b9a53646ceb2e8dd90a1f8b7dee205a68c36de6f
[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> // for block_sync_page
26 #include <linux/workqueue.h>
27 #include "crc32c.h"
28 #include "ctree.h"
29 #include "disk-io.h"
30 #include "transaction.h"
31 #include "btrfs_inode.h"
32 #include "volumes.h"
33 #include "print-tree.h"
34
35 #if 0
36 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
37 {
38         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
39                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
40                        (unsigned long long)extent_buffer_blocknr(buf),
41                        (unsigned long long)btrfs_header_blocknr(buf));
42                 return 1;
43         }
44         return 0;
45 }
46 #endif
47
48 static struct extent_io_ops btree_extent_io_ops;
49 static struct workqueue_struct *end_io_workqueue;
50 static struct workqueue_struct *async_submit_workqueue;
51
52 struct end_io_wq {
53         struct bio *bio;
54         bio_end_io_t *end_io;
55         void *private;
56         struct btrfs_fs_info *info;
57         int error;
58         int metadata;
59         struct list_head list;
60 };
61
62 struct async_submit_bio {
63         struct inode *inode;
64         struct bio *bio;
65         struct list_head list;
66         extent_submit_bio_hook_t *submit_bio_hook;
67         int rw;
68         int mirror_num;
69 };
70
71 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
72                                     size_t page_offset, u64 start, u64 len,
73                                     int create)
74 {
75         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
76         struct extent_map *em;
77         int ret;
78
79         spin_lock(&em_tree->lock);
80         em = lookup_extent_mapping(em_tree, start, len);
81         if (em) {
82                 em->bdev =
83                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
84                 spin_unlock(&em_tree->lock);
85                 goto out;
86         }
87         spin_unlock(&em_tree->lock);
88
89         em = alloc_extent_map(GFP_NOFS);
90         if (!em) {
91                 em = ERR_PTR(-ENOMEM);
92                 goto out;
93         }
94         em->start = 0;
95         em->len = (u64)-1;
96         em->block_start = 0;
97         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
98
99         spin_lock(&em_tree->lock);
100         ret = add_extent_mapping(em_tree, em);
101         if (ret == -EEXIST) {
102                 u64 failed_start = em->start;
103                 u64 failed_len = em->len;
104
105                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
106                        em->start, em->len, em->block_start);
107                 free_extent_map(em);
108                 em = lookup_extent_mapping(em_tree, start, len);
109                 if (em) {
110                         printk("after failing, found %Lu %Lu %Lu\n",
111                                em->start, em->len, em->block_start);
112                         ret = 0;
113                 } else {
114                         em = lookup_extent_mapping(em_tree, failed_start,
115                                                    failed_len);
116                         if (em) {
117                                 printk("double failure lookup gives us "
118                                        "%Lu %Lu -> %Lu\n", em->start,
119                                        em->len, em->block_start);
120                                 free_extent_map(em);
121                         }
122                         ret = -EIO;
123                 }
124         } else if (ret) {
125                 free_extent_map(em);
126                 em = NULL;
127         }
128         spin_unlock(&em_tree->lock);
129
130         if (ret)
131                 em = ERR_PTR(ret);
132 out:
133         return em;
134 }
135
136 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
137 {
138         return btrfs_crc32c(seed, data, len);
139 }
140
141 void btrfs_csum_final(u32 crc, char *result)
142 {
143         *(__le32 *)result = ~cpu_to_le32(crc);
144 }
145
146 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
147                            int verify)
148 {
149         char result[BTRFS_CRC32_SIZE];
150         unsigned long len;
151         unsigned long cur_len;
152         unsigned long offset = BTRFS_CSUM_SIZE;
153         char *map_token = NULL;
154         char *kaddr;
155         unsigned long map_start;
156         unsigned long map_len;
157         int err;
158         u32 crc = ~(u32)0;
159
160         len = buf->len - offset;
161         while(len > 0) {
162                 err = map_private_extent_buffer(buf, offset, 32,
163                                         &map_token, &kaddr,
164                                         &map_start, &map_len, KM_USER0);
165                 if (err) {
166                         printk("failed to map extent buffer! %lu\n",
167                                offset);
168                         return 1;
169                 }
170                 cur_len = min(len, map_len - (offset - map_start));
171                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
172                                       crc, cur_len);
173                 len -= cur_len;
174                 offset += cur_len;
175                 unmap_extent_buffer(buf, map_token, KM_USER0);
176         }
177         btrfs_csum_final(crc, result);
178
179         if (verify) {
180                 int from_this_trans = 0;
181
182                 if (root->fs_info->running_transaction &&
183                     btrfs_header_generation(buf) ==
184                     root->fs_info->running_transaction->transid)
185                         from_this_trans = 1;
186
187                 /* FIXME, this is not good */
188                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
189                         u32 val;
190                         u32 found = 0;
191                         memcpy(&found, result, BTRFS_CRC32_SIZE);
192
193                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
194                         printk("btrfs: %s checksum verify failed on %llu "
195                                "wanted %X found %X from_this_trans %d "
196                                "level %d\n",
197                                root->fs_info->sb->s_id,
198                                buf->start, val, found, from_this_trans,
199                                btrfs_header_level(buf));
200                         return 1;
201                 }
202         } else {
203                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
204         }
205         return 0;
206 }
207
208 static int verify_parent_transid(struct extent_io_tree *io_tree,
209                                  struct extent_buffer *eb, u64 parent_transid)
210 {
211         int ret;
212
213         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
214                 return 0;
215
216         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
217         if (extent_buffer_uptodate(io_tree, eb) &&
218             btrfs_header_generation(eb) == parent_transid) {
219                 ret = 0;
220                 goto out;
221         }
222         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
223                (unsigned long long)eb->start,
224                (unsigned long long)parent_transid,
225                (unsigned long long)btrfs_header_generation(eb));
226         ret = 1;
227 out:
228         clear_extent_buffer_uptodate(io_tree, eb);
229         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
230                       GFP_NOFS);
231         return ret;
232
233 }
234
235 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
236                                           struct extent_buffer *eb,
237                                           u64 start, u64 parent_transid)
238 {
239         struct extent_io_tree *io_tree;
240         int ret;
241         int num_copies = 0;
242         int mirror_num = 0;
243
244         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
245         while (1) {
246                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
247                                                btree_get_extent, mirror_num);
248                 if (!ret &&
249                     !verify_parent_transid(io_tree, eb, parent_transid))
250                         return ret;
251
252                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
253                                               eb->start, eb->len);
254                 if (num_copies == 1)
255                         return ret;
256
257                 mirror_num++;
258                 if (mirror_num > num_copies)
