Btrfs: Add async worker threads for pre and post IO checksumming
[linux-3.10.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/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h> // for block_sync_page
26 #include <linux/workqueue.h>
27 #include "crc32c.h"
28 #include "ctree.h"
29 #include "disk-io.h"
30 #include "transaction.h"
31 #include "btrfs_inode.h"
32 #include "volumes.h"
33 #include "print-tree.h"
34 #include "async-thread.h"
35
36 #if 0
37 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
38 {
39         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
40                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
41                        (unsigned long long)extent_buffer_blocknr(buf),
42                        (unsigned long long)btrfs_header_blocknr(buf));
43                 return 1;
44         }
45         return 0;
46 }
47 #endif
48
49 static struct extent_io_ops btree_extent_io_ops;
50 static void end_workqueue_fn(struct btrfs_work *work);
51
52 struct end_io_wq {
53         struct bio *bio;
54         bio_end_io_t *end_io;
55         void *private;
56         struct btrfs_fs_info *info;
57         int error;
58         int metadata;
59         struct list_head list;
60         struct btrfs_work work;
61 };
62
63 struct async_submit_bio {
64         struct inode *inode;
65         struct bio *bio;
66         struct list_head list;
67         extent_submit_bio_hook_t *submit_bio_hook;
68         int rw;
69         int mirror_num;
70         struct btrfs_work work;
71 };
72
73 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
74                                     size_t page_offset, u64 start, u64 len,
75                                     int create)
76 {
77         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
78         struct extent_map *em;
79         int ret;
80
81         spin_lock(&em_tree->lock);
82         em = lookup_extent_mapping(em_tree, start, len);
83         if (em) {
84                 em->bdev =
85                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
86                 spin_unlock(&em_tree->lock);
87                 goto out;
88         }
89         spin_unlock(&em_tree->lock);
90
91         em = alloc_extent_map(GFP_NOFS);
92         if (!em) {
93                 em = ERR_PTR(-ENOMEM);
94                 goto out;
95         }
96         em->start = 0;
97         em->len = (u64)-1;
98         em->block_start = 0;
99         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
100
101         spin_lock(&em_tree->lock);
102         ret = add_extent_mapping(em_tree, em);
103         if (ret == -EEXIST) {
104                 u64 failed_start = em->start;
105                 u64 failed_len = em->len;
106
107                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
108                        em->start, em->len, em->block_start);
109                 free_extent_map(em);
110                 em = lookup_extent_mapping(em_tree, start, len);
111                 if (em) {
112                         printk("after failing, found %Lu %Lu %Lu\n",
113                                em->start, em->len, em->block_start);
114                         ret = 0;
115                 } else {
116                         em = lookup_extent_mapping(em_tree, failed_start,
117                                                    failed_len);
118                         if (em) {
119                                 printk("double failure lookup gives us "
120                                        "%Lu %Lu -> %Lu\n", em->start,
121                                        em->len, em->block_start);
122                                 free_extent_map(em);
123                         }
124                         ret = -EIO;
125                 }
126         } else if (ret) {
127                 free_extent_map(em);
128                 em = NULL;
129         }
130         spin_unlock(&em_tree->lock);
131
132         if (ret)
133                 em = ERR_PTR(ret);
134 out:
135         return em;
136 }
137
138 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
139 {
140         return btrfs_crc32c(seed, data, len);
141 }
142
143 void btrfs_csum_final(u32 crc, char *result)
144 {
145         *(__le32 *)result = ~cpu_to_le32(crc);
146 }
147
148 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
149                            int verify)
150 {
151         char result[BTRFS_CRC32_SIZE];
152         unsigned long len;
153         unsigned long cur_len;
154         unsigned long offset = BTRFS_CSUM_SIZE;
155         char *map_token = NULL;
156         char *kaddr;
157         unsigned long map_start;
158         unsigned long map_len;
159         int err;
160         u32 crc = ~(u32)0;
161
162         len = buf->len - offset;
163         while(len > 0) {
164                 err = map_private_extent_buffer(buf, offset, 32,
165                                         &map_token, &kaddr,
166                                         &map_start, &map_len, KM_USER0);
167                 if (err) {
168                         printk("failed to map extent buffer! %lu\n",
169                                offset);
170                         return 1;
171                 }
172                 cur_len = min(len, map_len - (offset - map_start));
173                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
174                                       crc, cur_len);
175                 len -= cur_len;
176                 offset += cur_len;
177                 unmap_extent_buffer(buf, map_token, KM_USER0);
178         }
179         btrfs_csum_final(crc, result);
180
181         if (verify) {
182                 int from_this_trans = 0;
183
184                 if (root->fs_info->running_transaction &&
185                     btrfs_header_generation(buf) ==
186                     root->fs_info->running_transaction->transid)
187                         from_this_trans = 1;
188
189                 /* FIXME, this is not good */
190                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
191                         u32 val;
192                         u32 found = 0;
193                         memcpy(&found, result, BTRFS_CRC32_SIZE);
194
195                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
196                         printk("btrfs: %s checksum verify failed on %llu "
197                                "wanted %X found %X from_this_trans %d "
198                                "level %d\n",
199                                root->fs_info->sb->s_id,
200                                buf->start, val, found, from_this_trans,
201                                btrfs_header_level(buf));
202                         return 1;
203                 }
204         } else {
205                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
206         }
207         return 0;
208 }
209
210 static int verify_parent_transid(struct extent_io_tree *io_tree,
211                                  struct extent_buffer *eb, u64 parent_transid)
212 {
213         int ret;
214
215         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
216                 return 0;
217
218         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
219         if (extent_buffer_uptodate(io_tree, eb) &&
220             btrfs_header_generation(eb) == parent_transid) {
221                 ret = 0;
222                 goto out;
223         }
224         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
225                (unsigned long long)eb->start,
226                (unsigned long long)parent_transid,
227                (unsigned long long)btrfs_header_generation(eb));
228         ret = 1;
229 out:
230         clear_extent_buffer_uptodate(io_tree, eb);
231         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
232                       GFP_NOFS);
233         return ret;
234
235 }
236
237 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
238                                           struct extent_buffer *eb,
239                                           u64 start, u64 parent_transid)
240 {
241         struct extent_io_tree *io_tree;
242         int ret;
243         int num_copies = 0;
244         int mirror_num = 0;
245
246         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
247         while (1) {
248                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
