Btrfs: Add run time btree defrag, and an ioctl to force btree defrag
[firefly-linux-kernel-4.4.55.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         memset(&root->root_key, 0, sizeof(root->root_key));
298         memset(&root->root_item, 0, sizeof(root->root_item));
299         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
300         root->defrag_running = 0;
301         root->defrag_level = 0;
302         root->root_key.objectid = objectid;
303         return 0;
304 }
305
306 static int find_and_setup_root(int blocksize,
307                                struct btrfs_root *tree_root,
308                                struct btrfs_fs_info *fs_info,
309                                u64 objectid,
310                                struct btrfs_root *root)
311 {
312         int ret;
313
314         __setup_root(blocksize, root, fs_info, objectid);
315         ret = btrfs_find_last_root(tree_root, objectid,
316                                    &root->root_item, &root->root_key);
317         BUG_ON(ret);
318
319         root->node = read_tree_block(root,
320                                      btrfs_root_blocknr(&root->root_item));
321         BUG_ON(!root->node);
322         return 0;
323 }
324
325 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
326                                                struct btrfs_key *location)
327 {
328         struct btrfs_root *root;
329         struct btrfs_root *tree_root = fs_info->tree_root;
330         struct btrfs_path *path;
331         struct btrfs_leaf *l;
332         u64 highest_inode;
333         int ret = 0;
334
335         root = kzalloc(sizeof(*root), GFP_NOFS);
336         if (!root)
337                 return ERR_PTR(-ENOMEM);
338         if (location->offset == (u64)-1) {
339                 ret = find_and_setup_root(fs_info->sb->s_blocksize,
340                                           fs_info->tree_root, fs_info,
341                                           location->objectid, root);
342                 if (ret) {
343                         kfree(root);
344                         return ERR_PTR(ret);
345                 }
346                 goto insert;
347         }
348
349         __setup_root(fs_info->sb->s_blocksize, root, fs_info,
350                      location->objectid);
351
352         path = btrfs_alloc_path();
353         BUG_ON(!path);
354         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
355         if (ret != 0) {
356                 if (ret > 0)
357                         ret = -ENOENT;
358                 goto out;
359         }
360         l = btrfs_buffer_leaf(path->nodes[0]);
361         memcpy(&root->root_item,
362                btrfs_item_ptr(l, path->slots[0], struct btrfs_root_item),
363                sizeof(root->root_item));
364         memcpy(&root->root_key, location, sizeof(*location));
365         ret = 0;
366 out:
367         btrfs_release_path(root, path);
368         btrfs_free_path(path);
369         if (ret) {
370                 kfree(root);
371                 return ERR_PTR(ret);
372         }
373         root->node = read_tree_block(root,
374                                      btrfs_root_blocknr(&root->root_item));
375         BUG_ON(!root->node);
376 insert:
377         root->ref_cows = 1;
378         ret = btrfs_find_highest_inode(root, &highest_inode);
379         if (ret == 0) {
380                 root->highest_inode = highest_inode;
381                 root->last_inode_alloc = highest_inode;
382         }
383         return root;
384 }
385
386 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
387                                       struct btrfs_key *location)
388 {
389         struct btrfs_root *root;
390         int ret;
391
392         root = radix_tree_lookup(&fs_info->fs_roots_radix,
393                                  (unsigned long)location->objectid);
394         if (root)
395                 return root;
396
397         root = btrfs_read_fs_root_no_radix(fs_info, location);
398         if (IS_ERR(root))
399                 return root;
400         ret = radix_tree_insert(&fs_info->fs_roots_radix,
401                                 (unsigned long)root->root_key.objectid,
402                                 root);
403         if (ret) {
404                 brelse(root->node);
405                 kfree(root);
406                 return ERR_PTR(ret);
407         }
408         return root;
409 }
410
411 struct btrfs_root *open_ctree(struct super_block *sb)
412 {