259                         return ret;
260         }
261         return -EIO;
262 }
263
264 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
265 {
266         struct extent_io_tree *tree;
267         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
268         u64 found_start;
269         int found_level;
270         unsigned long len;
271         struct extent_buffer *eb;
272         int ret;
273
274         tree = &BTRFS_I(page->mapping->host)->io_tree;
275
276         if (page->private == EXTENT_PAGE_PRIVATE)
277                 goto out;
278         if (!page->private)
279                 goto out;
280         len = page->private >> 2;
281         if (len == 0) {
282                 WARN_ON(1);
283         }
284         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
285         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
286                                              btrfs_header_generation(eb));
287         BUG_ON(ret);
288         btrfs_clear_buffer_defrag(eb);
289         found_start = btrfs_header_bytenr(eb);
290         if (found_start != start) {
291                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
292                        start, found_start, len);
293                 WARN_ON(1);
294                 goto err;
295         }
296         if (eb->first_page != page) {
297                 printk("bad first page %lu %lu\n", eb->first_page->index,
298                        page->index);
299                 WARN_ON(1);
300                 goto err;
301         }
302         if (!PageUptodate(page)) {
303                 printk("csum not up to date page %lu\n", page->index);
304                 WARN_ON(1);
305                 goto err;
306         }
307         found_level = btrfs_header_level(eb);
308         spin_lock(&root->fs_info->hash_lock);
309         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
310         spin_unlock(&root->fs_info->hash_lock);
311         csum_tree_block(root, eb, 0);
312 err:
313         free_extent_buffer(eb);
314 out:
315         return 0;
316 }
317
318 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
319 {
320         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
321
322         csum_dirty_buffer(root, page);
323         return 0;
324 }
325
326 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
327                                struct extent_state *state)
328 {
329         struct extent_io_tree *tree;
330         u64 found_start;
331         int found_level;
332         unsigned long len;
333         struct extent_buffer *eb;
334         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
335         int ret = 0;
336
337         tree = &BTRFS_I(page->mapping->host)->io_tree;
338         if (page->private == EXTENT_PAGE_PRIVATE)
339                 goto out;
340         if (!page->private)
341                 goto out;
342         len = page->private >> 2;
343         if (len == 0) {
344                 WARN_ON(1);
345         }
346         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
347
348         btrfs_clear_buffer_defrag(eb);
349         found_start = btrfs_header_bytenr(eb);
350         if (found_start != start) {
351                 ret = -EIO;
352                 goto err;
353         }
354         if (eb->first_page != page) {
355                 printk("bad first page %lu %lu\n", eb->first_page->index,
356                        page->index);
357                 WARN_ON(1);
358                 ret = -EIO;
359                 goto err;
360         }
361         if (memcmp_extent_buffer(eb, root->fs_info->fsid,
362                                  (unsigned long)btrfs_header_fsid(eb),
363                                  BTRFS_FSID_SIZE)) {
364                 printk("bad fsid on block %Lu\n", eb->start);
365                 ret = -EIO;
366                 goto err;
367         }
368         found_level = btrfs_header_level(eb);
369
370         ret = csum_tree_block(root, eb, 1);
371         if (ret)
372                 ret = -EIO;
373
374         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
375         end = eb->start + end - 1;
376         release_extent_buffer_tail_pages(eb);
377 err:
378         free_extent_buffer(eb);
379 out:
380         return ret;
381 }
382
383 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
384 static void end_workqueue_bio(struct bio *bio, int err)
385 #else
386 static int end_workqueue_bio(struct bio *bio,
387                                    unsigned int bytes_done, int err)
388 #endif
389 {
390         struct end_io_wq *end_io_wq = bio->bi_private;
391         struct btrfs_fs_info *fs_info;
392         unsigned long flags;
393
394 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
395         if (bio->bi_size)
396                 return 1;
397 #endif
398
399         fs_info = end_io_wq->info;
400         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
401         end_io_wq->error = err;
402         list_add_tail(&end_io_wq->list, &fs_info->end_io_work_list);
403         spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
404         queue_work(end_io_workqueue, &fs_info->end_io_work);
405
406 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
407         return 0;
408 #endif
409 }
410
411 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
412                         int metadata)
413 {
414         struct end_io_wq *end_io_wq;
415         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
416         if (!end_io_wq)
417                 return -ENOMEM;
418
419         end_io_wq->private = bio->bi_private;
420         end_io_wq->end_io = bio->bi_end_io;
421         end_io_wq->info = info;
422         end_io_wq->error = 0;
423         end_io_wq->bio = bio;
424         end_io_wq->metadata = metadata;
425
426         bio->bi_private = end_io_wq;
427         bio->bi_end_io = end_workqueue_bio;
428         return 0;
429 }
430
431 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
432                         int rw, struct bio *bio, int mirror_num,
433                         extent_submit_bio_hook_t *submit_bio_hook)
434 {
435         struct async_submit_bio *async;
436
437         async = kmalloc(sizeof(*async), GFP_NOFS);
438         if (!async)
439                 return -ENOMEM;
440
441         async->inode = inode;
442         async->rw = rw;
443         async->bio = bio;
444         async->mirror_num = mirror_num;
445         async->submit_bio_hook = submit_bio_hook;
446
447         spin_lock(&fs_info->async_submit_work_lock);
448         list_add_tail(&async->list, &fs_info->async_submit_work_list);
449         atomic_inc(&fs_info->nr_async_submits);
450         spin_unlock(&fs_info->async_submit_work_lock);
451
452         queue_work(async_submit_workqueue, &fs_info->async_submit_work);
453         return 0;
454 }
455
456 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
457                                  int mirror_num)
458 {
459         struct btrfs_root *root = BTRFS_I(inode)->root;
460         u64 offset;
461         int ret;
462
463         offset = bio->bi_sector << 9;
464
465         if (rw & (1 << BIO_RW)) {
466                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
467         }
468
469         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
470         BUG_ON(ret);
471
472         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
473 }
474
475 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
476                                  int mirror_num)
477 {
478         if (!(rw & (1 << BIO_RW))) {
479                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
480         }
481         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
482                                    inode, rw, bio, mirror_num,
483                                    __btree_submit_bio_hook);
484 }
485
486 static int btree_writepage(struct page *page, struct writeback_control *wbc)
487 {
488         struct extent_io_tree *tree;
489         tree = &BTRFS_I(page->mapping->host)->io_tree;
490         return extent_write_full_page(tree, page, btree_get_extent, wbc);
491 }
492
493 static int btree_writepages(struct address_space *mapping,
494                             struct writeback_control *wbc)
495 {
496         struct extent_io_tree *tree;
497         tree = &BTRFS_I(mapping->host)->io_tree;
498         if (wbc->sync_mode == WB_SYNC_NONE) {
499                 u64 num_dirty;
500                 u64 start = 0;
501                 unsigned long thresh = 96 * 1024 * 1024;
502
503                 if (wbc->for_kupdate)
504                         return 0;
505
506                 if (current_is_pdflush()) {
507                         thresh = 96 * 1024 * 1024;
508                 } else {
509                         thresh = 8 * 1024 * 1024;
