2 * Copyright (C) 2007 Oracle. All rights reserved.
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.
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.
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.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/version.h>
38 #include <linux/xattr.h>
41 #include "transaction.h"
42 #include "btrfs_inode.h"
44 #include "print-tree.h"
46 #include "ordered-data.h"
48 struct btrfs_iget_args {
50 struct btrfs_root *root;
53 static struct inode_operations btrfs_dir_inode_operations;
54 static struct inode_operations btrfs_symlink_inode_operations;
55 static struct inode_operations btrfs_dir_ro_inode_operations;
56 static struct inode_operations btrfs_special_inode_operations;
57 static struct inode_operations btrfs_file_inode_operations;
58 static struct address_space_operations btrfs_aops;
59 static struct address_space_operations btrfs_symlink_aops;
60 static struct file_operations btrfs_dir_file_operations;
61 static struct extent_io_ops btrfs_extent_io_ops;
63 static struct kmem_cache *btrfs_inode_cachep;
64 struct kmem_cache *btrfs_trans_handle_cachep;
65 struct kmem_cache *btrfs_transaction_cachep;
66 struct kmem_cache *btrfs_bit_radix_cachep;
67 struct kmem_cache *btrfs_path_cachep;
70 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
71 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
72 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
73 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
74 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
75 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
76 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
77 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
80 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
89 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
90 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
91 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
99 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
101 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
105 static int cow_file_range(struct inode *inode, u64 start, u64 end)
107 struct btrfs_root *root = BTRFS_I(inode)->root;
108 struct btrfs_trans_handle *trans;
112 u64 blocksize = root->sectorsize;
114 struct btrfs_key ins;
115 struct extent_map *em;
116 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
119 trans = btrfs_join_transaction(root, 1);
121 btrfs_set_trans_block_group(trans, inode);
123 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
124 num_bytes = max(blocksize, num_bytes);
125 orig_num_bytes = num_bytes;
127 if (alloc_hint == EXTENT_MAP_INLINE)
130 BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
131 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1);
133 while(num_bytes > 0) {
134 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
135 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
136 root->sectorsize, 0, 0,
142 em = alloc_extent_map(GFP_NOFS);
144 em->len = ins.offset;
145 em->block_start = ins.objectid;
146 em->bdev = root->fs_info->fs_devices->latest_bdev;
148 spin_lock(&em_tree->lock);
149 ret = add_extent_mapping(em_tree, em);
150 spin_unlock(&em_tree->lock);
151 if (ret != -EEXIST) {
155 btrfs_drop_extent_cache(inode, start,
156 start + ins.offset - 1);
159 cur_alloc_size = ins.offset;
160 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
163 if (num_bytes < cur_alloc_size) {
164 printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
168 num_bytes -= cur_alloc_size;
169 alloc_hint = ins.objectid + ins.offset;
170 start += cur_alloc_size;
173 btrfs_end_transaction(trans, root);
177 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
185 struct btrfs_root *root = BTRFS_I(inode)->root;
186 struct btrfs_block_group_cache *block_group;
187 struct extent_buffer *leaf;
189 struct btrfs_path *path;
190 struct btrfs_file_extent_item *item;
193 struct btrfs_key found_key;
195 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
196 path = btrfs_alloc_path();
199 ret = btrfs_lookup_file_extent(NULL, root, path,
200 inode->i_ino, start, 0);
202 btrfs_free_path(path);
208 if (path->slots[0] == 0)
213 leaf = path->nodes[0];
214 item = btrfs_item_ptr(leaf, path->slots[0],
215 struct btrfs_file_extent_item);
217 /* are we inside the extent that was found? */
218 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
219 found_type = btrfs_key_type(&found_key);
220 if (found_key.objectid != inode->i_ino ||
221 found_type != BTRFS_EXTENT_DATA_KEY)
224 found_type = btrfs_file_extent_type(leaf, item);
225 extent_start = found_key.offset;
226 if (found_type == BTRFS_FILE_EXTENT_REG) {
227 u64 extent_num_bytes;
229 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
230 extent_end = extent_start + extent_num_bytes;
233 if (loops && start != extent_start)
236 if (start < extent_start || start >= extent_end)
239 cow_end = min(end, extent_end - 1);
240 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
244 if (btrfs_count_snapshots_in_path(root, path, inode->i_ino,
250 * we may be called by the resizer, make sure we're inside
251 * the limits of the FS
253 block_group = btrfs_lookup_block_group(root->fs_info,
255 if (!block_group || block_group->ro)
264 btrfs_free_path(path);
267 btrfs_release_path(root, path);
272 cow_file_range(inode, start, end);
277 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
279 struct btrfs_root *root = BTRFS_I(inode)->root;
282 if (btrfs_test_opt(root, NODATACOW) ||
283 btrfs_test_flag(inode, NODATACOW))
284 ret = run_delalloc_nocow(inode, start, end);
286 ret = cow_file_range(inode, start, end);
291 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
292 unsigned long old, unsigned long bits)
295 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
296 struct btrfs_root *root = BTRFS_I(inode)->root;
297 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
298 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
299 root->fs_info->delalloc_bytes += end - start + 1;
300 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
305 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
306 unsigned long old, unsigned long bits)
308 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
309 struct btrfs_root *root = BTRFS_I(inode)->root;
312 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
313 if (end - start + 1 > root->fs_info->delalloc_bytes) {
314 printk("warning: delalloc account %Lu %Lu\n",
315 end - start + 1, root->fs_info->delalloc_bytes);
316 root->fs_info->delalloc_bytes = 0;
317 BTRFS_I(inode)->delalloc_bytes = 0;
319 root->fs_info->delalloc_bytes -= end - start + 1;
320 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
322 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
327 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
328 size_t size, struct bio *bio)
330 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
331 struct btrfs_mapping_tree *map_tree;
332 u64 logical = bio->bi_sector << 9;
337 length = bio->bi_size;
338 map_tree = &root->fs_info->mapping_tree;
340 ret = btrfs_map_block(map_tree, READ, logical,
341 &map_length, NULL, 0);
343 if (map_length < length + size) {
349 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
352 struct btrfs_root *root = BTRFS_I(inode)->root;
355 ret = btrfs_csum_one_bio(root, inode, bio);
358 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
361 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
364 struct btrfs_root *root = BTRFS_I(inode)->root;
367 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
370 if (!(rw & (1 << BIO_RW))) {
374 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
375 inode, rw, bio, mirror_num,
376 __btrfs_submit_bio_hook);
378 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
381 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
382 struct inode *inode, u64 file_offset,
383 struct list_head *list)
385 struct list_head *cur;
386 struct btrfs_ordered_sum *sum;
388 btrfs_set_trans_block_group(trans, inode);
389 list_for_each(cur, list) {
390 sum = list_entry(cur, struct btrfs_ordered_sum, list);
391 mutex_lock(&BTRFS_I(inode)->csum_mutex);
392 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
394 mutex_unlock(&BTRFS_I(inode)->csum_mutex);
399 struct btrfs_writepage_fixup {
401 struct btrfs_work work;
404 /* see btrfs_writepage_start_hook for details on why this is required */
405 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
407 struct btrfs_writepage_fixup *fixup;
408 struct btrfs_ordered_extent *ordered;
414 fixup = container_of(work, struct btrfs_writepage_fixup, work);
418 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
419 ClearPageChecked(page);
423 inode = page->mapping->host;
424 page_start = page_offset(page);
425 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
427 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
428 ordered = btrfs_lookup_ordered_extent(inode, page_start);
432 set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start, page_end,
434 ClearPageChecked(page);
436 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
439 page_cache_release(page);
443 * There are a few paths in the higher layers of the kernel that directly
444 * set the page dirty bit without asking the filesystem if it is a
445 * good idea. This causes problems because we want to make sure COW
446 * properly happens and the data=ordered rules are followed.
448 * In our case any range that doesn't have the EXTENT_ORDERED bit set
449 * hasn't been properly setup for IO. We kick off an async process
450 * to fix it up. The async helper will wait for ordered extents, set
451 * the delalloc bit and make it safe to write the page.
