Btrfs: Add support for duplicate blocks on a single spindle
[linux-2.6] / fs / btrfs / disk-io.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/crc32c.h>
22 #include <linux/scatterlist.h>
23 #include <linux/swap.h>
24 #include <linux/radix-tree.h>
25 #include <linux/writeback.h>
26 #include <linux/buffer_head.h> // for block_sync_page
27 #include "ctree.h"
28 #include "disk-io.h"
29 #include "transaction.h"
30 #include "btrfs_inode.h"
31 #include "volumes.h"
32 #include "print-tree.h"
33
34 #if 0
35 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
36 {
37         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
38                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
39                        (unsigned long long)extent_buffer_blocknr(buf),
40                        (unsigned long long)btrfs_header_blocknr(buf));
41                 return 1;
42         }
43         return 0;
44 }
45 #endif
46
47 static struct extent_io_ops btree_extent_io_ops;
48
49 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
50                                     size_t page_offset, u64 start, u64 len,
51                                     int create)
52 {
53         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
54         struct extent_map *em;
55         int ret;
56
57 again:
58         spin_lock(&em_tree->lock);
59         em = lookup_extent_mapping(em_tree, start, len);
60         spin_unlock(&em_tree->lock);
61         if (em) {
62                 goto out;
63         }
64         em = alloc_extent_map(GFP_NOFS);
65         if (!em) {
66                 em = ERR_PTR(-ENOMEM);
67                 goto out;
68         }
69         em->start = 0;
70         em->len = i_size_read(inode);
71         em->block_start = 0;
72         em->bdev = inode->i_sb->s_bdev;
73
74         spin_lock(&em_tree->lock);
75         ret = add_extent_mapping(em_tree, em);
76         spin_unlock(&em_tree->lock);
77
78         if (ret == -EEXIST) {
79                 free_extent_map(em);
80                 em = NULL;
81                 goto again;
82         } else if (ret) {
83                 em = ERR_PTR(ret);
84         }
85 out:
86         return em;
87 }
88
89 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
90 {
91         return crc32c(seed, data, len);
92 }
93
94 void btrfs_csum_final(u32 crc, char *result)
95 {
96         *(__le32 *)result = ~cpu_to_le32(crc);
97 }
98
99 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
100                            int verify)
101 {
102         char result[BTRFS_CRC32_SIZE];
103         unsigned long len;
104         unsigned long cur_len;
105         unsigned long offset = BTRFS_CSUM_SIZE;
106         char *map_token = NULL;
107         char *kaddr;
108         unsigned long map_start;
109         unsigned long map_len;
110         int err;
111         u32 crc = ~(u32)0;
112
113         len = buf->len - offset;
114         while(len > 0) {
115                 err = map_private_extent_buffer(buf, offset, 32,
116                                         &map_token, &kaddr,
117                                         &map_start, &map_len, KM_USER0);
118                 if (err) {
119                         printk("failed to map extent buffer! %lu\n",
120                                offset);
121                         return 1;
122                 }
123                 cur_len = min(len, map_len - (offset - map_start));
124                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
125                                       crc, cur_len);
126                 len -= cur_len;
127                 offset += cur_len;
128                 unmap_extent_buffer(buf, map_token, KM_USER0);
129         }
130         btrfs_csum_final(crc, result);
131
132         if (verify) {
133                 int from_this_trans = 0;
134
135                 if (root->fs_info->running_transaction &&
136                     btrfs_header_generation(buf) ==
137                     root->fs_info->running_transaction->transid)
138                         from_this_trans = 1;
139
140                 /* FIXME, this is not good */
141                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
142                         u32 val;
143                         u32 found = 0;
144                         memcpy(&found, result, BTRFS_CRC32_SIZE);
145
146                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
147                         WARN_ON(1);
148                         printk("btrfs: %s checksum verify failed on %llu "
149                                "wanted %X found %X from_this_trans %d "
150                                "level %d\n",
151                                root->fs_info->sb->s_id,
152                                buf->start, val, found, from_this_trans,
153                                btrfs_header_level(buf));
154                         return 1;
155                 }
156         } else {
157                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
158         }
159         return 0;
160 }
161
162
163 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
164 {
165         struct extent_io_tree *tree;
166         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
167         u64 found_start;
168         int found_level;
169         unsigned long len;
170         struct extent_buffer *eb;
171         tree = &BTRFS_I(page->mapping->host)->io_tree;
172
173         if (page->private == EXTENT_PAGE_PRIVATE)
174                 goto out;
175         if (!page->private)
176                 goto out;
177         len = page->private >> 2;