510                 }
511                 num_dirty = count_range_bits(tree, &start, (u64)-1,
512                                              thresh, EXTENT_DIRTY);
513                 if (num_dirty < thresh) {
514                         return 0;
515                 }
516         }
517         return extent_writepages(tree, mapping, btree_get_extent, wbc);
518 }
519
520 int btree_readpage(struct file *file, struct page *page)
521 {
522         struct extent_io_tree *tree;
523         tree = &BTRFS_I(page->mapping->host)->io_tree;
524         return extent_read_full_page(tree, page, btree_get_extent);
525 }
526
527 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
528 {
529         struct extent_io_tree *tree;
530         struct extent_map_tree *map;
531         int ret;
532
533         if (page_count(page) > 3) {
534                 /* once for page->private, once for the caller, once
535                  * once for the page cache
536                  */
537                 return 0;
538         }
539         tree = &BTRFS_I(page->mapping->host)->io_tree;
540         map = &BTRFS_I(page->mapping->host)->extent_tree;
541         ret = try_release_extent_state(map, tree, page, gfp_flags);
542         if (ret == 1) {
543                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
544                 ClearPagePrivate(page);
545                 set_page_private(page, 0);
546                 page_cache_release(page);
547         }
548         return ret;
549 }
550
551 static void btree_invalidatepage(struct page *page, unsigned long offset)
552 {
553         struct extent_io_tree *tree;
554         tree = &BTRFS_I(page->mapping->host)->io_tree;
555         extent_invalidatepage(tree, page, offset);
556         btree_releasepage(page, GFP_NOFS);
557         if (PagePrivate(page)) {
558                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
559                 ClearPagePrivate(page);
560                 set_page_private(page, 0);
561                 page_cache_release(page);
562         }
563 }
564
565 #if 0
566 static int btree_writepage(struct page *page, struct writeback_control *wbc)
567 {
568         struct buffer_head *bh;
569         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
570         struct buffer_head *head;
571         if (!page_has_buffers(page)) {
572                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
573                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
574         }
575         head = page_buffers(page);
576         bh = head;
577         do {
578                 if (buffer_dirty(bh))
579                         csum_tree_block(root, bh, 0);
580                 bh = bh->b_this_page;
581         } while (bh != head);
582         return block_write_full_page(page, btree_get_block, wbc);
583 }
584 #endif
585
586 static struct address_space_operations btree_aops = {
587         .readpage       = btree_readpage,
588         .writepage      = btree_writepage,
589         .writepages     = btree_writepages,
590         .releasepage    = btree_releasepage,
591         .invalidatepage = btree_invalidatepage,
592         .sync_page      = block_sync_page,
593 };
594
595 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
596                          u64 parent_transid)
597 {
598         struct extent_buffer *buf = NULL;
599         struct inode *btree_inode = root->fs_info->btree_inode;
600         int ret = 0;
601
602         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
603         if (!buf)
604                 return 0;
605         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
606                                  buf, 0, 0, btree_get_extent, 0);
607         free_extent_buffer(buf);
608         return ret;
609 }
610
611 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
612                                             u64 bytenr, u32 blocksize)
613 {
614         struct inode *btree_inode = root->fs_info->btree_inode;
615         struct extent_buffer *eb;
616         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
617                                 bytenr, blocksize, GFP_NOFS);
618         return eb;
619 }
620
621 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
622                                                  u64 bytenr, u32 blocksize)
623 {
624         struct inode *btree_inode = root->fs_info->btree_inode;
625         struct extent_buffer *eb;
626
627         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
628                                  bytenr, blocksize, NULL, GFP_NOFS);
629         return eb;
630 }
631
632
633 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
634                                       u32 blocksize, u64 parent_transid)
635 {
636         struct extent_buffer *buf = NULL;
637         struct inode *btree_inode = root->fs_info->btree_inode;
638         struct extent_io_tree *io_tree;
639         int ret;
640
641         io_tree = &BTRFS_I(btree_inode)->io_tree;
642
643         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
644         if (!buf)
645                 return NULL;
646
647         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
648
649         if (ret == 0) {
650                 buf->flags |= EXTENT_UPTODATE;
651         }
652         return buf;
653
654 }
655
656 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
657                      struct extent_buffer *buf)
658 {
659         struct inode *btree_inode = root->fs_info->btree_inode;
660         if (btrfs_header_generation(buf) ==
661             root->fs_info->running_transaction->transid)
662                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
663                                           buf);
664         return 0;
665 }
666
667 int wait_on_tree_block_writeback(struct btrfs_root *root,
668                                  struct extent_buffer *buf)
669 {
670         struct inode *btree_inode = root->fs_info->btree_inode;
671         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
672                                         buf);
673         return 0;
674 }
675
676 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
677                         u32 stripesize, struct btrfs_root *root,
678                         struct btrfs_fs_info *fs_info,
679                         u64 objectid)
680 {
681         root->node = NULL;
682         root->inode = NULL;
683         root->commit_root = NULL;
684         root->sectorsize = sectorsize;
685         root->nodesize = nodesize;
686         root->leafsize = leafsize;
687         root->stripesize = stripesize;
688         root->ref_cows = 0;
689         root->track_dirty = 0;
690
691         root->fs_info = fs_info;
692         root->objectid = objectid;
693         root->last_trans = 0;
694         root->highest_inode = 0;
695         root->last_inode_alloc = 0;
696         root->name = NULL;
697         root->in_sysfs = 0;
698
699         INIT_LIST_HEAD(&root->dirty_list);
700         memset(&root->root_key, 0, sizeof(root->root_key));
701         memset(&root->root_item, 0, sizeof(root->root_item));
702         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
703         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
704         init_completion(&root->kobj_unregister);
705         root->defrag_running = 0;
706         root->defrag_level = 0;
707         root->root_key.objectid = objectid;
708         return 0;
709 }
710
711 static int find_and_setup_root(struct btrfs_root *tree_root,
712                                struct btrfs_fs_info *fs_info,
713                                u64 objectid,
714                                struct btrfs_root *root)
715 {
716         int ret;
717         u32 blocksize;
718
719         __setup_root(tree_root->nodesize, tree_root->leafsize,
720                      tree_root->sectorsize, tree_root->stripesize,
721                      root, fs_info, objectid);
722         ret = btrfs_find_last_root(tree_root, objectid,
723                                    &root->root_item, &root->root_key);
724         BUG_ON(ret);
725
726         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
727         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
728                                      blocksize, 0);
729         BUG_ON(!root->node);
730         return 0;
731 }
732
733 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
734                                                struct btrfs_key *location)
735 {
736         struct btrfs_root *root;
737         struct btrfs_root *tree_root = fs_info->tree_root;
738         struct btrfs_path *path;
739         struct extent_buffer *l;
740         u64 highest_inode;
741         u32 blocksize;
742         int ret = 0;
743
744         root = kzalloc(sizeof(*root), GFP_NOFS);
745         if (!root)