413         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
414                                                  GFP_NOFS);
415         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
416                                                GFP_NOFS);
417         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
418                                                 GFP_NOFS);
419         int ret;
420         int err = -EIO;
421         struct btrfs_super_block *disk_super;
422
423         if (!extent_root || !tree_root || !fs_info) {
424                 err = -ENOMEM;
425                 goto fail;
426         }
427         init_bit_radix(&fs_info->pinned_radix);
428         init_bit_radix(&fs_info->pending_del_radix);
429         init_bit_radix(&fs_info->extent_map_radix);
430         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
431         INIT_RADIX_TREE(&fs_info->block_group_radix, GFP_KERNEL);
432         INIT_RADIX_TREE(&fs_info->block_group_data_radix, GFP_KERNEL);
433         INIT_LIST_HEAD(&fs_info->trans_list);
434         INIT_LIST_HEAD(&fs_info->dead_roots);
435         sb_set_blocksize(sb, 4096);
436         fs_info->running_transaction = NULL;
437         fs_info->tree_root = tree_root;
438         fs_info->extent_root = extent_root;
439         fs_info->sb = sb;
440         fs_info->btree_inode = new_inode(sb);
441         fs_info->btree_inode->i_ino = 1;
442         fs_info->btree_inode->i_nlink = 1;
443         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
444         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
445         fs_info->do_barriers = 1;
446         fs_info->extent_tree_insert_nr = 0;
447         fs_info->extent_tree_prealloc_nr = 0;
448         fs_info->closing = 0;
449
450         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
451         BTRFS_I(fs_info->btree_inode)->root = tree_root;
452         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
453                sizeof(struct btrfs_key));
454         insert_inode_hash(fs_info->btree_inode);
455         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
456
457         mutex_init(&fs_info->trans_mutex);
458         mutex_init(&fs_info->fs_mutex);
459
460         __setup_root(sb->s_blocksize, tree_root,
461                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
462
463         fs_info->sb_buffer = read_tree_block(tree_root,
464                                              BTRFS_SUPER_INFO_OFFSET /
465                                              sb->s_blocksize);
466
467         if (!fs_info->sb_buffer)
468                 goto fail_iput;
469         disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
470         fs_info->disk_super = disk_super;
471         memcpy(&fs_info->super_copy, disk_super, sizeof(fs_info->super_copy));
472
473         if (!btrfs_super_root(disk_super))
474                 goto fail_sb_buffer;
475
476         i_size_write(fs_info->btree_inode,
477                      btrfs_super_total_blocks(disk_super) <<
478                      fs_info->btree_inode->i_blkbits);
479
480
481         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
482                     sizeof(disk_super->magic))) {
483                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
484                 goto fail_sb_buffer;
485         }
486         tree_root->node = read_tree_block(tree_root,
487                                           btrfs_super_root(disk_super));
488         if (!tree_root->node)
489                 goto fail_sb_buffer;
490
491         mutex_lock(&fs_info->fs_mutex);
492         ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
493                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
494         if (ret) {
495                 mutex_unlock(&fs_info->fs_mutex);
496                 goto fail_tree_root;
497         }
498
499         btrfs_read_block_groups(extent_root);
500
501         fs_info->generation = btrfs_super_generation(disk_super) + 1;
502         ret = btrfs_find_dead_roots(tree_root);
503         if (ret)
504                 goto fail_tree_root;
505         mutex_unlock(&fs_info->fs_mutex);
506         return tree_root;
507
508 fail_tree_root:
509         btrfs_block_release(tree_root, tree_root->node);
510 fail_sb_buffer:
511         btrfs_block_release(tree_root, fs_info->sb_buffer);
512 fail_iput:
513         iput(fs_info->btree_inode);
514 fail:
515         kfree(extent_root);
516         kfree(tree_root);
517         kfree(fs_info);
518         return ERR_PTR(err);
519 }
520