249                                                btree_get_extent, mirror_num);
250                 if (!ret &&
251                     !verify_parent_transid(io_tree, eb, parent_transid))
252                         return ret;
253
254                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
255                                               eb->start, eb->len);
256                 if (num_copies == 1)
257                         return ret;
258
259                 mirror_num++;
260                 if (mirror_num > num_copies)
261                         return ret;
262         }
263         return -EIO;
264 }
265
266 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
267 {
268         struct extent_io_tree *tree;
269         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
270         u64 found_start;
271         int found_level;
272         unsigned long len;
273         struct extent_buffer *eb;
274         int ret;
275
276         tree = &BTRFS_I(page->mapping->host)->io_tree;
277
278         if (page->private == EXTENT_PAGE_PRIVATE)
279                 goto out;
280         if (!page->private)
281                 goto out;
282         len = page->private >> 2;
283         if (len == 0) {
284                 WARN_ON(1);
285         }
286         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
287         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
288                                              btrfs_header_generation(eb));
289         BUG_ON(ret);
290         btrfs_clear_buffer_defrag(eb);
291         found_start = btrfs_header_bytenr(eb);
292         if (found_start != start) {
293                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
294                        start, found_start, len);
295                 WARN_ON(1);
296                 goto err;
297         }
298         if (eb->first_page != page) {
299                 printk("bad first page %lu %lu\n", eb->first_page->index,
300                        page->index);
301                 WARN_ON(1);
302                 goto err;
303         }
304         if (!PageUptodate(page)) {
305                 printk("csum not up to date page %lu\n", page->index);
306                 WARN_ON(1);
307                 goto err;
308         }
309         found_level = btrfs_header_level(eb);
310         spin_lock(&root->fs_info->hash_lock);
311         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
312         spin_unlock(&root->fs_info->hash_lock);
313         csum_tree_block(root, eb, 0);
314 err:
315         free_extent_buffer(eb);
316 out:
317         return 0;
318 }
319
320 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
321 {
322         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
323
324         csum_dirty_buffer(root, page);
325         return 0;
326 }
327
328 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
329                                struct extent_state *state)
330 {
331         struct extent_io_tree *tree;
332         u64 found_start;
333         int found_level;
334         unsigned long len;
335         struct extent_buffer *eb;
336         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
337         int ret = 0;
338
339         tree = &BTRFS_I(page->mapping->host)->io_tree;
340         if (page->private == EXTENT_PAGE_PRIVATE)
341                 goto out;
342         if (!page->private)
343                 goto out;
344         len = page->private >> 2;
345         if (len == 0) {
346                 WARN_ON(1);
347         }
348         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
349
350         btrfs_clear_buffer_defrag(eb);
351         found_start = btrfs_header_bytenr(eb);
352         if (found_start != start) {
353                 ret = -EIO;
354                 goto err;
355         }
356         if (eb->first_page != page) {
357                 printk("bad first page %lu %lu\n", eb->first_page->index,
358                        page->index);
359                 WARN_ON(1);
360                 ret = -EIO;
361                 goto err;
362         }
363         if (memcmp_extent_buffer(eb, root->fs_info->fsid,
364                                  (unsigned long)btrfs_header_fsid(eb),
365                                  BTRFS_FSID_SIZE)) {
366                 printk("bad fsid on block %Lu\n", eb->start);
367                 ret = -EIO;
368                 goto err;
369         }
370         found_level = btrfs_header_level(eb);
371
372         ret = csum_tree_block(root, eb, 1);
373         if (ret)
374                 ret = -EIO;
375
376         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
377         end = eb->start + end - 1;
378         release_extent_buffer_tail_pages(eb);
379 err:
380         free_extent_buffer(eb);
381 out:
382         return ret;
383 }
384
385 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
386 static void end_workqueue_bio(struct bio *bio, int err)
387 #else
388 static int end_workqueue_bio(struct bio *bio,
389                                    unsigned int bytes_done, int err)
390 #endif
391 {
392         struct end_io_wq *end_io_wq = bio->bi_private;
393         struct btrfs_fs_info *fs_info;
394
395 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
396         if (bio->bi_size)
397                 return 1;
398 #endif
399
400         fs_info = end_io_wq->info;
401         end_io_wq->error = err;
402         end_io_wq->work.func = end_workqueue_fn;
403         end_io_wq->work.flags = 0;
404         btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
405
406 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
407         return 0;
408 #endif
409 }
410
411 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
412                         int metadata)
413 {
414         struct end_io_wq *end_io_wq;
415         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
416         if (!end_io_wq)
417                 return -ENOMEM;
418
419         end_io_wq->private = bio->bi_private;
420         end_io_wq->end_io = bio->bi_end_io;
421         end_io_wq->info = info;
422         end_io_wq->error = 0;
423         end_io_wq->bio = bio;
424         end_io_wq->metadata = metadata;
425
426         bio->bi_private = end_io_wq;
427         bio->bi_end_io = end_workqueue_bio;
428         return 0;
429 }
430
431 static void run_one_async_submit(struct btrfs_work *work)
432 {
433         struct btrfs_fs_info *fs_info;
434         struct async_submit_bio *async;
435
436         async = container_of(work, struct  async_submit_bio, work);
437         fs_info = BTRFS_I(async->inode)->root->fs_info;
438         atomic_dec(&fs_info->nr_async_submits);
439         async->submit_bio_hook(async->inode, async->rw, async->bio,
440                                async->mirror_num);
441         kfree(async);
442 }
443
444 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
445                         int rw, struct bio *bio, int mirror_num,
446                         extent_submit_bio_hook_t *submit_bio_hook)
447 {
448         struct async_submit_bio *async;
449
450         async = kmalloc(sizeof(*async), GFP_NOFS);
451         if (!async)
452                 return -ENOMEM;
453
454         async->inode = inode;
455         async->rw = rw;
456         async->bio = bio;
457         async->mirror_num = mirror_num;
458         async->submit_bio_hook = submit_bio_hook;
459         async->work.func = run_one_async_submit;
460         async->work.flags = 0;
461         atomic_inc(&fs_info->nr_async_submits);
462         btrfs_queue_worker(&fs_info->workers, &async->work);
463         return 0;
464 }
465
466 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
467                                  int mirror_num)
468 {
469         struct btrfs_root *root = BTRFS_I(inode)->root;
470         u64 offset;
471         int ret;
472
473         offset = bio->bi_sector << 9;
474
475         /*
476          * when we're called for a write, we're already in the async
477          * submission context.  Just jump ingo btrfs_map_bio
478          */
479         if (rw & (1 << BIO_RW)) {
480                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
481                                      mirror_num, 0);
482         }
483
484         /*
485          * called for a read, do the setup so that checksum validation
486          * can happen in the async kernel threads
487          */
