Btrfs: Fine tune the btree writeback exclusion some more
[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 "print-tree.h"
32
33 #if 0
34 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
35 {
36         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
37                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
38                        (unsigned long long)extent_buffer_blocknr(buf),
39                        (unsigned long long)btrfs_header_blocknr(buf));
40                 return 1;
41         }
42         return 0;
43 }
44 #endif
45
46 static struct extent_map_ops btree_extent_map_ops;
47
48 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
49                                             u64 bytenr, u32 blocksize)
50 {
51         struct inode *btree_inode = root->fs_info->btree_inode;
52         struct extent_buffer *eb;
53         eb = find_extent_buffer(&BTRFS_I(btree_inode)->extent_tree,
54                                 bytenr, blocksize, GFP_NOFS);
55         return eb;
56 }
57
58 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
59                                                  u64 bytenr, u32 blocksize)
60 {
61         struct inode *btree_inode = root->fs_info->btree_inode;
62         struct extent_buffer *eb;
63
64         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->extent_tree,
65                                  bytenr, blocksize, NULL, GFP_NOFS);
66         return eb;
67 }
68
69 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
70                                     size_t page_offset, u64 start, u64 end,
71                                     int create)
72 {
73         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
74         struct extent_map *em;
75         int ret;
76
77 again:
78         em = lookup_extent_mapping(em_tree, start, end);
79         if (em) {
80                 goto out;
81         }
82         em = alloc_extent_map(GFP_NOFS);
83         if (!em) {
84                 em = ERR_PTR(-ENOMEM);
85                 goto out;
86         }
87         em->start = 0;
88         em->end = (i_size_read(inode) & ~((u64)PAGE_CACHE_SIZE -1)) - 1;
89         em->block_start = 0;
90         em->block_end = em->end;
91         em->bdev = inode->i_sb->s_bdev;
92         ret = add_extent_mapping(em_tree, em);
93         if (ret == -EEXIST) {
94                 free_extent_map(em);
95                 em = NULL;
96                 goto again;
97         } else if (ret) {
98                 em = ERR_PTR(ret);
99         }
100 out:
101         return em;
102 }
103
104 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
105 {
106         return crc32c(seed, data, len);
107 }
108
109 void btrfs_csum_final(u32 crc, char *result)
110 {
111         *(__le32 *)result = ~cpu_to_le32(crc);
112 }
113
114 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
115                            int verify)
116 {
117         char result[BTRFS_CRC32_SIZE];
118         unsigned long len;
119         unsigned long cur_len;
120         unsigned long offset = BTRFS_CSUM_SIZE;
121         char *map_token = NULL;
122         char *kaddr;
123         unsigned long map_start;
124         unsigned long map_len;
125         int err;
126         u32 crc = ~(u32)0;
127
128         len = buf->len - offset;
129         while(len > 0) {
130                 err = map_private_extent_buffer(buf, offset, 32,
131                                         &map_token, &kaddr,
132                                         &map_start, &map_len, KM_USER0);
133                 if (err) {
134                         printk("failed to map extent buffer! %lu\n",
135                                offset);
136                         return 1;
137                 }
138                 cur_len = min(len, map_len - (offset - map_start));
139                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
140                                       crc, cur_len);
141                 len -= cur_len;
142                 offset += cur_len;
143                 unmap_extent_buffer(buf, map_token, KM_USER0);
144         }
145         btrfs_csum_final(crc, result);
146
147         if (verify) {
148                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
149                         printk("btrfs: %s checksum verify failed on %llu\n",
150                                root->fs_info->sb->s_id,
151                                buf->start);
152                         return 1;
153                 }
154         } else {
155                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
156         }
157         return 0;
158 }
159
160
161 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
162 {
163         struct extent_map_tree *tree;
164         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
165         u64 found_start;
166         int found_level;
167         unsigned long len;
168         struct extent_buffer *eb;
169         tree = &BTRFS_I(page->mapping->host)->extent_tree;
170
171         if (page->private == EXTENT_PAGE_PRIVATE)
172                 goto out;
173         if (!page->private)
174                 goto out;
175         len = page->private >> 2;
176         if (len == 0) {
177                 WARN_ON(1);
178         }
179         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
180         read_extent_buffer_pages(tree, eb, start + PAGE_CACHE_SIZE, 1);