746                 return ERR_PTR(-ENOMEM);
747         if (location->offset == (u64)-1) {
748                 ret = find_and_setup_root(tree_root, fs_info,
749                                           location->objectid, root);
750                 if (ret) {
751                         kfree(root);
752                         return ERR_PTR(ret);
753                 }
754                 goto insert;
755         }
756
757         __setup_root(tree_root->nodesize, tree_root->leafsize,
758                      tree_root->sectorsize, tree_root->stripesize,
759                      root, fs_info, location->objectid);
760
761         path = btrfs_alloc_path();
762         BUG_ON(!path);
763         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
764         if (ret != 0) {
765                 if (ret > 0)
766                         ret = -ENOENT;
767                 goto out;
768         }
769         l = path->nodes[0];
770         read_extent_buffer(l, &root->root_item,
771                btrfs_item_ptr_offset(l, path->slots[0]),
772                sizeof(root->root_item));
773         memcpy(&root->root_key, location, sizeof(*location));
774         ret = 0;
775 out:
776         btrfs_release_path(root, path);
777         btrfs_free_path(path);
778         if (ret) {
779                 kfree(root);
780                 return ERR_PTR(ret);
781         }
782         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
783         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
784                                      blocksize, 0);
785         BUG_ON(!root->node);
786 insert:
787         root->ref_cows = 1;
788         ret = btrfs_find_highest_inode(root, &highest_inode);
789         if (ret == 0) {
790                 root->highest_inode = highest_inode;
791                 root->last_inode_alloc = highest_inode;
792         }
793         return root;
794 }
795
796 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
797                                         u64 root_objectid)
798 {
799         struct btrfs_root *root;
800
801         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
802                 return fs_info->tree_root;
803         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
804                 return fs_info->extent_root;
805
806         root = radix_tree_lookup(&fs_info->fs_roots_radix,
807                                  (unsigned long)root_objectid);
808         return root;
809 }
810
811 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
812                                               struct btrfs_key *location)
813 {
814         struct btrfs_root *root;
815         int ret;
816
817         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
818                 return fs_info->tree_root;
819         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
820                 return fs_info->extent_root;
821         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
822                 return fs_info->chunk_root;
823         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
824                 return fs_info->dev_root;
825
826         root = radix_tree_lookup(&fs_info->fs_roots_radix,
827                                  (unsigned long)location->objectid);
828         if (root)
829                 return root;
830
831         root = btrfs_read_fs_root_no_radix(fs_info, location);
832         if (IS_ERR(root))
833                 return root;
834         ret = radix_tree_insert(&fs_info->fs_roots_radix,
835                                 (unsigned long)root->root_key.objectid,
836                                 root);
837         if (ret) {
838                 free_extent_buffer(root->node);
839                 kfree(root);
840                 return ERR_PTR(ret);
841         }
842         ret = btrfs_find_dead_roots(fs_info->tree_root,
843                                     root->root_key.objectid, root);
844         BUG_ON(ret);
845
846         return root;
847 }
848
849 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
850                                       struct btrfs_key *location,
851                                       const char *name, int namelen)
852 {
853         struct btrfs_root *root;
854         int ret;
855
856         root = btrfs_read_fs_root_no_name(fs_info, location);
857         if (!root)
858                 return NULL;
859
860         if (root->in_sysfs)
861                 return root;
862
863         ret = btrfs_set_root_name(root, name, namelen);
864         if (ret) {
865                 free_extent_buffer(root->node);
866                 kfree(root);
867                 return ERR_PTR(ret);
868         }
869
870         ret = btrfs_sysfs_add_root(root);
871         if (ret) {
872                 free_extent_buffer(root->node);
873                 kfree(root->name);
874                 kfree(root);
875                 return ERR_PTR(ret);
876         }
877         root->in_sysfs = 1;
878         return root;
879 }
880 #if 0
881 static int add_hasher(struct btrfs_fs_info *info, char *type) {
882         struct btrfs_hasher *hasher;
883
884         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
885         if (!hasher)
886                 return -ENOMEM;
887         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
888         if (!hasher->hash_tfm) {
889                 kfree(hasher);
890                 return -EINVAL;
891         }
892         spin_lock(&info->hash_lock);
893         list_add(&hasher->list, &info->hashers);
894         spin_unlock(&info->hash_lock);
895         return 0;
896 }
897 #endif
898
899 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
900 {
901         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
902         int ret = 0;
903         int limit = 256 * info->fs_devices->open_devices;
904         struct list_head *cur;
905         struct btrfs_device *device;
906         struct backing_dev_info *bdi;
907
908         if ((bdi_bits & (1 << BDI_write_congested)) &&
909             atomic_read(&info->nr_async_submits) > limit) {
910                 return 1;
911         }
912
913         list_for_each(cur, &info->fs_devices->devices) {
914                 device = list_entry(cur, struct btrfs_device, dev_list);
915                 if (!device->bdev)
916                         continue;
917                 bdi = blk_get_backing_dev_info(device->bdev);
918                 if (bdi && bdi_congested(bdi, bdi_bits)) {
919                         ret = 1;
920                         break;
921                 }
922         }
923         return ret;
924 }
925
926 /*
927  * this unplugs every device on the box, and it is only used when page
928  * is null
929  */
930 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
931 {
932         struct list_head *cur;
933         struct btrfs_device *device;
934         struct btrfs_fs_info *info;
935
936         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
937         list_for_each(cur, &info->fs_devices->devices) {
938                 device = list_entry(cur, struct btrfs_device, dev_list);
939                 bdi = blk_get_backing_dev_info(device->bdev);
940                 if (bdi->unplug_io_fn) {
941                         bdi->unplug_io_fn(bdi, page);
942                 }
943         }
944 }
945
946 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
947 {
948         struct inode *inode;
949         struct extent_map_tree *em_tree;
950         struct extent_map *em;
951         struct address_space *mapping;
952         u64 offset;
953
954         /* the generic O_DIRECT read code does this */
955         if (!page) {
956                 __unplug_io_fn(bdi, page);
957                 return;
958         }
959
960         /*
961          * page->mapping may change at any time.  Get a consistent copy
962          * and use that for everything below
963          */
964         smp_mb();
965         mapping = page->mapping;
966         if (!mapping)
967                 return;
968
969         inode = mapping->host;
970         offset = page_offset(page);
971
972         em_tree = &BTRFS_I(inode)->extent_tree;
973         spin_lock(&em_tree->lock);
974         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
975         spin_unlock(&em_tree->lock);
976         if (!em)
977                 return;
978
979         offset = offset - em->start;
980         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
981                           em->block_start + offset, page);
982         free_extent_map(em);
983 }
984
985 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
986 {
987 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
988         bdi_init(bdi);
989 #endif
990         bdi->ra_pages   = default_backing_dev_info.ra_pages;
991         bdi->state              = 0;
