Btrfs: Add more synchronization before creating a snapshot
[linux-2.6-block.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 "ctree.h"
27 #include "disk-io.h"
28 #include "transaction.h"
29 #include "btrfs_inode.h"
30
31 u64 bh_blocknr(struct buffer_head *bh)
32 {
33         return bh->b_blocknr;
34 }
35
36 static int check_tree_block(struct btrfs_root *root, struct buffer_head *buf)
37 {
38         struct btrfs_node *node = btrfs_buffer_node(buf);
39         if (bh_blocknr(buf) != btrfs_header_blocknr(&node->header)) {
40                 printk(KERN_CRIT "bh_blocknr(buf) is %llu, header is %llu\n",
41                        (unsigned long long)bh_blocknr(buf),
42                        (unsigned long long)btrfs_header_blocknr(&node->header));
43                 return 1;
44         }
45         return 0;
46 }
47
48 struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr)
49 {
50         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
51         int blockbits = root->fs_info->sb->s_blocksize_bits;
52         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
53         struct page *page;
54         struct buffer_head *bh;
55         struct buffer_head *head;
56         struct buffer_head *ret = NULL;
57
58
59         page = find_lock_page(mapping, index);
60         if (!page)
61                 return NULL;
62
63         if (!page_has_buffers(page))
64                 goto out_unlock;
65
66         head = page_buffers(page);
67         bh = head;
68         do {
69                 if (buffer_mapped(bh) && bh_blocknr(bh) == blocknr) {
70                         ret = bh;
71                         get_bh(bh);
72                         goto out_unlock;
73                 }
74                 bh = bh->b_this_page;
75         } while (bh != head);
76 out_unlock:
77         unlock_page(page);
78         page_cache_release(page);
79         return ret;
80 }
81
82 int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
83                              u64 logical)
84 {
85         if (logical == 0) {
86                 bh->b_bdev = NULL;
87                 bh->b_blocknr = 0;
88                 set_buffer_mapped(bh);
89         } else {
90                 map_bh(bh, root->fs_info->sb, logical);
91         }
92         return 0;
93 }
94
95 struct buffer_head *btrfs_find_create_tree_block(struct btrfs_root *root,
96                                                  u64 blocknr)
97 {
98         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
99         int blockbits = root->fs_info->sb->s_blocksize_bits;
100         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
101         struct page *page;
102         struct buffer_head *bh;
103         struct buffer_head *head;
104         struct buffer_head *ret = NULL;
105         int err;
106         u64 first_block = index << (PAGE_CACHE_SHIFT - blockbits);
107
108         page = find_or_create_page(mapping, index, GFP_NOFS);
109         if (!page)
110                 return NULL;
111
112         if (!page_has_buffers(page))
113                 create_empty_buffers(page, root->fs_info->sb->s_blocksize, 0);
114         head = page_buffers(page);
115         bh = head;
116         do {
117                 if (!buffer_mapped(bh)) {
118                         err = btrfs_map_bh_to_logical(root, bh, first_block);
119                         BUG_ON(err);
120                 }
121                 if (bh_blocknr(bh) == blocknr) {
122                         ret = bh;
123                         get_bh(bh);
124                         goto out_unlock;
125                 }
126                 bh = bh->b_this_page;
127                 first_block++;
128         } while (bh != head);
129 out_unlock:
130         unlock_page(page);
131         if (ret)
132                 touch_buffer(ret);
133         page_cache_release(page);
134         return ret;
135 }
136
137 static int btree_get_block(struct inode *inode, sector_t iblock,
138                            struct buffer_head *bh, int create)
139 {
140         int err;
141         struct btrfs_root *root = BTRFS_I(bh->b_page->mapping->host)->root;
142         err = btrfs_map_bh_to_logical(root, bh, iblock);
143         return err;
144 }
145
146 int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
147                     char *result)
148 {
149         u32 crc;
150         crc = crc32c(0, data, len);
151         memcpy(result, &crc, BTRFS_CRC32_SIZE);
152         return 0;
153 }
154
155 static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
156                            int verify)
157 {
158         char result[BTRFS_CRC32_SIZE];
159         int ret;
160         struct btrfs_node *node;
161
162         ret = btrfs_csum_data(root, bh->b_data + BTRFS_CSUM_SIZE,