488         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
489         BUG_ON(ret);
490
491         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
492 }
493
494 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
495                                  int mirror_num)
496 {
497         /*
498          * kthread helpers are used to submit writes so that checksumming
499          * can happen in parallel across all CPUs
500          */
501         if (!(rw & (1 << BIO_RW))) {
502                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
503         }
504         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
505                                    inode, rw, bio, mirror_num,
506                                    __btree_submit_bio_hook);
507 }
508
509 static int btree_writepage(struct page *page, struct writeback_control *wbc)
510 {
511         struct extent_io_tree *tree;
512         tree = &BTRFS_I(page->mapping->host)->io_tree;
513         return extent_write_full_page(tree, page, btree_get_extent, wbc);
514 }
515
516 static int btree_writepages(struct address_space *mapping,
517                             struct writeback_control *wbc)
518 {
519         struct extent_io_tree *tree;
520         tree = &BTRFS_I(mapping->host)->io_tree;
521         if (wbc->sync_mode == WB_SYNC_NONE) {
522                 u64 num_dirty;
523                 u64 start = 0;
524                 unsigned long thresh = 96 * 1024 * 1024;
525
526                 if (wbc->for_kupdate)
527                         return 0;
528
529                 if (current_is_pdflush()) {
530                         thresh = 96 * 1024 * 1024;
531                 } else {
532                         thresh = 8 * 1024 * 1024;
533                 }
534                 num_dirty = count_range_bits(tree, &start, (u64)-1,
535                                              thresh, EXTENT_DIRTY);
536                 if (num_dirty < thresh) {
537                         return 0;
538                 }
539         }
540         return extent_writepages(tree, mapping, btree_get_extent, wbc);
541 }
542
543 int btree_readpage(struct file *file, struct page *page)
544 {
545         struct extent_io_tree *tree;
546         tree = &BTRFS_I(page->mapping->host)->io_tree;
547         return extent_read_full_page(tree, page, btree_get_extent);
548 }
549
550 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
551 {
552         struct extent_io_tree *tree;
553         struct extent_map_tree *map;
554         int ret;
555
556         if (page_count(page) > 3) {
557                 /* once for page->private, once for the caller, once
558                  * once for the page cache
559                  */
560                 return 0;
561         }
562         tree = &BTRFS_I(page->mapping->host)->io_tree;
563         map = &BTRFS_I(page->mapping->host)->extent_tree;
564         ret = try_release_extent_state(map, tree, page, gfp_flags);
565         if (ret == 1) {
566                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
567                 ClearPagePrivate(page);
568                 set_page_private(page, 0);
569                 page_cache_release(page);
570         }
571         return ret;
572 }
573
574 static void btree_invalidatepage(struct page *page, unsigned long offset)
575 {
576         struct extent_io_tree *tree;
577         tree = &BTRFS_I(page->mapping->host)->io_tree;
578         extent_invalidatepage(tree, page, offset);
579         btree_releasepage(page, GFP_NOFS);
580         if (PagePrivate(page)) {
581                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
582                 ClearPagePrivate(page);
583                 set_page_private(page, 0);
584                 page_cache_release(page);
585         }
586 }
587
588 #if 0
589 static int btree_writepage(struct page *page, struct writeback_control *wbc)
590 {
591         struct buffer_head *bh;
592         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
593         struct buffer_head *head;
594         if (!page_has_buffers(page)) {
595                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
596                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
597         }
598         head = page_buffers(page);
599         bh = head;
600         do {
601                 if (buffer_dirty(bh))
602                         csum_tree_block(root, bh, 0);
603                 bh = bh->b_this_page;
604         } while (bh != head);
605         return block_write_full_page(page, btree_get_block, wbc);
606 }
607 #endif
608
609 static struct address_space_operations btree_aops = {
610         .readpage       = btree_readpage,
611         .writepage      = btree_writepage,
612         .writepages     = btree_writepages,
613         .releasepage    = btree_releasepage,
614         .invalidatepage = btree_invalidatepage,
615         .sync_page      = block_sync_page,
616 };
617
618 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
619                          u64 parent_transid)
620 {
621         struct extent_buffer *buf = NULL;
622         struct inode *btree_inode = root->fs_info->btree_inode;
623         int ret = 0;
624
625         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
626         if (!buf)
627                 return 0;
628         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
629                                  buf, 0, 0, btree_get_extent, 0);
630         free_extent_buffer(buf);
631         return ret;
632 }
633
634 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
635                                             u64 bytenr, u32 blocksize)
636 {
637         struct inode *btree_inode = root->fs_info->btree_inode;
638         struct extent_buffer *eb;
639         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
640                                 bytenr, blocksize, GFP_NOFS);
641         return eb;
642 }
643
644 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
645                                                  u64 bytenr, u32 blocksize)
646 {
647         struct inode *btree_inode = root->fs_info->btree_inode;
648         struct extent_buffer *eb;
649
650         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
651                                  bytenr, blocksize, NULL, GFP_NOFS);
652         return eb;
653 }
654
655
656 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
657                                       u32 blocksize, u64 parent_transid)
658 {
659         struct extent_buffer *buf = NULL;
660         struct inode *btree_inode = root->fs_info->btree_inode;
661         struct extent_io_tree *io_tree;
662         int ret;
663
664         io_tree = &BTRFS_I(btree_inode)->io_tree;
665
666         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
667         if (!buf)
668                 return NULL;
669
670         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
671
672         if (ret == 0) {
673                 buf->flags |= EXTENT_UPTODATE;
674         }
675         return buf;
676
677 }
678
679 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
680                      struct extent_buffer *buf)
681 {
682         struct inode *btree_inode = root->fs_info->btree_inode;
683         if (btrfs_header_generation(buf) ==
684             root->fs_info->running_transaction->transid)
685                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
686                                           buf);
687         return 0;
688 }
689
690 int wait_on_tree_block_writeback(struct btrfs_root *root,
691                                  struct extent_buffer *buf)
692 {
693         struct inode *btree_inode = root->fs_info->btree_inode;
694         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
695                                         buf);
696         return 0;
697 }
698
699 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
700                         u32 stripesize, struct btrfs_root *root,
701                         struct btrfs_fs_info *fs_info,
702                         u64 objectid)
703 {
704         root->node = NULL;
705         root->inode = NULL;
706         root->commit_root = NULL;
707         root->sectorsize = sectorsize;
708         root->nodesize = nodesize;
709         root->leafsize = leafsize;
710         root->stripesize = stripesize;
711         root->ref_cows = 0;
712         root->track_dirty = 0;
713
714         root->fs_info = fs_info;