453 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
455 struct inode *inode = page->mapping->host;
456 struct btrfs_writepage_fixup *fixup;
457 struct btrfs_root *root = BTRFS_I(inode)->root;
460 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
465 if (PageChecked(page))
468 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
471 printk("queueing worker to fixup page %lu %Lu\n", inode->i_ino, page_offset(page));
472 SetPageChecked(page);
473 page_cache_get(page);
474 fixup->work.func = btrfs_writepage_fixup_worker;
476 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
480 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
481 struct extent_state *state, int uptodate)
483 struct inode *inode = page->mapping->host;
484 struct btrfs_root *root = BTRFS_I(inode)->root;
485 struct btrfs_trans_handle *trans;
486 struct btrfs_ordered_extent *ordered_extent;
487 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
489 struct list_head list;
490 struct btrfs_key ins;
493 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
497 trans = btrfs_join_transaction(root, 1);
499 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
500 BUG_ON(!ordered_extent);
502 lock_extent(io_tree, ordered_extent->file_offset,
503 ordered_extent->file_offset + ordered_extent->len - 1,
506 INIT_LIST_HEAD(&list);
508 ins.objectid = ordered_extent->start;
509 ins.offset = ordered_extent->len;
510 ins.type = BTRFS_EXTENT_ITEM_KEY;
511 ret = btrfs_alloc_reserved_extent(trans, root, root->root_key.objectid,
512 trans->transid, inode->i_ino,
513 ordered_extent->file_offset, &ins);
516 mutex_lock(&BTRFS_I(inode)->extent_mutex);
517 ret = btrfs_drop_extents(trans, root, inode,
518 ordered_extent->file_offset,
519 ordered_extent->file_offset +
521 ordered_extent->file_offset, &alloc_hint);
523 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
524 ordered_extent->file_offset,
525 ordered_extent->start,
527 ordered_extent->len, 0);
529 btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
530 ordered_extent->file_offset +
531 ordered_extent->len - 1);
532 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
534 inode->i_blocks += ordered_extent->len >> 9;
535 unlock_extent(io_tree, ordered_extent->file_offset,
536 ordered_extent->file_offset + ordered_extent->len - 1,
538 add_pending_csums(trans, inode, ordered_extent->file_offset,
539 &ordered_extent->list);
541 btrfs_ordered_update_i_size(inode, ordered_extent);
542 btrfs_remove_ordered_extent(inode, ordered_extent);
544 btrfs_put_ordered_extent(ordered_extent);
545 /* once for the tree */
546 btrfs_put_ordered_extent(ordered_extent);
548 btrfs_update_inode(trans, root, inode);
549 btrfs_end_transaction(trans, root);
553 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
556 struct inode *inode = page->mapping->host;
557 struct btrfs_root *root = BTRFS_I(inode)->root;
558 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
559 struct btrfs_csum_item *item;
560 struct btrfs_path *path = NULL;
563 if (btrfs_test_opt(root, NODATASUM) ||
564 btrfs_test_flag(inode, NODATASUM))
567 path = btrfs_alloc_path();
568 item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
571 * It is possible there is an ordered extent that has
572 * not yet finished for this range in the file. If so,
573 * that extent will have a csum cached, and it will insert
574 * the sum after all the blocks in the extent are fully
575 * on disk. So, look for an ordered extent and use the
578 ret = btrfs_find_ordered_sum(inode, start, &csum);
583 /* a csum that isn't present is a preallocated region. */
584 if (ret == -ENOENT || ret == -EFBIG)
587 printk("no csum found for inode %lu start %Lu\n", inode->i_ino,
591 read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
594 set_state_private(io_tree, start, csum);
597 btrfs_free_path(path);
601 struct io_failure_record {
609 int btrfs_io_failed_hook(struct bio *failed_bio,
610 struct page *page, u64 start, u64 end,
611 struct extent_state *state)
613 struct io_failure_record *failrec = NULL;
615 struct extent_map *em;
616 struct inode *inode = page->mapping->host;
617 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
618 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
625 ret = get_state_private(failure_tree, start, &private);
627 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
630 failrec->start = start;
631 failrec->len = end - start + 1;
632 failrec->last_mirror = 0;
634 spin_lock(&em_tree->lock);
635 em = lookup_extent_mapping(em_tree, start, failrec->len);
636 if (em->start > start || em->start + em->len < start) {
640 spin_unlock(&em_tree->lock);
642 if (!em || IS_ERR(em)) {
646 logical = start - em->start;
647 logical = em->block_start + logical;
648 failrec->logical = logical;
650 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
651 EXTENT_DIRTY, GFP_NOFS);
652 set_state_private(failure_tree, start,
653 (u64)(unsigned long)failrec);
655 failrec = (struct io_failure_record *)(unsigned long)private;
657 num_copies = btrfs_num_copies(
658 &BTRFS_I(inode)->root->fs_info->mapping_tree,
659 failrec->logical, failrec->len);
660 failrec->last_mirror++;
662 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
663 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
666 if (state && state->start != failrec->start)
668 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
670 if (!state || failrec->last_mirror > num_copies) {
671 set_state_private(failure_tree, failrec->start, 0);
672 clear_extent_bits(failure_tree, failrec->start,
673 failrec->start + failrec->len - 1,
674 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
678 bio = bio_alloc(GFP_NOFS, 1);
679 bio->bi_private = state;
680 bio->bi_end_io = failed_bio->bi_end_io;
681 bio->bi_sector = failrec->logical >> 9;
682 bio->bi_bdev = failed_bio->bi_bdev;
684 bio_add_page(bio, page, failrec->len, start - page_offset(page));
685 if (failed_bio->bi_rw & (1 << BIO_RW))
690 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
691 failrec->last_mirror);
695 int btrfs_clean_io_failures(struct inode *inode, u64 start)
699 struct io_failure_record *failure;
703 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
704 (u64)-1, 1, EXTENT_DIRTY)) {
705 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
706 start, &private_failure);
708 failure = (struct io_failure_record *)(unsigned long)
710 set_state_private(&BTRFS_I(inode)->io_failure_tree,
712 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
714 failure->start + failure->len - 1,
715 EXTENT_DIRTY | EXTENT_LOCKED,
723 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
724 struct extent_state *state)
726 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
727 struct inode *inode = page->mapping->host;
728 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
730 u64 private = ~(u32)0;
732 struct btrfs_root *root = BTRFS_I(inode)->root;
736 if (btrfs_test_opt(root, NODATASUM) ||
737 btrfs_test_flag(inode, NODATASUM))
739 if (state && state->start == start) {
740 private = state->private;
743 ret = get_state_private(io_tree, start, &private);
745 local_irq_save(flags);
746 kaddr = kmap_atomic(page, KM_IRQ0);
750 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
751 btrfs_csum_final(csum, (char *)&csum);
752 if (csum != private) {
755 kunmap_atomic(kaddr, KM_IRQ0);
756 local_irq_restore(flags);
758 /* if the io failure tree for this inode is non-empty,
759 * check to see if we've recovered from a failed IO
761 btrfs_clean_io_failures(inode, start);
765 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
766 page->mapping->host->i_ino, (unsigned long long)start, csum,
768 memset(kaddr + offset, 1, end - start + 1);
769 flush_dcache_page(page);
770 kunmap_atomic(kaddr, KM_IRQ0);
771 local_irq_restore(flags);
777 void btrfs_read_locked_inode(struct inode *inode)
779 struct btrfs_path *path;
780 struct extent_buffer *leaf;
781 struct btrfs_inode_item *inode_item;
782 struct btrfs_timespec *tspec;
783 struct btrfs_root *root = BTRFS_I(inode)->root;
784 struct btrfs_key location;
785 u64 alloc_group_block;
789 path = btrfs_alloc_path();
791 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
793 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
797 leaf = path->nodes[0];
798 inode_item = btrfs_item_ptr(leaf, path->slots[0],
799 struct btrfs_inode_item);
801 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
802 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
803 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
804 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
805 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
807 tspec = btrfs_inode_atime(inode_item);
808 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
809 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
811 tspec = btrfs_inode_mtime(inode_item);
812 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
813 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
815 tspec = btrfs_inode_ctime(inode_item);
816 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
817 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
819 inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
820 inode->i_generation = btrfs_inode_generation(leaf, inode_item);
822 rdev = btrfs_inode_rdev(leaf, inode_item);
824 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
825 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
827 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
828 if (!BTRFS_I(inode)->block_group) {
829 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
831 BTRFS_BLOCK_GROUP_METADATA, 0);
833 btrfs_free_path(path);
836 switch (inode->i_mode & S_IFMT) {
838 inode->i_mapping->a_ops = &btrfs_aops;
839 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