178         if (len == 0) {
179                 WARN_ON(1);
180         }
181         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
182         read_extent_buffer_pages(tree, eb, start + PAGE_CACHE_SIZE, 1,
183                                  btree_get_extent);
184         btrfs_clear_buffer_defrag(eb);
185         found_start = btrfs_header_bytenr(eb);
186         if (found_start != start) {
187                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
188                        start, found_start, len);
189                 WARN_ON(1);
190                 goto err;
191         }
192         if (eb->first_page != page) {
193                 printk("bad first page %lu %lu\n", eb->first_page->index,
194                        page->index);
195                 WARN_ON(1);
196                 goto err;
197         }
198         if (!PageUptodate(page)) {
199                 printk("csum not up to date page %lu\n", page->index);
200                 WARN_ON(1);
201                 goto err;
202         }
203         found_level = btrfs_header_level(eb);
204         spin_lock(&root->fs_info->hash_lock);
205         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
206         spin_unlock(&root->fs_info->hash_lock);
207         csum_tree_block(root, eb, 0);
208 err:
209         free_extent_buffer(eb);
210 out:
211         return 0;
212 }
213
214 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
215 {
216         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
217
218         csum_dirty_buffer(root, page);
219         return 0;
220 }
221
222 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio)
223 {
224         struct btrfs_root *root = BTRFS_I(inode)->root;
225         u64 offset;
226         offset = bio->bi_sector << 9;
227         if (offset == BTRFS_SUPER_INFO_OFFSET) {
228                 bio->bi_bdev = root->fs_info->sb->s_bdev;
229                 submit_bio(rw, bio);
230                 return 0;
231         }
232         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio);
233 }
234
235 static int btree_writepage(struct page *page, struct writeback_control *wbc)
236 {
237         struct extent_io_tree *tree;
238         tree = &BTRFS_I(page->mapping->host)->io_tree;
239         return extent_write_full_page(tree, page, btree_get_extent, wbc);
240 }
241
242 static int btree_writepages(struct address_space *mapping,
243                             struct writeback_control *wbc)
244 {
245         struct extent_io_tree *tree;
246         tree = &BTRFS_I(mapping->host)->io_tree;
247         if (wbc->sync_mode == WB_SYNC_NONE) {
248                 u64 num_dirty;
249                 u64 start = 0;
250                 unsigned long thresh = 96 * 1024 * 1024;
251
252                 if (wbc->for_kupdate)
253                         return 0;
254
255                 if (current_is_pdflush()) {
256                         thresh = 96 * 1024 * 1024;
257                 } else {
258                         thresh = 8 * 1024 * 1024;
259                 }
260                 num_dirty = count_range_bits(tree, &start, (u64)-1,
261                                              thresh, EXTENT_DIRTY);
262                 if (num_dirty < thresh) {
263                         return 0;
264                 }
265         }
266         return extent_writepages(tree, mapping, btree_get_extent, wbc);
267 }
268
269 int btree_readpage(struct file *file, struct page *page)
270 {
271         struct extent_io_tree *tree;
272         tree = &BTRFS_I(page->mapping->host)->io_tree;
273         return extent_read_full_page(tree, page, btree_get_extent);
274 }
275
276 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
277 {
278         struct extent_io_tree *tree;
279         struct extent_map_tree *map;
280         int ret;
281
282         tree = &BTRFS_I(page->mapping->host)->io_tree;
283         map = &BTRFS_I(page->mapping->host)->extent_tree;
284         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
285         if (ret == 1) {
286                 ClearPagePrivate(page);
287                 set_page_private(page, 0);
288                 page_cache_release(page);
289         }
290         return ret;
291 }
292
293 static void btree_invalidatepage(struct page *page, unsigned long offset)
294 {
295         struct extent_io_tree *tree;
296         tree = &BTRFS_I(page->mapping->host)->io_tree;
297         extent_invalidatepage(tree, page, offset);
298         btree_releasepage(page, GFP_NOFS);
299 }
300
301 #if 0
302 static int btree_writepage(struct page *page, struct writeback_control *wbc)
303 {
304         struct buffer_head *bh;
305         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
306         struct buffer_head *head;
307         if (!page_has_buffers(page)) {
308                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
309                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
310         }
311         head = page_buffers(page);
312         bh = head;
313         do {
314                 if (buffer_dirty(bh))
315                         csum_tree_block(root, bh, 0);
316                 bh = bh->b_this_page;
317         } while (bh != head);
318         return block_write_full_page(page, btree_get_block, wbc);
319 }
320 #endif
321
322 static struct address_space_operations btree_aops = {
323         .readpage       = btree_readpage,
324         .writepage      = btree_writepage,
325         .writepages     = btree_writepages,
326         .releasepage    = btree_releasepage,
327         .invalidatepage = btree_invalidatepage,
328         .sync_page      = block_sync_page,