715         root->objectid = objectid;
716         root->last_trans = 0;
717         root->highest_inode = 0;
718         root->last_inode_alloc = 0;
719         root->name = NULL;
720         root->in_sysfs = 0;
721
722         INIT_LIST_HEAD(&root->dirty_list);
723         memset(&root->root_key, 0, sizeof(root->root_key));
724         memset(&root->root_item, 0, sizeof(root->root_item));
725         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
726         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
727         init_completion(&root->kobj_unregister);
728         root->defrag_running = 0;
729         root->defrag_level = 0;
730         root->root_key.objectid = objectid;
731         return 0;
732 }
733
734 static int find_and_setup_root(struct btrfs_root *tree_root,
735                                struct btrfs_fs_info *fs_info,
736                                u64 objectid,
737                                struct btrfs_root *root)
738 {
739         int ret;
740         u32 blocksize;
741
742         __setup_root(tree_root->nodesize, tree_root->leafsize,
743                      tree_root->sectorsize, tree_root->stripesize,
744                      root, fs_info, objectid);
745         ret = btrfs_find_last_root(tree_root, objectid,
746                                    &root->root_item, &root->root_key);
747         BUG_ON(ret);
748
749         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
750         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
751                                      blocksize, 0);
752         BUG_ON(!root->node);
753         return 0;
754 }
755
756 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
757                                                struct btrfs_key *location)
758 {
759         struct btrfs_root *root;
760         struct btrfs_root *tree_root = fs_info->tree_root;
761         struct btrfs_path *path;
762         struct extent_buffer *l;
763         u64 highest_inode;
764         u32 blocksize;
765         int ret = 0;
766
767         root = kzalloc(sizeof(*root), GFP_NOFS);
768         if (!root)
769                 return ERR_PTR(-ENOMEM);
770         if (location->offset == (u64)-1) {
771                 ret = find_and_setup_root(tree_root, fs_info,
772                                           location->objectid, root);
773                 if (ret) {
774                         kfree(root);
775                         return ERR_PTR(ret);
776                 }
777                 goto insert;
778         }
779
780         __setup_root(tree_root->nodesize, tree_root->leafsize,
781                      tree_root->sectorsize, tree_root->stripesize,
782                      root, fs_info, location->objectid);
783
784         path = btrfs_alloc_path();
785         BUG_ON(!path);
786         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
787         if (ret != 0) {
788                 if (ret > 0)
789                         ret = -ENOENT;
790                 goto out;
791         }
792         l = path->nodes[0];
793         read_extent_buffer(l, &root->root_item,
794                btrfs_item_ptr_offset(l, path->slots[0]),
795                sizeof(root->root_item));
796         memcpy(&root->root_key, location, sizeof(*location));
797         ret = 0;
798 out:
799         btrfs_release_path(root, path);
800         btrfs_free_path(path);
801         if (ret) {
802                 kfree(root);
803                 return ERR_PTR(ret);
804         }
805         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
806         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
807                                      blocksize, 0);
808         BUG_ON(!root->node);
809 insert:
810         root->ref_cows = 1;
811         ret = btrfs_find_highest_inode(root, &highest_inode);
812         if (ret == 0) {
813                 root->highest_inode = highest_inode;
814                 root->last_inode_alloc = highest_inode;
815         }
816         return root;
817 }
818
819 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
820                                         u64 root_objectid)
821 {
822         struct btrfs_root *root;
823
824         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
825                 return fs_info->tree_root;
826         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
827                 return fs_info->extent_root;
828
829         root = radix_tree_lookup(&fs_info->fs_roots_radix,
830                                  (unsigned long)root_objectid);
831         return root;
832 }
833
834 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
835                                               struct btrfs_key *location)
836 {
837         struct btrfs_root *root;
838         int ret;
839
840         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
841                 return fs_info->tree_root;
842         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
843                 return fs_info->extent_root;
844         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
845                 return fs_info->chunk_root;
846         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
847                 return fs_info->dev_root;
848
849         root = radix_tree_lookup(&fs_info->fs_roots_radix,
850                                  (unsigned long)location->objectid);
851         if (root)
852                 return root;
853
854         root = btrfs_read_fs_root_no_radix(fs_info, location);
855         if (IS_ERR(root))
856                 return root;
857         ret = radix_tree_insert(&fs_info->fs_roots_radix,
858                                 (unsigned long)root->root_key.objectid,
859                                 root);
860         if (ret) {
861                 free_extent_buffer(root->node);
862                 kfree(root);
863                 return ERR_PTR(ret);
864         }
865         ret = btrfs_find_dead_roots(fs_info->tree_root,
866                                     root->root_key.objectid, root);
867         BUG_ON(ret);
868
869         return root;
870 }
871
872 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
873                                       struct btrfs_key *location,
874                                       const char *name, int namelen)
875 {
876         struct btrfs_root *root;
877         int ret;
878
879         root = btrfs_read_fs_root_no_name(fs_info, location);
880         if (!root)
881                 return NULL;
882
883         if (root->in_sysfs)
884                 return root;
885
886         ret = btrfs_set_root_name(root, name, namelen);
887         if (ret) {
888                 free_extent_buffer(root->node);
889                 kfree(root);
890                 return ERR_PTR(ret);
891         }
892
893         ret = btrfs_sysfs_add_root(root);
894         if (ret) {
895                 free_extent_buffer(root->node);
896                 kfree(root->name);
897                 kfree(root);
898                 return ERR_PTR(ret);
899         }
900         root->in_sysfs = 1;
901         return root;
902 }
903 #if 0
904 static int add_hasher(struct btrfs_fs_info *info, char *type) {
905         struct btrfs_hasher *hasher;
906
907         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
908         if (!hasher)
909                 return -ENOMEM;
910         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
911         if (!hasher->hash_tfm) {
912                 kfree(hasher);
913                 return -EINVAL;
914         }
915         spin_lock(&info->hash_lock);
916         list_add(&hasher->list, &info->hashers);
917         spin_unlock(&info->hash_lock);
918         return 0;
919 }
920 #endif
921
922 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
923 {
924         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
925         int ret = 0;
926         int limit = 256 * info->fs_devices->open_devices;
927         struct list_head *cur;
928         struct btrfs_device *device;
929         struct backing_dev_info *bdi;
930
931         if ((bdi_bits & (1 << BDI_write_congested)) &&
932             atomic_read(&info->nr_async_submits) > limit) {
933                 return 1;
934         }
935
936         list_for_each(cur, &info->fs_devices->devices) {
937                 device = list_entry(cur, struct btrfs_device, dev_list);
938                 if (!device->bdev)