840 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
841 inode->i_fop = &btrfs_file_operations;
842 inode->i_op = &btrfs_file_inode_operations;
845 inode->i_fop = &btrfs_dir_file_operations;
846 if (root == root->fs_info->tree_root)
847 inode->i_op = &btrfs_dir_ro_inode_operations;
849 inode->i_op = &btrfs_dir_inode_operations;
852 inode->i_op = &btrfs_symlink_inode_operations;
853 inode->i_mapping->a_ops = &btrfs_symlink_aops;
854 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
857 init_special_inode(inode, inode->i_mode, rdev);
863 btrfs_free_path(path);
864 make_bad_inode(inode);
867 static void fill_inode_item(struct extent_buffer *leaf,
868 struct btrfs_inode_item *item,
871 btrfs_set_inode_uid(leaf, item, inode->i_uid);
872 btrfs_set_inode_gid(leaf, item, inode->i_gid);
873 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
874 btrfs_set_inode_mode(leaf, item, inode->i_mode);
875 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
877 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
878 inode->i_atime.tv_sec);
879 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
880 inode->i_atime.tv_nsec);
882 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
883 inode->i_mtime.tv_sec);
884 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
885 inode->i_mtime.tv_nsec);
887 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
888 inode->i_ctime.tv_sec);
889 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
890 inode->i_ctime.tv_nsec);
892 btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
893 btrfs_set_inode_generation(leaf, item, inode->i_generation);
894 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
895 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
896 btrfs_set_inode_block_group(leaf, item,
897 BTRFS_I(inode)->block_group->key.objectid);
900 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
901 struct btrfs_root *root,
904 struct btrfs_inode_item *inode_item;
905 struct btrfs_path *path;
906 struct extent_buffer *leaf;
909 path = btrfs_alloc_path();
911 ret = btrfs_lookup_inode(trans, root, path,
912 &BTRFS_I(inode)->location, 1);
919 leaf = path->nodes[0];
920 inode_item = btrfs_item_ptr(leaf, path->slots[0],
921 struct btrfs_inode_item);
923 fill_inode_item(leaf, inode_item, inode);
924 btrfs_mark_buffer_dirty(leaf);
925 btrfs_set_inode_last_trans(trans, inode);
928 btrfs_free_path(path);
933 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
934 struct btrfs_root *root,
936 struct dentry *dentry)
938 struct btrfs_path *path;
939 const char *name = dentry->d_name.name;
940 int name_len = dentry->d_name.len;
942 struct extent_buffer *leaf;
943 struct btrfs_dir_item *di;
944 struct btrfs_key key;
946 path = btrfs_alloc_path();
952 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
962 leaf = path->nodes[0];
963 btrfs_dir_item_key_to_cpu(leaf, di, &key);
964 ret = btrfs_delete_one_dir_name(trans, root, path, di);
967 btrfs_release_path(root, path);
969 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
970 key.objectid, name, name_len, -1);
979 ret = btrfs_delete_one_dir_name(trans, root, path, di);
980 btrfs_release_path(root, path);
982 dentry->d_inode->i_ctime = dir->i_ctime;
983 ret = btrfs_del_inode_ref(trans, root, name, name_len,
984 dentry->d_inode->i_ino,
985 dentry->d_parent->d_inode->i_ino);
987 printk("failed to delete reference to %.*s, "
988 "inode %lu parent %lu\n", name_len, name,
989 dentry->d_inode->i_ino,
990 dentry->d_parent->d_inode->i_ino);
993 btrfs_free_path(path);
995 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
996 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
997 btrfs_update_inode(trans, root, dir);
998 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
999 dentry->d_inode->i_nlink--;
1001 drop_nlink(dentry->d_inode);
1003 ret = btrfs_update_inode(trans, root, dentry->d_inode);
1004 dir->i_sb->s_dirt = 1;
1009 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1011 struct btrfs_root *root;
1012 struct btrfs_trans_handle *trans;
1014 unsigned long nr = 0;
1016 root = BTRFS_I(dir)->root;
1018 ret = btrfs_check_free_space(root, 1, 1);
1022 trans = btrfs_start_transaction(root, 1);
1024 btrfs_set_trans_block_group(trans, dir);
1025 ret = btrfs_unlink_trans(trans, root, dir, dentry);
1026 nr = trans->blocks_used;
1028 btrfs_end_transaction_throttle(trans, root);
1030 btrfs_btree_balance_dirty(root, nr);
1034 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1036 struct inode *inode = dentry->d_inode;
1039 struct btrfs_root *root = BTRFS_I(dir)->root;
1040 struct btrfs_trans_handle *trans;
1041 unsigned long nr = 0;
1043 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1047 ret = btrfs_check_free_space(root, 1, 1);
1051 trans = btrfs_start_transaction(root, 1);
1052 btrfs_set_trans_block_group(trans, dir);
1054 /* now the directory is empty */
1055 err = btrfs_unlink_trans(trans, root, dir, dentry);
1057 btrfs_i_size_write(inode, 0);
1060 nr = trans->blocks_used;
1061 ret = btrfs_end_transaction_throttle(trans, root);
1063 btrfs_btree_balance_dirty(root, nr);
1071 * this can truncate away extent items, csum items and directory items.
1072 * It starts at a high offset and removes keys until it can't find
1073 * any higher than i_size.
1075 * csum items that cross the new i_size are truncated to the new size
1078 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
1079 struct btrfs_root *root,
1080 struct inode *inode,
1084 struct btrfs_path *path;
1085 struct btrfs_key key;
1086 struct btrfs_key found_key;
1088 struct extent_buffer *leaf;
1089 struct btrfs_file_extent_item *fi;
1090 u64 extent_start = 0;
1091 u64 extent_num_bytes = 0;
1097 int pending_del_nr = 0;
1098 int pending_del_slot = 0;
1099 int extent_type = -1;
1100 u64 mask = root->sectorsize - 1;
1102 btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
1103 path = btrfs_alloc_path();
1107 /* FIXME, add redo link to tree so we don't leak on crash */
1108 key.objectid = inode->i_ino;
1109 key.offset = (u64)-1;
1112 btrfs_init_path(path);
1114 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1119 BUG_ON(path->slots[0] == 0);
1125 leaf = path->nodes[0];
1126 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1127 found_type = btrfs_key_type(&found_key);
1129 if (found_key.objectid != inode->i_ino)
1132 if (found_type < min_type)
1135 item_end = found_key.offset;
1136 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1137 fi = btrfs_item_ptr(leaf, path->slots[0],
1138 struct btrfs_file_extent_item);
1139 extent_type = btrfs_file_extent_type(leaf, fi);
1140 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1142 btrfs_file_extent_num_bytes(leaf, fi);
1143 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1144 struct btrfs_item *item = btrfs_item_nr(leaf,
1146 item_end += btrfs_file_extent_inline_len(leaf,
1151 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1152 ret = btrfs_csum_truncate(trans, root, path,
1156 if (item_end < inode->i_size) {
1157 if (found_type == BTRFS_DIR_ITEM_KEY) {
1158 found_type = BTRFS_INODE_ITEM_KEY;
1159 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1160 found_type = BTRFS_CSUM_ITEM_KEY;
1161 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1162 found_type = BTRFS_XATTR_ITEM_KEY;
1163 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1164 found_type = BTRFS_INODE_REF_KEY;
1165 } else if (found_type) {
1170 btrfs_set_key_type(&key, found_type);
1173 if (found_key.offset >= inode->i_size)
1179 /* FIXME, shrink the extent if the ref count is only 1 */
1180 if (found_type != BTRFS_EXTENT_DATA_KEY)
1183 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1185 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1187 u64 orig_num_bytes =
1188 btrfs_file_extent_num_bytes(leaf, fi);
1189 extent_num_bytes = inode->i_size -
1190 found_key.offset + root->sectorsize - 1;
1191 extent_num_bytes = extent_num_bytes &
1192 ~((u64)root->sectorsize - 1);
1193 btrfs_set_file_extent_num_bytes(leaf, fi,
1195 num_dec = (orig_num_bytes -
1197 if (extent_start != 0)
1198 dec_i_blocks(inode, num_dec);
1199 btrfs_mark_buffer_dirty(leaf);
1202 btrfs_file_extent_disk_num_bytes(leaf,
1204 /* FIXME blocksize != 4096 */
1205 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1206 if (extent_start != 0) {
1208 dec_i_blocks(inode, num_dec);
1210 root_gen = btrfs_header_generation(leaf);
1211 root_owner = btrfs_header_owner(leaf);
1213 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1215 u32 newsize = inode->i_size - found_key.offset;
1216 dec_i_blocks(inode, item_end + 1 -
1217 found_key.offset - newsize);
1219 btrfs_file_extent_calc_inline_size(newsize);
1220 ret = btrfs_truncate_item(trans, root, path,
1224 dec_i_blocks(inode, item_end + 1 -
1230 if (!pending_del_nr) {
1231 /* no pending yet, add ourselves */
1232 pending_del_slot = path->slots[0];
1234 } else if (pending_del_nr &&
1235 path->slots[0] + 1 == pending_del_slot) {
1236 /* hop on the pending chunk */
1238 pending_del_slot = path->slots[0];
1240 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1246 ret = btrfs_free_extent(trans, root, extent_start,
1249 root_gen, inode->i_ino,
1250 found_key.offset, 0);
1254 if (path->slots[0] == 0) {
1257 btrfs_release_path(root, path);
1262 if (pending_del_nr &&
1263 path->slots[0] + 1 != pending_del_slot) {
1264 struct btrfs_key debug;
1266 btrfs_item_key_to_cpu(path->nodes[0], &debug,
1268 ret = btrfs_del_items(trans, root, path,
1273 btrfs_release_path(root, path);
1279 if (pending_del_nr) {
1280 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1283 btrfs_free_path(path);
1284 inode->i_sb->s_dirt = 1;
1289 * taken from block_truncate_page, but does cow as it zeros out
1290 * any bytes left in the last page in the file.