163                               bh->b_size - BTRFS_CSUM_SIZE, result);
164         if (ret)
165                 return ret;
166         if (verify) {
167                 if (memcmp(bh->b_data, result, BTRFS_CRC32_SIZE)) {
168                         printk("btrfs: %s checksum verify failed on %llu\n",
169                                root->fs_info->sb->s_id,
170                                (unsigned long long)bh_blocknr(bh));
171                         return 1;
172                 }
173         } else {
174                 node = btrfs_buffer_node(bh);
175                 memcpy(node->header.csum, result, BTRFS_CRC32_SIZE);
176         }
177         return 0;
178 }
179
180 static int btree_writepage(struct page *page, struct writeback_control *wbc)
181 {
182         struct buffer_head *bh;
183         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
184         struct buffer_head *head;
185         if (!page_has_buffers(page)) {
186                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
187                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
188         }
189         head = page_buffers(page);
190         bh = head;
191         do {
192                 if (buffer_dirty(bh))
193                         csum_tree_block(root, bh, 0);
194                 bh = bh->b_this_page;
195         } while (bh != head);
196         return block_write_full_page(page, btree_get_block, wbc);
197 }
198
199 static int btree_readpage(struct file * file, struct page * page)
200 {
201         return block_read_full_page(page, btree_get_block);
202 }
203
204 static struct address_space_operations btree_aops = {
205         .readpage       = btree_readpage,
206         .writepage      = btree_writepage,
207         .sync_page      = block_sync_page,
208 };
209
210 int readahead_tree_block(struct btrfs_root *root, u64 blocknr)
211 {
212         struct buffer_head *bh = NULL;
213         int ret = 0;
214
215         bh = btrfs_find_create_tree_block(root, blocknr);
216         if (!bh)
217                 return 0;
218         if (buffer_uptodate(bh)) {
219                 ret = 1;
220                 goto done;
221         }
222         if (test_set_buffer_locked(bh)) {
223                 ret = 1;
224                 goto done;
225         }
226         if (!buffer_uptodate(bh)) {
227                 get_bh(bh);
228                 bh->b_end_io = end_buffer_read_sync;
229                 submit_bh(READ, bh);
230         } else {
231                 unlock_buffer(bh);
232                 ret = 1;
233         }
234 done:
235         brelse(bh);
236         return ret;
237 }
238
239 struct buffer_head *read_tree_block(struct btrfs_root *root, u64 blocknr)
240 {
241         struct buffer_head *bh = NULL;
242
243         bh = btrfs_find_create_tree_block(root, blocknr);
244         if (!bh)
245                 return bh;
246         if (buffer_uptodate(bh))
247                 goto uptodate;
248         lock_buffer(bh);
249         if (!buffer_uptodate(bh)) {
250                 get_bh(bh);
251                 bh->b_end_io = end_buffer_read_sync;
252                 submit_bh(READ, bh);
253                 wait_on_buffer(bh);
254                 if (!buffer_uptodate(bh))
255                         goto fail;
256         } else {
257                 unlock_buffer(bh);
258         }
259 uptodate:
260         if (!buffer_checked(bh)) {
261                 csum_tree_block(root, bh, 1);
262                 set_buffer_checked(bh);
263         }
264         if (check_tree_block(root, bh))
265                 goto fail;
266         return bh;
267 fail:
268         brelse(bh);
269         return NULL;
270 }
271
272 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
273                      struct buffer_head *buf)
274 {
275         WARN_ON(atomic_read(&buf->b_count) == 0);
276         lock_buffer(buf);
277         clear_buffer_dirty(buf);
278         unlock_buffer(buf);
279         return 0;
280 }
281
282 static int __setup_root(int blocksize,
283                         struct btrfs_root *root,
284                         struct btrfs_fs_info *fs_info,
285                         u64 objectid)
286 {
287         root->node = NULL;
288         root->inode = NULL;
289         root->commit_root = NULL;
290         root->blocksize = blocksize;
291         root->ref_cows = 0;
292         root->fs_info = fs_info;
293         root->objectid = objectid;
294         root->last_trans = 0;
295         root->highest_inode = 0;
296         root->last_inode_alloc = 0;
297         root->name = NULL;
298         memset(&root->root_key, 0, sizeof(root->root_key));
299         memset(&root->root_item, 0, sizeof(root->root_item));
300         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
