2772         spin_unlock(&root->fs_info->ordered_extent_lock);
2773         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2774
2775         return 0;
2776 }
2777
2778 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2779 {
2780         struct list_head splice;
2781         struct btrfs_ordered_extent *ordered;
2782         struct inode *inode;
2783
2784         INIT_LIST_HEAD(&splice);
2785
2786         spin_lock(&root->fs_info->ordered_extent_lock);
2787
2788         list_splice_init(&root->fs_info->ordered_extents, &splice);
2789         while (!list_empty(&splice)) {
2790                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2791                                      root_extent_list);
2792
2793                 list_del_init(&ordered->root_extent_list);
2794                 atomic_inc(&ordered->refs);
2795
2796                 /* the inode may be getting freed (in sys_unlink path). */
2797                 inode = igrab(ordered->inode);
2798
2799                 spin_unlock(&root->fs_info->ordered_extent_lock);
2800                 if (inode)
2801                         iput(inode);
2802
2803                 atomic_set(&ordered->refs, 1);
2804                 btrfs_put_ordered_extent(ordered);
2805
2806                 spin_lock(&root->fs_info->ordered_extent_lock);
2807         }
2808
2809         spin_unlock(&root->fs_info->ordered_extent_lock);
2810
2811         return 0;
2812 }
2813
2814 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2815                                       struct btrfs_root *root)
2816 {
2817         struct rb_node *node;
2818         struct btrfs_delayed_ref_root *delayed_refs;
2819         struct btrfs_delayed_ref_node *ref;
2820         int ret = 0;
2821
2822         delayed_refs = &trans->delayed_refs;
2823
2824         spin_lock(&delayed_refs->lock);
2825         if (delayed_refs->num_entries == 0) {
2826                 spin_unlock(&delayed_refs->lock);
2827                 printk(KERN_INFO "delayed_refs has NO entry\n");
2828                 return ret;
2829         }
2830
2831         node = rb_first(&delayed_refs->root);
2832         while (node) {
2833                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2834                 node = rb_next(node);
2835
2836                 ref->in_tree = 0;
2837                 rb_erase(&ref->rb_node, &delayed_refs->root);
2838                 delayed_refs->num_entries--;
2839
2840                 atomic_set(&ref->refs, 1);
2841                 if (btrfs_delayed_ref_is_head(ref)) {
2842                         struct btrfs_delayed_ref_head *head;
2843
2844                         head = btrfs_delayed_node_to_head(ref);
2845                         mutex_lock(&head->mutex);
2846                         kfree(head->extent_op);
2847                         delayed_refs->num_heads--;
2848                         if (list_empty(&head->cluster))
2849                                 delayed_refs->num_heads_ready--;
2850                         list_del_init(&head->cluster);
2851                         mutex_unlock(&head->mutex);
2852                 }
2853
2854                 spin_unlock(&delayed_refs->lock);
2855                 btrfs_put_delayed_ref(ref);
2856
2857                 cond_resched();
2858                 spin_lock(&delayed_refs->lock);
2859         }
2860
2861         spin_unlock(&delayed_refs->lock);
2862
2863         return ret;
2864 }
2865
2866 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2867 {
2868         struct btrfs_pending_snapshot *snapshot;
2869         struct list_head splice;
2870
2871         INIT_LIST_HEAD(&splice);
2872
2873         list_splice_init(&t->pending_snapshots, &splice);
2874
2875         while (!list_empty(&splice)) {
2876                 snapshot = list_entry(splice.next,
2877                                       struct btrfs_pending_snapshot,
2878                                       list);
2879
2880                 list_del_init(&snapshot->list);
2881
2882                 kfree(snapshot);
2883         }
2884
2885         return 0;
2886 }
2887
2888 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2889 {
2890         struct btrfs_inode *btrfs_inode;
2891         struct list_head splice;
2892
2893         INIT_LIST_HEAD(&splice);
2894
2895         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2896
2897         spin_lock(&root->fs_info->delalloc_lock);
2898
2899         while (!list_empty(&splice)) {
2900                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2901                                     delalloc_inodes);
2902
2903                 list_del_init(&btrfs_inode->delalloc_inodes);
2904
2905                 btrfs_invalidate_inodes(btrfs_inode->root);
2906         }
2907
2908         spin_unlock(&root->fs_info->delalloc_lock);
2909
2910         return 0;
2911 }
2912
2913 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2914                                         struct extent_io_tree *dirty_pages,
2915                                         int mark)
2916 {
2917         int ret;
2918         struct page *page;
2919         struct inode *btree_inode = root->fs_info->btree_inode;
2920         struct extent_buffer *eb;
2921         u64 start = 0;
2922         u64 end;
2923         u64 offset;
2924         unsigned long index;
2925
2926         while (1) {
2927                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2928                                             mark);
2929                 if (ret)
2930                         break;
2931
2932                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2933                 while (start <= end) {
2934                         index = start >> PAGE_CACHE_SHIFT;
2935                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2936                         page = find_get_page(btree_inode->i_mapping, index);
2937                         if (!page)
2938                                 continue;
