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