992         bdi->capabilities       = default_backing_dev_info.capabilities;
993         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
994         bdi->unplug_io_data     = info;
995         bdi->congested_fn       = btrfs_congested_fn;
996         bdi->congested_data     = info;
997         return 0;
998 }
999
1000 static int bio_ready_for_csum(struct bio *bio)
1001 {
1002         u64 length = 0;
1003         u64 buf_len = 0;
1004         u64 start = 0;
1005         struct page *page;
1006         struct extent_io_tree *io_tree = NULL;
1007         struct btrfs_fs_info *info = NULL;
1008         struct bio_vec *bvec;
1009         int i;
1010         int ret;
1011
1012         bio_for_each_segment(bvec, bio, i) {
1013                 page = bvec->bv_page;
1014                 if (page->private == EXTENT_PAGE_PRIVATE) {
1015                         length += bvec->bv_len;
1016                         continue;
1017                 }
1018                 if (!page->private) {
1019                         length += bvec->bv_len;
1020                         continue;
1021                 }
1022                 length = bvec->bv_len;
1023                 buf_len = page->private >> 2;
1024                 start = page_offset(page) + bvec->bv_offset;
1025                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1026                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1027         }
1028         /* are we fully contained in this bio? */
1029         if (buf_len <= length)
1030                 return 1;
1031
1032         ret = extent_range_uptodate(io_tree, start + length,
1033                                     start + buf_len - 1);
1034         if (ret == 1)
1035                 return ret;
1036         return ret;
1037 }
1038
1039 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1040 static void btrfs_end_io_csum(void *p)
1041 #else
1042 static void btrfs_end_io_csum(struct work_struct *work)
1043 #endif
1044 {
1045 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1046         struct btrfs_fs_info *fs_info = p;
1047 #else
1048         struct btrfs_fs_info *fs_info = container_of(work,
1049                                                      struct btrfs_fs_info,
1050                                                      end_io_work);
1051 #endif
1052         unsigned long flags;
1053         struct end_io_wq *end_io_wq;
1054         struct bio *bio;
1055         struct list_head *next;
1056         int error;
1057         int was_empty;
1058
1059         while(1) {
1060                 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1061                 if (list_empty(&fs_info->end_io_work_list)) {
1062                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1063                                                flags);
1064                         return;
1065                 }
1066                 next = fs_info->end_io_work_list.next;
1067                 list_del(next);
1068                 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
1069
1070                 end_io_wq = list_entry(next, struct end_io_wq, list);
1071
1072                 bio = end_io_wq->bio;
1073                 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1074                         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1075                         was_empty = list_empty(&fs_info->end_io_work_list);
1076                         list_add_tail(&end_io_wq->list,
1077                                       &fs_info->end_io_work_list);
1078                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1079                                                flags);
1080                         if (was_empty)
1081                                 return;
1082                         continue;
1083                 }
1084                 error = end_io_wq->error;
1085                 bio->bi_private = end_io_wq->private;
1086                 bio->bi_end_io = end_io_wq->end_io;
1087                 kfree(end_io_wq);
1088 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1089                 bio_endio(bio, bio->bi_size, error);
1090 #else
1091                 bio_endio(bio, error);
1092 #endif
1093         }
1094 }
1095
1096 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1097 static void btrfs_async_submit_work(void *p)
1098 #else
1099 static void btrfs_async_submit_work(struct work_struct *work)
1100 #endif
1101 {
1102 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1103         struct btrfs_fs_info *fs_info = p;
1104 #else
1105         struct btrfs_fs_info *fs_info = container_of(work,
1106                                                      struct btrfs_fs_info,
1107                                                      async_submit_work);
1108 #endif
1109         struct async_submit_bio *async;
1110         struct list_head *next;
1111
1112         while(1) {
1113                 spin_lock(&fs_info->async_submit_work_lock);
1114                 if (list_empty(&fs_info->async_submit_work_list)) {
1115                         spin_unlock(&fs_info->async_submit_work_lock);
1116                         return;
1117                 }
1118                 next = fs_info->async_submit_work_list.next;
1119                 list_del(next);
1120                 atomic_dec(&fs_info->nr_async_submits);
1121                 spin_unlock(&fs_info->async_submit_work_lock);
1122
1123                 async = list_entry(next, struct async_submit_bio, list);
1124                 async->submit_bio_hook(async->inode, async->rw, async->bio,
1125                                        async->mirror_num);
1126                 kfree(async);
1127         }
1128 }
1129
1130 struct btrfs_root *open_ctree(struct super_block *sb,
1131                               struct btrfs_fs_devices *fs_devices,
1132                               char *options)
1133 {
1134         u32 sectorsize;
1135         u32 nodesize;
1136         u32 leafsize;
1137         u32 blocksize;
1138         u32 stripesize;
1139         struct buffer_head *bh;
1140         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1141                                                  GFP_NOFS);
1142         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1143                                                GFP_NOFS);
1144         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1145                                                 GFP_NOFS);
1146         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1147                                                 GFP_NOFS);
1148         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1149                                               GFP_NOFS);
1150         int ret;
1151         int err = -EINVAL;
1152         struct btrfs_super_block *disk_super;
1153
1154         if (!extent_root || !tree_root || !fs_info) {
1155                 err = -ENOMEM;
1156                 goto fail;
1157         }
1158         end_io_workqueue = create_workqueue("btrfs-end-io");
1159         BUG_ON(!end_io_workqueue);
1160         async_submit_workqueue = create_workqueue("btrfs-async-submit");
1161
1162         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1163         INIT_LIST_HEAD(&fs_info->trans_list);
1164         INIT_LIST_HEAD(&fs_info->dead_roots);
1165         INIT_LIST_HEAD(&fs_info->hashers);
1166         INIT_LIST_HEAD(&fs_info->end_io_work_list);
1167         INIT_LIST_HEAD(&fs_info->async_submit_work_list);
1168         spin_lock_init(&fs_info->hash_lock);
1169         spin_lock_init(&fs_info->end_io_work_lock);
1170         spin_lock_init(&fs_info->async_submit_work_lock);
1171         spin_lock_init(&fs_info->delalloc_lock);
1172         spin_lock_init(&fs_info->new_trans_lock);
1173
1174         init_completion(&fs_info->kobj_unregister);
1175         fs_info->tree_root = tree_root;
1176         fs_info->extent_root = extent_root;
1177         fs_info->chunk_root = chunk_root;
1178         fs_info->dev_root = dev_root;
1179         fs_info->fs_devices = fs_devices;
1180         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1181         INIT_LIST_HEAD(&fs_info->space_info);
1182         btrfs_mapping_init(&fs_info->mapping_tree);
1183         atomic_set(&fs_info->nr_async_submits, 0);
1184         fs_info->sb = sb;
1185         fs_info->max_extent = (u64)-1;
1186         fs_info->max_inline = 8192 * 1024;
1187         setup_bdi(fs_info, &fs_info->bdi);
1188         fs_info->btree_inode = new_inode(sb);
1189         fs_info->btree_inode->i_ino = 1;
1190         fs_info->btree_inode->i_nlink = 1;
1191
1192         sb->s_blocksize = 4096;
1193         sb->s_blocksize_bits = blksize_bits(4096);
1194
1195         /*
1196          * we set the i_size on the btree inode to the max possible int.