2939                         offset = page_offset(page);
2940
2941                         spin_lock(&dirty_pages->buffer_lock);
2942                         eb = radix_tree_lookup(
2943                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2944                                                offset >> PAGE_CACHE_SHIFT);
2945                         spin_unlock(&dirty_pages->buffer_lock);
2946                         if (eb) {
2947                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2948                                                          &eb->bflags);
2949                                 atomic_set(&eb->refs, 1);
2950                         }
2951                         if (PageWriteback(page))
2952                                 end_page_writeback(page);
2953
2954                         lock_page(page);
2955                         if (PageDirty(page)) {
2956                                 clear_page_dirty_for_io(page);
2957                                 spin_lock_irq(&page->mapping->tree_lock);
2958                                 radix_tree_tag_clear(&page->mapping->page_tree,
2959                                                         page_index(page),
2960                                                         PAGECACHE_TAG_DIRTY);
2961                                 spin_unlock_irq(&page->mapping->tree_lock);
2962                         }
2963
2964                         page->mapping->a_ops->invalidatepage(page, 0);
2965                         unlock_page(page);
2966                 }
2967         }
2968
2969         return ret;
2970 }
2971
2972 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
2973                                        struct extent_io_tree *pinned_extents)
2974 {
2975         struct extent_io_tree *unpin;
2976         u64 start;
2977         u64 end;
2978         int ret;
2979
2980         unpin = pinned_extents;
2981         while (1) {
2982                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2983                                             EXTENT_DIRTY);
2984                 if (ret)
2985                         break;
2986
2987                 /* opt_discard */
2988                 if (btrfs_test_opt(root, DISCARD))
2989                         ret = btrfs_error_discard_extent(root, start,
2990                                                          end + 1 - start,
2991                                                          NULL);
2992
2993                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2994                 btrfs_error_unpin_extent_range(root, start, end);
2995                 cond_resched();
2996         }
2997
2998         return 0;
2999 }
3000
3001 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3002 {
3003         struct btrfs_transaction *t;
3004         LIST_HEAD(list);
3005
3006         WARN_ON(1);
3007
3008         mutex_lock(&root->fs_info->trans_mutex);
3009         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3010
3011         list_splice_init(&root->fs_info->trans_list, &list);
3012         while (!list_empty(&list)) {
3013                 t = list_entry(list.next, struct btrfs_transaction, list);
3014                 if (!t)
3015                         break;
3016
3017                 btrfs_destroy_ordered_operations(root);
3018
3019                 btrfs_destroy_ordered_extents(root);
3020
3021                 btrfs_destroy_delayed_refs(t, root);
3022
3023                 btrfs_block_rsv_release(root,
3024                                         &root->fs_info->trans_block_rsv,
3025                                         t->dirty_pages.dirty_bytes);
3026
3027                 /* FIXME: cleanup wait for commit */
3028                 t->in_commit = 1;
3029                 t->blocked = 1;
3030                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3031                         wake_up(&root->fs_info->transaction_blocked_wait);
3032
3033                 t->blocked = 0;
3034                 if (waitqueue_active(&root->fs_info->transaction_wait))
3035                         wake_up(&root->fs_info->transaction_wait);
3036                 mutex_unlock(&root->fs_info->trans_mutex);
3037
3038                 mutex_lock(&root->fs_info->trans_mutex);
3039                 t->commit_done = 1;
3040                 if (waitqueue_active(&t->commit_wait))
3041                         wake_up(&t->commit_wait);
3042                 mutex_unlock(&root->fs_info->trans_mutex);
3043
3044                 mutex_lock(&root->fs_info->trans_mutex);
3045
3046                 btrfs_destroy_pending_snapshots(t);
3047
3048                 btrfs_destroy_delalloc_inodes(root);
3049
3050                 spin_lock(&root->fs_info->new_trans_lock);
3051                 root->fs_info->running_transaction = NULL;
3052                 spin_unlock(&root->fs_info->new_trans_lock);
3053
3054                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3055                                              EXTENT_DIRTY);
3056
3057                 btrfs_destroy_pinned_extent(root,
3058                                             root->fs_info->pinned_extents);
3059
3060                 atomic_set(&t->use_count, 0);
3061                 list_del_init(&t->list);
3062                 memset(t, 0, sizeof(*t));
3063                 kmem_cache_free(btrfs_transaction_cachep, t);
3064         }
3065
3066         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3067         mutex_unlock(&root->fs_info->trans_mutex);
3068
3069         return 0;
3070 }
3071
3072 static struct extent_io_ops btree_extent_io_ops = {
3073         .write_cache_pages_lock_hook = btree_lock_page_hook,
3074         .readpage_end_io_hook = btree_readpage_end_io_hook,
3075         .submit_bio_hook = btree_submit_bio_hook,
3076         /* note we're sharing with inode.c for the merge bio hook */
3077         .merge_bio_hook = btrfs_merge_bio_hook,
3078 };