939                         continue;
940                 bdi = blk_get_backing_dev_info(device->bdev);
941                 if (bdi && bdi_congested(bdi, bdi_bits)) {
942                         ret = 1;
943                         break;
944                 }
945         }
946         return ret;
947 }
948
949 /*
950  * this unplugs every device on the box, and it is only used when page
951  * is null
952  */
953 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
954 {
955         struct list_head *cur;
956         struct btrfs_device *device;
957         struct btrfs_fs_info *info;
958
959         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
960         list_for_each(cur, &info->fs_devices->devices) {
961                 device = list_entry(cur, struct btrfs_device, dev_list);
962                 bdi = blk_get_backing_dev_info(device->bdev);
963                 if (bdi->unplug_io_fn) {
964                         bdi->unplug_io_fn(bdi, page);
965                 }
966         }
967 }
968
969 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
970 {
971         struct inode *inode;
972         struct extent_map_tree *em_tree;
973         struct extent_map *em;
974         struct address_space *mapping;
975         u64 offset;
976
977         /* the generic O_DIRECT read code does this */
978         if (!page) {
979                 __unplug_io_fn(bdi, page);
980                 return;
981         }
982
983         /*
984          * page->mapping may change at any time.  Get a consistent copy
985          * and use that for everything below
986          */
987         smp_mb();
988         mapping = page->mapping;
989         if (!mapping)
990                 return;
991
992         inode = mapping->host;
993         offset = page_offset(page);
994
995         em_tree = &BTRFS_I(inode)->extent_tree;
996         spin_lock(&em_tree->lock);
997         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
998         spin_unlock(&em_tree->lock);
999         if (!em)
1000                 return;
1001
1002         offset = offset - em->start;
1003         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1004                           em->block_start + offset, page);
1005         free_extent_map(em);
1006 }
1007
1008 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1009 {
1010 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1011         bdi_init(bdi);
1012 #endif
1013         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1014         bdi->state              = 0;
1015         bdi->capabilities       = default_backing_dev_info.capabilities;
1016         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1017         bdi->unplug_io_data     = info;
1018         bdi->congested_fn       = btrfs_congested_fn;
1019         bdi->congested_data     = info;
1020         return 0;
1021 }
1022
1023 static int bio_ready_for_csum(struct bio *bio)
1024 {
1025         u64 length = 0;
1026         u64 buf_len = 0;
1027         u64 start = 0;
1028         struct page *page;
1029         struct extent_io_tree *io_tree = NULL;
1030         struct btrfs_fs_info *info = NULL;
1031         struct bio_vec *bvec;
1032         int i;
1033         int ret;
1034
1035         bio_for_each_segment(bvec, bio, i) {
1036                 page = bvec->bv_page;
1037                 if (page->private == EXTENT_PAGE_PRIVATE) {
1038                         length += bvec->bv_len;
1039                         continue;
1040                 }
1041                 if (!page->private) {
1042                         length += bvec->bv_len;
1043                         continue;
1044                 }
1045                 length = bvec->bv_len;
1046                 buf_len = page->private >> 2;
1047                 start = page_offset(page) + bvec->bv_offset;
1048                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1049                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1050         }
1051         /* are we fully contained in this bio? */
1052         if (buf_len <= length)
1053                 return 1;
1054
1055         ret = extent_range_uptodate(io_tree, start + length,
1056                                     start + buf_len - 1);
1057         if (ret == 1)
1058                 return ret;
1059         return ret;
1060 }
1061
1062 /*
1063  * called by the kthread helper functions to finally call the bio end_io
1064  * functions.  This is where read checksum verification actually happens
1065  */
1066 static void end_workqueue_fn(struct btrfs_work *work)
1067 {
1068         struct bio *bio;
1069         struct end_io_wq *end_io_wq;
1070         struct btrfs_fs_info *fs_info;
1071         int error;
1072
1073         end_io_wq = container_of(work, struct end_io_wq, work);
1074         bio = end_io_wq->bio;
1075         fs_info = end_io_wq->info;
1076
1077         /* metadata bios are special because the whole tree block must
1078          * be checksummed at once.  This makes sure the entire block is in
1079          * ram and up to date before trying to verify things.  For
1080          * blocksize <= pagesize, it is basically a noop
1081          */
1082         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1083                 btrfs_queue_worker(&fs_info->endio_workers,
1084                                    &end_io_wq->work);
1085                 return;
1086         }
1087         error = end_io_wq->error;
1088         bio->bi_private = end_io_wq->private;
1089         bio->bi_end_io = end_io_wq->end_io;
1090         kfree(end_io_wq);
1091 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1092         bio_endio(bio, bio->bi_size, error);
1093 #else
1094         bio_endio(bio, error);
1095 #endif
1096 }
1097
1098 struct btrfs_root *open_ctree(struct super_block *sb,
1099                               struct btrfs_fs_devices *fs_devices,
1100                               char *options)
1101 {
1102         u32 sectorsize;
1103         u32 nodesize;
1104         u32 leafsize;
1105         u32 blocksize;
1106         u32 stripesize;
1107         struct buffer_head *bh;
1108         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1109                                                  GFP_NOFS);
1110         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1111                                                GFP_NOFS);
1112         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1113                                                 GFP_NOFS);
1114         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1115                                                 GFP_NOFS);
1116         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1117                                               GFP_NOFS);
1118         int ret;
1119         int err = -EINVAL;
1120         struct btrfs_super_block *disk_super;
1121
1122         if (!extent_root || !tree_root || !fs_info) {
1123                 err = -ENOMEM;
1124                 goto fail;
1125         }
1126         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1127         INIT_LIST_HEAD(&fs_info->trans_list);
1128         INIT_LIST_HEAD(&fs_info->dead_roots);
1129         INIT_LIST_HEAD(&fs_info->hashers);
1130         spin_lock_init(&fs_info->hash_lock);
1131         spin_lock_init(&fs_info->delalloc_lock);
1132         spin_lock_init(&fs_info->new_trans_lock);
1133
1134         init_completion(&fs_info->kobj_unregister);
1135         fs_info->tree_root = tree_root;
1136         fs_info->extent_root = extent_root;
1137         fs_info->chunk_root = chunk_root;
1138         fs_info->dev_root = dev_root;
1139         fs_info->fs_devices = fs_devices;
1140         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1141         INIT_LIST_HEAD(&fs_info->space_info);
1142         btrfs_mapping_init(&fs_info->mapping_tree);
1143         atomic_set(&fs_info->nr_async_submits, 0);
1144         fs_info->sb = sb;
1145         fs_info->max_extent = (u64)-1;
1146         fs_info->max_inline = 8192 * 1024;
1147         setup_bdi(fs_info, &fs_info->bdi);
1148         fs_info->btree_inode = new_inode(sb);
1149         fs_info->btree_inode->i_ino = 1;
1150         fs_info->btree_inode->i_nlink = 1;
1151
1152         sb->s_blocksize = 4096;
1153         sb->s_blocksize_bits = blksize_bits(4096);
1154
1155         /*
1156          * we set the i_size on the btree inode to the max possible int.