1292 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1294 struct inode *inode = mapping->host;
1295 struct btrfs_root *root = BTRFS_I(inode)->root;
1296 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1297 struct btrfs_ordered_extent *ordered;
1299 u32 blocksize = root->sectorsize;
1300 pgoff_t index = from >> PAGE_CACHE_SHIFT;
1301 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1307 if ((offset & (blocksize - 1)) == 0)
1312 page = grab_cache_page(mapping, index);
1316 page_start = page_offset(page);
1317 page_end = page_start + PAGE_CACHE_SIZE - 1;
1319 if (!PageUptodate(page)) {
1320 ret = btrfs_readpage(NULL, page);
1322 if (page->mapping != mapping) {
1324 page_cache_release(page);
1327 if (!PageUptodate(page)) {
1332 wait_on_page_writeback(page);
1334 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1335 set_page_extent_mapped(page);
1337 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1339 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1341 page_cache_release(page);
1342 btrfs_start_ordered_extent(inode, ordered, 1);
1343 btrfs_put_ordered_extent(ordered);
1347 set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
1348 page_end, GFP_NOFS);
1350 if (offset != PAGE_CACHE_SIZE) {
1352 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1353 flush_dcache_page(page);
1356 ClearPageChecked(page);
1357 set_page_dirty(page);
1358 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1361 page_cache_release(page);
1366 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1368 struct inode *inode = dentry->d_inode;
1371 err = inode_change_ok(inode, attr);
1375 if (S_ISREG(inode->i_mode) &&
1376 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1377 struct btrfs_trans_handle *trans;
1378 struct btrfs_root *root = BTRFS_I(inode)->root;
1379 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1381 u64 mask = root->sectorsize - 1;
1382 u64 hole_start = (inode->i_size + mask) & ~mask;
1383 u64 block_end = (attr->ia_size + mask) & ~mask;
1387 if (attr->ia_size <= hole_start)
1390 err = btrfs_check_free_space(root, 1, 0);
1394 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1396 hole_size = block_end - hole_start;
1397 btrfs_wait_ordered_range(inode, hole_start, hole_size);
1398 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1400 trans = btrfs_start_transaction(root, 1);
1401 btrfs_set_trans_block_group(trans, inode);
1402 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1403 err = btrfs_drop_extents(trans, root, inode,
1404 hole_start, block_end, hole_start,
1407 if (alloc_hint != EXTENT_MAP_INLINE) {
1408 err = btrfs_insert_file_extent(trans, root,
1412 btrfs_drop_extent_cache(inode, hole_start,
1414 btrfs_check_file(root, inode);
1416 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1417 btrfs_end_transaction(trans, root);
1418 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1423 err = inode_setattr(inode, attr);
1428 void btrfs_delete_inode(struct inode *inode)
1430 struct btrfs_trans_handle *trans;
1431 struct btrfs_root *root = BTRFS_I(inode)->root;
1435 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1436 truncate_inode_pages(&inode->i_data, 0);
1437 if (is_bad_inode(inode)) {
1441 btrfs_i_size_write(inode, 0);
1442 trans = btrfs_start_transaction(root, 1);
1444 btrfs_set_trans_block_group(trans, inode);
1445 ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1447 goto no_delete_lock;
1449 nr = trans->blocks_used;
1452 btrfs_end_transaction(trans, root);
1453 btrfs_btree_balance_dirty(root, nr);
1457 nr = trans->blocks_used;
1458 btrfs_end_transaction(trans, root);
1459 btrfs_btree_balance_dirty(root, nr);
1465 * this returns the key found in the dir entry in the location pointer.
1466 * If no dir entries were found, location->objectid is 0.
1468 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1469 struct btrfs_key *location)
1471 const char *name = dentry->d_name.name;
1472 int namelen = dentry->d_name.len;
1473 struct btrfs_dir_item *di;
1474 struct btrfs_path *path;
1475 struct btrfs_root *root = BTRFS_I(dir)->root;
1478 if (namelen == 1 && strcmp(name, ".") == 0) {
1479 location->objectid = dir->i_ino;
1480 location->type = BTRFS_INODE_ITEM_KEY;
1481 location->offset = 0;
1484 path = btrfs_alloc_path();
1487 if (namelen == 2 && strcmp(name, "..") == 0) {
1488 struct btrfs_key key;
1489 struct extent_buffer *leaf;
1493 key.objectid = dir->i_ino;
1494 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1496 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1500 leaf = path->nodes[0];
1501 slot = path->slots[0];
1502 nritems = btrfs_header_nritems(leaf);
1503 if (slot >= nritems)
1506 btrfs_item_key_to_cpu(leaf, &key, slot);
1507 if (key.objectid != dir->i_ino ||
1508 key.type != BTRFS_INODE_REF_KEY) {
1511 location->objectid = key.offset;
1512 location->type = BTRFS_INODE_ITEM_KEY;
1513 location->offset = 0;
1517 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1521 if (!di || IS_ERR(di)) {
1524 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1526 btrfs_free_path(path);
1529 location->objectid = 0;
1534 * when we hit a tree root in a directory, the btrfs part of the inode
1535 * needs to be changed to reflect the root directory of the tree root. This
1536 * is kind of like crossing a mount point.
1538 static int fixup_tree_root_location(struct btrfs_root *root,
1539 struct btrfs_key *location,
1540 struct btrfs_root **sub_root,
1541 struct dentry *dentry)
1543 struct btrfs_path *path;
1544 struct btrfs_root_item *ri;
1546 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1548 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1551 path = btrfs_alloc_path();
1554 *sub_root = btrfs_read_fs_root(root->fs_info, location,
1555 dentry->d_name.name,
1556 dentry->d_name.len);
1557 if (IS_ERR(*sub_root))
1558 return PTR_ERR(*sub_root);
1560 ri = &(*sub_root)->root_item;
1561 location->objectid = btrfs_root_dirid(ri);
1562 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1563 location->offset = 0;
1565 btrfs_free_path(path);
1569 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1571 struct btrfs_iget_args *args = p;
1572 inode->i_ino = args->ino;
1573 BTRFS_I(inode)->root = args->root;
1574 BTRFS_I(inode)->delalloc_bytes = 0;
1575 BTRFS_I(inode)->disk_i_size = 0;
1576 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1577 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1578 inode->i_mapping, GFP_NOFS);
1579 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1580 inode->i_mapping, GFP_NOFS);
1581 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1582 mutex_init(&BTRFS_I(inode)->csum_mutex);
1583 mutex_init(&BTRFS_I(inode)->extent_mutex);
1587 static int btrfs_find_actor(struct inode *inode, void *opaque)
1589 struct btrfs_iget_args *args = opaque;
1590 return (args->ino == inode->i_ino &&
1591 args->root == BTRFS_I(inode)->root);
1594 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1597 struct btrfs_iget_args args;
1598 args.ino = objectid;
1599 args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1604 return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1607 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1608 struct btrfs_root *root)
1610 struct inode *inode;
1611 struct btrfs_iget_args args;
1612 args.ino = objectid;
1615 inode = iget5_locked(s, objectid, btrfs_find_actor,
1616 btrfs_init_locked_inode,
1621 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1622 struct nameidata *nd)
1624 struct inode * inode;
1625 struct btrfs_inode *bi = BTRFS_I(dir);
1626 struct btrfs_root *root = bi->root;
1627 struct btrfs_root *sub_root = root;
1628 struct btrfs_key location;
1631 if (dentry->d_name.len > BTRFS_NAME_LEN)
1632 return ERR_PTR(-ENAMETOOLONG);
1634 ret = btrfs_inode_by_name(dir, dentry, &location);
1637 return ERR_PTR(ret);
1640 if (location.objectid) {
1641 ret = fixup_tree_root_location(root, &location, &sub_root,
1644 return ERR_PTR(ret);
1646 return ERR_PTR(-ENOENT);
1647 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1650 return ERR_PTR(-EACCES);
1651 if (inode->i_state & I_NEW) {
1652 /* the inode and parent dir are two different roots */
1653 if (sub_root != root) {
1655 sub_root->inode = inode;
1657 BTRFS_I(inode)->root = sub_root;
1658 memcpy(&BTRFS_I(inode)->location, &location,
1660 btrfs_read_locked_inode(inode);
1661 unlock_new_inode(inode);
1664 return d_splice_alias(inode, dentry);
1667 static unsigned char btrfs_filetype_table[] = {
1668 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1671 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1673 struct inode *inode = filp->f_dentry->d_inode;
1674 struct btrfs_root *root = BTRFS_I(inode)->root;
1675 struct btrfs_item *item;
1676 struct btrfs_dir_item *di;
1677 struct btrfs_key key;
1678 struct btrfs_key found_key;
1679 struct btrfs_path *path;
1682 struct extent_buffer *leaf;
1685 unsigned char d_type;
1690 int key_type = BTRFS_DIR_INDEX_KEY;
1695 /* FIXME, use a real flag for deciding about the key type */
1696 if (root->fs_info->tree_root == root)
1697 key_type = BTRFS_DIR_ITEM_KEY;
1699 /* special case for "." */
1700 if (filp->f_pos == 0) {
1701 over = filldir(dirent, ".", 1,
1709 key.objectid = inode->i_ino;
1710 path = btrfs_alloc_path();
1713 /* special case for .., just use the back ref */
1714 if (filp->f_pos == 1) {
1715 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1717 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1719 leaf = path->nodes[0];
1720 slot = path->slots[0];
1721 nritems = btrfs_header_nritems(leaf);
1722 if (slot >= nritems) {
1723 btrfs_release_path(root, path);
1724 goto read_dir_items;
1726 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1727 btrfs_release_path(root, path);
1728 if (found_key.objectid != key.objectid ||
1729 found_key.type != BTRFS_INODE_REF_KEY)
1730 goto read_dir_items;
1731 over = filldir(dirent, "..", 2,
1732 2, found_key.offset, DT_DIR);
1739 btrfs_set_key_type(&key, key_type);
1740 key.offset = filp->f_pos;
1742 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1747 leaf = path->nodes[0];
1748 nritems = btrfs_header_nritems(leaf);
1749 slot = path->slots[0];
1750 if (advance || slot >= nritems) {
1751 if (slot >= nritems -1) {
1752 ret = btrfs_next_leaf(root, path);
1755 leaf = path->nodes[0];
1756 nritems = btrfs_header_nritems(leaf);
1757 slot = path->slots[0];
1764 item = btrfs_item_nr(leaf, slot);
1765 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1767 if (found_key.objectid != key.objectid)
1769 if (btrfs_key_type(&found_key) != key_type)
1771 if (found_key.offset < filp->f_pos)
1774 filp->f_pos = found_key.offset;
1776 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1778 di_total = btrfs_item_size(leaf, item);
1779 while(di_cur < di_total) {
1780 struct btrfs_key location;
1782 name_len = btrfs_dir_name_len(leaf, di);
1783 if (name_len < 32) {
1784 name_ptr = tmp_name;
1786 name_ptr = kmalloc(name_len, GFP_NOFS);
1789 read_extent_buffer(leaf, name_ptr,
1790 (unsigned long)(di + 1), name_len);
1792 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1793 btrfs_dir_item_key_to_cpu(leaf, di, &location);
1794 over = filldir(dirent, name_ptr, name_len,
1799 if (name_ptr != tmp_name)
1804 di_len = btrfs_dir_name_len(leaf, di) +
1805 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1807 di = (struct btrfs_dir_item *)((char *)di + di_len);
1810 if (key_type == BTRFS_DIR_INDEX_KEY)
1811 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1817 btrfs_free_path(path);
1821 int btrfs_write_inode(struct inode *inode, int wait)
1823 struct btrfs_root *root = BTRFS_I(inode)->root;
1824 struct btrfs_trans_handle *trans;
1828 trans = btrfs_join_transaction(root, 1);
1829 btrfs_set_trans_block_group(trans, inode);
1830 ret = btrfs_commit_transaction(trans, root);
1836 * This is somewhat expensive, updating the tree every time the
1837 * inode changes. But, it is most likely to find the inode in cache.