1197          * the real end of the address space is determined by all of
1198          * the devices in the system
1199          */
1200         fs_info->btree_inode->i_size = OFFSET_MAX;
1201         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1202         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1203
1204         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1205                              fs_info->btree_inode->i_mapping,
1206                              GFP_NOFS);
1207         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1208                              GFP_NOFS);
1209
1210         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1211
1212         extent_io_tree_init(&fs_info->free_space_cache,
1213                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1214         extent_io_tree_init(&fs_info->block_group_cache,
1215                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1216         extent_io_tree_init(&fs_info->pinned_extents,
1217                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1218         extent_io_tree_init(&fs_info->pending_del,
1219                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1220         extent_io_tree_init(&fs_info->extent_ins,
1221                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1222         fs_info->do_barriers = 1;
1223
1224 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1225         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum, fs_info);
1226         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work,
1227                   fs_info);
1228         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1229 #else
1230         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
1231         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work);
1232         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1233 #endif
1234         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1235         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1236                sizeof(struct btrfs_key));
1237         insert_inode_hash(fs_info->btree_inode);
1238         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1239
1240         mutex_init(&fs_info->trans_mutex);
1241         mutex_init(&fs_info->fs_mutex);
1242
1243 #if 0
1244         ret = add_hasher(fs_info, "crc32c");
1245         if (ret) {
1246                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1247                 err = -ENOMEM;
1248                 goto fail_iput;
1249         }
1250 #endif
1251         __setup_root(4096, 4096, 4096, 4096, tree_root,
1252                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1253
1254
1255         bh = __bread(fs_devices->latest_bdev,
1256                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1257         if (!bh)
1258                 goto fail_iput;
1259
1260         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1261         brelse(bh);
1262
1263         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1264
1265         disk_super = &fs_info->super_copy;
1266         if (!btrfs_super_root(disk_super))
1267                 goto fail_sb_buffer;
1268
1269         err = btrfs_parse_options(tree_root, options);
1270         if (err)
1271                 goto fail_sb_buffer;
1272
1273         err = -EINVAL;
1274         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1275                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1276                        (unsigned long long)btrfs_super_num_devices(disk_super),
1277                        (unsigned long long)fs_devices->open_devices);
1278                 if (btrfs_test_opt(tree_root, DEGRADED))
1279                         printk("continuing in degraded mode\n");
1280                 else {
1281                         goto fail_sb_buffer;
1282                 }
1283         }
1284
1285         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1286
1287         nodesize = btrfs_super_nodesize(disk_super);
1288         leafsize = btrfs_super_leafsize(disk_super);
1289         sectorsize = btrfs_super_sectorsize(disk_super);
1290         stripesize = btrfs_super_stripesize(disk_super);
1291         tree_root->nodesize = nodesize;
1292         tree_root->leafsize = leafsize;
1293         tree_root->sectorsize = sectorsize;
1294         tree_root->stripesize = stripesize;
1295
1296         sb->s_blocksize = sectorsize;
1297         sb->s_blocksize_bits = blksize_bits(sectorsize);
1298
1299         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1300                     sizeof(disk_super->magic))) {
1301                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1302                 goto fail_sb_buffer;
1303         }
1304
1305         mutex_lock(&fs_info->fs_mutex);
1306
1307         ret = btrfs_read_sys_array(tree_root);
1308         if (ret) {
1309                 printk("btrfs: failed to read the system array on %s\n",
1310                        sb->s_id);
1311                 goto fail_sys_array;
1312         }
1313
1314         blocksize = btrfs_level_size(tree_root,
1315                                      btrfs_super_chunk_root_level(disk_super));
1316
1317         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1318                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1319
1320         chunk_root->node = read_tree_block(chunk_root,
1321                                            btrfs_super_chunk_root(disk_super),
1322                                            blocksize, 0);
1323         BUG_ON(!chunk_root->node);
1324
1325         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1326                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1327                  BTRFS_UUID_SIZE);
1328
1329         ret = btrfs_read_chunk_tree(chunk_root);
1330         BUG_ON(ret);
1331
1332         btrfs_close_extra_devices(fs_devices);
1333
1334         blocksize = btrfs_level_size(tree_root,
1335                                      btrfs_super_root_level(disk_super));
1336
1337
1338         tree_root->node = read_tree_block(tree_root,
1339                                           btrfs_super_root(disk_super),
1340                                           blocksize, 0);
1341         if (!tree_root->node)
1342                 goto fail_sb_buffer;
1343
1344
1345         ret = find_and_setup_root(tree_root, fs_info,
1346                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1347         if (ret)
1348                 goto fail_tree_root;
1349         extent_root->track_dirty = 1;
1350
1351         ret = find_and_setup_root(tree_root, fs_info,
1352                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1353         dev_root->track_dirty = 1;
1354
1355         if (ret)
1356                 goto fail_extent_root;
1357
1358         btrfs_read_block_groups(extent_root);
1359
1360         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1361         fs_info->data_alloc_profile = (u64)-1;
1362         fs_info->metadata_alloc_profile = (u64)-1;
1363         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1364
1365         mutex_unlock(&fs_info->fs_mutex);
1366         return tree_root;
1367
1368 fail_extent_root:
1369         free_extent_buffer(extent_root->node);
1370 fail_tree_root:
1371         free_extent_buffer(tree_root->node);
1372 fail_sys_array:
1373         mutex_unlock(&fs_info->fs_mutex);
1374 fail_sb_buffer:
1375         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1376 fail_iput:
1377         iput(fs_info->btree_inode);
1378 fail:
1379         btrfs_close_devices(fs_info->fs_devices);
1380         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1381
1382         kfree(extent_root);
1383         kfree(tree_root);
1384 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1385         bdi_destroy(&fs_info->bdi);
1386 #endif
1387         kfree(fs_info);
1388         return ERR_PTR(err);
1389 }
1390
1391 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1392 {
1393         char b[BDEVNAME_SIZE];
1394
1395         if (uptodate) {
1396                 set_buffer_uptodate(bh);