329 };
330
331 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
332 {
333         struct extent_buffer *buf = NULL;
334         struct inode *btree_inode = root->fs_info->btree_inode;
335         int ret = 0;
336
337         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
338         if (!buf)
339                 return 0;
340         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
341                                  buf, 0, 0, btree_get_extent);
342         free_extent_buffer(buf);
343         return ret;
344 }
345
346 static int close_all_devices(struct btrfs_fs_info *fs_info)
347 {
348         struct list_head *list;
349         struct list_head *next;
350         struct btrfs_device *device;
351
352         list = &fs_info->fs_devices->devices;
353         list_for_each(next, list) {
354                 device = list_entry(next, struct btrfs_device, dev_list);
355                 if (device->bdev && device->bdev != fs_info->sb->s_bdev)
356                         close_bdev_excl(device->bdev);
357                 device->bdev = NULL;
358         }
359         return 0;
360 }
361
362 int btrfs_verify_block_csum(struct btrfs_root *root,
363                             struct extent_buffer *buf)
364 {
365         struct extent_io_tree *io_tree;
366         u64 end;
367         int ret;
368
369         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
370         if (buf->flags & EXTENT_CSUM)
371                 return 0;
372
373         end = min_t(u64, buf->len, PAGE_CACHE_SIZE);
374         end = buf->start + end - 1;
375         if (test_range_bit(io_tree, buf->start, end, EXTENT_CSUM, 1)) {
376                 buf->flags |= EXTENT_CSUM;
377                 return 0;
378         }
379
380         lock_extent(io_tree, buf->start, end, GFP_NOFS);
381
382         if (test_range_bit(io_tree, buf->start, end, EXTENT_CSUM, 1)) {
383                 buf->flags |= EXTENT_CSUM;
384                 ret = 0;
385                 goto out_unlock;
386         }
387
388         ret = csum_tree_block(root, buf, 1);
389         set_extent_bits(io_tree, buf->start, end, EXTENT_CSUM, GFP_NOFS);
390         buf->flags |= EXTENT_CSUM;
391
392 out_unlock:
393         unlock_extent(io_tree, buf->start, end, GFP_NOFS);
394         return ret;
395 }
396
397 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
398                                             u64 bytenr, u32 blocksize)
399 {
400         struct inode *btree_inode = root->fs_info->btree_inode;
401         struct extent_buffer *eb;
402         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
403                                 bytenr, blocksize, GFP_NOFS);
404         return eb;
405 }
406
407 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
408                                                  u64 bytenr, u32 blocksize)
409 {
410         struct inode *btree_inode = root->fs_info->btree_inode;
411         struct extent_buffer *eb;
412
413         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
414                                  bytenr, blocksize, NULL, GFP_NOFS);
415         return eb;
416 }
417
418
419 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
420                                       u32 blocksize)
421 {
422         struct extent_buffer *buf = NULL;
423         struct inode *btree_inode = root->fs_info->btree_inode;
424         struct extent_io_tree *io_tree;
425         int ret;
426
427         io_tree = &BTRFS_I(btree_inode)->io_tree;
428
429         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
430         if (!buf)
431                 return NULL;
432         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, buf, 0, 1,
433                                  btree_get_extent);
434
435         ret = btrfs_verify_block_csum(root, buf);
436         return buf;
437 }
438
439 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
440                      struct extent_buffer *buf)
441 {
442         struct inode *btree_inode = root->fs_info->btree_inode;
443         if (btrfs_header_generation(buf) ==
444             root->fs_info->running_transaction->transid)
445                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
446                                           buf);
447         return 0;
448 }
449
450 int wait_on_tree_block_writeback(struct btrfs_root *root,
451                                  struct extent_buffer *buf)
452 {
453         struct inode *btree_inode = root->fs_info->btree_inode;
454         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
455                                         buf);
456         return 0;
457 }
458
459 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
460                         u32 stripesize, struct btrfs_root *root,
461                         struct btrfs_fs_info *fs_info,
462                         u64 objectid)
463 {
464         root->node = NULL;
465         root->inode = NULL;
466         root->commit_root = NULL;
467         root->sectorsize = sectorsize;
468         root->nodesize = nodesize;
469         root->leafsize = leafsize;
470         root->stripesize = stripesize;
471         root->ref_cows = 0;
472         root->track_dirty = 0;
473
474         root->fs_info = fs_info;
475         root->objectid = objectid;
476         root->last_trans = 0;
477         root->highest_inode = 0;
478         root->last_inode_alloc = 0;
479         root->name = NULL;
480         root->in_sysfs = 0;
481
482         INIT_LIST_HEAD(&root->dirty_list);
483         memset(&root->root_key, 0, sizeof(root->root_key));
484         memset(&root->root_item, 0, sizeof(root->root_item));