1838 * FIXME, needs more benchmarking...there are no reasons other than performance
1839 * to keep or drop this code.
1841 void btrfs_dirty_inode(struct inode *inode)
1843 struct btrfs_root *root = BTRFS_I(inode)->root;
1844 struct btrfs_trans_handle *trans;
1846 trans = btrfs_join_transaction(root, 1);
1847 btrfs_set_trans_block_group(trans, inode);
1848 btrfs_update_inode(trans, root, inode);
1849 btrfs_end_transaction(trans, root);
1852 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1853 struct btrfs_root *root,
1854 const char *name, int name_len,
1857 struct btrfs_block_group_cache *group,
1860 struct inode *inode;
1861 struct btrfs_inode_item *inode_item;
1862 struct btrfs_block_group_cache *new_inode_group;
1863 struct btrfs_key *location;
1864 struct btrfs_path *path;
1865 struct btrfs_inode_ref *ref;
1866 struct btrfs_key key[2];
1872 path = btrfs_alloc_path();
1875 inode = new_inode(root->fs_info->sb);
1877 return ERR_PTR(-ENOMEM);
1879 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1880 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1881 inode->i_mapping, GFP_NOFS);
1882 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1883 inode->i_mapping, GFP_NOFS);
1884 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1885 mutex_init(&BTRFS_I(inode)->csum_mutex);
1886 mutex_init(&BTRFS_I(inode)->extent_mutex);
1887 BTRFS_I(inode)->delalloc_bytes = 0;
1888 BTRFS_I(inode)->disk_i_size = 0;
1889 BTRFS_I(inode)->root = root;
1895 new_inode_group = btrfs_find_block_group(root, group, 0,
1896 BTRFS_BLOCK_GROUP_METADATA, owner);
1897 if (!new_inode_group) {
1898 printk("find_block group failed\n");
1899 new_inode_group = group;
1901 BTRFS_I(inode)->block_group = new_inode_group;
1902 BTRFS_I(inode)->flags = 0;
1904 key[0].objectid = objectid;
1905 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
1908 key[1].objectid = objectid;
1909 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
1910 key[1].offset = ref_objectid;
1912 sizes[0] = sizeof(struct btrfs_inode_item);
1913 sizes[1] = name_len + sizeof(*ref);
1915 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
1919 if (objectid > root->highest_inode)
1920 root->highest_inode = objectid;
1922 inode->i_uid = current->fsuid;
1923 inode->i_gid = current->fsgid;
1924 inode->i_mode = mode;
1925 inode->i_ino = objectid;
1926 inode->i_blocks = 0;
1927 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1928 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1929 struct btrfs_inode_item);
1930 fill_inode_item(path->nodes[0], inode_item, inode);
1932 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1933 struct btrfs_inode_ref);
1934 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
1935 ptr = (unsigned long)(ref + 1);
1936 write_extent_buffer(path->nodes[0], name, ptr, name_len);
1938 btrfs_mark_buffer_dirty(path->nodes[0]);
1939 btrfs_free_path(path);
1941 location = &BTRFS_I(inode)->location;
1942 location->objectid = objectid;
1943 location->offset = 0;
1944 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1946 insert_inode_hash(inode);
1949 btrfs_free_path(path);
1950 return ERR_PTR(ret);
1953 static inline u8 btrfs_inode_type(struct inode *inode)
1955 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1958 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1959 struct dentry *dentry, struct inode *inode,
1963 struct btrfs_key key;
1964 struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1965 struct inode *parent_inode;
1967 key.objectid = inode->i_ino;
1968 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1971 ret = btrfs_insert_dir_item(trans, root,
1972 dentry->d_name.name, dentry->d_name.len,
1973 dentry->d_parent->d_inode->i_ino,
1974 &key, btrfs_inode_type(inode));
1977 ret = btrfs_insert_inode_ref(trans, root,
1978 dentry->d_name.name,
1981 dentry->d_parent->d_inode->i_ino);
1983 parent_inode = dentry->d_parent->d_inode;
1984 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1985 dentry->d_name.len * 2);
1986 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1987 ret = btrfs_update_inode(trans, root,
1988 dentry->d_parent->d_inode);
1993 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1994 struct dentry *dentry, struct inode *inode,
1997 int err = btrfs_add_link(trans, dentry, inode, backref);
1999 d_instantiate(dentry, inode);
2007 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
2008 int mode, dev_t rdev)
2010 struct btrfs_trans_handle *trans;
2011 struct btrfs_root *root = BTRFS_I(dir)->root;
2012 struct inode *inode = NULL;
2016 unsigned long nr = 0;
2018 if (!new_valid_dev(rdev))
2021 err = btrfs_check_free_space(root, 1, 0);
2025 trans = btrfs_start_transaction(root, 1);
2026 btrfs_set_trans_block_group(trans, dir);
2028 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2034 inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2036 dentry->d_parent->d_inode->i_ino, objectid,
2037 BTRFS_I(dir)->block_group, mode);
2038 err = PTR_ERR(inode);
2042 btrfs_set_trans_block_group(trans, inode);
2043 err = btrfs_add_nondir(trans, dentry, inode, 0);
2047 inode->i_op = &btrfs_special_inode_operations;
2048 init_special_inode(inode, inode->i_mode, rdev);
2049 btrfs_update_inode(trans, root, inode);
2051 dir->i_sb->s_dirt = 1;
2052 btrfs_update_inode_block_group(trans, inode);
2053 btrfs_update_inode_block_group(trans, dir);
2055 nr = trans->blocks_used;
2056 btrfs_end_transaction_throttle(trans, root);
2059 inode_dec_link_count(inode);
2062 btrfs_btree_balance_dirty(root, nr);
2066 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2067 int mode, struct nameidata *nd)
2069 struct btrfs_trans_handle *trans;
2070 struct btrfs_root *root = BTRFS_I(dir)->root;
2071 struct inode *inode = NULL;
2074 unsigned long nr = 0;
2077 err = btrfs_check_free_space(root, 1, 0);
2080 trans = btrfs_start_transaction(root, 1);
2081 btrfs_set_trans_block_group(trans, dir);
2083 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2089 inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2091 dentry->d_parent->d_inode->i_ino,
2092 objectid, BTRFS_I(dir)->block_group, mode);
2093 err = PTR_ERR(inode);
2097 btrfs_set_trans_block_group(trans, inode);
2098 err = btrfs_add_nondir(trans, dentry, inode, 0);
2102 inode->i_mapping->a_ops = &btrfs_aops;
2103 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2104 inode->i_fop = &btrfs_file_operations;
2105 inode->i_op = &btrfs_file_inode_operations;
2106 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2107 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2108 inode->i_mapping, GFP_NOFS);
2109 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2110 inode->i_mapping, GFP_NOFS);
2111 mutex_init(&BTRFS_I(inode)->csum_mutex);
2112 mutex_init(&BTRFS_I(inode)->extent_mutex);
2113 BTRFS_I(inode)->delalloc_bytes = 0;
2114 BTRFS_I(inode)->disk_i_size = 0;
2115 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2116 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2118 dir->i_sb->s_dirt = 1;
2119 btrfs_update_inode_block_group(trans, inode);
2120 btrfs_update_inode_block_group(trans, dir);
2122 nr = trans->blocks_used;
2123 btrfs_end_transaction_throttle(trans, root);
2126 inode_dec_link_count(inode);
2129 btrfs_btree_balance_dirty(root, nr);
2133 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2134 struct dentry *dentry)
2136 struct btrfs_trans_handle *trans;
2137 struct btrfs_root *root = BTRFS_I(dir)->root;
2138 struct inode *inode = old_dentry->d_inode;
2139 unsigned long nr = 0;
2143 if (inode->i_nlink == 0)
2146 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2151 err = btrfs_check_free_space(root, 1, 0);
2154 trans = btrfs_start_transaction(root, 1);
2156 btrfs_set_trans_block_group(trans, dir);
2157 atomic_inc(&inode->i_count);
2158 err = btrfs_add_nondir(trans, dentry, inode, 1);
2163 dir->i_sb->s_dirt = 1;
2164 btrfs_update_inode_block_group(trans, dir);
2165 err = btrfs_update_inode(trans, root, inode);
2170 nr = trans->blocks_used;
2171 btrfs_end_transaction_throttle(trans, root);
2174 inode_dec_link_count(inode);
2177 btrfs_btree_balance_dirty(root, nr);
2181 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2183 struct inode *inode = NULL;
2184 struct btrfs_trans_handle *trans;
2185 struct btrfs_root *root = BTRFS_I(dir)->root;
2187 int drop_on_err = 0;
2189 unsigned long nr = 1;
2191 err = btrfs_check_free_space(root, 1, 0);
2195 trans = btrfs_start_transaction(root, 1);