181         found_start = btrfs_header_bytenr(eb);
182         if (found_start != start) {
183                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
184                        start, found_start, len);
185         }
186         found_level = btrfs_header_level(eb);
187         csum_tree_block(root, eb, 0);
188         free_extent_buffer(eb);
189 out:
190         return 0;
191 }
192
193 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
194 {
195         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
196
197         csum_dirty_buffer(root, page);
198         return 0;
199 }
200
201 static int btree_writepage(struct page *page, struct writeback_control *wbc)
202 {
203         struct extent_map_tree *tree;
204         tree = &BTRFS_I(page->mapping->host)->extent_tree;
205         return extent_write_full_page(tree, page, btree_get_extent, wbc);
206 }
207
208 static int btree_writepages(struct address_space *mapping,
209                             struct writeback_control *wbc)
210 {
211         struct extent_map_tree *tree;
212         tree = &BTRFS_I(mapping->host)->extent_tree;
213         if (wbc->sync_mode == WB_SYNC_NONE && current_is_pdflush()) {
214                 u64 num_dirty;
215                 u64 start = 0;
216                 unsigned long thresh = 96 * 1024 * 1024;
217
218                 if (wbc->for_kupdate)
219                         return 0;
220
221                 num_dirty = count_range_bits(tree, &start, thresh, EXTENT_DIRTY);
222                 if (num_dirty < thresh) {
223                         return 0;
224                 }
225         }
226         return extent_writepages(tree, mapping, btree_get_extent, wbc);
227 }
228
229 int btree_readpage(struct file *file, struct page *page)
230 {
231         struct extent_map_tree *tree;
232         tree = &BTRFS_I(page->mapping->host)->extent_tree;
233         return extent_read_full_page(tree, page, btree_get_extent);
234 }
235
236 static int btree_releasepage(struct page *page, gfp_t unused_gfp_flags)
237 {
238         struct extent_map_tree *tree;
239         int ret;
240
241         tree = &BTRFS_I(page->mapping->host)->extent_tree;
242         ret = try_release_extent_mapping(tree, page);
243         if (ret == 1) {
244                 ClearPagePrivate(page);
245                 set_page_private(page, 0);
246                 page_cache_release(page);
247         }
248         return ret;
249 }
250
251 static void btree_invalidatepage(struct page *page, unsigned long offset)
252 {
253         struct extent_map_tree *tree;
254         tree = &BTRFS_I(page->mapping->host)->extent_tree;
255         extent_invalidatepage(tree, page, offset);
256         btree_releasepage(page, GFP_NOFS);
257 }
258
259 #if 0
260 static int btree_writepage(struct page *page, struct writeback_control *wbc)
261 {
262         struct buffer_head *bh;
263         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
264         struct buffer_head *head;
265         if (!page_has_buffers(page)) {
266                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
267                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
268         }
269         head = page_buffers(page);
270         bh = head;
271         do {
272                 if (buffer_dirty(bh))
273                         csum_tree_block(root, bh, 0);
274                 bh = bh->b_this_page;
275         } while (bh != head);
276         return block_write_full_page(page, btree_get_block, wbc);
277 }
278 #endif
279
280 static struct address_space_operations btree_aops = {
281         .readpage       = btree_readpage,
282         .writepage      = btree_writepage,
283         .writepages     = btree_writepages,
284         .releasepage    = btree_releasepage,
285         .invalidatepage = btree_invalidatepage,
286         .sync_page      = block_sync_page,
287 };
288
289 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
290 {
291         struct extent_buffer *buf = NULL;
292         struct inode *btree_inode = root->fs_info->btree_inode;
293         int ret = 0;
294
295         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
296         if (!buf)
297                 return 0;
298         read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
299                                  buf, 0, 0);
300         free_extent_buffer(buf);
301         return ret;
302 }
303
304 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
305                                       u32 blocksize)
306 {
307         struct extent_buffer *buf = NULL;
308         struct inode *btree_inode = root->fs_info->btree_inode;
309         struct extent_map_tree *extent_tree;
310         int ret;
311
312         extent_tree = &BTRFS_I(btree_inode)->extent_tree;
313
314         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
315         if (!buf)
316                 return NULL;
317         read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
318                                  buf, 0, 1);
319         if (buf->flags & EXTENT_CSUM) {
320                 return buf;
321         }
322         if (test_range_bit(extent_tree, buf->start, buf->start + buf->len - 1,
323                            EXTENT_CSUM, 1)) {
324                 buf->flags |= EXTENT_CSUM;
325                 return buf;
326         }
327         ret = csum_tree_block(root, buf, 1);