301         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
302         init_completion(&root->kobj_unregister);
303         init_rwsem(&root->snap_sem);
304         root->defrag_running = 0;
305         root->defrag_level = 0;
306         root->root_key.objectid = objectid;
307         return 0;
308 }
309
310 static int find_and_setup_root(int blocksize,
311                                struct btrfs_root *tree_root,
312                                struct btrfs_fs_info *fs_info,
313                                u64 objectid,
314                                struct btrfs_root *root)
315 {
316         int ret;
317
318         __setup_root(blocksize, root, fs_info, objectid);
319         ret = btrfs_find_last_root(tree_root, objectid,
320                                    &root->root_item, &root->root_key);
321         BUG_ON(ret);
322
323         root->node = read_tree_block(root,
324                                      btrfs_root_blocknr(&root->root_item));
325         BUG_ON(!root->node);
326         return 0;
327 }
328
329 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
330                                                struct btrfs_key *location)
331 {
332         struct btrfs_root *root;
333         struct btrfs_root *tree_root = fs_info->tree_root;
334         struct btrfs_path *path;
335         struct btrfs_leaf *l;
336         u64 highest_inode;
337         int ret = 0;
338
339         root = kzalloc(sizeof(*root), GFP_NOFS);
340         if (!root)
341                 return ERR_PTR(-ENOMEM);
342         if (location->offset == (u64)-1) {
343                 ret = find_and_setup_root(fs_info->sb->s_blocksize,
344                                           fs_info->tree_root, fs_info,
345                                           location->objectid, root);
346                 if (ret) {
347                         kfree(root);
348                         return ERR_PTR(ret);
349                 }
350                 goto insert;
351         }
352
353         __setup_root(fs_info->sb->s_blocksize, root, fs_info,
354                      location->objectid);
355
356         path = btrfs_alloc_path();
357         BUG_ON(!path);
358         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
359         if (ret != 0) {
360                 if (ret > 0)
361                         ret = -ENOENT;
362                 goto out;
363         }
364         l = btrfs_buffer_leaf(path->nodes[0]);
365         memcpy(&root->root_item,
366                btrfs_item_ptr(l, path->slots[0], struct btrfs_root_item),
367                sizeof(root->root_item));
368         memcpy(&root->root_key, location, sizeof(*location));
369         ret = 0;
370 out:
371         btrfs_release_path(root, path);
372         btrfs_free_path(path);
373         if (ret) {
374                 kfree(root);
375                 return ERR_PTR(ret);
376         }
377         root->node = read_tree_block(root,
378                                      btrfs_root_blocknr(&root->root_item));
379         BUG_ON(!root->node);
380 insert:
381         root->ref_cows = 1;
382         ret = btrfs_find_highest_inode(root, &highest_inode);
383         if (ret == 0) {
384                 root->highest_inode = highest_inode;
385                 root->last_inode_alloc = highest_inode;
386         }
387         return root;
388 }
389
390 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
391                                       struct btrfs_key *location,
392                                       const char *name, int namelen)
393 {
394         struct btrfs_root *root;
395         int ret;
396
397         root = radix_tree_lookup(&fs_info->fs_roots_radix,
398                                  (unsigned long)location->objectid);
399         if (root)
400                 return root;
401
402         root = btrfs_read_fs_root_no_radix(fs_info, location);
403         if (IS_ERR(root))
404                 return root;
405         ret = radix_tree_insert(&fs_info->fs_roots_radix,
406                                 (unsigned long)root->root_key.objectid,
407                                 root);
408         if (ret) {
409                 brelse(root->node);
410                 kfree(root);
411                 return ERR_PTR(ret);
412         }
413
414         ret = btrfs_set_root_name(root, name, namelen);
415         if (ret) {
416                 brelse(root->node);
417                 kfree(root);
418                 return ERR_PTR(ret);
419         }
420
421         ret = btrfs_sysfs_add_root(root);
422         if (ret) {
423                 brelse(root->node);
424                 kfree(root->name);
425                 kfree(root);
426                 return ERR_PTR(ret);
427         }