2196 btrfs_set_trans_block_group(trans, dir);
2198 if (IS_ERR(trans)) {
2199 err = PTR_ERR(trans);
2203 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2209 inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2211 dentry->d_parent->d_inode->i_ino, objectid,
2212 BTRFS_I(dir)->block_group, S_IFDIR | mode);
2213 if (IS_ERR(inode)) {
2214 err = PTR_ERR(inode);
2219 inode->i_op = &btrfs_dir_inode_operations;
2220 inode->i_fop = &btrfs_dir_file_operations;
2221 btrfs_set_trans_block_group(trans, inode);
2223 btrfs_i_size_write(inode, 0);
2224 err = btrfs_update_inode(trans, root, inode);
2228 err = btrfs_add_link(trans, dentry, inode, 0);
2232 d_instantiate(dentry, inode);
2234 dir->i_sb->s_dirt = 1;
2235 btrfs_update_inode_block_group(trans, inode);
2236 btrfs_update_inode_block_group(trans, dir);
2239 nr = trans->blocks_used;
2240 btrfs_end_transaction_throttle(trans, root);
2245 btrfs_btree_balance_dirty(root, nr);
2249 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2250 struct extent_map *existing,
2251 struct extent_map *em,
2252 u64 map_start, u64 map_len)
2256 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2257 start_diff = map_start - em->start;
2258 em->start = map_start;
2260 if (em->block_start < EXTENT_MAP_LAST_BYTE)
2261 em->block_start += start_diff;
2262 return add_extent_mapping(em_tree, em);
2265 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2266 size_t pg_offset, u64 start, u64 len,
2272 u64 extent_start = 0;
2274 u64 objectid = inode->i_ino;
2276 struct btrfs_path *path;
2277 struct btrfs_root *root = BTRFS_I(inode)->root;
2278 struct btrfs_file_extent_item *item;
2279 struct extent_buffer *leaf;
2280 struct btrfs_key found_key;
2281 struct extent_map *em = NULL;
2282 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2283 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2284 struct btrfs_trans_handle *trans = NULL;
2286 path = btrfs_alloc_path();
2290 spin_lock(&em_tree->lock);
2291 em = lookup_extent_mapping(em_tree, start, len);
2293 em->bdev = root->fs_info->fs_devices->latest_bdev;
2294 spin_unlock(&em_tree->lock);
2297 if (em->start > start || em->start + em->len <= start)
2298 free_extent_map(em);
2299 else if (em->block_start == EXTENT_MAP_INLINE && page)
2300 free_extent_map(em);
2304 em = alloc_extent_map(GFP_NOFS);
2309 em->bdev = root->fs_info->fs_devices->latest_bdev;
2310 em->start = EXTENT_MAP_HOLE;
2312 ret = btrfs_lookup_file_extent(trans, root, path,
2313 objectid, start, trans != NULL);
2320 if (path->slots[0] == 0)
2325 leaf = path->nodes[0];
2326 item = btrfs_item_ptr(leaf, path->slots[0],
2327 struct btrfs_file_extent_item);
2328 /* are we inside the extent that was found? */
2329 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2330 found_type = btrfs_key_type(&found_key);
2331 if (found_key.objectid != objectid ||
2332 found_type != BTRFS_EXTENT_DATA_KEY) {
2336 found_type = btrfs_file_extent_type(leaf, item);
2337 extent_start = found_key.offset;
2338 if (found_type == BTRFS_FILE_EXTENT_REG) {
2339 extent_end = extent_start +
2340 btrfs_file_extent_num_bytes(leaf, item);
2342 if (start < extent_start || start >= extent_end) {
2344 if (start < extent_start) {
2345 if (start + len <= extent_start)
2347 em->len = extent_end - extent_start;
2353 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2355 em->start = extent_start;
2356 em->len = extent_end - extent_start;
2357 em->block_start = EXTENT_MAP_HOLE;
2360 bytenr += btrfs_file_extent_offset(leaf, item);
2361 em->block_start = bytenr;
2362 em->start = extent_start;
2363 em->len = extent_end - extent_start;
2365 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2370 size_t extent_offset;
2373 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2375 extent_end = (extent_start + size + root->sectorsize - 1) &
2376 ~((u64)root->sectorsize - 1);
2377 if (start < extent_start || start >= extent_end) {
2379 if (start < extent_start) {
2380 if (start + len <= extent_start)
2382 em->len = extent_end - extent_start;
2388 em->block_start = EXTENT_MAP_INLINE;
2391 em->start = extent_start;
2396 page_start = page_offset(page) + pg_offset;
2397 extent_offset = page_start - extent_start;
2398 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2399 size - extent_offset);
2400 em->start = extent_start + extent_offset;
2401 em->len = (copy_size + root->sectorsize - 1) &
2402 ~((u64)root->sectorsize - 1);
2404 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2405 if (create == 0 && !PageUptodate(page)) {
2406 read_extent_buffer(leaf, map + pg_offset, ptr,
2408 flush_dcache_page(page);
2409 } else if (create && PageUptodate(page)) {
2412 free_extent_map(em);
2414 btrfs_release_path(root, path);
2415 trans = btrfs_join_transaction(root, 1);
2418 write_extent_buffer(leaf, map + pg_offset, ptr,
2420 btrfs_mark_buffer_dirty(leaf);
2423 set_extent_uptodate(io_tree, em->start,
2424 extent_map_end(em) - 1, GFP_NOFS);
2427 printk("unkknown found_type %d\n", found_type);
2434 em->block_start = EXTENT_MAP_HOLE;
2436 btrfs_release_path(root, path);
2437 if (em->start > start || extent_map_end(em) <= start) {
2438 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2444 spin_lock(&em_tree->lock);
2445 ret = add_extent_mapping(em_tree, em);
2446 /* it is possible that someone inserted the extent into the tree
2447 * while we had the lock dropped. It is also possible that
2448 * an overlapping map exists in the tree
2450 if (ret == -EEXIST) {
2451 struct extent_map *existing;
2455 existing = lookup_extent_mapping(em_tree, start, len);
2456 if (existing && (existing->start > start ||
2457 existing->start + existing->len <= start)) {
2458 free_extent_map(existing);
2462 existing = lookup_extent_mapping(em_tree, em->start,
2465 err = merge_extent_mapping(em_tree, existing,
2468 free_extent_map(existing);
2470 free_extent_map(em);
2475 printk("failing to insert %Lu %Lu\n",
2477 free_extent_map(em);
2481 free_extent_map(em);
2486 spin_unlock(&em_tree->lock);
2488 btrfs_free_path(path);
2490 ret = btrfs_end_transaction(trans, root);
2496 free_extent_map(em);
2498 return ERR_PTR(err);
2503 #if 0 /* waiting for O_DIRECT reads */
2504 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2505 struct buffer_head *bh_result, int create)
2507 struct extent_map *em;
2508 u64 start = (u64)iblock << inode->i_blkbits;
2509 struct btrfs_multi_bio *multi = NULL;
2510 struct btrfs_root *root = BTRFS_I(inode)->root;
2516 em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2518 if (!em || IS_ERR(em))
2521 if (em->start > start || em->start + em->len <= start) {
2525 if (em->block_start == EXTENT_MAP_INLINE) {
2530 len = em->start + em->len - start;
2531 len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2533 if (em->block_start == EXTENT_MAP_HOLE ||
2534 em->block_start == EXTENT_MAP_DELALLOC) {
2535 bh_result->b_size = len;
2539 logical = start - em->start;
2540 logical = em->block_start + logical;
2543 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2544 logical, &map_length, &multi, 0);
2546 bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2547 bh_result->b_size = min(map_length, len);
2549 bh_result->b_bdev = multi->stripes[0].dev->bdev;
2550 set_buffer_mapped(bh_result);
2553 free_extent_map(em);
2558 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2559 const struct iovec *iov, loff_t offset,
2560 unsigned long nr_segs)
2564 struct file *file = iocb->ki_filp;
2565 struct inode *inode = file->f_mapping->host;
2570 return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2571 offset, nr_segs, btrfs_get_block, NULL);
2575 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2577 return extent_bmap(mapping, iblock, btrfs_get_extent);
2580 int btrfs_readpage(struct file *file, struct page *page)
2582 struct extent_io_tree *tree;
2583 tree = &BTRFS_I(page->mapping->host)->io_tree;
2584 return extent_read_full_page(tree, page, btrfs_get_extent);
2587 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2589 struct extent_io_tree *tree;
2592 if (current->flags & PF_MEMALLOC) {
2593 redirty_page_for_writepage(wbc, page);
2597 tree = &BTRFS_I(page->mapping->host)->io_tree;
2598 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2601 static int btrfs_writepages(struct address_space *mapping,
2602 struct writeback_control *wbc)
2604 struct extent_io_tree *tree;
2605 tree = &BTRFS_I(mapping->host)->io_tree;