1157          * the real end of the address space is determined by all of
1158          * the devices in the system
1159          */
1160         fs_info->btree_inode->i_size = OFFSET_MAX;
1161         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1162         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1163
1164         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1165                              fs_info->btree_inode->i_mapping,
1166                              GFP_NOFS);
1167         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1168                              GFP_NOFS);
1169
1170         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1171
1172         extent_io_tree_init(&fs_info->free_space_cache,
1173                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1174         extent_io_tree_init(&fs_info->block_group_cache,
1175                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1176         extent_io_tree_init(&fs_info->pinned_extents,
1177                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1178         extent_io_tree_init(&fs_info->pending_del,
1179                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1180         extent_io_tree_init(&fs_info->extent_ins,
1181                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1182         fs_info->do_barriers = 1;
1183
1184 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1185         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1186 #else
1187         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1188 #endif
1189         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1190         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1191                sizeof(struct btrfs_key));
1192         insert_inode_hash(fs_info->btree_inode);
1193         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1194
1195         mutex_init(&fs_info->trans_mutex);
1196         mutex_init(&fs_info->fs_mutex);
1197
1198         /* we need to start all the end_io workers up front because the
1199          * queue work function gets called at interrupt time.  The endio
1200          * workers don't normally start IO, so some number of them <= the
1201          * number of cpus is fine.  They handle checksumming after a read.
1202          *
1203          * The other worker threads do start IO, so the max is larger than
1204          * the number of CPUs.  FIXME, tune this for huge machines
1205          */
1206         btrfs_init_workers(&fs_info->workers, num_online_cpus() * 2);
1207         btrfs_init_workers(&fs_info->endio_workers, num_online_cpus());
1208         btrfs_start_workers(&fs_info->workers, 1);
1209         btrfs_start_workers(&fs_info->endio_workers, num_online_cpus());
1210
1211 #if 0
1212         ret = add_hasher(fs_info, "crc32c");
1213         if (ret) {
1214                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1215                 err = -ENOMEM;
1216                 goto fail_iput;
1217         }
1218 #endif
1219         __setup_root(4096, 4096, 4096, 4096, tree_root,
1220                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1221
1222
1223         bh = __bread(fs_devices->latest_bdev,
1224                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1225         if (!bh)
1226                 goto fail_iput;
1227
1228         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1229         brelse(bh);
1230
1231         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1232
1233         disk_super = &fs_info->super_copy;
1234         if (!btrfs_super_root(disk_super))
1235                 goto fail_sb_buffer;
1236
1237         err = btrfs_parse_options(tree_root, options);
1238         if (err)
1239                 goto fail_sb_buffer;
1240
1241         err = -EINVAL;
1242         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1243                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1244                        (unsigned long long)btrfs_super_num_devices(disk_super),
1245                        (unsigned long long)fs_devices->open_devices);
1246                 if (btrfs_test_opt(tree_root, DEGRADED))
1247                         printk("continuing in degraded mode\n");
1248                 else {
1249                         goto fail_sb_buffer;
1250                 }
1251         }
1252
1253         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1254
1255         nodesize = btrfs_super_nodesize(disk_super);
1256         leafsize = btrfs_super_leafsize(disk_super);
1257         sectorsize = btrfs_super_sectorsize(disk_super);
1258         stripesize = btrfs_super_stripesize(disk_super);
1259         tree_root->nodesize = nodesize;
1260         tree_root->leafsize = leafsize;
1261         tree_root->sectorsize = sectorsize;
1262         tree_root->stripesize = stripesize;
1263
1264         sb->s_blocksize = sectorsize;
1265         sb->s_blocksize_bits = blksize_bits(sectorsize);
1266
1267         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1268                     sizeof(disk_super->magic))) {
1269                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1270                 goto fail_sb_buffer;
1271         }
1272
1273         mutex_lock(&fs_info->fs_mutex);
1274
1275         ret = btrfs_read_sys_array(tree_root);
1276         if (ret) {
1277                 printk("btrfs: failed to read the system array on %s\n",
1278                        sb->s_id);
1279                 goto fail_sys_array;
1280         }
1281
1282         blocksize = btrfs_level_size(tree_root,
1283                                      btrfs_super_chunk_root_level(disk_super));
1284
1285         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1286                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1287
1288         chunk_root->node = read_tree_block(chunk_root,
1289                                            btrfs_super_chunk_root(disk_super),
1290                                            blocksize, 0);
1291         BUG_ON(!chunk_root->node);
1292
1293         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1294                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1295                  BTRFS_UUID_SIZE);
1296
1297         ret = btrfs_read_chunk_tree(chunk_root);
1298         BUG_ON(ret);
1299
1300         btrfs_close_extra_devices(fs_devices);
1301
1302         blocksize = btrfs_level_size(tree_root,
1303                                      btrfs_super_root_level(disk_super));
1304
1305
1306         tree_root->node = read_tree_block(tree_root,
1307                                           btrfs_super_root(disk_super),
1308                                           blocksize, 0);
1309         if (!tree_root->node)
1310                 goto fail_sb_buffer;
1311
1312
1313         ret = find_and_setup_root(tree_root, fs_info,
1314                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1315         if (ret)
1316                 goto fail_tree_root;
1317         extent_root->track_dirty = 1;
1318
1319         ret = find_and_setup_root(tree_root, fs_info,
1320                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1321         dev_root->track_dirty = 1;
1322
1323         if (ret)
1324                 goto fail_extent_root;
1325
1326         btrfs_read_block_groups(extent_root);
1327
1328         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1329         fs_info->data_alloc_profile = (u64)-1;
1330         fs_info->metadata_alloc_profile = (u64)-1;
1331         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1332
1333         mutex_unlock(&fs_info->fs_mutex);
1334         return tree_root;
1335
1336 fail_extent_root:
1337         free_extent_buffer(extent_root->node);
1338 fail_tree_root:
1339         free_extent_buffer(tree_root->node);
1340 fail_sys_array:
1341         mutex_unlock(&fs_info->fs_mutex);
1342 fail_sb_buffer:
1343         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1344 fail_iput:
1345         iput(fs_info->btree_inode);
1346         btrfs_stop_workers(&fs_info->workers);
1347         btrfs_stop_workers(&fs_info->endio_workers);
1348 fail:
1349         btrfs_close_devices(fs_info->fs_devices);
1350         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1351