485         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
486         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
487         init_completion(&root->kobj_unregister);
488         root->defrag_running = 0;
489         root->defrag_level = 0;
490         root->root_key.objectid = objectid;
491         return 0;
492 }
493
494 static int find_and_setup_root(struct btrfs_root *tree_root,
495                                struct btrfs_fs_info *fs_info,
496                                u64 objectid,
497                                struct btrfs_root *root)
498 {
499         int ret;
500         u32 blocksize;
501
502         __setup_root(tree_root->nodesize, tree_root->leafsize,
503                      tree_root->sectorsize, tree_root->stripesize,
504                      root, fs_info, objectid);
505         ret = btrfs_find_last_root(tree_root, objectid,
506                                    &root->root_item, &root->root_key);
507         BUG_ON(ret);
508
509         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
510         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
511                                      blocksize);
512         BUG_ON(!root->node);
513         return 0;
514 }
515
516 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
517                                                struct btrfs_key *location)
518 {
519         struct btrfs_root *root;
520         struct btrfs_root *tree_root = fs_info->tree_root;
521         struct btrfs_path *path;
522         struct extent_buffer *l;
523         u64 highest_inode;
524         u32 blocksize;
525         int ret = 0;
526
527         root = kzalloc(sizeof(*root), GFP_NOFS);
528         if (!root)
529                 return ERR_PTR(-ENOMEM);
530         if (location->offset == (u64)-1) {
531                 ret = find_and_setup_root(tree_root, fs_info,
532                                           location->objectid, root);
533                 if (ret) {
534                         kfree(root);
535                         return ERR_PTR(ret);
536                 }
537                 goto insert;
538         }
539
540         __setup_root(tree_root->nodesize, tree_root->leafsize,
541                      tree_root->sectorsize, tree_root->stripesize,
542                      root, fs_info, location->objectid);
543
544         path = btrfs_alloc_path();
545         BUG_ON(!path);
546         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
547         if (ret != 0) {
548                 if (ret > 0)
549                         ret = -ENOENT;
550                 goto out;
551         }
552         l = path->nodes[0];
553         read_extent_buffer(l, &root->root_item,
554                btrfs_item_ptr_offset(l, path->slots[0]),
555                sizeof(root->root_item));
556         memcpy(&root->root_key, location, sizeof(*location));
557         ret = 0;
558 out:
559         btrfs_release_path(root, path);
560         btrfs_free_path(path);
561         if (ret) {
562                 kfree(root);
563                 return ERR_PTR(ret);
564         }
565         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
566         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
567                                      blocksize);
568         BUG_ON(!root->node);
569 insert:
570         root->ref_cows = 1;
571         ret = btrfs_find_highest_inode(root, &highest_inode);
572         if (ret == 0) {
573                 root->highest_inode = highest_inode;
574                 root->last_inode_alloc = highest_inode;
575         }
576         return root;
577 }
578
579 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
580                                         u64 root_objectid)
581 {
582         struct btrfs_root *root;
583
584         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
585                 return fs_info->tree_root;
586         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
587                 return fs_info->extent_root;
588
589         root = radix_tree_lookup(&fs_info->fs_roots_radix,
590                                  (unsigned long)root_objectid);
591         return root;
592 }
593
594 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
595                                               struct btrfs_key *location)
596 {
597         struct btrfs_root *root;
598         int ret;
599
600         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
601                 return fs_info->tree_root;
602         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
603                 return fs_info->extent_root;
604
605         root = radix_tree_lookup(&fs_info->fs_roots_radix,
606                                  (unsigned long)location->objectid);
607         if (root)
608                 return root;
609
610         root = btrfs_read_fs_root_no_radix(fs_info, location);
611         if (IS_ERR(root))
612                 return root;
613         ret = radix_tree_insert(&fs_info->fs_roots_radix,
614                                 (unsigned long)root->root_key.objectid,
615                                 root);
616         if (ret) {
617                 free_extent_buffer(root->node);
618                 kfree(root);
619                 return ERR_PTR(ret);
620         }
621         ret = btrfs_find_dead_roots(fs_info->tree_root,
622                                     root->root_key.objectid, root);
623         BUG_ON(ret);
624
625         return root;
626 }
627
628 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
629                                       struct btrfs_key *location,