328         set_extent_bits(extent_tree, buf->start,
329                         buf->start + buf->len - 1,
330                         EXTENT_CSUM, GFP_NOFS);
331         buf->flags |= EXTENT_CSUM;
332         return buf;
333 }
334
335 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
336                      struct extent_buffer *buf)
337 {
338         struct inode *btree_inode = root->fs_info->btree_inode;
339         clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, buf);
340         return 0;
341 }
342
343 int wait_on_tree_block_writeback(struct btrfs_root *root,
344                                  struct extent_buffer *buf)
345 {
346         struct inode *btree_inode = root->fs_info->btree_inode;
347         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->extent_tree,
348                                         buf);
349         return 0;
350 }
351
352 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
353                         struct btrfs_root *root,
354                         struct btrfs_fs_info *fs_info,
355                         u64 objectid)
356 {
357         root->node = NULL;
358         root->inode = NULL;
359         root->commit_root = NULL;
360         root->sectorsize = sectorsize;
361         root->nodesize = nodesize;
362         root->leafsize = leafsize;
363         root->ref_cows = 0;
364         root->fs_info = fs_info;
365         root->objectid = objectid;
366         root->last_trans = 0;
367         root->highest_inode = 0;
368         root->last_inode_alloc = 0;
369         root->name = NULL;
370         memset(&root->root_key, 0, sizeof(root->root_key));
371         memset(&root->root_item, 0, sizeof(root->root_item));
372         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
373         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
374         init_completion(&root->kobj_unregister);
375         init_rwsem(&root->snap_sem);
376         root->defrag_running = 0;
377         root->defrag_level = 0;
378         root->root_key.objectid = objectid;
379         return 0;
380 }
381
382 static int find_and_setup_root(struct btrfs_root *tree_root,
383                                struct btrfs_fs_info *fs_info,
384                                u64 objectid,
385                                struct btrfs_root *root)
386 {
387         int ret;
388         u32 blocksize;
389
390         __setup_root(tree_root->nodesize, tree_root->leafsize,
391                      tree_root->sectorsize, root, fs_info, objectid);
392         ret = btrfs_find_last_root(tree_root, objectid,
393                                    &root->root_item, &root->root_key);
394         BUG_ON(ret);
395
396         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
397         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
398                                      blocksize);
399         BUG_ON(!root->node);
400         return 0;
401 }
402
403 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
404                                                struct btrfs_key *location)
405 {
406         struct btrfs_root *root;
407         struct btrfs_root *tree_root = fs_info->tree_root;
408         struct btrfs_path *path;
409         struct extent_buffer *l;
410         u64 highest_inode;
411         u32 blocksize;
412         int ret = 0;
413
414         root = kzalloc(sizeof(*root), GFP_NOFS);
415         if (!root)
416                 return ERR_PTR(-ENOMEM);
417         if (location->offset == (u64)-1) {
418                 ret = find_and_setup_root(tree_root, fs_info,
419                                           location->objectid, root);
420                 if (ret) {
421                         kfree(root);
422                         return ERR_PTR(ret);
423                 }
424                 goto insert;
425         }
426
427         __setup_root(tree_root->nodesize, tree_root->leafsize,
428                      tree_root->sectorsize, root, fs_info,
429                      location->objectid);
430
431         path = btrfs_alloc_path();
432         BUG_ON(!path);
433         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
434         if (ret != 0) {
435                 if (ret > 0)
436                         ret = -ENOENT;
437                 goto out;
438         }
439         l = path->nodes[0];
440         read_extent_buffer(l, &root->root_item,
441                btrfs_item_ptr_offset(l, path->slots[0]),
442                sizeof(root->root_item));
443         memcpy(&root->root_key, location, sizeof(*location));
444         ret = 0;
445 out:
446         btrfs_release_path(root, path);
447         btrfs_free_path(path);
448         if (ret) {
449                 kfree(root);
450                 return ERR_PTR(ret);
451         }
452         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
453         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
454                                      blocksize);
455         BUG_ON(!root->node);
456 insert:
457         root->ref_cows = 1;
458         ret = btrfs_find_highest_inode(root, &highest_inode);
459         if (ret == 0) {
460                 root->highest_inode = highest_inode;
461                 root->last_inode_alloc = highest_inode;
462         }
463         return root;
464 }
465
466 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
467                                       struct btrfs_key *location,