521 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
522                       *root)
523 {
524         int ret;
525         struct buffer_head *bh = root->fs_info->sb_buffer;
526
527         lock_buffer(bh);
528         WARN_ON(atomic_read(&bh->b_count) < 1);
529         clear_buffer_dirty(bh);
530         csum_tree_block(root, bh, 0);
531         bh->b_end_io = end_buffer_write_sync;
532         get_bh(bh);
533         if (root->fs_info->do_barriers)
534                 ret = submit_bh(WRITE_BARRIER, bh);
535         else
536                 ret = submit_bh(WRITE, bh);
537         if (ret == -EOPNOTSUPP) {
538                 get_bh(bh);
539                 lock_buffer(bh);
540                 set_buffer_uptodate(bh);
541                 root->fs_info->do_barriers = 0;
542                 ret = submit_bh(WRITE, bh);
543         }
544         wait_on_buffer(bh);
545         if (!buffer_uptodate(bh)) {
546                 WARN_ON(1);
547                 return -EIO;
548         }
549         return 0;
550 }
551
552 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
553 {
554         radix_tree_delete(&fs_info->fs_roots_radix,
555                           (unsigned long)root->root_key.objectid);
556         if (root->inode)
557                 iput(root->inode);
558         if (root->node)
559                 brelse(root->node);
560         if (root->commit_root)
561                 brelse(root->commit_root);
562         kfree(root);
563         return 0;
564 }
565
566 static int del_fs_roots(struct btrfs_fs_info *fs_info)
567 {
568         int ret;
569         struct btrfs_root *gang[8];
570         int i;
571
572         while(1) {
573                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
574                                              (void **)gang, 0,
575                                              ARRAY_SIZE(gang));
576                 if (!ret)
577                         break;
578                 for (i = 0; i < ret; i++)
579                         btrfs_free_fs_root(fs_info, gang[i]);
580         }
581         return 0;
582 }
583
584 int close_ctree(struct btrfs_root *root)
585 {
586         int ret;
587         struct btrfs_trans_handle *trans;
588         struct btrfs_fs_info *fs_info = root->fs_info;
589
590         fs_info->closing = 1;
591         btrfs_transaction_flush_work(root);
592         mutex_lock(&fs_info->fs_mutex);
593         btrfs_defrag_dirty_roots(root->fs_info);
594         trans = btrfs_start_transaction(root, 1);
595         ret = btrfs_commit_transaction(trans, root);
596         /* run commit again to  drop the original snapshot */
597         trans = btrfs_start_transaction(root, 1);
598         btrfs_commit_transaction(trans, root);
599         ret = btrfs_write_and_wait_transaction(NULL, root);
600         BUG_ON(ret);
601         write_ctree_super(NULL, root);
602         mutex_unlock(&fs_info->fs_mutex);
603
604         if (fs_info->extent_root->node)
605                 btrfs_block_release(fs_info->extent_root,
606                                     fs_info->extent_root->node);
607         if (fs_info->tree_root->node)
608                 btrfs_block_release(fs_info->tree_root,
609                                     fs_info->tree_root->node);
610         btrfs_block_release(root, fs_info->sb_buffer);
611         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
612         iput(fs_info->btree_inode);
613
614         btrfs_free_block_groups(root->fs_info);
615         del_fs_roots(fs_info);
616         kfree(fs_info->extent_root);
617         kfree(fs_info->tree_root);
618         return 0;
619 }
620
621 void btrfs_mark_buffer_dirty(struct buffer_head *bh)
622 {
623         struct btrfs_root *root = BTRFS_I(bh->b_page->mapping->host)->root;
624         u64 transid = btrfs_header_generation(btrfs_buffer_header(bh));
625
626         WARN_ON(!atomic_read(&bh->b_count));
627
628         if (transid != root->fs_info->generation) {
629                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
630                         (unsigned long long)bh->b_blocknr,
631                         transid, root->fs_info->generation);
632                 WARN_ON(1);
633         }
634         mark_buffer_dirty(bh);
635 }
636
637 void btrfs_block_release(struct btrfs_root *root, struct buffer_head *buf)
638 {
639         brelse(buf);
640 }
641
642 void btrfs_btree_balance_dirty(struct btrfs_root *root)
643 {
644         balance_dirty_pages_ratelimited(root->fs_info->btree_inode->i_mapping);
645 }