630                                       const char *name, int namelen)
631 {
632         struct btrfs_root *root;
633         int ret;
634
635         root = btrfs_read_fs_root_no_name(fs_info, location);
636         if (!root)
637                 return NULL;
638
639         if (root->in_sysfs)
640                 return root;
641
642         ret = btrfs_set_root_name(root, name, namelen);
643         if (ret) {
644                 free_extent_buffer(root->node);
645                 kfree(root);
646                 return ERR_PTR(ret);
647         }
648
649         ret = btrfs_sysfs_add_root(root);
650         if (ret) {
651                 free_extent_buffer(root->node);
652                 kfree(root->name);
653                 kfree(root);
654                 return ERR_PTR(ret);
655         }
656         root->in_sysfs = 1;
657         return root;
658 }
659 #if 0
660 static int add_hasher(struct btrfs_fs_info *info, char *type) {
661         struct btrfs_hasher *hasher;
662
663         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
664         if (!hasher)
665                 return -ENOMEM;
666         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
667         if (!hasher->hash_tfm) {
668                 kfree(hasher);
669                 return -EINVAL;
670         }
671         spin_lock(&info->hash_lock);
672         list_add(&hasher->list, &info->hashers);
673         spin_unlock(&info->hash_lock);
674         return 0;
675 }
676 #endif
677
678 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
679 {
680         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
681         int ret = 0;
682         struct list_head *cur;
683         struct btrfs_device *device;
684         struct backing_dev_info *bdi;
685
686         list_for_each(cur, &info->fs_devices->devices) {
687                 device = list_entry(cur, struct btrfs_device, dev_list);
688                 bdi = blk_get_backing_dev_info(device->bdev);
689                 if (bdi && bdi_congested(bdi, bdi_bits)) {
690                         ret = 1;
691                         break;
692                 }
693         }
694         return ret;
695 }
696
697 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
698 {
699         struct list_head *cur;
700         struct btrfs_device *device;
701         struct btrfs_fs_info *info;
702
703         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
704         list_for_each(cur, &info->fs_devices->devices) {
705                 device = list_entry(cur, struct btrfs_device, dev_list);
706                 bdi = blk_get_backing_dev_info(device->bdev);
707                 if (bdi->unplug_io_fn) {
708                         bdi->unplug_io_fn(bdi, page);
709                 }
710         }
711 }
712
713 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
714 {
715         bdi_init(bdi);
716         bdi->ra_pages   = default_backing_dev_info.ra_pages * 4;
717         bdi->state              = 0;
718         bdi->capabilities       = default_backing_dev_info.capabilities;
719         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
720         bdi->unplug_io_data     = info;
721         bdi->congested_fn       = btrfs_congested_fn;
722         bdi->congested_data     = info;
723         return 0;
724 }
725
726 struct btrfs_root *open_ctree(struct super_block *sb,
727                               struct btrfs_fs_devices *fs_devices)
728 {
729         u32 sectorsize;
730         u32 nodesize;
731         u32 leafsize;
732         u32 blocksize;
733         u32 stripesize;
734         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
735                                                  GFP_NOFS);
736         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
737                                                GFP_NOFS);
738         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
739                                                 GFP_NOFS);
740         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
741                                                 GFP_NOFS);
742         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
743                                               GFP_NOFS);
744         int ret;
745         int err = -EINVAL;
746         struct btrfs_super_block *disk_super;
747
748         if (!extent_root || !tree_root || !fs_info) {
749                 err = -ENOMEM;
750                 goto fail;
751         }
752         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
753         INIT_LIST_HEAD(&fs_info->trans_list);
754         INIT_LIST_HEAD(&fs_info->dead_roots);
755         INIT_LIST_HEAD(&fs_info->hashers);
756         spin_lock_init(&fs_info->hash_lock);
757         spin_lock_init(&fs_info->delalloc_lock);
758         spin_lock_init(&fs_info->new_trans_lock);
759
760         init_completion(&fs_info->kobj_unregister);
761         sb_set_blocksize(sb, 4096);
762         fs_info->tree_root = tree_root;
763         fs_info->extent_root = extent_root;
764         fs_info->chunk_root = chunk_root;
765         fs_info->dev_root = dev_root;
766         fs_info->fs_devices = fs_devices;
767         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
768         INIT_LIST_HEAD(&fs_info->space_info);
769         btrfs_mapping_init(&fs_info->mapping_tree);
770         fs_info->sb = sb;
771         fs_info->max_extent = (u64)-1;
772         fs_info->max_inline = 8192 * 1024;
773         setup_bdi(fs_info, &fs_info->bdi);
774         fs_info->btree_inode = new_inode(sb);
775         fs_info->btree_inode->i_ino = 1;
776         fs_info->btree_inode->i_nlink = 1;