468                                       const char *name, int namelen)
469 {
470         struct btrfs_root *root;
471         int ret;
472
473         root = radix_tree_lookup(&fs_info->fs_roots_radix,
474                                  (unsigned long)location->objectid);
475         if (root)
476                 return root;
477
478         root = btrfs_read_fs_root_no_radix(fs_info, location);
479         if (IS_ERR(root))
480                 return root;
481         ret = radix_tree_insert(&fs_info->fs_roots_radix,
482                                 (unsigned long)root->root_key.objectid,
483                                 root);
484         if (ret) {
485                 free_extent_buffer(root->node);
486                 kfree(root);
487                 return ERR_PTR(ret);
488         }
489
490         ret = btrfs_set_root_name(root, name, namelen);
491         if (ret) {
492                 free_extent_buffer(root->node);
493                 kfree(root);
494                 return ERR_PTR(ret);
495         }
496
497         ret = btrfs_sysfs_add_root(root);
498         if (ret) {
499                 free_extent_buffer(root->node);
500                 kfree(root->name);
501                 kfree(root);
502                 return ERR_PTR(ret);
503         }
504
505         ret = btrfs_find_dead_roots(fs_info->tree_root,
506                                     root->root_key.objectid, root);
507         BUG_ON(ret);
508
509         return root;
510 }
511 #if 0
512 static int add_hasher(struct btrfs_fs_info *info, char *type) {
513         struct btrfs_hasher *hasher;
514
515         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
516         if (!hasher)
517                 return -ENOMEM;
518         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
519         if (!hasher->hash_tfm) {
520                 kfree(hasher);
521                 return -EINVAL;
522         }
523         spin_lock(&info->hash_lock);
524         list_add(&hasher->list, &info->hashers);
525         spin_unlock(&info->hash_lock);
526         return 0;
527 }
528 #endif
529 struct btrfs_root *open_ctree(struct super_block *sb)
530 {
531         u32 sectorsize;
532         u32 nodesize;
533         u32 leafsize;
534         u32 blocksize;
535         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
536                                                  GFP_NOFS);
537         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
538                                                GFP_NOFS);
539         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
540                                                 GFP_NOFS);
541         int ret;
542         int err = -EIO;
543         struct btrfs_super_block *disk_super;
544
545         if (!extent_root || !tree_root || !fs_info) {
546                 err = -ENOMEM;
547                 goto fail;
548         }
549         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
550         INIT_LIST_HEAD(&fs_info->trans_list);
551         INIT_LIST_HEAD(&fs_info->dead_roots);
552         INIT_LIST_HEAD(&fs_info->hashers);
553         spin_lock_init(&fs_info->hash_lock);
554
555         memset(&fs_info->super_kobj, 0, sizeof(fs_info->super_kobj));
556         init_completion(&fs_info->kobj_unregister);
557         sb_set_blocksize(sb, 4096);
558         fs_info->running_transaction = NULL;
559         fs_info->last_trans_committed = 0;
560         fs_info->tree_root = tree_root;
561         fs_info->extent_root = extent_root;
562         fs_info->sb = sb;
563         fs_info->btree_inode = new_inode(sb);
564         fs_info->btree_inode->i_ino = 1;
565         fs_info->btree_inode->i_nlink = 1;
566         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
567         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
568         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
569                              fs_info->btree_inode->i_mapping,
570                              GFP_NOFS);
571         BTRFS_I(fs_info->btree_inode)->extent_tree.ops = &btree_extent_map_ops;
572
573         extent_map_tree_init(&fs_info->free_space_cache,
574                              fs_info->btree_inode->i_mapping, GFP_NOFS);
575         extent_map_tree_init(&fs_info->block_group_cache,
576                              fs_info->btree_inode->i_mapping, GFP_NOFS);
577         extent_map_tree_init(&fs_info->pinned_extents,
578                              fs_info->btree_inode->i_mapping, GFP_NOFS);
579         extent_map_tree_init(&fs_info->pending_del,
580                              fs_info->btree_inode->i_mapping, GFP_NOFS);
581         extent_map_tree_init(&fs_info->extent_ins,
582                              fs_info->btree_inode->i_mapping, GFP_NOFS);
583         fs_info->do_barriers = 1;
584         fs_info->closing = 0;
585         fs_info->total_pinned = 0;
586
587         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
588         BTRFS_I(fs_info->btree_inode)->root = tree_root;
589         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
590                sizeof(struct btrfs_key));
591         insert_inode_hash(fs_info->btree_inode);
592         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
593
594         mutex_init(&fs_info->trans_mutex);
595         mutex_init(&fs_info->fs_mutex);
596
597 #if 0