1352         kfree(extent_root);
1353         kfree(tree_root);
1354 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1355         bdi_destroy(&fs_info->bdi);
1356 #endif
1357         kfree(fs_info);
1358         return ERR_PTR(err);
1359 }
1360
1361 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1362 {
1363         char b[BDEVNAME_SIZE];
1364
1365         if (uptodate) {
1366                 set_buffer_uptodate(bh);
1367         } else {
1368                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1369                         printk(KERN_WARNING "lost page write due to "
1370                                         "I/O error on %s\n",
1371                                        bdevname(bh->b_bdev, b));
1372                 }
1373                 /* note, we dont' set_buffer_write_io_error because we have
1374                  * our own ways of dealing with the IO errors
1375                  */
1376                 clear_buffer_uptodate(bh);
1377         }
1378         unlock_buffer(bh);
1379         put_bh(bh);
1380 }
1381
1382 int write_all_supers(struct btrfs_root *root)
1383 {
1384         struct list_head *cur;
1385         struct list_head *head = &root->fs_info->fs_devices->devices;
1386         struct btrfs_device *dev;
1387         struct btrfs_super_block *sb;
1388         struct btrfs_dev_item *dev_item;
1389         struct buffer_head *bh;
1390         int ret;
1391         int do_barriers;
1392         int max_errors;
1393         int total_errors = 0;
1394         u32 crc;
1395         u64 flags;
1396
1397         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1398         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1399
1400         sb = &root->fs_info->super_for_commit;
1401         dev_item = &sb->dev_item;
1402         list_for_each(cur, head) {
1403                 dev = list_entry(cur, struct btrfs_device, dev_list);
1404                 if (!dev->bdev) {
1405                         total_errors++;
1406                         continue;
1407                 }
1408                 if (!dev->in_fs_metadata)
1409                         continue;
1410
1411                 btrfs_set_stack_device_type(dev_item, dev->type);
1412                 btrfs_set_stack_device_id(dev_item, dev->devid);
1413                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1414                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1415                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1416                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1417                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1418                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1419                 flags = btrfs_super_flags(sb);
1420                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1421
1422
1423                 crc = ~(u32)0;
1424                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1425                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1426                 btrfs_csum_final(crc, sb->csum);
1427
1428                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1429                               BTRFS_SUPER_INFO_SIZE);
1430
1431                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1432                 dev->pending_io = bh;
1433
1434                 get_bh(bh);
1435                 set_buffer_uptodate(bh);
1436                 lock_buffer(bh);
1437                 bh->b_end_io = btrfs_end_buffer_write_sync;
1438
1439                 if (do_barriers && dev->barriers) {
1440                         ret = submit_bh(WRITE_BARRIER, bh);
1441                         if (ret == -EOPNOTSUPP) {
1442                                 printk("btrfs: disabling barriers on dev %s\n",
1443                                        dev->name);
1444                                 set_buffer_uptodate(bh);
1445                                 dev->barriers = 0;
1446                                 get_bh(bh);
1447                                 lock_buffer(bh);
1448                                 ret = submit_bh(WRITE, bh);
1449                         }
1450                 } else {
1451                         ret = submit_bh(WRITE, bh);
1452                 }
1453                 if (ret)
1454                         total_errors++;
1455         }
1456         if (total_errors > max_errors) {
1457                 printk("btrfs: %d errors while writing supers\n", total_errors);
1458                 BUG();
1459         }
1460         total_errors = 0;
1461
1462         list_for_each(cur, head) {
1463                 dev = list_entry(cur, struct btrfs_device, dev_list);
1464                 if (!dev->bdev)
1465                         continue;
1466                 if (!dev->in_fs_metadata)
1467                         continue;
1468
1469                 BUG_ON(!dev->pending_io);
1470                 bh = dev->pending_io;
1471                 wait_on_buffer(bh);
1472                 if (!buffer_uptodate(dev->pending_io)) {
1473                         if (do_barriers && dev->barriers) {
1474                                 printk("btrfs: disabling barriers on dev %s\n",
1475                                        dev->name);
1476                                 set_buffer_uptodate(bh);
1477                                 get_bh(bh);
1478                                 lock_buffer(bh);
1479                                 dev->barriers = 0;
1480                                 ret = submit_bh(WRITE, bh);
1481                                 BUG_ON(ret);
1482                                 wait_on_buffer(bh);
1483                                 if (!buffer_uptodate(bh))
1484                                         total_errors++;
1485                         } else {
1486                                 total_errors++;
1487                         }
1488
1489                 }
1490                 dev->pending_io = NULL;
1491                 brelse(bh);
1492         }
1493         if (total_errors > max_errors) {
1494                 printk("btrfs: %d errors while writing supers\n", total_errors);
1495                 BUG();
1496         }
1497         return 0;
1498 }
1499
1500 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1501                       *root)
1502 {
1503         int ret;
1504
1505         ret = write_all_supers(root);
1506         return ret;
1507 }
1508
1509 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1510 {
1511         radix_tree_delete(&fs_info->fs_roots_radix,
1512                           (unsigned long)root->root_key.objectid);
1513         if (root->in_sysfs)
1514                 btrfs_sysfs_del_root(root);
1515         if (root->inode)
1516                 iput(root->inode);
1517         if (root->node)
1518                 free_extent_buffer(root->node);
1519         if (root->commit_root)
1520                 free_extent_buffer(root->commit_root);
1521         if (root->name)
1522                 kfree(root->name);
1523         kfree(root);
1524         return 0;
1525 }
1526
1527 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1528 {
1529         int ret;
1530         struct btrfs_root *gang[8];
1531         int i;
1532
1533         while(1) {
1534                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1535                                              (void **)gang, 0,
1536                                              ARRAY_SIZE(gang));
1537                 if (!ret)
1538                         break;
1539                 for (i = 0; i < ret; i++)
1540                         btrfs_free_fs_root(fs_info, gang[i]);
1541         }
1542         return 0;
1543 }
1544
1545 int close_ctree(struct btrfs_root *root)
1546 {
1547         int ret;
1548         struct btrfs_trans_handle *trans;
1549         struct btrfs_fs_info *fs_info = root->fs_info;
1550
1551         fs_info->closing = 1;
1552         btrfs_transaction_flush_work(root);
1553         mutex_lock(&fs_info->fs_mutex);
1554         btrfs_defrag_dirty_roots(root->fs_info);
1555         trans = btrfs_start_transaction(root, 1);
1556         ret = btrfs_commit_transaction(trans, root);
1557         /* run commit again to  drop the original snapshot */
1558         trans = btrfs_start_transaction(root, 1);