777         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
778         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
779         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
780
781         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
782                              fs_info->btree_inode->i_mapping,
783                              GFP_NOFS);
784         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
785                              GFP_NOFS);
786
787         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
788
789         extent_io_tree_init(&fs_info->free_space_cache,
790                              fs_info->btree_inode->i_mapping, GFP_NOFS);
791         extent_io_tree_init(&fs_info->block_group_cache,
792                              fs_info->btree_inode->i_mapping, GFP_NOFS);
793         extent_io_tree_init(&fs_info->pinned_extents,
794                              fs_info->btree_inode->i_mapping, GFP_NOFS);
795         extent_io_tree_init(&fs_info->pending_del,
796                              fs_info->btree_inode->i_mapping, GFP_NOFS);
797         extent_io_tree_init(&fs_info->extent_ins,
798                              fs_info->btree_inode->i_mapping, GFP_NOFS);
799         fs_info->do_barriers = 1;
800
801 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
802         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
803 #else
804         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
805 #endif
806         BTRFS_I(fs_info->btree_inode)->root = tree_root;
807         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
808                sizeof(struct btrfs_key));
809         insert_inode_hash(fs_info->btree_inode);
810         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
811
812         mutex_init(&fs_info->trans_mutex);
813         mutex_init(&fs_info->fs_mutex);
814
815 #if 0
816         ret = add_hasher(fs_info, "crc32c");
817         if (ret) {
818                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
819                 err = -ENOMEM;
820                 goto fail_iput;
821         }
822 #endif
823         __setup_root(4096, 4096, 4096, 4096, tree_root,
824                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
825
826         fs_info->sb_buffer = read_tree_block(tree_root,
827                                              BTRFS_SUPER_INFO_OFFSET,
828                                              4096);
829
830         if (!fs_info->sb_buffer)
831                 goto fail_iput;
832
833         read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
834                            sizeof(fs_info->super_copy));
835
836         read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
837                            (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
838                            BTRFS_FSID_SIZE);
839
840         disk_super = &fs_info->super_copy;
841         if (!btrfs_super_root(disk_super))
842                 goto fail_sb_buffer;
843
844         if (btrfs_super_num_devices(disk_super) != fs_devices->num_devices) {
845                 printk("Btrfs: wanted %llu devices, but found %llu\n",
846                        (unsigned long long)btrfs_super_num_devices(disk_super),
847                        (unsigned long long)fs_devices->num_devices);
848                 goto fail_sb_buffer;
849         }
850         nodesize = btrfs_super_nodesize(disk_super);
851         leafsize = btrfs_super_leafsize(disk_super);
852         sectorsize = btrfs_super_sectorsize(disk_super);
853         stripesize = btrfs_super_stripesize(disk_super);
854         tree_root->nodesize = nodesize;
855         tree_root->leafsize = leafsize;
856         tree_root->sectorsize = sectorsize;
857         tree_root->stripesize = stripesize;
858         sb_set_blocksize(sb, sectorsize);
859
860         i_size_write(fs_info->btree_inode,
861                      btrfs_super_total_bytes(disk_super));
862
863         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
864                     sizeof(disk_super->magic))) {
865                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
866                 goto fail_sb_buffer;
867         }
868
869         mutex_lock(&fs_info->fs_mutex);
870
871         ret = btrfs_read_sys_array(tree_root);
872         BUG_ON(ret);
873
874         blocksize = btrfs_level_size(tree_root,
875                                      btrfs_super_chunk_root_level(disk_super));
876
877         __setup_root(nodesize, leafsize, sectorsize, stripesize,
878                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
879
880         chunk_root->node = read_tree_block(chunk_root,
881                                            btrfs_super_chunk_root(disk_super),
882                                            blocksize);
883         BUG_ON(!chunk_root->node);
884
885         ret = btrfs_read_chunk_tree(chunk_root);
886         BUG_ON(ret);
887
888         blocksize = btrfs_level_size(tree_root,
889                                      btrfs_super_root_level(disk_super));
890
891
892         tree_root->node = read_tree_block(tree_root,
893                                           btrfs_super_root(disk_super),
894                                           blocksize);
895         if (!tree_root->node)
896                 goto fail_sb_buffer;
897
898
899         ret = find_and_setup_root(tree_root, fs_info,
900                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
901         if (ret)
902                 goto fail_tree_root;
903         extent_root->track_dirty = 1;