598         ret = add_hasher(fs_info, "crc32c");
599         if (ret) {
600                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
601                 err = -ENOMEM;
602                 goto fail_iput;
603         }
604 #endif
605         __setup_root(512, 512, 512, tree_root,
606                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
607
608         fs_info->sb_buffer = read_tree_block(tree_root,
609                                              BTRFS_SUPER_INFO_OFFSET,
610                                              512);
611
612         if (!fs_info->sb_buffer)
613                 goto fail_iput;
614
615         read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
616                            sizeof(fs_info->super_copy));
617
618         read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
619                            (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
620                            BTRFS_FSID_SIZE);
621         disk_super = &fs_info->super_copy;
622         if (!btrfs_super_root(disk_super))
623                 goto fail_sb_buffer;
624
625         nodesize = btrfs_super_nodesize(disk_super);
626         leafsize = btrfs_super_leafsize(disk_super);
627         sectorsize = btrfs_super_sectorsize(disk_super);
628         tree_root->nodesize = nodesize;
629         tree_root->leafsize = leafsize;
630         tree_root->sectorsize = sectorsize;
631         sb_set_blocksize(sb, sectorsize);
632
633         i_size_write(fs_info->btree_inode,
634                      btrfs_super_total_bytes(disk_super));
635
636         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
637                     sizeof(disk_super->magic))) {
638                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
639                 goto fail_sb_buffer;
640         }
641
642         blocksize = btrfs_level_size(tree_root,
643                                      btrfs_super_root_level(disk_super));
644
645         tree_root->node = read_tree_block(tree_root,
646                                           btrfs_super_root(disk_super),
647                                           blocksize);
648         if (!tree_root->node)
649                 goto fail_sb_buffer;
650
651         mutex_lock(&fs_info->fs_mutex);
652
653         ret = find_and_setup_root(tree_root, fs_info,
654                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
655         if (ret) {
656                 mutex_unlock(&fs_info->fs_mutex);
657                 goto fail_tree_root;
658         }
659
660         btrfs_read_block_groups(extent_root);
661
662         fs_info->generation = btrfs_super_generation(disk_super) + 1;
663         mutex_unlock(&fs_info->fs_mutex);
664         return tree_root;
665
666 fail_tree_root:
667         free_extent_buffer(tree_root->node);
668 fail_sb_buffer:
669         free_extent_buffer(fs_info->sb_buffer);
670 fail_iput:
671         iput(fs_info->btree_inode);
672 fail:
673         kfree(extent_root);
674         kfree(tree_root);
675         kfree(fs_info);
676         return ERR_PTR(err);
677 }
678
679 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
680                       *root)
681 {
682         int ret;
683         struct extent_buffer *super = root->fs_info->sb_buffer;
684         struct inode *btree_inode = root->fs_info->btree_inode;
685
686         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, super);
687         ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
688                                      super->start, super->len);
689         return ret;
690 }
691
692 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
693 {
694         radix_tree_delete(&fs_info->fs_roots_radix,
695                           (unsigned long)root->root_key.objectid);
696         btrfs_sysfs_del_root(root);
697         if (root->inode)
698                 iput(root->inode);
699         if (root->node)
700                 free_extent_buffer(root->node);
701         if (root->commit_root)
702                 free_extent_buffer(root->commit_root);
703         if (root->name)
704                 kfree(root->name);
705         kfree(root);
706         return 0;
707 }
708
709 static int del_fs_roots(struct btrfs_fs_info *fs_info)
710 {
711         int ret;
712         struct btrfs_root *gang[8];
713         int i;
714
715         while(1) {
716                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
717                                              (void **)gang, 0,
718                                              ARRAY_SIZE(gang));
719                 if (!ret)
720                         break;
721                 for (i = 0; i < ret; i++)
722                         btrfs_free_fs_root(fs_info, gang[i]);
723         }
724         return 0;
725 }
726
727 int close_ctree(struct btrfs_root *root)
728 {
729         int ret;
730         struct btrfs_trans_handle *trans;
731         struct btrfs_fs_info *fs_info = root->fs_info;
732
733         fs_info->closing = 1;
734         btrfs_transaction_flush_work(root);
735         mutex_lock(&fs_info->fs_mutex);
736         btrfs_defrag_dirty_roots(root->fs_info);
737         trans = btrfs_start_transaction(root, 1);
738         ret = btrfs_commit_transaction(trans, root);
739         /* run commit again to  drop the original snapshot */
740         trans = btrfs_start_transaction(root, 1);