2606 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2610 btrfs_readpages(struct file *file, struct address_space *mapping,
2611 struct list_head *pages, unsigned nr_pages)
2613 struct extent_io_tree *tree;
2614 tree = &BTRFS_I(mapping->host)->io_tree;
2615 return extent_readpages(tree, mapping, pages, nr_pages,
2618 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2620 struct extent_io_tree *tree;
2621 struct extent_map_tree *map;
2624 tree = &BTRFS_I(page->mapping->host)->io_tree;
2625 map = &BTRFS_I(page->mapping->host)->extent_tree;
2626 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2628 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
2629 ClearPagePrivate(page);
2630 set_page_private(page, 0);
2631 page_cache_release(page);
2636 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2638 return __btrfs_releasepage(page, gfp_flags);
2641 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2643 struct extent_io_tree *tree;
2644 struct btrfs_ordered_extent *ordered;
2645 u64 page_start = page_offset(page);
2646 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2648 wait_on_page_writeback(page);
2649 tree = &BTRFS_I(page->mapping->host)->io_tree;
2651 btrfs_releasepage(page, GFP_NOFS);
2655 lock_extent(tree, page_start, page_end, GFP_NOFS);
2656 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
2660 * IO on this page will never be started, so we need
2661 * to account for any ordered extents now
2663 clear_extent_bit(tree, page_start, page_end,
2664 EXTENT_DIRTY | EXTENT_DELALLOC |
2665 EXTENT_LOCKED, 1, 0, GFP_NOFS);
2666 btrfs_writepage_end_io_hook(page, page_start,
2668 btrfs_put_ordered_extent(ordered);
2669 lock_extent(tree, page_start, page_end, GFP_NOFS);
2671 clear_extent_bit(tree, page_start, page_end,
2672 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2675 __btrfs_releasepage(page, GFP_NOFS);
2677 if (PagePrivate(page)) {
2678 invalidate_extent_lru(tree, page_offset(page),
2680 ClearPagePrivate(page);
2681 set_page_private(page, 0);
2682 page_cache_release(page);
2687 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2688 * called from a page fault handler when a page is first dirtied. Hence we must
2689 * be careful to check for EOF conditions here. We set the page up correctly
2690 * for a written page which means we get ENOSPC checking when writing into
2691 * holes and correct delalloc and unwritten extent mapping on filesystems that
2692 * support these features.
2694 * We are not allowed to take the i_mutex here so we have to play games to
2695 * protect against truncate races as the page could now be beyond EOF. Because
2696 * vmtruncate() writes the inode size before removing pages, once we have the
2697 * page lock we can determine safely if the page is beyond EOF. If it is not
2698 * beyond EOF, then the page is guaranteed safe against truncation until we
2701 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2703 struct inode *inode = fdentry(vma->vm_file)->d_inode;
2704 struct btrfs_root *root = BTRFS_I(inode)->root;
2705 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2706 struct btrfs_ordered_extent *ordered;
2708 unsigned long zero_start;
2714 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2721 size = i_size_read(inode);
2722 page_start = page_offset(page);
2723 page_end = page_start + PAGE_CACHE_SIZE - 1;
2725 if ((page->mapping != inode->i_mapping) ||
2726 (page_start >= size)) {
2727 /* page got truncated out from underneath us */
2730 wait_on_page_writeback(page);
2732 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2733 set_page_extent_mapped(page);
2736 * we can't set the delalloc bits if there are pending ordered
2737 * extents. Drop our locks and wait for them to finish
2739 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2741 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2743 btrfs_start_ordered_extent(inode, ordered, 1);
2744 btrfs_put_ordered_extent(ordered);
2748 set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
2749 page_end, GFP_NOFS);
2752 /* page is wholly or partially inside EOF */
2753 if (page_start + PAGE_CACHE_SIZE > size)
2754 zero_start = size & ~PAGE_CACHE_MASK;
2756 zero_start = PAGE_CACHE_SIZE;
2758 if (zero_start != PAGE_CACHE_SIZE) {
2760 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
2761 flush_dcache_page(page);
2764 ClearPageChecked(page);
2765 set_page_dirty(page);
2766 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2774 static void btrfs_truncate(struct inode *inode)
2776 struct btrfs_root *root = BTRFS_I(inode)->root;
2778 struct btrfs_trans_handle *trans;
2780 u64 mask = root->sectorsize - 1;
2782 if (!S_ISREG(inode->i_mode))
2784 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2787 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2789 trans = btrfs_start_transaction(root, 1);
2790 btrfs_set_trans_block_group(trans, inode);
2791 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
2792 btrfs_i_size_write(inode, inode->i_size);
2794 /* FIXME, add redo link to tree so we don't leak on crash */
2795 ret = btrfs_truncate_in_trans(trans, root, inode,
2796 BTRFS_EXTENT_DATA_KEY);
2797 btrfs_update_inode(trans, root, inode);
2798 nr = trans->blocks_used;
2800 ret = btrfs_end_transaction_throttle(trans, root);
2802 btrfs_btree_balance_dirty(root, nr);
2806 * Invalidate a single dcache entry at the root of the filesystem.
2807 * Needed after creation of snapshot or subvolume.
2809 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
2812 struct dentry *alias, *entry;
2815 alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
2819 /* change me if btrfs ever gets a d_hash operation */
2820 qstr.hash = full_name_hash(qstr.name, qstr.len);
2821 entry = d_lookup(alias, &qstr);
2824 d_invalidate(entry);
2830 int btrfs_create_subvol_root(struct btrfs_root *new_root,
2831 struct btrfs_trans_handle *trans, u64 new_dirid,
2832 struct btrfs_block_group_cache *block_group)
2834 struct inode *inode;
2837 inode = btrfs_new_inode(trans, new_root, "..", 2, new_dirid,
2838 new_dirid, block_group, S_IFDIR | 0700);
2840 return PTR_ERR(inode);
2841 inode->i_op = &btrfs_dir_inode_operations;
2842 inode->i_fop = &btrfs_dir_file_operations;
2843 new_root->inode = inode;
2845 ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2848 btrfs_i_size_write(inode, 0);
2850 return btrfs_update_inode(trans, new_root, inode);
2853 unsigned long btrfs_force_ra(struct address_space *mapping,
2854 struct file_ra_state *ra, struct file *file,
2855 pgoff_t offset, pgoff_t last_index)
2857 pgoff_t req_size = last_index - offset + 1;
2859 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2860 offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2863 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2864 return offset + req_size;
2868 struct inode *btrfs_alloc_inode(struct super_block *sb)
2870 struct btrfs_inode *ei;
2872 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2876 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2877 return &ei->vfs_inode;
2880 void btrfs_destroy_inode(struct inode *inode)
2882 struct btrfs_ordered_extent *ordered;
2883 WARN_ON(!list_empty(&inode->i_dentry));
2884 WARN_ON(inode->i_data.nrpages);
2887 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
2891 printk("found ordered extent %Lu %Lu\n",
2892 ordered->file_offset, ordered->len);
2893 btrfs_remove_ordered_extent(inode, ordered);
2894 btrfs_put_ordered_extent(ordered);
2895 btrfs_put_ordered_extent(ordered);
2898 btrfs_drop_extent_cache(inode, 0, (u64)-1);
2899 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2902 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2903 static void init_once(struct kmem_cache * cachep, void *foo)
2905 static void init_once(void * foo, struct kmem_cache * cachep,
2906 unsigned long flags)
2909 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2911 inode_init_once(&ei->vfs_inode);
2914 void btrfs_destroy_cachep(void)
2916 if (btrfs_inode_cachep)
2917 kmem_cache_destroy(btrfs_inode_cachep);
2918 if (btrfs_trans_handle_cachep)
2919 kmem_cache_destroy(btrfs_trans_handle_cachep);
2920 if (btrfs_transaction_cachep)
2921 kmem_cache_destroy(btrfs_transaction_cachep);
2922 if (btrfs_bit_radix_cachep)
2923 kmem_cache_destroy(btrfs_bit_radix_cachep);
2924 if (btrfs_path_cachep)
2925 kmem_cache_destroy(btrfs_path_cachep);
2928 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2929 unsigned long extra_flags,