904
905         ret = find_and_setup_root(tree_root, fs_info,
906                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
907         dev_root->track_dirty = 1;
908
909         if (ret)
910                 goto fail_extent_root;
911
912         btrfs_read_block_groups(extent_root);
913
914         fs_info->generation = btrfs_super_generation(disk_super) + 1;
915         if (btrfs_super_num_devices(disk_super) > 0) {
916                 fs_info->data_alloc_profile = BTRFS_BLOCK_GROUP_RAID0 |
917                         BTRFS_BLOCK_GROUP_RAID1;
918                 fs_info->metadata_alloc_profile = BTRFS_BLOCK_GROUP_RAID1 |
919                         BTRFS_BLOCK_GROUP_DUP;
920                 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
921         }
922         mutex_unlock(&fs_info->fs_mutex);
923         return tree_root;
924
925 fail_extent_root:
926         free_extent_buffer(extent_root->node);
927 fail_tree_root:
928         mutex_unlock(&fs_info->fs_mutex);
929         free_extent_buffer(tree_root->node);
930 fail_sb_buffer:
931         free_extent_buffer(fs_info->sb_buffer);
932         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
933 fail_iput:
934         iput(fs_info->btree_inode);
935 fail:
936         close_all_devices(fs_info);
937         kfree(extent_root);
938         kfree(tree_root);
939         bdi_destroy(&fs_info->bdi);
940         kfree(fs_info);
941         return ERR_PTR(err);
942 }
943
944 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
945                       *root)
946 {
947         int ret;
948         struct extent_buffer *super = root->fs_info->sb_buffer;
949         struct inode *btree_inode = root->fs_info->btree_inode;
950         struct super_block *sb = root->fs_info->sb;
951
952         if (!btrfs_test_opt(root, NOBARRIER))
953                 blkdev_issue_flush(sb->s_bdev, NULL);
954         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, super);
955         ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
956                                      super->start, super->len);
957         if (!btrfs_test_opt(root, NOBARRIER))
958                 blkdev_issue_flush(sb->s_bdev, NULL);
959         return ret;
960 }
961
962 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
963 {
964         radix_tree_delete(&fs_info->fs_roots_radix,
965                           (unsigned long)root->root_key.objectid);
966         if (root->in_sysfs)
967                 btrfs_sysfs_del_root(root);
968         if (root->inode)
969                 iput(root->inode);
970         if (root->node)
971                 free_extent_buffer(root->node);
972         if (root->commit_root)
973                 free_extent_buffer(root->commit_root);
974         if (root->name)
975                 kfree(root->name);
976         kfree(root);
977         return 0;
978 }
979
980 static int del_fs_roots(struct btrfs_fs_info *fs_info)
981 {
982         int ret;
983         struct btrfs_root *gang[8];
984         int i;
985
986         while(1) {
987                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
988                                              (void **)gang, 0,
989                                              ARRAY_SIZE(gang));
990                 if (!ret)
991                         break;
992                 for (i = 0; i < ret; i++)
993                         btrfs_free_fs_root(fs_info, gang[i]);
994         }
995         return 0;
996 }
997
998 int close_ctree(struct btrfs_root *root)
999 {
1000         int ret;
1001         struct btrfs_trans_handle *trans;
1002         struct btrfs_fs_info *fs_info = root->fs_info;
1003
1004         fs_info->closing = 1;
1005         btrfs_transaction_flush_work(root);
1006         mutex_lock(&fs_info->fs_mutex);
1007         btrfs_defrag_dirty_roots(root->fs_info);
1008         trans = btrfs_start_transaction(root, 1);
1009         ret = btrfs_commit_transaction(trans, root);
1010         /* run commit again to  drop the original snapshot */
1011         trans = btrfs_start_transaction(root, 1);
1012         btrfs_commit_transaction(trans, root);
1013         ret = btrfs_write_and_wait_transaction(NULL, root);
1014         BUG_ON(ret);
1015         write_ctree_super(NULL, root);
1016         mutex_unlock(&fs_info->fs_mutex);
1017
1018         if (fs_info->delalloc_bytes) {
1019                 printk("btrfs: at unmount delalloc count %Lu\n",
1020                        fs_info->delalloc_bytes);
1021         }
1022         if (fs_info->extent_root->node)
1023                 free_extent_buffer(fs_info->extent_root->node);
1024
1025         if (fs_info->tree_root->node)
1026                 free_extent_buffer(fs_info->tree_root->node);
1027
1028         if (root->fs_info->chunk_root->node);
1029                 free_extent_buffer(root->fs_info->chunk_root->node);
1030
1031         if (root->fs_info->dev_root->node);
1032                 free_extent_buffer(root->fs_info->dev_root->node);
1033
1034         free_extent_buffer(fs_info->sb_buffer);
1035
1036         btrfs_free_block_groups(root->fs_info);
1037         del_fs_roots(fs_info);
1038
1039         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1040
1041         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1042         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1043         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1044         extent_io_tree_empty_lru(&fs_info->pending_del);
1045         extent_io_tree_empty_lru(&fs_info->extent_ins);
1046         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1047