1559         btrfs_commit_transaction(trans, root);
1560         ret = btrfs_write_and_wait_transaction(NULL, root);
1561         BUG_ON(ret);
1562
1563         write_ctree_super(NULL, root);
1564         mutex_unlock(&fs_info->fs_mutex);
1565
1566         btrfs_transaction_flush_work(root);
1567
1568         if (fs_info->delalloc_bytes) {
1569                 printk("btrfs: at unmount delalloc count %Lu\n",
1570                        fs_info->delalloc_bytes);
1571         }
1572         if (fs_info->extent_root->node)
1573                 free_extent_buffer(fs_info->extent_root->node);
1574
1575         if (fs_info->tree_root->node)
1576                 free_extent_buffer(fs_info->tree_root->node);
1577
1578         if (root->fs_info->chunk_root->node);
1579                 free_extent_buffer(root->fs_info->chunk_root->node);
1580
1581         if (root->fs_info->dev_root->node);
1582                 free_extent_buffer(root->fs_info->dev_root->node);
1583
1584         btrfs_free_block_groups(root->fs_info);
1585         del_fs_roots(fs_info);
1586
1587         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1588
1589         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1590         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1591         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1592         extent_io_tree_empty_lru(&fs_info->pending_del);
1593         extent_io_tree_empty_lru(&fs_info->extent_ins);
1594         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1595
1596         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1597
1598         btrfs_stop_workers(&fs_info->workers);
1599         btrfs_stop_workers(&fs_info->endio_workers);
1600
1601         iput(fs_info->btree_inode);
1602 #if 0
1603         while(!list_empty(&fs_info->hashers)) {
1604                 struct btrfs_hasher *hasher;
1605                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1606                                     hashers);
1607                 list_del(&hasher->hashers);
1608                 crypto_free_hash(&fs_info->hash_tfm);
1609                 kfree(hasher);
1610         }
1611 #endif
1612         btrfs_close_devices(fs_info->fs_devices);
1613         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1614
1615 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1616         bdi_destroy(&fs_info->bdi);
1617 #endif
1618
1619         kfree(fs_info->extent_root);
1620         kfree(fs_info->tree_root);
1621         kfree(fs_info->chunk_root);
1622         kfree(fs_info->dev_root);
1623         return 0;
1624 }
1625
1626 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1627 {
1628         int ret;
1629         struct inode *btree_inode = buf->first_page->mapping->host;
1630
1631         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1632         if (!ret)
1633                 return ret;
1634
1635         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1636                                     parent_transid);
1637         return !ret;
1638 }
1639
1640 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1641 {
1642         struct inode *btree_inode = buf->first_page->mapping->host;
1643         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1644                                           buf);
1645 }
1646
1647 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1648 {
1649         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1650         u64 transid = btrfs_header_generation(buf);
1651         struct inode *btree_inode = root->fs_info->btree_inode;
1652
1653         if (transid != root->fs_info->generation) {
1654                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1655                         (unsigned long long)buf->start,
1656                         transid, root->fs_info->generation);
1657                 WARN_ON(1);
1658         }
1659         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1660 }
1661
1662 void btrfs_throttle(struct btrfs_root *root)
1663 {
1664         struct backing_dev_info *bdi;
1665
1666         bdi = &root->fs_info->bdi;
1667         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1668 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1669                 congestion_wait(WRITE, HZ/20);
1670 #else
1671                 blk_congestion_wait(WRITE, HZ/20);
1672 #endif
1673         }
1674 }
1675
1676 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1677 {
1678         /*
1679          * looks as though older kernels can get into trouble with
1680          * this code, they end up stuck in balance_dirty_pages forever
1681          */
1682         struct extent_io_tree *tree;
1683         u64 num_dirty;
1684         u64 start = 0;
1685         unsigned long thresh = 16 * 1024 * 1024;
1686         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1687
1688         if (current_is_pdflush())
1689                 return;
1690
1691         num_dirty = count_range_bits(tree, &start, (u64)-1,
1692                                      thresh, EXTENT_DIRTY);
1693         if (num_dirty > thresh) {
1694                 balance_dirty_pages_ratelimited_nr(
1695                                    root->fs_info->btree_inode->i_mapping, 1);
1696         }
1697         return;
1698 }
1699
1700 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1701 {
1702         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1703         struct inode *btree_inode = root->fs_info->btree_inode;
1704         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1705                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1706 }
1707
1708 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1709 {
1710         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1711         struct inode *btree_inode = root->fs_info->btree_inode;
1712         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1713                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1714                         GFP_NOFS);
1715 }
1716
1717 int btrfs_buffer_defrag(struct extent_buffer *buf)
1718 {
1719         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1720         struct inode *btree_inode = root->fs_info->btree_inode;
1721         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1722                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1723 }
1724
1725 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1726 {
1727         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1728         struct inode *btree_inode = root->fs_info->btree_inode;
1729         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1730                      buf->start, buf->start + buf->len - 1,
1731                      EXTENT_DEFRAG_DONE, 0);
1732 }
1733
1734 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1735 {
1736         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1737         struct inode *btree_inode = root->fs_info->btree_inode;
1738         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1739                      buf->start, buf->start + buf->len - 1,
1740                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1741 }
1742
1743 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1744 {
1745         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1746         struct inode *btree_inode = root->fs_info->btree_inode;
1747         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1748                      buf->start, buf->start + buf->len - 1,
1749                      EXTENT_DEFRAG, GFP_NOFS);
1750 }
1751
1752 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1753 {
1754         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1755         int ret;
1756         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1757         if (ret == 0) {
1758                 buf->flags |= EXTENT_UPTODATE;
1759         }
1760         return ret;
1761 }
1762
1763 static struct extent_io_ops btree_extent_io_ops = {
1764         .writepage_io_hook = btree_writepage_io_hook,
1765         .readpage_end_io_hook = btree_readpage_end_io_hook,
1766         .submit_bio_hook = btree_submit_bio_hook,
1767         /* note we're sharing with inode.c for the merge bio hook */
1768         .merge_bio_hook = btrfs_merge_bio_hook,
1769 };