1397         } else {
1398                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1399                         printk(KERN_WARNING "lost page write due to "
1400                                         "I/O error on %s\n",
1401                                        bdevname(bh->b_bdev, b));
1402                 }
1403                 /* note, we dont' set_buffer_write_io_error because we have
1404                  * our own ways of dealing with the IO errors
1405                  */
1406                 clear_buffer_uptodate(bh);
1407         }
1408         unlock_buffer(bh);
1409         put_bh(bh);
1410 }
1411
1412 int write_all_supers(struct btrfs_root *root)
1413 {
1414         struct list_head *cur;
1415         struct list_head *head = &root->fs_info->fs_devices->devices;
1416         struct btrfs_device *dev;
1417         struct btrfs_super_block *sb;
1418         struct btrfs_dev_item *dev_item;
1419         struct buffer_head *bh;
1420         int ret;
1421         int do_barriers;
1422         int max_errors;
1423         int total_errors = 0;
1424         u32 crc;
1425         u64 flags;
1426
1427         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1428         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1429
1430         sb = &root->fs_info->super_for_commit;
1431         dev_item = &sb->dev_item;
1432         list_for_each(cur, head) {
1433                 dev = list_entry(cur, struct btrfs_device, dev_list);
1434                 if (!dev->bdev) {
1435                         total_errors++;
1436                         continue;
1437                 }
1438                 if (!dev->in_fs_metadata)
1439                         continue;
1440
1441                 btrfs_set_stack_device_type(dev_item, dev->type);
1442                 btrfs_set_stack_device_id(dev_item, dev->devid);
1443                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1444                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1445                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1446                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1447                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1448                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1449                 flags = btrfs_super_flags(sb);
1450                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1451
1452
1453                 crc = ~(u32)0;
1454                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1455                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1456                 btrfs_csum_final(crc, sb->csum);
1457
1458                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1459                               BTRFS_SUPER_INFO_SIZE);
1460
1461                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1462                 dev->pending_io = bh;
1463
1464                 get_bh(bh);
1465                 set_buffer_uptodate(bh);
1466                 lock_buffer(bh);
1467                 bh->b_end_io = btrfs_end_buffer_write_sync;
1468
1469                 if (do_barriers && dev->barriers) {
1470                         ret = submit_bh(WRITE_BARRIER, bh);
1471                         if (ret == -EOPNOTSUPP) {
1472                                 printk("btrfs: disabling barriers on dev %s\n",
1473                                        dev->name);
1474                                 set_buffer_uptodate(bh);
1475                                 dev->barriers = 0;
1476                                 get_bh(bh);
1477                                 lock_buffer(bh);
1478                                 ret = submit_bh(WRITE, bh);
1479                         }
1480                 } else {
1481                         ret = submit_bh(WRITE, bh);
1482                 }
1483                 if (ret)
1484                         total_errors++;
1485         }
1486         if (total_errors > max_errors) {
1487                 printk("btrfs: %d errors while writing supers\n", total_errors);
1488                 BUG();
1489         }
1490         total_errors = 0;
1491
1492         list_for_each(cur, head) {
1493                 dev = list_entry(cur, struct btrfs_device, dev_list);
1494                 if (!dev->bdev)
1495                         continue;
1496                 if (!dev->in_fs_metadata)
1497                         continue;
1498
1499                 BUG_ON(!dev->pending_io);
1500                 bh = dev->pending_io;
1501                 wait_on_buffer(bh);
1502                 if (!buffer_uptodate(dev->pending_io)) {
1503                         if (do_barriers && dev->barriers) {
1504                                 printk("btrfs: disabling barriers on dev %s\n",
1505                                        dev->name);
1506                                 set_buffer_uptodate(bh);
1507                                 get_bh(bh);
1508                                 lock_buffer(bh);
1509                                 dev->barriers = 0;
1510                                 ret = submit_bh(WRITE, bh);
1511                                 BUG_ON(ret);
1512                                 wait_on_buffer(bh);
1513                                 if (!buffer_uptodate(bh))
1514                                         total_errors++;
1515                         } else {
1516                                 total_errors++;
1517                         }
1518
1519                 }
1520                 dev->pending_io = NULL;
1521                 brelse(bh);
1522         }
1523         if (total_errors > max_errors) {
1524                 printk("btrfs: %d errors while writing supers\n", total_errors);
1525                 BUG();
1526         }
1527         return 0;
1528 }
1529
1530 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1531                       *root)
1532 {
1533         int ret;
1534
1535         ret = write_all_supers(root);
1536         return ret;
1537 }
1538
1539 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1540 {
1541         radix_tree_delete(&fs_info->fs_roots_radix,
1542                           (unsigned long)root->root_key.objectid);
1543         if (root->in_sysfs)
1544                 btrfs_sysfs_del_root(root);
1545         if (root->inode)
1546                 iput(root->inode);
1547         if (root->node)
1548                 free_extent_buffer(root->node);
1549         if (root->commit_root)
1550                 free_extent_buffer(root->commit_root);
1551         if (root->name)
1552                 kfree(root->name);
1553         kfree(root);
1554         return 0;
1555 }
1556
1557 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1558 {
1559         int ret;
1560         struct btrfs_root *gang[8];
1561         int i;
1562
1563         while(1) {
1564                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1565                                              (void **)gang, 0,
1566                                              ARRAY_SIZE(gang));
1567                 if (!ret)
1568                         break;
1569                 for (i = 0; i < ret; i++)
1570                         btrfs_free_fs_root(fs_info, gang[i]);
1571         }
1572         return 0;
1573 }
1574
1575 int close_ctree(struct btrfs_root *root)
1576 {
1577         int ret;
1578         struct btrfs_trans_handle *trans;
1579         struct btrfs_fs_info *fs_info = root->fs_info;
1580
1581         fs_info->closing = 1;
1582         btrfs_transaction_flush_work(root);
1583         mutex_lock(&fs_info->fs_mutex);
1584         btrfs_defrag_dirty_roots(root->fs_info);
1585         trans = btrfs_start_transaction(root, 1);
1586         ret = btrfs_commit_transaction(trans, root);
1587         /* run commit again to  drop the original snapshot */
1588         trans = btrfs_start_transaction(root, 1);
1589         btrfs_commit_transaction(trans, root);
1590         ret = btrfs_write_and_wait_transaction(NULL, root);
1591         BUG_ON(ret);
1592
1593         write_ctree_super(NULL, root);
1594         mutex_unlock(&fs_info->fs_mutex);
1595
1596         btrfs_transaction_flush_work(root);
1597