1048         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1049
1050         iput(fs_info->btree_inode);
1051 #if 0
1052         while(!list_empty(&fs_info->hashers)) {
1053                 struct btrfs_hasher *hasher;
1054                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1055                                     hashers);
1056                 list_del(&hasher->hashers);
1057                 crypto_free_hash(&fs_info->hash_tfm);
1058                 kfree(hasher);
1059         }
1060 #endif
1061         close_all_devices(fs_info);
1062         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1063         bdi_destroy(&fs_info->bdi);
1064
1065         kfree(fs_info->extent_root);
1066         kfree(fs_info->tree_root);
1067         kfree(fs_info->chunk_root);
1068         kfree(fs_info->dev_root);
1069         return 0;
1070 }
1071
1072 int btrfs_buffer_uptodate(struct extent_buffer *buf)
1073 {
1074         struct inode *btree_inode = buf->first_page->mapping->host;
1075         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1076 }
1077
1078 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1079 {
1080         struct inode *btree_inode = buf->first_page->mapping->host;
1081         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1082                                           buf);
1083 }
1084
1085 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1086 {
1087         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1088         u64 transid = btrfs_header_generation(buf);
1089         struct inode *btree_inode = root->fs_info->btree_inode;
1090
1091         if (transid != root->fs_info->generation) {
1092                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1093                         (unsigned long long)buf->start,
1094                         transid, root->fs_info->generation);
1095                 WARN_ON(1);
1096         }
1097         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1098 }
1099
1100 void btrfs_throttle(struct btrfs_root *root)
1101 {
1102         struct backing_dev_info *bdi;
1103
1104         bdi = root->fs_info->sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1105         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1106 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1107                 congestion_wait(WRITE, HZ/20);
1108 #else
1109                 blk_congestion_wait(WRITE, HZ/20);
1110 #endif
1111         }
1112 }
1113
1114 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1115 {
1116         balance_dirty_pages_ratelimited_nr(
1117                                    root->fs_info->btree_inode->i_mapping, 1);
1118 }
1119
1120 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1121 {
1122         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1123         struct inode *btree_inode = root->fs_info->btree_inode;
1124         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1125                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1126 }
1127
1128 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1129 {
1130         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1131         struct inode *btree_inode = root->fs_info->btree_inode;
1132         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1133                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1134                         GFP_NOFS);
1135 }
1136
1137 int btrfs_buffer_defrag(struct extent_buffer *buf)
1138 {
1139         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1140         struct inode *btree_inode = root->fs_info->btree_inode;
1141         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1142                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1143 }
1144
1145 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1146 {
1147         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1148         struct inode *btree_inode = root->fs_info->btree_inode;
1149         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1150                      buf->start, buf->start + buf->len - 1,
1151                      EXTENT_DEFRAG_DONE, 0);
1152 }
1153
1154 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1155 {
1156         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1157         struct inode *btree_inode = root->fs_info->btree_inode;
1158         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1159                      buf->start, buf->start + buf->len - 1,
1160                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1161 }
1162
1163 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1164 {
1165         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1166         struct inode *btree_inode = root->fs_info->btree_inode;
1167         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1168                      buf->start, buf->start + buf->len - 1,
1169                      EXTENT_DEFRAG, GFP_NOFS);
1170 }
1171
1172 int btrfs_read_buffer(struct extent_buffer *buf)
1173 {
1174         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1175         struct inode *btree_inode = root->fs_info->btree_inode;
1176         return read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1177                                         buf, 0, 1, btree_get_extent);
1178 }
1179
1180 static struct extent_io_ops btree_extent_io_ops = {
1181         .writepage_io_hook = btree_writepage_io_hook,
1182         .submit_bio_hook = btree_submit_bio_hook,
1183         /* note we're sharing with inode.c for the merge bio hook */
1184         .merge_bio_hook = btrfs_merge_bio_hook,
1185 };