428
429         return root;
430 }
431
432 struct btrfs_root *open_ctree(struct super_block *sb)
433 {
434         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
435                                                  GFP_NOFS);
436         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
437                                                GFP_NOFS);
438         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
439                                                 GFP_NOFS);
440         int ret;
441         int err = -EIO;
442         struct btrfs_super_block *disk_super;
443
444         if (!extent_root || !tree_root || !fs_info) {
445                 err = -ENOMEM;
446                 goto fail;
447         }
448         init_bit_radix(&fs_info->pinned_radix);
449         init_bit_radix(&fs_info->pending_del_radix);
450         init_bit_radix(&fs_info->extent_map_radix);
451         init_bit_radix(&fs_info->extent_ins_radix);
452         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
453         INIT_RADIX_TREE(&fs_info->block_group_radix, GFP_KERNEL);
454         INIT_RADIX_TREE(&fs_info->block_group_data_radix, GFP_KERNEL);
455         INIT_LIST_HEAD(&fs_info->trans_list);
456         INIT_LIST_HEAD(&fs_info->dead_roots);
457         memset(&fs_info->super_kobj, 0, sizeof(fs_info->super_kobj));
458         init_completion(&fs_info->kobj_unregister);
459         sb_set_blocksize(sb, 4096);
460         fs_info->running_transaction = NULL;
461         fs_info->last_trans_committed = 0;
462         fs_info->tree_root = tree_root;
463         fs_info->extent_root = extent_root;
464         fs_info->sb = sb;
465         fs_info->btree_inode = new_inode(sb);
466         fs_info->btree_inode->i_ino = 1;
467         fs_info->btree_inode->i_nlink = 1;
468         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
469         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
470         fs_info->do_barriers = 1;
471         fs_info->closing = 0;
472
473         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
474         BTRFS_I(fs_info->btree_inode)->root = tree_root;
475         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
476                sizeof(struct btrfs_key));
477         insert_inode_hash(fs_info->btree_inode);
478         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
479
480         mutex_init(&fs_info->trans_mutex);
481         mutex_init(&fs_info->fs_mutex);
482
483         __setup_root(sb->s_blocksize, tree_root,
484                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
485
486         fs_info->sb_buffer = read_tree_block(tree_root,
487                                              BTRFS_SUPER_INFO_OFFSET /
488                                              sb->s_blocksize);
489
490         if (!fs_info->sb_buffer)
491                 goto fail_iput;
492         disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
493         fs_info->disk_super = disk_super;
494         memcpy(&fs_info->super_copy, disk_super, sizeof(fs_info->super_copy));
495
496         if (!btrfs_super_root(disk_super))
497                 goto fail_sb_buffer;
498
499         i_size_write(fs_info->btree_inode,
500                      btrfs_super_total_blocks(disk_super) <<
501                      fs_info->btree_inode->i_blkbits);
502
503
504         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
505                     sizeof(disk_super->magic))) {
506                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
507                 goto fail_sb_buffer;
508         }
509         tree_root->node = read_tree_block(tree_root,
510                                           btrfs_super_root(disk_super));
511         if (!tree_root->node)
512                 goto fail_sb_buffer;
513
514         mutex_lock(&fs_info->fs_mutex);
515         ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
516                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
517         if (ret) {
518                 mutex_unlock(&fs_info->fs_mutex);
519                 goto fail_tree_root;
520         }
521
522         btrfs_read_block_groups(extent_root);
523
524         fs_info->generation = btrfs_super_generation(disk_super) + 1;
525         ret = btrfs_find_dead_roots(tree_root);
526         if (ret) {
527                 mutex_unlock(&fs_info->fs_mutex);
528                 goto fail_tree_root;
529         }
530         mutex_unlock(&fs_info->fs_mutex);
531         return tree_root;
532
533 fail_tree_root:
534         btrfs_block_release(tree_root, tree_root->node);
535 fail_sb_buffer:
536         btrfs_block_release(tree_root, fs_info->sb_buffer);
537 fail_iput:
538         iput(fs_info->btree_inode);
539 fail:
540         kfree(extent_root);