2930 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2931 void (*ctor)(struct kmem_cache *, void *)
2933 void (*ctor)(void *, struct kmem_cache *,
2938 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2939 SLAB_MEM_SPREAD | extra_flags), ctor
2940 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2946 int btrfs_init_cachep(void)
2948 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2949 sizeof(struct btrfs_inode),
2951 if (!btrfs_inode_cachep)
2953 btrfs_trans_handle_cachep =
2954 btrfs_cache_create("btrfs_trans_handle_cache",
2955 sizeof(struct btrfs_trans_handle),
2957 if (!btrfs_trans_handle_cachep)
2959 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2960 sizeof(struct btrfs_transaction),
2962 if (!btrfs_transaction_cachep)
2964 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2965 sizeof(struct btrfs_path),
2967 if (!btrfs_path_cachep)
2969 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2970 SLAB_DESTROY_BY_RCU, NULL);
2971 if (!btrfs_bit_radix_cachep)
2975 btrfs_destroy_cachep();
2979 static int btrfs_getattr(struct vfsmount *mnt,
2980 struct dentry *dentry, struct kstat *stat)
2982 struct inode *inode = dentry->d_inode;
2983 generic_fillattr(inode, stat);
2984 stat->blksize = PAGE_CACHE_SIZE;
2985 stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
2989 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2990 struct inode * new_dir,struct dentry *new_dentry)
2992 struct btrfs_trans_handle *trans;
2993 struct btrfs_root *root = BTRFS_I(old_dir)->root;
2994 struct inode *new_inode = new_dentry->d_inode;
2995 struct inode *old_inode = old_dentry->d_inode;
2996 struct timespec ctime = CURRENT_TIME;
2999 if (S_ISDIR(old_inode->i_mode) && new_inode &&
3000 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3004 ret = btrfs_check_free_space(root, 1, 0);
3008 trans = btrfs_start_transaction(root, 1);
3010 btrfs_set_trans_block_group(trans, new_dir);
3012 old_dentry->d_inode->i_nlink++;
3013 old_dir->i_ctime = old_dir->i_mtime = ctime;
3014 new_dir->i_ctime = new_dir->i_mtime = ctime;
3015 old_inode->i_ctime = ctime;
3017 ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
3022 new_inode->i_ctime = CURRENT_TIME;
3023 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
3027 ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
3032 btrfs_end_transaction(trans, root);
3037 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3038 const char *symname)
3040 struct btrfs_trans_handle *trans;
3041 struct btrfs_root *root = BTRFS_I(dir)->root;
3042 struct btrfs_path *path;
3043 struct btrfs_key key;
3044 struct inode *inode = NULL;
3051 struct btrfs_file_extent_item *ei;
3052 struct extent_buffer *leaf;
3053 unsigned long nr = 0;
3055 name_len = strlen(symname) + 1;
3056 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3057 return -ENAMETOOLONG;
3059 err = btrfs_check_free_space(root, 1, 0);
3063 trans = btrfs_start_transaction(root, 1);
3064 btrfs_set_trans_block_group(trans, dir);
3066 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3072 inode = btrfs_new_inode(trans, root, dentry->d_name.name,
3074 dentry->d_parent->d_inode->i_ino, objectid,
3075 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
3076 err = PTR_ERR(inode);
3080 btrfs_set_trans_block_group(trans, inode);
3081 err = btrfs_add_nondir(trans, dentry, inode, 0);
3085 inode->i_mapping->a_ops = &btrfs_aops;
3086 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3087 inode->i_fop = &btrfs_file_operations;
3088 inode->i_op = &btrfs_file_inode_operations;
3089 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3090 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3091 inode->i_mapping, GFP_NOFS);
3092 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3093 inode->i_mapping, GFP_NOFS);
3094 mutex_init(&BTRFS_I(inode)->csum_mutex);
3095 mutex_init(&BTRFS_I(inode)->extent_mutex);
3096 BTRFS_I(inode)->delalloc_bytes = 0;
3097 BTRFS_I(inode)->disk_i_size = 0;
3098 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3099 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3101 dir->i_sb->s_dirt = 1;
3102 btrfs_update_inode_block_group(trans, inode);
3103 btrfs_update_inode_block_group(trans, dir);
3107 path = btrfs_alloc_path();
3109 key.objectid = inode->i_ino;
3111 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3112 datasize = btrfs_file_extent_calc_inline_size(name_len);
3113 err = btrfs_insert_empty_item(trans, root, path, &key,
3119 leaf = path->nodes[0];
3120 ei = btrfs_item_ptr(leaf, path->slots[0],
3121 struct btrfs_file_extent_item);
3122 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3123 btrfs_set_file_extent_type(leaf, ei,
3124 BTRFS_FILE_EXTENT_INLINE);
3125 ptr = btrfs_file_extent_inline_start(ei);
3126 write_extent_buffer(leaf, symname, ptr, name_len);
3127 btrfs_mark_buffer_dirty(leaf);
3128 btrfs_free_path(path);
3130 inode->i_op = &btrfs_symlink_inode_operations;
3131 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3132 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3133 btrfs_i_size_write(inode, name_len - 1);
3134 err = btrfs_update_inode(trans, root, inode);
3139 nr = trans->blocks_used;
3140 btrfs_end_transaction_throttle(trans, root);
3143 inode_dec_link_count(inode);
3146 btrfs_btree_balance_dirty(root, nr);
3150 static int btrfs_set_page_dirty(struct page *page)
3152 return __set_page_dirty_nobuffers(page);
3155 static int btrfs_permission(struct inode *inode, int mask,
3156 struct nameidata *nd)
3158 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3160 return generic_permission(inode, mask, NULL);
3163 static struct inode_operations btrfs_dir_inode_operations = {
3164 .lookup = btrfs_lookup,
3165 .create = btrfs_create,
3166 .unlink = btrfs_unlink,
3168 .mkdir = btrfs_mkdir,
3169 .rmdir = btrfs_rmdir,
3170 .rename = btrfs_rename,
3171 .symlink = btrfs_symlink,
3172 .setattr = btrfs_setattr,
3173 .mknod = btrfs_mknod,
3174 .setxattr = generic_setxattr,
3175 .getxattr = generic_getxattr,
3176 .listxattr = btrfs_listxattr,
3177 .removexattr = generic_removexattr,
3178 .permission = btrfs_permission,
3180 static struct inode_operations btrfs_dir_ro_inode_operations = {
3181 .lookup = btrfs_lookup,
3182 .permission = btrfs_permission,
3184 static struct file_operations btrfs_dir_file_operations = {
3185 .llseek = generic_file_llseek,
3186 .read = generic_read_dir,
3187 .readdir = btrfs_readdir,
3188 .unlocked_ioctl = btrfs_ioctl,
3189 #ifdef CONFIG_COMPAT
3190 .compat_ioctl = btrfs_ioctl,
3192 .release = btrfs_release_file,
3195 static struct extent_io_ops btrfs_extent_io_ops = {
3196 .fill_delalloc = run_delalloc_range,
3197 .submit_bio_hook = btrfs_submit_bio_hook,
3198 .merge_bio_hook = btrfs_merge_bio_hook,
3199 .readpage_io_hook = btrfs_readpage_io_hook,
3200 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3201 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3202 .writepage_start_hook = btrfs_writepage_start_hook,
3203 .readpage_io_failed_hook = btrfs_io_failed_hook,
3204 .set_bit_hook = btrfs_set_bit_hook,
3205 .clear_bit_hook = btrfs_clear_bit_hook,
3208 static struct address_space_operations btrfs_aops = {
3209 .readpage = btrfs_readpage,
3210 .writepage = btrfs_writepage,
3211 .writepages = btrfs_writepages,
3212 .readpages = btrfs_readpages,
3213 .sync_page = block_sync_page,
3215 .direct_IO = btrfs_direct_IO,
3216 .invalidatepage = btrfs_invalidatepage,
3217 .releasepage = btrfs_releasepage,
3218 .set_page_dirty = btrfs_set_page_dirty,
3221 static struct address_space_operations btrfs_symlink_aops = {
3222 .readpage = btrfs_readpage,
3223 .writepage = btrfs_writepage,
3224 .invalidatepage = btrfs_invalidatepage,
3225 .releasepage = btrfs_releasepage,
3228 static struct inode_operations btrfs_file_inode_operations = {
3229 .truncate = btrfs_truncate,
3230 .getattr = btrfs_getattr,
3231 .setattr = btrfs_setattr,
3232 .setxattr = generic_setxattr,
3233 .getxattr = generic_getxattr,
3234 .listxattr = btrfs_listxattr,
3235 .removexattr = generic_removexattr,
3236 .permission = btrfs_permission,
3238 static struct inode_operations btrfs_special_inode_operations = {
3239 .getattr = btrfs_getattr,
3240 .setattr = btrfs_setattr,
3241 .permission = btrfs_permission,
3243 static struct inode_operations btrfs_symlink_inode_operations = {
3244 .readlink = generic_readlink,
3245 .follow_link = page_follow_link_light,
3246 .put_link = page_put_link,
3247 .permission = btrfs_permission,