1598         if (fs_info->delalloc_bytes) {
1599                 printk("btrfs: at unmount delalloc count %Lu\n",
1600                        fs_info->delalloc_bytes);
1601         }
1602         if (fs_info->extent_root->node)
1603                 free_extent_buffer(fs_info->extent_root->node);
1604
1605         if (fs_info->tree_root->node)
1606                 free_extent_buffer(fs_info->tree_root->node);
1607
1608         if (root->fs_info->chunk_root->node);
1609                 free_extent_buffer(root->fs_info->chunk_root->node);
1610
1611         if (root->fs_info->dev_root->node);
1612                 free_extent_buffer(root->fs_info->dev_root->node);
1613
1614         btrfs_free_block_groups(root->fs_info);
1615         del_fs_roots(fs_info);
1616
1617         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1618
1619         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1620         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1621         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1622         extent_io_tree_empty_lru(&fs_info->pending_del);
1623         extent_io_tree_empty_lru(&fs_info->extent_ins);
1624         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1625
1626         flush_workqueue(async_submit_workqueue);
1627         flush_workqueue(end_io_workqueue);
1628
1629         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1630
1631         flush_workqueue(async_submit_workqueue);
1632         destroy_workqueue(async_submit_workqueue);
1633
1634         flush_workqueue(end_io_workqueue);
1635         destroy_workqueue(end_io_workqueue);
1636
1637         iput(fs_info->btree_inode);
1638 #if 0
1639         while(!list_empty(&fs_info->hashers)) {
1640                 struct btrfs_hasher *hasher;
1641                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1642                                     hashers);
1643                 list_del(&hasher->hashers);
1644                 crypto_free_hash(&fs_info->hash_tfm);
1645                 kfree(hasher);
1646         }
1647 #endif
1648         btrfs_close_devices(fs_info->fs_devices);
1649         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1650
1651 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1652         bdi_destroy(&fs_info->bdi);
1653 #endif
1654
1655         kfree(fs_info->extent_root);
1656         kfree(fs_info->tree_root);
1657         kfree(fs_info->chunk_root);
1658         kfree(fs_info->dev_root);
1659         return 0;
1660 }
1661
1662 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1663 {
1664         int ret;
1665         struct inode *btree_inode = buf->first_page->mapping->host;
1666
1667         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1668         if (!ret)
1669                 return ret;
1670
1671         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1672                                     parent_transid);
1673         return !ret;
1674 }
1675
1676 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1677 {
1678         struct inode *btree_inode = buf->first_page->mapping->host;
1679         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1680                                           buf);
1681 }
1682
1683 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1684 {
1685         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1686         u64 transid = btrfs_header_generation(buf);
1687         struct inode *btree_inode = root->fs_info->btree_inode;
1688
1689         if (transid != root->fs_info->generation) {
1690                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1691                         (unsigned long long)buf->start,
1692                         transid, root->fs_info->generation);
1693                 WARN_ON(1);
1694         }
1695         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1696 }
1697
1698 void btrfs_throttle(struct btrfs_root *root)
1699 {
1700         struct backing_dev_info *bdi;
1701
1702         bdi = &root->fs_info->bdi;
1703         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1704 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1705                 congestion_wait(WRITE, HZ/20);
1706 #else
1707                 blk_congestion_wait(WRITE, HZ/20);
1708 #endif
1709         }
1710 }
1711
1712 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1713 {
1714         /*
1715          * looks as though older kernels can get into trouble with
1716          * this code, they end up stuck in balance_dirty_pages forever
1717          */
1718         struct extent_io_tree *tree;
1719         u64 num_dirty;
1720         u64 start = 0;
1721         unsigned long thresh = 16 * 1024 * 1024;
1722         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1723
1724         if (current_is_pdflush())
1725                 return;
1726
1727         num_dirty = count_range_bits(tree, &start, (u64)-1,
1728                                      thresh, EXTENT_DIRTY);
1729         if (num_dirty > thresh) {
1730                 balance_dirty_pages_ratelimited_nr(
1731                                    root->fs_info->btree_inode->i_mapping, 1);
1732         }
1733         return;
1734 }
1735
1736 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1737 {
1738         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1739         struct inode *btree_inode = root->fs_info->btree_inode;
1740         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1741                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1742 }
1743
1744 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1745 {
1746         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1747         struct inode *btree_inode = root->fs_info->btree_inode;
1748         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1749                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1750                         GFP_NOFS);
1751 }
1752
1753 int btrfs_buffer_defrag(struct extent_buffer *buf)
1754 {
1755         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1756         struct inode *btree_inode = root->fs_info->btree_inode;
1757         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1758                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1759 }
1760
1761 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1762 {
1763         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1764         struct inode *btree_inode = root->fs_info->btree_inode;
1765         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1766                      buf->start, buf->start + buf->len - 1,
1767                      EXTENT_DEFRAG_DONE, 0);
1768 }
1769
1770 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1771 {
1772         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1773         struct inode *btree_inode = root->fs_info->btree_inode;
1774         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1775                      buf->start, buf->start + buf->len - 1,
1776                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1777 }
1778
1779 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1780 {
1781         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1782         struct inode *btree_inode = root->fs_info->btree_inode;
1783         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1784                      buf->start, buf->start + buf->len - 1,
1785                      EXTENT_DEFRAG, GFP_NOFS);
1786 }
1787
1788 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1789 {
1790         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1791         int ret;
1792         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1793         if (ret == 0) {
1794                 buf->flags |= EXTENT_UPTODATE;
1795         }
1796         return ret;
1797 }
1798
1799 static struct extent_io_ops btree_extent_io_ops = {
1800         .writepage_io_hook = btree_writepage_io_hook,
1801         .readpage_end_io_hook = btree_readpage_end_io_hook,
1802         .submit_bio_hook = btree_submit_bio_hook,
1803         /* note we're sharing with inode.c for the merge bio hook */
1804         .merge_bio_hook = btrfs_merge_bio_hook,
1805 };