741         btrfs_commit_transaction(trans, root);
742         ret = btrfs_write_and_wait_transaction(NULL, root);
743         BUG_ON(ret);
744         write_ctree_super(NULL, root);
745         mutex_unlock(&fs_info->fs_mutex);
746
747         if (fs_info->extent_root->node)
748                 free_extent_buffer(fs_info->extent_root->node);
749
750         if (fs_info->tree_root->node)
751                 free_extent_buffer(fs_info->tree_root->node);
752
753         free_extent_buffer(fs_info->sb_buffer);
754
755         btrfs_free_block_groups(root->fs_info);
756         del_fs_roots(fs_info);
757         extent_map_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->extent_tree);
758         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
759         iput(fs_info->btree_inode);
760 #if 0
761         while(!list_empty(&fs_info->hashers)) {
762                 struct btrfs_hasher *hasher;
763                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
764                                     hashers);
765                 list_del(&hasher->hashers);
766                 crypto_free_hash(&fs_info->hash_tfm);
767                 kfree(hasher);
768         }
769 #endif
770         kfree(fs_info->extent_root);
771         kfree(fs_info->tree_root);
772         return 0;
773 }
774
775 int btrfs_buffer_uptodate(struct extent_buffer *buf)
776 {
777         struct inode *btree_inode = buf->first_page->mapping->host;
778         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->extent_tree, buf);
779 }
780
781 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
782 {
783         struct inode *btree_inode = buf->first_page->mapping->host;
784         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->extent_tree,
785                                           buf);
786 }
787
788 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
789 {
790         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
791         u64 transid = btrfs_header_generation(buf);
792         struct inode *btree_inode = root->fs_info->btree_inode;
793
794         if (transid != root->fs_info->generation) {
795                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
796                         (unsigned long long)buf->start,
797                         transid, root->fs_info->generation);
798                 WARN_ON(1);
799         }
800         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, buf);
801 }
802
803 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
804 {
805         balance_dirty_pages_ratelimited_nr(
806                         root->fs_info->btree_inode->i_mapping, 1);
807 }
808
809 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
810 {
811         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
812         struct inode *btree_inode = root->fs_info->btree_inode;
813         set_extent_bits(&BTRFS_I(btree_inode)->extent_tree, buf->start,
814                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
815 }
816
817 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
818 {
819         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
820         struct inode *btree_inode = root->fs_info->btree_inode;
821         set_extent_bits(&BTRFS_I(btree_inode)->extent_tree, buf->start,
822                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
823                         GFP_NOFS);
824 }
825
826 int btrfs_buffer_defrag(struct extent_buffer *buf)
827 {
828         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
829         struct inode *btree_inode = root->fs_info->btree_inode;
830         return test_range_bit(&BTRFS_I(btree_inode)->extent_tree,
831                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
832 }
833
834 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
835 {
836         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
837         struct inode *btree_inode = root->fs_info->btree_inode;
838         return test_range_bit(&BTRFS_I(btree_inode)->extent_tree,
839                      buf->start, buf->start + buf->len - 1,
840                      EXTENT_DEFRAG_DONE, 0);
841 }
842
843 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
844 {
845         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
846         struct inode *btree_inode = root->fs_info->btree_inode;
847         return clear_extent_bits(&BTRFS_I(btree_inode)->extent_tree,
848                      buf->start, buf->start + buf->len - 1,
849                      EXTENT_DEFRAG_DONE, GFP_NOFS);
850 }
851
852 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
853 {
854         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
855         struct inode *btree_inode = root->fs_info->btree_inode;
856         return clear_extent_bits(&BTRFS_I(btree_inode)->extent_tree,
857                      buf->start, buf->start + buf->len - 1,
858                      EXTENT_DEFRAG, GFP_NOFS);
859 }
860
861 int btrfs_read_buffer(struct extent_buffer *buf)
862 {
863         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
864         struct inode *btree_inode = root->fs_info->btree_inode;
865         return read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
866                                         buf, 0, 1);
867 }
868
869 static struct extent_map_ops btree_extent_map_ops = {
870         .writepage_io_hook = btree_writepage_io_hook,
871 };