541         kfree(tree_root);
542         kfree(fs_info);
543         return ERR_PTR(err);
544 }
545
546 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
547                       *root)
548 {
549         int ret;
550         struct buffer_head *bh = root->fs_info->sb_buffer;
551
552         lock_buffer(bh);
553         WARN_ON(atomic_read(&bh->b_count) < 1);
554         clear_buffer_dirty(bh);
555         csum_tree_block(root, bh, 0);
556         bh->b_end_io = end_buffer_write_sync;
557         get_bh(bh);
558         if (root->fs_info->do_barriers)
559                 ret = submit_bh(WRITE_BARRIER, bh);
560         else
561                 ret = submit_bh(WRITE, bh);
562         if (ret == -EOPNOTSUPP) {
563                 get_bh(bh);
564                 lock_buffer(bh);
565                 set_buffer_uptodate(bh);
566                 root->fs_info->do_barriers = 0;
567                 ret = submit_bh(WRITE, bh);
568         }
569         wait_on_buffer(bh);
570         if (!buffer_uptodate(bh)) {
571                 WARN_ON(1);
572                 return -EIO;
573         }
574         return 0;
575 }
576
577 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
578 {
579         radix_tree_delete(&fs_info->fs_roots_radix,
580                           (unsigned long)root->root_key.objectid);
581         btrfs_sysfs_del_root(root);
582         if (root->inode)
583                 iput(root->inode);
584         if (root->node)
585                 brelse(root->node);
586         if (root->commit_root)
587                 brelse(root->commit_root);
588         if (root->name)
589                 kfree(root->name);
590         kfree(root);
591         return 0;
592 }
593
594 static int del_fs_roots(struct btrfs_fs_info *fs_info)
595 {
596         int ret;
597         struct btrfs_root *gang[8];
598         int i;
599
600         while(1) {
601                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
602                                              (void **)gang, 0,
603                                              ARRAY_SIZE(gang));
604                 if (!ret)
605                         break;
606                 for (i = 0; i < ret; i++)
607                         btrfs_free_fs_root(fs_info, gang[i]);
608         }
609         return 0;
610 }
611
612 int close_ctree(struct btrfs_root *root)
613 {
614         int ret;
615         struct btrfs_trans_handle *trans;
616         struct btrfs_fs_info *fs_info = root->fs_info;
617
618         fs_info->closing = 1;
619         btrfs_transaction_flush_work(root);
620         mutex_lock(&fs_info->fs_mutex);
621         btrfs_defrag_dirty_roots(root->fs_info);
622         trans = btrfs_start_transaction(root, 1);
623         ret = btrfs_commit_transaction(trans, root);
624         /* run commit again to  drop the original snapshot */
625         trans = btrfs_start_transaction(root, 1);
626         btrfs_commit_transaction(trans, root);
627         ret = btrfs_write_and_wait_transaction(NULL, root);
628         BUG_ON(ret);
629         write_ctree_super(NULL, root);
630         mutex_unlock(&fs_info->fs_mutex);
631
632         if (fs_info->extent_root->node)
633                 btrfs_block_release(fs_info->extent_root,
634                                     fs_info->extent_root->node);
635         if (fs_info->tree_root->node)
636                 btrfs_block_release(fs_info->tree_root,
637                                     fs_info->tree_root->node);
638         btrfs_block_release(root, fs_info->sb_buffer);
639         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
640         iput(fs_info->btree_inode);
641
642         btrfs_free_block_groups(root->fs_info);
643         del_fs_roots(fs_info);
644         kfree(fs_info->extent_root);
645         kfree(fs_info->tree_root);
646         return 0;
647 }
648
649 void btrfs_mark_buffer_dirty(struct buffer_head *bh)
650 {
651         struct btrfs_root *root = BTRFS_I(bh->b_page->mapping->host)->root;
652         u64 transid = btrfs_header_generation(btrfs_buffer_header(bh));
653
654         WARN_ON(!atomic_read(&bh->b_count));
655
656         if (transid != root->fs_info->generation) {
657                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
658                         (unsigned long long)bh->b_blocknr,
659                         transid, root->fs_info->generation);
660                 WARN_ON(1);
661         }
662         mark_buffer_dirty(bh);
663 }
664
665 void btrfs_block_release(struct btrfs_root *root, struct buffer_head *buf)
666 {
667         brelse(buf);
668 }
669
670 void btrfs_btree_balance_dirty(struct btrfs_root *root)
671 {
672         balance_dirty_pages_ratelimited(root->fs_info->btree_inode->i_mapping);
673 }